2 SDL - Simple DirectMedia Layer
3 Copyright (C) 1997-2009 Sam Lantinga
5 This library is free software; you can redistribute it and/or
6 modify it under the terms of the GNU Lesser General Public
7 License as published by the Free Software Foundation; either
8 version 2.1 of the License, or (at your option) any later version.
10 This library is distributed in the hope that it will be useful,
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 Lesser General Public License for more details.
15 You should have received a copy of the GNU Lesser General Public
16 License along with this library; if not, write to the Free Software
17 Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
22 #include "SDL_config.h"
24 /* This file contains portable memory management functions for SDL */
26 #include "SDL_stdinc.h"
30 #define LACKS_SYS_TYPES_H
32 #define LACKS_STRINGS_H
33 #define LACKS_STRING_H
34 #define LACKS_STDLIB_H
38 This is a version (aka dlmalloc) of malloc/free/realloc written by
39 Doug Lea and released to the public domain, as explained at
40 http://creativecommons.org/licenses/publicdomain. Send questions,
41 comments, complaints, performance data, etc to dl@cs.oswego.edu
43 * Version 2.8.3 Thu Sep 22 11:16:15 2005 Doug Lea (dl at gee)
45 Note: There may be an updated version of this malloc obtainable at
46 ftp://gee.cs.oswego.edu/pub/misc/malloc.c
47 Check before installing!
51 This library is all in one file to simplify the most common usage:
52 ftp it, compile it (-O3), and link it into another program. All of
53 the compile-time options default to reasonable values for use on
54 most platforms. You might later want to step through various
55 compile-time and dynamic tuning options.
57 For convenience, an include file for code using this malloc is at:
58 ftp://gee.cs.oswego.edu/pub/misc/malloc-2.8.3.h
59 You don't really need this .h file unless you call functions not
60 defined in your system include files. The .h file contains only the
61 excerpts from this file needed for using this malloc on ANSI C/C++
62 systems, so long as you haven't changed compile-time options about
63 naming and tuning parameters. If you do, then you can create your
64 own malloc.h that does include all settings by cutting at the point
65 indicated below. Note that you may already by default be using a C
66 library containing a malloc that is based on some version of this
67 malloc (for example in linux). You might still want to use the one
68 in this file to customize settings or to avoid overheads associated
69 with library versions.
73 Supported pointer/size_t representation: 4 or 8 bytes
74 size_t MUST be an unsigned type of the same width as
75 pointers. (If you are using an ancient system that declares
76 size_t as a signed type, or need it to be a different width
77 than pointers, you can use a previous release of this malloc
78 (e.g. 2.7.2) supporting these.)
80 Alignment: 8 bytes (default)
81 This suffices for nearly all current machines and C compilers.
82 However, you can define MALLOC_ALIGNMENT to be wider than this
83 if necessary (up to 128bytes), at the expense of using more space.
85 Minimum overhead per allocated chunk: 4 or 8 bytes (if 4byte sizes)
86 8 or 16 bytes (if 8byte sizes)
87 Each malloced chunk has a hidden word of overhead holding size
88 and status information, and additional cross-check word
89 if FOOTERS is defined.
91 Minimum allocated size: 4-byte ptrs: 16 bytes (including overhead)
92 8-byte ptrs: 32 bytes (including overhead)
94 Even a request for zero bytes (i.e., malloc(0)) returns a
95 pointer to something of the minimum allocatable size.
96 The maximum overhead wastage (i.e., number of extra bytes
97 allocated than were requested in malloc) is less than or equal
98 to the minimum size, except for requests >= mmap_threshold that
99 are serviced via mmap(), where the worst case wastage is about
100 32 bytes plus the remainder from a system page (the minimal
101 mmap unit); typically 4096 or 8192 bytes.
103 Security: static-safe; optionally more or less
104 The "security" of malloc refers to the ability of malicious
105 code to accentuate the effects of errors (for example, freeing
106 space that is not currently malloc'ed or overwriting past the
107 ends of chunks) in code that calls malloc. This malloc
108 guarantees not to modify any memory locations below the base of
109 heap, i.e., static variables, even in the presence of usage
110 errors. The routines additionally detect most improper frees
111 and reallocs. All this holds as long as the static bookkeeping
112 for malloc itself is not corrupted by some other means. This
113 is only one aspect of security -- these checks do not, and
114 cannot, detect all possible programming errors.
116 If FOOTERS is defined nonzero, then each allocated chunk
117 carries an additional check word to verify that it was malloced
118 from its space. These check words are the same within each
119 execution of a program using malloc, but differ across
120 executions, so externally crafted fake chunks cannot be
121 freed. This improves security by rejecting frees/reallocs that
122 could corrupt heap memory, in addition to the checks preventing
123 writes to statics that are always on. This may further improve
124 security at the expense of time and space overhead. (Note that
125 FOOTERS may also be worth using with MSPACES.)
127 By default detected errors cause the program to abort (calling
128 "abort()"). You can override this to instead proceed past
129 errors by defining PROCEED_ON_ERROR. In this case, a bad free
130 has no effect, and a malloc that encounters a bad address
131 caused by user overwrites will ignore the bad address by
132 dropping pointers and indices to all known memory. This may
133 be appropriate for programs that should continue if at all
134 possible in the face of programming errors, although they may
135 run out of memory because dropped memory is never reclaimed.
137 If you don't like either of these options, you can define
138 CORRUPTION_ERROR_ACTION and USAGE_ERROR_ACTION to do anything
139 else. And if if you are sure that your program using malloc has
140 no errors or vulnerabilities, you can define INSECURE to 1,
141 which might (or might not) provide a small performance improvement.
143 Thread-safety: NOT thread-safe unless USE_LOCKS defined
144 When USE_LOCKS is defined, each public call to malloc, free,
145 etc is surrounded with either a pthread mutex or a win32
146 spinlock (depending on WIN32). This is not especially fast, and
147 can be a major bottleneck. It is designed only to provide
148 minimal protection in concurrent environments, and to provide a
149 basis for extensions. If you are using malloc in a concurrent
150 program, consider instead using ptmalloc, which is derived from
151 a version of this malloc. (See http://www.malloc.de).
153 System requirements: Any combination of MORECORE and/or MMAP/MUNMAP
154 This malloc can use unix sbrk or any emulation (invoked using
155 the CALL_MORECORE macro) and/or mmap/munmap or any emulation
156 (invoked using CALL_MMAP/CALL_MUNMAP) to get and release system
157 memory. On most unix systems, it tends to work best if both
158 MORECORE and MMAP are enabled. On Win32, it uses emulations
159 based on VirtualAlloc. It also uses common C library functions
162 Compliance: I believe it is compliant with the Single Unix Specification
163 (See http://www.unix.org). Also SVID/XPG, ANSI C, and probably
166 * Overview of algorithms
168 This is not the fastest, most space-conserving, most portable, or
169 most tunable malloc ever written. However it is among the fastest
170 while also being among the most space-conserving, portable and
171 tunable. Consistent balance across these factors results in a good
172 general-purpose allocator for malloc-intensive programs.
174 In most ways, this malloc is a best-fit allocator. Generally, it
175 chooses the best-fitting existing chunk for a request, with ties
176 broken in approximately least-recently-used order. (This strategy
177 normally maintains low fragmentation.) However, for requests less
178 than 256bytes, it deviates from best-fit when there is not an
179 exactly fitting available chunk by preferring to use space adjacent
180 to that used for the previous small request, as well as by breaking
181 ties in approximately most-recently-used order. (These enhance
182 locality of series of small allocations.) And for very large requests
183 (>= 256Kb by default), it relies on system memory mapping
184 facilities, if supported. (This helps avoid carrying around and
185 possibly fragmenting memory used only for large chunks.)
187 All operations (except malloc_stats and mallinfo) have execution
188 times that are bounded by a constant factor of the number of bits in
189 a size_t, not counting any clearing in calloc or copying in realloc,
190 or actions surrounding MORECORE and MMAP that have times
191 proportional to the number of non-contiguous regions returned by
192 system allocation routines, which is often just 1.
194 The implementation is not very modular and seriously overuses
195 macros. Perhaps someday all C compilers will do as good a job
196 inlining modular code as can now be done by brute-force expansion,
197 but now, enough of them seem not to.
199 Some compilers issue a lot of warnings about code that is
200 dead/unreachable only on some platforms, and also about intentional
201 uses of negation on unsigned types. All known cases of each can be
204 For a longer but out of date high-level description, see
205 http://gee.cs.oswego.edu/dl/html/malloc.html
208 If MSPACES is defined, then in addition to malloc, free, etc.,
209 this file also defines mspace_malloc, mspace_free, etc. These
210 are versions of malloc routines that take an "mspace" argument
211 obtained using create_mspace, to control all internal bookkeeping.
212 If ONLY_MSPACES is defined, only these versions are compiled.
213 So if you would like to use this allocator for only some allocations,
214 and your system malloc for others, you can compile with
215 ONLY_MSPACES and then do something like...
216 static mspace mymspace = create_mspace(0,0); // for example
217 #define mymalloc(bytes) mspace_malloc(mymspace, bytes)
219 (Note: If you only need one instance of an mspace, you can instead
220 use "USE_DL_PREFIX" to relabel the global malloc.)
222 You can similarly create thread-local allocators by storing
223 mspaces as thread-locals. For example:
224 static __thread mspace tlms = 0;
225 void* tlmalloc(size_t bytes) {
226 if (tlms == 0) tlms = create_mspace(0, 0);
227 return mspace_malloc(tlms, bytes);
229 void tlfree(void* mem) { mspace_free(tlms, mem); }
231 Unless FOOTERS is defined, each mspace is completely independent.
232 You cannot allocate from one and free to another (although
233 conformance is only weakly checked, so usage errors are not always
234 caught). If FOOTERS is defined, then each chunk carries around a tag
235 indicating its originating mspace, and frees are directed to their
238 ------------------------- Compile-time options ---------------------------
240 Be careful in setting #define values for numerical constants of type
241 size_t. On some systems, literal values are not automatically extended
242 to size_t precision unless they are explicitly casted.
244 WIN32 default: defined if _WIN32 defined
245 Defining WIN32 sets up defaults for MS environment and compilers.
246 Otherwise defaults are for unix.
248 MALLOC_ALIGNMENT default: (size_t)8
249 Controls the minimum alignment for malloc'ed chunks. It must be a
250 power of two and at least 8, even on machines for which smaller
251 alignments would suffice. It may be defined as larger than this
252 though. Note however that code and data structures are optimized for
253 the case of 8-byte alignment.
255 MSPACES default: 0 (false)
256 If true, compile in support for independent allocation spaces.
257 This is only supported if HAVE_MMAP is true.
259 ONLY_MSPACES default: 0 (false)
260 If true, only compile in mspace versions, not regular versions.
262 USE_LOCKS default: 0 (false)
263 Causes each call to each public routine to be surrounded with
264 pthread or WIN32 mutex lock/unlock. (If set true, this can be
265 overridden on a per-mspace basis for mspace versions.)
268 If true, provide extra checking and dispatching by placing
269 information in the footers of allocated chunks. This adds
270 space and time overhead.
273 If true, omit checks for usage errors and heap space overwrites.
275 USE_DL_PREFIX default: NOT defined
276 Causes compiler to prefix all public routines with the string 'dl'.
277 This can be useful when you only want to use this malloc in one part
278 of a program, using your regular system malloc elsewhere.
280 ABORT default: defined as abort()
281 Defines how to abort on failed checks. On most systems, a failed
282 check cannot die with an "assert" or even print an informative
283 message, because the underlying print routines in turn call malloc,
284 which will fail again. Generally, the best policy is to simply call
285 abort(). It's not very useful to do more than this because many
286 errors due to overwriting will show up as address faults (null, odd
287 addresses etc) rather than malloc-triggered checks, so will also
288 abort. Also, most compilers know that abort() does not return, so
289 can better optimize code conditionally calling it.
291 PROCEED_ON_ERROR default: defined as 0 (false)
292 Controls whether detected bad addresses cause them to bypassed
293 rather than aborting. If set, detected bad arguments to free and
294 realloc are ignored. And all bookkeeping information is zeroed out
295 upon a detected overwrite of freed heap space, thus losing the
296 ability to ever return it from malloc again, but enabling the
297 application to proceed. If PROCEED_ON_ERROR is defined, the
298 static variable malloc_corruption_error_count is compiled in
299 and can be examined to see if errors have occurred. This option
300 generates slower code than the default abort policy.
302 DEBUG default: NOT defined
303 The DEBUG setting is mainly intended for people trying to modify
304 this code or diagnose problems when porting to new platforms.
305 However, it may also be able to better isolate user errors than just
306 using runtime checks. The assertions in the check routines spell
307 out in more detail the assumptions and invariants underlying the
308 algorithms. The checking is fairly extensive, and will slow down
309 execution noticeably. Calling malloc_stats or mallinfo with DEBUG
310 set will attempt to check every non-mmapped allocated and free chunk
311 in the course of computing the summaries.
313 ABORT_ON_ASSERT_FAILURE default: defined as 1 (true)
314 Debugging assertion failures can be nearly impossible if your
315 version of the assert macro causes malloc to be called, which will
316 lead to a cascade of further failures, blowing the runtime stack.
317 ABORT_ON_ASSERT_FAILURE cause assertions failures to call abort(),
318 which will usually make debugging easier.
320 MALLOC_FAILURE_ACTION default: sets errno to ENOMEM, or no-op on win32
321 The action to take before "return 0" when malloc fails to be able to
322 return memory because there is none available.
324 HAVE_MORECORE default: 1 (true) unless win32 or ONLY_MSPACES
325 True if this system supports sbrk or an emulation of it.
327 MORECORE default: sbrk
328 The name of the sbrk-style system routine to call to obtain more
329 memory. See below for guidance on writing custom MORECORE
330 functions. The type of the argument to sbrk/MORECORE varies across
331 systems. It cannot be size_t, because it supports negative
332 arguments, so it is normally the signed type of the same width as
333 size_t (sometimes declared as "intptr_t"). It doesn't much matter
334 though. Internally, we only call it with arguments less than half
335 the max value of a size_t, which should work across all reasonable
336 possibilities, although sometimes generating compiler warnings. See
337 near the end of this file for guidelines for creating a custom
340 MORECORE_CONTIGUOUS default: 1 (true)
341 If true, take advantage of fact that consecutive calls to MORECORE
342 with positive arguments always return contiguous increasing
343 addresses. This is true of unix sbrk. It does not hurt too much to
344 set it true anyway, since malloc copes with non-contiguities.
345 Setting it false when definitely non-contiguous saves time
346 and possibly wasted space it would take to discover this though.
348 MORECORE_CANNOT_TRIM default: NOT defined
349 True if MORECORE cannot release space back to the system when given
350 negative arguments. This is generally necessary only if you are
351 using a hand-crafted MORECORE function that cannot handle negative
354 HAVE_MMAP default: 1 (true)
355 True if this system supports mmap or an emulation of it. If so, and
356 HAVE_MORECORE is not true, MMAP is used for all system
357 allocation. If set and HAVE_MORECORE is true as well, MMAP is
358 primarily used to directly allocate very large blocks. It is also
359 used as a backup strategy in cases where MORECORE fails to provide
360 space from system. Note: A single call to MUNMAP is assumed to be
361 able to unmap memory that may have be allocated using multiple calls
362 to MMAP, so long as they are adjacent.
364 HAVE_MREMAP default: 1 on linux, else 0
365 If true realloc() uses mremap() to re-allocate large blocks and
366 extend or shrink allocation spaces.
368 MMAP_CLEARS default: 1 on unix
369 True if mmap clears memory so calloc doesn't need to. This is true
370 for standard unix mmap using /dev/zero.
372 USE_BUILTIN_FFS default: 0 (i.e., not used)
373 Causes malloc to use the builtin ffs() function to compute indices.
374 Some compilers may recognize and intrinsify ffs to be faster than the
375 supplied C version. Also, the case of x86 using gcc is special-cased
376 to an asm instruction, so is already as fast as it can be, and so
377 this setting has no effect. (On most x86s, the asm version is only
378 slightly faster than the C version.)
380 malloc_getpagesize default: derive from system includes, or 4096.
381 The system page size. To the extent possible, this malloc manages
382 memory from the system in page-size units. This may be (and
383 usually is) a function rather than a constant. This is ignored
384 if WIN32, where page size is determined using getSystemInfo during
387 USE_DEV_RANDOM default: 0 (i.e., not used)
388 Causes malloc to use /dev/random to initialize secure magic seed for
389 stamping footers. Otherwise, the current time is used.
391 NO_MALLINFO default: 0
392 If defined, don't compile "mallinfo". This can be a simple way
393 of dealing with mismatches between system declarations and
396 MALLINFO_FIELD_TYPE default: size_t
397 The type of the fields in the mallinfo struct. This was originally
398 defined as "int" in SVID etc, but is more usefully defined as
399 size_t. The value is used only if HAVE_USR_INCLUDE_MALLOC_H is not set
401 REALLOC_ZERO_BYTES_FREES default: not defined
402 This should be set if a call to realloc with zero bytes should
403 be the same as a call to free. Some people think it should. Otherwise,
404 since this malloc returns a unique pointer for malloc(0), so does
407 LACKS_UNISTD_H, LACKS_FCNTL_H, LACKS_SYS_PARAM_H, LACKS_SYS_MMAN_H
408 LACKS_STRINGS_H, LACKS_STRING_H, LACKS_SYS_TYPES_H, LACKS_ERRNO_H
409 LACKS_STDLIB_H default: NOT defined unless on WIN32
410 Define these if your system does not have these header files.
411 You might need to manually insert some of the declarations they provide.
413 DEFAULT_GRANULARITY default: page size if MORECORE_CONTIGUOUS,
414 system_info.dwAllocationGranularity in WIN32,
416 Also settable using mallopt(M_GRANULARITY, x)
417 The unit for allocating and deallocating memory from the system. On
418 most systems with contiguous MORECORE, there is no reason to
419 make this more than a page. However, systems with MMAP tend to
420 either require or encourage larger granularities. You can increase
421 this value to prevent system allocation functions to be called so
422 often, especially if they are slow. The value must be at least one
423 page and must be a power of two. Setting to 0 causes initialization
424 to either page size or win32 region size. (Note: In previous
425 versions of malloc, the equivalent of this option was called
428 DEFAULT_TRIM_THRESHOLD default: 2MB
429 Also settable using mallopt(M_TRIM_THRESHOLD, x)
430 The maximum amount of unused top-most memory to keep before
431 releasing via malloc_trim in free(). Automatic trimming is mainly
432 useful in long-lived programs using contiguous MORECORE. Because
433 trimming via sbrk can be slow on some systems, and can sometimes be
434 wasteful (in cases where programs immediately afterward allocate
435 more large chunks) the value should be high enough so that your
436 overall system performance would improve by releasing this much
437 memory. As a rough guide, you might set to a value close to the
438 average size of a process (program) running on your system.
439 Releasing this much memory would allow such a process to run in
440 memory. Generally, it is worth tuning trim thresholds when a
441 program undergoes phases where several large chunks are allocated
442 and released in ways that can reuse each other's storage, perhaps
443 mixed with phases where there are no such chunks at all. The trim
444 value must be greater than page size to have any useful effect. To
445 disable trimming completely, you can set to MAX_SIZE_T. Note that the trick
446 some people use of mallocing a huge space and then freeing it at
447 program startup, in an attempt to reserve system memory, doesn't
448 have the intended effect under automatic trimming, since that memory
449 will immediately be returned to the system.
451 DEFAULT_MMAP_THRESHOLD default: 256K
452 Also settable using mallopt(M_MMAP_THRESHOLD, x)
453 The request size threshold for using MMAP to directly service a
454 request. Requests of at least this size that cannot be allocated
455 using already-existing space will be serviced via mmap. (If enough
456 normal freed space already exists it is used instead.) Using mmap
457 segregates relatively large chunks of memory so that they can be
458 individually obtained and released from the host system. A request
459 serviced through mmap is never reused by any other request (at least
460 not directly; the system may just so happen to remap successive
461 requests to the same locations). Segregating space in this way has
462 the benefits that: Mmapped space can always be individually released
463 back to the system, which helps keep the system level memory demands
464 of a long-lived program low. Also, mapped memory doesn't become
465 `locked' between other chunks, as can happen with normally allocated
466 chunks, which means that even trimming via malloc_trim would not
467 release them. However, it has the disadvantage that the space
468 cannot be reclaimed, consolidated, and then used to service later
469 requests, as happens with normal chunks. The advantages of mmap
470 nearly always outweigh disadvantages for "large" chunks, but the
471 value of "large" may vary across systems. The default is an
472 empirically derived value that works well in most systems. You can
473 disable mmap by setting to MAX_SIZE_T.
