2 This is a version (aka dlmalloc) of malloc/free/realloc written by
3 Doug Lea and released to the public domain, as explained at
4 http://creativecommons.org/licenses/publicdomain. Send questions,
5 comments, complaints, performance data, etc to dl@cs.oswego.edu
7 * Version 2.8.3 Thu Sep 22 11:16:15 2005 Doug Lea (dl at gee)
9 Note: There may be an updated version of this malloc obtainable at
10 ftp://gee.cs.oswego.edu/pub/misc/malloc.c
11 Check before installing!
15 This library is all in one file to simplify the most common usage:
16 ftp it, compile it (-O3), and link it into another program. All of
17 the compile-time options default to reasonable values for use on
18 most platforms. You might later want to step through various
19 compile-time and dynamic tuning options.
21 For convenience, an include file for code using this malloc is at:
22 ftp://gee.cs.oswego.edu/pub/misc/malloc-2.8.3.h
23 You don't really need this .h file unless you call functions not
24 defined in your system include files. The .h file contains only the
25 excerpts from this file needed for using this malloc on ANSI C/C++
26 systems, so long as you haven't changed compile-time options about
27 naming and tuning parameters. If you do, then you can create your
28 own malloc.h that does include all settings by cutting at the point
29 indicated below. Note that you may already by default be using a C
30 library containing a malloc that is based on some version of this
31 malloc (for example in linux). You might still want to use the one
32 in this file to customize settings or to avoid overheads associated
33 with library versions.
37 Supported pointer/size_t representation: 4 or 8 bytes
38 size_t MUST be an unsigned type of the same width as
39 pointers. (If you are using an ancient system that declares
40 size_t as a signed type, or need it to be a different width
41 than pointers, you can use a previous release of this malloc
42 (e.g. 2.7.2) supporting these.)
44 Alignment: 8 bytes (default)
45 This suffices for nearly all current machines and C compilers.
46 However, you can define MALLOC_ALIGNMENT to be wider than this
47 if necessary (up to 128bytes), at the expense of using more space.
49 Minimum overhead per allocated chunk: 4 or 8 bytes (if 4byte sizes)
50 8 or 16 bytes (if 8byte sizes)
51 Each malloced chunk has a hidden word of overhead holding size
52 and status information, and additional cross-check word
53 if FOOTERS is defined.
55 Minimum allocated size: 4-byte ptrs: 16 bytes (including overhead)
56 8-byte ptrs: 32 bytes (including overhead)
58 Even a request for zero bytes (i.e., malloc(0)) returns a
59 pointer to something of the minimum allocatable size.
60 The maximum overhead wastage (i.e., number of extra bytes
61 allocated than were requested in malloc) is less than or equal
62 to the minimum size, except for requests >= mmap_threshold that
63 are serviced via mmap(), where the worst case wastage is about
64 32 bytes plus the remainder from a system page (the minimal
65 mmap unit); typically 4096 or 8192 bytes.
67 Security: static-safe; optionally more or less
68 The "security" of malloc refers to the ability of malicious
69 code to accentuate the effects of errors (for example, freeing
70 space that is not currently malloc'ed or overwriting past the
71 ends of chunks) in code that calls malloc. This malloc
72 guarantees not to modify any memory locations below the base of
73 heap, i.e., static variables, even in the presence of usage
74 errors. The routines additionally detect most improper frees
75 and reallocs. All this holds as long as the static bookkeeping
76 for malloc itself is not corrupted by some other means. This
77 is only one aspect of security -- these checks do not, and
78 cannot, detect all possible programming errors.
80 If FOOTERS is defined nonzero, then each allocated chunk
81 carries an additional check word to verify that it was malloced
82 from its space. These check words are the same within each
83 execution of a program using malloc, but differ across
84 executions, so externally crafted fake chunks cannot be
85 freed. This improves security by rejecting frees/reallocs that
86 could corrupt heap memory, in addition to the checks preventing
87 writes to statics that are always on. This may further improve
88 security at the expense of time and space overhead. (Note that
89 FOOTERS may also be worth using with MSPACES.)
91 By default detected errors cause the program to abort (calling
92 "abort()"). You can override this to instead proceed past
93 errors by defining PROCEED_ON_ERROR. In this case, a bad free
94 has no effect, and a malloc that encounters a bad address
95 caused by user overwrites will ignore the bad address by
96 dropping pointers and indices to all known memory. This may
97 be appropriate for programs that should continue if at all
98 possible in the face of programming errors, although they may
99 run out of memory because dropped memory is never reclaimed.
101 If you don't like either of these options, you can define
102 CORRUPTION_ERROR_ACTION and USAGE_ERROR_ACTION to do anything
103 else. And if if you are sure that your program using malloc has
104 no errors or vulnerabilities, you can define INSECURE to 1,
105 which might (or might not) provide a small performance improvement.
107 Thread-safety: NOT thread-safe unless USE_LOCKS defined
108 When USE_LOCKS is defined, each public call to malloc, free,
109 etc is surrounded with either a pthread mutex or a win32
110 spinlock (depending on WIN32). This is not especially fast, and
111 can be a major bottleneck. It is designed only to provide
112 minimal protection in concurrent environments, and to provide a
113 basis for extensions. If you are using malloc in a concurrent
114 program, consider instead using ptmalloc, which is derived from
115 a version of this malloc. (See http://www.malloc.de).
117 System requirements: Any combination of MORECORE and/or MMAP/MUNMAP
118 This malloc can use unix sbrk or any emulation (invoked using
119 the CALL_MORECORE macro) and/or mmap/munmap or any emulation
120 (invoked using CALL_MMAP/CALL_MUNMAP) to get and release system
121 memory. On most unix systems, it tends to work best if both
122 MORECORE and MMAP are enabled. On Win32, it uses emulations
123 based on VirtualAlloc. It also uses common C library functions
126 Compliance: I believe it is compliant with the Single Unix Specification
127 (See http://www.unix.org). Also SVID/XPG, ANSI C, and probably
130 * Overview of algorithms
132 This is not the fastest, most space-conserving, most portable, or
133 most tunable malloc ever written. However it is among the fastest
134 while also being among the most space-conserving, portable and
135 tunable. Consistent balance across these factors results in a good
136 general-purpose allocator for malloc-intensive programs.
138 In most ways, this malloc is a best-fit allocator. Generally, it
139 chooses the best-fitting existing chunk for a request, with ties
140 broken in approximately least-recently-used order. (This strategy
141 normally maintains low fragmentation.) However, for requests less
142 than 256bytes, it deviates from best-fit when there is not an
143 exactly fitting available chunk by preferring to use space adjacent
144 to that used for the previous small request, as well as by breaking
145 ties in approximately most-recently-used order. (These enhance
146 locality of series of small allocations.) And for very large requests
147 (>= 256Kb by default), it relies on system memory mapping
148 facilities, if supported. (This helps avoid carrying around and
149 possibly fragmenting memory used only for large chunks.)
151 All operations (except malloc_stats and mallinfo) have execution
152 times that are bounded by a constant factor of the number of bits in
153 a size_t, not counting any clearing in calloc or copying in realloc,
154 or actions surrounding MORECORE and MMAP that have times
155 proportional to the number of non-contiguous regions returned by
156 system allocation routines, which is often just 1.
158 The implementation is not very modular and seriously overuses
159 macros. Perhaps someday all C compilers will do as good a job
160 inlining modular code as can now be done by brute-force expansion,
161 but now, enough of them seem not to.
163 Some compilers issue a lot of warnings about code that is
164 dead/unreachable only on some platforms, and also about intentional
165 uses of negation on unsigned types. All known cases of each can be
168 For a longer but out of date high-level description, see
169 http://gee.cs.oswego.edu/dl/html/malloc.html
172 If MSPACES is defined, then in addition to malloc, free, etc.,
173 this file also defines mspace_malloc, mspace_free, etc. These
174 are versions of malloc routines that take an "mspace" argument
175 obtained using create_mspace, to control all internal bookkeeping.
176 If ONLY_MSPACES is defined, only these versions are compiled.
177 So if you would like to use this allocator for only some allocations,
178 and your system malloc for others, you can compile with
179 ONLY_MSPACES and then do something like...
180 static mspace mymspace = create_mspace(0,0); // for example
181 #define mymalloc(bytes) mspace_malloc(mymspace, bytes)
183 (Note: If you only need one instance of an mspace, you can instead
184 use "USE_DL_PREFIX" to relabel the global malloc.)
186 You can similarly create thread-local allocators by storing
187 mspaces as thread-locals. For example:
188 static __thread mspace tlms = 0;
189 void* tlmalloc(size_t bytes) {
190 if (tlms == 0) tlms = create_mspace(0, 0);
191 return mspace_malloc(tlms, bytes);
193 void tlfree(void* mem) { mspace_free(tlms, mem); }
195 Unless FOOTERS is defined, each mspace is completely independent.
196 You cannot allocate from one and free to another (although
197 conformance is only weakly checked, so usage errors are not always
198 caught). If FOOTERS is defined, then each chunk carries around a tag
199 indicating its originating mspace, and frees are directed to their
202 ------------------------- Compile-time options ---------------------------
204 Be careful in setting #define values for numerical constants of type
205 size_t. On some systems, literal values are not automatically extended
206 to size_t precision unless they are explicitly casted.
208 WIN32 default: defined if _WIN32 defined
209 Defining WIN32 sets up defaults for MS environment and compilers.
210 Otherwise defaults are for unix.
212 MALLOC_ALIGNMENT default: (size_t)8
213 Controls the minimum alignment for malloc'ed chunks. It must be a
214 power of two and at least 8, even on machines for which smaller
215 alignments would suffice. It may be defined as larger than this
216 though. Note however that code and data structures are optimized for
217 the case of 8-byte alignment.
219 MSPACES default: 0 (false)
220 If true, compile in support for independent allocation spaces.
221 This is only supported if HAVE_MMAP is true.
223 ONLY_MSPACES default: 0 (false)
224 If true, only compile in mspace versions, not regular versions.
226 USE_LOCKS default: 0 (false)
227 Causes each call to each public routine to be surrounded with
228 pthread or WIN32 mutex lock/unlock. (If set true, this can be
229 overridden on a per-mspace basis for mspace versions.)
232 If true, provide extra checking and dispatching by placing
233 information in the footers of allocated chunks. This adds
234 space and time overhead.
237 If true, omit checks for usage errors and heap space overwrites.
239 USE_DL_PREFIX default: NOT defined
240 Causes compiler to prefix all public routines with the string 'dl'.
241 This can be useful when you only want to use this malloc in one part
242 of a program, using your regular system malloc elsewhere.
244 ABORT default: defined as abort()
245 Defines how to abort on failed checks. On most systems, a failed
246 check cannot die with an "assert" or even print an informative
247 message, because the underlying print routines in turn call malloc,
248 which will fail again. Generally, the best policy is to simply call
249 abort(). It's not very useful to do more than this because many
250 errors due to overwriting will show up as address faults (null, odd
251 addresses etc) rather than malloc-triggered checks, so will also
252 abort. Also, most compilers know that abort() does not return, so
253 can better optimize code conditionally calling it.
255 PROCEED_ON_ERROR default: defined as 0 (false)
256 Controls whether detected bad addresses cause them to bypassed
257 rather than aborting. If set, detected bad arguments to free and
258 realloc are ignored. And all bookkeeping information is zeroed out
259 upon a detected overwrite of freed heap space, thus losing the
260 ability to ever return it from malloc again, but enabling the
261 application to proceed. If PROCEED_ON_ERROR is defined, the
262 static variable malloc_corruption_error_count is compiled in
263 and can be examined to see if errors have occurred. This option
264 generates slower code than the default abort policy.
266 DEBUG default: NOT defined
267 The DEBUG setting is mainly intended for people trying to modify
268 this code or diagnose problems when porting to new platforms.
269 However, it may also be able to better isolate user errors than just
270 using runtime checks. The assertions in the check routines spell
271 out in more detail the assumptions and invariants underlying the
272 algorithms. The checking is fairly extensive, and will slow down
273 execution noticeably. Calling malloc_stats or mallinfo with DEBUG
274 set will attempt to check every non-mmapped allocated and free chunk
275 in the course of computing the summaries.
277 ABORT_ON_ASSERT_FAILURE default: defined as 1 (true)
278 Debugging assertion failures can be nearly impossible if your
279 version of the assert macro causes malloc to be called, which will
280 lead to a cascade of further failures, blowing the runtime stack.
281 ABORT_ON_ASSERT_FAILURE cause assertions failures to call abort(),
282 which will usually make debugging easier.
284 MALLOC_FAILURE_ACTION default: sets errno to ENOMEM, or no-op on win32
285 The action to take before "return 0" when malloc fails to be able to
286 return memory because there is none available.
288 HAVE_MORECORE default: 1 (true) unless win32 or ONLY_MSPACES
289 True if this system supports sbrk or an emulation of it.
291 MORECORE default: sbrk
292 The name of the sbrk-style system routine to call to obtain more
293 memory. See below for guidance on writing custom MORECORE
294 functions. The type of the argument to sbrk/MORECORE varies across
295 systems. It cannot be size_t, because it supports negative
296 arguments, so it is normally the signed type of the same width as
297 size_t (sometimes declared as "intptr_t"). It doesn't much matter
298 though. Internally, we only call it with arguments less than half
299 the max value of a size_t, which should work across all reasonable
300 possibilities, although sometimes generating compiler warnings. See
301 near the end of this file for guidelines for creating a custom
304 MORECORE_CONTIGUOUS default: 1 (true)
305 If true, take advantage of fact that consecutive calls to MORECORE
306 with positive arguments always return contiguous increasing
307 addresses. This is true of unix sbrk. It does not hurt too much to
308 set it true anyway, since malloc copes with non-contiguities.
309 Setting it false when definitely non-contiguous saves time
310 and possibly wasted space it would take to discover this though.
312 MORECORE_CANNOT_TRIM default: NOT defined
313 True if MORECORE cannot release space back to the system when given
314 negative arguments. This is generally necessary only if you are
315 using a hand-crafted MORECORE function that cannot handle negative
318 HAVE_MMAP default: 1 (true)
319 True if this system supports mmap or an emulation of it. If so, and
320 HAVE_MORECORE is not true, MMAP is used for all system
321 allocation. If set and HAVE_MORECORE is true as well, MMAP is
322 primarily used to directly allocate very large blocks. It is also
323 used as a backup strategy in cases where MORECORE fails to provide
324 space from system. Note: A single call to MUNMAP is assumed to be
325 able to unmap memory that may have be allocated using multiple calls
326 to MMAP, so long as they are adjacent.
328 HAVE_MREMAP default: 1 on linux, else 0
329 If true realloc() uses mremap() to re-allocate large blocks and
330 extend or shrink allocation spaces.
332 MMAP_CLEARS default: 1 on unix
333 True if mmap clears memory so calloc doesn't need to. This is true
334 for standard unix mmap using /dev/zero.
336 USE_BUILTIN_FFS default: 0 (i.e., not used)
337 Causes malloc to use the builtin ffs() function to compute indices.
338 Some compilers may recognize and intrinsify ffs to be faster than the
339 supplied C version. Also, the case of x86 using gcc is special-cased
340 to an asm instruction, so is already as fast as it can be, and so
341 this setting has no effect. (On most x86s, the asm version is only
342 slightly faster than the C version.)
344 malloc_getpagesize default: derive from system includes, or 4096.
345 The system page size. To the extent possible, this malloc manages
346 memory from the system in page-size units. This may be (and
347 usually is) a function rather than a constant. This is ignored
348 if WIN32, where page size is determined using getSystemInfo during
351 USE_DEV_RANDOM default: 0 (i.e., not used)
352 Causes malloc to use /dev/random to initialize secure magic seed for
353 stamping footers. Otherwise, the current time is used.
355 NO_MALLINFO default: 0
356 If defined, don't compile "mallinfo". This can be a simple way
357 of dealing with mismatches between system declarations and
360 MALLINFO_FIELD_TYPE default: size_t
361 The type of the fields in the mallinfo struct. This was originally
362 defined as "int" in SVID etc, but is more usefully defined as
363 size_t. The value is used only if HAVE_USR_INCLUDE_MALLOC_H is not set
365 REALLOC_ZERO_BYTES_FREES default: not defined
366 This should be set if a call to realloc with zero bytes should
367 be the same as a call to free. Some people think it should. Otherwise,
368 since this malloc returns a unique pointer for malloc(0), so does
371 LACKS_UNISTD_H, LACKS_FCNTL_H, LACKS_SYS_PARAM_H, LACKS_SYS_MMAN_H
372 LACKS_STRINGS_H, LACKS_STRING_H, LACKS_SYS_TYPES_H, LACKS_ERRNO_H
373 LACKS_STDLIB_H default: NOT defined unless on WIN32
374 Define these if your system does not have these header files.
375 You might need to manually insert some of the declarations they provide.
377 DEFAULT_GRANULARITY default: page size if MORECORE_CONTIGUOUS,
378 system_info.dwAllocationGranularity in WIN32,
380 Also settable using mallopt(M_GRANULARITY, x)
381 The unit for allocating and deallocating memory from the system. On
382 most systems with contiguous MORECORE, there is no reason to
383 make this more than a page. However, systems with MMAP tend to
384 either require or encourage larger granularities. You can increase
385 this value to prevent system allocation functions to be called so
386 often, especially if they are slow. The value must be at least one
387 page and must be a power of two. Setting to 0 causes initialization
388 to either page size or win32 region size. (Note: In previous
389 versions of malloc, the equivalent of this option was called
392 DEFAULT_TRIM_THRESHOLD default: 2MB
393 Also settable using mallopt(M_TRIM_THRESHOLD, x)
394 The maximum amount of unused top-most memory to keep before
395 releasing via malloc_trim in free(). Automatic trimming is mainly
396 useful in long-lived programs using contiguous MORECORE. Because
397 trimming via sbrk can be slow on some systems, and can sometimes be
398 wasteful (in cases where programs immediately afterward allocate
399 more large chunks) the value should be high enough so that your
400 overall system performance would improve by releasing this much
401 memory. As a rough guide, you might set to a value close to the
402 average size of a process (program) running on your system.
403 Releasing this much memory would allow such a process to run in
404 memory. Generally, it is worth tuning trim thresholds when a
405 program undergoes phases where several large chunks are allocated
406 and released in ways that can reuse each other's storage, perhaps
407 mixed with phases where there are no such chunks at all. The trim
408 value must be greater than page size to have any useful effect. To
409 disable trimming completely, you can set to MAX_SIZE_T. Note that the trick
410 some people use of mallocing a huge space and then freeing it at
411 program startup, in an attempt to reserve system memory, doesn't
412 have the intended effect under automatic trimming, since that memory
413 will immediately be returned to the system.
415 DEFAULT_MMAP_THRESHOLD default: 256K
416 Also settable using mallopt(M_MMAP_THRESHOLD, x)
417 The request size threshold for using MMAP to directly service a
418 request. Requests of at least this size that cannot be allocated
419 using already-existing space will be serviced via mmap. (If enough
420 normal freed space already exists it is used instead.) Using mmap
421 segregates relatively large chunks of memory so that they can be
422 individually obtained and released from the host system. A request
423 serviced through mmap is never reused by any other request (at least
424 not directly; the system may just so happen to remap successive
425 requests to the same locations). Segregating space in this way has
426 the benefits that: Mmapped space can always be individually released
427 back to the system, which helps keep the system level memory demands
428 of a long-lived program low. Also, mapped memory doesn't become
429 `locked' between other chunks, as can happen with normally allocated
430 chunks, which means that even trimming via malloc_trim would not
431 release them. However, it has the disadvantage that the space
432 cannot be reclaimed, consolidated, and then used to service later
433 requests, as happens with normal chunks. The advantages of mmap
434 nearly always outweigh disadvantages for "large" chunks, but the
435 value of "large" may vary across systems. The default is an
436 empirically derived value that works well in most systems. You can
437 disable mmap by setting to MAX_SIZE_T.
447 #define WIN32_LEAN_AND_MEAN
450 #define HAVE_MORECORE 0
451 #define LACKS_UNISTD_H
452 #define LACKS_SYS_PARAM_H
453 #define LACKS_SYS_MMAN_H
454 #define LACKS_STRING_H
455 #define LACKS_STRINGS_H
456 #define LACKS_SYS_TYPES_H
457 #define LACKS_ERRNO_H
458 #define MALLOC_FAILURE_ACTION
459 #define MMAP_CLEARS 0 /* WINCE and some others apparently don't clear */
466 #define HAVE_MORECORE 0
467 #define LACKS_SYS_MMAN_H
470 #if defined(DARWIN) || defined(_DARWIN)
471 /* Mac OSX docs advise not to use sbrk; it seems better to use mmap */
472 #ifndef HAVE_MORECORE
473 #define HAVE_MORECORE 0
475 #endif /* HAVE_MORECORE */
478 #ifndef LACKS_SYS_TYPES_H
479 #include <sys/types.h> /* For size_t */
480 #endif /* LACKS_SYS_TYPES_H */
482 /* The maximum possible size_t value has all bits set */
483 #define MAX_SIZE_T (~(size_t)0)
486 #define ONLY_MSPACES 0
487 #endif /* ONLY_MSPACES */
491 #else /* ONLY_MSPACES */
493 #endif /* ONLY_MSPACES */
495 #ifndef MALLOC_ALIGNMENT
496 #define MALLOC_ALIGNMENT ((size_t)8U)
497 #endif /* MALLOC_ALIGNMENT */
502 #define ABORT abort()
504 #ifndef ABORT_ON_ASSERT_FAILURE
505 #define ABORT_ON_ASSERT_FAILURE 1
506 #endif /* ABORT_ON_ASSERT_FAILURE */
507 #ifndef PROCEED_ON_ERROR
508 #define PROCEED_ON_ERROR 0
509 #endif /* PROCEED_ON_ERROR */
512 #endif /* USE_LOCKS */
515 #endif /* INSECURE */
518 #endif /* HAVE_MMAP */
520 #define MMAP_CLEARS 1
521 #endif /* MMAP_CLEARS */
524 #define HAVE_MREMAP 1
526 #define HAVE_MREMAP 0
528 #endif /* HAVE_MREMAP */
529 #ifndef MALLOC_FAILURE_ACTION
530 #define MALLOC_FAILURE_ACTION errno = ENOMEM;
531 #endif /* MALLOC_FAILURE_ACTION */
532 #ifndef HAVE_MORECORE
534 #define HAVE_MORECORE 0
535 #else /* ONLY_MSPACES */
536 #define HAVE_MORECORE 1
537 #endif /* ONLY_MSPACES */
538 #endif /* HAVE_MORECORE */
540 #define MORECORE_CONTIGUOUS 0
541 #else /* !HAVE_MORECORE */
543 #define MORECORE sbrk
544 #endif /* MORECORE */
545 #ifndef MORECORE_CONTIGUOUS
546 #define MORECORE_CONTIGUOUS 1
547 #endif /* MORECORE_CONTIGUOUS */
548 #endif /* HAVE_MORECORE */
549 #ifndef DEFAULT_GRANULARITY
550 #if MORECORE_CONTIGUOUS
551 #define DEFAULT_GRANULARITY (0) /* 0 means to compute in init_mparams */
552 #else /* MORECORE_CONTIGUOUS */
553 #define DEFAULT_GRANULARITY ((size_t)64U * (size_t)1024U)
554 #endif /* MORECORE_CONTIGUOUS */
555 #endif /* DEFAULT_GRANULARITY */
556 #ifndef DEFAULT_TRIM_THRESHOLD
557 #ifndef MORECORE_CANNOT_TRIM
558 #define DEFAULT_TRIM_THRESHOLD ((size_t)2U * (size_t)1024U * (size_t)1024U)
559 #else /* MORECORE_CANNOT_TRIM */
560 #define DEFAULT_TRIM_THRESHOLD MAX_SIZE_T
561 #endif /* MORECORE_CANNOT_TRIM */
562 #endif /* DEFAULT_TRIM_THRESHOLD */
563 #ifndef DEFAULT_MMAP_THRESHOLD
565 #define DEFAULT_MMAP_THRESHOLD ((size_t)256U * (size_t)1024U)
566 #else /* HAVE_MMAP */
567 #define DEFAULT_MMAP_THRESHOLD MAX_SIZE_T
568 #endif /* HAVE_MMAP */
569 #endif /* DEFAULT_MMAP_THRESHOLD */
570 #ifndef USE_BUILTIN_FFS
571 #define USE_BUILTIN_FFS 0
572 #endif /* USE_BUILTIN_FFS */
573 #ifndef USE_DEV_RANDOM
574 #define USE_DEV_RANDOM 0
575 #endif /* USE_DEV_RANDOM */
577 #define NO_MALLINFO 0
578 #endif /* NO_MALLINFO */
579 #ifndef MALLINFO_FIELD_TYPE
580 #define MALLINFO_FIELD_TYPE size_t
581 #endif /* MALLINFO_FIELD_TYPE */
584 mallopt tuning options. SVID/XPG defines four standard parameter
585 numbers for mallopt, normally defined in malloc.h. None of these
586 are used in this malloc, so setting them has no effect. But this
587 malloc does support the following options.
