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 * Modifications made to the original version for mono:
16 * - added PROT_EXEC to MMAP_PROT
17 * - added PAGE_EXECUTE_READWRITE to the win32mmap and win32direct_mmap
18 * - a large portion of functions is #ifdef'ed out to make the native code smaller
22 #define USE_DL_PREFIX 1
24 /* Use mmap for allocating memory */
25 #define HAVE_MORECORE 0
31 This library is all in one file to simplify the most common usage:
32 ftp it, compile it (-O3), and link it into another program. All of
33 the compile-time options default to reasonable values for use on
34 most platforms. You might later want to step through various
35 compile-time and dynamic tuning options.
37 For convenience, an include file for code using this malloc is at:
38 ftp://gee.cs.oswego.edu/pub/misc/malloc-2.8.3.h
39 You don't really need this .h file unless you call functions not
40 defined in your system include files. The .h file contains only the
41 excerpts from this file needed for using this malloc on ANSI C/C++
42 systems, so long as you haven't changed compile-time options about
43 naming and tuning parameters. If you do, then you can create your
44 own malloc.h that does include all settings by cutting at the point
45 indicated below. Note that you may already by default be using a C
46 library containing a malloc that is based on some version of this
47 malloc (for example in linux). You might still want to use the one
48 in this file to customize settings or to avoid overheads associated
49 with library versions.
53 Supported pointer/size_t representation: 4 or 8 bytes
54 size_t MUST be an unsigned type of the same width as
55 pointers. (If you are using an ancient system that declares
56 size_t as a signed type, or need it to be a different width
57 than pointers, you can use a previous release of this malloc
58 (e.g. 2.7.2) supporting these.)
60 Alignment: 8 bytes (default)
61 This suffices for nearly all current machines and C compilers.
62 However, you can define MALLOC_ALIGNMENT to be wider than this
63 if necessary (up to 128bytes), at the expense of using more space.
65 Minimum overhead per allocated chunk: 4 or 8 bytes (if 4byte sizes)
66 8 or 16 bytes (if 8byte sizes)
67 Each malloced chunk has a hidden word of overhead holding size
68 and status information, and additional cross-check word
69 if FOOTERS is defined.
71 Minimum allocated size: 4-byte ptrs: 16 bytes (including overhead)
72 8-byte ptrs: 32 bytes (including overhead)
74 Even a request for zero bytes (i.e., malloc(0)) returns a
75 pointer to something of the minimum allocatable size.
76 The maximum overhead wastage (i.e., number of extra bytes
77 allocated than were requested in malloc) is less than or equal
78 to the minimum size, except for requests >= mmap_threshold that
79 are serviced via mmap(), where the worst case wastage is about
80 32 bytes plus the remainder from a system page (the minimal
81 mmap unit); typically 4096 or 8192 bytes.
83 Security: static-safe; optionally more or less
84 The "security" of malloc refers to the ability of malicious
85 code to accentuate the effects of errors (for example, freeing
86 space that is not currently malloc'ed or overwriting past the
87 ends of chunks) in code that calls malloc. This malloc
88 guarantees not to modify any memory locations below the base of
89 heap, i.e., static variables, even in the presence of usage
90 errors. The routines additionally detect most improper frees
91 and reallocs. All this holds as long as the static bookkeeping
92 for malloc itself is not corrupted by some other means. This
93 is only one aspect of security -- these checks do not, and
94 cannot, detect all possible programming errors.
96 If FOOTERS is defined nonzero, then each allocated chunk
97 carries an additional check word to verify that it was malloced
98 from its space. These check words are the same within each
99 execution of a program using malloc, but differ across
100 executions, so externally crafted fake chunks cannot be
101 freed. This improves security by rejecting frees/reallocs that
102 could corrupt heap memory, in addition to the checks preventing
103 writes to statics that are always on. This may further improve
104 security at the expense of time and space overhead. (Note that
105 FOOTERS may also be worth using with MSPACES.)
107 By default detected errors cause the program to abort (calling
108 "abort()"). You can override this to instead proceed past
109 errors by defining PROCEED_ON_ERROR. In this case, a bad free
110 has no effect, and a malloc that encounters a bad address
111 caused by user overwrites will ignore the bad address by
112 dropping pointers and indices to all known memory. This may
113 be appropriate for programs that should continue if at all
114 possible in the face of programming errors, although they may
115 run out of memory because dropped memory is never reclaimed.
117 If you don't like either of these options, you can define
118 CORRUPTION_ERROR_ACTION and USAGE_ERROR_ACTION to do anything
119 else. And if if you are sure that your program using malloc has
120 no errors or vulnerabilities, you can define INSECURE to 1,
121 which might (or might not) provide a small performance improvement.
123 Thread-safety: NOT thread-safe unless USE_LOCKS defined
124 When USE_LOCKS is defined, each public call to malloc, free,
125 etc is surrounded with either a pthread mutex or a win32
126 spinlock (depending on WIN32). This is not especially fast, and
127 can be a major bottleneck. It is designed only to provide
128 minimal protection in concurrent environments, and to provide a
129 basis for extensions. If you are using malloc in a concurrent
130 program, consider instead using ptmalloc, which is derived from
131 a version of this malloc. (See http://www.malloc.de).
133 System requirements: Any combination of MORECORE and/or MMAP/MUNMAP
134 This malloc can use unix sbrk or any emulation (invoked using
135 the CALL_MORECORE macro) and/or mmap/munmap or any emulation
136 (invoked using CALL_MMAP/CALL_MUNMAP) to get and release system
137 memory. On most unix systems, it tends to work best if both
138 MORECORE and MMAP are enabled. On Win32, it uses emulations
139 based on VirtualAlloc. It also uses common C library functions
142 Compliance: I believe it is compliant with the Single Unix Specification
143 (See http://www.unix.org). Also SVID/XPG, ANSI C, and probably
146 * Overview of algorithms
148 This is not the fastest, most space-conserving, most portable, or
149 most tunable malloc ever written. However it is among the fastest
150 while also being among the most space-conserving, portable and
151 tunable. Consistent balance across these factors results in a good
152 general-purpose allocator for malloc-intensive programs.
154 In most ways, this malloc is a best-fit allocator. Generally, it
155 chooses the best-fitting existing chunk for a request, with ties
156 broken in approximately least-recently-used order. (This strategy
157 normally maintains low fragmentation.) However, for requests less
158 than 256bytes, it deviates from best-fit when there is not an
159 exactly fitting available chunk by preferring to use space adjacent
160 to that used for the previous small request, as well as by breaking
161 ties in approximately most-recently-used order. (These enhance
162 locality of series of small allocations.) And for very large requests
163 (>= 256Kb by default), it relies on system memory mapping
164 facilities, if supported. (This helps avoid carrying around and
165 possibly fragmenting memory used only for large chunks.)
167 All operations (except malloc_stats and mallinfo) have execution
168 times that are bounded by a constant factor of the number of bits in
169 a size_t, not counting any clearing in calloc or copying in realloc,
170 or actions surrounding MORECORE and MMAP that have times
171 proportional to the number of non-contiguous regions returned by
172 system allocation routines, which is often just 1.
174 The implementation is not very modular and seriously overuses
175 macros. Perhaps someday all C compilers will do as good a job
176 inlining modular code as can now be done by brute-force expansion,
177 but now, enough of them seem not to.
179 Some compilers issue a lot of warnings about code that is
180 dead/unreachable only on some platforms, and also about intentional
181 uses of negation on unsigned types. All known cases of each can be
184 For a longer but out of date high-level description, see
185 http://gee.cs.oswego.edu/dl/html/malloc.html
188 If MSPACES is defined, then in addition to malloc, free, etc.,
189 this file also defines mspace_malloc, mspace_free, etc. These
190 are versions of malloc routines that take an "mspace" argument
191 obtained using create_mspace, to control all internal bookkeeping.
192 If ONLY_MSPACES is defined, only these versions are compiled.
193 So if you would like to use this allocator for only some allocations,
194 and your system malloc for others, you can compile with
195 ONLY_MSPACES and then do something like...
196 static mspace mymspace = create_mspace(0,0); // for example
197 #define mymalloc(bytes) mspace_malloc(mymspace, bytes)
199 (Note: If you only need one instance of an mspace, you can instead
200 use "USE_DL_PREFIX" to relabel the global malloc.)
202 You can similarly create thread-local allocators by storing
203 mspaces as thread-locals. For example:
204 static __thread mspace tlms = 0;
205 void* tlmalloc(size_t bytes) {
206 if (tlms == 0) tlms = create_mspace(0, 0);
207 return mspace_malloc(tlms, bytes);
209 void tlfree(void* mem) { mspace_free(tlms, mem); }
211 Unless FOOTERS is defined, each mspace is completely independent.
212 You cannot allocate from one and free to another (although
213 conformance is only weakly checked, so usage errors are not always
214 caught). If FOOTERS is defined, then each chunk carries around a tag
215 indicating its originating mspace, and frees are directed to their
218 ------------------------- Compile-time options ---------------------------
220 Be careful in setting #define values for numerical constants of type
221 size_t. On some systems, literal values are not automatically extended
222 to size_t precision unless they are explicitly casted.
224 WIN32 default: defined if _WIN32 defined
225 Defining WIN32 sets up defaults for MS environment and compilers.
226 Otherwise defaults are for unix.
228 MALLOC_ALIGNMENT default: (size_t)8
229 Controls the minimum alignment for malloc'ed chunks. It must be a
230 power of two and at least 8, even on machines for which smaller
231 alignments would suffice. It may be defined as larger than this
232 though. Note however that code and data structures are optimized for
233 the case of 8-byte alignment.
235 MSPACES default: 0 (false)
236 If true, compile in support for independent allocation spaces.
237 This is only supported if HAVE_MMAP is true.
239 ONLY_MSPACES default: 0 (false)
240 If true, only compile in mspace versions, not regular versions.
242 USE_LOCKS default: 0 (false)
243 Causes each call to each public routine to be surrounded with
244 pthread or WIN32 mutex lock/unlock. (If set true, this can be
245 overridden on a per-mspace basis for mspace versions.)
248 If true, provide extra checking and dispatching by placing
249 information in the footers of allocated chunks. This adds
250 space and time overhead.
253 If true, omit checks for usage errors and heap space overwrites.
255 USE_DL_PREFIX default: NOT defined
256 Causes compiler to prefix all public routines with the string 'dl'.
257 This can be useful when you only want to use this malloc in one part
258 of a program, using your regular system malloc elsewhere.
260 ABORT default: defined as abort()
261 Defines how to abort on failed checks. On most systems, a failed
262 check cannot die with an "assert" or even print an informative
263 message, because the underlying print routines in turn call malloc,
264 which will fail again. Generally, the best policy is to simply call
265 abort(). It's not very useful to do more than this because many
266 errors due to overwriting will show up as address faults (null, odd
267 addresses etc) rather than malloc-triggered checks, so will also
268 abort. Also, most compilers know that abort() does not return, so
269 can better optimize code conditionally calling it.
271 PROCEED_ON_ERROR default: defined as 0 (false)
272 Controls whether detected bad addresses cause them to bypassed
273 rather than aborting. If set, detected bad arguments to free and
274 realloc are ignored. And all bookkeeping information is zeroed out
275 upon a detected overwrite of freed heap space, thus losing the
276 ability to ever return it from malloc again, but enabling the
277 application to proceed. If PROCEED_ON_ERROR is defined, the
278 static variable malloc_corruption_error_count is compiled in
279 and can be examined to see if errors have occurred. This option
280 generates slower code than the default abort policy.
282 DEBUG default: NOT defined
283 The DEBUG setting is mainly intended for people trying to modify
284 this code or diagnose problems when porting to new platforms.
285 However, it may also be able to better isolate user errors than just
286 using runtime checks. The assertions in the check routines spell
287 out in more detail the assumptions and invariants underlying the
288 algorithms. The checking is fairly extensive, and will slow down
289 execution noticeably. Calling malloc_stats or mallinfo with DEBUG
290 set will attempt to check every non-mmapped allocated and free chunk
291 in the course of computing the summaries.
293 ABORT_ON_ASSERT_FAILURE default: defined as 1 (true)
294 Debugging assertion failures can be nearly impossible if your
295 version of the assert macro causes malloc to be called, which will
296 lead to a cascade of further failures, blowing the runtime stack.
297 ABORT_ON_ASSERT_FAILURE cause assertions failures to call abort(),
298 which will usually make debugging easier.
300 MALLOC_FAILURE_ACTION default: sets errno to ENOMEM, or no-op on win32
301 The action to take before "return 0" when malloc fails to be able to
302 return memory because there is none available.
304 HAVE_MORECORE default: 1 (true) unless win32 or ONLY_MSPACES
305 True if this system supports sbrk or an emulation of it.
307 MORECORE default: sbrk
308 The name of the sbrk-style system routine to call to obtain more
309 memory. See below for guidance on writing custom MORECORE
310 functions. The type of the argument to sbrk/MORECORE varies across
311 systems. It cannot be size_t, because it supports negative
312 arguments, so it is normally the signed type of the same width as
313 size_t (sometimes declared as "intptr_t"). It doesn't much matter
314 though. Internally, we only call it with arguments less than half
315 the max value of a size_t, which should work across all reasonable
316 possibilities, although sometimes generating compiler warnings. See
317 near the end of this file for guidelines for creating a custom
320 MORECORE_CONTIGUOUS default: 1 (true)
321 If true, take advantage of fact that consecutive calls to MORECORE
322 with positive arguments always return contiguous increasing
323 addresses. This is true of unix sbrk. It does not hurt too much to
324 set it true anyway, since malloc copes with non-contiguities.