483 #define WIN32_LEAN_AND_MEAN
486 #define HAVE_MORECORE 0
487 #define LACKS_UNISTD_H
488 #define LACKS_SYS_PARAM_H
489 #define LACKS_SYS_MMAN_H
490 #define LACKS_STRING_H
491 #define LACKS_STRINGS_H
492 #define LACKS_SYS_TYPES_H
493 #define LACKS_ERRNO_H
494 #define LACKS_FCNTL_H
495 #define MALLOC_FAILURE_ACTION
496 #define MMAP_CLEARS 0 /* WINCE and some others apparently don't clear */
499 #if defined(DARWIN) || defined(_DARWIN)
500 /* Mac OSX docs advise not to use sbrk; it seems better to use mmap */
501 #ifndef HAVE_MORECORE
502 #define HAVE_MORECORE 0
504 #endif /* HAVE_MORECORE */
507 #ifndef LACKS_SYS_TYPES_H
508 #include <sys/types.h> /* For size_t */
509 #endif /* LACKS_SYS_TYPES_H */
511 /* The maximum possible size_t value has all bits set */
512 #define MAX_SIZE_T (~(size_t)0)
515 #define ONLY_MSPACES 0
516 #endif /* ONLY_MSPACES */
520 #else /* ONLY_MSPACES */
522 #endif /* ONLY_MSPACES */
524 #ifndef MALLOC_ALIGNMENT
525 #define MALLOC_ALIGNMENT ((size_t)8U)
526 #endif /* MALLOC_ALIGNMENT */
531 #define ABORT abort()
533 #ifndef ABORT_ON_ASSERT_FAILURE
534 #define ABORT_ON_ASSERT_FAILURE 1
535 #endif /* ABORT_ON_ASSERT_FAILURE */
536 #ifndef PROCEED_ON_ERROR
537 #define PROCEED_ON_ERROR 0
538 #endif /* PROCEED_ON_ERROR */
541 #endif /* USE_LOCKS */
544 #endif /* INSECURE */
547 #endif /* HAVE_MMAP */
549 #define MMAP_CLEARS 1
550 #endif /* MMAP_CLEARS */
553 #define HAVE_MREMAP 1
555 #define HAVE_MREMAP 0
557 #endif /* HAVE_MREMAP */
558 #ifndef MALLOC_FAILURE_ACTION
559 #define MALLOC_FAILURE_ACTION errno = ENOMEM;
560 #endif /* MALLOC_FAILURE_ACTION */
561 #ifndef HAVE_MORECORE
563 #define HAVE_MORECORE 0
564 #else /* ONLY_MSPACES */
565 #define HAVE_MORECORE 1
566 #endif /* ONLY_MSPACES */
567 #endif /* HAVE_MORECORE */
569 #define MORECORE_CONTIGUOUS 0
570 #else /* !HAVE_MORECORE */
572 #define MORECORE sbrk
573 #endif /* MORECORE */
574 #ifndef MORECORE_CONTIGUOUS
575 #define MORECORE_CONTIGUOUS 1
576 #endif /* MORECORE_CONTIGUOUS */
577 #endif /* HAVE_MORECORE */
578 #ifndef DEFAULT_GRANULARITY
579 #if MORECORE_CONTIGUOUS
580 #define DEFAULT_GRANULARITY (0) /* 0 means to compute in init_mparams */
581 #else /* MORECORE_CONTIGUOUS */
582 #define DEFAULT_GRANULARITY ((size_t)64U * (size_t)1024U)
583 #endif /* MORECORE_CONTIGUOUS */
584 #endif /* DEFAULT_GRANULARITY */
585 #ifndef DEFAULT_TRIM_THRESHOLD
586 #ifndef MORECORE_CANNOT_TRIM
587 #define DEFAULT_TRIM_THRESHOLD ((size_t)2U * (size_t)1024U * (size_t)1024U)
588 #else /* MORECORE_CANNOT_TRIM */
589 #define DEFAULT_TRIM_THRESHOLD MAX_SIZE_T
590 #endif /* MORECORE_CANNOT_TRIM */
591 #endif /* DEFAULT_TRIM_THRESHOLD */
592 #ifndef DEFAULT_MMAP_THRESHOLD
594 #define DEFAULT_MMAP_THRESHOLD ((size_t)256U * (size_t)1024U)
595 #else /* HAVE_MMAP */
596 #define DEFAULT_MMAP_THRESHOLD MAX_SIZE_T
597 #endif /* HAVE_MMAP */
598 #endif /* DEFAULT_MMAP_THRESHOLD */
599 #ifndef USE_BUILTIN_FFS
600 #define USE_BUILTIN_FFS 0
601 #endif /* USE_BUILTIN_FFS */
602 #ifndef USE_DEV_RANDOM
603 #define USE_DEV_RANDOM 0
604 #endif /* USE_DEV_RANDOM */
606 #define NO_MALLINFO 0
607 #endif /* NO_MALLINFO */
608 #ifndef MALLINFO_FIELD_TYPE
609 #define MALLINFO_FIELD_TYPE size_t
610 #endif /* MALLINFO_FIELD_TYPE */
612 #define memset SDL_memset
613 #define memcpy SDL_memcpy
614 #define malloc SDL_malloc
615 #define calloc SDL_calloc
616 #define realloc SDL_realloc
617 #define free SDL_free
620 mallopt tuning options. SVID/XPG defines four standard parameter
621 numbers for mallopt, normally defined in malloc.h. None of these
622 are used in this malloc, so setting them has no effect. But this
623 malloc does support the following options.
626 #define M_TRIM_THRESHOLD (-1)
627 #define M_GRANULARITY (-2)
628 #define M_MMAP_THRESHOLD (-3)
630 /* ------------------------ Mallinfo declarations ------------------------ */
634 This version of malloc supports the standard SVID/XPG mallinfo
635 routine that returns a struct containing usage properties and
636 statistics. It should work on any system that has a
637 /usr/include/malloc.h defining struct mallinfo. The main
638 declaration needed is the mallinfo struct that is returned (by-copy)
639 by mallinfo(). The malloinfo struct contains a bunch of fields that
640 are not even meaningful in this version of malloc. These fields are
641 are instead filled by mallinfo() with other numbers that might be of
644 HAVE_USR_INCLUDE_MALLOC_H should be set if you have a
645 /usr/include/malloc.h file that includes a declaration of struct
646 mallinfo. If so, it is included; else a compliant version is
647 declared below. These must be precisely the same for mallinfo() to
648 work. The original SVID version of this struct, defined on most
649 systems with mallinfo, declares all fields as ints. But some others
650 define as unsigned long. If your system defines the fields using a
651 type of different width than listed here, you MUST #include your
652 system version and #define HAVE_USR_INCLUDE_MALLOC_H.
655 /* #define HAVE_USR_INCLUDE_MALLOC_H */
657 #ifdef HAVE_USR_INCLUDE_MALLOC_H
658 #include "/usr/include/malloc.h"
659 #else /* HAVE_USR_INCLUDE_MALLOC_H */
662 MALLINFO_FIELD_TYPE arena; /* non-mmapped space allocated from system */
663 MALLINFO_FIELD_TYPE ordblks; /* number of free chunks */
664 MALLINFO_FIELD_TYPE smblks; /* always 0 */
665 MALLINFO_FIELD_TYPE hblks; /* always 0 */
666 MALLINFO_FIELD_TYPE hblkhd; /* space in mmapped regions */
667 MALLINFO_FIELD_TYPE usmblks; /* maximum total allocated space */
668 MALLINFO_FIELD_TYPE fsmblks; /* always 0 */
669 MALLINFO_FIELD_TYPE uordblks; /* total allocated space */
670 MALLINFO_FIELD_TYPE fordblks; /* total free space */
671 MALLINFO_FIELD_TYPE keepcost; /* releasable (via malloc_trim) space */
674 #endif /* HAVE_USR_INCLUDE_MALLOC_H */
675 #endif /* NO_MALLINFO */
679 #endif /* __cplusplus */
683 /* ------------------- Declarations of public routines ------------------- */
685 #ifndef USE_DL_PREFIX
686 #define dlcalloc calloc
688 #define dlmalloc malloc
689 #define dlmemalign memalign
690 #define dlrealloc realloc
691 #define dlvalloc valloc
692 #define dlpvalloc pvalloc
693 #define dlmallinfo mallinfo
694 #define dlmallopt mallopt
695 #define dlmalloc_trim malloc_trim
696 #define dlmalloc_stats malloc_stats
697 #define dlmalloc_usable_size malloc_usable_size
698 #define dlmalloc_footprint malloc_footprint
699 #define dlmalloc_max_footprint malloc_max_footprint
700 #define dlindependent_calloc independent_calloc
701 #define dlindependent_comalloc independent_comalloc
702 #endif /* USE_DL_PREFIX */
707 Returns a pointer to a newly allocated chunk of at least n bytes, or
708 null if no space is available, in which case errno is set to ENOMEM
711 If n is zero, malloc returns a minimum-sized chunk. (The minimum
712 size is 16 bytes on most 32bit systems, and 32 bytes on 64bit
713 systems.) Note that size_t is an unsigned type, so calls with
714 arguments that would be negative if signed are interpreted as
715 requests for huge amounts of space, which will often fail. The
716 maximum supported value of n differs across systems, but is in all
717 cases less than the maximum representable value of a size_t.
719 void* dlmalloc(size_t);
723 Releases the chunk of memory pointed to by p, that had been previously
724 allocated using malloc or a related routine such as realloc.
725 It has no effect if p is null. If p was not malloced or already
726 freed, free(p) will by default cause the current program to abort.
731 calloc(size_t n_elements, size_t element_size);
732 Returns a pointer to n_elements * element_size bytes, with all locations
735 void* dlcalloc(size_t, size_t);
738 realloc(void* p, size_t n)
739 Returns a pointer to a chunk of size n that contains the same data
740 as does chunk p up to the minimum of (n, p's size) bytes, or null
741 if no space is available.
743 The returned pointer may or may not be the same as p. The algorithm
744 prefers extending p in most cases when possible, otherwise it
745 employs the equivalent of a malloc-copy-free sequence.
747 If p is null, realloc is equivalent to malloc.
749 If space is not available, realloc returns null, errno is set (if on
750 ANSI) and p is NOT freed.
752 if n is for fewer bytes than already held by p, the newly unused
753 space is lopped off and freed if possible. realloc with a size
754 argument of zero (re)allocates a minimum-sized chunk.
756 The old unix realloc convention of allowing the last-free'd chunk
757 to be used as an argument to realloc is not supported.
760 void* dlrealloc(void*, size_t);
763 memalign(size_t alignment, size_t n);
764 Returns a pointer to a newly allocated chunk of n bytes, aligned
765 in accord with the alignment argument.
767 The alignment argument should be a power of two. If the argument is
768 not a power of two, the nearest greater power is used.
769 8-byte alignment is guaranteed by normal malloc calls, so don't
770 bother calling memalign with an argument of 8 or less.
772 Overreliance on memalign is a sure way to fragment space.
774 void* dlmemalign(size_t, size_t);
778 Equivalent to memalign(pagesize, n), where pagesize is the page
779 size of the system. If the pagesize is unknown, 4096 is used.
781 void* dlvalloc(size_t);
784 mallopt(int parameter_number, int parameter_value)
785 Sets tunable parameters The format is to provide a
786 (parameter-number, parameter-value) pair. mallopt then sets the
787 corresponding parameter to the argument value if it can (i.e., so
788 long as the value is meaningful), and returns 1 if successful else
789 0. SVID/XPG/ANSI defines four standard param numbers for mallopt,
790 normally defined in malloc.h. None of these are use in this malloc,
791 so setting them has no effect. But this malloc also supports other
792 options in mallopt. See below for details. Briefly, supported
793 parameters are as follows (listed defaults are for "typical"
796 Symbol param # default allowed param values
797 M_TRIM_THRESHOLD -1 2*1024*1024 any (MAX_SIZE_T disables)
798 M_GRANULARITY -2 page size any power of 2 >= page size
799 M_MMAP_THRESHOLD -3 256*1024 any (or 0 if no MMAP support)
801 int dlmallopt(int, int);
805 Returns the number of bytes obtained from the system. The total
806 number of bytes allocated by malloc, realloc etc., is less than this
807 value. Unlike mallinfo, this function returns only a precomputed
808 result, so can be called frequently to monitor memory consumption.
809 Even if locks are otherwise defined, this function does not use them,
810 so results might not be up to date.
812 size_t dlmalloc_footprint(void);
815 malloc_max_footprint();
816 Returns the maximum number of bytes obtained from the system. This
817 value will be greater than current footprint if deallocated space
818 has been reclaimed by the system. The peak number of bytes allocated
819 by malloc, realloc etc., is less than this value. Unlike mallinfo,
820 this function returns only a precomputed result, so can be called
821 frequently to monitor memory consumption. Even if locks are
822 otherwise defined, this function does not use them, so results might
825 size_t dlmalloc_max_footprint(void);
830 Returns (by copy) a struct containing various summary statistics:
832 arena: current total non-mmapped bytes allocated from system
833 ordblks: the number of free chunks
835 hblks: current number of mmapped regions
836 hblkhd: total bytes held in mmapped regions
837 usmblks: the maximum total allocated space. This will be greater
838 than current total if trimming has occurred.
840 uordblks: current total allocated space (normal or mmapped)
841 fordblks: total free space
842 keepcost: the maximum number of bytes that could ideally be released
843 back to system via malloc_trim. ("ideally" means that
844 it ignores page restrictions etc.)
846 Because these fields are ints, but internal bookkeeping may
847 be kept as longs, the reported values may wrap around zero and
850 struct mallinfo dlmallinfo(void);
851 #endif /* NO_MALLINFO */
854 independent_calloc(size_t n_elements, size_t element_size, void* chunks[]);
856 independent_calloc is similar to calloc, but instead of returning a
857 single cleared space, it returns an array of pointers to n_elements
858 independent elements that can hold contents of size elem_size, each
859 of which starts out cleared, and can be independently freed,
860 realloc'ed etc. The elements are guaranteed to be adjacently
861 allocated (this is not guaranteed to occur with multiple callocs or
862 mallocs), which may also improve cache locality in some
865 The "chunks" argument is optional (i.e., may be null, which is
866 probably the most typical usage). If it is null, the returned array
867 is itself dynamically allocated and should also be freed when it is
868 no longer needed. Otherwise, the chunks array must be of at least
869 n_elements in length. It is filled in with the pointers to the
872 In either case, independent_calloc returns this pointer array, or
873 null if the allocation failed. If n_elements is zero and "chunks"
874 is null, it returns a chunk representing an array with zero elements
875 (which should be freed if not wanted).
877 Each element must be individually freed when it is no longer
878 needed. If you'd like to instead be able to free all at once, you
879 should instead use regular calloc and assign pointers into this
880 space to represent elements. (In this case though, you cannot
881 independently free elements.)
883 independent_calloc simplifies and speeds up implementations of many
884 kinds of pools. It may also be useful when constructing large data
885 structures that initially have a fixed number of fixed-sized nodes,
886 but the number is not known at compile time, and some of the nodes
887 may later need to be freed. For example:
889 struct Node { int item; struct Node* next; };
891 struct Node* build_list() {
893 int n = read_number_of_nodes_needed();
894 if (n <= 0) return 0;
895 pool = (struct Node**)(independent_calloc(n, sizeof(struct Node), 0);
896 if (pool == 0) die();
897 // organize into a linked list...
898 struct Node* first = pool[0];
899 for (i = 0; i < n-1; ++i)
900 pool[i]->next = pool[i+1];
901 free(pool); // Can now free the array (or not, if it is needed later)
905 void** dlindependent_calloc(size_t, size_t, void**);
908 independent_comalloc(size_t n_elements, size_t sizes[], void* chunks[]);
910 independent_comalloc allocates, all at once, a set of n_elements
911 chunks with sizes indicated in the "sizes" array. It returns
912 an array of pointers to these elements, each of which can be
913 independently freed, realloc'ed etc. The elements are guaranteed to
914 be adjacently allocated (this is not guaranteed to occur with
915 multiple callocs or mallocs), which may also improve cache locality
916 in some applications.
918 The "chunks" argument is optional (i.e., may be null). If it is null
919 the returned array is itself dynamically allocated and should also
920 be freed when it is no longer needed. Otherwise, the chunks array
921 must be of at least n_elements in length. It is filled in with the
922 pointers to the chunks.
924 In either case, independent_comalloc returns this pointer array, or
925 null if the allocation failed. If n_elements is zero and chunks is
926 null, it returns a chunk representing an array with zero elements
927 (which should be freed if not wanted).
929 Each element must be individually freed when it is no longer
930 needed. If you'd like to instead be able to free all at once, you
931 should instead use a single regular malloc, and assign pointers at
932 particular offsets in the aggregate space. (In this case though, you
933 cannot independently free elements.)
935 independent_comallac differs from independent_calloc in that each
936 element may have a different size, and also that it does not
937 automatically clear elements.
939 independent_comalloc can be used to speed up allocation in cases
940 where several structs or objects must always be allocated at the
941 same time. For example:
946 void send_message(char* msg) {
947 int msglen = strlen(msg);
948 size_t sizes[3] = { sizeof(struct Head), msglen, sizeof(struct Foot) };
950 if (independent_comalloc(3, sizes, chunks) == 0)
952 struct Head* head = (struct Head*)(chunks[0]);
953 char* body = (char*)(chunks[1]);
954 struct Foot* foot = (struct Foot*)(chunks[2]);
958 In general though, independent_comalloc is worth using only for
959 larger values of n_elements. For small values, you probably won't
960 detect enough difference from series of malloc calls to bother.
962 Overuse of independent_comalloc can increase overall memory usage,
963 since it cannot reuse existing noncontiguous small chunks that
964 might be available for some of the elements.
966 void** dlindependent_comalloc(size_t, size_t*, void**);
971 Equivalent to valloc(minimum-page-that-holds(n)), that is,
972 round up n to nearest pagesize.
974 void* dlpvalloc(size_t);
977 malloc_trim(size_t pad);
979 If possible, gives memory back to the system (via negative arguments
980 to sbrk) if there is unused memory at the `high' end of the malloc
981 pool or in unused MMAP segments. You can call this after freeing
982 large blocks of memory to potentially reduce the system-level memory
983 requirements of a program. However, it cannot guarantee to reduce
984 memory. Under some allocation patterns, some large free blocks of
985 memory will be locked between two used chunks, so they cannot be
986 given back to the system.
988 The `pad' argument to malloc_trim represents the amount of free
989 trailing space to leave untrimmed. If this argument is zero, only
990 the minimum amount of memory to maintain internal data structures
991 will be left. Non-zero arguments can be supplied to maintain enough
992 trailing space to service future expected allocations without having
993 to re-obtain memory from the system.
995 Malloc_trim returns 1 if it actually released any memory, else 0.
997 int dlmalloc_trim(size_t);
1000 malloc_usable_size(void* p);
1002 Returns the number of bytes you can actually use in
1003 an allocated chunk, which may be more than you requested (although
1004 often not) due to alignment and minimum size constraints.
1005 You can use this many bytes without worrying about
1006 overwriting other allocated objects. This is not a particularly great
1007 programming practice. malloc_usable_size can be more useful in
1008 debugging and assertions, for example:
1011 assert(malloc_usable_size(p) >= 256);
1013 size_t dlmalloc_usable_size(void*);
1017 Prints on stderr the amount of space obtained from the system (both
1018 via sbrk and mmap), the maximum amount (which may be more than
1019 current if malloc_trim and/or munmap got called), and the current
1020 number of bytes allocated via malloc (or realloc, etc) but not yet
1021 freed. Note that this is the number of bytes allocated, not the
1022 number requested. It will be larger than the number requested
1023 because of alignment and bookkeeping overhead. Because it includes
1024 alignment wastage as being in use, this figure may be greater than
1025 zero even when no user-level chunks are allocated.
1027 The reported current and maximum system memory can be inaccurate if
1028 a program makes other calls to system memory allocation functions
1029 (normally sbrk) outside of malloc.
1031 malloc_stats prints only the most commonly interesting statistics.
1032 More information can be obtained by calling mallinfo.
1034 void dlmalloc_stats(void);
1036 #endif /* ONLY_MSPACES */
1041 mspace is an opaque type representing an independent
1042 region of space that supports mspace_malloc, etc.
1044 typedef void* mspace;
1047 create_mspace creates and returns a new independent space with the
1048 given initial capacity, or, if 0, the default granularity size. It
1049 returns null if there is no system memory available to create the
1050 space. If argument locked is non-zero, the space uses a separate
1051 lock to control access. The capacity of the space will grow
1052 dynamically as needed to service mspace_malloc requests. You can
1053 control the sizes of incremental increases of this space by
1054 compiling with a different DEFAULT_GRANULARITY or dynamically
1055 setting with mallopt(M_GRANULARITY, value).
1057 mspace create_mspace(size_t capacity, int locked);
1060 destroy_mspace destroys the given space, and attempts to return all
1061 of its memory back to the system, returning the total number of
1062 bytes freed. After destruction, the results of access to all memory
1063 used by the space become undefined.
1065 size_t destroy_mspace(mspace msp);
1068 create_mspace_with_base uses the memory supplied as the initial base
1069 of a new mspace. Part (less than 128*sizeof(size_t) bytes) of this
1070 space is used for bookkeeping, so the capacity must be at least this
1071 large. (Otherwise 0 is returned.) When this initial space is
1072 exhausted, additional memory will be obtained from the system.
1073 Destroying this space will deallocate all additionally allocated
1074 space (if possible) but not the initial base.
1076 mspace create_mspace_with_base(void* base, size_t capacity, int locked);
1079 mspace_malloc behaves as malloc, but operates within
1082 void* mspace_malloc(mspace msp, size_t bytes);
1085 mspace_free behaves as free, but operates within
1088 If compiled with FOOTERS==1, mspace_free is not actually needed.
1089 free may be called instead of mspace_free because freed chunks from
1090 any space are handled by their originating spaces.
1092 void mspace_free(mspace msp, void* mem);
1095 mspace_realloc behaves as realloc, but operates within
1098 If compiled with FOOTERS==1, mspace_realloc is not actually
1099 needed. realloc may be called instead of mspace_realloc because
1100 realloced chunks from any space are handled by their originating
1103 void* mspace_realloc(mspace msp, void* mem, size_t newsize);
1106 mspace_calloc behaves as calloc, but operates within
1109 void* mspace_calloc(mspace msp, size_t n_elements, size_t elem_size);
1112 mspace_memalign behaves as memalign, but operates within
1115 void* mspace_memalign(mspace msp, size_t alignment, size_t bytes);
1118 mspace_independent_calloc behaves as independent_calloc, but
1119 operates within the given space.
1121 void** mspace_independent_calloc(mspace msp, size_t n_elements,
1122 size_t elem_size, void* chunks[]);
1125 mspace_independent_comalloc behaves as independent_comalloc, but
1126 operates within the given space.
1128 void** mspace_independent_comalloc(mspace msp, size_t n_elements,
1129 size_t sizes[], void* chunks[]);
1132 mspace_footprint() returns the number of bytes obtained from the
1133 system for this space.
1135 size_t mspace_footprint(mspace msp);
1138 mspace_max_footprint() returns the peak number of bytes obtained from the
1139 system for this space.
1141 size_t mspace_max_footprint(mspace msp);
1146 mspace_mallinfo behaves as mallinfo, but reports properties of
1149 struct mallinfo mspace_mallinfo(mspace msp);
1150 #endif /* NO_MALLINFO */
1153 mspace_malloc_stats behaves as malloc_stats, but reports
1154 properties of the given space.
1156 void mspace_malloc_stats(mspace msp);
1159 mspace_trim behaves as malloc_trim, but
1160 operates within the given space.
1162 int mspace_trim(mspace msp, size_t pad);
1165 An alias for mallopt.
1167 int mspace_mallopt(int, int);
1169 #endif /* MSPACES */
1172 }; /* end of extern "C" */
1173 #endif /* __cplusplus */
1176 ========================================================================
1177 To make a fully customizable malloc.h header file, cut everything
1178 above this line, put into file malloc.h, edit to suit, and #include it
1179 on the next line, as well as in programs that use this malloc.