590 #define M_TRIM_THRESHOLD (-1)
591 #define M_GRANULARITY (-2)
592 #define M_MMAP_THRESHOLD (-3)
594 /* ------------------------ Mallinfo declarations ------------------------ */
598 This version of malloc supports the standard SVID/XPG mallinfo
599 routine that returns a struct containing usage properties and
600 statistics. It should work on any system that has a
601 /usr/include/malloc.h defining struct mallinfo. The main
602 declaration needed is the mallinfo struct that is returned (by-copy)
603 by mallinfo(). The malloinfo struct contains a bunch of fields that
604 are not even meaningful in this version of malloc. These fields are
605 are instead filled by mallinfo() with other numbers that might be of
608 HAVE_USR_INCLUDE_MALLOC_H should be set if you have a
609 /usr/include/malloc.h file that includes a declaration of struct
610 mallinfo. If so, it is included; else a compliant version is
611 declared below. These must be precisely the same for mallinfo() to
612 work. The original SVID version of this struct, defined on most
613 systems with mallinfo, declares all fields as ints. But some others
614 define as unsigned long. If your system defines the fields using a
615 type of different width than listed here, you MUST #include your
616 system version and #define HAVE_USR_INCLUDE_MALLOC_H.
619 /* #define HAVE_USR_INCLUDE_MALLOC_H */
621 #ifdef HAVE_USR_INCLUDE_MALLOC_H
622 #include "/usr/include/malloc.h"
623 #else /* HAVE_USR_INCLUDE_MALLOC_H */
626 MALLINFO_FIELD_TYPE arena; /* non-mmapped space allocated from system */
627 MALLINFO_FIELD_TYPE ordblks; /* number of free chunks */
628 MALLINFO_FIELD_TYPE smblks; /* always 0 */
629 MALLINFO_FIELD_TYPE hblks; /* always 0 */
630 MALLINFO_FIELD_TYPE hblkhd; /* space in mmapped regions */
631 MALLINFO_FIELD_TYPE usmblks; /* maximum total allocated space */
632 MALLINFO_FIELD_TYPE fsmblks; /* always 0 */
633 MALLINFO_FIELD_TYPE uordblks; /* total allocated space */
634 MALLINFO_FIELD_TYPE fordblks; /* total free space */
635 MALLINFO_FIELD_TYPE keepcost; /* releasable (via malloc_trim) space */
638 #endif /* HAVE_USR_INCLUDE_MALLOC_H */
639 #endif /* NO_MALLINFO */
643 #endif /* __cplusplus */
647 /* ------------------- Declarations of public routines ------------------- */
649 #ifndef USE_DL_PREFIX
650 #define dlcalloc calloc
652 #define dlmalloc malloc
653 #define dlmemalign memalign
654 #define dlrealloc realloc
655 #define dlvalloc valloc
656 #define dlpvalloc pvalloc
657 #define dlmallinfo mallinfo
658 #define dlmallopt mallopt
659 #define dlmalloc_trim malloc_trim
660 #define dlmalloc_stats malloc_stats
661 #define dlmalloc_usable_size malloc_usable_size
662 #define dlmalloc_footprint malloc_footprint
663 #define dlmalloc_max_footprint malloc_max_footprint
664 #define dlindependent_calloc independent_calloc
665 #define dlindependent_comalloc independent_comalloc
666 #endif /* USE_DL_PREFIX */
671 Returns a pointer to a newly allocated chunk of at least n bytes, or
672 null if no space is available, in which case errno is set to ENOMEM
675 If n is zero, malloc returns a minimum-sized chunk. (The minimum
676 size is 16 bytes on most 32bit systems, and 32 bytes on 64bit
677 systems.) Note that size_t is an unsigned type, so calls with
678 arguments that would be negative if signed are interpreted as
679 requests for huge amounts of space, which will often fail. The
680 maximum supported value of n differs across systems, but is in all
681 cases less than the maximum representable value of a size_t.
683 void* dlmalloc(size_t);
687 Releases the chunk of memory pointed to by p, that had been previously
688 allocated using malloc or a related routine such as realloc.
689 It has no effect if p is null. If p was not malloced or already
690 freed, free(p) will by default cause the current program to abort.
695 calloc(size_t n_elements, size_t element_size);
696 Returns a pointer to n_elements * element_size bytes, with all locations
699 void* dlcalloc(size_t, size_t);
702 realloc(void* p, size_t n)
703 Returns a pointer to a chunk of size n that contains the same data
704 as does chunk p up to the minimum of (n, p's size) bytes, or null
705 if no space is available.
707 The returned pointer may or may not be the same as p. The algorithm
708 prefers extending p in most cases when possible, otherwise it
709 employs the equivalent of a malloc-copy-free sequence.
711 If p is null, realloc is equivalent to malloc.
713 If space is not available, realloc returns null, errno is set (if on
714 ANSI) and p is NOT freed.
716 if n is for fewer bytes than already held by p, the newly unused
717 space is lopped off and freed if possible. realloc with a size
718 argument of zero (re)allocates a minimum-sized chunk.
720 The old unix realloc convention of allowing the last-free'd chunk
721 to be used as an argument to realloc is not supported.
724 void* dlrealloc(void*, size_t);
727 memalign(size_t alignment, size_t n);
728 Returns a pointer to a newly allocated chunk of n bytes, aligned
729 in accord with the alignment argument.
731 The alignment argument should be a power of two. If the argument is
732 not a power of two, the nearest greater power is used.
733 8-byte alignment is guaranteed by normal malloc calls, so don't
734 bother calling memalign with an argument of 8 or less.
736 Overreliance on memalign is a sure way to fragment space.
738 void* dlmemalign(size_t, size_t);
742 Equivalent to memalign(pagesize, n), where pagesize is the page
743 size of the system. If the pagesize is unknown, 4096 is used.
745 void* dlvalloc(size_t);
748 mallopt(int parameter_number, int parameter_value)
749 Sets tunable parameters The format is to provide a
750 (parameter-number, parameter-value) pair. mallopt then sets the
751 corresponding parameter to the argument value if it can (i.e., so
752 long as the value is meaningful), and returns 1 if successful else
753 0. SVID/XPG/ANSI defines four standard param numbers for mallopt,
754 normally defined in malloc.h. None of these are use in this malloc,
755 so setting them has no effect. But this malloc also supports other
756 options in mallopt. See below for details. Briefly, supported
757 parameters are as follows (listed defaults are for "typical"
760 Symbol param # default allowed param values
761 M_TRIM_THRESHOLD -1 2*1024*1024 any (MAX_SIZE_T disables)
762 M_GRANULARITY -2 page size any power of 2 >= page size
763 M_MMAP_THRESHOLD -3 256*1024 any (or 0 if no MMAP support)
765 int dlmallopt(int, int);
769 Returns the number of bytes obtained from the system. The total
770 number of bytes allocated by malloc, realloc etc., is less than this
771 value. Unlike mallinfo, this function returns only a precomputed
772 result, so can be called frequently to monitor memory consumption.
773 Even if locks are otherwise defined, this function does not use them,
774 so results might not be up to date.
776 size_t dlmalloc_footprint(void);
779 malloc_max_footprint();
780 Returns the maximum number of bytes obtained from the system. This
781 value will be greater than current footprint if deallocated space
782 has been reclaimed by the system. The peak number of bytes allocated
783 by malloc, realloc etc., is less than this value. Unlike mallinfo,
784 this function returns only a precomputed result, so can be called
785 frequently to monitor memory consumption. Even if locks are
786 otherwise defined, this function does not use them, so results might
789 size_t dlmalloc_max_footprint(void);
794 Returns (by copy) a struct containing various summary statistics:
796 arena: current total non-mmapped bytes allocated from system
797 ordblks: the number of free chunks
799 hblks: current number of mmapped regions
800 hblkhd: total bytes held in mmapped regions
801 usmblks: the maximum total allocated space. This will be greater
802 than current total if trimming has occurred.
804 uordblks: current total allocated space (normal or mmapped)
805 fordblks: total free space
806 keepcost: the maximum number of bytes that could ideally be released
807 back to system via malloc_trim. ("ideally" means that
808 it ignores page restrictions etc.)
810 Because these fields are ints, but internal bookkeeping may
811 be kept as longs, the reported values may wrap around zero and
814 struct mallinfo dlmallinfo(void);
815 #endif /* NO_MALLINFO */
818 independent_calloc(size_t n_elements, size_t element_size, void* chunks[]);
820 independent_calloc is similar to calloc, but instead of returning a
821 single cleared space, it returns an array of pointers to n_elements
822 independent elements that can hold contents of size elem_size, each
823 of which starts out cleared, and can be independently freed,
824 realloc'ed etc. The elements are guaranteed to be adjacently
825 allocated (this is not guaranteed to occur with multiple callocs or
826 mallocs), which may also improve cache locality in some
829 The "chunks" argument is optional (i.e., may be null, which is
830 probably the most typical usage). If it is null, the returned array
831 is itself dynamically allocated and should also be freed when it is
832 no longer needed. Otherwise, the chunks array must be of at least
833 n_elements in length. It is filled in with the pointers to the
836 In either case, independent_calloc returns this pointer array, or
837 null if the allocation failed. If n_elements is zero and "chunks"
838 is null, it returns a chunk representing an array with zero elements
839 (which should be freed if not wanted).
841 Each element must be individually freed when it is no longer
842 needed. If you'd like to instead be able to free all at once, you
843 should instead use regular calloc and assign pointers into this
844 space to represent elements. (In this case though, you cannot
845 independently free elements.)
847 independent_calloc simplifies and speeds up implementations of many
848 kinds of pools. It may also be useful when constructing large data
849 structures that initially have a fixed number of fixed-sized nodes,
850 but the number is not known at compile time, and some of the nodes
851 may later need to be freed. For example:
853 struct Node { int item; struct Node* next; };
855 struct Node* build_list() {
857 int n = read_number_of_nodes_needed();
858 if (n <= 0) return 0;
859 pool = (struct Node**)(independent_calloc(n, sizeof(struct Node), 0);
860 if (pool == 0) die();
861 // organize into a linked list...
862 struct Node* first = pool[0];
863 for (i = 0; i < n-1; ++i)
864 pool[i]->next = pool[i+1];
865 free(pool); // Can now free the array (or not, if it is needed later)
869 void** dlindependent_calloc(size_t, size_t, void**);
872 independent_comalloc(size_t n_elements, size_t sizes[], void* chunks[]);
874 independent_comalloc allocates, all at once, a set of n_elements
875 chunks with sizes indicated in the "sizes" array. It returns
876 an array of pointers to these elements, each of which can be
877 independently freed, realloc'ed etc. The elements are guaranteed to
878 be adjacently allocated (this is not guaranteed to occur with
879 multiple callocs or mallocs), which may also improve cache locality
880 in some applications.
882 The "chunks" argument is optional (i.e., may be null). If it is null
883 the returned array is itself dynamically allocated and should also
884 be freed when it is no longer needed. Otherwise, the chunks array
885 must be of at least n_elements in length. It is filled in with the
886 pointers to the chunks.
888 In either case, independent_comalloc returns this pointer array, or
889 null if the allocation failed. If n_elements is zero and chunks is
890 null, it returns a chunk representing an array with zero elements
891 (which should be freed if not wanted).
893 Each element must be individually freed when it is no longer
894 needed. If you'd like to instead be able to free all at once, you
895 should instead use a single regular malloc, and assign pointers at
896 particular offsets in the aggregate space. (In this case though, you
897 cannot independently free elements.)
899 independent_comallac differs from independent_calloc in that each
900 element may have a different size, and also that it does not
901 automatically clear elements.
903 independent_comalloc can be used to speed up allocation in cases
904 where several structs or objects must always be allocated at the
905 same time. For example:
910 void send_message(char* msg) {
911 int msglen = strlen(msg);
912 size_t sizes[3] = { sizeof(struct Head), msglen, sizeof(struct Foot) };
914 if (independent_comalloc(3, sizes, chunks) == 0)
916 struct Head* head = (struct Head*)(chunks[0]);
917 char* body = (char*)(chunks[1]);
918 struct Foot* foot = (struct Foot*)(chunks[2]);
922 In general though, independent_comalloc is worth using only for
923 larger values of n_elements. For small values, you probably won't
924 detect enough difference from series of malloc calls to bother.
926 Overuse of independent_comalloc can increase overall memory usage,
927 since it cannot reuse existing noncontiguous small chunks that
928 might be available for some of the elements.
930 void** dlindependent_comalloc(size_t, size_t*, void**);
935 Equivalent to valloc(minimum-page-that-holds(n)), that is,
936 round up n to nearest pagesize.
938 void* dlpvalloc(size_t);
941 malloc_trim(size_t pad);
943 If possible, gives memory back to the system (via negative arguments
944 to sbrk) if there is unused memory at the `high' end of the malloc
945 pool or in unused MMAP segments. You can call this after freeing
946 large blocks of memory to potentially reduce the system-level memory
947 requirements of a program. However, it cannot guarantee to reduce
948 memory. Under some allocation patterns, some large free blocks of
949 memory will be locked between two used chunks, so they cannot be
950 given back to the system.
952 The `pad' argument to malloc_trim represents the amount of free
953 trailing space to leave untrimmed. If this argument is zero, only
954 the minimum amount of memory to maintain internal data structures
955 will be left. Non-zero arguments can be supplied to maintain enough
956 trailing space to service future expected allocations without having
957 to re-obtain memory from the system.
959 Malloc_trim returns 1 if it actually released any memory, else 0.
961 int dlmalloc_trim(size_t);
964 malloc_usable_size(void* p);
966 Returns the number of bytes you can actually use in
967 an allocated chunk, which may be more than you requested (although
968 often not) due to alignment and minimum size constraints.
969 You can use this many bytes without worrying about
970 overwriting other allocated objects. This is not a particularly great
971 programming practice. malloc_usable_size can be more useful in
972 debugging and assertions, for example:
975 assert(malloc_usable_size(p) >= 256);
977 size_t dlmalloc_usable_size(void*);
981 Prints on stderr the amount of space obtained from the system (both
982 via sbrk and mmap), the maximum amount (which may be more than
983 current if malloc_trim and/or munmap got called), and the current
984 number of bytes allocated via malloc (or realloc, etc) but not yet
985 freed. Note that this is the number of bytes allocated, not the
986 number requested. It will be larger than the number requested
987 because of alignment and bookkeeping overhead. Because it includes
988 alignment wastage as being in use, this figure may be greater than
989 zero even when no user-level chunks are allocated.
991 The reported current and maximum system memory can be inaccurate if
992 a program makes other calls to system memory allocation functions
993 (normally sbrk) outside of malloc.
995 malloc_stats prints only the most commonly interesting statistics.
996 More information can be obtained by calling mallinfo.
998 void dlmalloc_stats(void);
1000 #endif /* ONLY_MSPACES */
1005 mspace is an opaque type representing an independent
1006 region of space that supports mspace_malloc, etc.
1008 typedef void* mspace;
1011 create_mspace creates and returns a new independent space with the
1012 given initial capacity, or, if 0, the default granularity size. It
1013 returns null if there is no system memory available to create the
1014 space. If argument locked is non-zero, the space uses a separate
1015 lock to control access. The capacity of the space will grow
1016 dynamically as needed to service mspace_malloc requests. You can
1017 control the sizes of incremental increases of this space by
1018 compiling with a different DEFAULT_GRANULARITY or dynamically
1019 setting with mallopt(M_GRANULARITY, value).
1021 mspace create_mspace(size_t capacity, int locked);
1024 destroy_mspace destroys the given space, and attempts to return all
1025 of its memory back to the system, returning the total number of
1026 bytes freed. After destruction, the results of access to all memory
1027 used by the space become undefined.
1029 size_t destroy_mspace(mspace msp);
1032 create_mspace_with_base uses the memory supplied as the initial base
1033 of a new mspace. Part (less than 128*sizeof(size_t) bytes) of this
1034 space is used for bookkeeping, so the capacity must be at least this
1035 large. (Otherwise 0 is returned.) When this initial space is
1036 exhausted, additional memory will be obtained from the system.
1037 Destroying this space will deallocate all additionally allocated
1038 space (if possible) but not the initial base.
1040 mspace create_mspace_with_base(void* base, size_t capacity, int locked);
1043 mspace_malloc behaves as malloc, but operates within
1046 void* mspace_malloc(mspace msp, size_t bytes);
1049 mspace_free behaves as free, but operates within
1052 If compiled with FOOTERS==1, mspace_free is not actually needed.
1053 free may be called instead of mspace_free because freed chunks from
1054 any space are handled by their originating spaces.
1056 void mspace_free(mspace msp, void* mem);
1059 mspace_realloc behaves as realloc, but operates within
1062 If compiled with FOOTERS==1, mspace_realloc is not actually
1063 needed. realloc may be called instead of mspace_realloc because
1064 realloced chunks from any space are handled by their originating
1067 void* mspace_realloc(mspace msp, void* mem, size_t newsize);
1070 mspace_calloc behaves as calloc, but operates within
1073 void* mspace_calloc(mspace msp, size_t n_elements, size_t elem_size);
1076 mspace_memalign behaves as memalign, but operates within
1079 void* mspace_memalign(mspace msp, size_t alignment, size_t bytes);
1082 mspace_independent_calloc behaves as independent_calloc, but
1083 operates within the given space.
1085 void** mspace_independent_calloc(mspace msp, size_t n_elements,
1086 size_t elem_size, void* chunks[]);
1089 mspace_independent_comalloc behaves as independent_comalloc, but
1090 operates within the given space.
1092 void** mspace_independent_comalloc(mspace msp, size_t n_elements,
1093 size_t sizes[], void* chunks[]);
1096 mspace_footprint() returns the number of bytes obtained from the
1097 system for this space.
1099 size_t mspace_footprint(mspace msp);
1102 mspace_max_footprint() returns the peak number of bytes obtained from the
1103 system for this space.
1105 size_t mspace_max_footprint(mspace msp);
1110 mspace_mallinfo behaves as mallinfo, but reports properties of
1113 struct mallinfo mspace_mallinfo(mspace msp);
1114 #endif /* NO_MALLINFO */
1117 mspace_malloc_stats behaves as malloc_stats, but reports
1118 properties of the given space.
1120 void mspace_malloc_stats(mspace msp);
1123 mspace_trim behaves as malloc_trim, but
1124 operates within the given space.
1126 int mspace_trim(mspace msp, size_t pad);
1129 An alias for mallopt.
1131 int mspace_mallopt(int, int);
1133 #endif /* MSPACES */
1136 }; /* end of extern "C" */
1137 #endif /* __cplusplus */
1140 ========================================================================
1141 To make a fully customizable malloc.h header file, cut everything
1142 above this line, put into file malloc.h, edit to suit, and #include it
1143 on the next line, as well as in programs that use this malloc.
1144 ========================================================================
1147 /* #include "malloc.h" */
1149 /*------------------------------ internal #includes ---------------------- */
1152 #pragma warning( disable : 4146 ) /* no "unsigned" warnings */
1153 #endif /* _MSC_VER */
1155 #include <stdio.h> /* for printing in malloc_stats */
1157 #ifndef LACKS_ERRNO_H
1158 #include <errno.h> /* for MALLOC_FAILURE_ACTION */
1159 #endif /* LACKS_ERRNO_H */
1161 #include <time.h> /* for magic initialization */
1162 #endif /* FOOTERS */
1163 #ifndef LACKS_STDLIB_H
1164 #include <stdlib.h> /* for abort() */
1165 #endif /* LACKS_STDLIB_H */
1167 #if ABORT_ON_ASSERT_FAILURE
1168 #define assert(x) if(!(x)) ABORT
1169 #else /* ABORT_ON_ASSERT_FAILURE */
1171 #endif /* ABORT_ON_ASSERT_FAILURE */
1175 #ifndef LACKS_STRING_H
1176 #include <string.h> /* for memset etc */
1177 #endif /* LACKS_STRING_H */
1179 #ifndef LACKS_STRINGS_H
1180 #include <strings.h> /* for ffs */
1181 #endif /* LACKS_STRINGS_H */
1182 #endif /* USE_BUILTIN_FFS */
1184 #ifndef LACKS_SYS_MMAN_H
1185 #include <sys/mman.h> /* for mmap */
1186 #endif /* LACKS_SYS_MMAN_H */
1187 #ifndef LACKS_FCNTL_H
1189 #endif /* LACKS_FCNTL_H */
1190 #endif /* HAVE_MMAP */
1192 #ifndef LACKS_UNISTD_H
1193 #include <unistd.h> /* for sbrk */
1194 #else /* LACKS_UNISTD_H */
1195 #if !defined(__FreeBSD__) && !defined(__OpenBSD__) && !defined(__NetBSD__)
1196 extern void* sbrk(ptrdiff_t);
1197 #endif /* FreeBSD etc */
1198 #endif /* LACKS_UNISTD_H */
1199 #endif /* HAVE_MMAP */
1202 #ifndef malloc_getpagesize
1203 # ifdef _SC_PAGESIZE /* some SVR4 systems omit an underscore */
1204 # ifndef _SC_PAGE_SIZE
1205 # define _SC_PAGE_SIZE _SC_PAGESIZE
1208 # ifdef _SC_PAGE_SIZE
1209 # define malloc_getpagesize sysconf(_SC_PAGE_SIZE)
1211 # if defined(BSD) || defined(DGUX) || defined(HAVE_GETPAGESIZE)
1212 extern size_t getpagesize();
1213 # define malloc_getpagesize getpagesize()
1215 # ifdef WIN32 /* use supplied emulation of getpagesize */
1216 # define malloc_getpagesize getpagesize()
1218 # ifndef LACKS_SYS_PARAM_H
1219 # include <sys/param.h>
1221 # ifdef EXEC_PAGESIZE
1222 # define malloc_getpagesize EXEC_PAGESIZE
1226 # define malloc_getpagesize NBPG
1228 # define malloc_getpagesize (NBPG * CLSIZE)
1232 # define malloc_getpagesize NBPC
1235 # define malloc_getpagesize PAGESIZE
1236 # else /* just guess */
1237 # define malloc_getpagesize ((size_t)4096U)
1248 /* ------------------- size_t and alignment properties -------------------- */
1250 /* The byte and bit size of a size_t */
1251 #define SIZE_T_SIZE (sizeof(size_t))
1252 #define SIZE_T_BITSIZE (sizeof(size_t) << 3)
1254 /* Some constants coerced to size_t */
1255 /* Annoying but necessary to avoid errors on some plaftorms */
1256 #define SIZE_T_ZERO ((size_t)0)
1257 #define SIZE_T_ONE ((size_t)1)
1258 #define SIZE_T_TWO ((size_t)2)
1259 #define TWO_SIZE_T_SIZES (SIZE_T_SIZE<<1)
1260 #define FOUR_SIZE_T_SIZES (SIZE_T_SIZE<<2)
1261 #define SIX_SIZE_T_SIZES (FOUR_SIZE_T_SIZES+TWO_SIZE_T_SIZES)
1262 #define HALF_MAX_SIZE_T (MAX_SIZE_T / 2U)
1264 /* The bit mask value corresponding to MALLOC_ALIGNMENT */
1265 #define CHUNK_ALIGN_MASK (MALLOC_ALIGNMENT - SIZE_T_ONE)
1267 /* True if address a has acceptable alignment */
1268 #define is_aligned(A) (((size_t)((A)) & (CHUNK_ALIGN_MASK)) == 0)
1270 /* the number of bytes to offset an address to align it */
1271 #define align_offset(A)\
1272 ((((size_t)(A) & CHUNK_ALIGN_MASK) == 0)? 0 :\
1273 ((MALLOC_ALIGNMENT - ((size_t)(A) & CHUNK_ALIGN_MASK)) & CHUNK_ALIGN_MASK))
1275 /* -------------------------- MMAP preliminaries ------------------------- */
1278 If HAVE_MORECORE or HAVE_MMAP are false, we just define calls and
1279 checks to fail so compiler optimizer can delete code rather than
1280 using so many "#if"s.