325 Setting it false when definitely non-contiguous saves time
326 and possibly wasted space it would take to discover this though.
328 MORECORE_CANNOT_TRIM default: NOT defined
329 True if MORECORE cannot release space back to the system when given
330 negative arguments. This is generally necessary only if you are
331 using a hand-crafted MORECORE function that cannot handle negative
334 HAVE_MMAP default: 1 (true)
335 True if this system supports mmap or an emulation of it. If so, and
336 HAVE_MORECORE is not true, MMAP is used for all system
337 allocation. If set and HAVE_MORECORE is true as well, MMAP is
338 primarily used to directly allocate very large blocks. It is also
339 used as a backup strategy in cases where MORECORE fails to provide
340 space from system. Note: A single call to MUNMAP is assumed to be
341 able to unmap memory that may have be allocated using multiple calls
342 to MMAP, so long as they are adjacent.
344 HAVE_MREMAP default: 1 on linux, else 0
345 If true realloc() uses mremap() to re-allocate large blocks and
346 extend or shrink allocation spaces.
348 MMAP_CLEARS default: 1 on unix
349 True if mmap clears memory so calloc doesn't need to. This is true
350 for standard unix mmap using /dev/zero.
352 USE_BUILTIN_FFS default: 0 (i.e., not used)
353 Causes malloc to use the builtin ffs() function to compute indices.
354 Some compilers may recognize and intrinsify ffs to be faster than the
355 supplied C version. Also, the case of x86 using gcc is special-cased
356 to an asm instruction, so is already as fast as it can be, and so
357 this setting has no effect. (On most x86s, the asm version is only
358 slightly faster than the C version.)
360 malloc_getpagesize default: derive from system includes, or 4096.
361 The system page size. To the extent possible, this malloc manages
362 memory from the system in page-size units. This may be (and
363 usually is) a function rather than a constant. This is ignored
364 if WIN32, where page size is determined using getSystemInfo during
367 USE_DEV_RANDOM default: 0 (i.e., not used)
368 Causes malloc to use /dev/random to initialize secure magic seed for
369 stamping footers. Otherwise, the current time is used.
371 NO_MALLINFO default: 0
372 If defined, don't compile "mallinfo". This can be a simple way
373 of dealing with mismatches between system declarations and
376 MALLINFO_FIELD_TYPE default: size_t
377 The type of the fields in the mallinfo struct. This was originally
378 defined as "int" in SVID etc, but is more usefully defined as
379 size_t. The value is used only if HAVE_USR_INCLUDE_MALLOC_H is not set
381 REALLOC_ZERO_BYTES_FREES default: not defined
382 This should be set if a call to realloc with zero bytes should
383 be the same as a call to free. Some people think it should. Otherwise,
384 since this malloc returns a unique pointer for malloc(0), so does
387 LACKS_UNISTD_H, LACKS_FCNTL_H, LACKS_SYS_PARAM_H, LACKS_SYS_MMAN_H
388 LACKS_STRINGS_H, LACKS_STRING_H, LACKS_SYS_TYPES_H, LACKS_ERRNO_H
389 LACKS_STDLIB_H default: NOT defined unless on WIN32
390 Define these if your system does not have these header files.
391 You might need to manually insert some of the declarations they provide.
393 DEFAULT_GRANULARITY default: page size if MORECORE_CONTIGUOUS,
394 system_info.dwAllocationGranularity in WIN32,
396 Also settable using mallopt(M_GRANULARITY, x)
397 The unit for allocating and deallocating memory from the system. On
398 most systems with contiguous MORECORE, there is no reason to
399 make this more than a page. However, systems with MMAP tend to
400 either require or encourage larger granularities. You can increase
401 this value to prevent system allocation functions to be called so
402 often, especially if they are slow. The value must be at least one
403 page and must be a power of two. Setting to 0 causes initialization
404 to either page size or win32 region size. (Note: In previous
405 versions of malloc, the equivalent of this option was called
408 DEFAULT_TRIM_THRESHOLD default: 2MB
409 Also settable using mallopt(M_TRIM_THRESHOLD, x)
410 The maximum amount of unused top-most memory to keep before
411 releasing via malloc_trim in free(). Automatic trimming is mainly
412 useful in long-lived programs using contiguous MORECORE. Because
413 trimming via sbrk can be slow on some systems, and can sometimes be
414 wasteful (in cases where programs immediately afterward allocate
415 more large chunks) the value should be high enough so that your
416 overall system performance would improve by releasing this much
417 memory. As a rough guide, you might set to a value close to the
418 average size of a process (program) running on your system.
419 Releasing this much memory would allow such a process to run in
420 memory. Generally, it is worth tuning trim thresholds when a
421 program undergoes phases where several large chunks are allocated
422 and released in ways that can reuse each other's storage, perhaps
423 mixed with phases where there are no such chunks at all. The trim
424 value must be greater than page size to have any useful effect. To
425 disable trimming completely, you can set to MAX_SIZE_T. Note that the trick
426 some people use of mallocing a huge space and then freeing it at
427 program startup, in an attempt to reserve system memory, doesn't
428 have the intended effect under automatic trimming, since that memory
429 will immediately be returned to the system.
431 DEFAULT_MMAP_THRESHOLD default: 256K
432 Also settable using mallopt(M_MMAP_THRESHOLD, x)
433 The request size threshold for using MMAP to directly service a
434 request. Requests of at least this size that cannot be allocated
435 using already-existing space will be serviced via mmap. (If enough
436 normal freed space already exists it is used instead.) Using mmap
437 segregates relatively large chunks of memory so that they can be
438 individually obtained and released from the host system. A request
439 serviced through mmap is never reused by any other request (at least
440 not directly; the system may just so happen to remap successive
441 requests to the same locations). Segregating space in this way has
442 the benefits that: Mmapped space can always be individually released
443 back to the system, which helps keep the system level memory demands
444 of a long-lived program low. Also, mapped memory doesn't become
445 `locked' between other chunks, as can happen with normally allocated
446 chunks, which means that even trimming via malloc_trim would not
447 release them. However, it has the disadvantage that the space
448 cannot be reclaimed, consolidated, and then used to service later
449 requests, as happens with normal chunks. The advantages of mmap
450 nearly always outweigh disadvantages for "large" chunks, but the
451 value of "large" may vary across systems. The default is an
452 empirically derived value that works well in most systems. You can
453 disable mmap by setting to MAX_SIZE_T.
463 #define WIN32_LEAN_AND_MEAN
466 #define HAVE_MORECORE 0
467 #define LACKS_UNISTD_H
468 #define LACKS_SYS_PARAM_H
469 #define LACKS_SYS_MMAN_H
470 #define LACKS_STRING_H
471 #define LACKS_STRINGS_H
472 #define LACKS_SYS_TYPES_H
473 #define LACKS_ERRNO_H
474 #define MALLOC_FAILURE_ACTION
475 #define MMAP_CLEARS 0 /* WINCE and some others apparently don't clear */
478 #if defined(DARWIN) || defined(_DARWIN)
479 /* Mac OSX docs advise not to use sbrk; it seems better to use mmap */
480 #ifndef HAVE_MORECORE
481 #define HAVE_MORECORE 0
483 #endif /* HAVE_MORECORE */
486 #if defined(__native_client__)
490 #define HAVE_MREMAP 0
493 #ifndef LACKS_SYS_TYPES_H
494 #include <sys/types.h> /* For size_t */
495 #endif /* LACKS_SYS_TYPES_H */
497 /* The maximum possible size_t value has all bits set */
498 #define MAX_SIZE_T (~(size_t)0)
501 #define ONLY_MSPACES 0
502 #endif /* ONLY_MSPACES */
506 #else /* ONLY_MSPACES */
508 #endif /* ONLY_MSPACES */
510 #ifndef MALLOC_ALIGNMENT
511 #define MALLOC_ALIGNMENT ((size_t)8U)
512 #endif /* MALLOC_ALIGNMENT */
517 #define ABORT abort()
519 #ifndef ABORT_ON_ASSERT_FAILURE
520 #define ABORT_ON_ASSERT_FAILURE 1
521 #endif /* ABORT_ON_ASSERT_FAILURE */
522 #ifndef PROCEED_ON_ERROR
523 #define PROCEED_ON_ERROR 0
524 #endif /* PROCEED_ON_ERROR */
527 #endif /* USE_LOCKS */
530 #endif /* INSECURE */
533 #endif /* HAVE_MMAP */
535 #define MMAP_CLEARS 1
536 #endif /* MMAP_CLEARS */
539 #define HAVE_MREMAP 1
541 #define HAVE_MREMAP 0
543 #endif /* HAVE_MREMAP */
544 #ifndef MALLOC_FAILURE_ACTION
545 #define MALLOC_FAILURE_ACTION errno = ENOMEM;
546 #endif /* MALLOC_FAILURE_ACTION */
547 #ifndef HAVE_MORECORE
549 #define HAVE_MORECORE 0
550 #else /* ONLY_MSPACES */
551 #define HAVE_MORECORE 1
552 #endif /* ONLY_MSPACES */
553 #endif /* HAVE_MORECORE */
555 #define MORECORE_CONTIGUOUS 0
556 #else /* !HAVE_MORECORE */
558 #define MORECORE sbrk
559 #endif /* MORECORE */
560 #ifndef MORECORE_CONTIGUOUS
561 #define MORECORE_CONTIGUOUS 1
562 #endif /* MORECORE_CONTIGUOUS */
563 #endif /* HAVE_MORECORE */
564 #ifndef DEFAULT_GRANULARITY
565 #if MORECORE_CONTIGUOUS
566 #define DEFAULT_GRANULARITY (0) /* 0 means to compute in init_mparams */
567 #else /* MORECORE_CONTIGUOUS */
568 #define DEFAULT_GRANULARITY ((size_t)64U * (size_t)1024U)
569 #endif /* MORECORE_CONTIGUOUS */
570 #endif /* DEFAULT_GRANULARITY */
571 #ifndef DEFAULT_TRIM_THRESHOLD
572 #ifndef MORECORE_CANNOT_TRIM
573 #define DEFAULT_TRIM_THRESHOLD ((size_t)2U * (size_t)1024U * (size_t)1024U)
574 #else /* MORECORE_CANNOT_TRIM */
575 #define DEFAULT_TRIM_THRESHOLD MAX_SIZE_T
576 #endif /* MORECORE_CANNOT_TRIM */
577 #endif /* DEFAULT_TRIM_THRESHOLD */
578 #ifndef DEFAULT_MMAP_THRESHOLD
580 #define DEFAULT_MMAP_THRESHOLD ((size_t)256U * (size_t)1024U)
581 #else /* HAVE_MMAP */
582 #define DEFAULT_MMAP_THRESHOLD MAX_SIZE_T
583 #endif /* HAVE_MMAP */
584 #endif /* DEFAULT_MMAP_THRESHOLD */
585 #ifndef USE_BUILTIN_FFS
586 #define USE_BUILTIN_FFS 0
587 #endif /* USE_BUILTIN_FFS */
588 #ifndef USE_DEV_RANDOM
589 #define USE_DEV_RANDOM 0
590 #endif /* USE_DEV_RANDOM */
592 #define NO_MALLINFO 0
593 #endif /* NO_MALLINFO */
594 #ifndef MALLINFO_FIELD_TYPE
595 #define MALLINFO_FIELD_TYPE size_t
596 #endif /* MALLINFO_FIELD_TYPE */
599 mallopt tuning options. SVID/XPG defines four standard parameter
600 numbers for mallopt, normally defined in malloc.h. None of these
601 are used in this malloc, so setting them has no effect. But this
602 malloc does support the following options.
605 #define M_TRIM_THRESHOLD (-1)
606 #define M_GRANULARITY (-2)
607 #define M_MMAP_THRESHOLD (-3)
609 /* ------------------------ Mallinfo declarations ------------------------ */
613 This version of malloc supports the standard SVID/XPG mallinfo
614 routine that returns a struct containing usage properties and
615 statistics. It should work on any system that has a
616 /usr/include/malloc.h defining struct mallinfo. The main
617 declaration needed is the mallinfo struct that is returned (by-copy)
618 by mallinfo(). The malloinfo struct contains a bunch of fields that
619 are not even meaningful in this version of malloc. These fields are
620 are instead filled by mallinfo() with other numbers that might be of
623 HAVE_USR_INCLUDE_MALLOC_H should be set if you have a
624 /usr/include/malloc.h file that includes a declaration of struct
625 mallinfo. If so, it is included; else a compliant version is
626 declared below. These must be precisely the same for mallinfo() to
627 work. The original SVID version of this struct, defined on most
628 systems with mallinfo, declares all fields as ints. But some others
629 define as unsigned long. If your system defines the fields using a
630 type of different width than listed here, you MUST #include your
631 system version and #define HAVE_USR_INCLUDE_MALLOC_H.