1180 ========================================================================
1183 /* #include "malloc.h" */
1185 /*------------------------------ internal #includes ---------------------- */
1188 #pragma warning( disable : 4146 ) /* no "unsigned" warnings */
1189 #endif /* _MSC_VER */
1191 #ifndef LACKS_STDIO_H
1192 #include <stdio.h> /* for printing in malloc_stats */
1195 #ifndef LACKS_ERRNO_H
1196 #include <errno.h> /* for MALLOC_FAILURE_ACTION */
1197 #endif /* LACKS_ERRNO_H */
1199 #include <time.h> /* for magic initialization */
1200 #endif /* FOOTERS */
1201 #ifndef LACKS_STDLIB_H
1202 #include <stdlib.h> /* for abort() */
1203 #endif /* LACKS_STDLIB_H */
1205 #if ABORT_ON_ASSERT_FAILURE
1206 #define assert(x) if(!(x)) ABORT
1207 #else /* ABORT_ON_ASSERT_FAILURE */
1209 #endif /* ABORT_ON_ASSERT_FAILURE */
1213 #ifndef LACKS_STRING_H
1214 #include <string.h> /* for memset etc */
1215 #endif /* LACKS_STRING_H */
1217 #ifndef LACKS_STRINGS_H
1218 #include <strings.h> /* for ffs */
1219 #endif /* LACKS_STRINGS_H */
1220 #endif /* USE_BUILTIN_FFS */
1222 #ifndef LACKS_SYS_MMAN_H
1223 #include <sys/mman.h> /* for mmap */
1224 #endif /* LACKS_SYS_MMAN_H */
1225 #ifndef LACKS_FCNTL_H
1227 #endif /* LACKS_FCNTL_H */
1228 #endif /* HAVE_MMAP */
1230 #ifndef LACKS_UNISTD_H
1231 #include <unistd.h> /* for sbrk */
1232 #else /* LACKS_UNISTD_H */
1233 #if !defined(__FreeBSD__) && !defined(__OpenBSD__) && !defined(__NetBSD__)
1234 extern void* sbrk(ptrdiff_t);
1235 #endif /* FreeBSD etc */
1236 #endif /* LACKS_UNISTD_H */
1237 #endif /* HAVE_MMAP */
1240 #ifndef malloc_getpagesize
1241 # ifdef _SC_PAGESIZE /* some SVR4 systems omit an underscore */
1242 # ifndef _SC_PAGE_SIZE
1243 # define _SC_PAGE_SIZE _SC_PAGESIZE
1246 # ifdef _SC_PAGE_SIZE
1247 # define malloc_getpagesize sysconf(_SC_PAGE_SIZE)
1249 # if defined(BSD) || defined(DGUX) || defined(HAVE_GETPAGESIZE)
1250 extern size_t getpagesize();
1251 # define malloc_getpagesize getpagesize()
1253 # ifdef WIN32 /* use supplied emulation of getpagesize */
1254 # define malloc_getpagesize getpagesize()
1256 # ifndef LACKS_SYS_PARAM_H
1257 # include <sys/param.h>
1259 # ifdef EXEC_PAGESIZE
1260 # define malloc_getpagesize EXEC_PAGESIZE
1264 # define malloc_getpagesize NBPG
1266 # define malloc_getpagesize (NBPG * CLSIZE)
1270 # define malloc_getpagesize NBPC
1273 # define malloc_getpagesize PAGESIZE
1274 # else /* just guess */
1275 # define malloc_getpagesize ((size_t)4096U)
1286 /* ------------------- size_t and alignment properties -------------------- */
1288 /* The byte and bit size of a size_t */
1289 #define SIZE_T_SIZE (sizeof(size_t))
1290 #define SIZE_T_BITSIZE (sizeof(size_t) << 3)
1292 /* Some constants coerced to size_t */
1293 /* Annoying but necessary to avoid errors on some plaftorms */
1294 #define SIZE_T_ZERO ((size_t)0)
1295 #define SIZE_T_ONE ((size_t)1)
1296 #define SIZE_T_TWO ((size_t)2)
1297 #define TWO_SIZE_T_SIZES (SIZE_T_SIZE<<1)
1298 #define FOUR_SIZE_T_SIZES (SIZE_T_SIZE<<2)
1299 #define SIX_SIZE_T_SIZES (FOUR_SIZE_T_SIZES+TWO_SIZE_T_SIZES)
1300 #define HALF_MAX_SIZE_T (MAX_SIZE_T / 2U)
1302 /* The bit mask value corresponding to MALLOC_ALIGNMENT */
1303 #define CHUNK_ALIGN_MASK (MALLOC_ALIGNMENT - SIZE_T_ONE)
1305 /* True if address a has acceptable alignment */
1306 #define is_aligned(A) (((size_t)((A)) & (CHUNK_ALIGN_MASK)) == 0)
1308 /* the number of bytes to offset an address to align it */
1309 #define align_offset(A)\
1310 ((((size_t)(A) & CHUNK_ALIGN_MASK) == 0)? 0 :\
1311 ((MALLOC_ALIGNMENT - ((size_t)(A) & CHUNK_ALIGN_MASK)) & CHUNK_ALIGN_MASK))
1313 /* -------------------------- MMAP preliminaries ------------------------- */
1316 If HAVE_MORECORE or HAVE_MMAP are false, we just define calls and
1317 checks to fail so compiler optimizer can delete code rather than
1318 using so many "#if"s.
1322 /* MORECORE and MMAP must return MFAIL on failure */
1323 #define MFAIL ((void*)(MAX_SIZE_T))
1324 #define CMFAIL ((char*)(MFAIL)) /* defined for convenience */
1327 #define IS_MMAPPED_BIT (SIZE_T_ZERO)
1328 #define USE_MMAP_BIT (SIZE_T_ZERO)
1329 #define CALL_MMAP(s) MFAIL
1330 #define CALL_MUNMAP(a, s) (-1)
1331 #define DIRECT_MMAP(s) MFAIL
1333 #else /* HAVE_MMAP */
1334 #define IS_MMAPPED_BIT (SIZE_T_ONE)
1335 #define USE_MMAP_BIT (SIZE_T_ONE)
1338 #define CALL_MUNMAP(a, s) munmap((a), (s))
1339 #define MMAP_PROT (PROT_READ|PROT_WRITE)
1340 #if !defined(MAP_ANONYMOUS) && defined(MAP_ANON)
1341 #define MAP_ANONYMOUS MAP_ANON
1342 #endif /* MAP_ANON */
1343 #ifdef MAP_ANONYMOUS
1344 #define MMAP_FLAGS (MAP_PRIVATE|MAP_ANONYMOUS)
1345 #define CALL_MMAP(s) mmap(0, (s), MMAP_PROT, MMAP_FLAGS, -1, 0)
1346 #else /* MAP_ANONYMOUS */
1348 Nearly all versions of mmap support MAP_ANONYMOUS, so the following
1349 is unlikely to be needed, but is supplied just in case.
1351 #define MMAP_FLAGS (MAP_PRIVATE)
1352 static int dev_zero_fd = -1; /* Cached file descriptor for /dev/zero. */
1353 #define CALL_MMAP(s) ((dev_zero_fd < 0) ? \
1354 (dev_zero_fd = open("/dev/zero", O_RDWR), \
1355 mmap(0, (s), MMAP_PROT, MMAP_FLAGS, dev_zero_fd, 0)) : \
1356 mmap(0, (s), MMAP_PROT, MMAP_FLAGS, dev_zero_fd, 0))
1357 #endif /* MAP_ANONYMOUS */
1359 #define DIRECT_MMAP(s) CALL_MMAP(s)
1362 /* Win32 MMAP via VirtualAlloc */
1363 static void* win32mmap(size_t size) {
1364 void* ptr = VirtualAlloc(0, size, MEM_RESERVE|MEM_COMMIT, PAGE_READWRITE);
1365 return (ptr != 0)? ptr: MFAIL;
1368 /* For direct MMAP, use MEM_TOP_DOWN to minimize interference */
1369 static void* win32direct_mmap(size_t size) {
1370 void* ptr = VirtualAlloc(0, size, MEM_RESERVE|MEM_COMMIT|MEM_TOP_DOWN,
1372 return (ptr != 0)? ptr: MFAIL;
1375 /* This function supports releasing coalesed segments */
1376 static int win32munmap(void* ptr, size_t size) {
1377 MEMORY_BASIC_INFORMATION minfo;
1380 if (VirtualQuery(cptr, &minfo, sizeof(minfo)) == 0)
1382 if (minfo.BaseAddress != cptr || minfo.AllocationBase != cptr ||
1383 minfo.State != MEM_COMMIT || minfo.RegionSize > size)
1385 if (VirtualFree(cptr, 0, MEM_RELEASE) == 0)
1387 cptr += minfo.RegionSize;
1388 size -= minfo.RegionSize;
1393 #define CALL_MMAP(s) win32mmap(s)
1394 #define CALL_MUNMAP(a, s) win32munmap((a), (s))
1395 #define DIRECT_MMAP(s) win32direct_mmap(s)
1397 #endif /* HAVE_MMAP */
1399 #if HAVE_MMAP && HAVE_MREMAP
1400 #define CALL_MREMAP(addr, osz, nsz, mv) mremap((addr), (osz), (nsz), (mv))
1401 #else /* HAVE_MMAP && HAVE_MREMAP */
1402 #define CALL_MREMAP(addr, osz, nsz, mv) MFAIL
1403 #endif /* HAVE_MMAP && HAVE_MREMAP */
1406 #define CALL_MORECORE(S) MORECORE(S)
1407 #else /* HAVE_MORECORE */
1408 #define CALL_MORECORE(S) MFAIL
1409 #endif /* HAVE_MORECORE */
1411 /* mstate bit set if continguous morecore disabled or failed */
1412 #define USE_NONCONTIGUOUS_BIT (4U)
1414 /* segment bit set in create_mspace_with_base */
1415 #define EXTERN_BIT (8U)
1418 /* --------------------------- Lock preliminaries ------------------------ */
1423 When locks are defined, there are up to two global locks:
1425 * If HAVE_MORECORE, morecore_mutex protects sequences of calls to
1426 MORECORE. In many cases sys_alloc requires two calls, that should
1427 not be interleaved with calls by other threads. This does not
1428 protect against direct calls to MORECORE by other threads not
1429 using this lock, so there is still code to cope the best we can on
1432 * magic_init_mutex ensures that mparams.magic and other
1433 unique mparams values are initialized only once.
1437 /* By default use posix locks */
1438 #include <pthread.h>
1439 #define MLOCK_T pthread_mutex_t
1440 #define INITIAL_LOCK(l) pthread_mutex_init(l, NULL)
1441 #define ACQUIRE_LOCK(l) pthread_mutex_lock(l)
1442 #define RELEASE_LOCK(l) pthread_mutex_unlock(l)
1445 static MLOCK_T morecore_mutex = PTHREAD_MUTEX_INITIALIZER;
1446 #endif /* HAVE_MORECORE */
1448 static MLOCK_T magic_init_mutex = PTHREAD_MUTEX_INITIALIZER;
1452 Because lock-protected regions have bounded times, and there
1453 are no recursive lock calls, we can use simple spinlocks.
1456 #define MLOCK_T long
1457 static int win32_acquire_lock (MLOCK_T *sl) {
1459 #ifdef InterlockedCompareExchangePointer
1460 if (!InterlockedCompareExchange(sl, 1, 0))
1462 #else /* Use older void* version */
1463 if (!InterlockedCompareExchange((void**)sl, (void*)1, (void*)0))
1465 #endif /* InterlockedCompareExchangePointer */
1470 static void win32_release_lock (MLOCK_T *sl) {
1471 InterlockedExchange (sl, 0);
1474 #define INITIAL_LOCK(l) *(l)=0
1475 #define ACQUIRE_LOCK(l) win32_acquire_lock(l)
1476 #define RELEASE_LOCK(l) win32_release_lock(l)
1478 static MLOCK_T morecore_mutex;
1479 #endif /* HAVE_MORECORE */
1480 static MLOCK_T magic_init_mutex;
1483 #define USE_LOCK_BIT (2U)
1484 #else /* USE_LOCKS */
1485 #define USE_LOCK_BIT (0U)
1486 #define INITIAL_LOCK(l)
1487 #endif /* USE_LOCKS */
1489 #if USE_LOCKS && HAVE_MORECORE
1490 #define ACQUIRE_MORECORE_LOCK() ACQUIRE_LOCK(&morecore_mutex);
1491 #define RELEASE_MORECORE_LOCK() RELEASE_LOCK(&morecore_mutex);
1492 #else /* USE_LOCKS && HAVE_MORECORE */
1493 #define ACQUIRE_MORECORE_LOCK()
1494 #define RELEASE_MORECORE_LOCK()
1495 #endif /* USE_LOCKS && HAVE_MORECORE */
1498 #define ACQUIRE_MAGIC_INIT_LOCK() ACQUIRE_LOCK(&magic_init_mutex);
1499 #define RELEASE_MAGIC_INIT_LOCK() RELEASE_LOCK(&magic_init_mutex);
1500 #else /* USE_LOCKS */
1501 #define ACQUIRE_MAGIC_INIT_LOCK()
1502 #define RELEASE_MAGIC_INIT_LOCK()
1503 #endif /* USE_LOCKS */
1506 /* ----------------------- Chunk representations ------------------------ */
1509 (The following includes lightly edited explanations by Colin Plumb.)
1511 The malloc_chunk declaration below is misleading (but accurate and
1512 necessary). It declares a "view" into memory allowing access to
1513 necessary fields at known offsets from a given base.
1515 Chunks of memory are maintained using a `boundary tag' method as
1516 originally described by Knuth. (See the paper by Paul Wilson
1517 ftp://ftp.cs.utexas.edu/pub/garbage/allocsrv.ps for a survey of such
1518 techniques.) Sizes of free chunks are stored both in the front of
1519 each chunk and at the end. This makes consolidating fragmented
1520 chunks into bigger chunks fast. The head fields also hold bits
1521 representing whether chunks are free or in use.
1523 Here are some pictures to make it clearer. They are "exploded" to
1524 show that the state of a chunk can be thought of as extending from
1525 the high 31 bits of the head field of its header through the
1526 prev_foot and PINUSE_BIT bit of the following chunk header.
1528 A chunk that's in use looks like:
1530 chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1531 | Size of previous chunk (if P = 1) |
1532 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1533 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |P|
1534 | Size of this chunk 1| +-+
1535 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1541 +- size - sizeof(size_t) available payload bytes -+
1545 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1546 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |1|
1547 | Size of next chunk (may or may not be in use) | +-+
1548 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1550 And if it's free, it looks like this:
1553 | User payload (must be in use, or we would have merged!) |
1554 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1555 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |P|
1556 | Size of this chunk 0| +-+
1557 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1559 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1561 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1563 +- size - sizeof(struct chunk) unused bytes -+
1565 chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1566 | Size of this chunk |
1567 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1568 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |0|
1569 | Size of next chunk (must be in use, or we would have merged)| +-+
1570 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1574 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1577 Note that since we always merge adjacent free chunks, the chunks
1578 adjacent to a free chunk must be in use.
1580 Given a pointer to a chunk (which can be derived trivially from the
1581 payload pointer) we can, in O(1) time, find out whether the adjacent
1582 chunks are free, and if so, unlink them from the lists that they
1583 are on and merge them with the current chunk.
1585 Chunks always begin on even word boundaries, so the mem portion
1586 (which is returned to the user) is also on an even word boundary, and
1587 thus at least double-word aligned.
1589 The P (PINUSE_BIT) bit, stored in the unused low-order bit of the
1590 chunk size (which is always a multiple of two words), is an in-use
1591 bit for the *previous* chunk. If that bit is *clear*, then the
1592 word before the current chunk size contains the previous chunk
1593 size, and can be used to find the front of the previous chunk.
1594 The very first chunk allocated always has this bit set, preventing
1595 access to non-existent (or non-owned) memory. If pinuse is set for
1596 any given chunk, then you CANNOT determine the size of the
1597 previous chunk, and might even get a memory addressing fault when
1600 The C (CINUSE_BIT) bit, stored in the unused second-lowest bit of
1601 the chunk size redundantly records whether the current chunk is
1602 inuse. This redundancy enables usage checks within free and realloc,
1603 and reduces indirection when freeing and consolidating chunks.
1605 Each freshly allocated chunk must have both cinuse and pinuse set.
1606 That is, each allocated chunk borders either a previously allocated
1607 and still in-use chunk, or the base of its memory arena. This is
1608 ensured by making all allocations from the the `lowest' part of any
1609 found chunk. Further, no free chunk physically borders another one,
1610 so each free chunk is known to be preceded and followed by either
1611 inuse chunks or the ends of memory.
1613 Note that the `foot' of the current chunk is actually represented
1614 as the prev_foot of the NEXT chunk. This makes it easier to
1615 deal with alignments etc but can be very confusing when trying
1616 to extend or adapt this code.
1618 The exceptions to all this are
1620 1. The special chunk `top' is the top-most available chunk (i.e.,
1621 the one bordering the end of available memory). It is treated
1622 specially. Top is never included in any bin, is used only if
1623 no other chunk is available, and is released back to the
1624 system if it is very large (see M_TRIM_THRESHOLD). In effect,
1625 the top chunk is treated as larger (and thus less well
1626 fitting) than any other available chunk. The top chunk
1627 doesn't update its trailing size field since there is no next
1628 contiguous chunk that would have to index off it. However,
1629 space is still allocated for it (TOP_FOOT_SIZE) to enable
1630 separation or merging when space is extended.
1632 3. Chunks allocated via mmap, which have the lowest-order bit
1633 (IS_MMAPPED_BIT) set in their prev_foot fields, and do not set
1634 PINUSE_BIT in their head fields. Because they are allocated
1635 one-by-one, each must carry its own prev_foot field, which is
1636 also used to hold the offset this chunk has within its mmapped
1637 region, which is needed to preserve alignment. Each mmapped
1638 chunk is trailed by the first two fields of a fake next-chunk
1639 for sake of usage checks.
1643 struct malloc_chunk {
1644 size_t prev_foot; /* Size of previous chunk (if free). */
1645 size_t head; /* Size and inuse bits. */
1646 struct malloc_chunk* fd; /* double links -- used only if free. */
1647 struct malloc_chunk* bk;
1650 typedef struct malloc_chunk mchunk;
1651 typedef struct malloc_chunk* mchunkptr;
1652 typedef struct malloc_chunk* sbinptr; /* The type of bins of chunks */
1653 typedef size_t bindex_t; /* Described below */
1654 typedef unsigned int binmap_t; /* Described below */
1655 typedef unsigned int flag_t; /* The type of various bit flag sets */
1657 /* ------------------- Chunks sizes and alignments ----------------------- */
1659 #define MCHUNK_SIZE (sizeof(mchunk))
1662 #define CHUNK_OVERHEAD (TWO_SIZE_T_SIZES)
1664 #define CHUNK_OVERHEAD (SIZE_T_SIZE)
1665 #endif /* FOOTERS */
1667 /* MMapped chunks need a second word of overhead ... */
1668 #define MMAP_CHUNK_OVERHEAD (TWO_SIZE_T_SIZES)
1669 /* ... and additional padding for fake next-chunk at foot */
1670 #define MMAP_FOOT_PAD (FOUR_SIZE_T_SIZES)
1672 /* The smallest size we can malloc is an aligned minimal chunk */
1673 #define MIN_CHUNK_SIZE\
1674 ((MCHUNK_SIZE + CHUNK_ALIGN_MASK) & ~CHUNK_ALIGN_MASK)
1676 /* conversion from malloc headers to user pointers, and back */
1677 #define chunk2mem(p) ((void*)((char*)(p) + TWO_SIZE_T_SIZES))
1678 #define mem2chunk(mem) ((mchunkptr)((char*)(mem) - TWO_SIZE_T_SIZES))
1679 /* chunk associated with aligned address A */
1680 #define align_as_chunk(A) (mchunkptr)((A) + align_offset(chunk2mem(A)))
1682 /* Bounds on request (not chunk) sizes. */
1683 #define MAX_REQUEST ((-MIN_CHUNK_SIZE) << 2)
1684 #define MIN_REQUEST (MIN_CHUNK_SIZE - CHUNK_OVERHEAD - SIZE_T_ONE)
1686 /* pad request bytes into a usable size */
1687 #define pad_request(req) \
1688 (((req) + CHUNK_OVERHEAD + CHUNK_ALIGN_MASK) & ~CHUNK_ALIGN_MASK)
1690 /* pad request, checking for minimum (but not maximum) */
1691 #define request2size(req) \
1692 (((req) < MIN_REQUEST)? MIN_CHUNK_SIZE : pad_request(req))
1695 /* ------------------ Operations on head and foot fields ----------------- */
1698 The head field of a chunk is or'ed with PINUSE_BIT when previous
1699 adjacent chunk in use, and or'ed with CINUSE_BIT if this chunk is in
1700 use. If the chunk was obtained with mmap, the prev_foot field has
1701 IS_MMAPPED_BIT set, otherwise holding the offset of the base of the
1702 mmapped region to the base of the chunk.
1705 #define PINUSE_BIT (SIZE_T_ONE)
1706 #define CINUSE_BIT (SIZE_T_TWO)
1707 #define INUSE_BITS (PINUSE_BIT|CINUSE_BIT)
1709 /* Head value for fenceposts */
1710 #define FENCEPOST_HEAD (INUSE_BITS|SIZE_T_SIZE)
1712 /* extraction of fields from head words */
1713 #define cinuse(p) ((p)->head & CINUSE_BIT)
1714 #define pinuse(p) ((p)->head & PINUSE_BIT)
1715 #define chunksize(p) ((p)->head & ~(INUSE_BITS))
1717 #define clear_pinuse(p) ((p)->head &= ~PINUSE_BIT)
1718 #define clear_cinuse(p) ((p)->head &= ~CINUSE_BIT)
1720 /* Treat space at ptr +/- offset as a chunk */
1721 #define chunk_plus_offset(p, s) ((mchunkptr)(((char*)(p)) + (s)))
1722 #define chunk_minus_offset(p, s) ((mchunkptr)(((char*)(p)) - (s)))
1724 /* Ptr to next or previous physical malloc_chunk. */
1725 #define next_chunk(p) ((mchunkptr)( ((char*)(p)) + ((p)->head & ~INUSE_BITS)))
1726 #define prev_chunk(p) ((mchunkptr)( ((char*)(p)) - ((p)->prev_foot) ))
1728 /* extract next chunk's pinuse bit */
1729 #define next_pinuse(p) ((next_chunk(p)->head) & PINUSE_BIT)
1731 /* Get/set size at footer */
1732 #define get_foot(p, s) (((mchunkptr)((char*)(p) + (s)))->prev_foot)
1733 #define set_foot(p, s) (((mchunkptr)((char*)(p) + (s)))->prev_foot = (s))
1735 /* Set size, pinuse bit, and foot */
1736 #define set_size_and_pinuse_of_free_chunk(p, s)\
1737 ((p)->head = (s|PINUSE_BIT), set_foot(p, s))
1739 /* Set size, pinuse bit, foot, and clear next pinuse */
1740 #define set_free_with_pinuse(p, s, n)\
1741 (clear_pinuse(n), set_size_and_pinuse_of_free_chunk(p, s))
1743 #define is_mmapped(p)\
1744 (!((p)->head & PINUSE_BIT) && ((p)->prev_foot & IS_MMAPPED_BIT))
1746 /* Get the internal overhead associated with chunk p */
1747 #define overhead_for(p)\
1748 (is_mmapped(p)? MMAP_CHUNK_OVERHEAD : CHUNK_OVERHEAD)
1750 /* Return true if malloced space is not necessarily cleared */
1752 #define calloc_must_clear(p) (!is_mmapped(p))
1753 #else /* MMAP_CLEARS */
1754 #define calloc_must_clear(p) (1)
1755 #endif /* MMAP_CLEARS */
1757 /* ---------------------- Overlaid data structures ----------------------- */
1760 When chunks are not in use, they are treated as nodes of either
1763 "Small" chunks are stored in circular doubly-linked lists, and look
1766 chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1767 | Size of previous chunk |
1768 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1769 `head:' | Size of chunk, in bytes |P|
1770 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1771 | Forward pointer to next chunk in list |
1772 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1773 | Back pointer to previous chunk in list |
1774 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1775 | Unused space (may be 0 bytes long) .