1284 /* MORECORE and MMAP must return MFAIL on failure */
1285 #define MFAIL ((void*)(MAX_SIZE_T))
1286 #define CMFAIL ((char*)(MFAIL)) /* defined for convenience */
1289 #define IS_MMAPPED_BIT (SIZE_T_ZERO)
1290 #define USE_MMAP_BIT (SIZE_T_ZERO)
1291 #define CALL_MMAP(s) MFAIL
1292 #define CALL_MUNMAP(a, s) (-1)
1293 #define DIRECT_MMAP(s) MFAIL
1295 #else /* HAVE_MMAP */
1296 #define IS_MMAPPED_BIT (SIZE_T_ONE)
1297 #define USE_MMAP_BIT (SIZE_T_ONE)
1299 #if !defined(WIN32) && !defined (__OS2__)
1300 #define CALL_MUNMAP(a, s) munmap((a), (s))
1301 #define MMAP_PROT (PROT_READ|PROT_WRITE)
1302 #if !defined(MAP_ANONYMOUS) && defined(MAP_ANON)
1303 #define MAP_ANONYMOUS MAP_ANON
1304 #endif /* MAP_ANON */
1305 #ifdef MAP_ANONYMOUS
1306 #define MMAP_FLAGS (MAP_PRIVATE|MAP_ANONYMOUS)
1307 #define CALL_MMAP(s) mmap(0, (s), MMAP_PROT, MMAP_FLAGS, -1, 0)
1308 #else /* MAP_ANONYMOUS */
1310 Nearly all versions of mmap support MAP_ANONYMOUS, so the following
1311 is unlikely to be needed, but is supplied just in case.
1313 #define MMAP_FLAGS (MAP_PRIVATE)
1314 static int dev_zero_fd = -1; /* Cached file descriptor for /dev/zero. */
1315 #define CALL_MMAP(s) ((dev_zero_fd < 0) ? \
1316 (dev_zero_fd = open("/dev/zero", O_RDWR), \
1317 mmap(0, (s), MMAP_PROT, MMAP_FLAGS, dev_zero_fd, 0)) : \
1318 mmap(0, (s), MMAP_PROT, MMAP_FLAGS, dev_zero_fd, 0))
1319 #endif /* MAP_ANONYMOUS */
1321 #define DIRECT_MMAP(s) CALL_MMAP(s)
1323 #elif defined(__OS2__)
1325 /* OS/2 MMAP via DosAllocMem */
1326 static void* os2mmap(size_t size) {
1328 if (DosAllocMem(&ptr, size, OBJ_ANY|PAG_COMMIT|PAG_READ|PAG_WRITE) &&
1329 DosAllocMem(&ptr, size, PAG_COMMIT|PAG_READ|PAG_WRITE))
1334 #define os2direct_mmap(n) os2mmap(n)
1336 /* This function supports releasing coalesed segments */
1337 static int os2munmap(void* ptr, size_t size) {
1339 ULONG ulSize = size;
1341 if (DosQueryMem(ptr, &ulSize, &ulFlags) != 0)
1343 if ((ulFlags & PAG_BASE) == 0 ||(ulFlags & PAG_COMMIT) == 0 ||
1346 if (DosFreeMem(ptr) != 0)
1348 ptr = ( void * ) ( ( char * ) ptr + ulSize );
1354 #define CALL_MMAP(s) os2mmap(s)
1355 #define CALL_MUNMAP(a, s) os2munmap((a), (s))
1356 #define DIRECT_MMAP(s) os2direct_mmap(s)
1360 /* Win32 MMAP via VirtualAlloc */
1361 static void* win32mmap(size_t size) {
1362 void* ptr = VirtualAlloc(0, size, MEM_RESERVE|MEM_COMMIT, PAGE_EXECUTE_READWRITE);
1363 return (ptr != 0)? ptr: MFAIL;
1366 /* For direct MMAP, use MEM_TOP_DOWN to minimize interference */
1367 static void* win32direct_mmap(size_t size) {
1368 void* ptr = VirtualAlloc(0, size, MEM_RESERVE|MEM_COMMIT|MEM_TOP_DOWN,
1369 PAGE_EXECUTE_READWRITE);
1370 return (ptr != 0)? ptr: MFAIL;
1373 /* This function supports releasing coalesed segments */
1374 static int win32munmap(void* ptr, size_t size) {
1375 MEMORY_BASIC_INFORMATION minfo;
1378 if (VirtualQuery(cptr, &minfo, sizeof(minfo)) == 0)
1380 if (minfo.BaseAddress != cptr || minfo.AllocationBase != cptr ||
1381 minfo.State != MEM_COMMIT || minfo.RegionSize > size)
1383 if (VirtualFree(cptr, 0, MEM_RELEASE) == 0)
1385 cptr += minfo.RegionSize;
1386 size -= minfo.RegionSize;
1391 #define CALL_MMAP(s) win32mmap(s)
1392 #define CALL_MUNMAP(a, s) win32munmap((a), (s))
1393 #define DIRECT_MMAP(s) win32direct_mmap(s)
1395 #endif /* HAVE_MMAP */
1397 #if HAVE_MMAP && HAVE_MREMAP
1398 #define CALL_MREMAP(addr, osz, nsz, mv) mremap((addr), (osz), (nsz), (mv))
1399 #else /* HAVE_MMAP && HAVE_MREMAP */
1400 #define CALL_MREMAP(addr, osz, nsz, mv) MFAIL
1401 #endif /* HAVE_MMAP && HAVE_MREMAP */
1404 #define CALL_MORECORE(S) MORECORE(S)
1405 #else /* HAVE_MORECORE */
1406 #define CALL_MORECORE(S) MFAIL
1407 #endif /* HAVE_MORECORE */
1409 /* mstate bit set if continguous morecore disabled or failed */
1410 #define USE_NONCONTIGUOUS_BIT (4U)
1412 /* segment bit set in create_mspace_with_base */
1413 #define EXTERN_BIT (8U)
1416 /* --------------------------- Lock preliminaries ------------------------ */
1421 When locks are defined, there are up to two global locks:
1423 * If HAVE_MORECORE, morecore_mutex protects sequences of calls to
1424 MORECORE. In many cases sys_alloc requires two calls, that should
1425 not be interleaved with calls by other threads. This does not
1426 protect against direct calls to MORECORE by other threads not
1427 using this lock, so there is still code to cope the best we can on
1430 * magic_init_mutex ensures that mparams.magic and other
1431 unique mparams values are initialized only once.
1434 #if !defined(WIN32) && !defined(__OS2__)
1435 /* By default use posix locks */
1436 #include <pthread.h>
1437 #define MLOCK_T pthread_mutex_t
1438 #define INITIAL_LOCK(l) pthread_mutex_init(l, NULL)
1439 #define ACQUIRE_LOCK(l) pthread_mutex_lock(l)
1440 #define RELEASE_LOCK(l) pthread_mutex_unlock(l)
1443 static MLOCK_T morecore_mutex = PTHREAD_MUTEX_INITIALIZER;
1444 #endif /* HAVE_MORECORE */
1446 static MLOCK_T magic_init_mutex = PTHREAD_MUTEX_INITIALIZER;
1448 #elif defined(__OS2__)
1449 #define MLOCK_T HMTX
1450 #define INITIAL_LOCK(l) DosCreateMutexSem(0, l, 0, FALSE)
1451 #define ACQUIRE_LOCK(l) DosRequestMutexSem(*l, SEM_INDEFINITE_WAIT)
1452 #define RELEASE_LOCK(l) DosReleaseMutexSem(*l)
1454 static MLOCK_T morecore_mutex;
1455 #endif /* HAVE_MORECORE */
1456 static MLOCK_T magic_init_mutex;
1460 Because lock-protected regions have bounded times, and there
1461 are no recursive lock calls, we can use simple spinlocks.
1464 #define MLOCK_T long
1465 static int win32_acquire_lock (MLOCK_T *sl) {
1467 #ifdef InterlockedCompareExchangePointer
1468 if (!InterlockedCompareExchange(sl, 1, 0))
1470 #else /* Use older void* version */
1471 if (!InterlockedCompareExchange((void**)sl, (void*)1, (void*)0))
1473 #endif /* InterlockedCompareExchangePointer */
1478 static void win32_release_lock (MLOCK_T *sl) {
1479 InterlockedExchange (sl, 0);
1482 #define INITIAL_LOCK(l) *(l)=0
1483 #define ACQUIRE_LOCK(l) win32_acquire_lock(l)
1484 #define RELEASE_LOCK(l) win32_release_lock(l)
1486 static MLOCK_T morecore_mutex;
1487 #endif /* HAVE_MORECORE */
1488 static MLOCK_T magic_init_mutex;
1491 #define USE_LOCK_BIT (2U)
1492 #else /* USE_LOCKS */
1493 #define USE_LOCK_BIT (0U)
1494 #define INITIAL_LOCK(l)
1495 #endif /* USE_LOCKS */
1497 #if USE_LOCKS && HAVE_MORECORE
1498 #define ACQUIRE_MORECORE_LOCK() ACQUIRE_LOCK(&morecore_mutex);
1499 #define RELEASE_MORECORE_LOCK() RELEASE_LOCK(&morecore_mutex);
1500 #else /* USE_LOCKS && HAVE_MORECORE */
1501 #define ACQUIRE_MORECORE_LOCK()
1502 #define RELEASE_MORECORE_LOCK()
1503 #endif /* USE_LOCKS && HAVE_MORECORE */
1506 #define ACQUIRE_MAGIC_INIT_LOCK() ACQUIRE_LOCK(&magic_init_mutex);
1507 #define RELEASE_MAGIC_INIT_LOCK() RELEASE_LOCK(&magic_init_mutex);
1508 #else /* USE_LOCKS */
1509 #define ACQUIRE_MAGIC_INIT_LOCK()
1510 #define RELEASE_MAGIC_INIT_LOCK()
1511 #endif /* USE_LOCKS */
1514 /* ----------------------- Chunk representations ------------------------ */
1517 (The following includes lightly edited explanations by Colin Plumb.)
1519 The malloc_chunk declaration below is misleading (but accurate and
1520 necessary). It declares a "view" into memory allowing access to
1521 necessary fields at known offsets from a given base.
1523 Chunks of memory are maintained using a `boundary tag' method as
1524 originally described by Knuth. (See the paper by Paul Wilson
1525 ftp://ftp.cs.utexas.edu/pub/garbage/allocsrv.ps for a survey of such
1526 techniques.) Sizes of free chunks are stored both in the front of
1527 each chunk and at the end. This makes consolidating fragmented
1528 chunks into bigger chunks fast. The head fields also hold bits
1529 representing whether chunks are free or in use.
1531 Here are some pictures to make it clearer. They are "exploded" to
1532 show that the state of a chunk can be thought of as extending from
1533 the high 31 bits of the head field of its header through the
1534 prev_foot and PINUSE_BIT bit of the following chunk header.
1536 A chunk that's in use looks like:
1538 chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1539 | Size of previous chunk (if P = 1) |
1540 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1541 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |P|
1542 | Size of this chunk 1| +-+
1543 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1549 +- size - sizeof(size_t) available payload bytes -+
1553 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1554 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |1|
1555 | Size of next chunk (may or may not be in use) | +-+
1556 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1558 And if it's free, it looks like this:
1561 | User payload (must be in use, or we would have merged!) |
1562 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1563 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |P|
1564 | Size of this chunk 0| +-+
1565 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1567 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1569 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1571 +- size - sizeof(struct chunk) unused bytes -+
1573 chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1574 | Size of this chunk |
1575 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1576 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |0|
1577 | Size of next chunk (must be in use, or we would have merged)| +-+
1578 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1582 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1585 Note that since we always merge adjacent free chunks, the chunks
1586 adjacent to a free chunk must be in use.
1588 Given a pointer to a chunk (which can be derived trivially from the
1589 payload pointer) we can, in O(1) time, find out whether the adjacent
1590 chunks are free, and if so, unlink them from the lists that they
1591 are on and merge them with the current chunk.
1593 Chunks always begin on even word boundaries, so the mem portion
1594 (which is returned to the user) is also on an even word boundary, and
1595 thus at least double-word aligned.
1597 The P (PINUSE_BIT) bit, stored in the unused low-order bit of the
1598 chunk size (which is always a multiple of two words), is an in-use
1599 bit for the *previous* chunk. If that bit is *clear*, then the
1600 word before the current chunk size contains the previous chunk
1601 size, and can be used to find the front of the previous chunk.
1602 The very first chunk allocated always has this bit set, preventing
1603 access to non-existent (or non-owned) memory. If pinuse is set for
1604 any given chunk, then you CANNOT determine the size of the
1605 previous chunk, and might even get a memory addressing fault when
1608 The C (CINUSE_BIT) bit, stored in the unused second-lowest bit of
1609 the chunk size redundantly records whether the current chunk is
1610 inuse. This redundancy enables usage checks within free and realloc,
1611 and reduces indirection when freeing and consolidating chunks.
1613 Each freshly allocated chunk must have both cinuse and pinuse set.
1614 That is, each allocated chunk borders either a previously allocated
1615 and still in-use chunk, or the base of its memory arena. This is
1616 ensured by making all allocations from the the `lowest' part of any
1617 found chunk. Further, no free chunk physically borders another one,
1618 so each free chunk is known to be preceded and followed by either
1619 inuse chunks or the ends of memory.
1621 Note that the `foot' of the current chunk is actually represented
1622 as the prev_foot of the NEXT chunk. This makes it easier to
1623 deal with alignments etc but can be very confusing when trying
1624 to extend or adapt this code.
1626 The exceptions to all this are
1628 1. The special chunk `top' is the top-most available chunk (i.e.,
1629 the one bordering the end of available memory). It is treated
1630 specially. Top is never included in any bin, is used only if
1631 no other chunk is available, and is released back to the
1632 system if it is very large (see M_TRIM_THRESHOLD). In effect,
1633 the top chunk is treated as larger (and thus less well
1634 fitting) than any other available chunk. The top chunk
1635 doesn't update its trailing size field since there is no next
1636 contiguous chunk that would have to index off it. However,
1637 space is still allocated for it (TOP_FOOT_SIZE) to enable
1638 separation or merging when space is extended.
1640 3. Chunks allocated via mmap, which have the lowest-order bit
1641 (IS_MMAPPED_BIT) set in their prev_foot fields, and do not set
1642 PINUSE_BIT in their head fields. Because they are allocated
1643 one-by-one, each must carry its own prev_foot field, which is
1644 also used to hold the offset this chunk has within its mmapped
1645 region, which is needed to preserve alignment. Each mmapped
1646 chunk is trailed by the first two fields of a fake next-chunk
1647 for sake of usage checks.
1651 struct malloc_chunk {
1652 size_t prev_foot; /* Size of previous chunk (if free). */
1653 size_t head; /* Size and inuse bits. */
1654 struct malloc_chunk* fd; /* double links -- used only if free. */
1655 struct malloc_chunk* bk;
1658 typedef struct malloc_chunk mchunk;
1659 typedef struct malloc_chunk* mchunkptr;
1660 typedef struct malloc_chunk* sbinptr; /* The type of bins of chunks */
1661 typedef unsigned int bindex_t; /* Described below */
1662 typedef unsigned int binmap_t; /* Described below */
1663 typedef unsigned int flag_t; /* The type of various bit flag sets */
1665 /* ------------------- Chunks sizes and alignments ----------------------- */
1667 #define MCHUNK_SIZE (sizeof(mchunk))
1670 #define CHUNK_OVERHEAD (TWO_SIZE_T_SIZES)
1672 #define CHUNK_OVERHEAD (SIZE_T_SIZE)
1673 #endif /* FOOTERS */
1675 /* MMapped chunks need a second word of overhead ... */
1676 #define MMAP_CHUNK_OVERHEAD (TWO_SIZE_T_SIZES)
1677 /* ... and additional padding for fake next-chunk at foot */
1678 #define MMAP_FOOT_PAD (FOUR_SIZE_T_SIZES)
1680 /* The smallest size we can malloc is an aligned minimal chunk */
1681 #define MIN_CHUNK_SIZE\
1682 ((MCHUNK_SIZE + CHUNK_ALIGN_MASK) & ~CHUNK_ALIGN_MASK)
1684 /* conversion from malloc headers to user pointers, and back */
1685 #define chunk2mem(p) ((void*)((char*)(p) + TWO_SIZE_T_SIZES))
1686 #define mem2chunk(mem) ((mchunkptr)((char*)(mem) - TWO_SIZE_T_SIZES))
1687 /* chunk associated with aligned address A */
1688 #define align_as_chunk(A) (mchunkptr)((A) + align_offset(chunk2mem(A)))
1690 /* Bounds on request (not chunk) sizes. */
1691 #define MAX_REQUEST ((-MIN_CHUNK_SIZE) << 2)
1692 #define MIN_REQUEST (MIN_CHUNK_SIZE - CHUNK_OVERHEAD - SIZE_T_ONE)
1694 /* pad request bytes into a usable size */
1695 #define pad_request(req) \
1696 (((req) + CHUNK_OVERHEAD + CHUNK_ALIGN_MASK) & ~CHUNK_ALIGN_MASK)
1698 /* pad request, checking for minimum (but not maximum) */
1699 #define request2size(req) \
1700 (((req) < MIN_REQUEST)? MIN_CHUNK_SIZE : pad_request(req))
1703 /* ------------------ Operations on head and foot fields ----------------- */
1706 The head field of a chunk is or'ed with PINUSE_BIT when previous
1707 adjacent chunk in use, and or'ed with CINUSE_BIT if this chunk is in
1708 use. If the chunk was obtained with mmap, the prev_foot field has
1709 IS_MMAPPED_BIT set, otherwise holding the offset of the base of the
1710 mmapped region to the base of the chunk.
1713 #define PINUSE_BIT (SIZE_T_ONE)
1714 #define CINUSE_BIT (SIZE_T_TWO)
1715 #define INUSE_BITS (PINUSE_BIT|CINUSE_BIT)
1717 /* Head value for fenceposts */
1718 #define FENCEPOST_HEAD (INUSE_BITS|SIZE_T_SIZE)
1720 /* extraction of fields from head words */
1721 #define cinuse(p) ((p)->head & CINUSE_BIT)
1722 #define pinuse(p) ((p)->head & PINUSE_BIT)
1723 #define chunksize(p) ((p)->head & ~(INUSE_BITS))
1725 #define clear_pinuse(p) ((p)->head &= ~PINUSE_BIT)
1726 #define clear_cinuse(p) ((p)->head &= ~CINUSE_BIT)
1728 /* Treat space at ptr +/- offset as a chunk */
1729 #define chunk_plus_offset(p, s) ((mchunkptr)(((char*)(p)) + (s)))
1730 #define chunk_minus_offset(p, s) ((mchunkptr)(((char*)(p)) - (s)))
1732 /* Ptr to next or previous physical malloc_chunk. */
1733 #define next_chunk(p) ((mchunkptr)( ((char*)(p)) + ((p)->head & ~INUSE_BITS)))
1734 #define prev_chunk(p) ((mchunkptr)( ((char*)(p)) - ((p)->prev_foot) ))
1736 /* extract next chunk's pinuse bit */
1737 #define next_pinuse(p) ((next_chunk(p)->head) & PINUSE_BIT)
1739 /* Get/set size at footer */
1740 #define get_foot(p, s) (((mchunkptr)((char*)(p) + (s)))->prev_foot)
1741 #define set_foot(p, s) (((mchunkptr)((char*)(p) + (s)))->prev_foot = (s))
1743 /* Set size, pinuse bit, and foot */
1744 #define set_size_and_pinuse_of_free_chunk(p, s)\
1745 ((p)->head = (s|PINUSE_BIT), set_foot(p, s))
1747 /* Set size, pinuse bit, foot, and clear next pinuse */
1748 #define set_free_with_pinuse(p, s, n)\
1749 (clear_pinuse(n), set_size_and_pinuse_of_free_chunk(p, s))
1751 #define is_mmapped(p)\
1752 (!((p)->head & PINUSE_BIT) && ((p)->prev_foot & IS_MMAPPED_BIT))
1754 /* Get the internal overhead associated with chunk p */
1755 #define overhead_for(p)\
1756 (is_mmapped(p)? MMAP_CHUNK_OVERHEAD : CHUNK_OVERHEAD)
1758 /* Return true if malloced space is not necessarily cleared */
1760 #define calloc_must_clear(p) (!is_mmapped(p))
1761 #else /* MMAP_CLEARS */
1762 #define calloc_must_clear(p) (1)
1763 #endif /* MMAP_CLEARS */
1765 /* ---------------------- Overlaid data structures ----------------------- */
1768 When chunks are not in use, they are treated as nodes of either
1771 "Small" chunks are stored in circular doubly-linked lists, and look
1774 chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1775 | Size of previous chunk |
1776 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1777 `head:' | Size of chunk, in bytes |P|
1778 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1779 | Forward pointer to next chunk in list |
1780 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1781 | Back pointer to previous chunk in list |
1782 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1783 | Unused space (may be 0 bytes long) .