634 /* #define HAVE_USR_INCLUDE_MALLOC_H */
636 #ifdef HAVE_USR_INCLUDE_MALLOC_H
637 #include "/usr/include/malloc.h"
638 #else /* HAVE_USR_INCLUDE_MALLOC_H */
641 MALLINFO_FIELD_TYPE arena; /* non-mmapped space allocated from system */
642 MALLINFO_FIELD_TYPE ordblks; /* number of free chunks */
643 MALLINFO_FIELD_TYPE smblks; /* always 0 */
644 MALLINFO_FIELD_TYPE hblks; /* always 0 */
645 MALLINFO_FIELD_TYPE hblkhd; /* space in mmapped regions */
646 MALLINFO_FIELD_TYPE usmblks; /* maximum total allocated space */
647 MALLINFO_FIELD_TYPE fsmblks; /* always 0 */
648 MALLINFO_FIELD_TYPE uordblks; /* total allocated space */
649 MALLINFO_FIELD_TYPE fordblks; /* total free space */
650 MALLINFO_FIELD_TYPE keepcost; /* releasable (via malloc_trim) space */
653 #endif /* HAVE_USR_INCLUDE_MALLOC_H */
654 #endif /* NO_MALLINFO */
658 #endif /* __cplusplus */
662 /* ------------------- Declarations of public routines ------------------- */
664 #ifndef USE_DL_PREFIX
665 #define dlcalloc calloc
667 #define dlmalloc malloc
668 #define dlmemalign memalign
669 #define dlrealloc realloc
670 #define dlvalloc valloc
671 #define dlpvalloc pvalloc
672 #define dlmallinfo mallinfo
673 #define dlmallopt mallopt
674 #define dlmalloc_trim malloc_trim
675 #define dlmalloc_stats malloc_stats
676 #define dlmalloc_usable_size malloc_usable_size
677 #define dlmalloc_footprint malloc_footprint
678 #define dlmalloc_max_footprint malloc_max_footprint
679 #define dlindependent_calloc independent_calloc
680 #define dlindependent_comalloc independent_comalloc
681 #endif /* USE_DL_PREFIX */
686 Returns a pointer to a newly allocated chunk of at least n bytes, or
687 null if no space is available, in which case errno is set to ENOMEM
690 If n is zero, malloc returns a minimum-sized chunk. (The minimum
691 size is 16 bytes on most 32bit systems, and 32 bytes on 64bit
692 systems.) Note that size_t is an unsigned type, so calls with
693 arguments that would be negative if signed are interpreted as
694 requests for huge amounts of space, which will often fail. The
695 maximum supported value of n differs across systems, but is in all
696 cases less than the maximum representable value of a size_t.
698 void* dlmalloc(size_t);
702 Releases the chunk of memory pointed to by p, that had been previously
703 allocated using malloc or a related routine such as realloc.
704 It has no effect if p is null. If p was not malloced or already
705 freed, free(p) will by default cause the current program to abort.
710 calloc(size_t n_elements, size_t element_size);
711 Returns a pointer to n_elements * element_size bytes, with all locations
714 void* dlcalloc(size_t, size_t);
717 realloc(void* p, size_t n)
718 Returns a pointer to a chunk of size n that contains the same data
719 as does chunk p up to the minimum of (n, p's size) bytes, or null
720 if no space is available.
722 The returned pointer may or may not be the same as p. The algorithm
723 prefers extending p in most cases when possible, otherwise it
724 employs the equivalent of a malloc-copy-free sequence.
726 If p is null, realloc is equivalent to malloc.
728 If space is not available, realloc returns null, errno is set (if on
729 ANSI) and p is NOT freed.
731 if n is for fewer bytes than already held by p, the newly unused
732 space is lopped off and freed if possible. realloc with a size
733 argument of zero (re)allocates a minimum-sized chunk.
735 The old unix realloc convention of allowing the last-free'd chunk
736 to be used as an argument to realloc is not supported.
739 void* dlrealloc(void*, size_t);
742 memalign(size_t alignment, size_t n);
743 Returns a pointer to a newly allocated chunk of n bytes, aligned
744 in accord with the alignment argument.
746 The alignment argument should be a power of two. If the argument is
747 not a power of two, the nearest greater power is used.
748 8-byte alignment is guaranteed by normal malloc calls, so don't
749 bother calling memalign with an argument of 8 or less.
751 Overreliance on memalign is a sure way to fragment space.
753 void* dlmemalign(size_t, size_t);
757 Equivalent to memalign(pagesize, n), where pagesize is the page
758 size of the system. If the pagesize is unknown, 4096 is used.
760 void* dlvalloc(size_t);
763 mallopt(int parameter_number, int parameter_value)
764 Sets tunable parameters The format is to provide a
765 (parameter-number, parameter-value) pair. mallopt then sets the
766 corresponding parameter to the argument value if it can (i.e., so
767 long as the value is meaningful), and returns 1 if successful else
768 0. SVID/XPG/ANSI defines four standard param numbers for mallopt,
769 normally defined in malloc.h. None of these are use in this malloc,
770 so setting them has no effect. But this malloc also supports other
771 options in mallopt. See below for details. Briefly, supported
772 parameters are as follows (listed defaults are for "typical"
775 Symbol param # default allowed param values
776 M_TRIM_THRESHOLD -1 2*1024*1024 any (MAX_SIZE_T disables)
777 M_GRANULARITY -2 page size any power of 2 >= page size
778 M_MMAP_THRESHOLD -3 256*1024 any (or 0 if no MMAP support)
780 int dlmallopt(int, int);
784 Returns the number of bytes obtained from the system. The total
785 number of bytes allocated by malloc, realloc etc., is less than this
786 value. Unlike mallinfo, this function returns only a precomputed
787 result, so can be called frequently to monitor memory consumption.
788 Even if locks are otherwise defined, this function does not use them,
789 so results might not be up to date.
791 size_t dlmalloc_footprint(void);
794 malloc_max_footprint();
795 Returns the maximum number of bytes obtained from the system. This
796 value will be greater than current footprint if deallocated space
797 has been reclaimed by the system. The peak number of bytes allocated
798 by malloc, realloc etc., is less than this value. Unlike mallinfo,
799 this function returns only a precomputed result, so can be called
800 frequently to monitor memory consumption. Even if locks are
801 otherwise defined, this function does not use them, so results might
804 size_t dlmalloc_max_footprint(void);
809 Returns (by copy) a struct containing various summary statistics:
811 arena: current total non-mmapped bytes allocated from system
812 ordblks: the number of free chunks
814 hblks: current number of mmapped regions
815 hblkhd: total bytes held in mmapped regions
816 usmblks: the maximum total allocated space. This will be greater
817 than current total if trimming has occurred.
819 uordblks: current total allocated space (normal or mmapped)
820 fordblks: total free space
821 keepcost: the maximum number of bytes that could ideally be released
822 back to system via malloc_trim. ("ideally" means that
823 it ignores page restrictions etc.)
825 Because these fields are ints, but internal bookkeeping may
826 be kept as longs, the reported values may wrap around zero and
829 struct mallinfo dlmallinfo(void);
830 #endif /* NO_MALLINFO */
833 independent_calloc(size_t n_elements, size_t element_size, void* chunks[]);
835 independent_calloc is similar to calloc, but instead of returning a
836 single cleared space, it returns an array of pointers to n_elements
837 independent elements that can hold contents of size elem_size, each
838 of which starts out cleared, and can be independently freed,
839 realloc'ed etc. The elements are guaranteed to be adjacently
840 allocated (this is not guaranteed to occur with multiple callocs or
841 mallocs), which may also improve cache locality in some
844 The "chunks" argument is optional (i.e., may be null, which is
845 probably the most typical usage). If it is null, the returned array
846 is itself dynamically allocated and should also be freed when it is
847 no longer needed. Otherwise, the chunks array must be of at least
848 n_elements in length. It is filled in with the pointers to the
851 In either case, independent_calloc returns this pointer array, or
852 null if the allocation failed. If n_elements is zero and "chunks"
853 is null, it returns a chunk representing an array with zero elements
854 (which should be freed if not wanted).
856 Each element must be individually freed when it is no longer
857 needed. If you'd like to instead be able to free all at once, you
858 should instead use regular calloc and assign pointers into this
859 space to represent elements. (In this case though, you cannot
860 independently free elements.)
862 independent_calloc simplifies and speeds up implementations of many
863 kinds of pools. It may also be useful when constructing large data
864 structures that initially have a fixed number of fixed-sized nodes,
865 but the number is not known at compile time, and some of the nodes
866 may later need to be freed. For example:
868 struct Node { int item; struct Node* next; };
870 struct Node* build_list() {
872 int n = read_number_of_nodes_needed();
873 if (n <= 0) return 0;
874 pool = (struct Node**)(independent_calloc(n, sizeof(struct Node), 0);
875 if (pool == 0) die();
876 // organize into a linked list...
877 struct Node* first = pool[0];
878 for (i = 0; i < n-1; ++i)
879 pool[i]->next = pool[i+1];
880 free(pool); // Can now free the array (or not, if it is needed later)
884 void** dlindependent_calloc(size_t, size_t, void**);
887 independent_comalloc(size_t n_elements, size_t sizes[], void* chunks[]);
889 independent_comalloc allocates, all at once, a set of n_elements
890 chunks with sizes indicated in the "sizes" array. It returns
891 an array of pointers to these elements, each of which can be
892 independently freed, realloc'ed etc. The elements are guaranteed to
893 be adjacently allocated (this is not guaranteed to occur with
894 multiple callocs or mallocs), which may also improve cache locality
895 in some applications.
897 The "chunks" argument is optional (i.e., may be null). If it is null
898 the returned array is itself dynamically allocated and should also
899 be freed when it is no longer needed. Otherwise, the chunks array
900 must be of at least n_elements in length. It is filled in with the
901 pointers to the chunks.
903 In either case, independent_comalloc returns this pointer array, or
904 null if the allocation failed. If n_elements is zero and chunks is
905 null, it returns a chunk representing an array with zero elements
906 (which should be freed if not wanted).
908 Each element must be individually freed when it is no longer
909 needed. If you'd like to instead be able to free all at once, you
910 should instead use a single regular malloc, and assign pointers at
911 particular offsets in the aggregate space. (In this case though, you
912 cannot independently free elements.)
914 independent_comallac differs from independent_calloc in that each
915 element may have a different size, and also that it does not
916 automatically clear elements.
918 independent_comalloc can be used to speed up allocation in cases
919 where several structs or objects must always be allocated at the
920 same time. For example:
925 void send_message(char* msg) {
926 int msglen = strlen(msg);
927 size_t sizes[3] = { sizeof(struct Head), msglen, sizeof(struct Foot) };
929 if (independent_comalloc(3, sizes, chunks) == 0)
931 struct Head* head = (struct Head*)(chunks[0]);
932 char* body = (char*)(chunks[1]);
933 struct Foot* foot = (struct Foot*)(chunks[2]);
937 In general though, independent_comalloc is worth using only for
938 larger values of n_elements. For small values, you probably won't
939 detect enough difference from series of malloc calls to bother.
941 Overuse of independent_comalloc can increase overall memory usage,
942 since it cannot reuse existing noncontiguous small chunks that
943 might be available for some of the elements.
945 void** dlindependent_comalloc(size_t, size_t*, void**);
950 Equivalent to valloc(minimum-page-that-holds(n)), that is,
951 round up n to nearest pagesize.
953 void* dlpvalloc(size_t);
956 malloc_trim(size_t pad);
958 If possible, gives memory back to the system (via negative arguments
959 to sbrk) if there is unused memory at the `high' end of the malloc
960 pool or in unused MMAP segments. You can call this after freeing
961 large blocks of memory to potentially reduce the system-level memory
962 requirements of a program. However, it cannot guarantee to reduce
963 memory. Under some allocation patterns, some large free blocks of
964 memory will be locked between two used chunks, so they cannot be
965 given back to the system.
967 The `pad' argument to malloc_trim represents the amount of free
968 trailing space to leave untrimmed. If this argument is zero, only
969 the minimum amount of memory to maintain internal data structures
970 will be left. Non-zero arguments can be supplied to maintain enough
971 trailing space to service future expected allocations without having
972 to re-obtain memory from the system.
974 Malloc_trim returns 1 if it actually released any memory, else 0.
976 int dlmalloc_trim(size_t);
979 malloc_usable_size(void* p);
981 Returns the number of bytes you can actually use in
982 an allocated chunk, which may be more than you requested (although
983 often not) due to alignment and minimum size constraints.
984 You can use this many bytes without worrying about
985 overwriting other allocated objects. This is not a particularly great
986 programming practice. malloc_usable_size can be more useful in
987 debugging and assertions, for example:
990 assert(malloc_usable_size(p) >= 256);
992 size_t dlmalloc_usable_size(void*);
996 Prints on stderr the amount of space obtained from the system (both
997 via sbrk and mmap), the maximum amount (which may be more than
998 current if malloc_trim and/or munmap got called), and the current
999 number of bytes allocated via malloc (or realloc, etc) but not yet
1000 freed. Note that this is the number of bytes allocated, not the
1001 number requested. It will be larger than the number requested
1002 because of alignment and bookkeeping overhead. Because it includes
1003 alignment wastage as being in use, this figure may be greater than
1004 zero even when no user-level chunks are allocated.
1006 The reported current and maximum system memory can be inaccurate if
1007 a program makes other calls to system memory allocation functions
1008 (normally sbrk) outside of malloc.
1010 malloc_stats prints only the most commonly interesting statistics.
1011 More information can be obtained by calling mallinfo.
1013 void dlmalloc_stats(void);
1015 #endif /* ONLY_MSPACES */
1020 mspace is an opaque type representing an independent
1021 region of space that supports mspace_malloc, etc.