1778 nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1779 `foot:' | Size of chunk, in bytes |
1780 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1782 Larger chunks are kept in a form of bitwise digital trees (aka
1783 tries) keyed on chunksizes. Because malloc_tree_chunks are only for
1784 free chunks greater than 256 bytes, their size doesn't impose any
1785 constraints on user chunk sizes. Each node looks like:
1787 chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1788 | Size of previous chunk |
1789 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1790 `head:' | Size of chunk, in bytes |P|
1791 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1792 | Forward pointer to next chunk of same size |
1793 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1794 | Back pointer to previous chunk of same size |
1795 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1796 | Pointer to left child (child[0]) |
1797 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1798 | Pointer to right child (child[1]) |
1799 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1800 | Pointer to parent |
1801 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1802 | bin index of this chunk |
1803 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1806 nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1807 `foot:' | Size of chunk, in bytes |
1808 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1810 Each tree holding treenodes is a tree of unique chunk sizes. Chunks
1811 of the same size are arranged in a circularly-linked list, with only
1812 the oldest chunk (the next to be used, in our FIFO ordering)
1813 actually in the tree. (Tree members are distinguished by a non-null
1814 parent pointer.) If a chunk with the same size an an existing node
1815 is inserted, it is linked off the existing node using pointers that
1816 work in the same way as fd/bk pointers of small chunks.
1818 Each tree contains a power of 2 sized range of chunk sizes (the
1819 smallest is 0x100 <= x < 0x180), which is is divided in half at each
1820 tree level, with the chunks in the smaller half of the range (0x100
1821 <= x < 0x140 for the top nose) in the left subtree and the larger
1822 half (0x140 <= x < 0x180) in the right subtree. This is, of course,
1823 done by inspecting individual bits.
1825 Using these rules, each node's left subtree contains all smaller
1826 sizes than its right subtree. However, the node at the root of each
1827 subtree has no particular ordering relationship to either. (The
1828 dividing line between the subtree sizes is based on trie relation.)
1829 If we remove the last chunk of a given size from the interior of the
1830 tree, we need to replace it with a leaf node. The tree ordering
1831 rules permit a node to be replaced by any leaf below it.
1833 The smallest chunk in a tree (a common operation in a best-fit
1834 allocator) can be found by walking a path to the leftmost leaf in
1835 the tree. Unlike a usual binary tree, where we follow left child
1836 pointers until we reach a null, here we follow the right child
1837 pointer any time the left one is null, until we reach a leaf with
1838 both child pointers null. The smallest chunk in the tree will be
1839 somewhere along that path.
1841 The worst case number of steps to add, find, or remove a node is
1842 bounded by the number of bits differentiating chunks within
1843 bins. Under current bin calculations, this ranges from 6 up to 21
1844 (for 32 bit sizes) or up to 53 (for 64 bit sizes). The typical case
1845 is of course much better.
1848 struct malloc_tree_chunk {
1849 /* The first four fields must be compatible with malloc_chunk */
1852 struct malloc_tree_chunk* fd;
1853 struct malloc_tree_chunk* bk;
1855 struct malloc_tree_chunk* child[2];
1856 struct malloc_tree_chunk* parent;
1860 typedef struct malloc_tree_chunk tchunk;
1861 typedef struct malloc_tree_chunk* tchunkptr;
1862 typedef struct malloc_tree_chunk* tbinptr; /* The type of bins of trees */
1864 /* A little helper macro for trees */
1865 #define leftmost_child(t) ((t)->child[0] != 0? (t)->child[0] : (t)->child[1])
1867 /* ----------------------------- Segments -------------------------------- */
1870 Each malloc space may include non-contiguous segments, held in a
1871 list headed by an embedded malloc_segment record representing the
1872 top-most space. Segments also include flags holding properties of
1873 the space. Large chunks that are directly allocated by mmap are not
1874 included in this list. They are instead independently created and
1875 destroyed without otherwise keeping track of them.
1877 Segment management mainly comes into play for spaces allocated by
1878 MMAP. Any call to MMAP might or might not return memory that is
1879 adjacent to an existing segment. MORECORE normally contiguously
1880 extends the current space, so this space is almost always adjacent,
1881 which is simpler and faster to deal with. (This is why MORECORE is
1882 used preferentially to MMAP when both are available -- see
1883 sys_alloc.) When allocating using MMAP, we don't use any of the
1884 hinting mechanisms (inconsistently) supported in various
1885 implementations of unix mmap, or distinguish reserving from
1886 committing memory. Instead, we just ask for space, and exploit
1887 contiguity when we get it. It is probably possible to do
1888 better than this on some systems, but no general scheme seems
1889 to be significantly better.
1891 Management entails a simpler variant of the consolidation scheme
1892 used for chunks to reduce fragmentation -- new adjacent memory is
1893 normally prepended or appended to an existing segment. However,
1894 there are limitations compared to chunk consolidation that mostly
1895 reflect the fact that segment processing is relatively infrequent
1896 (occurring only when getting memory from system) and that we
1897 don't expect to have huge numbers of segments:
1899 * Segments are not indexed, so traversal requires linear scans. (It
1900 would be possible to index these, but is not worth the extra
1901 overhead and complexity for most programs on most platforms.)
1902 * New segments are only appended to old ones when holding top-most
1903 memory; if they cannot be prepended to others, they are held in
1906 Except for the top-most segment of an mstate, each segment record
1907 is kept at the tail of its segment. Segments are added by pushing
1908 segment records onto the list headed by &mstate.seg for the
1911 Segment flags control allocation/merge/deallocation policies:
1912 * If EXTERN_BIT set, then we did not allocate this segment,
1913 and so should not try to deallocate or merge with others.
1914 (This currently holds only for the initial segment passed
1915 into create_mspace_with_base.)
1916 * If IS_MMAPPED_BIT set, the segment may be merged with
1917 other surrounding mmapped segments and trimmed/de-allocated
1919 * If neither bit is set, then the segment was obtained using
1920 MORECORE so can be merged with surrounding MORECORE'd segments
1921 and deallocated/trimmed using MORECORE with negative arguments.
1924 struct malloc_segment {
1925 char* base; /* base address */
1926 size_t size; /* allocated size */
1927 struct malloc_segment* next; /* ptr to next segment */
1928 flag_t sflags; /* mmap and extern flag */
1931 #define is_mmapped_segment(S) ((S)->sflags & IS_MMAPPED_BIT)
1932 #define is_extern_segment(S) ((S)->sflags & EXTERN_BIT)
1934 typedef struct malloc_segment msegment;
1935 typedef struct malloc_segment* msegmentptr;
1937 /* ---------------------------- malloc_state ----------------------------- */
1940 A malloc_state holds all of the bookkeeping for a space.
1941 The main fields are:
1944 The topmost chunk of the currently active segment. Its size is
1945 cached in topsize. The actual size of topmost space is
1946 topsize+TOP_FOOT_SIZE, which includes space reserved for adding
1947 fenceposts and segment records if necessary when getting more
1948 space from the system. The size at which to autotrim top is
1949 cached from mparams in trim_check, except that it is disabled if
1952 Designated victim (dv)
1953 This is the preferred chunk for servicing small requests that
1954 don't have exact fits. It is normally the chunk split off most
1955 recently to service another small request. Its size is cached in
1956 dvsize. The link fields of this chunk are not maintained since it
1957 is not kept in a bin.
1960 An array of bin headers for free chunks. These bins hold chunks
1961 with sizes less than MIN_LARGE_SIZE bytes. Each bin contains
1962 chunks of all the same size, spaced 8 bytes apart. To simplify
1963 use in double-linked lists, each bin header acts as a malloc_chunk
1964 pointing to the real first node, if it exists (else pointing to
1965 itself). This avoids special-casing for headers. But to avoid
1966 waste, we allocate only the fd/bk pointers of bins, and then use
1967 repositioning tricks to treat these as the fields of a chunk.
1970 Treebins are pointers to the roots of trees holding a range of
1971 sizes. There are 2 equally spaced treebins for each power of two
1972 from TREE_SHIFT to TREE_SHIFT+16. The last bin holds anything
1976 There is one bit map for small bins ("smallmap") and one for
1977 treebins ("treemap). Each bin sets its bit when non-empty, and
1978 clears the bit when empty. Bit operations are then used to avoid
1979 bin-by-bin searching -- nearly all "search" is done without ever
1980 looking at bins that won't be selected. The bit maps
1981 conservatively use 32 bits per map word, even if on 64bit system.
1982 For a good description of some of the bit-based techniques used
1983 here, see Henry S. Warren Jr's book "Hacker's Delight" (and
1984 supplement at http://hackersdelight.org/). Many of these are
1985 intended to reduce the branchiness of paths through malloc etc, as
1986 well as to reduce the number of memory locations read or written.
1989 A list of segments headed by an embedded malloc_segment record
1990 representing the initial space.
1992 Address check support
1993 The least_addr field is the least address ever obtained from
1994 MORECORE or MMAP. Attempted frees and reallocs of any address less
1995 than this are trapped (unless INSECURE is defined).
1998 A cross-check field that should always hold same value as mparams.magic.
2001 Bits recording whether to use MMAP, locks, or contiguous MORECORE
2004 Each space keeps track of current and maximum system memory
2005 obtained via MORECORE or MMAP.
2008 If USE_LOCKS is defined, the "mutex" lock is acquired and released
2009 around every public call using this mspace.
2012 /* Bin types, widths and sizes */
2013 #define NSMALLBINS (32U)
2014 #define NTREEBINS (32U)
2015 #define SMALLBIN_SHIFT (3U)
2016 #define SMALLBIN_WIDTH (SIZE_T_ONE << SMALLBIN_SHIFT)
2017 #define TREEBIN_SHIFT (8U)
2018 #define MIN_LARGE_SIZE (SIZE_T_ONE << TREEBIN_SHIFT)
2019 #define MAX_SMALL_SIZE (MIN_LARGE_SIZE - SIZE_T_ONE)
2020 #define MAX_SMALL_REQUEST (MAX_SMALL_SIZE - CHUNK_ALIGN_MASK - CHUNK_OVERHEAD)
2022 struct malloc_state {
2032 mchunkptr smallbins[(NSMALLBINS+1)*2];
2033 tbinptr treebins[NTREEBINS];
2035 size_t max_footprint;
2038 MLOCK_T mutex; /* locate lock among fields that rarely change */
2039 #endif /* USE_LOCKS */
2043 typedef struct malloc_state* mstate;
2045 /* ------------- Global malloc_state and malloc_params ------------------- */
2048 malloc_params holds global properties, including those that can be
2049 dynamically set using mallopt. There is a single instance, mparams,
2050 initialized in init_mparams.
2053 struct malloc_params {
2057 size_t mmap_threshold;
2058 size_t trim_threshold;
2059 flag_t default_mflags;
2062 static struct malloc_params mparams;
2064 /* The global malloc_state used for all non-"mspace" calls */
2065 static struct malloc_state _gm_;
2067 #define is_global(M) ((M) == &_gm_)
2068 #define is_initialized(M) ((M)->top != 0)
2070 /* -------------------------- system alloc setup ------------------------- */
2072 /* Operations on mflags */
2074 #define use_lock(M) ((M)->mflags & USE_LOCK_BIT)
2075 #define enable_lock(M) ((M)->mflags |= USE_LOCK_BIT)
2076 #define disable_lock(M) ((M)->mflags &= ~USE_LOCK_BIT)
2078 #define use_mmap(M) ((M)->mflags & USE_MMAP_BIT)
2079 #define enable_mmap(M) ((M)->mflags |= USE_MMAP_BIT)
2080 #define disable_mmap(M) ((M)->mflags &= ~USE_MMAP_BIT)
2082 #define use_noncontiguous(M) ((M)->mflags & USE_NONCONTIGUOUS_BIT)
2083 #define disable_contiguous(M) ((M)->mflags |= USE_NONCONTIGUOUS_BIT)
2085 #define set_lock(M,L)\
2086 ((M)->mflags = (L)?\
2087 ((M)->mflags | USE_LOCK_BIT) :\
2088 ((M)->mflags & ~USE_LOCK_BIT))
2090 /* page-align a size */
2091 #define page_align(S)\
2092 (((S) + (mparams.page_size)) & ~(mparams.page_size - SIZE_T_ONE))
2094 /* granularity-align a size */
2095 #define granularity_align(S)\
2096 (((S) + (mparams.granularity)) & ~(mparams.granularity - SIZE_T_ONE))
2098 #define is_page_aligned(S)\
2099 (((size_t)(S) & (mparams.page_size - SIZE_T_ONE)) == 0)
2100 #define is_granularity_aligned(S)\
2101 (((size_t)(S) & (mparams.granularity - SIZE_T_ONE)) == 0)
2103 /* True if segment S holds address A */
2104 #define segment_holds(S, A)\
2105 ((char*)(A) >= S->base && (char*)(A) < S->base + S->size)
2107 /* Return segment holding given address */
2108 static msegmentptr segment_holding(mstate m, char* addr) {
2109 msegmentptr sp = &m->seg;
2111 if (addr >= sp->base && addr < sp->base + sp->size)
2113 if ((sp = sp->next) == 0)
2118 /* Return true if segment contains a segment link */
2119 static int has_segment_link(mstate m, msegmentptr ss) {
2120 msegmentptr sp = &m->seg;
2122 if ((char*)sp >= ss->base && (char*)sp < ss->base + ss->size)
2124 if ((sp = sp->next) == 0)
2129 #ifndef MORECORE_CANNOT_TRIM
2130 #define should_trim(M,s) ((s) > (M)->trim_check)
2131 #else /* MORECORE_CANNOT_TRIM */
2132 #define should_trim(M,s) (0)
2133 #endif /* MORECORE_CANNOT_TRIM */
2136 TOP_FOOT_SIZE is padding at the end of a segment, including space
2137 that may be needed to place segment records and fenceposts when new
2138 noncontiguous segments are added.
2140 #define TOP_FOOT_SIZE\
2141 (align_offset(chunk2mem(0))+pad_request(sizeof(struct malloc_segment))+MIN_CHUNK_SIZE)
2144 /* ------------------------------- Hooks -------------------------------- */
2147 PREACTION should be defined to return 0 on success, and nonzero on
2148 failure. If you are not using locking, you can redefine these to do
2154 /* Ensure locks are initialized */
2155 #define GLOBALLY_INITIALIZE() (mparams.page_size == 0 && init_mparams())
2157 #define PREACTION(M) ((GLOBALLY_INITIALIZE() || use_lock(M))? ACQUIRE_LOCK(&(M)->mutex) : 0)
2158 #define POSTACTION(M) { if (use_lock(M)) RELEASE_LOCK(&(M)->mutex); }
2159 #else /* USE_LOCKS */
2162 #define PREACTION(M) (0)
2163 #endif /* PREACTION */
2166 #define POSTACTION(M)
2167 #endif /* POSTACTION */
2169 #endif /* USE_LOCKS */
2172 CORRUPTION_ERROR_ACTION is triggered upon detected bad addresses.
2173 USAGE_ERROR_ACTION is triggered on detected bad frees and
2174 reallocs. The argument p is an address that might have triggered the
2175 fault. It is ignored by the two predefined actions, but might be
2176 useful in custom actions that try to help diagnose errors.
2179 #if PROCEED_ON_ERROR
2181 /* A count of the number of corruption errors causing resets */
2182 int malloc_corruption_error_count;
2184 /* default corruption action */
2185 static void reset_on_error(mstate m);
2187 #define CORRUPTION_ERROR_ACTION(m) reset_on_error(m)
2188 #define USAGE_ERROR_ACTION(m, p)
2190 #else /* PROCEED_ON_ERROR */
2192 #ifndef CORRUPTION_ERROR_ACTION
2193 #define CORRUPTION_ERROR_ACTION(m) ABORT
2194 #endif /* CORRUPTION_ERROR_ACTION */
2196 #ifndef USAGE_ERROR_ACTION
2197 #define USAGE_ERROR_ACTION(m,p) ABORT
2198 #endif /* USAGE_ERROR_ACTION */
2200 #endif /* PROCEED_ON_ERROR */
2202 /* -------------------------- Debugging setup ---------------------------- */
2206 #define check_free_chunk(M,P)
2207 #define check_inuse_chunk(M,P)
2208 #define check_malloced_chunk(M,P,N)
2209 #define check_mmapped_chunk(M,P)
2210 #define check_malloc_state(M)
2211 #define check_top_chunk(M,P)
2214 #define check_free_chunk(M,P) do_check_free_chunk(M,P)
2215 #define check_inuse_chunk(M,P) do_check_inuse_chunk(M,P)
2216 #define check_top_chunk(M,P) do_check_top_chunk(M,P)
2217 #define check_malloced_chunk(M,P,N) do_check_malloced_chunk(M,P,N)
2218 #define check_mmapped_chunk(M,P) do_check_mmapped_chunk(M,P)
2219 #define check_malloc_state(M) do_check_malloc_state(M)
2221 static void do_check_any_chunk(mstate m, mchunkptr p);
2222 static void do_check_top_chunk(mstate m, mchunkptr p);
2223 static void do_check_mmapped_chunk(mstate m, mchunkptr p);
2224 static void do_check_inuse_chunk(mstate m, mchunkptr p);
2225 static void do_check_free_chunk(mstate m, mchunkptr p);
2226 static void do_check_malloced_chunk(mstate m, void* mem, size_t s);
2227 static void do_check_tree(mstate m, tchunkptr t);
2228 static void do_check_treebin(mstate m, bindex_t i);
2229 static void do_check_smallbin(mstate m, bindex_t i);
2230 static void do_check_malloc_state(mstate m);
2231 static int bin_find(mstate m, mchunkptr x);
2232 static size_t traverse_and_check(mstate m);
2235 /* ---------------------------- Indexing Bins ---------------------------- */
2237 #define is_small(s) (((s) >> SMALLBIN_SHIFT) < NSMALLBINS)
2238 #define small_index(s) ((s) >> SMALLBIN_SHIFT)
2239 #define small_index2size(i) ((i) << SMALLBIN_SHIFT)
2240 #define MIN_SMALL_INDEX (small_index(MIN_CHUNK_SIZE))
2242 /* addressing by index. See above about smallbin repositioning */
2243 #define smallbin_at(M, i) ((sbinptr)((char*)&((M)->smallbins[(i)<<1])))
2244 #define treebin_at(M,i) (&((M)->treebins[i]))
2246 /* assign tree index for size S to variable I */
2247 #if defined(__GNUC__) && defined(i386)
2248 #define compute_tree_index(S, I)\
2250 size_t X = S >> TREEBIN_SHIFT;\
2253 else if (X > 0xFFFF)\
2257 __asm__("bsrl %1,%0\n\t" : "=r" (K) : "rm" (X));\
2258 I = (bindex_t)((K << 1) + ((S >> (K + (TREEBIN_SHIFT-1)) & 1)));\
2262 #define compute_tree_index(S, I)\
2264 size_t X = S >> TREEBIN_SHIFT;\
2267 else if (X > 0xFFFF)\
2270 unsigned int Y = (unsigned int)X;\
2271 unsigned int N = ((Y - 0x100) >> 16) & 8;\
2272 unsigned int K = (((Y <<= N) - 0x1000) >> 16) & 4;\
2274 N += K = (((Y <<= K) - 0x4000) >> 16) & 2;\
2275 K = 14 - N + ((Y <<= K) >> 15);\
2276 I = (K << 1) + ((S >> (K + (TREEBIN_SHIFT-1)) & 1));\
2281 /* Bit representing maximum resolved size in a treebin at i */
2282 #define bit_for_tree_index(i) \
2283 (i == NTREEBINS-1)? (SIZE_T_BITSIZE-1) : (((i) >> 1) + TREEBIN_SHIFT - 2)
2285 /* Shift placing maximum resolved bit in a treebin at i as sign bit */
2286 #define leftshift_for_tree_index(i) \
2287 ((i == NTREEBINS-1)? 0 : \
2288 ((SIZE_T_BITSIZE-SIZE_T_ONE) - (((i) >> 1) + TREEBIN_SHIFT - 2)))
2290 /* The size of the smallest chunk held in bin with index i */
2291 #define minsize_for_tree_index(i) \
2292 ((SIZE_T_ONE << (((i) >> 1) + TREEBIN_SHIFT)) | \
2293 (((size_t)((i) & SIZE_T_ONE)) << (((i) >> 1) + TREEBIN_SHIFT - 1)))
2296 /* ------------------------ Operations on bin maps ----------------------- */
2298 /* bit corresponding to given index */
2299 #define idx2bit(i) ((binmap_t)(1) << (i))
2301 /* Mark/Clear bits with given index */
2302 #define mark_smallmap(M,i) ((M)->smallmap |= idx2bit(i))
2303 #define clear_smallmap(M,i) ((M)->smallmap &= ~idx2bit(i))
2304 #define smallmap_is_marked(M,i) ((M)->smallmap & idx2bit(i))
2306 #define mark_treemap(M,i) ((M)->treemap |= idx2bit(i))
2307 #define clear_treemap(M,i) ((M)->treemap &= ~idx2bit(i))
2308 #define treemap_is_marked(M,i) ((M)->treemap & idx2bit(i))
2310 /* index corresponding to given bit */
2312 #if defined(__GNUC__) && defined(i386)
2313 #define compute_bit2idx(X, I)\
2316 __asm__("bsfl %1,%0\n\t" : "=r" (J) : "rm" (X));\
2322 #define compute_bit2idx(X, I) I = ffs(X)-1
2324 #else /* USE_BUILTIN_FFS */
2325 #define compute_bit2idx(X, I)\
2327 unsigned int Y = X - 1;\
2328 unsigned int K = Y >> (16-4) & 16;\
2329 unsigned int N = K; Y >>= K;\
2330 N += K = Y >> (8-3) & 8; Y >>= K;\
2331 N += K = Y >> (4-2) & 4; Y >>= K;\
2332 N += K = Y >> (2-1) & 2; Y >>= K;\
2333 N += K = Y >> (1-0) & 1; Y >>= K;\
2334 I = (bindex_t)(N + Y);\
2336 #endif /* USE_BUILTIN_FFS */
2339 /* isolate the least set bit of a bitmap */
2340 #define least_bit(x) ((x) & -(x))
2342 /* mask with all bits to left of least bit of x on */
2343 #define left_bits(x) ((x<<1) | -(x<<1))
2345 /* mask with all bits to left of or equal to least bit of x on */
2346 #define same_or_left_bits(x) ((x) | -(x))
2349 /* ----------------------- Runtime Check Support ------------------------- */
2352 For security, the main invariant is that malloc/free/etc never
2353 writes to a static address other than malloc_state, unless static
2354 malloc_state itself has been corrupted, which cannot occur via
2355 malloc (because of these checks). In essence this means that we
2356 believe all pointers, sizes, maps etc held in malloc_state, but
2357 check all of those linked or offsetted from other embedded data
2358 structures. These checks are interspersed with main code in a way
2359 that tends to minimize their run-time cost.