1786 nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1787 `foot:' | Size of chunk, in bytes |
1788 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1790 Larger chunks are kept in a form of bitwise digital trees (aka
1791 tries) keyed on chunksizes. Because malloc_tree_chunks are only for
1792 free chunks greater than 256 bytes, their size doesn't impose any
1793 constraints on user chunk sizes. Each node looks like:
1795 chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1796 | Size of previous chunk |
1797 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1798 `head:' | Size of chunk, in bytes |P|
1799 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1800 | Forward pointer to next chunk of same size |
1801 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1802 | Back pointer to previous chunk of same size |
1803 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1804 | Pointer to left child (child[0]) |
1805 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1806 | Pointer to right child (child[1]) |
1807 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1808 | Pointer to parent |
1809 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1810 | bin index of this chunk |
1811 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1814 nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1815 `foot:' | Size of chunk, in bytes |
1816 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1818 Each tree holding treenodes is a tree of unique chunk sizes. Chunks
1819 of the same size are arranged in a circularly-linked list, with only
1820 the oldest chunk (the next to be used, in our FIFO ordering)
1821 actually in the tree. (Tree members are distinguished by a non-null
1822 parent pointer.) If a chunk with the same size an an existing node
1823 is inserted, it is linked off the existing node using pointers that
1824 work in the same way as fd/bk pointers of small chunks.
1826 Each tree contains a power of 2 sized range of chunk sizes (the
1827 smallest is 0x100 <= x < 0x180), which is is divided in half at each
1828 tree level, with the chunks in the smaller half of the range (0x100
1829 <= x < 0x140 for the top nose) in the left subtree and the larger
1830 half (0x140 <= x < 0x180) in the right subtree. This is, of course,
1831 done by inspecting individual bits.
1833 Using these rules, each node's left subtree contains all smaller
1834 sizes than its right subtree. However, the node at the root of each
1835 subtree has no particular ordering relationship to either. (The
1836 dividing line between the subtree sizes is based on trie relation.)
1837 If we remove the last chunk of a given size from the interior of the
1838 tree, we need to replace it with a leaf node. The tree ordering
1839 rules permit a node to be replaced by any leaf below it.
1841 The smallest chunk in a tree (a common operation in a best-fit
1842 allocator) can be found by walking a path to the leftmost leaf in
1843 the tree. Unlike a usual binary tree, where we follow left child
1844 pointers until we reach a null, here we follow the right child
1845 pointer any time the left one is null, until we reach a leaf with
1846 both child pointers null. The smallest chunk in the tree will be
1847 somewhere along that path.
1849 The worst case number of steps to add, find, or remove a node is
1850 bounded by the number of bits differentiating chunks within
1851 bins. Under current bin calculations, this ranges from 6 up to 21
1852 (for 32 bit sizes) or up to 53 (for 64 bit sizes). The typical case
1853 is of course much better.
1856 struct malloc_tree_chunk {
1857 /* The first four fields must be compatible with malloc_chunk */
1860 struct malloc_tree_chunk* fd;
1861 struct malloc_tree_chunk* bk;
1863 struct malloc_tree_chunk* child[2];
1864 struct malloc_tree_chunk* parent;
1868 typedef struct malloc_tree_chunk tchunk;
1869 typedef struct malloc_tree_chunk* tchunkptr;
1870 typedef struct malloc_tree_chunk* tbinptr; /* The type of bins of trees */
1872 /* A little helper macro for trees */
1873 #define leftmost_child(t) ((t)->child[0] != 0? (t)->child[0] : (t)->child[1])
1875 /* ----------------------------- Segments -------------------------------- */
1878 Each malloc space may include non-contiguous segments, held in a
1879 list headed by an embedded malloc_segment record representing the
1880 top-most space. Segments also include flags holding properties of
1881 the space. Large chunks that are directly allocated by mmap are not
1882 included in this list. They are instead independently created and
1883 destroyed without otherwise keeping track of them.
1885 Segment management mainly comes into play for spaces allocated by
1886 MMAP. Any call to MMAP might or might not return memory that is
1887 adjacent to an existing segment. MORECORE normally contiguously
1888 extends the current space, so this space is almost always adjacent,
1889 which is simpler and faster to deal with. (This is why MORECORE is
1890 used preferentially to MMAP when both are available -- see
1891 sys_alloc.) When allocating using MMAP, we don't use any of the
1892 hinting mechanisms (inconsistently) supported in various
1893 implementations of unix mmap, or distinguish reserving from
1894 committing memory. Instead, we just ask for space, and exploit
1895 contiguity when we get it. It is probably possible to do
1896 better than this on some systems, but no general scheme seems
1897 to be significantly better.
1899 Management entails a simpler variant of the consolidation scheme
1900 used for chunks to reduce fragmentation -- new adjacent memory is
1901 normally prepended or appended to an existing segment. However,
1902 there are limitations compared to chunk consolidation that mostly
1903 reflect the fact that segment processing is relatively infrequent
1904 (occurring only when getting memory from system) and that we
1905 don't expect to have huge numbers of segments:
1907 * Segments are not indexed, so traversal requires linear scans. (It
1908 would be possible to index these, but is not worth the extra
1909 overhead and complexity for most programs on most platforms.)
1910 * New segments are only appended to old ones when holding top-most
1911 memory; if they cannot be prepended to others, they are held in
1914 Except for the top-most segment of an mstate, each segment record
1915 is kept at the tail of its segment. Segments are added by pushing
1916 segment records onto the list headed by &mstate.seg for the
1919 Segment flags control allocation/merge/deallocation policies:
1920 * If EXTERN_BIT set, then we did not allocate this segment,
1921 and so should not try to deallocate or merge with others.
1922 (This currently holds only for the initial segment passed
1923 into create_mspace_with_base.)
1924 * If IS_MMAPPED_BIT set, the segment may be merged with
1925 other surrounding mmapped segments and trimmed/de-allocated
1927 * If neither bit is set, then the segment was obtained using
1928 MORECORE so can be merged with surrounding MORECORE'd segments
1929 and deallocated/trimmed using MORECORE with negative arguments.
1932 struct malloc_segment {
1933 char* base; /* base address */
1934 size_t size; /* allocated size */
1935 struct malloc_segment* next; /* ptr to next segment */
1936 #if FFI_MMAP_EXEC_WRIT
1937 /* The mmap magic is supposed to store the address of the executable
1938 segment at the very end of the requested block. */
1940 # define mmap_exec_offset(b,s) (*(ptrdiff_t*)((b)+(s)-sizeof(ptrdiff_t)))
1942 /* We can only merge segments if their corresponding executable
1943 segments are at identical offsets. */
1944 # define check_segment_merge(S,b,s) \
1945 (mmap_exec_offset((b),(s)) == (S)->exec_offset)
1947 # define add_segment_exec_offset(p,S) ((char*)(p) + (S)->exec_offset)
1948 # define sub_segment_exec_offset(p,S) ((char*)(p) - (S)->exec_offset)
1950 /* The removal of sflags only works with HAVE_MORECORE == 0. */
1952 # define get_segment_flags(S) (IS_MMAPPED_BIT)
1953 # define set_segment_flags(S,v) \
1954 (((v) != IS_MMAPPED_BIT) ? (ABORT, (v)) : \
1955 (((S)->exec_offset = \
1956 mmap_exec_offset((S)->base, (S)->size)), \
1957 (mmap_exec_offset((S)->base + (S)->exec_offset, (S)->size) != \
1958 (S)->exec_offset) ? (ABORT, (v)) : \
1959 (mmap_exec_offset((S)->base, (S)->size) = 0), (v)))
1961 /* We use an offset here, instead of a pointer, because then, when
1962 base changes, we don't have to modify this. On architectures
1963 with segmented addresses, this might not work. */
1964 ptrdiff_t exec_offset;
1967 # define get_segment_flags(S) ((S)->sflags)
1968 # define set_segment_flags(S,v) ((S)->sflags = (v))
1969 # define check_segment_merge(S,b,s) (1)
1971 flag_t sflags; /* mmap and extern flag */
1975 #define is_mmapped_segment(S) (get_segment_flags(S) & IS_MMAPPED_BIT)
1976 #define is_extern_segment(S) (get_segment_flags(S) & EXTERN_BIT)
1978 typedef struct malloc_segment msegment;
1979 typedef struct malloc_segment* msegmentptr;
1981 /* ---------------------------- malloc_state ----------------------------- */
1984 A malloc_state holds all of the bookkeeping for a space.
1985 The main fields are:
1988 The topmost chunk of the currently active segment. Its size is
1989 cached in topsize. The actual size of topmost space is
1990 topsize+TOP_FOOT_SIZE, which includes space reserved for adding
1991 fenceposts and segment records if necessary when getting more
1992 space from the system. The size at which to autotrim top is
1993 cached from mparams in trim_check, except that it is disabled if
1996 Designated victim (dv)
1997 This is the preferred chunk for servicing small requests that
1998 don't have exact fits. It is normally the chunk split off most
1999 recently to service another small request. Its size is cached in
2000 dvsize. The link fields of this chunk are not maintained since it
2001 is not kept in a bin.
2004 An array of bin headers for free chunks. These bins hold chunks
2005 with sizes less than MIN_LARGE_SIZE bytes. Each bin contains
2006 chunks of all the same size, spaced 8 bytes apart. To simplify
2007 use in double-linked lists, each bin header acts as a malloc_chunk
2008 pointing to the real first node, if it exists (else pointing to
2009 itself). This avoids special-casing for headers. But to avoid
2010 waste, we allocate only the fd/bk pointers of bins, and then use
2011 repositioning tricks to treat these as the fields of a chunk.
2014 Treebins are pointers to the roots of trees holding a range of
2015 sizes. There are 2 equally spaced treebins for each power of two
2016 from TREE_SHIFT to TREE_SHIFT+16. The last bin holds anything
2020 There is one bit map for small bins ("smallmap") and one for
2021 treebins ("treemap). Each bin sets its bit when non-empty, and
2022 clears the bit when empty. Bit operations are then used to avoid
2023 bin-by-bin searching -- nearly all "search" is done without ever
2024 looking at bins that won't be selected. The bit maps
2025 conservatively use 32 bits per map word, even if on 64bit system.
2026 For a good description of some of the bit-based techniques used
2027 here, see Henry S. Warren Jr's book "Hacker's Delight" (and
2028 supplement at http://hackersdelight.org/). Many of these are
2029 intended to reduce the branchiness of paths through malloc etc, as
2030 well as to reduce the number of memory locations read or written.
2033 A list of segments headed by an embedded malloc_segment record
2034 representing the initial space.
2036 Address check support
2037 The least_addr field is the least address ever obtained from
2038 MORECORE or MMAP. Attempted frees and reallocs of any address less
2039 than this are trapped (unless INSECURE is defined).
2042 A cross-check field that should always hold same value as mparams.magic.
2045 Bits recording whether to use MMAP, locks, or contiguous MORECORE
2048 Each space keeps track of current and maximum system memory
2049 obtained via MORECORE or MMAP.
2052 If USE_LOCKS is defined, the "mutex" lock is acquired and released
2053 around every public call using this mspace.
2056 /* Bin types, widths and sizes */
2057 #define NSMALLBINS (32U)
2058 #define NTREEBINS (32U)
2059 #define SMALLBIN_SHIFT (3U)
2060 #define SMALLBIN_WIDTH (SIZE_T_ONE << SMALLBIN_SHIFT)
2061 #define TREEBIN_SHIFT (8U)
2062 #define MIN_LARGE_SIZE (SIZE_T_ONE << TREEBIN_SHIFT)
2063 #define MAX_SMALL_SIZE (MIN_LARGE_SIZE - SIZE_T_ONE)
2064 #define MAX_SMALL_REQUEST (MAX_SMALL_SIZE - CHUNK_ALIGN_MASK - CHUNK_OVERHEAD)
2066 struct malloc_state {
2076 mchunkptr smallbins[(NSMALLBINS+1)*2];
2077 tbinptr treebins[NTREEBINS];
2079 size_t max_footprint;
2082 MLOCK_T mutex; /* locate lock among fields that rarely change */
2083 #endif /* USE_LOCKS */
2087 typedef struct malloc_state* mstate;
2089 /* ------------- Global malloc_state and malloc_params ------------------- */
2092 malloc_params holds global properties, including those that can be
2093 dynamically set using mallopt. There is a single instance, mparams,
2094 initialized in init_mparams.
2097 struct malloc_params {
2101 size_t mmap_threshold;
2102 size_t trim_threshold;
2103 flag_t default_mflags;
2106 static struct malloc_params mparams;
2108 /* The global malloc_state used for all non-"mspace" calls */
2109 static struct malloc_state _gm_;
2111 #define is_global(M) ((M) == &_gm_)
2112 #define is_initialized(M) ((M)->top != 0)
2114 /* -------------------------- system alloc setup ------------------------- */
2116 /* Operations on mflags */
2118 #define use_lock(M) ((M)->mflags & USE_LOCK_BIT)
2119 #define enable_lock(M) ((M)->mflags |= USE_LOCK_BIT)
2120 #define disable_lock(M) ((M)->mflags &= ~USE_LOCK_BIT)
2122 #define use_mmap(M) ((M)->mflags & USE_MMAP_BIT)
2123 #define enable_mmap(M) ((M)->mflags |= USE_MMAP_BIT)
2124 #define disable_mmap(M) ((M)->mflags &= ~USE_MMAP_BIT)
2126 #define use_noncontiguous(M) ((M)->mflags & USE_NONCONTIGUOUS_BIT)
2127 #define disable_contiguous(M) ((M)->mflags |= USE_NONCONTIGUOUS_BIT)
2129 #define set_lock(M,L)\
2130 ((M)->mflags = (L)?\
2131 ((M)->mflags | USE_LOCK_BIT) :\
2132 ((M)->mflags & ~USE_LOCK_BIT))
2134 /* page-align a size */
2135 #define page_align(S)\
2136 (((S) + (mparams.page_size)) & ~(mparams.page_size - SIZE_T_ONE))
2138 /* granularity-align a size */
2139 #define granularity_align(S)\
2140 (((S) + (mparams.granularity)) & ~(mparams.granularity - SIZE_T_ONE))
2142 #define is_page_aligned(S)\
2143 (((size_t)(S) & (mparams.page_size - SIZE_T_ONE)) == 0)
2144 #define is_granularity_aligned(S)\
2145 (((size_t)(S) & (mparams.granularity - SIZE_T_ONE)) == 0)
2147 /* True if segment S holds address A */
2148 #define segment_holds(S, A)\
2149 ((char*)(A) >= S->base && (char*)(A) < S->base + S->size)
2151 /* Return segment holding given address */
2152 static msegmentptr segment_holding(mstate m, char* addr) {
2153 msegmentptr sp = &m->seg;
2155 if (addr >= sp->base && addr < sp->base + sp->size)
2157 if ((sp = sp->next) == 0)
2162 /* Return true if segment contains a segment link */
2163 static int has_segment_link(mstate m, msegmentptr ss) {
2164 msegmentptr sp = &m->seg;
2166 if ((char*)sp >= ss->base && (char*)sp < ss->base + ss->size)
2168 if ((sp = sp->next) == 0)
2173 #ifndef MORECORE_CANNOT_TRIM
2174 #define should_trim(M,s) ((s) > (M)->trim_check)
2175 #else /* MORECORE_CANNOT_TRIM */
2176 #define should_trim(M,s) (0)
2177 #endif /* MORECORE_CANNOT_TRIM */
2180 TOP_FOOT_SIZE is padding at the end of a segment, including space
2181 that may be needed to place segment records and fenceposts when new
2182 noncontiguous segments are added.
2184 #define TOP_FOOT_SIZE\
2185 (align_offset(chunk2mem(0))+pad_request(sizeof(struct malloc_segment))+MIN_CHUNK_SIZE)
2188 /* ------------------------------- Hooks -------------------------------- */
2191 PREACTION should be defined to return 0 on success, and nonzero on
2192 failure. If you are not using locking, you can redefine these to do
2198 /* Ensure locks are initialized */
2199 #define GLOBALLY_INITIALIZE() (mparams.page_size == 0 && init_mparams())
2201 #define PREACTION(M) ((GLOBALLY_INITIALIZE() || use_lock(M))? ACQUIRE_LOCK(&(M)->mutex) : 0)
2202 #define POSTACTION(M) { if (use_lock(M)) RELEASE_LOCK(&(M)->mutex); }
2203 #else /* USE_LOCKS */
2206 #define PREACTION(M) (0)
2207 #endif /* PREACTION */
2210 #define POSTACTION(M)
2211 #endif /* POSTACTION */
2213 #endif /* USE_LOCKS */
2216 CORRUPTION_ERROR_ACTION is triggered upon detected bad addresses.
2217 USAGE_ERROR_ACTION is triggered on detected bad frees and
2218 reallocs. The argument p is an address that might have triggered the
2219 fault. It is ignored by the two predefined actions, but might be
2220 useful in custom actions that try to help diagnose errors.
2223 #if PROCEED_ON_ERROR
2225 /* A count of the number of corruption errors causing resets */
2226 int malloc_corruption_error_count;
2228 /* default corruption action */
2229 static void reset_on_error(mstate m);
2231 #define CORRUPTION_ERROR_ACTION(m) reset_on_error(m)
2232 #define USAGE_ERROR_ACTION(m, p)
2234 #else /* PROCEED_ON_ERROR */
2236 #ifndef CORRUPTION_ERROR_ACTION
2237 #define CORRUPTION_ERROR_ACTION(m) ABORT
2238 #endif /* CORRUPTION_ERROR_ACTION */
2240 #ifndef USAGE_ERROR_ACTION
2241 #define USAGE_ERROR_ACTION(m,p) ABORT
2242 #endif /* USAGE_ERROR_ACTION */
2244 #endif /* PROCEED_ON_ERROR */
2246 /* -------------------------- Debugging setup ---------------------------- */
2250 #define check_free_chunk(M,P)
2251 #define check_inuse_chunk(M,P)
2252 #define check_malloced_chunk(M,P,N)
2253 #define check_mmapped_chunk(M,P)
2254 #define check_malloc_state(M)
2255 #define check_top_chunk(M,P)
2258 #define check_free_chunk(M,P) do_check_free_chunk(M,P)
2259 #define check_inuse_chunk(M,P) do_check_inuse_chunk(M,P)
2260 #define check_top_chunk(M,P) do_check_top_chunk(M,P)
2261 #define check_malloced_chunk(M,P,N) do_check_malloced_chunk(M,P,N)
2262 #define check_mmapped_chunk(M,P) do_check_mmapped_chunk(M,P)
2263 #define check_malloc_state(M) do_check_malloc_state(M)
2265 static void do_check_any_chunk(mstate m, mchunkptr p);
2266 static void do_check_top_chunk(mstate m, mchunkptr p);
2267 static void do_check_mmapped_chunk(mstate m, mchunkptr p);
2268 static void do_check_inuse_chunk(mstate m, mchunkptr p);
2269 static void do_check_free_chunk(mstate m, mchunkptr p);
2270 static void do_check_malloced_chunk(mstate m, void* mem, size_t s);
2271 static void do_check_tree(mstate m, tchunkptr t);
2272 static void do_check_treebin(mstate m, bindex_t i);
2273 static void do_check_smallbin(mstate m, bindex_t i);
2274 static void do_check_malloc_state(mstate m);
2275 static int bin_find(mstate m, mchunkptr x);
2276 static size_t traverse_and_check(mstate m);
2279 /* ---------------------------- Indexing Bins ---------------------------- */
2281 #define is_small(s) (((s) >> SMALLBIN_SHIFT) < NSMALLBINS)
2282 #define small_index(s) ((s) >> SMALLBIN_SHIFT)
2283 #define small_index2size(i) ((i) << SMALLBIN_SHIFT)
2284 #define MIN_SMALL_INDEX (small_index(MIN_CHUNK_SIZE))
2286 /* addressing by index. See above about smallbin repositioning */
2287 #define smallbin_at(M, i) ((sbinptr)((char*)&((M)->smallbins[(i)<<1])))
2288 #define treebin_at(M,i) (&((M)->treebins[i]))
2290 /* assign tree index for size S to variable I */
2291 #if defined(__GNUC__) && defined(i386)
2292 #define compute_tree_index(S, I)\
2294 size_t X = S >> TREEBIN_SHIFT;\
2297 else if (X > 0xFFFF)\
2301 __asm__("bsrl %1,%0\n\t" : "=r" (K) : "rm" (X));\
2302 I = (bindex_t)((K << 1) + ((S >> (K + (TREEBIN_SHIFT-1)) & 1)));\
2306 #define compute_tree_index(S, I)\
2308 size_t X = S >> TREEBIN_SHIFT;\
2311 else if (X > 0xFFFF)\
2314 unsigned int Y = (unsigned int)X;\
2315 unsigned int N = ((Y - 0x100) >> 16) & 8;\
2316 unsigned int K = (((Y <<= N) - 0x1000) >> 16) & 4;\
2318 N += K = (((Y <<= K) - 0x4000) >> 16) & 2;\
2319 K = 14 - N + ((Y <<= K) >> 15);\
2320 I = (K << 1) + ((S >> (K + (TREEBIN_SHIFT-1)) & 1));\
2325 /* Bit representing maximum resolved size in a treebin at i */
2326 #define bit_for_tree_index(i) \
2327 (i == NTREEBINS-1)? (SIZE_T_BITSIZE-1) : (((i) >> 1) + TREEBIN_SHIFT - 2)
2329 /* Shift placing maximum resolved bit in a treebin at i as sign bit */
2330 #define leftshift_for_tree_index(i) \
2331 ((i == NTREEBINS-1)? 0 : \
2332 ((SIZE_T_BITSIZE-SIZE_T_ONE) - (((i) >> 1) + TREEBIN_SHIFT - 2)))
2334 /* The size of the smallest chunk held in bin with index i */
2335 #define minsize_for_tree_index(i) \
2336 ((SIZE_T_ONE << (((i) >> 1) + TREEBIN_SHIFT)) | \
2337 (((size_t)((i) & SIZE_T_ONE)) << (((i) >> 1) + TREEBIN_SHIFT - 1)))
2340 /* ------------------------ Operations on bin maps ----------------------- */
2342 /* bit corresponding to given index */
2343 #define idx2bit(i) ((binmap_t)(1) << (i))
2345 /* Mark/Clear bits with given index */
2346 #define mark_smallmap(M,i) ((M)->smallmap |= idx2bit(i))
2347 #define clear_smallmap(M,i) ((M)->smallmap &= ~idx2bit(i))
2348 #define smallmap_is_marked(M,i) ((M)->smallmap & idx2bit(i))
2350 #define mark_treemap(M,i) ((M)->treemap |= idx2bit(i))
2351 #define clear_treemap(M,i) ((M)->treemap &= ~idx2bit(i))
2352 #define treemap_is_marked(M,i) ((M)->treemap & idx2bit(i))
2354 /* index corresponding to given bit */
2356 #if defined(__GNUC__) && defined(i386)
2357 #define compute_bit2idx(X, I)\
2360 __asm__("bsfl %1,%0\n\t" : "=r" (J) : "rm" (X));\
2366 #define compute_bit2idx(X, I) I = ffs(X)-1
2368 #else /* USE_BUILTIN_FFS */
2369 #define compute_bit2idx(X, I)\
2371 unsigned int Y = X - 1;\
2372 unsigned int K = Y >> (16-4) & 16;\
2373 unsigned int N = K; Y >>= K;\
2374 N += K = Y >> (8-3) & 8; Y >>= K;\
2375 N += K = Y >> (4-2) & 4; Y >>= K;\
2376 N += K = Y >> (2-1) & 2; Y >>= K;\
2377 N += K = Y >> (1-0) & 1; Y >>= K;\
2378 I = (bindex_t)(N + Y);\
2380 #endif /* USE_BUILTIN_FFS */
2383 /* isolate the least set bit of a bitmap */
2384 #define least_bit(x) ((x) & -(x))
2386 /* mask with all bits to left of least bit of x on */
2387 #define left_bits(x) ((x<<1) | -(x<<1))
2389 /* mask with all bits to left of or equal to least bit of x on */
2390 #define same_or_left_bits(x) ((x) | -(x))
2393 /* ----------------------- Runtime Check Support ------------------------- */
2396 For security, the main invariant is that malloc/free/etc never
2397 writes to a static address other than malloc_state, unless static
2398 malloc_state itself has been corrupted, which cannot occur via
2399 malloc (because of these checks). In essence this means that we
2400 believe all pointers, sizes, maps etc held in malloc_state, but
2401 check all of those linked or offsetted from other embedded data
2402 structures. These checks are interspersed with main code in a way
2403 that tends to minimize their run-time cost.