1023 typedef void* mspace;
1026 create_mspace creates and returns a new independent space with the
1027 given initial capacity, or, if 0, the default granularity size. It
1028 returns null if there is no system memory available to create the
1029 space. If argument locked is non-zero, the space uses a separate
1030 lock to control access. The capacity of the space will grow
1031 dynamically as needed to service mspace_malloc requests. You can
1032 control the sizes of incremental increases of this space by
1033 compiling with a different DEFAULT_GRANULARITY or dynamically
1034 setting with mallopt(M_GRANULARITY, value).
1036 mspace create_mspace(size_t capacity, int locked);
1039 destroy_mspace destroys the given space, and attempts to return all
1040 of its memory back to the system, returning the total number of
1041 bytes freed. After destruction, the results of access to all memory
1042 used by the space become undefined.
1044 size_t destroy_mspace(mspace msp);
1047 create_mspace_with_base uses the memory supplied as the initial base
1048 of a new mspace. Part (less than 128*sizeof(size_t) bytes) of this
1049 space is used for bookkeeping, so the capacity must be at least this
1050 large. (Otherwise 0 is returned.) When this initial space is
1051 exhausted, additional memory will be obtained from the system.
1052 Destroying this space will deallocate all additionally allocated
1053 space (if possible) but not the initial base.
1055 mspace create_mspace_with_base(void* base, size_t capacity, int locked);
1058 mspace_malloc behaves as malloc, but operates within
1061 void* mspace_malloc(mspace msp, size_t bytes);
1064 mspace_free behaves as free, but operates within
1067 If compiled with FOOTERS==1, mspace_free is not actually needed.
1068 free may be called instead of mspace_free because freed chunks from
1069 any space are handled by their originating spaces.
1071 void mspace_free(mspace msp, void* mem);
1074 mspace_realloc behaves as realloc, but operates within
1077 If compiled with FOOTERS==1, mspace_realloc is not actually
1078 needed. realloc may be called instead of mspace_realloc because
1079 realloced chunks from any space are handled by their originating
1082 void* mspace_realloc(mspace msp, void* mem, size_t newsize);
1085 mspace_calloc behaves as calloc, but operates within
1088 void* mspace_calloc(mspace msp, size_t n_elements, size_t elem_size);
1091 mspace_memalign behaves as memalign, but operates within
1094 void* mspace_memalign(mspace msp, size_t alignment, size_t bytes);
1097 mspace_independent_calloc behaves as independent_calloc, but
1098 operates within the given space.
1100 void** mspace_independent_calloc(mspace msp, size_t n_elements,
1101 size_t elem_size, void* chunks[]);
1104 mspace_independent_comalloc behaves as independent_comalloc, but
1105 operates within the given space.
1107 void** mspace_independent_comalloc(mspace msp, size_t n_elements,
1108 size_t sizes[], void* chunks[]);
1111 mspace_footprint() returns the number of bytes obtained from the
1112 system for this space.
1114 size_t mspace_footprint(mspace msp);
1117 mspace_max_footprint() returns the peak number of bytes obtained from the
1118 system for this space.
1120 size_t mspace_max_footprint(mspace msp);
1125 mspace_mallinfo behaves as mallinfo, but reports properties of
1128 struct mallinfo mspace_mallinfo(mspace msp);
1129 #endif /* NO_MALLINFO */
1132 mspace_malloc_stats behaves as malloc_stats, but reports
1133 properties of the given space.
1135 void mspace_malloc_stats(mspace msp);
1138 mspace_trim behaves as malloc_trim, but
1139 operates within the given space.
1141 int mspace_trim(mspace msp, size_t pad);
1144 An alias for mallopt.
1146 int mspace_mallopt(int, int);
1148 #endif /* MSPACES */
1151 }; /* end of extern "C" */
1152 #endif /* __cplusplus */
1155 ========================================================================
1156 To make a fully customizable malloc.h header file, cut everything
1157 above this line, put into file malloc.h, edit to suit, and #include it
1158 on the next line, as well as in programs that use this malloc.
1159 ========================================================================
1162 /* #include "malloc.h" */
1164 /*------------------------------ internal #includes ---------------------- */
1167 #pragma warning( disable : 4146 ) /* no "unsigned" warnings */
1170 #include <stdio.h> /* for printing in malloc_stats */
1172 #ifndef LACKS_ERRNO_H
1173 #include <errno.h> /* for MALLOC_FAILURE_ACTION */
1174 #endif /* LACKS_ERRNO_H */
1176 #include <time.h> /* for magic initialization */
1177 #endif /* FOOTERS */
1178 #ifndef LACKS_STDLIB_H
1179 #include <stdlib.h> /* for abort() */
1180 #endif /* LACKS_STDLIB_H */
1182 #if ABORT_ON_ASSERT_FAILURE
1183 #define assert(x) if(!(x)) ABORT
1184 #else /* ABORT_ON_ASSERT_FAILURE */
1186 #endif /* ABORT_ON_ASSERT_FAILURE */
1190 #ifndef LACKS_STRING_H
1191 #include <string.h> /* for memset etc */
1192 #endif /* LACKS_STRING_H */
1194 #ifndef LACKS_STRINGS_H
1195 #include <strings.h> /* for ffs */
1196 #endif /* LACKS_STRINGS_H */
1197 #endif /* USE_BUILTIN_FFS */
1199 #ifndef LACKS_SYS_MMAN_H
1200 #include <sys/mman.h> /* for mmap */
1201 #endif /* LACKS_SYS_MMAN_H */
1202 #ifndef LACKS_FCNTL_H
1204 #endif /* LACKS_FCNTL_H */
1205 #endif /* HAVE_MMAP */
1207 #ifndef LACKS_UNISTD_H
1208 #include <unistd.h> /* for sbrk */
1209 #else /* LACKS_UNISTD_H */
1210 #if !defined(__FreeBSD__) && !defined(__OpenBSD__) && !defined(__NetBSD__)
1211 extern void* sbrk(ptrdiff_t);
1212 #endif /* FreeBSD etc */
1213 #endif /* LACKS_UNISTD_H */
1214 #endif /* HAVE_MMAP */
1217 #ifndef malloc_getpagesize
1218 # ifdef _SC_PAGESIZE /* some SVR4 systems omit an underscore */
1219 # ifndef _SC_PAGE_SIZE
1220 # define _SC_PAGE_SIZE _SC_PAGESIZE
1223 # ifdef _SC_PAGE_SIZE
1224 # define malloc_getpagesize sysconf(_SC_PAGE_SIZE)
1226 # if defined(BSD) || defined(DGUX) || defined(HAVE_GETPAGESIZE)
1227 extern size_t getpagesize();
1228 # define malloc_getpagesize getpagesize()
1230 # ifdef WIN32 /* use supplied emulation of getpagesize */
1231 # define malloc_getpagesize getpagesize()
1233 # ifndef LACKS_SYS_PARAM_H
1234 # include <sys/param.h>
1236 # ifdef EXEC_PAGESIZE
1237 # define malloc_getpagesize EXEC_PAGESIZE
1241 # define malloc_getpagesize NBPG
1243 # define malloc_getpagesize (NBPG * CLSIZE)
1247 # define malloc_getpagesize NBPC
1250 # define malloc_getpagesize PAGESIZE
1251 # else /* just guess */
1252 # define malloc_getpagesize ((size_t)4096U)
1263 /* ------------------- size_t and alignment properties -------------------- */
1265 /* The byte and bit size of a size_t */
1266 #define SIZE_T_SIZE (sizeof(size_t))
1267 #define SIZE_T_BITSIZE (sizeof(size_t) << 3)
1269 /* Some constants coerced to size_t */
1270 /* Annoying but necessary to avoid errors on some plaftorms */
1271 #define SIZE_T_ZERO ((size_t)0)
1272 #define SIZE_T_ONE ((size_t)1)
1273 #define SIZE_T_TWO ((size_t)2)
1274 #define TWO_SIZE_T_SIZES (SIZE_T_SIZE<<1)
1275 #define FOUR_SIZE_T_SIZES (SIZE_T_SIZE<<2)
1276 #define SIX_SIZE_T_SIZES (FOUR_SIZE_T_SIZES+TWO_SIZE_T_SIZES)
1277 #define HALF_MAX_SIZE_T (MAX_SIZE_T / 2U)
1279 /* The bit mask value corresponding to MALLOC_ALIGNMENT */
1280 #define CHUNK_ALIGN_MASK (MALLOC_ALIGNMENT - SIZE_T_ONE)
1282 /* True if address a has acceptable alignment */
1283 #define is_aligned(A) (((size_t)((A)) & (CHUNK_ALIGN_MASK)) == 0)
1285 /* the number of bytes to offset an address to align it */
1286 #define align_offset(A)\
1287 ((((size_t)(A) & CHUNK_ALIGN_MASK) == 0)? 0 :\
1288 ((MALLOC_ALIGNMENT - ((size_t)(A) & CHUNK_ALIGN_MASK)) & CHUNK_ALIGN_MASK))
1290 /* -------------------------- MMAP preliminaries ------------------------- */
1293 If HAVE_MORECORE or HAVE_MMAP are false, we just define calls and
1294 checks to fail so compiler optimizer can delete code rather than
1295 using so many "#if"s.
1299 /* MORECORE and MMAP must return MFAIL on failure */
1300 #define MFAIL ((void*)(MAX_SIZE_T))
1301 #define CMFAIL ((char*)(MFAIL)) /* defined for convenience */
1304 #define IS_MMAPPED_BIT (SIZE_T_ZERO)
1305 #define USE_MMAP_BIT (SIZE_T_ZERO)
1306 #define CALL_MMAP(s) MFAIL
1307 #define CALL_MUNMAP(a, s) (-1)
1308 #define DIRECT_MMAP(s) MFAIL
1310 #else /* HAVE_MMAP */
1311 #define IS_MMAPPED_BIT (SIZE_T_ONE)
1312 #define USE_MMAP_BIT (SIZE_T_ONE)
1315 #define CALL_MUNMAP(a, s) munmap((a), (s))
1316 #define MMAP_PROT (PROT_READ|PROT_WRITE|PROT_EXEC)
1317 #if !defined(MAP_ANONYMOUS) && defined(MAP_ANON)
1318 #define MAP_ANONYMOUS MAP_ANON
1319 #endif /* MAP_ANON */
1320 #ifdef MAP_ANONYMOUS
1321 #define MMAP_FLAGS (MAP_PRIVATE|MAP_ANONYMOUS)
1322 #define CALL_MMAP(s) mmap(0, (s), MMAP_PROT, MMAP_FLAGS, -1, 0)
1323 #else /* MAP_ANONYMOUS */
1325 Nearly all versions of mmap support MAP_ANONYMOUS, so the following
1326 is unlikely to be needed, but is supplied just in case.
1328 #define MMAP_FLAGS (MAP_PRIVATE)
1329 static int dev_zero_fd = -1; /* Cached file descriptor for /dev/zero. */
1330 #define CALL_MMAP(s) ((dev_zero_fd < 0) ? \
1331 (dev_zero_fd = open("/dev/zero", O_RDWR), \
1332 mmap(0, (s), MMAP_PROT, MMAP_FLAGS, dev_zero_fd, 0)) : \
1333 mmap(0, (s), MMAP_PROT, MMAP_FLAGS, dev_zero_fd, 0))
1334 #endif /* MAP_ANONYMOUS */
1336 #define DIRECT_MMAP(s) CALL_MMAP(s)
1339 /* Win32 MMAP via VirtualAlloc */
1340 static void* win32mmap(size_t size) {
1341 void* ptr = VirtualAlloc(0, size, MEM_RESERVE|MEM_COMMIT, PAGE_EXECUTE_READWRITE);
1342 return (ptr != 0)? ptr: MFAIL;
1345 /* For direct MMAP, use MEM_TOP_DOWN to minimize interference */
1346 static void* win32direct_mmap(size_t size) {
1347 void* ptr = VirtualAlloc(0, size, MEM_RESERVE|MEM_COMMIT|MEM_TOP_DOWN,
1348 PAGE_EXECUTE_READWRITE);
1349 return (ptr != 0)? ptr: MFAIL;
1352 /* This function supports releasing coalesed segments */
1353 static int win32munmap(void* ptr, size_t size) {
1354 MEMORY_BASIC_INFORMATION minfo;
1357 if (VirtualQuery(cptr, &minfo, sizeof(minfo)) == 0)
1359 if (minfo.BaseAddress != cptr || minfo.AllocationBase != cptr ||
1360 minfo.State != MEM_COMMIT || minfo.RegionSize > size)
1362 if (VirtualFree(cptr, 0, MEM_RELEASE) == 0)
1364 cptr += minfo.RegionSize;
1365 size -= minfo.RegionSize;
1370 #define CALL_MMAP(s) win32mmap(s)
1371 #define CALL_MUNMAP(a, s) win32munmap((a), (s))
1372 #define DIRECT_MMAP(s) win32direct_mmap(s)
1374 #endif /* HAVE_MMAP */
1376 #if HAVE_MMAP && HAVE_MREMAP
1377 #define CALL_MREMAP(addr, osz, nsz, mv) mremap((addr), (osz), (nsz), (mv))
1378 #else /* HAVE_MMAP && HAVE_MREMAP */
1379 #define CALL_MREMAP(addr, osz, nsz, mv) MFAIL
1380 #endif /* HAVE_MMAP && HAVE_MREMAP */
1383 #define CALL_MORECORE(S) MORECORE(S)
1384 #else /* HAVE_MORECORE */
1385 #define CALL_MORECORE(S) MFAIL
1386 #endif /* HAVE_MORECORE */
1388 /* mstate bit set if continguous morecore disabled or failed */
1389 #define USE_NONCONTIGUOUS_BIT (4U)
1391 /* segment bit set in create_mspace_with_base */
1392 #define EXTERN_BIT (8U)
1395 /* --------------------------- Lock preliminaries ------------------------ */
1400 When locks are defined, there are up to two global locks:
1402 * If HAVE_MORECORE, morecore_mutex protects sequences of calls to
1403 MORECORE. In many cases sys_alloc requires two calls, that should
1404 not be interleaved with calls by other threads. This does not
1405 protect against direct calls to MORECORE by other threads not
1406 using this lock, so there is still code to cope the best we can on
1409 * magic_init_mutex ensures that mparams.magic and other
1410 unique mparams values are initialized only once.