2361 When FOOTERS is defined, in addition to range checking, we also
2362 verify footer fields of inuse chunks, which can be used guarantee
2363 that the mstate controlling malloc/free is intact. This is a
2364 streamlined version of the approach described by William Robertson
2365 et al in "Run-time Detection of Heap-based Overflows" LISA'03
2366 http://www.usenix.org/events/lisa03/tech/robertson.html The footer
2367 of an inuse chunk holds the xor of its mstate and a random seed,
2368 that is checked upon calls to free() and realloc(). This is
2369 (probablistically) unguessable from outside the program, but can be
2370 computed by any code successfully malloc'ing any chunk, so does not
2371 itself provide protection against code that has already broken
2372 security through some other means. Unlike Robertson et al, we
2373 always dynamically check addresses of all offset chunks (previous,
2374 next, etc). This turns out to be cheaper than relying on hashes.
2378 /* Check if address a is at least as high as any from MORECORE or MMAP */
2379 #define ok_address(M, a) ((char*)(a) >= (M)->least_addr)
2380 /* Check if address of next chunk n is higher than base chunk p */
2381 #define ok_next(p, n) ((char*)(p) < (char*)(n))
2382 /* Check if p has its cinuse bit on */
2383 #define ok_cinuse(p) cinuse(p)
2384 /* Check if p has its pinuse bit on */
2385 #define ok_pinuse(p) pinuse(p)
2387 #else /* !INSECURE */
2388 #define ok_address(M, a) (1)
2389 #define ok_next(b, n) (1)
2390 #define ok_cinuse(p) (1)
2391 #define ok_pinuse(p) (1)
2392 #endif /* !INSECURE */
2394 #if (FOOTERS && !INSECURE)
2395 /* Check if (alleged) mstate m has expected magic field */
2396 #define ok_magic(M) ((M)->magic == mparams.magic)
2397 #else /* (FOOTERS && !INSECURE) */
2398 #define ok_magic(M) (1)
2399 #endif /* (FOOTERS && !INSECURE) */
2402 /* In gcc, use __builtin_expect to minimize impact of checks */
2404 #if defined(__GNUC__) && __GNUC__ >= 3
2405 #define RTCHECK(e) __builtin_expect(e, 1)
2407 #define RTCHECK(e) (e)
2409 #else /* !INSECURE */
2410 #define RTCHECK(e) (1)
2411 #endif /* !INSECURE */
2413 /* macros to set up inuse chunks with or without footers */
2417 #define mark_inuse_foot(M,p,s)
2419 /* Set cinuse bit and pinuse bit of next chunk */
2420 #define set_inuse(M,p,s)\
2421 ((p)->head = (((p)->head & PINUSE_BIT)|s|CINUSE_BIT),\
2422 ((mchunkptr)(((char*)(p)) + (s)))->head |= PINUSE_BIT)
2424 /* Set cinuse and pinuse of this chunk and pinuse of next chunk */
2425 #define set_inuse_and_pinuse(M,p,s)\
2426 ((p)->head = (s|PINUSE_BIT|CINUSE_BIT),\
2427 ((mchunkptr)(((char*)(p)) + (s)))->head |= PINUSE_BIT)
2429 /* Set size, cinuse and pinuse bit of this chunk */
2430 #define set_size_and_pinuse_of_inuse_chunk(M, p, s)\
2431 ((p)->head = (s|PINUSE_BIT|CINUSE_BIT))
2435 /* Set foot of inuse chunk to be xor of mstate and seed */
2436 #define mark_inuse_foot(M,p,s)\
2437 (((mchunkptr)((char*)(p) + (s)))->prev_foot = ((size_t)(M) ^ mparams.magic))
2439 #define get_mstate_for(p)\
2440 ((mstate)(((mchunkptr)((char*)(p) +\
2441 (chunksize(p))))->prev_foot ^ mparams.magic))
2443 #define set_inuse(M,p,s)\
2444 ((p)->head = (((p)->head & PINUSE_BIT)|s|CINUSE_BIT),\
2445 (((mchunkptr)(((char*)(p)) + (s)))->head |= PINUSE_BIT), \
2446 mark_inuse_foot(M,p,s))
2448 #define set_inuse_and_pinuse(M,p,s)\
2449 ((p)->head = (s|PINUSE_BIT|CINUSE_BIT),\
2450 (((mchunkptr)(((char*)(p)) + (s)))->head |= PINUSE_BIT),\
2451 mark_inuse_foot(M,p,s))
2453 #define set_size_and_pinuse_of_inuse_chunk(M, p, s)\
2454 ((p)->head = (s|PINUSE_BIT|CINUSE_BIT),\
2455 mark_inuse_foot(M, p, s))
2457 #endif /* !FOOTERS */
2459 /* ---------------------------- setting mparams -------------------------- */
2461 /* Initialize mparams */
2462 static int init_mparams(void) {
2463 if (mparams.page_size == 0) {
2466 mparams.mmap_threshold = DEFAULT_MMAP_THRESHOLD;
2467 mparams.trim_threshold = DEFAULT_TRIM_THRESHOLD;
2468 #if MORECORE_CONTIGUOUS
2469 mparams.default_mflags = USE_LOCK_BIT|USE_MMAP_BIT;
2470 #else /* MORECORE_CONTIGUOUS */
2471 mparams.default_mflags = USE_LOCK_BIT|USE_MMAP_BIT|USE_NONCONTIGUOUS_BIT;
2472 #endif /* MORECORE_CONTIGUOUS */
2474 #if (FOOTERS && !INSECURE)
2478 unsigned char buf[sizeof(size_t)];
2479 /* Try to use /dev/urandom, else fall back on using time */
2480 if ((fd = open("/dev/urandom", O_RDONLY)) >= 0 &&
2481 read(fd, buf, sizeof(buf)) == sizeof(buf)) {
2482 s = *((size_t *) buf);
2486 #endif /* USE_DEV_RANDOM */
2487 s = (size_t)(time(0) ^ (size_t)0x55555555U);
2489 s |= (size_t)8U; /* ensure nonzero */
2490 s &= ~(size_t)7U; /* improve chances of fault for bad values */
2493 #else /* (FOOTERS && !INSECURE) */
2494 s = (size_t)0x58585858U;
2495 #endif /* (FOOTERS && !INSECURE) */
2496 ACQUIRE_MAGIC_INIT_LOCK();
2497 if (mparams.magic == 0) {
2499 /* Set up lock for main malloc area */
2500 INITIAL_LOCK(&gm->mutex);
2501 gm->mflags = mparams.default_mflags;
2503 RELEASE_MAGIC_INIT_LOCK();
2506 mparams.page_size = malloc_getpagesize;
2507 mparams.granularity = ((DEFAULT_GRANULARITY != 0)?
2508 DEFAULT_GRANULARITY : mparams.page_size);
2511 SYSTEM_INFO system_info;
2512 GetSystemInfo(&system_info);
2513 mparams.page_size = system_info.dwPageSize;
2514 mparams.granularity = system_info.dwAllocationGranularity;
2518 /* Sanity-check configuration:
2519 size_t must be unsigned and as wide as pointer type.
2520 ints must be at least 4 bytes.
2521 alignment must be at least 8.
2522 Alignment, min chunk size, and page size must all be powers of 2.
2524 if ((sizeof(size_t) != sizeof(char*)) ||
2525 (MAX_SIZE_T < MIN_CHUNK_SIZE) ||
2526 (sizeof(int) < 4) ||
2527 (MALLOC_ALIGNMENT < (size_t)8U) ||
2528 ((MALLOC_ALIGNMENT & (MALLOC_ALIGNMENT-SIZE_T_ONE)) != 0) ||
2529 ((MCHUNK_SIZE & (MCHUNK_SIZE-SIZE_T_ONE)) != 0) ||
2530 ((mparams.granularity & (mparams.granularity-SIZE_T_ONE)) != 0) ||
2531 ((mparams.page_size & (mparams.page_size-SIZE_T_ONE)) != 0))
2537 /* support for mallopt */
2538 static int change_mparam(int param_number, int value) {
2539 size_t val = (size_t)value;
2541 switch(param_number) {
2542 case M_TRIM_THRESHOLD:
2543 mparams.trim_threshold = val;
2546 if (val >= mparams.page_size && ((val & (val-1)) == 0)) {
2547 mparams.granularity = val;
2552 case M_MMAP_THRESHOLD:
2553 mparams.mmap_threshold = val;
2561 /* ------------------------- Debugging Support --------------------------- */
2563 /* Check properties of any chunk, whether free, inuse, mmapped etc */
2564 static void do_check_any_chunk(mstate m, mchunkptr p) {
2565 assert((is_aligned(chunk2mem(p))) || (p->head == FENCEPOST_HEAD));
2566 assert(ok_address(m, p));
2569 /* Check properties of top chunk */
2570 static void do_check_top_chunk(mstate m, mchunkptr p) {
2571 msegmentptr sp = segment_holding(m, (char*)p);
2572 size_t sz = chunksize(p);
2574 assert((is_aligned(chunk2mem(p))) || (p->head == FENCEPOST_HEAD));
2575 assert(ok_address(m, p));
2576 assert(sz == m->topsize);
2578 assert(sz == ((sp->base + sp->size) - (char*)p) - TOP_FOOT_SIZE);
2580 assert(!next_pinuse(p));
2583 /* Check properties of (inuse) mmapped chunks */
2584 static void do_check_mmapped_chunk(mstate m, mchunkptr p) {
2585 size_t sz = chunksize(p);
2586 size_t len = (sz + (p->prev_foot & ~IS_MMAPPED_BIT) + MMAP_FOOT_PAD);
2587 assert(is_mmapped(p));
2588 assert(use_mmap(m));
2589 assert((is_aligned(chunk2mem(p))) || (p->head == FENCEPOST_HEAD));
2590 assert(ok_address(m, p));
2591 assert(!is_small(sz));
2592 assert((len & (mparams.page_size-SIZE_T_ONE)) == 0);
2593 assert(chunk_plus_offset(p, sz)->head == FENCEPOST_HEAD);
2594 assert(chunk_plus_offset(p, sz+SIZE_T_SIZE)->head == 0);
2597 /* Check properties of inuse chunks */
2598 static void do_check_inuse_chunk(mstate m, mchunkptr p) {
2599 do_check_any_chunk(m, p);
2601 assert(next_pinuse(p));
2602 /* If not pinuse and not mmapped, previous chunk has OK offset */
2603 assert(is_mmapped(p) || pinuse(p) || next_chunk(prev_chunk(p)) == p);
2605 do_check_mmapped_chunk(m, p);
2608 /* Check properties of free chunks */
2609 static void do_check_free_chunk(mstate m, mchunkptr p) {
2610 size_t sz = p->head & ~(PINUSE_BIT|CINUSE_BIT);
2611 mchunkptr next = chunk_plus_offset(p, sz);
2612 do_check_any_chunk(m, p);
2614 assert(!next_pinuse(p));
2615 assert (!is_mmapped(p));
2616 if (p != m->dv && p != m->top) {
2617 if (sz >= MIN_CHUNK_SIZE) {
2618 assert((sz & CHUNK_ALIGN_MASK) == 0);
2619 assert(is_aligned(chunk2mem(p)));
2620 assert(next->prev_foot == sz);
2622 assert (next == m->top || cinuse(next));
2623 assert(p->fd->bk == p);
2624 assert(p->bk->fd == p);
2626 else /* markers are always of size SIZE_T_SIZE */
2627 assert(sz == SIZE_T_SIZE);
2631 /* Check properties of malloced chunks at the point they are malloced */
2632 static void do_check_malloced_chunk(mstate m, void* mem, size_t s) {
2634 mchunkptr p = mem2chunk(mem);
2635 size_t sz = p->head & ~(PINUSE_BIT|CINUSE_BIT);
2636 do_check_inuse_chunk(m, p);
2637 assert((sz & CHUNK_ALIGN_MASK) == 0);
2638 assert(sz >= MIN_CHUNK_SIZE);
2640 /* unless mmapped, size is less than MIN_CHUNK_SIZE more than request */
2641 assert(is_mmapped(p) || sz < (s + MIN_CHUNK_SIZE));
2645 /* Check a tree and its subtrees. */
2646 static void do_check_tree(mstate m, tchunkptr t) {
2649 bindex_t tindex = t->index;
2650 size_t tsize = chunksize(t);
2652 compute_tree_index(tsize, idx);
2653 assert(tindex == idx);
2654 assert(tsize >= MIN_LARGE_SIZE);
2655 assert(tsize >= minsize_for_tree_index(idx));
2656 assert((idx == NTREEBINS-1) || (tsize < minsize_for_tree_index((idx+1))));
2658 do { /* traverse through chain of same-sized nodes */
2659 do_check_any_chunk(m, ((mchunkptr)u));
2660 assert(u->index == tindex);
2661 assert(chunksize(u) == tsize);
2663 assert(!next_pinuse(u));
2664 assert(u->fd->bk == u);
2665 assert(u->bk->fd == u);
2666 if (u->parent == 0) {
2667 assert(u->child[0] == 0);
2668 assert(u->child[1] == 0);
2671 assert(head == 0); /* only one node on chain has parent */
2673 assert(u->parent != u);
2674 assert (u->parent->child[0] == u ||
2675 u->parent->child[1] == u ||
2676 *((tbinptr*)(u->parent)) == u);
2677 if (u->child[0] != 0) {
2678 assert(u->child[0]->parent == u);
2679 assert(u->child[0] != u);
2680 do_check_tree(m, u->child[0]);
2682 if (u->child[1] != 0) {
2683 assert(u->child[1]->parent == u);
2684 assert(u->child[1] != u);
2685 do_check_tree(m, u->child[1]);
2687 if (u->child[0] != 0 && u->child[1] != 0) {
2688 assert(chunksize(u->child[0]) < chunksize(u->child[1]));
2696 /* Check all the chunks in a treebin. */
2697 static void do_check_treebin(mstate m, bindex_t i) {
2698 tbinptr* tb = treebin_at(m, i);
2700 int empty = (m->treemap & (1U << i)) == 0;
2704 do_check_tree(m, t);
2707 /* Check all the chunks in a smallbin. */
2708 static void do_check_smallbin(mstate m, bindex_t i) {
2709 sbinptr b = smallbin_at(m, i);
2710 mchunkptr p = b->bk;
2711 unsigned int empty = (m->smallmap & (1U << i)) == 0;
2715 for (; p != b; p = p->bk) {
2716 size_t size = chunksize(p);
2718 /* each chunk claims to be free */
2719 do_check_free_chunk(m, p);
2720 /* chunk belongs in bin */
2721 assert(small_index(size) == i);
2722 assert(p->bk == b || chunksize(p->bk) == chunksize(p));
2723 /* chunk is followed by an inuse chunk */
2725 if (q->head != FENCEPOST_HEAD)
2726 do_check_inuse_chunk(m, q);
2731 /* Find x in a bin. Used in other check functions. */
2732 static int bin_find(mstate m, mchunkptr x) {
2733 size_t size = chunksize(x);
2734 if (is_small(size)) {
2735 bindex_t sidx = small_index(size);
2736 sbinptr b = smallbin_at(m, sidx);
2737 if (smallmap_is_marked(m, sidx)) {
2742 } while ((p = p->fd) != b);
2747 compute_tree_index(size, tidx);
2748 if (treemap_is_marked(m, tidx)) {
2749 tchunkptr t = *treebin_at(m, tidx);
2750 size_t sizebits = size << leftshift_for_tree_index(tidx);
2751 while (t != 0 && chunksize(t) != size) {
2752 t = t->child[(sizebits >> (SIZE_T_BITSIZE-SIZE_T_ONE)) & 1];
2758 if (u == (tchunkptr)x)
2760 } while ((u = u->fd) != t);
2767 /* Traverse each chunk and check it; return total */
2768 static size_t traverse_and_check(mstate m) {
2770 if (is_initialized(m)) {
2771 msegmentptr s = &m->seg;
2772 sum += m->topsize + TOP_FOOT_SIZE;
2774 mchunkptr q = align_as_chunk(s->base);
2775 mchunkptr lastq = 0;
2777 while (segment_holds(s, q) &&
2778 q != m->top && q->head != FENCEPOST_HEAD) {
2779 sum += chunksize(q);
2781 assert(!bin_find(m, q));
2782 do_check_inuse_chunk(m, q);
2785 assert(q == m->dv || bin_find(m, q));
2786 assert(lastq == 0 || cinuse(lastq)); /* Not 2 consecutive free */
2787 do_check_free_chunk(m, q);
2798 /* Check all properties of malloc_state. */
2799 static void do_check_malloc_state(mstate m) {
2803 for (i = 0; i < NSMALLBINS; ++i)
2804 do_check_smallbin(m, i);
2805 for (i = 0; i < NTREEBINS; ++i)
2806 do_check_treebin(m, i);
2808 if (m->dvsize != 0) { /* check dv chunk */
2809 do_check_any_chunk(m, m->dv);
2810 assert(m->dvsize == chunksize(m->dv));
2811 assert(m->dvsize >= MIN_CHUNK_SIZE);
2812 assert(bin_find(m, m->dv) == 0);
2815 if (m->top != 0) { /* check top chunk */
2816 do_check_top_chunk(m, m->top);
2817 assert(m->topsize == chunksize(m->top));
2818 assert(m->topsize > 0);
2819 assert(bin_find(m, m->top) == 0);
2822 total = traverse_and_check(m);
2823 assert(total <= m->footprint);
2824 assert(m->footprint <= m->max_footprint);
2828 /* ----------------------------- statistics ------------------------------ */
2831 static struct mallinfo internal_mallinfo(mstate m) {
2832 struct mallinfo nm = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
2833 if (!PREACTION(m)) {
2834 check_malloc_state(m);
2835 if (is_initialized(m)) {
2836 size_t nfree = SIZE_T_ONE; /* top always free */
2837 size_t mfree = m->topsize + TOP_FOOT_SIZE;
2839 msegmentptr s = &m->seg;
2841 mchunkptr q = align_as_chunk(s->base);
2842 while (segment_holds(s, q) &&
2843 q != m->top && q->head != FENCEPOST_HEAD) {
2844 size_t sz = chunksize(q);
2857 nm.hblkhd = m->footprint - sum;
2858 nm.usmblks = m->max_footprint;
2859 nm.uordblks = m->footprint - mfree;
2860 nm.fordblks = mfree;
2861 nm.keepcost = m->topsize;
2868 #endif /* !NO_MALLINFO */
2870 static void internal_malloc_stats(mstate m) {
2871 if (!PREACTION(m)) {
2875 check_malloc_state(m);
2876 if (is_initialized(m)) {
2877 msegmentptr s = &m->seg;
2878 maxfp = m->max_footprint;
2880 used = fp - (m->topsize + TOP_FOOT_SIZE);
2883 mchunkptr q = align_as_chunk(s->base);
2884 while (segment_holds(s, q) &&
2885 q != m->top && q->head != FENCEPOST_HEAD) {
2887 used -= chunksize(q);
2894 #ifndef LACKS_STDIO_H
2895 fprintf(stderr, "max system bytes = %10lu\n", (unsigned long)(maxfp));
2896 fprintf(stderr, "system bytes = %10lu\n", (unsigned long)(fp));
2897 fprintf(stderr, "in use bytes = %10lu\n", (unsigned long)(used));
2904 /* ----------------------- Operations on smallbins ----------------------- */
2907 Various forms of linking and unlinking are defined as macros. Even
2908 the ones for trees, which are very long but have very short typical
2909 paths. This is ugly but reduces reliance on inlining support of
2913 /* Link a free chunk into a smallbin */
2914 #define insert_small_chunk(M, P, S) {\
2915 bindex_t I = small_index(S);\
2916 mchunkptr B = smallbin_at(M, I);\
2918 assert(S >= MIN_CHUNK_SIZE);\
2919 if (!smallmap_is_marked(M, I))\
2920 mark_smallmap(M, I);\
2921 else if (RTCHECK(ok_address(M, B->fd)))\
2924 CORRUPTION_ERROR_ACTION(M);\
2932 /* Unlink a chunk from a smallbin */
2933 #define unlink_small_chunk(M, P, S) {\
2934 mchunkptr F = P->fd;\
2935 mchunkptr B = P->bk;\
2936 bindex_t I = small_index(S);\
2939 assert(chunksize(P) == small_index2size(I));\
2941 clear_smallmap(M, I);\
2942 else if (RTCHECK((F == smallbin_at(M,I) || ok_address(M, F)) &&\
2943 (B == smallbin_at(M,I) || ok_address(M, B)))) {\
2948 CORRUPTION_ERROR_ACTION(M);\
2952 /* Unlink the first chunk from a smallbin */
2953 #define unlink_first_small_chunk(M, B, P, I) {\
2954 mchunkptr F = P->fd;\
2957 assert(chunksize(P) == small_index2size(I));\
2959 clear_smallmap(M, I);\
2960 else if (RTCHECK(ok_address(M, F))) {\
2965 CORRUPTION_ERROR_ACTION(M);\
2969 /* Replace dv node, binning the old one */
2970 /* Used only when dvsize known to be small */
2971 #define replace_dv(M, P, S) {\
2972 size_t DVS = M->dvsize;\
2974 mchunkptr DV = M->dv;\
2975 assert(is_small(DVS));\
2976 insert_small_chunk(M, DV, DVS);\
2982 /* ------------------------- Operations on trees ------------------------- */
2984 /* Insert chunk into tree */
2985 #define insert_large_chunk(M, X, S) {\
2988 compute_tree_index(S, I);\
2989 H = treebin_at(M, I);\
2991 X->child[0] = X->child[1] = 0;\
2992 if (!treemap_is_marked(M, I)) {\
2993 mark_treemap(M, I);\
2995 X->parent = (tchunkptr)H;\
3000 size_t K = S << leftshift_for_tree_index(I);\
3002 if (chunksize(T) != S) {\
3003 tchunkptr* C = &(T->child[(K >> (SIZE_T_BITSIZE-SIZE_T_ONE)) & 1]);\
3007 else if (RTCHECK(ok_address(M, C))) {\
3014 CORRUPTION_ERROR_ACTION(M);\
3019 tchunkptr F = T->fd;\
3020 if (RTCHECK(ok_address(M, T) && ok_address(M, F))) {\
3028 CORRUPTION_ERROR_ACTION(M);\
3039 1. If x is a chained node, unlink it from its same-sized fd/bk links
3040 and choose its bk node as its replacement.