2405 When FOOTERS is defined, in addition to range checking, we also
2406 verify footer fields of inuse chunks, which can be used guarantee
2407 that the mstate controlling malloc/free is intact. This is a
2408 streamlined version of the approach described by William Robertson
2409 et al in "Run-time Detection of Heap-based Overflows" LISA'03
2410 http://www.usenix.org/events/lisa03/tech/robertson.html The footer
2411 of an inuse chunk holds the xor of its mstate and a random seed,
2412 that is checked upon calls to free() and realloc(). This is
2413 (probablistically) unguessable from outside the program, but can be
2414 computed by any code successfully malloc'ing any chunk, so does not
2415 itself provide protection against code that has already broken
2416 security through some other means. Unlike Robertson et al, we
2417 always dynamically check addresses of all offset chunks (previous,
2418 next, etc). This turns out to be cheaper than relying on hashes.
2422 /* Check if address a is at least as high as any from MORECORE or MMAP */
2423 #define ok_address(M, a) ((char*)(a) >= (M)->least_addr)
2424 /* Check if address of next chunk n is higher than base chunk p */
2425 #define ok_next(p, n) ((char*)(p) < (char*)(n))
2426 /* Check if p has its cinuse bit on */
2427 #define ok_cinuse(p) cinuse(p)
2428 /* Check if p has its pinuse bit on */
2429 #define ok_pinuse(p) pinuse(p)
2431 #else /* !INSECURE */
2432 #define ok_address(M, a) (1)
2433 #define ok_next(b, n) (1)
2434 #define ok_cinuse(p) (1)
2435 #define ok_pinuse(p) (1)
2436 #endif /* !INSECURE */
2438 #if (FOOTERS && !INSECURE)
2439 /* Check if (alleged) mstate m has expected magic field */
2440 #define ok_magic(M) ((M)->magic == mparams.magic)
2441 #else /* (FOOTERS && !INSECURE) */
2442 #define ok_magic(M) (1)
2443 #endif /* (FOOTERS && !INSECURE) */
2446 /* In gcc, use __builtin_expect to minimize impact of checks */
2448 #if defined(__GNUC__) && __GNUC__ >= 3
2449 #define RTCHECK(e) __builtin_expect(e, 1)
2451 #define RTCHECK(e) (e)
2453 #else /* !INSECURE */
2454 #define RTCHECK(e) (1)
2455 #endif /* !INSECURE */
2457 /* macros to set up inuse chunks with or without footers */
2461 #define mark_inuse_foot(M,p,s)
2463 /* Set cinuse bit and pinuse bit of next chunk */
2464 #define set_inuse(M,p,s)\
2465 ((p)->head = (((p)->head & PINUSE_BIT)|s|CINUSE_BIT),\
2466 ((mchunkptr)(((char*)(p)) + (s)))->head |= PINUSE_BIT)
2468 /* Set cinuse and pinuse of this chunk and pinuse of next chunk */
2469 #define set_inuse_and_pinuse(M,p,s)\
2470 ((p)->head = (s|PINUSE_BIT|CINUSE_BIT),\
2471 ((mchunkptr)(((char*)(p)) + (s)))->head |= PINUSE_BIT)
2473 /* Set size, cinuse and pinuse bit of this chunk */
2474 #define set_size_and_pinuse_of_inuse_chunk(M, p, s)\
2475 ((p)->head = (s|PINUSE_BIT|CINUSE_BIT))
2479 /* Set foot of inuse chunk to be xor of mstate and seed */
2480 #define mark_inuse_foot(M,p,s)\
2481 (((mchunkptr)((char*)(p) + (s)))->prev_foot = ((size_t)(M) ^ mparams.magic))
2483 #define get_mstate_for(p)\
2484 ((mstate)(((mchunkptr)((char*)(p) +\
2485 (chunksize(p))))->prev_foot ^ mparams.magic))
2487 #define set_inuse(M,p,s)\
2488 ((p)->head = (((p)->head & PINUSE_BIT)|s|CINUSE_BIT),\
2489 (((mchunkptr)(((char*)(p)) + (s)))->head |= PINUSE_BIT), \
2490 mark_inuse_foot(M,p,s))
2492 #define set_inuse_and_pinuse(M,p,s)\
2493 ((p)->head = (s|PINUSE_BIT|CINUSE_BIT),\
2494 (((mchunkptr)(((char*)(p)) + (s)))->head |= PINUSE_BIT),\
2495 mark_inuse_foot(M,p,s))
2497 #define set_size_and_pinuse_of_inuse_chunk(M, p, s)\
2498 ((p)->head = (s|PINUSE_BIT|CINUSE_BIT),\
2499 mark_inuse_foot(M, p, s))
2501 #endif /* !FOOTERS */
2503 /* ---------------------------- setting mparams -------------------------- */
2505 /* Initialize mparams */
2506 static int init_mparams(void) {
2507 if (mparams.page_size == 0) {
2510 mparams.mmap_threshold = DEFAULT_MMAP_THRESHOLD;
2511 mparams.trim_threshold = DEFAULT_TRIM_THRESHOLD;
2512 #if MORECORE_CONTIGUOUS
2513 mparams.default_mflags = USE_LOCK_BIT|USE_MMAP_BIT;
2514 #else /* MORECORE_CONTIGUOUS */
2515 mparams.default_mflags = USE_LOCK_BIT|USE_MMAP_BIT|USE_NONCONTIGUOUS_BIT;
2516 #endif /* MORECORE_CONTIGUOUS */
2518 #if (FOOTERS && !INSECURE)
2522 unsigned char buf[sizeof(size_t)];
2523 /* Try to use /dev/urandom, else fall back on using time */
2524 if ((fd = open("/dev/urandom", O_RDONLY)) >= 0 &&
2525 read(fd, buf, sizeof(buf)) == sizeof(buf)) {
2526 s = *((size_t *) buf);
2530 #endif /* USE_DEV_RANDOM */
2531 s = (size_t)(time(0) ^ (size_t)0x55555555U);
2533 s |= (size_t)8U; /* ensure nonzero */
2534 s &= ~(size_t)7U; /* improve chances of fault for bad values */
2537 #else /* (FOOTERS && !INSECURE) */
2538 s = (size_t)0x58585858U;
2539 #endif /* (FOOTERS && !INSECURE) */
2540 ACQUIRE_MAGIC_INIT_LOCK();
2541 if (mparams.magic == 0) {
2543 /* Set up lock for main malloc area */
2544 INITIAL_LOCK(&gm->mutex);
2545 gm->mflags = mparams.default_mflags;
2547 RELEASE_MAGIC_INIT_LOCK();
2549 #if !defined(WIN32) && !defined(__OS2__)
2550 mparams.page_size = malloc_getpagesize;
2551 mparams.granularity = ((DEFAULT_GRANULARITY != 0)?
2552 DEFAULT_GRANULARITY : mparams.page_size);
2553 #elif defined (__OS2__)
2554 /* if low-memory is used, os2munmap() would break
2555 if it were anything other than 64k */
2556 mparams.page_size = 4096u;
2557 mparams.granularity = 65536u;
2560 SYSTEM_INFO system_info;
2561 GetSystemInfo(&system_info);
2562 mparams.page_size = system_info.dwPageSize;
2563 mparams.granularity = system_info.dwAllocationGranularity;
2567 /* Sanity-check configuration:
2568 size_t must be unsigned and as wide as pointer type.
2569 ints must be at least 4 bytes.
2570 alignment must be at least 8.
2571 Alignment, min chunk size, and page size must all be powers of 2.
2573 if ((sizeof(size_t) != sizeof(char*)) ||
2574 (MAX_SIZE_T < MIN_CHUNK_SIZE) ||
2575 (sizeof(int) < 4) ||
2576 (MALLOC_ALIGNMENT < (size_t)8U) ||
2577 ((MALLOC_ALIGNMENT & (MALLOC_ALIGNMENT-SIZE_T_ONE)) != 0) ||
2578 ((MCHUNK_SIZE & (MCHUNK_SIZE-SIZE_T_ONE)) != 0) ||
2579 ((mparams.granularity & (mparams.granularity-SIZE_T_ONE)) != 0) ||
2580 ((mparams.page_size & (mparams.page_size-SIZE_T_ONE)) != 0))
2586 /* support for mallopt */
2587 static int change_mparam(int param_number, int value) {
2588 size_t val = (size_t)value;
2590 switch(param_number) {
2591 case M_TRIM_THRESHOLD:
2592 mparams.trim_threshold = val;
2595 if (val >= mparams.page_size && ((val & (val-1)) == 0)) {
2596 mparams.granularity = val;
2601 case M_MMAP_THRESHOLD:
2602 mparams.mmap_threshold = val;
2610 /* ------------------------- Debugging Support --------------------------- */
2612 /* Check properties of any chunk, whether free, inuse, mmapped etc */
2613 static void do_check_any_chunk(mstate m, mchunkptr p) {
2614 assert((is_aligned(chunk2mem(p))) || (p->head == FENCEPOST_HEAD));
2615 assert(ok_address(m, p));
2618 /* Check properties of top chunk */
2619 static void do_check_top_chunk(mstate m, mchunkptr p) {
2620 msegmentptr sp = segment_holding(m, (char*)p);
2621 size_t sz = chunksize(p);
2623 assert((is_aligned(chunk2mem(p))) || (p->head == FENCEPOST_HEAD));
2624 assert(ok_address(m, p));
2625 assert(sz == m->topsize);
2627 assert(sz == ((sp->base + sp->size) - (char*)p) - TOP_FOOT_SIZE);
2629 assert(!next_pinuse(p));
2632 /* Check properties of (inuse) mmapped chunks */
2633 static void do_check_mmapped_chunk(mstate m, mchunkptr p) {
2634 size_t sz = chunksize(p);
2635 size_t len = (sz + (p->prev_foot & ~IS_MMAPPED_BIT) + MMAP_FOOT_PAD);
2636 assert(is_mmapped(p));
2637 assert(use_mmap(m));
2638 assert((is_aligned(chunk2mem(p))) || (p->head == FENCEPOST_HEAD));
2639 assert(ok_address(m, p));
2640 assert(!is_small(sz));
2641 assert((len & (mparams.page_size-SIZE_T_ONE)) == 0);
2642 assert(chunk_plus_offset(p, sz)->head == FENCEPOST_HEAD);
2643 assert(chunk_plus_offset(p, sz+SIZE_T_SIZE)->head == 0);
2646 /* Check properties of inuse chunks */
2647 static void do_check_inuse_chunk(mstate m, mchunkptr p) {
2648 do_check_any_chunk(m, p);
2650 assert(next_pinuse(p));
2651 /* If not pinuse and not mmapped, previous chunk has OK offset */
2652 assert(is_mmapped(p) || pinuse(p) || next_chunk(prev_chunk(p)) == p);
2654 do_check_mmapped_chunk(m, p);
2657 /* Check properties of free chunks */
2658 static void do_check_free_chunk(mstate m, mchunkptr p) {
2659 size_t sz = p->head & ~(PINUSE_BIT|CINUSE_BIT);
2660 mchunkptr next = chunk_plus_offset(p, sz);
2661 do_check_any_chunk(m, p);
2663 assert(!next_pinuse(p));
2664 assert (!is_mmapped(p));
2665 if (p != m->dv && p != m->top) {
2666 if (sz >= MIN_CHUNK_SIZE) {
2667 assert((sz & CHUNK_ALIGN_MASK) == 0);
2668 assert(is_aligned(chunk2mem(p)));
2669 assert(next->prev_foot == sz);
2671 assert (next == m->top || cinuse(next));
2672 assert(p->fd->bk == p);
2673 assert(p->bk->fd == p);
2675 else /* markers are always of size SIZE_T_SIZE */
2676 assert(sz == SIZE_T_SIZE);
2680 /* Check properties of malloced chunks at the point they are malloced */
2681 static void do_check_malloced_chunk(mstate m, void* mem, size_t s) {
2683 mchunkptr p = mem2chunk(mem);
2684 size_t sz = p->head & ~(PINUSE_BIT|CINUSE_BIT);
2685 do_check_inuse_chunk(m, p);
2686 assert((sz & CHUNK_ALIGN_MASK) == 0);
2687 assert(sz >= MIN_CHUNK_SIZE);
2689 /* unless mmapped, size is less than MIN_CHUNK_SIZE more than request */
2690 assert(is_mmapped(p) || sz < (s + MIN_CHUNK_SIZE));
2694 /* Check a tree and its subtrees. */
2695 static void do_check_tree(mstate m, tchunkptr t) {
2698 bindex_t tindex = t->index;
2699 size_t tsize = chunksize(t);
2701 compute_tree_index(tsize, idx);
2702 assert(tindex == idx);
2703 assert(tsize >= MIN_LARGE_SIZE);
2704 assert(tsize >= minsize_for_tree_index(idx));
2705 assert((idx == NTREEBINS-1) || (tsize < minsize_for_tree_index((idx+1))));
2707 do { /* traverse through chain of same-sized nodes */
2708 do_check_any_chunk(m, ((mchunkptr)u));
2709 assert(u->index == tindex);
2710 assert(chunksize(u) == tsize);
2712 assert(!next_pinuse(u));
2713 assert(u->fd->bk == u);
2714 assert(u->bk->fd == u);
2715 if (u->parent == 0) {
2716 assert(u->child[0] == 0);
2717 assert(u->child[1] == 0);
2720 assert(head == 0); /* only one node on chain has parent */
2722 assert(u->parent != u);
2723 assert (u->parent->child[0] == u ||
2724 u->parent->child[1] == u ||
2725 *((tbinptr*)(u->parent)) == u);
2726 if (u->child[0] != 0) {
2727 assert(u->child[0]->parent == u);
2728 assert(u->child[0] != u);
2729 do_check_tree(m, u->child[0]);
2731 if (u->child[1] != 0) {
2732 assert(u->child[1]->parent == u);
2733 assert(u->child[1] != u);
2734 do_check_tree(m, u->child[1]);
2736 if (u->child[0] != 0 && u->child[1] != 0) {
2737 assert(chunksize(u->child[0]) < chunksize(u->child[1]));
2745 /* Check all the chunks in a treebin. */
2746 static void do_check_treebin(mstate m, bindex_t i) {
2747 tbinptr* tb = treebin_at(m, i);
2749 int empty = (m->treemap & (1U << i)) == 0;
2753 do_check_tree(m, t);
2756 /* Check all the chunks in a smallbin. */
2757 static void do_check_smallbin(mstate m, bindex_t i) {
2758 sbinptr b = smallbin_at(m, i);
2759 mchunkptr p = b->bk;
2760 unsigned int empty = (m->smallmap & (1U << i)) == 0;
2764 for (; p != b; p = p->bk) {
2765 size_t size = chunksize(p);
2767 /* each chunk claims to be free */
2768 do_check_free_chunk(m, p);
2769 /* chunk belongs in bin */
2770 assert(small_index(size) == i);
2771 assert(p->bk == b || chunksize(p->bk) == chunksize(p));
2772 /* chunk is followed by an inuse chunk */
2774 if (q->head != FENCEPOST_HEAD)
2775 do_check_inuse_chunk(m, q);
2780 /* Find x in a bin. Used in other check functions. */
2781 static int bin_find(mstate m, mchunkptr x) {
2782 size_t size = chunksize(x);
2783 if (is_small(size)) {
2784 bindex_t sidx = small_index(size);
2785 sbinptr b = smallbin_at(m, sidx);
2786 if (smallmap_is_marked(m, sidx)) {
2791 } while ((p = p->fd) != b);
2796 compute_tree_index(size, tidx);
2797 if (treemap_is_marked(m, tidx)) {
2798 tchunkptr t = *treebin_at(m, tidx);
2799 size_t sizebits = size << leftshift_for_tree_index(tidx);
2800 while (t != 0 && chunksize(t) != size) {
2801 t = t->child[(sizebits >> (SIZE_T_BITSIZE-SIZE_T_ONE)) & 1];
2807 if (u == (tchunkptr)x)
2809 } while ((u = u->fd) != t);
2816 /* Traverse each chunk and check it; return total */
2817 static size_t traverse_and_check(mstate m) {
2819 if (is_initialized(m)) {
2820 msegmentptr s = &m->seg;
2821 sum += m->topsize + TOP_FOOT_SIZE;
2823 mchunkptr q = align_as_chunk(s->base);
2824 mchunkptr lastq = 0;
2826 while (segment_holds(s, q) &&
2827 q != m->top && q->head != FENCEPOST_HEAD) {
2828 sum += chunksize(q);
2830 assert(!bin_find(m, q));
2831 do_check_inuse_chunk(m, q);
2834 assert(q == m->dv || bin_find(m, q));
2835 assert(lastq == 0 || cinuse(lastq)); /* Not 2 consecutive free */
2836 do_check_free_chunk(m, q);
2847 /* Check all properties of malloc_state. */
2848 static void do_check_malloc_state(mstate m) {
2852 for (i = 0; i < NSMALLBINS; ++i)
2853 do_check_smallbin(m, i);
2854 for (i = 0; i < NTREEBINS; ++i)
2855 do_check_treebin(m, i);
2857 if (m->dvsize != 0) { /* check dv chunk */
2858 do_check_any_chunk(m, m->dv);
2859 assert(m->dvsize == chunksize(m->dv));
2860 assert(m->dvsize >= MIN_CHUNK_SIZE);
2861 assert(bin_find(m, m->dv) == 0);
2864 if (m->top != 0) { /* check top chunk */
2865 do_check_top_chunk(m, m->top);
2866 assert(m->topsize == chunksize(m->top));
2867 assert(m->topsize > 0);
2868 assert(bin_find(m, m->top) == 0);
2871 total = traverse_and_check(m);
2872 assert(total <= m->footprint);
2873 assert(m->footprint <= m->max_footprint);
2877 /* ----------------------------- statistics ------------------------------ */
2880 static struct mallinfo internal_mallinfo(mstate m) {
2881 struct mallinfo nm = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
2882 if (!PREACTION(m)) {
2883 check_malloc_state(m);
2884 if (is_initialized(m)) {
2885 size_t nfree = SIZE_T_ONE; /* top always free */
2886 size_t mfree = m->topsize + TOP_FOOT_SIZE;
2888 msegmentptr s = &m->seg;
2890 mchunkptr q = align_as_chunk(s->base);
2891 while (segment_holds(s, q) &&
2892 q != m->top && q->head != FENCEPOST_HEAD) {
2893 size_t sz = chunksize(q);
2906 nm.hblkhd = m->footprint - sum;
2907 nm.usmblks = m->max_footprint;
2908 nm.uordblks = m->footprint - mfree;
2909 nm.fordblks = mfree;
2910 nm.keepcost = m->topsize;
2917 #endif /* !NO_MALLINFO */
2919 static void internal_malloc_stats(mstate m) {
2920 if (!PREACTION(m)) {
2924 check_malloc_state(m);
2925 if (is_initialized(m)) {
2926 msegmentptr s = &m->seg;
2927 maxfp = m->max_footprint;
2929 used = fp - (m->topsize + TOP_FOOT_SIZE);
2932 mchunkptr q = align_as_chunk(s->base);
2933 while (segment_holds(s, q) &&
2934 q != m->top && q->head != FENCEPOST_HEAD) {
2936 used -= chunksize(q);
2943 fprintf(stderr, "max system bytes = %10lu\n", (unsigned long)(maxfp));
2944 fprintf(stderr, "system bytes = %10lu\n", (unsigned long)(fp));
2945 fprintf(stderr, "in use bytes = %10lu\n", (unsigned long)(used));
2951 /* ----------------------- Operations on smallbins ----------------------- */
2954 Various forms of linking and unlinking are defined as macros. Even
2955 the ones for trees, which are very long but have very short typical
2956 paths. This is ugly but reduces reliance on inlining support of
2960 /* Link a free chunk into a smallbin */
2961 #define insert_small_chunk(M, P, S) {\
2962 bindex_t I = small_index(S);\
2963 mchunkptr B = smallbin_at(M, I);\
2965 assert(S >= MIN_CHUNK_SIZE);\
2966 if (!smallmap_is_marked(M, I))\
2967 mark_smallmap(M, I);\
2968 else if (RTCHECK(ok_address(M, B->fd)))\
2971 CORRUPTION_ERROR_ACTION(M);\
2979 /* Unlink a chunk from a smallbin */
2980 #define unlink_small_chunk(M, P, S) {\
2981 mchunkptr F = P->fd;\
2982 mchunkptr B = P->bk;\
2983 bindex_t I = small_index(S);\
2986 assert(chunksize(P) == small_index2size(I));\
2988 clear_smallmap(M, I);\
2989 else if (RTCHECK((F == smallbin_at(M,I) || ok_address(M, F)) &&\
2990 (B == smallbin_at(M,I) || ok_address(M, B)))) {\
2995 CORRUPTION_ERROR_ACTION(M);\
2999 /* Unlink the first chunk from a smallbin */
3000 #define unlink_first_small_chunk(M, B, P, I) {\
3001 mchunkptr F = P->fd;\
3004 assert(chunksize(P) == small_index2size(I));\
3006 clear_smallmap(M, I);\
3007 else if (RTCHECK(ok_address(M, F))) {\
3012 CORRUPTION_ERROR_ACTION(M);\
3016 /* Replace dv node, binning the old one */
3017 /* Used only when dvsize known to be small */
3018 #define replace_dv(M, P, S) {\
3019 size_t DVS = M->dvsize;\
3021 mchunkptr DV = M->dv;\
3022 assert(is_small(DVS));\
3023 insert_small_chunk(M, DV, DVS);\
3029 /* ------------------------- Operations on trees ------------------------- */
3031 /* Insert chunk into tree */
3032 #define insert_large_chunk(M, X, S) {\
3035 compute_tree_index(S, I);\
3036 H = treebin_at(M, I);\
3038 X->child[0] = X->child[1] = 0;\
3039 if (!treemap_is_marked(M, I)) {\
3040 mark_treemap(M, I);\
3042 X->parent = (tchunkptr)H;\
3047 size_t K = S << leftshift_for_tree_index(I);\
3049 if (chunksize(T) != S) {\
3050 tchunkptr* C = &(T->child[(K >> (SIZE_T_BITSIZE-SIZE_T_ONE)) & 1]);\
3054 else if (RTCHECK(ok_address(M, C))) {\
3061 CORRUPTION_ERROR_ACTION(M);\
3066 tchunkptr F = T->fd;\
3067 if (RTCHECK(ok_address(M, T) && ok_address(M, F))) {\
3075 CORRUPTION_ERROR_ACTION(M);\
3086 1. If x is a chained node, unlink it from its same-sized fd/bk links
3087 and choose its bk node as its replacement.