1414 /* By default use posix locks */
1415 #include <pthread.h>
1416 #define MLOCK_T pthread_mutex_t
1417 #define INITIAL_LOCK(l) pthread_mutex_init(l, NULL)
1418 #define ACQUIRE_LOCK(l) pthread_mutex_lock(l)
1419 #define RELEASE_LOCK(l) pthread_mutex_unlock(l)
1422 static MLOCK_T morecore_mutex = PTHREAD_MUTEX_INITIALIZER;
1423 #endif /* HAVE_MORECORE */
1425 static MLOCK_T magic_init_mutex = PTHREAD_MUTEX_INITIALIZER;
1429 Because lock-protected regions have bounded times, and there
1430 are no recursive lock calls, we can use simple spinlocks.
1433 #define MLOCK_T long
1434 static int win32_acquire_lock (MLOCK_T *sl) {
1436 #ifdef InterlockedCompareExchangePointer
1437 if (!InterlockedCompareExchange(sl, 1, 0))
1439 #else /* Use older void* version */
1440 if (!InterlockedCompareExchange((void**)sl, (void*)1, (void*)0))
1442 #endif /* InterlockedCompareExchangePointer */
1447 static void win32_release_lock (MLOCK_T *sl) {
1448 InterlockedExchange (sl, 0);
1451 #define INITIAL_LOCK(l) *(l)=0
1452 #define ACQUIRE_LOCK(l) win32_acquire_lock(l)
1453 #define RELEASE_LOCK(l) win32_release_lock(l)
1455 static MLOCK_T morecore_mutex;
1456 #endif /* HAVE_MORECORE */
1457 static MLOCK_T magic_init_mutex;
1460 #define USE_LOCK_BIT (2U)
1461 #else /* USE_LOCKS */
1462 #define USE_LOCK_BIT (0U)
1463 #define INITIAL_LOCK(l)
1464 #endif /* USE_LOCKS */
1466 #if USE_LOCKS && HAVE_MORECORE
1467 #define ACQUIRE_MORECORE_LOCK() ACQUIRE_LOCK(&morecore_mutex);
1468 #define RELEASE_MORECORE_LOCK() RELEASE_LOCK(&morecore_mutex);
1469 #else /* USE_LOCKS && HAVE_MORECORE */
1470 #define ACQUIRE_MORECORE_LOCK()
1471 #define RELEASE_MORECORE_LOCK()
1472 #endif /* USE_LOCKS && HAVE_MORECORE */
1475 #define ACQUIRE_MAGIC_INIT_LOCK() ACQUIRE_LOCK(&magic_init_mutex);
1476 #define RELEASE_MAGIC_INIT_LOCK() RELEASE_LOCK(&magic_init_mutex);
1477 #else /* USE_LOCKS */
1478 #define ACQUIRE_MAGIC_INIT_LOCK()
1479 #define RELEASE_MAGIC_INIT_LOCK()
1480 #endif /* USE_LOCKS */
1483 /* ----------------------- Chunk representations ------------------------ */
1486 (The following includes lightly edited explanations by Colin Plumb.)
1488 The malloc_chunk declaration below is misleading (but accurate and
1489 necessary). It declares a "view" into memory allowing access to
1490 necessary fields at known offsets from a given base.
1492 Chunks of memory are maintained using a `boundary tag' method as
1493 originally described by Knuth. (See the paper by Paul Wilson
1494 ftp://ftp.cs.utexas.edu/pub/garbage/allocsrv.ps for a survey of such
1495 techniques.) Sizes of free chunks are stored both in the front of
1496 each chunk and at the end. This makes consolidating fragmented
1497 chunks into bigger chunks fast. The head fields also hold bits
1498 representing whether chunks are free or in use.
1500 Here are some pictures to make it clearer. They are "exploded" to
1501 show that the state of a chunk can be thought of as extending from
1502 the high 31 bits of the head field of its header through the
1503 prev_foot and PINUSE_BIT bit of the following chunk header.
1505 A chunk that's in use looks like:
1507 chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1508 | Size of previous chunk (if P = 1) |
1509 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1510 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |P|
1511 | Size of this chunk 1| +-+
1512 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1518 +- size - sizeof(size_t) available payload bytes -+
1522 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1523 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |1|
1524 | Size of next chunk (may or may not be in use) | +-+
1525 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1527 And if it's free, it looks like this:
1530 | User payload (must be in use, or we would have merged!) |
1531 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1532 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |P|
1533 | Size of this chunk 0| +-+
1534 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1536 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1538 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1540 +- size - sizeof(struct chunk) unused bytes -+
1542 chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1543 | Size of this chunk |
1544 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1545 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |0|
1546 | Size of next chunk (must be in use, or we would have merged)| +-+
1547 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1551 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1554 Note that since we always merge adjacent free chunks, the chunks
1555 adjacent to a free chunk must be in use.
1557 Given a pointer to a chunk (which can be derived trivially from the
1558 payload pointer) we can, in O(1) time, find out whether the adjacent
1559 chunks are free, and if so, unlink them from the lists that they
1560 are on and merge them with the current chunk.
1562 Chunks always begin on even word boundaries, so the mem portion
1563 (which is returned to the user) is also on an even word boundary, and
1564 thus at least double-word aligned.
1566 The P (PINUSE_BIT) bit, stored in the unused low-order bit of the
1567 chunk size (which is always a multiple of two words), is an in-use
1568 bit for the *previous* chunk. If that bit is *clear*, then the
1569 word before the current chunk size contains the previous chunk
1570 size, and can be used to find the front of the previous chunk.
1571 The very first chunk allocated always has this bit set, preventing
1572 access to non-existent (or non-owned) memory. If pinuse is set for
1573 any given chunk, then you CANNOT determine the size of the
1574 previous chunk, and might even get a memory addressing fault when
1577 The C (CINUSE_BIT) bit, stored in the unused second-lowest bit of
1578 the chunk size redundantly records whether the current chunk is
1579 inuse. This redundancy enables usage checks within free and realloc,
1580 and reduces indirection when freeing and consolidating chunks.
1582 Each freshly allocated chunk must have both cinuse and pinuse set.
1583 That is, each allocated chunk borders either a previously allocated
1584 and still in-use chunk, or the base of its memory arena. This is
1585 ensured by making all allocations from the the `lowest' part of any
1586 found chunk. Further, no free chunk physically borders another one,
1587 so each free chunk is known to be preceded and followed by either
1588 inuse chunks or the ends of memory.
1590 Note that the `foot' of the current chunk is actually represented
1591 as the prev_foot of the NEXT chunk. This makes it easier to
1592 deal with alignments etc but can be very confusing when trying
1593 to extend or adapt this code.
1595 The exceptions to all this are
1597 1. The special chunk `top' is the top-most available chunk (i.e.,
1598 the one bordering the end of available memory). It is treated
1599 specially. Top is never included in any bin, is used only if
1600 no other chunk is available, and is released back to the
1601 system if it is very large (see M_TRIM_THRESHOLD). In effect,
1602 the top chunk is treated as larger (and thus less well
1603 fitting) than any other available chunk. The top chunk
1604 doesn't update its trailing size field since there is no next
1605 contiguous chunk that would have to index off it. However,
1606 space is still allocated for it (TOP_FOOT_SIZE) to enable
1607 separation or merging when space is extended.
1609 3. Chunks allocated via mmap, which have the lowest-order bit
1610 (IS_MMAPPED_BIT) set in their prev_foot fields, and do not set
1611 PINUSE_BIT in their head fields. Because they are allocated
1612 one-by-one, each must carry its own prev_foot field, which is
1613 also used to hold the offset this chunk has within its mmapped
1614 region, which is needed to preserve alignment. Each mmapped
1615 chunk is trailed by the first two fields of a fake next-chunk
1616 for sake of usage checks.
1620 struct malloc_chunk {
1621 size_t prev_foot; /* Size of previous chunk (if free). */
1622 size_t head; /* Size and inuse bits. */
1623 struct malloc_chunk* fd; /* double links -- used only if free. */
1624 struct malloc_chunk* bk;
1627 typedef struct malloc_chunk mchunk;
1628 typedef struct malloc_chunk* mchunkptr;
1629 typedef struct malloc_chunk* sbinptr; /* The type of bins of chunks */
1630 typedef unsigned int bindex_t; /* Described below */
1631 typedef unsigned int binmap_t; /* Described below */
1632 typedef unsigned int flag_t; /* The type of various bit flag sets */
1634 /* ------------------- Chunks sizes and alignments ----------------------- */
1636 #define MCHUNK_SIZE (sizeof(mchunk))
1639 #define CHUNK_OVERHEAD (TWO_SIZE_T_SIZES)
1641 #define CHUNK_OVERHEAD (SIZE_T_SIZE)
1642 #endif /* FOOTERS */
1644 /* MMapped chunks need a second word of overhead ... */
1645 #define MMAP_CHUNK_OVERHEAD (TWO_SIZE_T_SIZES)
1646 /* ... and additional padding for fake next-chunk at foot */
1647 #define MMAP_FOOT_PAD (FOUR_SIZE_T_SIZES)
1649 /* The smallest size we can malloc is an aligned minimal chunk */
1650 #define MIN_CHUNK_SIZE\
1651 ((MCHUNK_SIZE + CHUNK_ALIGN_MASK) & ~CHUNK_ALIGN_MASK)
1653 /* conversion from malloc headers to user pointers, and back */
1654 #define chunk2mem(p) ((void*)((char*)(p) + TWO_SIZE_T_SIZES))
1655 #define mem2chunk(mem) ((mchunkptr)((char*)(mem) - TWO_SIZE_T_SIZES))
1656 /* chunk associated with aligned address A */
1657 #define align_as_chunk(A) (mchunkptr)((A) + align_offset(chunk2mem(A)))
1659 /* Bounds on request (not chunk) sizes. */
1660 #define MAX_REQUEST ((-MIN_CHUNK_SIZE) << 2)
1661 #define MIN_REQUEST (MIN_CHUNK_SIZE - CHUNK_OVERHEAD - SIZE_T_ONE)
1663 /* pad request bytes into a usable size */
1664 #define pad_request(req) \
1665 (((req) + CHUNK_OVERHEAD + CHUNK_ALIGN_MASK) & ~CHUNK_ALIGN_MASK)
1667 /* pad request, checking for minimum (but not maximum) */
1668 #define request2size(req) \
1669 (((req) < MIN_REQUEST)? MIN_CHUNK_SIZE : pad_request(req))
1672 /* ------------------ Operations on head and foot fields ----------------- */
1675 The head field of a chunk is or'ed with PINUSE_BIT when previous
1676 adjacent chunk in use, and or'ed with CINUSE_BIT if this chunk is in
1677 use. If the chunk was obtained with mmap, the prev_foot field has
1678 IS_MMAPPED_BIT set, otherwise holding the offset of the base of the
1679 mmapped region to the base of the chunk.
1682 #define PINUSE_BIT (SIZE_T_ONE)
1683 #define CINUSE_BIT (SIZE_T_TWO)
1684 #define INUSE_BITS (PINUSE_BIT|CINUSE_BIT)
1686 /* Head value for fenceposts */
1687 #define FENCEPOST_HEAD (INUSE_BITS|SIZE_T_SIZE)
1689 /* extraction of fields from head words */
1690 #define cinuse(p) ((p)->head & CINUSE_BIT)
1691 #define pinuse(p) ((p)->head & PINUSE_BIT)
1692 #define chunksize(p) ((p)->head & ~(INUSE_BITS))
1694 #define clear_pinuse(p) ((p)->head &= ~PINUSE_BIT)
1695 #define clear_cinuse(p) ((p)->head &= ~CINUSE_BIT)
1697 /* Treat space at ptr +/- offset as a chunk */
1698 #define chunk_plus_offset(p, s) ((mchunkptr)(((char*)(p)) + (s)))
1699 #define chunk_minus_offset(p, s) ((mchunkptr)(((char*)(p)) - (s)))
1701 /* Ptr to next or previous physical malloc_chunk. */
1702 #define next_chunk(p) ((mchunkptr)( ((char*)(p)) + ((p)->head & ~INUSE_BITS)))
1703 #define prev_chunk(p) ((mchunkptr)( ((char*)(p)) - ((p)->prev_foot) ))
1705 /* extract next chunk's pinuse bit */
1706 #define next_pinuse(p) ((next_chunk(p)->head) & PINUSE_BIT)
1708 /* Get/set size at footer */
1709 #define get_foot(p, s) (((mchunkptr)((char*)(p) + (s)))->prev_foot)
1710 #define set_foot(p, s) (((mchunkptr)((char*)(p) + (s)))->prev_foot = (s))
1712 /* Set size, pinuse bit, and foot */
1713 #define set_size_and_pinuse_of_free_chunk(p, s)\
1714 ((p)->head = (s|PINUSE_BIT), set_foot(p, s))
1716 /* Set size, pinuse bit, foot, and clear next pinuse */
1717 #define set_free_with_pinuse(p, s, n)\
1718 (clear_pinuse(n), set_size_and_pinuse_of_free_chunk(p, s))
1720 #define is_mmapped(p)\
1721 (!((p)->head & PINUSE_BIT) && ((p)->prev_foot & IS_MMAPPED_BIT))
1723 /* Get the internal overhead associated with chunk p */
1724 #define overhead_for(p)\
1725 (is_mmapped(p)? MMAP_CHUNK_OVERHEAD : CHUNK_OVERHEAD)
1727 /* Return true if malloced space is not necessarily cleared */
1729 #define calloc_must_clear(p) (!is_mmapped(p))
1730 #else /* MMAP_CLEARS */
1731 #define calloc_must_clear(p) (1)
1732 #endif /* MMAP_CLEARS */
1734 /* ---------------------- Overlaid data structures ----------------------- */
1737 When chunks are not in use, they are treated as nodes of either
1740 "Small" chunks are stored in circular doubly-linked lists, and look
1743 chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1744 | Size of previous chunk |
1745 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1746 `head:' | Size of chunk, in bytes |P|
1747 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1748 | Forward pointer to next chunk in list |
1749 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1750 | Back pointer to previous chunk in list |
1751 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1752 | Unused space (may be 0 bytes long) .