3041 2. If x was the last node of its size, but not a leaf node, it must
3042 be replaced with a leaf node (not merely one with an open left or
3043 right), to make sure that lefts and rights of descendents
3044 correspond properly to bit masks. We use the rightmost descendent
3045 of x. We could use any other leaf, but this is easy to locate and
3046 tends to counteract removal of leftmosts elsewhere, and so keeps
3047 paths shorter than minimally guaranteed. This doesn't loop much
3048 because on average a node in a tree is near the bottom.
3049 3. If x is the base of a chain (i.e., has parent links) relink
3050 x's parent and children to x's replacement (or null if none).
3053 #define unlink_large_chunk(M, X) {\
3054 tchunkptr XP = X->parent;\
3057 tchunkptr F = X->fd;\
3059 if (RTCHECK(ok_address(M, F))) {\
3064 CORRUPTION_ERROR_ACTION(M);\
3069 if (((R = *(RP = &(X->child[1]))) != 0) ||\
3070 ((R = *(RP = &(X->child[0]))) != 0)) {\
3072 while ((*(CP = &(R->child[1])) != 0) ||\
3073 (*(CP = &(R->child[0])) != 0)) {\
3076 if (RTCHECK(ok_address(M, RP)))\
3079 CORRUPTION_ERROR_ACTION(M);\
3084 tbinptr* H = treebin_at(M, X->index);\
3086 if ((*H = R) == 0) \
3087 clear_treemap(M, X->index);\
3089 else if (RTCHECK(ok_address(M, XP))) {\
3090 if (XP->child[0] == X) \
3096 CORRUPTION_ERROR_ACTION(M);\
3098 if (RTCHECK(ok_address(M, R))) {\
3101 if ((C0 = X->child[0]) != 0) {\
3102 if (RTCHECK(ok_address(M, C0))) {\
3107 CORRUPTION_ERROR_ACTION(M);\
3109 if ((C1 = X->child[1]) != 0) {\
3110 if (RTCHECK(ok_address(M, C1))) {\
3115 CORRUPTION_ERROR_ACTION(M);\
3119 CORRUPTION_ERROR_ACTION(M);\
3124 /* Relays to large vs small bin operations */
3126 #define insert_chunk(M, P, S)\
3127 if (is_small(S)) insert_small_chunk(M, P, S)\
3128 else { tchunkptr TP = (tchunkptr)(P); insert_large_chunk(M, TP, S); }
3130 #define unlink_chunk(M, P, S)\
3131 if (is_small(S)) unlink_small_chunk(M, P, S)\
3132 else { tchunkptr TP = (tchunkptr)(P); unlink_large_chunk(M, TP); }
3135 /* Relays to internal calls to malloc/free from realloc, memalign etc */
3138 #define internal_malloc(m, b) mspace_malloc(m, b)
3139 #define internal_free(m, mem) mspace_free(m,mem);
3140 #else /* ONLY_MSPACES */
3142 #define internal_malloc(m, b)\
3143 (m == gm)? dlmalloc(b) : mspace_malloc(m, b)
3144 #define internal_free(m, mem)\
3145 if (m == gm) dlfree(mem); else mspace_free(m,mem);
3147 #define internal_malloc(m, b) dlmalloc(b)
3148 #define internal_free(m, mem) dlfree(mem)
3149 #endif /* MSPACES */
3150 #endif /* ONLY_MSPACES */
3152 /* ----------------------- Direct-mmapping chunks ----------------------- */
3155 Directly mmapped chunks are set up with an offset to the start of
3156 the mmapped region stored in the prev_foot field of the chunk. This
3157 allows reconstruction of the required argument to MUNMAP when freed,
3158 and also allows adjustment of the returned chunk to meet alignment
3159 requirements (especially in memalign). There is also enough space
3160 allocated to hold a fake next chunk of size SIZE_T_SIZE to maintain
3161 the PINUSE bit so frees can be checked.
3164 /* Malloc using mmap */
3165 static void* mmap_alloc(mstate m, size_t nb) {
3166 size_t mmsize = granularity_align(nb + SIX_SIZE_T_SIZES + CHUNK_ALIGN_MASK);
3167 if (mmsize > nb) { /* Check for wrap around 0 */
3168 char* mm = (char*)(DIRECT_MMAP(mmsize));
3170 size_t offset = align_offset(chunk2mem(mm));
3171 size_t psize = mmsize - offset - MMAP_FOOT_PAD;
3172 mchunkptr p = (mchunkptr)(mm + offset);
3173 p->prev_foot = offset | IS_MMAPPED_BIT;
3174 (p)->head = (psize|CINUSE_BIT);
3175 mark_inuse_foot(m, p, psize);
3176 chunk_plus_offset(p, psize)->head = FENCEPOST_HEAD;
3177 chunk_plus_offset(p, psize+SIZE_T_SIZE)->head = 0;
3179 if (mm < m->least_addr)
3181 if ((m->footprint += mmsize) > m->max_footprint)
3182 m->max_footprint = m->footprint;
3183 assert(is_aligned(chunk2mem(p)));
3184 check_mmapped_chunk(m, p);
3185 return chunk2mem(p);
3191 /* Realloc using mmap */
3192 static mchunkptr mmap_resize(mstate m, mchunkptr oldp, size_t nb) {
3193 size_t oldsize = chunksize(oldp);
3194 if (is_small(nb)) /* Can't shrink mmap regions below small size */
3196 /* Keep old chunk if big enough but not too big */
3197 if (oldsize >= nb + SIZE_T_SIZE &&
3198 (oldsize - nb) <= (mparams.granularity << 1))
3201 size_t offset = oldp->prev_foot & ~IS_MMAPPED_BIT;
3202 size_t oldmmsize = oldsize + offset + MMAP_FOOT_PAD;
3203 size_t newmmsize = granularity_align(nb + SIX_SIZE_T_SIZES +
3205 char* cp = (char*)CALL_MREMAP((char*)oldp - offset,
3206 oldmmsize, newmmsize, 1);
3208 mchunkptr newp = (mchunkptr)(cp + offset);
3209 size_t psize = newmmsize - offset - MMAP_FOOT_PAD;
3210 newp->head = (psize|CINUSE_BIT);
3211 mark_inuse_foot(m, newp, psize);
3212 chunk_plus_offset(newp, psize)->head = FENCEPOST_HEAD;
3213 chunk_plus_offset(newp, psize+SIZE_T_SIZE)->head = 0;
3215 if (cp < m->least_addr)
3217 if ((m->footprint += newmmsize - oldmmsize) > m->max_footprint)
3218 m->max_footprint = m->footprint;
3219 check_mmapped_chunk(m, newp);
3226 /* -------------------------- mspace management -------------------------- */
3228 /* Initialize top chunk and its size */
3229 static void init_top(mstate m, mchunkptr p, size_t psize) {
3230 /* Ensure alignment */
3231 size_t offset = align_offset(chunk2mem(p));
3232 p = (mchunkptr)((char*)p + offset);
3237 p->head = psize | PINUSE_BIT;
3238 /* set size of fake trailing chunk holding overhead space only once */
3239 chunk_plus_offset(p, psize)->head = TOP_FOOT_SIZE;
3240 m->trim_check = mparams.trim_threshold; /* reset on each update */
3243 /* Initialize bins for a new mstate that is otherwise zeroed out */
3244 static void init_bins(mstate m) {
3245 /* Establish circular links for smallbins */
3247 for (i = 0; i < NSMALLBINS; ++i) {
3248 sbinptr bin = smallbin_at(m,i);
3249 bin->fd = bin->bk = bin;
3253 #if PROCEED_ON_ERROR
3255 /* default corruption action */
3256 static void reset_on_error(mstate m) {
3258 ++malloc_corruption_error_count;
3259 /* Reinitialize fields to forget about all memory */
3260 m->smallbins = m->treebins = 0;
3261 m->dvsize = m->topsize = 0;
3266 for (i = 0; i < NTREEBINS; ++i)
3267 *treebin_at(m, i) = 0;
3270 #endif /* PROCEED_ON_ERROR */
3272 /* Allocate chunk and prepend remainder with chunk in successor base. */
3273 static void* prepend_alloc(mstate m, char* newbase, char* oldbase,
3275 mchunkptr p = align_as_chunk(newbase);
3276 mchunkptr oldfirst = align_as_chunk(oldbase);
3277 size_t psize = (char*)oldfirst - (char*)p;
3278 mchunkptr q = chunk_plus_offset(p, nb);
3279 size_t qsize = psize - nb;
3280 set_size_and_pinuse_of_inuse_chunk(m, p, nb);
3282 assert((char*)oldfirst > (char*)q);
3283 assert(pinuse(oldfirst));
3284 assert(qsize >= MIN_CHUNK_SIZE);
3286 /* consolidate remainder with first chunk of old base */
3287 if (oldfirst == m->top) {
3288 size_t tsize = m->topsize += qsize;
3290 q->head = tsize | PINUSE_BIT;
3291 check_top_chunk(m, q);
3293 else if (oldfirst == m->dv) {
3294 size_t dsize = m->dvsize += qsize;
3296 set_size_and_pinuse_of_free_chunk(q, dsize);
3299 if (!cinuse(oldfirst)) {
3300 size_t nsize = chunksize(oldfirst);
3301 unlink_chunk(m, oldfirst, nsize);
3302 oldfirst = chunk_plus_offset(oldfirst, nsize);
3305 set_free_with_pinuse(q, qsize, oldfirst);
3306 insert_chunk(m, q, qsize);
3307 check_free_chunk(m, q);
3310 check_malloced_chunk(m, chunk2mem(p), nb);
3311 return chunk2mem(p);
3315 /* Add a segment to hold a new noncontiguous region */
3316 static void add_segment(mstate m, char* tbase, size_t tsize, flag_t mmapped) {
3317 /* Determine locations and sizes of segment, fenceposts, old top */
3318 char* old_top = (char*)m->top;
3319 msegmentptr oldsp = segment_holding(m, old_top);
3320 char* old_end = oldsp->base + oldsp->size;
3321 size_t ssize = pad_request(sizeof(struct malloc_segment));
3322 char* rawsp = old_end - (ssize + FOUR_SIZE_T_SIZES + CHUNK_ALIGN_MASK);
3323 size_t offset = align_offset(chunk2mem(rawsp));
3324 char* asp = rawsp + offset;
3325 char* csp = (asp < (old_top + MIN_CHUNK_SIZE))? old_top : asp;
3326 mchunkptr sp = (mchunkptr)csp;
3327 msegmentptr ss = (msegmentptr)(chunk2mem(sp));
3328 mchunkptr tnext = chunk_plus_offset(sp, ssize);
3329 mchunkptr p = tnext;
3332 /* reset top to new space */
3333 init_top(m, (mchunkptr)tbase, tsize - TOP_FOOT_SIZE);
3335 /* Set up segment record */
3336 assert(is_aligned(ss));
3337 set_size_and_pinuse_of_inuse_chunk(m, sp, ssize);
3338 *ss = m->seg; /* Push current record */
3339 m->seg.base = tbase;
3340 m->seg.size = tsize;
3341 m->seg.sflags = mmapped;
3344 /* Insert trailing fenceposts */
3346 mchunkptr nextp = chunk_plus_offset(p, SIZE_T_SIZE);
3347 p->head = FENCEPOST_HEAD;
3349 if ((char*)(&(nextp->head)) < old_end)
3354 assert(nfences >= 2);
3356 /* Insert the rest of old top into a bin as an ordinary free chunk */
3357 if (csp != old_top) {
3358 mchunkptr q = (mchunkptr)old_top;
3359 size_t psize = csp - old_top;
3360 mchunkptr tn = chunk_plus_offset(q, psize);
3361 set_free_with_pinuse(q, psize, tn);
3362 insert_chunk(m, q, psize);
3365 check_top_chunk(m, m->top);
3368 /* -------------------------- System allocation -------------------------- */
3370 /* Get memory from system using MORECORE or MMAP */
3371 static void* sys_alloc(mstate m, size_t nb) {
3372 char* tbase = CMFAIL;
3374 flag_t mmap_flag = 0;
3378 /* Directly map large chunks */
3379 if (use_mmap(m) && nb >= mparams.mmap_threshold) {
3380 void* mem = mmap_alloc(m, nb);
3386 Try getting memory in any of three ways (in most-preferred to
3387 least-preferred order):
3388 1. A call to MORECORE that can normally contiguously extend memory.
3389 (disabled if not MORECORE_CONTIGUOUS or not HAVE_MORECORE or
3390 or main space is mmapped or a previous contiguous call failed)
3391 2. A call to MMAP new space (disabled if not HAVE_MMAP).
3392 Note that under the default settings, if MORECORE is unable to
3393 fulfill a request, and HAVE_MMAP is true, then mmap is
3394 used as a noncontiguous system allocator. This is a useful backup
3395 strategy for systems with holes in address spaces -- in this case
3396 sbrk cannot contiguously expand the heap, but mmap may be able to
3398 3. A call to MORECORE that cannot usually contiguously extend memory.
3399 (disabled if not HAVE_MORECORE)
3402 if (MORECORE_CONTIGUOUS && !use_noncontiguous(m)) {
3404 msegmentptr ss = (m->top == 0)? 0 : segment_holding(m, (char*)m->top);
3406 ACQUIRE_MORECORE_LOCK();
3408 if (ss == 0) { /* First time through or recovery */
3409 char* base = (char*)CALL_MORECORE(0);
3410 if (base != CMFAIL) {
3411 asize = granularity_align(nb + TOP_FOOT_SIZE + MALLOC_ALIGNMENT + SIZE_T_ONE);
3412 /* Adjust to end on a page boundary */
3413 if (!is_page_aligned(base))
3414 asize += (page_align((size_t)base) - (size_t)base);
3415 /* Can't call MORECORE if size is negative when treated as signed */
3416 if (asize < HALF_MAX_SIZE_T &&
3417 (br = (char*)(CALL_MORECORE(asize))) == base) {
3424 /* Subtract out existing available top space from MORECORE request. */
3425 asize = granularity_align(nb - m->topsize + TOP_FOOT_SIZE + MALLOC_ALIGNMENT + SIZE_T_ONE);
3426 /* Use mem here only if it did continuously extend old space */
3427 if (asize < HALF_MAX_SIZE_T &&
3428 (br = (char*)(CALL_MORECORE(asize))) == ss->base+ss->size) {
3434 if (tbase == CMFAIL) { /* Cope with partial failure */
3435 if (br != CMFAIL) { /* Try to use/extend the space we did get */
3436 if (asize < HALF_MAX_SIZE_T &&
3437 asize < nb + TOP_FOOT_SIZE + SIZE_T_ONE) {
3438 size_t esize = granularity_align(nb + TOP_FOOT_SIZE + MALLOC_ALIGNMENT + SIZE_T_ONE - asize);
3439 if (esize < HALF_MAX_SIZE_T) {
3440 char* end = (char*)CALL_MORECORE(esize);
3443 else { /* Can't use; try to release */
3444 end = (char*)CALL_MORECORE(-asize);
3450 if (br != CMFAIL) { /* Use the space we did get */
3455 disable_contiguous(m); /* Don't try contiguous path in the future */
3458 RELEASE_MORECORE_LOCK();
3461 if (HAVE_MMAP && tbase == CMFAIL) { /* Try MMAP */
3462 size_t req = nb + TOP_FOOT_SIZE + MALLOC_ALIGNMENT + SIZE_T_ONE;
3463 size_t rsize = granularity_align(req);
3464 if (rsize > nb) { /* Fail if wraps around zero */
3465 char* mp = (char*)(CALL_MMAP(rsize));
3469 mmap_flag = IS_MMAPPED_BIT;
3474 if (HAVE_MORECORE && tbase == CMFAIL) { /* Try noncontiguous MORECORE */
3475 size_t asize = granularity_align(nb + TOP_FOOT_SIZE + MALLOC_ALIGNMENT + SIZE_T_ONE);
3476 if (asize < HALF_MAX_SIZE_T) {
3479 ACQUIRE_MORECORE_LOCK();
3480 br = (char*)(CALL_MORECORE(asize));
3481 end = (char*)(CALL_MORECORE(0));
3482 RELEASE_MORECORE_LOCK();
3483 if (br != CMFAIL && end != CMFAIL && br < end) {
3484 size_t ssize = end - br;
3485 if (ssize > nb + TOP_FOOT_SIZE) {
3493 if (tbase != CMFAIL) {
3495 if ((m->footprint += tsize) > m->max_footprint)
3496 m->max_footprint = m->footprint;
3498 if (!is_initialized(m)) { /* first-time initialization */
3499 m->seg.base = m->least_addr = tbase;
3500 m->seg.size = tsize;
3501 m->seg.sflags = mmap_flag;
3502 m->magic = mparams.magic;
3505 init_top(m, (mchunkptr)tbase, tsize - TOP_FOOT_SIZE);
3507 /* Offset top by embedded malloc_state */
3508 mchunkptr mn = next_chunk(mem2chunk(m));
3509 init_top(m, mn, (size_t)((tbase + tsize) - (char*)mn) -TOP_FOOT_SIZE);
3514 /* Try to merge with an existing segment */
3515 msegmentptr sp = &m->seg;
3516 while (sp != 0 && tbase != sp->base + sp->size)
3519 !is_extern_segment(sp) &&
3520 (sp->sflags & IS_MMAPPED_BIT) == mmap_flag &&
3521 segment_holds(sp, m->top)) { /* append */
3523 init_top(m, m->top, m->topsize + tsize);
3526 if (tbase < m->least_addr)
3527 m->least_addr = tbase;
3529 while (sp != 0 && sp->base != tbase + tsize)
3532 !is_extern_segment(sp) &&
3533 (sp->sflags & IS_MMAPPED_BIT) == mmap_flag) {
3534 char* oldbase = sp->base;
3537 return prepend_alloc(m, tbase, oldbase, nb);
3540 add_segment(m, tbase, tsize, mmap_flag);
3544 if (nb < m->topsize) { /* Allocate from new or extended top space */
3545 size_t rsize = m->topsize -= nb;
3546 mchunkptr p = m->top;
3547 mchunkptr r = m->top = chunk_plus_offset(p, nb);
3548 r->head = rsize | PINUSE_BIT;
3549 set_size_and_pinuse_of_inuse_chunk(m, p, nb);
3550 check_top_chunk(m, m->top);
3551 check_malloced_chunk(m, chunk2mem(p), nb);
3552 return chunk2mem(p);
3556 MALLOC_FAILURE_ACTION;
3560 /* ----------------------- system deallocation -------------------------- */
3562 /* Unmap and unlink any mmapped segments that don't contain used chunks */
3563 static size_t release_unused_segments(mstate m) {
3564 size_t released = 0;
3565 msegmentptr pred = &m->seg;
3566 msegmentptr sp = pred->next;
3568 char* base = sp->base;
3569 size_t size = sp->size;
3570 msegmentptr next = sp->next;
3571 if (is_mmapped_segment(sp) && !is_extern_segment(sp)) {
3572 mchunkptr p = align_as_chunk(base);
3573 size_t psize = chunksize(p);
3574 /* Can unmap if first chunk holds entire segment and not pinned */
3575 if (!cinuse(p) && (char*)p + psize >= base + size - TOP_FOOT_SIZE) {
3576 tchunkptr tp = (tchunkptr)p;
3577 assert(segment_holds(sp, (char*)sp));
3583 unlink_large_chunk(m, tp);
3585 if (CALL_MUNMAP(base, size) == 0) {
3587 m->footprint -= size;
3588 /* unlink obsoleted record */
3592 else { /* back out if cannot unmap */
3593 insert_large_chunk(m, tp, psize);
3603 static int sys_trim(mstate m, size_t pad) {
3604 size_t released = 0;
3605 if (pad < MAX_REQUEST && is_initialized(m)) {
3606 pad += TOP_FOOT_SIZE; /* ensure enough room for segment overhead */
3608 if (m->topsize > pad) {
3609 /* Shrink top space in granularity-size units, keeping at least one */
3610 size_t unit = mparams.granularity;
3611 size_t extra = ((m->topsize - pad + (unit - SIZE_T_ONE)) / unit -
3613 msegmentptr sp = segment_holding(m, (char*)m->top);
3615 if (!is_extern_segment(sp)) {
3616 if (is_mmapped_segment(sp)) {
3618 sp->size >= extra &&
3619 !has_segment_link(m, sp)) { /* can't shrink if pinned */
3620 size_t newsize = sp->size - extra;
3621 /* Prefer mremap, fall back to munmap */
3622 if ((CALL_MREMAP(sp->base, sp->size, newsize, 0) != MFAIL) ||
3623 (CALL_MUNMAP(sp->base + newsize, extra) == 0)) {
3628 else if (HAVE_MORECORE) {
3629 if (extra >= HALF_MAX_SIZE_T) /* Avoid wrapping negative */
3630 extra = (HALF_MAX_SIZE_T) + SIZE_T_ONE - unit;
3631 ACQUIRE_MORECORE_LOCK();
3633 /* Make sure end of memory is where we last set it. */
3634 char* old_br = (char*)(CALL_MORECORE(0));
3635 if (old_br == sp->base + sp->size) {
3636 char* rel_br = (char*)(CALL_MORECORE(-extra));
3637 char* new_br = (char*)(CALL_MORECORE(0));
3638 if (rel_br != CMFAIL && new_br < old_br)
3639 released = old_br - new_br;
3642 RELEASE_MORECORE_LOCK();
3646 if (released != 0) {
3647 sp->size -= released;
3648 m->footprint -= released;
3649 init_top(m, m->top, m->topsize - released);
3650 check_top_chunk(m, m->top);
3654 /* Unmap any unused mmapped segments */
3656 released += release_unused_segments(m);
3658 /* On failure, disable autotrim to avoid repeated failed future calls */
3660 m->trim_check = MAX_SIZE_T;
3663 return (released != 0)? 1 : 0;
3666 /* ---------------------------- malloc support --------------------------- */
3668 /* allocate a large request from the best fitting chunk in a treebin */
3669 static void* tmalloc_large(mstate m, size_t nb) {
3671 size_t rsize = -nb; /* Unsigned negation */
3674 compute_tree_index(nb, idx);
3676 if ((t = *treebin_at(m, idx)) != 0) {
3677 /* Traverse tree for this bin looking for node with size == nb */
3678 size_t sizebits = nb << leftshift_for_tree_index(idx);
3679 tchunkptr rst = 0; /* The deepest untaken right subtree */
3682 size_t trem = chunksize(t) - nb;
3685 if ((rsize = trem) == 0)
3689 t = t->child[(sizebits >> (SIZE_T_BITSIZE-SIZE_T_ONE)) & 1];
3690 if (rt != 0 && rt != t)
3693 t = rst; /* set t to least subtree holding sizes > nb */
3700 if (t == 0 && v == 0) { /* set t to root of next non-empty treebin */
3701 binmap_t leftbits = left_bits(idx2bit(idx)) & m->treemap;
3702 if (leftbits != 0) {
3704 binmap_t leastbit = least_bit(leftbits);
3705 compute_bit2idx(leastbit, i);
3706 t = *treebin_at(m, i);
3710 while (t != 0) { /* find smallest of tree or subtree */
3711 size_t trem = chunksize(t) - nb;
3716 t = leftmost_child(t);
3719 /* If dv is a better fit, return 0 so malloc will use it */
3720 if (v != 0 && rsize < (size_t)(m->dvsize - nb)) {
3721 if (RTCHECK(ok_address(m, v))) { /* split */
3722 mchunkptr r = chunk_plus_offset(v, nb);
3723 assert(chunksize(v) == rsize + nb);
3724 if (RTCHECK(ok_next(v, r))) {
3725 unlink_large_chunk(m, v);
3726 if (rsize < MIN_CHUNK_SIZE)
3727 set_inuse_and_pinuse(m, v, (rsize + nb));
3729 set_size_and_pinuse_of_inuse_chunk(m, v, nb);
3730 set_size_and_pinuse_of_free_chunk(r, rsize);
3731 insert_chunk(m, r, rsize);
3733 return chunk2mem(v);
3736 CORRUPTION_ERROR_ACTION(m);
3741 /* allocate a small request from the best fitting chunk in a treebin */
3742 static void* tmalloc_small(mstate m, size_t nb) {
3746 binmap_t leastbit = least_bit(m->treemap);
3747 compute_bit2idx(leastbit, i);
3749 v = t = *treebin_at(m, i);
3750 rsize = chunksize(t) - nb;
3752 while ((t = leftmost_child(t)) != 0) {
3753 size_t trem = chunksize(t) - nb;
3760 if (RTCHECK(ok_address(m, v))) {
3761 mchunkptr r = chunk_plus_offset(v, nb);
3762 assert(chunksize(v) == rsize + nb);
3763 if (RTCHECK(ok_next(v, r))) {
3764 unlink_large_chunk(m, v);
3765 if (rsize < MIN_CHUNK_SIZE)
3766 set_inuse_and_pinuse(m, v, (rsize + nb));
3768 set_size_and_pinuse_of_inuse_chunk(m, v, nb);
3769 set_size_and_pinuse_of_free_chunk(r, rsize);
3770 replace_dv(m, r, rsize);
3772 return chunk2mem(v);
3776 CORRUPTION_ERROR_ACTION(m);
3780 /* --------------------------- realloc support --------------------------- */
3782 static void* internal_realloc(mstate m, void* oldmem, size_t bytes) {
3783 if (bytes >= MAX_REQUEST) {
3784 MALLOC_FAILURE_ACTION;
3787 if (!PREACTION(m)) {
3788 mchunkptr oldp = mem2chunk(oldmem);
3789 size_t oldsize = chunksize(oldp);
3790 mchunkptr next = chunk_plus_offset(oldp, oldsize);
3794 /* Try to either shrink or extend into top. Else malloc-copy-free */
3796 if (RTCHECK(ok_address(m, oldp) && ok_cinuse(oldp) &&
3797 ok_next(oldp, next) && ok_pinuse(next))) {
3798 size_t nb = request2size(bytes);
3799 if (is_mmapped(oldp))
3800 newp = mmap_resize(m, oldp, nb);
3801 else if (oldsize >= nb) { /* already big enough */
3802 size_t rsize = oldsize - nb;
3804 if (rsize >= MIN_CHUNK_SIZE) {
3805 mchunkptr remainder = chunk_plus_offset(newp, nb);
3806 set_inuse(m, newp, nb);
3807 set_inuse(m, remainder, rsize);
3808 extra = chunk2mem(remainder);
3811 else if (next == m->top && oldsize + m->topsize > nb) {
3812 /* Expand into top */
3813 size_t newsize = oldsize + m->topsize;
3814 size_t newtopsize = newsize - nb;
3815 mchunkptr newtop = chunk_plus_offset(oldp, nb);
3816 set_inuse(m, oldp, nb);
3817 newtop->head = newtopsize |PINUSE_BIT;
3819 m->topsize = newtopsize;
3824 USAGE_ERROR_ACTION(m, oldmem);
3833 internal_free(m, extra);
3835 check_inuse_chunk(m, newp);
3836 return chunk2mem(newp);
3839 void* newmem = internal_malloc(m, bytes);
3841 size_t oc = oldsize - overhead_for(oldp);
3842 memcpy(newmem, oldmem, (oc < bytes)? oc : bytes);
3843 internal_free(m, oldmem);
3851 /* --------------------------- memalign support -------------------------- */
3853 static void* internal_memalign(mstate m, size_t alignment, size_t bytes) {
3854 if (alignment <= MALLOC_ALIGNMENT) /* Can just use malloc */
3855 return internal_malloc(m, bytes);
3856 if (alignment < MIN_CHUNK_SIZE) /* must be at least a minimum chunk size */
3857 alignment = MIN_CHUNK_SIZE;
3858 if ((alignment & (alignment-SIZE_T_ONE)) != 0) {/* Ensure a power of 2 */
3859 size_t a = MALLOC_ALIGNMENT << 1;
3860 while (a < alignment) a <<= 1;
3864 if (bytes >= MAX_REQUEST - alignment) {
3865 if (m != 0) { /* Test isn't needed but avoids compiler warning */
3866 MALLOC_FAILURE_ACTION;
3870 size_t nb = request2size(bytes);
3871 size_t req = nb + alignment + MIN_CHUNK_SIZE - CHUNK_OVERHEAD;
3872 char* mem = (char*)internal_malloc(m, req);
3876 mchunkptr p = mem2chunk(mem);
3878 if (PREACTION(m)) return 0;