3088 2. If x was the last node of its size, but not a leaf node, it must
3089 be replaced with a leaf node (not merely one with an open left or
3090 right), to make sure that lefts and rights of descendents
3091 correspond properly to bit masks. We use the rightmost descendent
3092 of x. We could use any other leaf, but this is easy to locate and
3093 tends to counteract removal of leftmosts elsewhere, and so keeps
3094 paths shorter than minimally guaranteed. This doesn't loop much
3095 because on average a node in a tree is near the bottom.
3096 3. If x is the base of a chain (i.e., has parent links) relink
3097 x's parent and children to x's replacement (or null if none).
3100 #define unlink_large_chunk(M, X) {\
3101 tchunkptr XP = X->parent;\
3104 tchunkptr F = X->fd;\
3106 if (RTCHECK(ok_address(M, F))) {\
3111 CORRUPTION_ERROR_ACTION(M);\
3116 if (((R = *(RP = &(X->child[1]))) != 0) ||\
3117 ((R = *(RP = &(X->child[0]))) != 0)) {\
3119 while ((*(CP = &(R->child[1])) != 0) ||\
3120 (*(CP = &(R->child[0])) != 0)) {\
3123 if (RTCHECK(ok_address(M, RP)))\
3126 CORRUPTION_ERROR_ACTION(M);\
3131 tbinptr* H = treebin_at(M, X->index);\
3133 if ((*H = R) == 0) \
3134 clear_treemap(M, X->index);\
3136 else if (RTCHECK(ok_address(M, XP))) {\
3137 if (XP->child[0] == X) \
3143 CORRUPTION_ERROR_ACTION(M);\
3145 if (RTCHECK(ok_address(M, R))) {\
3148 if ((C0 = X->child[0]) != 0) {\
3149 if (RTCHECK(ok_address(M, C0))) {\
3154 CORRUPTION_ERROR_ACTION(M);\
3156 if ((C1 = X->child[1]) != 0) {\
3157 if (RTCHECK(ok_address(M, C1))) {\
3162 CORRUPTION_ERROR_ACTION(M);\
3166 CORRUPTION_ERROR_ACTION(M);\
3171 /* Relays to large vs small bin operations */
3173 #define insert_chunk(M, P, S)\
3174 if (is_small(S)) insert_small_chunk(M, P, S)\
3175 else { tchunkptr TP = (tchunkptr)(P); insert_large_chunk(M, TP, S); }
3177 #define unlink_chunk(M, P, S)\
3178 if (is_small(S)) unlink_small_chunk(M, P, S)\
3179 else { tchunkptr TP = (tchunkptr)(P); unlink_large_chunk(M, TP); }
3182 /* Relays to internal calls to malloc/free from realloc, memalign etc */
3185 #define internal_malloc(m, b) mspace_malloc(m, b)
3186 #define internal_free(m, mem) mspace_free(m,mem);
3187 #else /* ONLY_MSPACES */
3189 #define internal_malloc(m, b)\
3190 (m == gm)? dlmalloc(b) : mspace_malloc(m, b)
3191 #define internal_free(m, mem)\
3192 if (m == gm) dlfree(mem); else mspace_free(m,mem);
3194 #define internal_malloc(m, b) dlmalloc(b)
3195 #define internal_free(m, mem) dlfree(mem)
3196 #endif /* MSPACES */
3197 #endif /* ONLY_MSPACES */
3199 /* ----------------------- Direct-mmapping chunks ----------------------- */
3202 Directly mmapped chunks are set up with an offset to the start of
3203 the mmapped region stored in the prev_foot field of the chunk. This
3204 allows reconstruction of the required argument to MUNMAP when freed,
3205 and also allows adjustment of the returned chunk to meet alignment
3206 requirements (especially in memalign). There is also enough space
3207 allocated to hold a fake next chunk of size SIZE_T_SIZE to maintain
3208 the PINUSE bit so frees can be checked.
3211 /* Malloc using mmap */
3212 static void* mmap_alloc(mstate m, size_t nb) {
3213 size_t mmsize = granularity_align(nb + SIX_SIZE_T_SIZES + CHUNK_ALIGN_MASK);
3214 if (mmsize > nb) { /* Check for wrap around 0 */
3215 char* mm = (char*)(DIRECT_MMAP(mmsize));
3217 size_t offset = align_offset(chunk2mem(mm));
3218 size_t psize = mmsize - offset - MMAP_FOOT_PAD;
3219 mchunkptr p = (mchunkptr)(mm + offset);
3220 p->prev_foot = offset | IS_MMAPPED_BIT;
3221 (p)->head = (psize|CINUSE_BIT);
3222 mark_inuse_foot(m, p, psize);
3223 chunk_plus_offset(p, psize)->head = FENCEPOST_HEAD;
3224 chunk_plus_offset(p, psize+SIZE_T_SIZE)->head = 0;
3226 if (mm < m->least_addr)
3228 if ((m->footprint += mmsize) > m->max_footprint)
3229 m->max_footprint = m->footprint;
3230 assert(is_aligned(chunk2mem(p)));
3231 check_mmapped_chunk(m, p);
3232 return chunk2mem(p);
3238 /* Realloc using mmap */
3239 static mchunkptr mmap_resize(mstate m, mchunkptr oldp, size_t nb) {
3240 size_t oldsize = chunksize(oldp);
3241 if (is_small(nb)) /* Can't shrink mmap regions below small size */
3243 /* Keep old chunk if big enough but not too big */
3244 if (oldsize >= nb + SIZE_T_SIZE &&
3245 (oldsize - nb) <= (mparams.granularity << 1))
3248 size_t offset = oldp->prev_foot & ~IS_MMAPPED_BIT;
3249 size_t oldmmsize = oldsize + offset + MMAP_FOOT_PAD;
3250 size_t newmmsize = granularity_align(nb + SIX_SIZE_T_SIZES +
3252 char* cp = (char*)CALL_MREMAP((char*)oldp - offset,
3253 oldmmsize, newmmsize, 1);
3255 mchunkptr newp = (mchunkptr)(cp + offset);
3256 size_t psize = newmmsize - offset - MMAP_FOOT_PAD;
3257 newp->head = (psize|CINUSE_BIT);
3258 mark_inuse_foot(m, newp, psize);
3259 chunk_plus_offset(newp, psize)->head = FENCEPOST_HEAD;
3260 chunk_plus_offset(newp, psize+SIZE_T_SIZE)->head = 0;
3262 if (cp < m->least_addr)
3264 if ((m->footprint += newmmsize - oldmmsize) > m->max_footprint)
3265 m->max_footprint = m->footprint;
3266 check_mmapped_chunk(m, newp);
3273 /* -------------------------- mspace management -------------------------- */
3275 /* Initialize top chunk and its size */
3276 static void init_top(mstate m, mchunkptr p, size_t psize) {
3277 /* Ensure alignment */
3278 size_t offset = align_offset(chunk2mem(p));
3279 p = (mchunkptr)((char*)p + offset);
3284 p->head = psize | PINUSE_BIT;
3285 /* set size of fake trailing chunk holding overhead space only once */
3286 chunk_plus_offset(p, psize)->head = TOP_FOOT_SIZE;
3287 m->trim_check = mparams.trim_threshold; /* reset on each update */
3290 /* Initialize bins for a new mstate that is otherwise zeroed out */
3291 static void init_bins(mstate m) {
3292 /* Establish circular links for smallbins */
3294 for (i = 0; i < NSMALLBINS; ++i) {
3295 sbinptr bin = smallbin_at(m,i);
3296 bin->fd = bin->bk = bin;
3300 #if PROCEED_ON_ERROR
3302 /* default corruption action */
3303 static void reset_on_error(mstate m) {
3305 ++malloc_corruption_error_count;
3306 /* Reinitialize fields to forget about all memory */
3307 m->smallbins = m->treebins = 0;
3308 m->dvsize = m->topsize = 0;
3313 for (i = 0; i < NTREEBINS; ++i)
3314 *treebin_at(m, i) = 0;
3317 #endif /* PROCEED_ON_ERROR */
3319 /* Allocate chunk and prepend remainder with chunk in successor base. */
3320 static void* prepend_alloc(mstate m, char* newbase, char* oldbase,
3322 mchunkptr p = align_as_chunk(newbase);
3323 mchunkptr oldfirst = align_as_chunk(oldbase);
3324 size_t psize = (char*)oldfirst - (char*)p;
3325 mchunkptr q = chunk_plus_offset(p, nb);
3326 size_t qsize = psize - nb;
3327 set_size_and_pinuse_of_inuse_chunk(m, p, nb);
3329 assert((char*)oldfirst > (char*)q);
3330 assert(pinuse(oldfirst));
3331 assert(qsize >= MIN_CHUNK_SIZE);
3333 /* consolidate remainder with first chunk of old base */
3334 if (oldfirst == m->top) {
3335 size_t tsize = m->topsize += qsize;
3337 q->head = tsize | PINUSE_BIT;
3338 check_top_chunk(m, q);
3340 else if (oldfirst == m->dv) {
3341 size_t dsize = m->dvsize += qsize;
3343 set_size_and_pinuse_of_free_chunk(q, dsize);
3346 if (!cinuse(oldfirst)) {
3347 size_t nsize = chunksize(oldfirst);
3348 unlink_chunk(m, oldfirst, nsize);
3349 oldfirst = chunk_plus_offset(oldfirst, nsize);
3352 set_free_with_pinuse(q, qsize, oldfirst);
3353 insert_chunk(m, q, qsize);
3354 check_free_chunk(m, q);
3357 check_malloced_chunk(m, chunk2mem(p), nb);
3358 return chunk2mem(p);
3362 /* Add a segment to hold a new noncontiguous region */
3363 static void add_segment(mstate m, char* tbase, size_t tsize, flag_t mmapped) {
3364 /* Determine locations and sizes of segment, fenceposts, old top */
3365 char* old_top = (char*)m->top;
3366 msegmentptr oldsp = segment_holding(m, old_top);
3367 char* old_end = oldsp->base + oldsp->size;
3368 size_t ssize = pad_request(sizeof(struct malloc_segment));
3369 char* rawsp = old_end - (ssize + FOUR_SIZE_T_SIZES + CHUNK_ALIGN_MASK);
3370 size_t offset = align_offset(chunk2mem(rawsp));
3371 char* asp = rawsp + offset;
3372 char* csp = (asp < (old_top + MIN_CHUNK_SIZE))? old_top : asp;
3373 mchunkptr sp = (mchunkptr)csp;
3374 msegmentptr ss = (msegmentptr)(chunk2mem(sp));
3375 mchunkptr tnext = chunk_plus_offset(sp, ssize);
3376 mchunkptr p = tnext;
3379 /* reset top to new space */
3380 init_top(m, (mchunkptr)tbase, tsize - TOP_FOOT_SIZE);
3382 /* Set up segment record */
3383 assert(is_aligned(ss));
3384 set_size_and_pinuse_of_inuse_chunk(m, sp, ssize);
3385 *ss = m->seg; /* Push current record */
3386 m->seg.base = tbase;
3387 m->seg.size = tsize;
3388 set_segment_flags(&m->seg, mmapped);
3391 /* Insert trailing fenceposts */
3393 mchunkptr nextp = chunk_plus_offset(p, SIZE_T_SIZE);
3394 p->head = FENCEPOST_HEAD;
3396 if ((char*)(&(nextp->head)) < old_end)
3401 assert(nfences >= 2);
3403 /* Insert the rest of old top into a bin as an ordinary free chunk */
3404 if (csp != old_top) {
3405 mchunkptr q = (mchunkptr)old_top;
3406 size_t psize = csp - old_top;
3407 mchunkptr tn = chunk_plus_offset(q, psize);
3408 set_free_with_pinuse(q, psize, tn);
3409 insert_chunk(m, q, psize);
3412 check_top_chunk(m, m->top);
3415 /* -------------------------- System allocation -------------------------- */
3417 /* Get memory from system using MORECORE or MMAP */
3418 static void* sys_alloc(mstate m, size_t nb) {
3419 char* tbase = CMFAIL;
3421 flag_t mmap_flag = 0;
3425 /* Directly map large chunks */
3426 if (use_mmap(m) && nb >= mparams.mmap_threshold) {
3427 void* mem = mmap_alloc(m, nb);
3433 Try getting memory in any of three ways (in most-preferred to
3434 least-preferred order):
3435 1. A call to MORECORE that can normally contiguously extend memory.
3436 (disabled if not MORECORE_CONTIGUOUS or not HAVE_MORECORE or
3437 or main space is mmapped or a previous contiguous call failed)
3438 2. A call to MMAP new space (disabled if not HAVE_MMAP).
3439 Note that under the default settings, if MORECORE is unable to
3440 fulfill a request, and HAVE_MMAP is true, then mmap is
3441 used as a noncontiguous system allocator. This is a useful backup
3442 strategy for systems with holes in address spaces -- in this case
3443 sbrk cannot contiguously expand the heap, but mmap may be able to
3445 3. A call to MORECORE that cannot usually contiguously extend memory.
3446 (disabled if not HAVE_MORECORE)
3449 if (MORECORE_CONTIGUOUS && !use_noncontiguous(m)) {
3451 msegmentptr ss = (m->top == 0)? 0 : segment_holding(m, (char*)m->top);
3453 ACQUIRE_MORECORE_LOCK();
3455 if (ss == 0) { /* First time through or recovery */
3456 char* base = (char*)CALL_MORECORE(0);
3457 if (base != CMFAIL) {
3458 asize = granularity_align(nb + TOP_FOOT_SIZE + SIZE_T_ONE);
3459 /* Adjust to end on a page boundary */
3460 if (!is_page_aligned(base))
3461 asize += (page_align((size_t)base) - (size_t)base);
3462 /* Can't call MORECORE if size is negative when treated as signed */
3463 if (asize < HALF_MAX_SIZE_T &&
3464 (br = (char*)(CALL_MORECORE(asize))) == base) {
3471 /* Subtract out existing available top space from MORECORE request. */
3472 asize = granularity_align(nb - m->topsize + TOP_FOOT_SIZE + SIZE_T_ONE);
3473 /* Use mem here only if it did continuously extend old space */
3474 if (asize < HALF_MAX_SIZE_T &&
3475 (br = (char*)(CALL_MORECORE(asize))) == ss->base+ss->size) {
3481 if (tbase == CMFAIL) { /* Cope with partial failure */
3482 if (br != CMFAIL) { /* Try to use/extend the space we did get */
3483 if (asize < HALF_MAX_SIZE_T &&
3484 asize < nb + TOP_FOOT_SIZE + SIZE_T_ONE) {
3485 size_t esize = granularity_align(nb + TOP_FOOT_SIZE + SIZE_T_ONE - asize);
3486 if (esize < HALF_MAX_SIZE_T) {
3487 char* end = (char*)CALL_MORECORE(esize);
3490 else { /* Can't use; try to release */
3491 (void)CALL_MORECORE(-asize);
3497 if (br != CMFAIL) { /* Use the space we did get */
3502 disable_contiguous(m); /* Don't try contiguous path in the future */
3505 RELEASE_MORECORE_LOCK();
3508 if (HAVE_MMAP && tbase == CMFAIL) { /* Try MMAP */
3509 size_t req = nb + TOP_FOOT_SIZE + SIZE_T_ONE;
3510 size_t rsize = granularity_align(req);
3511 if (rsize > nb) { /* Fail if wraps around zero */
3512 char* mp = (char*)(CALL_MMAP(rsize));
3516 mmap_flag = IS_MMAPPED_BIT;
3521 if (HAVE_MORECORE && tbase == CMFAIL) { /* Try noncontiguous MORECORE */
3522 size_t asize = granularity_align(nb + TOP_FOOT_SIZE + SIZE_T_ONE);
3523 if (asize < HALF_MAX_SIZE_T) {
3526 ACQUIRE_MORECORE_LOCK();
3527 br = (char*)(CALL_MORECORE(asize));
3528 end = (char*)(CALL_MORECORE(0));
3529 RELEASE_MORECORE_LOCK();
3530 if (br != CMFAIL && end != CMFAIL && br < end) {
3531 size_t ssize = end - br;
3532 if (ssize > nb + TOP_FOOT_SIZE) {
3540 if (tbase != CMFAIL) {
3542 if ((m->footprint += tsize) > m->max_footprint)
3543 m->max_footprint = m->footprint;
3545 if (!is_initialized(m)) { /* first-time initialization */
3546 m->seg.base = m->least_addr = tbase;
3547 m->seg.size = tsize;
3548 set_segment_flags(&m->seg, mmap_flag);
3549 m->magic = mparams.magic;
3552 init_top(m, (mchunkptr)tbase, tsize - TOP_FOOT_SIZE);
3554 /* Offset top by embedded malloc_state */
3555 mchunkptr mn = next_chunk(mem2chunk(m));
3556 init_top(m, mn, (size_t)((tbase + tsize) - (char*)mn) -TOP_FOOT_SIZE);
3561 /* Try to merge with an existing segment */
3562 msegmentptr sp = &m->seg;
3563 while (sp != 0 && tbase != sp->base + sp->size)
3566 !is_extern_segment(sp) &&
3567 check_segment_merge(sp, tbase, tsize) &&
3568 (get_segment_flags(sp) & IS_MMAPPED_BIT) == mmap_flag &&
3569 segment_holds(sp, m->top)) { /* append */
3571 init_top(m, m->top, m->topsize + tsize);
3574 if (tbase < m->least_addr)
3575 m->least_addr = tbase;
3577 while (sp != 0 && sp->base != tbase + tsize)
3580 !is_extern_segment(sp) &&
3581 check_segment_merge(sp, tbase, tsize) &&
3582 (get_segment_flags(sp) & IS_MMAPPED_BIT) == mmap_flag) {
3583 char* oldbase = sp->base;
3586 return prepend_alloc(m, tbase, oldbase, nb);
3589 add_segment(m, tbase, tsize, mmap_flag);
3593 if (nb < m->topsize) { /* Allocate from new or extended top space */
3594 size_t rsize = m->topsize -= nb;
3595 mchunkptr p = m->top;
3596 mchunkptr r = m->top = chunk_plus_offset(p, nb);
3597 r->head = rsize | PINUSE_BIT;
3598 set_size_and_pinuse_of_inuse_chunk(m, p, nb);
3599 check_top_chunk(m, m->top);
3600 check_malloced_chunk(m, chunk2mem(p), nb);
3601 return chunk2mem(p);
3605 MALLOC_FAILURE_ACTION;
3609 /* ----------------------- system deallocation -------------------------- */
3611 /* Unmap and unlink any mmapped segments that don't contain used chunks */
3612 static size_t release_unused_segments(mstate m) {
3613 size_t released = 0;
3614 msegmentptr pred = &m->seg;
3615 msegmentptr sp = pred->next;
3617 char* base = sp->base;
3618 size_t size = sp->size;
3619 msegmentptr next = sp->next;
3620 if (is_mmapped_segment(sp) && !is_extern_segment(sp)) {
3621 mchunkptr p = align_as_chunk(base);
3622 size_t psize = chunksize(p);
3623 /* Can unmap if first chunk holds entire segment and not pinned */
3624 if (!cinuse(p) && (char*)p + psize >= base + size - TOP_FOOT_SIZE) {
3625 tchunkptr tp = (tchunkptr)p;
3626 assert(segment_holds(sp, (char*)sp));
3632 unlink_large_chunk(m, tp);
3634 if (CALL_MUNMAP(base, size) == 0) {
3636 m->footprint -= size;
3637 /* unlink obsoleted record */
3641 else { /* back out if cannot unmap */
3642 insert_large_chunk(m, tp, psize);
3652 static int sys_trim(mstate m, size_t pad) {
3653 size_t released = 0;
3654 if (pad < MAX_REQUEST && is_initialized(m)) {
3655 pad += TOP_FOOT_SIZE; /* ensure enough room for segment overhead */
3657 if (m->topsize > pad) {
3658 /* Shrink top space in granularity-size units, keeping at least one */
3659 size_t unit = mparams.granularity;
3660 size_t extra = ((m->topsize - pad + (unit - SIZE_T_ONE)) / unit -
3662 msegmentptr sp = segment_holding(m, (char*)m->top);
3664 if (!is_extern_segment(sp)) {
3665 if (is_mmapped_segment(sp)) {
3667 sp->size >= extra &&
3668 !has_segment_link(m, sp)) { /* can't shrink if pinned */
3669 size_t newsize = sp->size - extra;
3670 /* Prefer mremap, fall back to munmap */
3671 if ((CALL_MREMAP(sp->base, sp->size, newsize, 0) != MFAIL) ||
3672 (CALL_MUNMAP(sp->base + newsize, extra) == 0)) {
3677 else if (HAVE_MORECORE) {
3678 if (extra >= HALF_MAX_SIZE_T) /* Avoid wrapping negative */
3679 extra = (HALF_MAX_SIZE_T) + SIZE_T_ONE - unit;
3680 ACQUIRE_MORECORE_LOCK();
3682 /* Make sure end of memory is where we last set it. */
3683 char* old_br = (char*)(CALL_MORECORE(0));
3684 if (old_br == sp->base + sp->size) {
3685 char* rel_br = (char*)(CALL_MORECORE(-extra));
3686 char* new_br = (char*)(CALL_MORECORE(0));
3687 if (rel_br != CMFAIL && new_br < old_br)
3688 released = old_br - new_br;
3691 RELEASE_MORECORE_LOCK();
3695 if (released != 0) {
3696 sp->size -= released;
3697 m->footprint -= released;
3698 init_top(m, m->top, m->topsize - released);
3699 check_top_chunk(m, m->top);
3703 /* Unmap any unused mmapped segments */
3705 released += release_unused_segments(m);
3707 /* On failure, disable autotrim to avoid repeated failed future calls */
3709 m->trim_check = MAX_SIZE_T;
3712 return (released != 0)? 1 : 0;
3715 /* ---------------------------- malloc support --------------------------- */
3717 /* allocate a large request from the best fitting chunk in a treebin */
3718 static void* tmalloc_large(mstate m, size_t nb) {
3720 size_t rsize = -nb; /* Unsigned negation */
3723 compute_tree_index(nb, idx);
3725 if ((t = *treebin_at(m, idx)) != 0) {
3726 /* Traverse tree for this bin looking for node with size == nb */
3727 size_t sizebits = nb << leftshift_for_tree_index(idx);
3728 tchunkptr rst = 0; /* The deepest untaken right subtree */
3731 size_t trem = chunksize(t) - nb;
3734 if ((rsize = trem) == 0)
3738 t = t->child[(sizebits >> (SIZE_T_BITSIZE-SIZE_T_ONE)) & 1];
3739 if (rt != 0 && rt != t)
3742 t = rst; /* set t to least subtree holding sizes > nb */
3749 if (t == 0 && v == 0) { /* set t to root of next non-empty treebin */
3750 binmap_t leftbits = left_bits(idx2bit(idx)) & m->treemap;
3751 if (leftbits != 0) {
3753 binmap_t leastbit = least_bit(leftbits);
3754 compute_bit2idx(leastbit, i);
3755 t = *treebin_at(m, i);
3759 while (t != 0) { /* find smallest of tree or subtree */
3760 size_t trem = chunksize(t) - nb;
3765 t = leftmost_child(t);
3768 /* If dv is a better fit, return 0 so malloc will use it */
3769 if (v != 0 && rsize < (size_t)(m->dvsize - nb)) {
3770 if (RTCHECK(ok_address(m, v))) { /* split */
3771 mchunkptr r = chunk_plus_offset(v, nb);
3772 assert(chunksize(v) == rsize + nb);
3773 if (RTCHECK(ok_next(v, r))) {
3774 unlink_large_chunk(m, v);
3775 if (rsize < MIN_CHUNK_SIZE)
3776 set_inuse_and_pinuse(m, v, (rsize + nb));
3778 set_size_and_pinuse_of_inuse_chunk(m, v, nb);
3779 set_size_and_pinuse_of_free_chunk(r, rsize);
3780 insert_chunk(m, r, rsize);
3782 return chunk2mem(v);
3785 CORRUPTION_ERROR_ACTION(m);
3790 /* allocate a small request from the best fitting chunk in a treebin */
3791 static void* tmalloc_small(mstate m, size_t nb) {
3795 binmap_t leastbit = least_bit(m->treemap);
3796 compute_bit2idx(leastbit, i);
3798 v = t = *treebin_at(m, i);
3799 rsize = chunksize(t) - nb;
3801 while ((t = leftmost_child(t)) != 0) {
3802 size_t trem = chunksize(t) - nb;
3809 if (RTCHECK(ok_address(m, v))) {
3810 mchunkptr r = chunk_plus_offset(v, nb);
3811 assert(chunksize(v) == rsize + nb);
3812 if (RTCHECK(ok_next(v, r))) {
3813 unlink_large_chunk(m, v);
3814 if (rsize < MIN_CHUNK_SIZE)
3815 set_inuse_and_pinuse(m, v, (rsize + nb));
3817 set_size_and_pinuse_of_inuse_chunk(m, v, nb);
3818 set_size_and_pinuse_of_free_chunk(r, rsize);
3819 replace_dv(m, r, rsize);
3821 return chunk2mem(v);
3825 CORRUPTION_ERROR_ACTION(m);
3829 /* --------------------------- realloc support --------------------------- */
3831 static void* internal_realloc(mstate m, void* oldmem, size_t bytes) {
3832 if (bytes >= MAX_REQUEST) {
3833 MALLOC_FAILURE_ACTION;
3836 if (!PREACTION(m)) {
3837 mchunkptr oldp = mem2chunk(oldmem);
3838 size_t oldsize = chunksize(oldp);
3839 mchunkptr next = chunk_plus_offset(oldp, oldsize);
3843 /* Try to either shrink or extend into top. Else malloc-copy-free */
3845 if (RTCHECK(ok_address(m, oldp) && ok_cinuse(oldp) &&
3846 ok_next(oldp, next) && ok_pinuse(next))) {
3847 size_t nb = request2size(bytes);
3848 if (is_mmapped(oldp))
3849 newp = mmap_resize(m, oldp, nb);
3850 else if (oldsize >= nb) { /* already big enough */
3851 size_t rsize = oldsize - nb;
3853 if (rsize >= MIN_CHUNK_SIZE) {
3854 mchunkptr remainder = chunk_plus_offset(newp, nb);
3855 set_inuse(m, newp, nb);
3856 set_inuse(m, remainder, rsize);
3857 extra = chunk2mem(remainder);
3860 else if (next == m->top && oldsize + m->topsize > nb) {
3861 /* Expand into top */
3862 size_t newsize = oldsize + m->topsize;
3863 size_t newtopsize = newsize - nb;
3864 mchunkptr newtop = chunk_plus_offset(oldp, nb);
3865 set_inuse(m, oldp, nb);
3866 newtop->head = newtopsize |PINUSE_BIT;
3868 m->topsize = newtopsize;
3873 USAGE_ERROR_ACTION(m, oldmem);
3882 internal_free(m, extra);
3884 check_inuse_chunk(m, newp);
3885 return chunk2mem(newp);
3888 void* newmem = internal_malloc(m, bytes);
3890 size_t oc = oldsize - overhead_for(oldp);
3891 memcpy(newmem, oldmem, (oc < bytes)? oc : bytes);
3892 internal_free(m, oldmem);
3900 /* --------------------------- memalign support -------------------------- */
3902 static void* internal_memalign(mstate m, size_t alignment, size_t bytes) {
3903 if (alignment <= MALLOC_ALIGNMENT) /* Can just use malloc */
3904 return internal_malloc(m, bytes);
3905 if (alignment < MIN_CHUNK_SIZE) /* must be at least a minimum chunk size */
3906 alignment = MIN_CHUNK_SIZE;
3907 if ((alignment & (alignment-SIZE_T_ONE)) != 0) {/* Ensure a power of 2 */
3908 size_t a = MALLOC_ALIGNMENT << 1;
3909 while (a < alignment) a <<= 1;
3913 if (bytes >= MAX_REQUEST - alignment) {
3914 if (m != 0) { /* Test isn't needed but avoids compiler warning */
3915 MALLOC_FAILURE_ACTION;
3919 size_t nb = request2size(bytes);
3920 size_t req = nb + alignment + MIN_CHUNK_SIZE - CHUNK_OVERHEAD;
3921 char* mem = (char*)internal_malloc(m, req);
3925 mchunkptr p = mem2chunk(mem);
3927 if (PREACTION(m)) return 0;