1755 nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1756 `foot:' | Size of chunk, in bytes |
1757 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1759 Larger chunks are kept in a form of bitwise digital trees (aka
1760 tries) keyed on chunksizes. Because malloc_tree_chunks are only for
1761 free chunks greater than 256 bytes, their size doesn't impose any
1762 constraints on user chunk sizes. Each node looks like:
1764 chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1765 | Size of previous chunk |
1766 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1767 `head:' | Size of chunk, in bytes |P|
1768 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1769 | Forward pointer to next chunk of same size |
1770 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1771 | Back pointer to previous chunk of same size |
1772 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1773 | Pointer to left child (child[0]) |
1774 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1775 | Pointer to right child (child[1]) |
1776 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1777 | Pointer to parent |
1778 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1779 | bin index of this chunk |
1780 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1783 nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1784 `foot:' | Size of chunk, in bytes |
1785 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1787 Each tree holding treenodes is a tree of unique chunk sizes. Chunks
1788 of the same size are arranged in a circularly-linked list, with only
1789 the oldest chunk (the next to be used, in our FIFO ordering)
1790 actually in the tree. (Tree members are distinguished by a non-null
1791 parent pointer.) If a chunk with the same size an an existing node
1792 is inserted, it is linked off the existing node using pointers that
1793 work in the same way as fd/bk pointers of small chunks.
1795 Each tree contains a power of 2 sized range of chunk sizes (the
1796 smallest is 0x100 <= x < 0x180), which is is divided in half at each
1797 tree level, with the chunks in the smaller half of the range (0x100
1798 <= x < 0x140 for the top nose) in the left subtree and the larger
1799 half (0x140 <= x < 0x180) in the right subtree. This is, of course,
1800 done by inspecting individual bits.
1802 Using these rules, each node's left subtree contains all smaller
1803 sizes than its right subtree. However, the node at the root of each
1804 subtree has no particular ordering relationship to either. (The
1805 dividing line between the subtree sizes is based on trie relation.)
1806 If we remove the last chunk of a given size from the interior of the
1807 tree, we need to replace it with a leaf node. The tree ordering
1808 rules permit a node to be replaced by any leaf below it.
1810 The smallest chunk in a tree (a common operation in a best-fit
1811 allocator) can be found by walking a path to the leftmost leaf in
1812 the tree. Unlike a usual binary tree, where we follow left child
1813 pointers until we reach a null, here we follow the right child
1814 pointer any time the left one is null, until we reach a leaf with
1815 both child pointers null. The smallest chunk in the tree will be
1816 somewhere along that path.
1818 The worst case number of steps to add, find, or remove a node is
1819 bounded by the number of bits differentiating chunks within
1820 bins. Under current bin calculations, this ranges from 6 up to 21
1821 (for 32 bit sizes) or up to 53 (for 64 bit sizes). The typical case
1822 is of course much better.
1825 struct malloc_tree_chunk {
1826 /* The first four fields must be compatible with malloc_chunk */
1829 struct malloc_tree_chunk* fd;
1830 struct malloc_tree_chunk* bk;
1832 struct malloc_tree_chunk* child[2];
1833 struct malloc_tree_chunk* parent;
1837 typedef struct malloc_tree_chunk tchunk;
1838 typedef struct malloc_tree_chunk* tchunkptr;
1839 typedef struct malloc_tree_chunk* tbinptr; /* The type of bins of trees */
1841 /* A little helper macro for trees */
1842 #define leftmost_child(t) ((t)->child[0] != 0? (t)->child[0] : (t)->child[1])
1844 /* ----------------------------- Segments -------------------------------- */
1847 Each malloc space may include non-contiguous segments, held in a
1848 list headed by an embedded malloc_segment record representing the
1849 top-most space. Segments also include flags holding properties of
1850 the space. Large chunks that are directly allocated by mmap are not
1851 included in this list. They are instead independently created and
1852 destroyed without otherwise keeping track of them.
1854 Segment management mainly comes into play for spaces allocated by
1855 MMAP. Any call to MMAP might or might not return memory that is
1856 adjacent to an existing segment. MORECORE normally contiguously
1857 extends the current space, so this space is almost always adjacent,
1858 which is simpler and faster to deal with. (This is why MORECORE is
1859 used preferentially to MMAP when both are available -- see
1860 sys_alloc.) When allocating using MMAP, we don't use any of the
1861 hinting mechanisms (inconsistently) supported in various
1862 implementations of unix mmap, or distinguish reserving from
1863 committing memory. Instead, we just ask for space, and exploit
1864 contiguity when we get it. It is probably possible to do
1865 better than this on some systems, but no general scheme seems
1866 to be significantly better.
1868 Management entails a simpler variant of the consolidation scheme
1869 used for chunks to reduce fragmentation -- new adjacent memory is
1870 normally prepended or appended to an existing segment. However,
1871 there are limitations compared to chunk consolidation that mostly
1872 reflect the fact that segment processing is relatively infrequent
1873 (occurring only when getting memory from system) and that we
1874 don't expect to have huge numbers of segments:
1876 * Segments are not indexed, so traversal requires linear scans. (It
1877 would be possible to index these, but is not worth the extra
1878 overhead and complexity for most programs on most platforms.)
1879 * New segments are only appended to old ones when holding top-most
1880 memory; if they cannot be prepended to others, they are held in
1883 Except for the top-most segment of an mstate, each segment record
1884 is kept at the tail of its segment. Segments are added by pushing
1885 segment records onto the list headed by &mstate.seg for the
1888 Segment flags control allocation/merge/deallocation policies:
1889 * If EXTERN_BIT set, then we did not allocate this segment,
1890 and so should not try to deallocate or merge with others.
1891 (This currently holds only for the initial segment passed
1892 into create_mspace_with_base.)
1893 * If IS_MMAPPED_BIT set, the segment may be merged with
1894 other surrounding mmapped segments and trimmed/de-allocated
1896 * If neither bit is set, then the segment was obtained using
1897 MORECORE so can be merged with surrounding MORECORE'd segments
1898 and deallocated/trimmed using MORECORE with negative arguments.
1901 struct malloc_segment {
1902 char* base; /* base address */
1903 size_t size; /* allocated size */
1904 struct malloc_segment* next; /* ptr to next segment */
1905 flag_t sflags; /* mmap and extern flag */
1908 #define is_mmapped_segment(S) ((S)->sflags & IS_MMAPPED_BIT)
1909 #define is_extern_segment(S) ((S)->sflags & EXTERN_BIT)
1911 typedef struct malloc_segment msegment;
1912 typedef struct malloc_segment* msegmentptr;
1914 /* ---------------------------- malloc_state ----------------------------- */
1917 A malloc_state holds all of the bookkeeping for a space.
1918 The main fields are:
1921 The topmost chunk of the currently active segment. Its size is
1922 cached in topsize. The actual size of topmost space is
1923 topsize+TOP_FOOT_SIZE, which includes space reserved for adding
1924 fenceposts and segment records if necessary when getting more
1925 space from the system. The size at which to autotrim top is
1926 cached from mparams in trim_check, except that it is disabled if
1929 Designated victim (dv)
1930 This is the preferred chunk for servicing small requests that
1931 don't have exact fits. It is normally the chunk split off most
1932 recently to service another small request. Its size is cached in
1933 dvsize. The link fields of this chunk are not maintained since it
1934 is not kept in a bin.
1937 An array of bin headers for free chunks. These bins hold chunks
1938 with sizes less than MIN_LARGE_SIZE bytes. Each bin contains
1939 chunks of all the same size, spaced 8 bytes apart. To simplify
1940 use in double-linked lists, each bin header acts as a malloc_chunk
1941 pointing to the real first node, if it exists (else pointing to
1942 itself). This avoids special-casing for headers. But to avoid
1943 waste, we allocate only the fd/bk pointers of bins, and then use
1944 repositioning tricks to treat these as the fields of a chunk.
1947 Treebins are pointers to the roots of trees holding a range of
1948 sizes. There are 2 equally spaced treebins for each power of two
1949 from TREE_SHIFT to TREE_SHIFT+16. The last bin holds anything
1953 There is one bit map for small bins ("smallmap") and one for
1954 treebins ("treemap). Each bin sets its bit when non-empty, and
1955 clears the bit when empty. Bit operations are then used to avoid
1956 bin-by-bin searching -- nearly all "search" is done without ever
1957 looking at bins that won't be selected. The bit maps
1958 conservatively use 32 bits per map word, even if on 64bit system.
1959 For a good description of some of the bit-based techniques used
1960 here, see Henry S. Warren Jr's book "Hacker's Delight" (and
1961 supplement at http://hackersdelight.org/). Many of these are
1962 intended to reduce the branchiness of paths through malloc etc, as
1963 well as to reduce the number of memory locations read or written.
1966 A list of segments headed by an embedded malloc_segment record
1967 representing the initial space.
1969 Address check support
1970 The least_addr field is the least address ever obtained from
1971 MORECORE or MMAP. Attempted frees and reallocs of any address less
1972 than this are trapped (unless INSECURE is defined).
1975 A cross-check field that should always hold same value as mparams.magic.
1978 Bits recording whether to use MMAP, locks, or contiguous MORECORE
1981 Each space keeps track of current and maximum system memory
1982 obtained via MORECORE or MMAP.
1985 If USE_LOCKS is defined, the "mutex" lock is acquired and released
1986 around every public call using this mspace.
1989 /* Bin types, widths and sizes */
1990 #define NSMALLBINS (32U)
1991 #define NTREEBINS (32U)
1992 #define SMALLBIN_SHIFT (3U)
1993 #define SMALLBIN_WIDTH (SIZE_T_ONE << SMALLBIN_SHIFT)
1994 #define TREEBIN_SHIFT (8U)
1995 #define MIN_LARGE_SIZE (SIZE_T_ONE << TREEBIN_SHIFT)
1996 #define MAX_SMALL_SIZE (MIN_LARGE_SIZE - SIZE_T_ONE)
1997 #define MAX_SMALL_REQUEST (MAX_SMALL_SIZE - CHUNK_ALIGN_MASK - CHUNK_OVERHEAD)
1999 struct malloc_state {
2009 mchunkptr smallbins[(NSMALLBINS+1)*2];
2010 tbinptr treebins[NTREEBINS];
2012 size_t max_footprint;
2015 MLOCK_T mutex; /* locate lock among fields that rarely change */
2016 #endif /* USE_LOCKS */
2020 typedef struct malloc_state* mstate;
2022 /* ------------- Global malloc_state and malloc_params ------------------- */
2025 malloc_params holds global properties, including those that can be
2026 dynamically set using mallopt. There is a single instance, mparams,
2027 initialized in init_mparams.