3879 if ((((size_t)(mem)) % alignment) != 0) { /* misaligned */
3881 Find an aligned spot inside chunk. Since we need to give
3882 back leading space in a chunk of at least MIN_CHUNK_SIZE, if
3883 the first calculation places us at a spot with less than
3884 MIN_CHUNK_SIZE leader, we can move to the next aligned spot.
3885 We've allocated enough total room so that this is always
3888 char* br = (char*)mem2chunk((size_t)(((size_t)(mem +
3892 char* pos = ((size_t)(br - (char*)(p)) >= MIN_CHUNK_SIZE)?
3894 mchunkptr newp = (mchunkptr)pos;
3895 size_t leadsize = pos - (char*)(p);
3896 size_t newsize = chunksize(p) - leadsize;
3898 if (is_mmapped(p)) { /* For mmapped chunks, just adjust offset */
3899 newp->prev_foot = p->prev_foot + leadsize;
3900 newp->head = (newsize|CINUSE_BIT);
3902 else { /* Otherwise, give back leader, use the rest */
3903 set_inuse(m, newp, newsize);
3904 set_inuse(m, p, leadsize);
3905 leader = chunk2mem(p);
3910 /* Give back spare room at the end */
3911 if (!is_mmapped(p)) {
3912 size_t size = chunksize(p);
3913 if (size > nb + MIN_CHUNK_SIZE) {
3914 size_t remainder_size = size - nb;
3915 mchunkptr remainder = chunk_plus_offset(p, nb);
3916 set_inuse(m, p, nb);
3917 set_inuse(m, remainder, remainder_size);
3918 trailer = chunk2mem(remainder);
3922 assert (chunksize(p) >= nb);
3923 assert((((size_t)(chunk2mem(p))) % alignment) == 0);
3924 check_inuse_chunk(m, p);
3927 internal_free(m, leader);
3930 internal_free(m, trailer);
3932 return chunk2mem(p);
3938 /* ------------------------ comalloc/coalloc support --------------------- */
3940 static void** ialloc(mstate m,
3946 This provides common support for independent_X routines, handling
3947 all of the combinations that can result.
3950 bit 0 set if all elements are same size (using sizes[0])
3951 bit 1 set if elements should be zeroed
3954 size_t element_size; /* chunksize of each element, if all same */
3955 size_t contents_size; /* total size of elements */
3956 size_t array_size; /* request size of pointer array */
3957 void* mem; /* malloced aggregate space */
3958 mchunkptr p; /* corresponding chunk */
3959 size_t remainder_size; /* remaining bytes while splitting */
3960 void** marray; /* either "chunks" or malloced ptr array */
3961 mchunkptr array_chunk; /* chunk for malloced ptr array */
3962 flag_t was_enabled; /* to disable mmap */
3966 /* compute array length, if needed */
3968 if (n_elements == 0)
3969 return chunks; /* nothing to do */
3974 /* if empty req, must still return chunk representing empty array */
3975 if (n_elements == 0)
3976 return (void**)internal_malloc(m, 0);
3978 array_size = request2size(n_elements * (sizeof(void*)));
3981 /* compute total element size */
3982 if (opts & 0x1) { /* all-same-size */
3983 element_size = request2size(*sizes);
3984 contents_size = n_elements * element_size;
3986 else { /* add up all the sizes */
3989 for (i = 0; i != n_elements; ++i)
3990 contents_size += request2size(sizes[i]);
3993 size = contents_size + array_size;
3996 Allocate the aggregate chunk. First disable direct-mmapping so
3997 malloc won't use it, since we would not be able to later
3998 free/realloc space internal to a segregated mmap region.
4000 was_enabled = use_mmap(m);
4002 mem = internal_malloc(m, size - CHUNK_OVERHEAD);
4008 if (PREACTION(m)) return 0;
4010 remainder_size = chunksize(p);
4012 assert(!is_mmapped(p));
4014 if (opts & 0x2) { /* optionally clear the elements */
4015 memset((size_t*)mem, 0, remainder_size - SIZE_T_SIZE - array_size);
4018 /* If not provided, allocate the pointer array as final part of chunk */
4020 size_t array_chunk_size;
4021 array_chunk = chunk_plus_offset(p, contents_size);
4022 array_chunk_size = remainder_size - contents_size;
4023 marray = (void**) (chunk2mem(array_chunk));
4024 set_size_and_pinuse_of_inuse_chunk(m, array_chunk, array_chunk_size);
4025 remainder_size = contents_size;
4028 /* split out elements */
4029 for (i = 0; ; ++i) {
4030 marray[i] = chunk2mem(p);
4031 if (i != n_elements-1) {
4032 if (element_size != 0)
4033 size = element_size;
4035 size = request2size(sizes[i]);
4036 remainder_size -= size;
4037 set_size_and_pinuse_of_inuse_chunk(m, p, size);
4038 p = chunk_plus_offset(p, size);
4040 else { /* the final element absorbs any overallocation slop */
4041 set_size_and_pinuse_of_inuse_chunk(m, p, remainder_size);
4047 if (marray != chunks) {
4048 /* final element must have exactly exhausted chunk */
4049 if (element_size != 0) {
4050 assert(remainder_size == element_size);
4053 assert(remainder_size == request2size(sizes[i]));
4055 check_inuse_chunk(m, mem2chunk(marray));
4057 for (i = 0; i != n_elements; ++i)
4058 check_inuse_chunk(m, mem2chunk(marray[i]));
4067 /* -------------------------- public routines ---------------------------- */
4071 void* dlmalloc(size_t bytes) {
4074 If a small request (< 256 bytes minus per-chunk overhead):
4075 1. If one exists, use a remainderless chunk in associated smallbin.
4076 (Remainderless means that there are too few excess bytes to
4077 represent as a chunk.)
4078 2. If it is big enough, use the dv chunk, which is normally the
4079 chunk adjacent to the one used for the most recent small request.
4080 3. If one exists, split the smallest available chunk in a bin,
4081 saving remainder in dv.
4082 4. If it is big enough, use the top chunk.
4083 5. If available, get memory from system and use it
4084 Otherwise, for a large request:
4085 1. Find the smallest available binned chunk that fits, and use it
4086 if it is better fitting than dv chunk, splitting if necessary.
4087 2. If better fitting than any binned chunk, use the dv chunk.
4088 3. If it is big enough, use the top chunk.
4089 4. If request size >= mmap threshold, try to directly mmap this chunk.
4090 5. If available, get memory from system and use it
4092 The ugly goto's here ensure that postaction occurs along all paths.
4095 if (!PREACTION(gm)) {
4098 if (bytes <= MAX_SMALL_REQUEST) {
4101 nb = (bytes < MIN_REQUEST)? MIN_CHUNK_SIZE : pad_request(bytes);
4102 idx = small_index(nb);
4103 smallbits = gm->smallmap >> idx;
4105 if ((smallbits & 0x3U) != 0) { /* Remainderless fit to a smallbin. */
4107 idx += ~smallbits & 1; /* Uses next bin if idx empty */
4108 b = smallbin_at(gm, idx);
4110 assert(chunksize(p) == small_index2size(idx));
4111 unlink_first_small_chunk(gm, b, p, idx);
4112 set_inuse_and_pinuse(gm, p, small_index2size(idx));
4114 check_malloced_chunk(gm, mem, nb);
4118 else if (nb > gm->dvsize) {
4119 if (smallbits != 0) { /* Use chunk in next nonempty smallbin */
4123 binmap_t leftbits = (smallbits << idx) & left_bits(idx2bit(idx));
4124 binmap_t leastbit = least_bit(leftbits);
4125 compute_bit2idx(leastbit, i);
4126 b = smallbin_at(gm, i);
4128 assert(chunksize(p) == small_index2size(i));
4129 unlink_first_small_chunk(gm, b, p, i);
4130 rsize = small_index2size(i) - nb;
4131 /* Fit here cannot be remainderless if 4byte sizes */
4132 if (SIZE_T_SIZE != 4 && rsize < MIN_CHUNK_SIZE)
4133 set_inuse_and_pinuse(gm, p, small_index2size(i));
4135 set_size_and_pinuse_of_inuse_chunk(gm, p, nb);
4136 r = chunk_plus_offset(p, nb);
4137 set_size_and_pinuse_of_free_chunk(r, rsize);
4138 replace_dv(gm, r, rsize);
4141 check_malloced_chunk(gm, mem, nb);
4145 else if (gm->treemap != 0 && (mem = tmalloc_small(gm, nb)) != 0) {
4146 check_malloced_chunk(gm, mem, nb);
4151 else if (bytes >= MAX_REQUEST)
4152 nb = MAX_SIZE_T; /* Too big to allocate. Force failure (in sys alloc) */
4154 nb = pad_request(bytes);
4155 if (gm->treemap != 0 && (mem = tmalloc_large(gm, nb)) != 0) {
4156 check_malloced_chunk(gm, mem, nb);
4161 if (nb <= gm->dvsize) {
4162 size_t rsize = gm->dvsize - nb;
4163 mchunkptr p = gm->dv;
4164 if (rsize >= MIN_CHUNK_SIZE) { /* split dv */
4165 mchunkptr r = gm->dv = chunk_plus_offset(p, nb);
4167 set_size_and_pinuse_of_free_chunk(r, rsize);
4168 set_size_and_pinuse_of_inuse_chunk(gm, p, nb);
4170 else { /* exhaust dv */
4171 size_t dvs = gm->dvsize;
4174 set_inuse_and_pinuse(gm, p, dvs);
4177 check_malloced_chunk(gm, mem, nb);
4181 else if (nb < gm->topsize) { /* Split top */
4182 size_t rsize = gm->topsize -= nb;
4183 mchunkptr p = gm->top;
4184 mchunkptr r = gm->top = chunk_plus_offset(p, nb);
4185 r->head = rsize | PINUSE_BIT;
4186 set_size_and_pinuse_of_inuse_chunk(gm, p, nb);
4188 check_top_chunk(gm, gm->top);
4189 check_malloced_chunk(gm, mem, nb);
4193 mem = sys_alloc(gm, nb);
4203 void dlfree(void* mem) {
4205 Consolidate freed chunks with preceeding or succeeding bordering
4206 free chunks, if they exist, and then place in a bin. Intermixed
4207 with special cases for top, dv, mmapped chunks, and usage errors.
4211 mchunkptr p = mem2chunk(mem);
4213 mstate fm = get_mstate_for(p);
4214 if (!ok_magic(fm)) {
4215 USAGE_ERROR_ACTION(fm, p);
4220 #endif /* FOOTERS */
4221 if (!PREACTION(fm)) {
4222 check_inuse_chunk(fm, p);
4223 if (RTCHECK(ok_address(fm, p) && ok_cinuse(p))) {
4224 size_t psize = chunksize(p);
4225 mchunkptr next = chunk_plus_offset(p, psize);
4227 size_t prevsize = p->prev_foot;
4228 if ((prevsize & IS_MMAPPED_BIT) != 0) {
4229 prevsize &= ~IS_MMAPPED_BIT;
4230 psize += prevsize + MMAP_FOOT_PAD;
4231 if (CALL_MUNMAP((char*)p - prevsize, psize) == 0)
4232 fm->footprint -= psize;
4236 mchunkptr prev = chunk_minus_offset(p, prevsize);
4239 if (RTCHECK(ok_address(fm, prev))) { /* consolidate backward */
4241 unlink_chunk(fm, p, prevsize);
4243 else if ((next->head & INUSE_BITS) == INUSE_BITS) {
4245 set_free_with_pinuse(p, psize, next);
4254 if (RTCHECK(ok_next(p, next) && ok_pinuse(next))) {
4255 if (!cinuse(next)) { /* consolidate forward */
4256 if (next == fm->top) {
4257 size_t tsize = fm->topsize += psize;
4259 p->head = tsize | PINUSE_BIT;
4264 if (should_trim(fm, tsize))
4268 else if (next == fm->dv) {
4269 size_t dsize = fm->dvsize += psize;
4271 set_size_and_pinuse_of_free_chunk(p, dsize);
4275 size_t nsize = chunksize(next);
4277 unlink_chunk(fm, next, nsize);
4278 set_size_and_pinuse_of_free_chunk(p, psize);
4286 set_free_with_pinuse(p, psize, next);
4287 insert_chunk(fm, p, psize);
4288 check_free_chunk(fm, p);
4293 USAGE_ERROR_ACTION(fm, p);
4300 #endif /* FOOTERS */
4303 void* dlcalloc(size_t n_elements, size_t elem_size) {
4306 if (n_elements != 0) {
4307 req = n_elements * elem_size;
4308 if (((n_elements | elem_size) & ~(size_t)0xffff) &&
4309 (req / n_elements != elem_size))
4310 req = MAX_SIZE_T; /* force downstream failure on overflow */
4312 mem = dlmalloc(req);
4313 if (mem != 0 && calloc_must_clear(mem2chunk(mem)))
4314 memset(mem, 0, req);
4318 void* dlrealloc(void* oldmem, size_t bytes) {
4320 return dlmalloc(bytes);
4321 #ifdef REALLOC_ZERO_BYTES_FREES
4326 #endif /* REALLOC_ZERO_BYTES_FREES */
4331 mstate m = get_mstate_for(mem2chunk(oldmem));
4333 USAGE_ERROR_ACTION(m, oldmem);
4336 #endif /* FOOTERS */
4337 return internal_realloc(m, oldmem, bytes);
4341 void* dlmemalign(size_t alignment, size_t bytes) {
4342 return internal_memalign(gm, alignment, bytes);
4345 void** dlindependent_calloc(size_t n_elements, size_t elem_size,
4347 size_t sz = elem_size; /* serves as 1-element array */
4348 return ialloc(gm, n_elements, &sz, 3, chunks);
4351 void** dlindependent_comalloc(size_t n_elements, size_t sizes[],
4353 return ialloc(gm, n_elements, sizes, 0, chunks);
4356 void* dlvalloc(size_t bytes) {
4359 pagesz = mparams.page_size;
4360 return dlmemalign(pagesz, bytes);
4363 void* dlpvalloc(size_t bytes) {
4366 pagesz = mparams.page_size;
4367 return dlmemalign(pagesz, (bytes + pagesz - SIZE_T_ONE) & ~(pagesz - SIZE_T_ONE));
4370 int dlmalloc_trim(size_t pad) {
4372 if (!PREACTION(gm)) {
4373 result = sys_trim(gm, pad);
4379 size_t dlmalloc_footprint(void) {
4380 return gm->footprint;
4383 size_t dlmalloc_max_footprint(void) {
4384 return gm->max_footprint;
4388 struct mallinfo dlmallinfo(void) {
4389 return internal_mallinfo(gm);
4391 #endif /* NO_MALLINFO */
4393 void dlmalloc_stats() {
4394 internal_malloc_stats(gm);
4397 size_t dlmalloc_usable_size(void* mem) {
4399 mchunkptr p = mem2chunk(mem);
4401 return chunksize(p) - overhead_for(p);
4406 int dlmallopt(int param_number, int value) {
4407 return change_mparam(param_number, value);
4410 #endif /* !ONLY_MSPACES */
4412 /* ----------------------------- user mspaces ---------------------------- */
4416 static mstate init_user_mstate(char* tbase, size_t tsize) {
4417 size_t msize = pad_request(sizeof(struct malloc_state));
4419 mchunkptr msp = align_as_chunk(tbase);
4420 mstate m = (mstate)(chunk2mem(msp));
4421 memset(m, 0, msize);
4422 INITIAL_LOCK(&m->mutex);
4423 msp->head = (msize|PINUSE_BIT|CINUSE_BIT);
4424 m->seg.base = m->least_addr = tbase;
4425 m->seg.size = m->footprint = m->max_footprint = tsize;
4426 m->magic = mparams.magic;
4427 m->mflags = mparams.default_mflags;
4428 disable_contiguous(m);
4430 mn = next_chunk(mem2chunk(m));
4431 init_top(m, mn, (size_t)((tbase + tsize) - (char*)mn) - TOP_FOOT_SIZE);
4432 check_top_chunk(m, m->top);
4436 mspace create_mspace(size_t capacity, int locked) {
4438 size_t msize = pad_request(sizeof(struct malloc_state));
4439 init_mparams(); /* Ensure pagesize etc initialized */
4441 if (capacity < (size_t) -(msize + TOP_FOOT_SIZE + mparams.page_size)) {
4442 size_t rs = ((capacity == 0)? mparams.granularity :
4443 (capacity + TOP_FOOT_SIZE + msize));
4444 size_t tsize = granularity_align(rs);
4445 char* tbase = (char*)(CALL_MMAP(tsize));
4446 if (tbase != CMFAIL) {
4447 m = init_user_mstate(tbase, tsize);
4448 m->seg.sflags = IS_MMAPPED_BIT;
4449 set_lock(m, locked);
4455 mspace create_mspace_with_base(void* base, size_t capacity, int locked) {
4457 size_t msize = pad_request(sizeof(struct malloc_state));
4458 init_mparams(); /* Ensure pagesize etc initialized */
4460 if (capacity > msize + TOP_FOOT_SIZE &&
4461 capacity < (size_t) -(msize + TOP_FOOT_SIZE + mparams.page_size)) {
4462 m = init_user_mstate((char*)base, capacity);
4463 m->seg.sflags = EXTERN_BIT;
4464 set_lock(m, locked);
4469 size_t destroy_mspace(mspace msp) {
4471 mstate ms = (mstate)msp;
4473 msegmentptr sp = &ms->seg;
4475 char* base = sp->base;
4476 size_t size = sp->size;
4477 flag_t flag = sp->sflags;
4479 if ((flag & IS_MMAPPED_BIT) && !(flag & EXTERN_BIT) &&
4480 CALL_MUNMAP(base, size) == 0)
4485 USAGE_ERROR_ACTION(ms,ms);
4491 mspace versions of routines are near-clones of the global
4492 versions. This is not so nice but better than the alternatives.