3928 if ((((size_t)(mem)) % alignment) != 0) { /* misaligned */
3930 Find an aligned spot inside chunk. Since we need to give
3931 back leading space in a chunk of at least MIN_CHUNK_SIZE, if
3932 the first calculation places us at a spot with less than
3933 MIN_CHUNK_SIZE leader, we can move to the next aligned spot.
3934 We've allocated enough total room so that this is always
3937 char* br = (char*)mem2chunk((size_t)(((size_t)(mem +
3941 char* pos = ((size_t)(br - (char*)(p)) >= MIN_CHUNK_SIZE)?
3943 mchunkptr newp = (mchunkptr)pos;
3944 size_t leadsize = pos - (char*)(p);
3945 size_t newsize = chunksize(p) - leadsize;
3947 if (is_mmapped(p)) { /* For mmapped chunks, just adjust offset */
3948 newp->prev_foot = p->prev_foot + leadsize;
3949 newp->head = (newsize|CINUSE_BIT);
3951 else { /* Otherwise, give back leader, use the rest */
3952 set_inuse(m, newp, newsize);
3953 set_inuse(m, p, leadsize);
3954 leader = chunk2mem(p);
3959 /* Give back spare room at the end */
3960 if (!is_mmapped(p)) {
3961 size_t size = chunksize(p);
3962 if (size > nb + MIN_CHUNK_SIZE) {
3963 size_t remainder_size = size - nb;
3964 mchunkptr remainder = chunk_plus_offset(p, nb);
3965 set_inuse(m, p, nb);
3966 set_inuse(m, remainder, remainder_size);
3967 trailer = chunk2mem(remainder);
3971 assert (chunksize(p) >= nb);
3972 assert((((size_t)(chunk2mem(p))) % alignment) == 0);
3973 check_inuse_chunk(m, p);
3976 internal_free(m, leader);
3979 internal_free(m, trailer);
3981 return chunk2mem(p);
3987 /* ------------------------ comalloc/coalloc support --------------------- */
3989 static void** ialloc(mstate m,
3995 This provides common support for independent_X routines, handling
3996 all of the combinations that can result.
3999 bit 0 set if all elements are same size (using sizes[0])
4000 bit 1 set if elements should be zeroed
4003 size_t element_size; /* chunksize of each element, if all same */
4004 size_t contents_size; /* total size of elements */
4005 size_t array_size; /* request size of pointer array */
4006 void* mem; /* malloced aggregate space */
4007 mchunkptr p; /* corresponding chunk */
4008 size_t remainder_size; /* remaining bytes while splitting */
4009 void** marray; /* either "chunks" or malloced ptr array */
4010 mchunkptr array_chunk; /* chunk for malloced ptr array */
4011 flag_t was_enabled; /* to disable mmap */
4015 /* compute array length, if needed */
4017 if (n_elements == 0)
4018 return chunks; /* nothing to do */
4023 /* if empty req, must still return chunk representing empty array */
4024 if (n_elements == 0)
4025 return (void**)internal_malloc(m, 0);
4027 array_size = request2size(n_elements * (sizeof(void*)));
4030 /* compute total element size */
4031 if (opts & 0x1) { /* all-same-size */
4032 element_size = request2size(*sizes);
4033 contents_size = n_elements * element_size;
4035 else { /* add up all the sizes */
4038 for (i = 0; i != n_elements; ++i)
4039 contents_size += request2size(sizes[i]);
4042 size = contents_size + array_size;
4045 Allocate the aggregate chunk. First disable direct-mmapping so
4046 malloc won't use it, since we would not be able to later
4047 free/realloc space internal to a segregated mmap region.
4049 was_enabled = use_mmap(m);
4051 mem = internal_malloc(m, size - CHUNK_OVERHEAD);
4057 if (PREACTION(m)) return 0;
4059 remainder_size = chunksize(p);
4061 assert(!is_mmapped(p));
4063 if (opts & 0x2) { /* optionally clear the elements */
4064 memset((size_t*)mem, 0, remainder_size - SIZE_T_SIZE - array_size);
4067 /* If not provided, allocate the pointer array as final part of chunk */
4069 size_t array_chunk_size;
4070 array_chunk = chunk_plus_offset(p, contents_size);
4071 array_chunk_size = remainder_size - contents_size;
4072 marray = (void**) (chunk2mem(array_chunk));
4073 set_size_and_pinuse_of_inuse_chunk(m, array_chunk, array_chunk_size);
4074 remainder_size = contents_size;
4077 /* split out elements */
4078 for (i = 0; ; ++i) {
4079 marray[i] = chunk2mem(p);
4080 if (i != n_elements-1) {
4081 if (element_size != 0)
4082 size = element_size;
4084 size = request2size(sizes[i]);
4085 remainder_size -= size;
4086 set_size_and_pinuse_of_inuse_chunk(m, p, size);
4087 p = chunk_plus_offset(p, size);
4089 else { /* the final element absorbs any overallocation slop */
4090 set_size_and_pinuse_of_inuse_chunk(m, p, remainder_size);
4096 if (marray != chunks) {
4097 /* final element must have exactly exhausted chunk */
4098 if (element_size != 0) {
4099 assert(remainder_size == element_size);
4102 assert(remainder_size == request2size(sizes[i]));
4104 check_inuse_chunk(m, mem2chunk(marray));
4106 for (i = 0; i != n_elements; ++i)
4107 check_inuse_chunk(m, mem2chunk(marray[i]));
4116 /* -------------------------- public routines ---------------------------- */
4120 void* dlmalloc(size_t bytes) {
4123 If a small request (< 256 bytes minus per-chunk overhead):
4124 1. If one exists, use a remainderless chunk in associated smallbin.
4125 (Remainderless means that there are too few excess bytes to
4126 represent as a chunk.)
4127 2. If it is big enough, use the dv chunk, which is normally the
4128 chunk adjacent to the one used for the most recent small request.
4129 3. If one exists, split the smallest available chunk in a bin,
4130 saving remainder in dv.
4131 4. If it is big enough, use the top chunk.
4132 5. If available, get memory from system and use it
4133 Otherwise, for a large request:
4134 1. Find the smallest available binned chunk that fits, and use it
4135 if it is better fitting than dv chunk, splitting if necessary.
4136 2. If better fitting than any binned chunk, use the dv chunk.
4137 3. If it is big enough, use the top chunk.
4138 4. If request size >= mmap threshold, try to directly mmap this chunk.
4139 5. If available, get memory from system and use it
4141 The ugly goto's here ensure that postaction occurs along all paths.
4144 if (!PREACTION(gm)) {
4147 if (bytes <= MAX_SMALL_REQUEST) {
4150 nb = (bytes < MIN_REQUEST)? MIN_CHUNK_SIZE : pad_request(bytes);
4151 idx = small_index(nb);
4152 smallbits = gm->smallmap >> idx;
4154 if ((smallbits & 0x3U) != 0) { /* Remainderless fit to a smallbin. */
4156 idx += ~smallbits & 1; /* Uses next bin if idx empty */
4157 b = smallbin_at(gm, idx);
4159 assert(chunksize(p) == small_index2size(idx));
4160 unlink_first_small_chunk(gm, b, p, idx);
4161 set_inuse_and_pinuse(gm, p, small_index2size(idx));
4163 check_malloced_chunk(gm, mem, nb);
4167 else if (nb > gm->dvsize) {
4168 if (smallbits != 0) { /* Use chunk in next nonempty smallbin */
4172 binmap_t leftbits = (smallbits << idx) & left_bits(idx2bit(idx));
4173 binmap_t leastbit = least_bit(leftbits);
4174 compute_bit2idx(leastbit, i);
4175 b = smallbin_at(gm, i);
4177 assert(chunksize(p) == small_index2size(i));
4178 unlink_first_small_chunk(gm, b, p, i);
4179 rsize = small_index2size(i) - nb;
4180 /* Fit here cannot be remainderless if 4byte sizes */
4181 if (SIZE_T_SIZE != 4 && rsize < MIN_CHUNK_SIZE)
4182 set_inuse_and_pinuse(gm, p, small_index2size(i));
4184 set_size_and_pinuse_of_inuse_chunk(gm, p, nb);
4185 r = chunk_plus_offset(p, nb);
4186 set_size_and_pinuse_of_free_chunk(r, rsize);
4187 replace_dv(gm, r, rsize);
4190 check_malloced_chunk(gm, mem, nb);
4194 else if (gm->treemap != 0 && (mem = tmalloc_small(gm, nb)) != 0) {
4195 check_malloced_chunk(gm, mem, nb);
4200 else if (bytes >= MAX_REQUEST)
4201 nb = MAX_SIZE_T; /* Too big to allocate. Force failure (in sys alloc) */
4203 nb = pad_request(bytes);
4204 if (gm->treemap != 0 && (mem = tmalloc_large(gm, nb)) != 0) {
4205 check_malloced_chunk(gm, mem, nb);
4210 if (nb <= gm->dvsize) {
4211 size_t rsize = gm->dvsize - nb;
4212 mchunkptr p = gm->dv;
4213 if (rsize >= MIN_CHUNK_SIZE) { /* split dv */
4214 mchunkptr r = gm->dv = chunk_plus_offset(p, nb);
4216 set_size_and_pinuse_of_free_chunk(r, rsize);
4217 set_size_and_pinuse_of_inuse_chunk(gm, p, nb);
4219 else { /* exhaust dv */
4220 size_t dvs = gm->dvsize;
4223 set_inuse_and_pinuse(gm, p, dvs);
4226 check_malloced_chunk(gm, mem, nb);
4230 else if (nb < gm->topsize) { /* Split top */
4231 size_t rsize = gm->topsize -= nb;
4232 mchunkptr p = gm->top;
4233 mchunkptr r = gm->top = chunk_plus_offset(p, nb);
4234 r->head = rsize | PINUSE_BIT;
4235 set_size_and_pinuse_of_inuse_chunk(gm, p, nb);
4237 check_top_chunk(gm, gm->top);
4238 check_malloced_chunk(gm, mem, nb);
4242 mem = sys_alloc(gm, nb);
4252 void dlfree(void* mem) {
4254 Consolidate freed chunks with preceding or succeeding bordering
4255 free chunks, if they exist, and then place in a bin. Intermixed
4256 with special cases for top, dv, mmapped chunks, and usage errors.
4260 mchunkptr p = mem2chunk(mem);
4262 mstate fm = get_mstate_for(p);
4263 if (!ok_magic(fm)) {
4264 USAGE_ERROR_ACTION(fm, p);
4269 #endif /* FOOTERS */
4270 if (!PREACTION(fm)) {
4271 check_inuse_chunk(fm, p);
4272 if (RTCHECK(ok_address(fm, p) && ok_cinuse(p))) {
4273 size_t psize = chunksize(p);
4274 mchunkptr next = chunk_plus_offset(p, psize);
4276 size_t prevsize = p->prev_foot;
4277 if ((prevsize & IS_MMAPPED_BIT) != 0) {
4278 prevsize &= ~IS_MMAPPED_BIT;
4279 psize += prevsize + MMAP_FOOT_PAD;
4280 if (CALL_MUNMAP((char*)p - prevsize, psize) == 0)
4281 fm->footprint -= psize;
4285 mchunkptr prev = chunk_minus_offset(p, prevsize);
4288 if (RTCHECK(ok_address(fm, prev))) { /* consolidate backward */
4290 unlink_chunk(fm, p, prevsize);
4292 else if ((next->head & INUSE_BITS) == INUSE_BITS) {
4294 set_free_with_pinuse(p, psize, next);
4303 if (RTCHECK(ok_next(p, next) && ok_pinuse(next))) {
4304 if (!cinuse(next)) { /* consolidate forward */
4305 if (next == fm->top) {
4306 size_t tsize = fm->topsize += psize;
4308 p->head = tsize | PINUSE_BIT;
4313 if (should_trim(fm, tsize))
4317 else if (next == fm->dv) {
4318 size_t dsize = fm->dvsize += psize;
4320 set_size_and_pinuse_of_free_chunk(p, dsize);
4324 size_t nsize = chunksize(next);
4326 unlink_chunk(fm, next, nsize);
4327 set_size_and_pinuse_of_free_chunk(p, psize);
4335 set_free_with_pinuse(p, psize, next);
4336 insert_chunk(fm, p, psize);
4337 check_free_chunk(fm, p);
4342 USAGE_ERROR_ACTION(fm, p);
4349 #endif /* FOOTERS */
4352 void* dlcalloc(size_t n_elements, size_t elem_size) {
4355 if (n_elements != 0) {
4356 req = n_elements * elem_size;
4357 if (((n_elements | elem_size) & ~(size_t)0xffff) &&
4358 (req / n_elements != elem_size))
4359 req = MAX_SIZE_T; /* force downstream failure on overflow */
4361 mem = dlmalloc(req);
4362 if (mem != 0 && calloc_must_clear(mem2chunk(mem)))
4363 memset(mem, 0, req);
4367 void* dlrealloc(void* oldmem, size_t bytes) {
4369 return dlmalloc(bytes);
4370 #ifdef REALLOC_ZERO_BYTES_FREES
4375 #endif /* REALLOC_ZERO_BYTES_FREES */
4380 mstate m = get_mstate_for(mem2chunk(oldmem));
4382 USAGE_ERROR_ACTION(m, oldmem);
4385 #endif /* FOOTERS */
4386 return internal_realloc(m, oldmem, bytes);
4390 void* dlmemalign(size_t alignment, size_t bytes) {
4391 return internal_memalign(gm, alignment, bytes);
4394 void** dlindependent_calloc(size_t n_elements, size_t elem_size,
4396 size_t sz = elem_size; /* serves as 1-element array */
4397 return ialloc(gm, n_elements, &sz, 3, chunks);
4400 void** dlindependent_comalloc(size_t n_elements, size_t sizes[],
4402 return ialloc(gm, n_elements, sizes, 0, chunks);
4405 void* dlvalloc(size_t bytes) {
4408 pagesz = mparams.page_size;
4409 return dlmemalign(pagesz, bytes);
4412 void* dlpvalloc(size_t bytes) {
4415 pagesz = mparams.page_size;
4416 return dlmemalign(pagesz, (bytes + pagesz - SIZE_T_ONE) & ~(pagesz - SIZE_T_ONE));
4419 int dlmalloc_trim(size_t pad) {
4421 if (!PREACTION(gm)) {
4422 result = sys_trim(gm, pad);
4428 size_t dlmalloc_footprint(void) {
4429 return gm->footprint;
4432 size_t dlmalloc_max_footprint(void) {
4433 return gm->max_footprint;
4437 struct mallinfo dlmallinfo(void) {
4438 return internal_mallinfo(gm);
4440 #endif /* NO_MALLINFO */
4442 void dlmalloc_stats() {
4443 internal_malloc_stats(gm);
4446 size_t dlmalloc_usable_size(void* mem) {
4448 mchunkptr p = mem2chunk(mem);
4450 return chunksize(p) - overhead_for(p);
4455 int dlmallopt(int param_number, int value) {
4456 return change_mparam(param_number, value);
4459 #endif /* !ONLY_MSPACES */
4461 /* ----------------------------- user mspaces ---------------------------- */
4465 static mstate init_user_mstate(char* tbase, size_t tsize) {
4466 size_t msize = pad_request(sizeof(struct malloc_state));
4468 mchunkptr msp = align_as_chunk(tbase);
4469 mstate m = (mstate)(chunk2mem(msp));
4470 memset(m, 0, msize);
4471 INITIAL_LOCK(&m->mutex);
4472 msp->head = (msize|PINUSE_BIT|CINUSE_BIT);
4473 m->seg.base = m->least_addr = tbase;
4474 m->seg.size = m->footprint = m->max_footprint = tsize;
4475 m->magic = mparams.magic;
4476 m->mflags = mparams.default_mflags;
4477 disable_contiguous(m);
4479 mn = next_chunk(mem2chunk(m));
4480 init_top(m, mn, (size_t)((tbase + tsize) - (char*)mn) - TOP_FOOT_SIZE);
4481 check_top_chunk(m, m->top);
4485 mspace create_mspace(size_t capacity, int locked) {
4487 size_t msize = pad_request(sizeof(struct malloc_state));
4488 init_mparams(); /* Ensure pagesize etc initialized */
4490 if (capacity < (size_t) -(msize + TOP_FOOT_SIZE + mparams.page_size)) {
4491 size_t rs = ((capacity == 0)? mparams.granularity :
4492 (capacity + TOP_FOOT_SIZE + msize));
4493 size_t tsize = granularity_align(rs);
4494 char* tbase = (char*)(CALL_MMAP(tsize));
4495 if (tbase != CMFAIL) {
4496 m = init_user_mstate(tbase, tsize);
4497 set_segment_flags(&m->seg, IS_MMAPPED_BIT);
4498 set_lock(m, locked);
4504 mspace create_mspace_with_base(void* base, size_t capacity, int locked) {
4506 size_t msize = pad_request(sizeof(struct malloc_state));
4507 init_mparams(); /* Ensure pagesize etc initialized */
4509 if (capacity > msize + TOP_FOOT_SIZE &&
4510 capacity < (size_t) -(msize + TOP_FOOT_SIZE + mparams.page_size)) {
4511 m = init_user_mstate((char*)base, capacity);
4512 set_segment_flags(&m->seg, EXTERN_BIT);
4513 set_lock(m, locked);
4518 size_t destroy_mspace(mspace msp) {
4520 mstate ms = (mstate)msp;
4522 msegmentptr sp = &ms->seg;
4524 char* base = sp->base;
4525 size_t size = sp->size;
4526 flag_t flag = get_segment_flags(sp);
4528 if ((flag & IS_MMAPPED_BIT) && !(flag & EXTERN_BIT) &&
4529 CALL_MUNMAP(base, size) == 0)
4534 USAGE_ERROR_ACTION(ms,ms);
4540 mspace versions of routines are near-clones of the global
4541 versions. This is not so nice but better than the alternatives.