2030 struct malloc_params {
2034 size_t mmap_threshold;
2035 size_t trim_threshold;
2036 flag_t default_mflags;
2039 static struct malloc_params mparams;
2041 /* The global malloc_state used for all non-"mspace" calls */
2042 static struct malloc_state _gm_;
2044 #define is_global(M) ((M) == &_gm_)
2045 #define is_initialized(M) ((M)->top != 0)
2047 /* -------------------------- system alloc setup ------------------------- */
2049 /* Operations on mflags */
2051 #define use_lock(M) ((M)->mflags & USE_LOCK_BIT)
2052 #define enable_lock(M) ((M)->mflags |= USE_LOCK_BIT)
2053 #define disable_lock(M) ((M)->mflags &= ~USE_LOCK_BIT)
2055 #define use_mmap(M) ((M)->mflags & USE_MMAP_BIT)
2056 #define enable_mmap(M) ((M)->mflags |= USE_MMAP_BIT)
2057 #define disable_mmap(M) ((M)->mflags &= ~USE_MMAP_BIT)
2059 #define use_noncontiguous(M) ((M)->mflags & USE_NONCONTIGUOUS_BIT)
2060 #define disable_contiguous(M) ((M)->mflags |= USE_NONCONTIGUOUS_BIT)
2062 #define set_lock(M,L)\
2063 ((M)->mflags = (L)?\
2064 ((M)->mflags | USE_LOCK_BIT) :\
2065 ((M)->mflags & ~USE_LOCK_BIT))
2067 /* page-align a size */
2068 #define page_align(S)\
2069 (((S) + (mparams.page_size)) & ~(mparams.page_size - SIZE_T_ONE))
2071 /* granularity-align a size */
2072 #define granularity_align(S)\
2073 (((S) + (mparams.granularity)) & ~(mparams.granularity - SIZE_T_ONE))
2075 #define is_page_aligned(S)\
2076 (((size_t)(S) & (mparams.page_size - SIZE_T_ONE)) == 0)
2077 #define is_granularity_aligned(S)\
2078 (((size_t)(S) & (mparams.granularity - SIZE_T_ONE)) == 0)
2080 /* True if segment S holds address A */
2081 #define segment_holds(S, A)\
2082 ((char*)(A) >= S->base && (char*)(A) < S->base + S->size)
2084 /* Return segment holding given address */
2085 static msegmentptr segment_holding(mstate m, char* addr) {
2086 msegmentptr sp = &m->seg;
2088 if (addr >= sp->base && addr < sp->base + sp->size)
2090 if ((sp = sp->next) == 0)
2095 /* Return true if segment contains a segment link */
2096 static int has_segment_link(mstate m, msegmentptr ss) {
2097 msegmentptr sp = &m->seg;
2099 if ((char*)sp >= ss->base && (char*)sp < ss->base + ss->size)
2101 if ((sp = sp->next) == 0)
2106 #ifndef MORECORE_CANNOT_TRIM
2107 #define should_trim(M,s) ((s) > (M)->trim_check)
2108 #else /* MORECORE_CANNOT_TRIM */
2109 #define should_trim(M,s) (0)
2110 #endif /* MORECORE_CANNOT_TRIM */
2113 TOP_FOOT_SIZE is padding at the end of a segment, including space
2114 that may be needed to place segment records and fenceposts when new
2115 noncontiguous segments are added.
2117 #define TOP_FOOT_SIZE\
2118 (align_offset(chunk2mem(0))+pad_request(sizeof(struct malloc_segment))+MIN_CHUNK_SIZE)
2121 /* ------------------------------- Hooks -------------------------------- */
2124 PREACTION should be defined to return 0 on success, and nonzero on
2125 failure. If you are not using locking, you can redefine these to do
2131 /* Ensure locks are initialized */
2132 #define GLOBALLY_INITIALIZE() (mparams.page_size == 0 && init_mparams())
2134 #define PREACTION(M) ((GLOBALLY_INITIALIZE() || use_lock(M))? ACQUIRE_LOCK(&(M)->mutex) : 0)
2135 #define POSTACTION(M) { if (use_lock(M)) RELEASE_LOCK(&(M)->mutex); }
2136 #else /* USE_LOCKS */
2139 #define PREACTION(M) (0)
2140 #endif /* PREACTION */
2143 #define POSTACTION(M)
2144 #endif /* POSTACTION */
2146 #endif /* USE_LOCKS */
2149 CORRUPTION_ERROR_ACTION is triggered upon detected bad addresses.
2150 USAGE_ERROR_ACTION is triggered on detected bad frees and
2151 reallocs. The argument p is an address that might have triggered the
2152 fault. It is ignored by the two predefined actions, but might be
2153 useful in custom actions that try to help diagnose errors.
2156 #if PROCEED_ON_ERROR
2158 /* A count of the number of corruption errors causing resets */
2159 int malloc_corruption_error_count;
2161 /* default corruption action */
2162 static void reset_on_error(mstate m);
2164 #define CORRUPTION_ERROR_ACTION(m) reset_on_error(m)
2165 #define USAGE_ERROR_ACTION(m, p)
2167 #else /* PROCEED_ON_ERROR */
2169 #ifndef CORRUPTION_ERROR_ACTION
2170 #define CORRUPTION_ERROR_ACTION(m) ABORT
2171 #endif /* CORRUPTION_ERROR_ACTION */
2173 #ifndef USAGE_ERROR_ACTION
2174 #define USAGE_ERROR_ACTION(m,p) ABORT
2175 #endif /* USAGE_ERROR_ACTION */
2177 #endif /* PROCEED_ON_ERROR */
2179 /* -------------------------- Debugging setup ---------------------------- */
2183 #define check_free_chunk(M,P)
2184 #define check_inuse_chunk(M,P)
2185 #define check_malloced_chunk(M,P,N)
2186 #define check_mmapped_chunk(M,P)
2187 #define check_malloc_state(M)
2188 #define check_top_chunk(M,P)
2191 #define check_free_chunk(M,P) do_check_free_chunk(M,P)
2192 #define check_inuse_chunk(M,P) do_check_inuse_chunk(M,P)
2193 #define check_top_chunk(M,P) do_check_top_chunk(M,P)
2194 #define check_malloced_chunk(M,P,N) do_check_malloced_chunk(M,P,N)
2195 #define check_mmapped_chunk(M,P) do_check_mmapped_chunk(M,P)
2196 #define check_malloc_state(M) do_check_malloc_state(M)
2198 static void do_check_any_chunk(mstate m, mchunkptr p);
2199 static void do_check_top_chunk(mstate m, mchunkptr p);
2200 static void do_check_mmapped_chunk(mstate m, mchunkptr p);
2201 static void do_check_inuse_chunk(mstate m, mchunkptr p);
2202 static void do_check_free_chunk(mstate m, mchunkptr p);
2203 static void do_check_malloced_chunk(mstate m, void* mem, size_t s);
2204 static void do_check_tree(mstate m, tchunkptr t);
2205 static void do_check_treebin(mstate m, bindex_t i);
2206 static void do_check_smallbin(mstate m, bindex_t i);
2207 static void do_check_malloc_state(mstate m);
2208 static int bin_find(mstate m, mchunkptr x);
2209 static size_t traverse_and_check(mstate m);
2212 /* ---------------------------- Indexing Bins ---------------------------- */
2214 #define is_small(s) (((s) >> SMALLBIN_SHIFT) < NSMALLBINS)
2215 #define small_index(s) ((s) >> SMALLBIN_SHIFT)
2216 #define small_index2size(i) ((i) << SMALLBIN_SHIFT)
2217 #define MIN_SMALL_INDEX (small_index(MIN_CHUNK_SIZE))
2219 /* addressing by index. See above about smallbin repositioning */
2220 #define smallbin_at(M, i) ((sbinptr)((char*)&((M)->smallbins[(i)<<1])))
2221 #define treebin_at(M,i) (&((M)->treebins[i]))
2223 /* assign tree index for size S to variable I */
2224 #if defined(__GNUC__) && defined(i386)
2225 #define compute_tree_index(S, I)\
2227 size_t X = S >> TREEBIN_SHIFT;\
2230 else if (X > 0xFFFF)\
2234 __asm__("bsrl %1,%0\n\t" : "=r" (K) : "rm" (X));\
2235 I = (bindex_t)((K << 1) + ((S >> (K + (TREEBIN_SHIFT-1)) & 1)));\
2239 #define compute_tree_index(S, I)\
2241 size_t X = S >> TREEBIN_SHIFT;\
2244 else if (X > 0xFFFF)\
2247 unsigned int Y = (unsigned int)X;\
2248 unsigned int N = ((Y - 0x100) >> 16) & 8;\
2249 unsigned int K = (((Y <<= N) - 0x1000) >> 16) & 4;\
2251 N += K = (((Y <<= K) - 0x4000) >> 16) & 2;\
2252 K = 14 - N + ((Y <<= K) >> 15);\
2253 I = (K << 1) + ((S >> (K + (TREEBIN_SHIFT-1)) & 1));\
2258 /* Bit representing maximum resolved size in a treebin at i */
2259 #define bit_for_tree_index(i) \
2260 (i == NTREEBINS-1)? (SIZE_T_BITSIZE-1) : (((i) >> 1) + TREEBIN_SHIFT - 2)
2262 /* Shift placing maximum resolved bit in a treebin at i as sign bit */
2263 #define leftshift_for_tree_index(i) \
2264 ((i == NTREEBINS-1)? 0 : \
2265 ((SIZE_T_BITSIZE-SIZE_T_ONE) - (((i) >> 1) + TREEBIN_SHIFT - 2)))
2267 /* The size of the smallest chunk held in bin with index i */
2268 #define minsize_for_tree_index(i) \
2269 ((SIZE_T_ONE << (((i) >> 1) + TREEBIN_SHIFT)) | \
2270 (((size_t)((i) & SIZE_T_ONE)) << (((i) >> 1) + TREEBIN_SHIFT - 1)))
2273 /* ------------------------ Operations on bin maps ----------------------- */
2275 /* bit corresponding to given index */
2276 #define idx2bit(i) ((binmap_t)(1) << (i))
2278 /* Mark/Clear bits with given index */
2279 #define mark_smallmap(M,i) ((M)->smallmap |= idx2bit(i))
2280 #define clear_smallmap(M,i) ((M)->smallmap &= ~idx2bit(i))
2281 #define smallmap_is_marked(M,i) ((M)->smallmap & idx2bit(i))
2283 #define mark_treemap(M,i) ((M)->treemap |= idx2bit(i))
2284 #define clear_treemap(M,i) ((M)->treemap &= ~idx2bit(i))
2285 #define treemap_is_marked(M,i) ((M)->treemap & idx2bit(i))
2287 /* index corresponding to given bit */
2289 #if defined(__GNUC__) && defined(i386)
2290 #define compute_bit2idx(X, I)\
2293 __asm__("bsfl %1,%0\n\t" : "=r" (J) : "rm" (X));\
2299 #define compute_bit2idx(X, I) I = ffs(X)-1
2301 #else /* USE_BUILTIN_FFS */
2302 #define compute_bit2idx(X, I)\
2304 unsigned int Y = X - 1;\
2305 unsigned int K = Y >> (16-4) & 16;\
2306 unsigned int N = K; Y >>= K;\
2307 N += K = Y >> (8-3) & 8; Y >>= K;\
2308 N += K = Y >> (4-2) & 4; Y >>= K;\
2309 N += K = Y >> (2-1) & 2; Y >>= K;\
2310 N += K = Y >> (1-0) & 1; Y >>= K;\
2311 I = (bindex_t)(N + Y);\
2313 #endif /* USE_BUILTIN_FFS */
2316 /* isolate the least set bit of a bitmap */
2317 #define least_bit(x) ((x) & -(x))
2319 /* mask with all bits to left of least bit of x on */
2320 #define left_bits(x) ((x<<1) | -(x<<1))
2322 /* mask with all bits to left of or equal to least bit of x on */
2323 #define same_or_left_bits(x) ((x) | -(x))
2326 /* ----------------------- Runtime Check Support ------------------------- */
2329 For security, the main invariant is that malloc/free/etc never
2330 writes to a static address other than malloc_state, unless static
2331 malloc_state itself has been corrupted, which cannot occur via
2332 malloc (because of these checks). In essence this means that we
2333 believe all pointers, sizes, maps etc held in malloc_state, but
2334 check all of those linked or offsetted from other embedded data
2335 structures. These checks are interspersed with main code in a way
2336 that tends to minimize their run-time cost.
2338 When FOOTERS is defined, in addition to range checking, we also
2339 verify footer fields of inuse chunks, which can be used guarantee
2340 that the mstate controlling malloc/free is intact. This is a
2341 streamlined version of the approach described by William Robertson
2342 et al in "Run-time Detection of Heap-based Overflows" LISA'03
2343 http://www.usenix.org/events/lisa03/tech/robertson.html The footer
2344 of an inuse chunk holds the xor of its mstate and a random seed,
2345 that is checked upon calls to free() and realloc(). This is
2346 (probablistically) unguessable from outside the program, but can be
2347 computed by any code successfully malloc'ing any chunk, so does not
2348 itself provide protection against code that has already broken
2349 security through some other means. Unlike Robertson et al, we
2350 always dynamically check addresses of all offset chunks (previous,
2351 next, etc). This turns out to be cheaper than relying on hashes.