4496 void* mspace_malloc(mspace msp, size_t bytes) {
4497 mstate ms = (mstate)msp;
4498 if (!ok_magic(ms)) {
4499 USAGE_ERROR_ACTION(ms,ms);
4502 if (!PREACTION(ms)) {
4505 if (bytes <= MAX_SMALL_REQUEST) {
4508 nb = (bytes < MIN_REQUEST)? MIN_CHUNK_SIZE : pad_request(bytes);
4509 idx = small_index(nb);
4510 smallbits = ms->smallmap >> idx;
4512 if ((smallbits & 0x3U) != 0) { /* Remainderless fit to a smallbin. */
4514 idx += ~smallbits & 1; /* Uses next bin if idx empty */
4515 b = smallbin_at(ms, idx);
4517 assert(chunksize(p) == small_index2size(idx));
4518 unlink_first_small_chunk(ms, b, p, idx);
4519 set_inuse_and_pinuse(ms, p, small_index2size(idx));
4521 check_malloced_chunk(ms, mem, nb);
4525 else if (nb > ms->dvsize) {
4526 if (smallbits != 0) { /* Use chunk in next nonempty smallbin */
4530 binmap_t leftbits = (smallbits << idx) & left_bits(idx2bit(idx));
4531 binmap_t leastbit = least_bit(leftbits);
4532 compute_bit2idx(leastbit, i);
4533 b = smallbin_at(ms, i);
4535 assert(chunksize(p) == small_index2size(i));
4536 unlink_first_small_chunk(ms, b, p, i);
4537 rsize = small_index2size(i) - nb;
4538 /* Fit here cannot be remainderless if 4byte sizes */
4539 if (SIZE_T_SIZE != 4 && rsize < MIN_CHUNK_SIZE)
4540 set_inuse_and_pinuse(ms, p, small_index2size(i));
4542 set_size_and_pinuse_of_inuse_chunk(ms, p, nb);
4543 r = chunk_plus_offset(p, nb);
4544 set_size_and_pinuse_of_free_chunk(r, rsize);
4545 replace_dv(ms, r, rsize);
4548 check_malloced_chunk(ms, mem, nb);
4552 else if (ms->treemap != 0 && (mem = tmalloc_small(ms, nb)) != 0) {
4553 check_malloced_chunk(ms, mem, nb);
4558 else if (bytes >= MAX_REQUEST)
4559 nb = MAX_SIZE_T; /* Too big to allocate. Force failure (in sys alloc) */
4561 nb = pad_request(bytes);
4562 if (ms->treemap != 0 && (mem = tmalloc_large(ms, nb)) != 0) {
4563 check_malloced_chunk(ms, mem, nb);
4568 if (nb <= ms->dvsize) {
4569 size_t rsize = ms->dvsize - nb;
4570 mchunkptr p = ms->dv;
4571 if (rsize >= MIN_CHUNK_SIZE) { /* split dv */
4572 mchunkptr r = ms->dv = chunk_plus_offset(p, nb);
4574 set_size_and_pinuse_of_free_chunk(r, rsize);
4575 set_size_and_pinuse_of_inuse_chunk(ms, p, nb);
4577 else { /* exhaust dv */
4578 size_t dvs = ms->dvsize;
4581 set_inuse_and_pinuse(ms, p, dvs);
4584 check_malloced_chunk(ms, mem, nb);
4588 else if (nb < ms->topsize) { /* Split top */
4589 size_t rsize = ms->topsize -= nb;
4590 mchunkptr p = ms->top;
4591 mchunkptr r = ms->top = chunk_plus_offset(p, nb);
4592 r->head = rsize | PINUSE_BIT;
4593 set_size_and_pinuse_of_inuse_chunk(ms, p, nb);
4595 check_top_chunk(ms, ms->top);
4596 check_malloced_chunk(ms, mem, nb);
4600 mem = sys_alloc(ms, nb);
4610 void mspace_free(mspace msp, void* mem) {
4612 mchunkptr p = mem2chunk(mem);
4614 mstate fm = get_mstate_for(p);
4616 mstate fm = (mstate)msp;
4617 #endif /* FOOTERS */
4618 if (!ok_magic(fm)) {
4619 USAGE_ERROR_ACTION(fm, p);
4622 if (!PREACTION(fm)) {
4623 check_inuse_chunk(fm, p);
4624 if (RTCHECK(ok_address(fm, p) && ok_cinuse(p))) {
4625 size_t psize = chunksize(p);
4626 mchunkptr next = chunk_plus_offset(p, psize);
4628 size_t prevsize = p->prev_foot;
4629 if ((prevsize & IS_MMAPPED_BIT) != 0) {
4630 prevsize &= ~IS_MMAPPED_BIT;
4631 psize += prevsize + MMAP_FOOT_PAD;
4632 if (CALL_MUNMAP((char*)p - prevsize, psize) == 0)
4633 fm->footprint -= psize;
4637 mchunkptr prev = chunk_minus_offset(p, prevsize);
4640 if (RTCHECK(ok_address(fm, prev))) { /* consolidate backward */
4642 unlink_chunk(fm, p, prevsize);
4644 else if ((next->head & INUSE_BITS) == INUSE_BITS) {
4646 set_free_with_pinuse(p, psize, next);
4655 if (RTCHECK(ok_next(p, next) && ok_pinuse(next))) {
4656 if (!cinuse(next)) { /* consolidate forward */
4657 if (next == fm->top) {
4658 size_t tsize = fm->topsize += psize;
4660 p->head = tsize | PINUSE_BIT;
4665 if (should_trim(fm, tsize))
4669 else if (next == fm->dv) {
4670 size_t dsize = fm->dvsize += psize;
4672 set_size_and_pinuse_of_free_chunk(p, dsize);
4676 size_t nsize = chunksize(next);
4678 unlink_chunk(fm, next, nsize);
4679 set_size_and_pinuse_of_free_chunk(p, psize);
4687 set_free_with_pinuse(p, psize, next);
4688 insert_chunk(fm, p, psize);
4689 check_free_chunk(fm, p);
4694 USAGE_ERROR_ACTION(fm, p);
4701 void* mspace_calloc(mspace msp, size_t n_elements, size_t elem_size) {
4704 mstate ms = (mstate)msp;
4705 if (!ok_magic(ms)) {
4706 USAGE_ERROR_ACTION(ms,ms);
4709 if (n_elements != 0) {
4710 req = n_elements * elem_size;
4711 if (((n_elements | elem_size) & ~(size_t)0xffff) &&
4712 (req / n_elements != elem_size))
4713 req = MAX_SIZE_T; /* force downstream failure on overflow */
4715 mem = internal_malloc(ms, req);
4716 if (mem != 0 && calloc_must_clear(mem2chunk(mem)))
4717 memset(mem, 0, req);
4721 void* mspace_realloc(mspace msp, void* oldmem, size_t bytes) {
4723 return mspace_malloc(msp, bytes);
4724 #ifdef REALLOC_ZERO_BYTES_FREES
4726 mspace_free(msp, oldmem);
4729 #endif /* REALLOC_ZERO_BYTES_FREES */
4732 mchunkptr p = mem2chunk(oldmem);
4733 mstate ms = get_mstate_for(p);
4735 mstate ms = (mstate)msp;
4736 #endif /* FOOTERS */
4737 if (!ok_magic(ms)) {
4738 USAGE_ERROR_ACTION(ms,ms);
4741 return internal_realloc(ms, oldmem, bytes);
4745 void* mspace_memalign(mspace msp, size_t alignment, size_t bytes) {
4746 mstate ms = (mstate)msp;
4747 if (!ok_magic(ms)) {
4748 USAGE_ERROR_ACTION(ms,ms);
4751 return internal_memalign(ms, alignment, bytes);
4754 void** mspace_independent_calloc(mspace msp, size_t n_elements,
4755 size_t elem_size, void* chunks[]) {
4756 size_t sz = elem_size; /* serves as 1-element array */
4757 mstate ms = (mstate)msp;
4758 if (!ok_magic(ms)) {
4759 USAGE_ERROR_ACTION(ms,ms);
4762 return ialloc(ms, n_elements, &sz, 3, chunks);
4765 void** mspace_independent_comalloc(mspace msp, size_t n_elements,
4766 size_t sizes[], void* chunks[]) {
4767 mstate ms = (mstate)msp;
4768 if (!ok_magic(ms)) {
4769 USAGE_ERROR_ACTION(ms,ms);
4772 return ialloc(ms, n_elements, sizes, 0, chunks);
4775 int mspace_trim(mspace msp, size_t pad) {
4777 mstate ms = (mstate)msp;
4779 if (!PREACTION(ms)) {
4780 result = sys_trim(ms, pad);
4785 USAGE_ERROR_ACTION(ms,ms);
4790 void mspace_malloc_stats(mspace msp) {
4791 mstate ms = (mstate)msp;
4793 internal_malloc_stats(ms);
4796 USAGE_ERROR_ACTION(ms,ms);
4800 size_t mspace_footprint(mspace msp) {
4802 mstate ms = (mstate)msp;
4804 result = ms->footprint;
4806 USAGE_ERROR_ACTION(ms,ms);
4811 size_t mspace_max_footprint(mspace msp) {
4813 mstate ms = (mstate)msp;
4815 result = ms->max_footprint;
4817 USAGE_ERROR_ACTION(ms,ms);
4823 struct mallinfo mspace_mallinfo(mspace msp) {
4824 mstate ms = (mstate)msp;
4825 if (!ok_magic(ms)) {
4826 USAGE_ERROR_ACTION(ms,ms);
4828 return internal_mallinfo(ms);
4830 #endif /* NO_MALLINFO */
4832 int mspace_mallopt(int param_number, int value) {
4833 return change_mparam(param_number, value);
4836 #endif /* MSPACES */
4838 /* -------------------- Alternative MORECORE functions ------------------- */
4841 Guidelines for creating a custom version of MORECORE:
4843 * For best performance, MORECORE should allocate in multiples of pagesize.
4844 * MORECORE may allocate more memory than requested. (Or even less,
4845 but this will usually result in a malloc failure.)
4846 * MORECORE must not allocate memory when given argument zero, but
4847 instead return one past the end address of memory from previous
4849 * For best performance, consecutive calls to MORECORE with positive
4850 arguments should return increasing addresses, indicating that
4851 space has been contiguously extended.
4852 * Even though consecutive calls to MORECORE need not return contiguous
4853 addresses, it must be OK for malloc'ed chunks to span multiple
4854 regions in those cases where they do happen to be contiguous.
4855 * MORECORE need not handle negative arguments -- it may instead
4856 just return MFAIL when given negative arguments.
4857 Negative arguments are always multiples of pagesize. MORECORE
4858 must not misinterpret negative args as large positive unsigned
4859 args. You can suppress all such calls from even occurring by defining
4860 MORECORE_CANNOT_TRIM,
4862 As an example alternative MORECORE, here is a custom allocator
4863 kindly contributed for pre-OSX macOS. It uses virtually but not
4864 necessarily physically contiguous non-paged memory (locked in,
4865 present and won't get swapped out). You can use it by uncommenting
4866 this section, adding some #includes, and setting up the appropriate
4869 #define MORECORE osMoreCore
4871 There is also a shutdown routine that should somehow be called for
4872 cleanup upon program exit.
4874 #define MAX_POOL_ENTRIES 100
4875 #define MINIMUM_MORECORE_SIZE (64 * 1024U)
4876 static int next_os_pool;
4877 void *our_os_pools[MAX_POOL_ENTRIES];
4879 void *osMoreCore(int size)
4882 static void *sbrk_top = 0;
4886 if (size < MINIMUM_MORECORE_SIZE)
4887 size = MINIMUM_MORECORE_SIZE;
4888 if (CurrentExecutionLevel() == kTaskLevel)
4889 ptr = PoolAllocateResident(size + RM_PAGE_SIZE, 0);
4892 return (void *) MFAIL;
4894 // save ptrs so they can be freed during cleanup
4895 our_os_pools[next_os_pool] = ptr;
4897 ptr = (void *) ((((size_t) ptr) + RM_PAGE_MASK) & ~RM_PAGE_MASK);
4898 sbrk_top = (char *) ptr + size;
4903 // we don't currently support shrink behavior
4904 return (void *) MFAIL;
4912 // cleanup any allocated memory pools
4913 // called as last thing before shutting down driver
4915 void osCleanupMem(void)
4919 for (ptr = our_os_pools; ptr < &our_os_pools[MAX_POOL_ENTRIES]; ptr++)
4922 PoolDeallocate(*ptr);
4930 /* -----------------------------------------------------------------------
4932 V2.8.3 Thu Sep 22 11:16:32 2005 Doug Lea (dl at gee)
4933 * Add max_footprint functions
4934 * Ensure all appropriate literals are size_t
4935 * Fix conditional compilation problem for some #define settings
4936 * Avoid concatenating segments with the one provided
4937 in create_mspace_with_base
4938 * Rename some variables to avoid compiler shadowing warnings
4939 * Use explicit lock initialization.
4940 * Better handling of sbrk interference.
4941 * Simplify and fix segment insertion, trimming and mspace_destroy
4942 * Reinstate REALLOC_ZERO_BYTES_FREES option from 2.7.x
4943 * Thanks especially to Dennis Flanagan for help on these.
4945 V2.8.2 Sun Jun 12 16:01:10 2005 Doug Lea (dl at gee)
4946 * Fix memalign brace error.
4948 V2.8.1 Wed Jun 8 16:11:46 2005 Doug Lea (dl at gee)
4949 * Fix improper #endif nesting in C++
4950 * Add explicit casts needed for C++
4952 V2.8.0 Mon May 30 14:09:02 2005 Doug Lea (dl at gee)
4953 * Use trees for large bins
4955 * Use segments to unify sbrk-based and mmap-based system allocation,
4956 removing need for emulation on most platforms without sbrk.
4957 * Default safety checks
4958 * Optional footer checks. Thanks to William Robertson for the idea.
4959 * Internal code refactoring
4960 * Incorporate suggestions and platform-specific changes.
4961 Thanks to Dennis Flanagan, Colin Plumb, Niall Douglas,
4962 Aaron Bachmann, Emery Berger, and others.
4963 * Speed up non-fastbin processing enough to remove fastbins.
4964 * Remove useless cfree() to avoid conflicts with other apps.
4965 * Remove internal memcpy, memset. Compilers handle builtins better.
4966 * Remove some options that no one ever used and rename others.
4968 V2.7.2 Sat Aug 17 09:07:30 2002 Doug Lea (dl at gee)
4969 * Fix malloc_state bitmap array misdeclaration
4971 V2.7.1 Thu Jul 25 10:58:03 2002 Doug Lea (dl at gee)
4972 * Allow tuning of FIRST_SORTED_BIN_SIZE
4973 * Use PTR_UINT as type for all ptr->int casts. Thanks to John Belmonte.
4974 * Better detection and support for non-contiguousness of MORECORE.
4975 Thanks to Andreas Mueller, Conal Walsh, and Wolfram Gloger
4976 * Bypass most of malloc if no frees. Thanks To Emery Berger.
4977 * Fix freeing of old top non-contiguous chunk im sysmalloc.
4978 * Raised default trim and map thresholds to 256K.
4979 * Fix mmap-related #defines. Thanks to Lubos Lunak.
4980 * Fix copy macros; added LACKS_FCNTL_H. Thanks to Neal Walfield.
4981 * Branch-free bin calculation
4982 * Default trim and mmap thresholds now 256K.
4984 V2.7.0 Sun Mar 11 14:14:06 2001 Doug Lea (dl at gee)
4985 * Introduce independent_comalloc and independent_calloc.
4986 Thanks to Michael Pachos for motivation and help.
4987 * Make optional .h file available
4988 * Allow > 2GB requests on 32bit systems.
4989 * new WIN32 sbrk, mmap, munmap, lock code from <Walter@GeNeSys-e.de>.
4990 Thanks also to Andreas Mueller <a.mueller at paradatec.de>,
4992 * Allow override of MALLOC_ALIGNMENT (Thanks to Ruud Waij for
4994 * memalign: check alignment arg
4995 * realloc: don't try to shift chunks backwards, since this
4996 leads to more fragmentation in some programs and doesn't
4997 seem to help in any others.
4998 * Collect all cases in malloc requiring system memory into sysmalloc
4999 * Use mmap as backup to sbrk
5000 * Place all internal state in malloc_state
5001 * Introduce fastbins (although similar to 2.5.1)
5002 * Many minor tunings and cosmetic improvements
5003 * Introduce USE_PUBLIC_MALLOC_WRAPPERS, USE_MALLOC_LOCK
5004 * Introduce MALLOC_FAILURE_ACTION, MORECORE_CONTIGUOUS
5005 Thanks to Tony E. Bennett <tbennett@nvidia.com> and others.
5006 * Include errno.h to support default failure action.
5008 V2.6.6 Sun Dec 5 07:42:19 1999 Doug Lea (dl at gee)
5009 * return null for negative arguments
5010 * Added Several WIN32 cleanups from Martin C. Fong <mcfong at yahoo.com>
5011 * Add 'LACKS_SYS_PARAM_H' for those systems without 'sys/param.h'
5012 (e.g. WIN32 platforms)
5013 * Cleanup header file inclusion for WIN32 platforms
5014 * Cleanup code to avoid Microsoft Visual C++ compiler complaints
5015 * Add 'USE_DL_PREFIX' to quickly allow co-existence with existing
5016 memory allocation routines
5017 * Set 'malloc_getpagesize' for WIN32 platforms (needs more work)
5018 * Use 'assert' rather than 'ASSERT' in WIN32 code to conform to
5019 usage of 'assert' in non-WIN32 code
5020 * Improve WIN32 'sbrk()' emulation's 'findRegion()' routine to
5022 * Always call 'fREe()' rather than 'free()'
5024 V2.6.5 Wed Jun 17 15:57:31 1998 Doug Lea (dl at gee)
5025 * Fixed ordering problem with boundary-stamping
5027 V2.6.3 Sun May 19 08:17:58 1996 Doug Lea (dl at gee)
5028 * Added pvalloc, as recommended by H.J. Liu
5029 * Added 64bit pointer support mainly from Wolfram Gloger
5030 * Added anonymously donated WIN32 sbrk emulation
5031 * Malloc, calloc, getpagesize: add optimizations from Raymond Nijssen
5032 * malloc_extend_top: fix mask error that caused wastage after
5034 * Add linux mremap support code from HJ Liu
5036 V2.6.2 Tue Dec 5 06:52:55 1995 Doug Lea (dl at gee)
5037 * Integrated most documentation with the code.
5038 * Add support for mmap, with help from
5039 Wolfram Gloger (Gloger@lrz.uni-muenchen.de).
5040 * Use last_remainder in more cases.
5041 * Pack bins using idea from colin@nyx10.cs.du.edu
5042 * Use ordered bins instead of best-fit threshhold
5043 * Eliminate block-local decls to simplify tracing and debugging.
5044 * Support another case of realloc via move into top
5045 * Fix error occuring when initial sbrk_base not word-aligned.
5046 * Rely on page size for units instead of SBRK_UNIT to
5047 avoid surprises about sbrk alignment conventions.
5048 * Add mallinfo, mallopt. Thanks to Raymond Nijssen
5049 (raymond@es.ele.tue.nl) for the suggestion.
5050 * Add `pad' argument to malloc_trim and top_pad mallopt parameter.
5051 * More precautions for cases where other routines call sbrk,
5052 courtesy of Wolfram Gloger (Gloger@lrz.uni-muenchen.de).
5053 * Added macros etc., allowing use in linux libc from
5054 H.J. Lu (hjl@gnu.ai.mit.edu)
5055 * Inverted this history list
5057 V2.6.1 Sat Dec 2 14:10:57 1995 Doug Lea (dl at gee)
5058 * Re-tuned and fixed to behave more nicely with V2.6.0 changes.
5059 * Removed all preallocation code since under current scheme
5060 the work required to undo bad preallocations exceeds
5061 the work saved in good cases for most test programs.
5062 * No longer use return list or unconsolidated bins since
5063 no scheme using them consistently outperforms those that don't
5064 given above changes.
5065 * Use best fit for very large chunks to prevent some worst-cases.
5066 * Added some support for debugging
5068 V2.6.0 Sat Nov 4 07:05:23 1995 Doug Lea (dl at gee)
5069 * Removed footers when chunks are in use. Thanks to
5070 Paul Wilson (wilson@cs.texas.edu) for the suggestion.
5072 V2.5.4 Wed Nov 1 07:54:51 1995 Doug Lea (dl at gee)
5073 * Added malloc_trim, with help from Wolfram Gloger
5074 (wmglo@Dent.MED.Uni-Muenchen.DE).
5076 V2.5.3 Tue Apr 26 10:16:01 1994 Doug Lea (dl at g)
5078 V2.5.2 Tue Apr 5 16:20:40 1994 Doug Lea (dl at g)
5079 * realloc: try to expand in both directions
5080 * malloc: swap order of clean-bin strategy;
5081 * realloc: only conditionally expand backwards
5082 * Try not to scavenge used bins
5083 * Use bin counts as a guide to preallocation
5084 * Occasionally bin return list chunks in first scan
5085 * Add a few optimizations from colin@nyx10.cs.du.edu
5087 V2.5.1 Sat Aug 14 15:40:43 1993 Doug Lea (dl at g)
5088 * faster bin computation & slightly different binning
5089 * merged all consolidations to one part of malloc proper
5090 (eliminating old malloc_find_space & malloc_clean_bin)
5091 * Scan 2 returns chunks (not just 1)
5092 * Propagate failure in realloc if malloc returns 0
5093 * Add stuff to allow compilation on non-ANSI compilers
5094 from kpv@research.att.com
5096 V2.5 Sat Aug 7 07:41:59 1993 Doug Lea (dl at g.oswego.edu)
5097 * removed potential for odd address access in prev_chunk
5098 * removed dependency on getpagesize.h
5099 * misc cosmetics and a bit more internal documentation
5100 * anticosmetics: mangled names in macros to evade debugger strangeness
5101 * tested on sparc, hp-700, dec-mips, rs6000
5102 with gcc & native cc (hp, dec only) allowing
5103 Detlefs & Zorn comparison study (in SIGPLAN Notices.)
5105 Trial version Fri Aug 28 13:14:29 1992 Doug Lea (dl at g.oswego.edu)
5106 * Based loosely on libg++-1.2X malloc. (It retains some of the overall
5107 structure of old version, but most details differ.)
5111 #endif /* !HAVE_MALLOC */