4545 void* mspace_malloc(mspace msp, size_t bytes) {
4546 mstate ms = (mstate)msp;
4547 if (!ok_magic(ms)) {
4548 USAGE_ERROR_ACTION(ms,ms);
4551 if (!PREACTION(ms)) {
4554 if (bytes <= MAX_SMALL_REQUEST) {
4557 nb = (bytes < MIN_REQUEST)? MIN_CHUNK_SIZE : pad_request(bytes);
4558 idx = small_index(nb);
4559 smallbits = ms->smallmap >> idx;
4561 if ((smallbits & 0x3U) != 0) { /* Remainderless fit to a smallbin. */
4563 idx += ~smallbits & 1; /* Uses next bin if idx empty */
4564 b = smallbin_at(ms, idx);
4566 assert(chunksize(p) == small_index2size(idx));
4567 unlink_first_small_chunk(ms, b, p, idx);
4568 set_inuse_and_pinuse(ms, p, small_index2size(idx));
4570 check_malloced_chunk(ms, mem, nb);
4574 else if (nb > ms->dvsize) {
4575 if (smallbits != 0) { /* Use chunk in next nonempty smallbin */
4579 binmap_t leftbits = (smallbits << idx) & left_bits(idx2bit(idx));
4580 binmap_t leastbit = least_bit(leftbits);
4581 compute_bit2idx(leastbit, i);
4582 b = smallbin_at(ms, i);
4584 assert(chunksize(p) == small_index2size(i));
4585 unlink_first_small_chunk(ms, b, p, i);
4586 rsize = small_index2size(i) - nb;
4587 /* Fit here cannot be remainderless if 4byte sizes */
4588 if (SIZE_T_SIZE != 4 && rsize < MIN_CHUNK_SIZE)
4589 set_inuse_and_pinuse(ms, p, small_index2size(i));
4591 set_size_and_pinuse_of_inuse_chunk(ms, p, nb);
4592 r = chunk_plus_offset(p, nb);
4593 set_size_and_pinuse_of_free_chunk(r, rsize);
4594 replace_dv(ms, r, rsize);
4597 check_malloced_chunk(ms, mem, nb);
4601 else if (ms->treemap != 0 && (mem = tmalloc_small(ms, nb)) != 0) {
4602 check_malloced_chunk(ms, mem, nb);
4607 else if (bytes >= MAX_REQUEST)
4608 nb = MAX_SIZE_T; /* Too big to allocate. Force failure (in sys alloc) */
4610 nb = pad_request(bytes);
4611 if (ms->treemap != 0 && (mem = tmalloc_large(ms, nb)) != 0) {
4612 check_malloced_chunk(ms, mem, nb);
4617 if (nb <= ms->dvsize) {
4618 size_t rsize = ms->dvsize - nb;
4619 mchunkptr p = ms->dv;
4620 if (rsize >= MIN_CHUNK_SIZE) { /* split dv */
4621 mchunkptr r = ms->dv = chunk_plus_offset(p, nb);
4623 set_size_and_pinuse_of_free_chunk(r, rsize);
4624 set_size_and_pinuse_of_inuse_chunk(ms, p, nb);
4626 else { /* exhaust dv */
4627 size_t dvs = ms->dvsize;
4630 set_inuse_and_pinuse(ms, p, dvs);
4633 check_malloced_chunk(ms, mem, nb);
4637 else if (nb < ms->topsize) { /* Split top */
4638 size_t rsize = ms->topsize -= nb;
4639 mchunkptr p = ms->top;
4640 mchunkptr r = ms->top = chunk_plus_offset(p, nb);
4641 r->head = rsize | PINUSE_BIT;
4642 set_size_and_pinuse_of_inuse_chunk(ms, p, nb);
4644 check_top_chunk(ms, ms->top);
4645 check_malloced_chunk(ms, mem, nb);
4649 mem = sys_alloc(ms, nb);
4659 void mspace_free(mspace msp, void* mem) {
4661 mchunkptr p = mem2chunk(mem);
4663 mstate fm = get_mstate_for(p);
4665 mstate fm = (mstate)msp;
4666 #endif /* FOOTERS */
4667 if (!ok_magic(fm)) {
4668 USAGE_ERROR_ACTION(fm, p);
4671 if (!PREACTION(fm)) {
4672 check_inuse_chunk(fm, p);
4673 if (RTCHECK(ok_address(fm, p) && ok_cinuse(p))) {
4674 size_t psize = chunksize(p);
4675 mchunkptr next = chunk_plus_offset(p, psize);
4677 size_t prevsize = p->prev_foot;
4678 if ((prevsize & IS_MMAPPED_BIT) != 0) {
4679 prevsize &= ~IS_MMAPPED_BIT;
4680 psize += prevsize + MMAP_FOOT_PAD;
4681 if (CALL_MUNMAP((char*)p - prevsize, psize) == 0)
4682 fm->footprint -= psize;
4686 mchunkptr prev = chunk_minus_offset(p, prevsize);
4689 if (RTCHECK(ok_address(fm, prev))) { /* consolidate backward */
4691 unlink_chunk(fm, p, prevsize);
4693 else if ((next->head & INUSE_BITS) == INUSE_BITS) {
4695 set_free_with_pinuse(p, psize, next);
4704 if (RTCHECK(ok_next(p, next) && ok_pinuse(next))) {
4705 if (!cinuse(next)) { /* consolidate forward */
4706 if (next == fm->top) {
4707 size_t tsize = fm->topsize += psize;
4709 p->head = tsize | PINUSE_BIT;
4714 if (should_trim(fm, tsize))
4718 else if (next == fm->dv) {
4719 size_t dsize = fm->dvsize += psize;
4721 set_size_and_pinuse_of_free_chunk(p, dsize);
4725 size_t nsize = chunksize(next);
4727 unlink_chunk(fm, next, nsize);
4728 set_size_and_pinuse_of_free_chunk(p, psize);
4736 set_free_with_pinuse(p, psize, next);
4737 insert_chunk(fm, p, psize);
4738 check_free_chunk(fm, p);
4743 USAGE_ERROR_ACTION(fm, p);
4750 void* mspace_calloc(mspace msp, size_t n_elements, size_t elem_size) {
4753 mstate ms = (mstate)msp;
4754 if (!ok_magic(ms)) {
4755 USAGE_ERROR_ACTION(ms,ms);
4758 if (n_elements != 0) {
4759 req = n_elements * elem_size;
4760 if (((n_elements | elem_size) & ~(size_t)0xffff) &&
4761 (req / n_elements != elem_size))
4762 req = MAX_SIZE_T; /* force downstream failure on overflow */
4764 mem = internal_malloc(ms, req);
4765 if (mem != 0 && calloc_must_clear(mem2chunk(mem)))
4766 memset(mem, 0, req);
4770 void* mspace_realloc(mspace msp, void* oldmem, size_t bytes) {
4772 return mspace_malloc(msp, bytes);
4773 #ifdef REALLOC_ZERO_BYTES_FREES
4775 mspace_free(msp, oldmem);
4778 #endif /* REALLOC_ZERO_BYTES_FREES */
4781 mchunkptr p = mem2chunk(oldmem);
4782 mstate ms = get_mstate_for(p);
4784 mstate ms = (mstate)msp;
4785 #endif /* FOOTERS */
4786 if (!ok_magic(ms)) {
4787 USAGE_ERROR_ACTION(ms,ms);
4790 return internal_realloc(ms, oldmem, bytes);
4794 void* mspace_memalign(mspace msp, size_t alignment, size_t bytes) {
4795 mstate ms = (mstate)msp;
4796 if (!ok_magic(ms)) {
4797 USAGE_ERROR_ACTION(ms,ms);
4800 return internal_memalign(ms, alignment, bytes);
4803 void** mspace_independent_calloc(mspace msp, size_t n_elements,
4804 size_t elem_size, void* chunks[]) {
4805 size_t sz = elem_size; /* serves as 1-element array */
4806 mstate ms = (mstate)msp;
4807 if (!ok_magic(ms)) {
4808 USAGE_ERROR_ACTION(ms,ms);
4811 return ialloc(ms, n_elements, &sz, 3, chunks);
4814 void** mspace_independent_comalloc(mspace msp, size_t n_elements,
4815 size_t sizes[], void* chunks[]) {
4816 mstate ms = (mstate)msp;
4817 if (!ok_magic(ms)) {
4818 USAGE_ERROR_ACTION(ms,ms);
4821 return ialloc(ms, n_elements, sizes, 0, chunks);
4824 int mspace_trim(mspace msp, size_t pad) {
4826 mstate ms = (mstate)msp;
4828 if (!PREACTION(ms)) {
4829 result = sys_trim(ms, pad);
4834 USAGE_ERROR_ACTION(ms,ms);
4839 void mspace_malloc_stats(mspace msp) {
4840 mstate ms = (mstate)msp;
4842 internal_malloc_stats(ms);
4845 USAGE_ERROR_ACTION(ms,ms);
4849 size_t mspace_footprint(mspace msp) {
4851 mstate ms = (mstate)msp;
4853 result = ms->footprint;
4855 USAGE_ERROR_ACTION(ms,ms);
4860 size_t mspace_max_footprint(mspace msp) {
4862 mstate ms = (mstate)msp;
4864 result = ms->max_footprint;
4866 USAGE_ERROR_ACTION(ms,ms);
4872 struct mallinfo mspace_mallinfo(mspace msp) {
4873 mstate ms = (mstate)msp;
4874 if (!ok_magic(ms)) {
4875 USAGE_ERROR_ACTION(ms,ms);
4877 return internal_mallinfo(ms);
4879 #endif /* NO_MALLINFO */
4881 int mspace_mallopt(int param_number, int value) {
4882 return change_mparam(param_number, value);
4885 #endif /* MSPACES */
4887 /* -------------------- Alternative MORECORE functions ------------------- */
4890 Guidelines for creating a custom version of MORECORE:
4892 * For best performance, MORECORE should allocate in multiples of pagesize.
4893 * MORECORE may allocate more memory than requested. (Or even less,
4894 but this will usually result in a malloc failure.)
4895 * MORECORE must not allocate memory when given argument zero, but
4896 instead return one past the end address of memory from previous
4898 * For best performance, consecutive calls to MORECORE with positive
4899 arguments should return increasing addresses, indicating that
4900 space has been contiguously extended.
4901 * Even though consecutive calls to MORECORE need not return contiguous
4902 addresses, it must be OK for malloc'ed chunks to span multiple
4903 regions in those cases where they do happen to be contiguous.
4904 * MORECORE need not handle negative arguments -- it may instead
4905 just return MFAIL when given negative arguments.
4906 Negative arguments are always multiples of pagesize. MORECORE
4907 must not misinterpret negative args as large positive unsigned
4908 args. You can suppress all such calls from even occurring by defining
4909 MORECORE_CANNOT_TRIM,
4911 As an example alternative MORECORE, here is a custom allocator
4912 kindly contributed for pre-OSX macOS. It uses virtually but not
4913 necessarily physically contiguous non-paged memory (locked in,
4914 present and won't get swapped out). You can use it by uncommenting
4915 this section, adding some #includes, and setting up the appropriate
4918 #define MORECORE osMoreCore
4920 There is also a shutdown routine that should somehow be called for
4921 cleanup upon program exit.
4923 #define MAX_POOL_ENTRIES 100
4924 #define MINIMUM_MORECORE_SIZE (64 * 1024U)
4925 static int next_os_pool;
4926 void *our_os_pools[MAX_POOL_ENTRIES];
4928 void *osMoreCore(int size)
4931 static void *sbrk_top = 0;
4935 if (size < MINIMUM_MORECORE_SIZE)
4936 size = MINIMUM_MORECORE_SIZE;
4937 if (CurrentExecutionLevel() == kTaskLevel)
4938 ptr = PoolAllocateResident(size + RM_PAGE_SIZE, 0);
4941 return (void *) MFAIL;
4943 // save ptrs so they can be freed during cleanup
4944 our_os_pools[next_os_pool] = ptr;
4946 ptr = (void *) ((((size_t) ptr) + RM_PAGE_MASK) & ~RM_PAGE_MASK);
4947 sbrk_top = (char *) ptr + size;
4952 // we don't currently support shrink behavior
4953 return (void *) MFAIL;
4961 // cleanup any allocated memory pools
4962 // called as last thing before shutting down driver
4964 void osCleanupMem(void)
4968 for (ptr = our_os_pools; ptr < &our_os_pools[MAX_POOL_ENTRIES]; ptr++)
4971 PoolDeallocate(*ptr);
4979 /* -----------------------------------------------------------------------
4981 V2.8.3 Thu Sep 22 11:16:32 2005 Doug Lea (dl at gee)
4982 * Add max_footprint functions
4983 * Ensure all appropriate literals are size_t
4984 * Fix conditional compilation problem for some #define settings
4985 * Avoid concatenating segments with the one provided
4986 in create_mspace_with_base
4987 * Rename some variables to avoid compiler shadowing warnings
4988 * Use explicit lock initialization.
4989 * Better handling of sbrk interference.
4990 * Simplify and fix segment insertion, trimming and mspace_destroy
4991 * Reinstate REALLOC_ZERO_BYTES_FREES option from 2.7.x
4992 * Thanks especially to Dennis Flanagan for help on these.
4994 V2.8.2 Sun Jun 12 16:01:10 2005 Doug Lea (dl at gee)
4995 * Fix memalign brace error.
4997 V2.8.1 Wed Jun 8 16:11:46 2005 Doug Lea (dl at gee)
4998 * Fix improper #endif nesting in C++
4999 * Add explicit casts needed for C++
5001 V2.8.0 Mon May 30 14:09:02 2005 Doug Lea (dl at gee)
5002 * Use trees for large bins
5004 * Use segments to unify sbrk-based and mmap-based system allocation,
5005 removing need for emulation on most platforms without sbrk.
5006 * Default safety checks
5007 * Optional footer checks. Thanks to William Robertson for the idea.
5008 * Internal code refactoring
5009 * Incorporate suggestions and platform-specific changes.
5010 Thanks to Dennis Flanagan, Colin Plumb, Niall Douglas,
5011 Aaron Bachmann, Emery Berger, and others.
5012 * Speed up non-fastbin processing enough to remove fastbins.
5013 * Remove useless cfree() to avoid conflicts with other apps.
5014 * Remove internal memcpy, memset. Compilers handle builtins better.
5015 * Remove some options that no one ever used and rename others.
5017 V2.7.2 Sat Aug 17 09:07:30 2002 Doug Lea (dl at gee)
5018 * Fix malloc_state bitmap array misdeclaration
5020 V2.7.1 Thu Jul 25 10:58:03 2002 Doug Lea (dl at gee)
5021 * Allow tuning of FIRST_SORTED_BIN_SIZE
5022 * Use PTR_UINT as type for all ptr->int casts. Thanks to John Belmonte.
5023 * Better detection and support for non-contiguousness of MORECORE.
5024 Thanks to Andreas Mueller, Conal Walsh, and Wolfram Gloger
5025 * Bypass most of malloc if no frees. Thanks To Emery Berger.
5026 * Fix freeing of old top non-contiguous chunk im sysmalloc.
5027 * Raised default trim and map thresholds to 256K.
5028 * Fix mmap-related #defines. Thanks to Lubos Lunak.
5029 * Fix copy macros; added LACKS_FCNTL_H. Thanks to Neal Walfield.
5030 * Branch-free bin calculation
5031 * Default trim and mmap thresholds now 256K.
5033 V2.7.0 Sun Mar 11 14:14:06 2001 Doug Lea (dl at gee)
5034 * Introduce independent_comalloc and independent_calloc.
5035 Thanks to Michael Pachos for motivation and help.
5036 * Make optional .h file available
5037 * Allow > 2GB requests on 32bit systems.
5038 * new WIN32 sbrk, mmap, munmap, lock code from <Walter@GeNeSys-e.de>.
5039 Thanks also to Andreas Mueller <a.mueller at paradatec.de>,
5041 * Allow override of MALLOC_ALIGNMENT (Thanks to Ruud Waij for
5043 * memalign: check alignment arg
5044 * realloc: don't try to shift chunks backwards, since this
5045 leads to more fragmentation in some programs and doesn't
5046 seem to help in any others.
5047 * Collect all cases in malloc requiring system memory into sysmalloc
5048 * Use mmap as backup to sbrk
5049 * Place all internal state in malloc_state
5050 * Introduce fastbins (although similar to 2.5.1)
5051 * Many minor tunings and cosmetic improvements
5052 * Introduce USE_PUBLIC_MALLOC_WRAPPERS, USE_MALLOC_LOCK
5053 * Introduce MALLOC_FAILURE_ACTION, MORECORE_CONTIGUOUS
5054 Thanks to Tony E. Bennett <tbennett@nvidia.com> and others.
5055 * Include errno.h to support default failure action.
5057 V2.6.6 Sun Dec 5 07:42:19 1999 Doug Lea (dl at gee)
5058 * return null for negative arguments
5059 * Added Several WIN32 cleanups from Martin C. Fong <mcfong at yahoo.com>
5060 * Add 'LACKS_SYS_PARAM_H' for those systems without 'sys/param.h'
5061 (e.g. WIN32 platforms)
5062 * Cleanup header file inclusion for WIN32 platforms
5063 * Cleanup code to avoid Microsoft Visual C++ compiler complaints
5064 * Add 'USE_DL_PREFIX' to quickly allow co-existence with existing
5065 memory allocation routines
5066 * Set 'malloc_getpagesize' for WIN32 platforms (needs more work)
5067 * Use 'assert' rather than 'ASSERT' in WIN32 code to conform to
5068 usage of 'assert' in non-WIN32 code
5069 * Improve WIN32 'sbrk()' emulation's 'findRegion()' routine to
5071 * Always call 'fREe()' rather than 'free()'
5073 V2.6.5 Wed Jun 17 15:57:31 1998 Doug Lea (dl at gee)
5074 * Fixed ordering problem with boundary-stamping
5076 V2.6.3 Sun May 19 08:17:58 1996 Doug Lea (dl at gee)
5077 * Added pvalloc, as recommended by H.J. Liu
5078 * Added 64bit pointer support mainly from Wolfram Gloger
5079 * Added anonymously donated WIN32 sbrk emulation
5080 * Malloc, calloc, getpagesize: add optimizations from Raymond Nijssen
5081 * malloc_extend_top: fix mask error that caused wastage after
5083 * Add linux mremap support code from HJ Liu
5085 V2.6.2 Tue Dec 5 06:52:55 1995 Doug Lea (dl at gee)
5086 * Integrated most documentation with the code.
5087 * Add support for mmap, with help from
5088 Wolfram Gloger (Gloger@lrz.uni-muenchen.de).
5089 * Use last_remainder in more cases.
5090 * Pack bins using idea from colin@nyx10.cs.du.edu
5091 * Use ordered bins instead of best-fit threshhold
5092 * Eliminate block-local decls to simplify tracing and debugging.
5093 * Support another case of realloc via move into top
5094 * Fix error occuring when initial sbrk_base not word-aligned.
5095 * Rely on page size for units instead of SBRK_UNIT to
5096 avoid surprises about sbrk alignment conventions.
5097 * Add mallinfo, mallopt. Thanks to Raymond Nijssen
5098 (raymond@es.ele.tue.nl) for the suggestion.
5099 * Add `pad' argument to malloc_trim and top_pad mallopt parameter.
5100 * More precautions for cases where other routines call sbrk,
5101 courtesy of Wolfram Gloger (Gloger@lrz.uni-muenchen.de).
5102 * Added macros etc., allowing use in linux libc from
5103 H.J. Lu (hjl@gnu.ai.mit.edu)
5104 * Inverted this history list
5106 V2.6.1 Sat Dec 2 14:10:57 1995 Doug Lea (dl at gee)
5107 * Re-tuned and fixed to behave more nicely with V2.6.0 changes.
5108 * Removed all preallocation code since under current scheme
5109 the work required to undo bad preallocations exceeds
5110 the work saved in good cases for most test programs.
5111 * No longer use return list or unconsolidated bins since
5112 no scheme using them consistently outperforms those that don't
5113 given above changes.
5114 * Use best fit for very large chunks to prevent some worst-cases.
5115 * Added some support for debugging
5117 V2.6.0 Sat Nov 4 07:05:23 1995 Doug Lea (dl at gee)
5118 * Removed footers when chunks are in use. Thanks to
5119 Paul Wilson (wilson@cs.texas.edu) for the suggestion.
5121 V2.5.4 Wed Nov 1 07:54:51 1995 Doug Lea (dl at gee)
5122 * Added malloc_trim, with help from Wolfram Gloger
5123 (wmglo@Dent.MED.Uni-Muenchen.DE).
5125 V2.5.3 Tue Apr 26 10:16:01 1994 Doug Lea (dl at g)
5127 V2.5.2 Tue Apr 5 16:20:40 1994 Doug Lea (dl at g)
5128 * realloc: try to expand in both directions
5129 * malloc: swap order of clean-bin strategy;
5130 * realloc: only conditionally expand backwards
5131 * Try not to scavenge used bins
5132 * Use bin counts as a guide to preallocation
5133 * Occasionally bin return list chunks in first scan
5134 * Add a few optimizations from colin@nyx10.cs.du.edu
5136 V2.5.1 Sat Aug 14 15:40:43 1993 Doug Lea (dl at g)
5137 * faster bin computation & slightly different binning
5138 * merged all consolidations to one part of malloc proper
5139 (eliminating old malloc_find_space & malloc_clean_bin)
5140 * Scan 2 returns chunks (not just 1)
5141 * Propagate failure in realloc if malloc returns 0
5142 * Add stuff to allow compilation on non-ANSI compilers
5143 from kpv@research.att.com
5145 V2.5 Sat Aug 7 07:41:59 1993 Doug Lea (dl at g.oswego.edu)
5146 * removed potential for odd address access in prev_chunk
5147 * removed dependency on getpagesize.h
5148 * misc cosmetics and a bit more internal documentation
5149 * anticosmetics: mangled names in macros to evade debugger strangeness
5150 * tested on sparc, hp-700, dec-mips, rs6000
5151 with gcc & native cc (hp, dec only) allowing
5152 Detlefs & Zorn comparison study (in SIGPLAN Notices.)
5154 Trial version Fri Aug 28 13:14:29 1992 Doug Lea (dl at g.oswego.edu)
5155 * Based loosely on libg++-1.2X malloc. (It retains some of the overall
5156 structure of old version, but most details differ.)
OSDN Git Service