2355 /* Check if address a is at least as high as any from MORECORE or MMAP */
2356 #define ok_address(M, a) ((char*)(a) >= (M)->least_addr)
2357 /* Check if address of next chunk n is higher than base chunk p */
2358 #define ok_next(p, n) ((char*)(p) < (char*)(n))
2359 /* Check if p has its cinuse bit on */
2360 #define ok_cinuse(p) cinuse(p)
2361 /* Check if p has its pinuse bit on */
2362 #define ok_pinuse(p) pinuse(p)
2364 #else /* !INSECURE */
2365 #define ok_address(M, a) (1)
2366 #define ok_next(b, n) (1)
2367 #define ok_cinuse(p) (1)
2368 #define ok_pinuse(p) (1)
2369 #endif /* !INSECURE */
2371 #if (FOOTERS && !INSECURE)
2372 /* Check if (alleged) mstate m has expected magic field */
2373 #define ok_magic(M) ((M)->magic == mparams.magic)
2374 #else /* (FOOTERS && !INSECURE) */
2375 #define ok_magic(M) (1)
2376 #endif /* (FOOTERS && !INSECURE) */
2379 /* In gcc, use __builtin_expect to minimize impact of checks */
2381 #if defined(__GNUC__) && __GNUC__ >= 3
2382 #define RTCHECK(e) __builtin_expect(e, 1)
2384 #define RTCHECK(e) (e)
2386 #else /* !INSECURE */
2387 #define RTCHECK(e) (1)
2388 #endif /* !INSECURE */
2390 /* macros to set up inuse chunks with or without footers */
2394 #define mark_inuse_foot(M,p,s)
2396 /* Set cinuse bit and pinuse bit of next chunk */
2397 #define set_inuse(M,p,s)\
2398 ((p)->head = (((p)->head & PINUSE_BIT)|s|CINUSE_BIT),\
2399 ((mchunkptr)(((char*)(p)) + (s)))->head |= PINUSE_BIT)
2401 /* Set cinuse and pinuse of this chunk and pinuse of next chunk */
2402 #define set_inuse_and_pinuse(M,p,s)\
2403 ((p)->head = (s|PINUSE_BIT|CINUSE_BIT),\
2404 ((mchunkptr)(((char*)(p)) + (s)))->head |= PINUSE_BIT)
2406 /* Set size, cinuse and pinuse bit of this chunk */
2407 #define set_size_and_pinuse_of_inuse_chunk(M, p, s)\
2408 ((p)->head = (s|PINUSE_BIT|CINUSE_BIT))
2412 /* Set foot of inuse chunk to be xor of mstate and seed */
2413 #define mark_inuse_foot(M,p,s)\
2414 (((mchunkptr)((char*)(p) + (s)))->prev_foot = ((size_t)(M) ^ mparams.magic))
2416 #define get_mstate_for(p)\
2417 ((mstate)(((mchunkptr)((char*)(p) +\
2418 (chunksize(p))))->prev_foot ^ mparams.magic))
2420 #define set_inuse(M,p,s)\
2421 ((p)->head = (((p)->head & PINUSE_BIT)|s|CINUSE_BIT),\
2422 (((mchunkptr)(((char*)(p)) + (s)))->head |= PINUSE_BIT), \
2423 mark_inuse_foot(M,p,s))
2425 #define set_inuse_and_pinuse(M,p,s)\
2426 ((p)->head = (s|PINUSE_BIT|CINUSE_BIT),\
2427 (((mchunkptr)(((char*)(p)) + (s)))->head |= PINUSE_BIT),\
2428 mark_inuse_foot(M,p,s))
2430 #define set_size_and_pinuse_of_inuse_chunk(M, p, s)\
2431 ((p)->head = (s|PINUSE_BIT|CINUSE_BIT),\
2432 mark_inuse_foot(M, p, s))
2434 #endif /* !FOOTERS */
2436 /* ---------------------------- setting mparams -------------------------- */
2438 /* Initialize mparams */
2439 static int init_mparams(void) {
2440 if (mparams.page_size == 0) {
2443 mparams.mmap_threshold = DEFAULT_MMAP_THRESHOLD;
2444 mparams.trim_threshold = DEFAULT_TRIM_THRESHOLD;
2445 #if MORECORE_CONTIGUOUS
2446 mparams.default_mflags = USE_LOCK_BIT|USE_MMAP_BIT;
2447 #else /* MORECORE_CONTIGUOUS */
2448 mparams.default_mflags = USE_LOCK_BIT|USE_MMAP_BIT|USE_NONCONTIGUOUS_BIT;
2449 #endif /* MORECORE_CONTIGUOUS */
2451 #if (FOOTERS && !INSECURE)
2455 unsigned char buf[sizeof(size_t)];
2456 /* Try to use /dev/urandom, else fall back on using time */
2457 if ((fd = open("/dev/urandom", O_RDONLY)) >= 0 &&
2458 read(fd, buf, sizeof(buf)) == sizeof(buf)) {
2459 s = *((size_t *) buf);
2463 #endif /* USE_DEV_RANDOM */
2464 s = (size_t)(time(0) ^ (size_t)0x55555555U);
2466 s |= (size_t)8U; /* ensure nonzero */
2467 s &= ~(size_t)7U; /* improve chances of fault for bad values */
2470 #else /* (FOOTERS && !INSECURE) */
2471 s = (size_t)0x58585858U;
2472 #endif /* (FOOTERS && !INSECURE) */
2473 ACQUIRE_MAGIC_INIT_LOCK();
2474 if (mparams.magic == 0) {
2476 /* Set up lock for main malloc area */
2477 INITIAL_LOCK(&gm->mutex);
2478 gm->mflags = mparams.default_mflags;
2480 RELEASE_MAGIC_INIT_LOCK();
2483 mparams.page_size = malloc_getpagesize;
2484 mparams.granularity = ((DEFAULT_GRANULARITY != 0)?
2485 DEFAULT_GRANULARITY : mparams.page_size);
2488 SYSTEM_INFO system_info;
2489 GetSystemInfo(&system_info);
2490 mparams.page_size = system_info.dwPageSize;
2491 mparams.granularity = system_info.dwAllocationGranularity;
2495 /* Sanity-check configuration:
2496 size_t must be unsigned and as wide as pointer type.
2497 ints must be at least 4 bytes.
2498 alignment must be at least 8.
2499 Alignment, min chunk size, and page size must all be powers of 2.
2501 if ((sizeof(size_t) != sizeof(char*)) ||
2502 (MAX_SIZE_T < MIN_CHUNK_SIZE) ||
2503 (sizeof(int) < 4) ||
2504 (MALLOC_ALIGNMENT < (size_t)8U) ||
2505 ((MALLOC_ALIGNMENT & (MALLOC_ALIGNMENT-SIZE_T_ONE)) != 0) ||
2506 ((MCHUNK_SIZE & (MCHUNK_SIZE-SIZE_T_ONE)) != 0) ||
2507 ((mparams.granularity & (mparams.granularity-SIZE_T_ONE)) != 0) ||
2508 ((mparams.page_size & (mparams.page_size-SIZE_T_ONE)) != 0))
2515 /* support for mallopt */
2516 static int change_mparam(int param_number, int value) {
2517 size_t val = (size_t)value;
2519 switch(param_number) {
2520 case M_TRIM_THRESHOLD:
2521 mparams.trim_threshold = val;
2524 if (val >= mparams.page_size && ((val & (val-1)) == 0)) {
2525 mparams.granularity = val;
2530 case M_MMAP_THRESHOLD:
2531 mparams.mmap_threshold = val;
2540 /* ------------------------- Debugging Support --------------------------- */
2542 /* Check properties of any chunk, whether free, inuse, mmapped etc */
2543 static void do_check_any_chunk(mstate m, mchunkptr p) {
2544 assert((is_aligned(chunk2mem(p))) || (p->head == FENCEPOST_HEAD));
2545 assert(ok_address(m, p));
2548 /* Check properties of top chunk */
2549 static void do_check_top_chunk(mstate m, mchunkptr p) {
2550 msegmentptr sp = segment_holding(m, (char*)p);
2551 size_t sz = chunksize(p);
2553 assert((is_aligned(chunk2mem(p))) || (p->head == FENCEPOST_HEAD));
2554 assert(ok_address(m, p));
2555 assert(sz == m->topsize);
2557 assert(sz == ((sp->base + sp->size) - (char*)p) - TOP_FOOT_SIZE);
2559 assert(!next_pinuse(p));
2562 /* Check properties of (inuse) mmapped chunks */
2563 static void do_check_mmapped_chunk(mstate m, mchunkptr p) {
2564 size_t sz = chunksize(p);
2565 size_t len = (sz + (p->prev_foot & ~IS_MMAPPED_BIT) + MMAP_FOOT_PAD);
2566 assert(is_mmapped(p));
2567 assert(use_mmap(m));
2568 assert((is_aligned(chunk2mem(p))) || (p->head == FENCEPOST_HEAD));
2569 assert(ok_address(m, p));
2570 assert(!is_small(sz));
2571 assert((len & (mparams.page_size-SIZE_T_ONE)) == 0);
2572 assert(chunk_plus_offset(p, sz)->head == FENCEPOST_HEAD);
2573 assert(chunk_plus_offset(p, sz+SIZE_T_SIZE)->head == 0);
2576 /* Check properties of inuse chunks */
2577 static void do_check_inuse_chunk(mstate m, mchunkptr p) {
2578 do_check_any_chunk(m, p);
2580 assert(next_pinuse(p));
2581 /* If not pinuse and not mmapped, previous chunk has OK offset */
2582 assert(is_mmapped(p) || pinuse(p) || next_chunk(prev_chunk(p)) == p);
2584 do_check_mmapped_chunk(m, p);
2587 /* Check properties of free chunks */
2588 static void do_check_free_chunk(mstate m, mchunkptr p) {
2589 size_t sz = p->head & ~(PINUSE_BIT|CINUSE_BIT);
2590 mchunkptr next = chunk_plus_offset(p, sz);
2591 do_check_any_chunk(m, p);
2593 assert(!next_pinuse(p));
2594 assert (!is_mmapped(p));
2595 if (p != m->dv && p != m->top) {
2596 if (sz >= MIN_CHUNK_SIZE) {
2597 assert((sz & CHUNK_ALIGN_MASK) == 0);
2598 assert(is_aligned(chunk2mem(p)));
2599 assert(next->prev_foot == sz);
2601 assert (next == m->top || cinuse(next));
2602 assert(p->fd->bk == p);
2603 assert(p->bk->fd == p);
2605 else /* markers are always of size SIZE_T_SIZE */
2606 assert(sz == SIZE_T_SIZE);
2610 /* Check properties of malloced chunks at the point they are malloced */
2611 static void do_check_malloced_chunk(mstate m, void* mem, size_t s) {
2613 mchunkptr p = mem2chunk(mem);
2614 size_t sz = p->head & ~(PINUSE_BIT|CINUSE_BIT);
2615 do_check_inuse_chunk(m, p);
2616 assert((sz & CHUNK_ALIGN_MASK) == 0);
2617 assert(sz >= MIN_CHUNK_SIZE);
2619 /* unless mmapped, size is less than MIN_CHUNK_SIZE more than request */
2620 assert(is_mmapped(p) || sz < (s + MIN_CHUNK_SIZE));
2624 /* Check a tree and its subtrees. */
2625 static void do_check_tree(mstate m, tchunkptr t) {
2628 bindex_t tindex = t->index;
2629 size_t tsize = chunksize(t);
2631 compute_tree_index(tsize, idx);
2632 assert(tindex == idx);
2633 assert(tsize >= MIN_LARGE_SIZE);
2634 assert(tsize >= minsize_for_tree_index(idx));
2635 assert((idx == NTREEBINS-1) || (tsize < minsize_for_tree_index((idx+1))));
2637 do { /* traverse through chain of same-sized nodes */
2638 do_check_any_chunk(m, ((mchunkptr)u));
2639 assert(u->index == tindex);
2640 assert(chunksize(u) == tsize);
2642 assert(!next_pinuse(u));
2643 assert(u->fd->bk == u);
2644 assert(u->bk->fd == u);
2645 if (u->parent == 0) {
2646 assert(u->child[0] == 0);
2647 assert(u->child[1] == 0);
2650 assert(head == 0); /* only one node on chain has parent */
2652 assert(u->parent != u);
2653 assert (u->parent->child[0] == u ||
2654 u->parent->child[1] == u ||
2655 *((tbinptr*)(u->parent)) == u);
2656 if (u->child[0] != 0) {
2657 assert(u->child[0]->parent == u);
2658 assert(u->child[0] != u);
2659 do_check_tree(m, u->child[0]);
2661 if (u->child[1] != 0) {
2662 assert(u->child[1]->parent == u);
2663 assert(u->child[1] != u);
2664 do_check_tree(m, u->child[1]);
2666 if (u->child[0] != 0 && u->child[1] != 0) {
2667 assert(chunksize(u->child[0]) < chunksize(u->child[1]));
2675 /* Check all the chunks in a treebin. */
2676 static void do_check_treebin(mstate m, bindex_t i) {
2677 tbinptr* tb = treebin_at(m, i);
2679 int empty = (m->treemap & (1U << i)) == 0;
2683 do_check_tree(m, t);
2686 /* Check all the chunks in a smallbin. */
2687 static void do_check_smallbin(mstate m, bindex_t i) {
2688 sbinptr b = smallbin_at(m, i);
2689 mchunkptr p = b->bk;
2690 unsigned int empty = (m->smallmap & (1U << i)) == 0;
2694 for (; p != b; p = p->bk) {
2695 size_t size = chunksize(p);
2697 /* each chunk claims to be free */
2698 do_check_free_chunk(m, p);
2699 /* chunk belongs in bin */
2700 assert(small_index(size) == i);
2701 assert(p->bk == b || chunksize(p->bk) == chunksize(p));
2702 /* chunk is followed by an inuse chunk */
2704 if (q->head != FENCEPOST_HEAD)
2705 do_check_inuse_chunk(m, q);
2710 /* Find x in a bin. Used in other check functions. */
2711 static int bin_find(mstate m, mchunkptr x) {
2712 size_t size = chunksize(x);
2713 if (is_small(size)) {
2714 bindex_t sidx = small_index(size);
2715 sbinptr b = smallbin_at(m, sidx);
2716 if (smallmap_is_marked(m, sidx)) {
2721 } while ((p = p->fd) != b);
2726 compute_tree_index(size, tidx);
2727 if (treemap_is_marked(m, tidx)) {
2728 tchunkptr t = *treebin_at(m, tidx);
2729 size_t sizebits = size << leftshift_for_tree_index(tidx);
2730 while (t != 0 && chunksize(t) != size) {
2731 t = t->child[(sizebits >> (SIZE_T_BITSIZE-SIZE_T_ONE)) & 1];
2737 if (u == (tchunkptr)x)
2739 } while ((u = u->fd) != t);
2746 /* Traverse each chunk and check it; return total */
2747 static size_t traverse_and_check(mstate m) {
2749 if (is_initialized(m)) {
2750 msegmentptr s = &m->seg;
2751 sum += m->topsize + TOP_FOOT_SIZE;
2753 mchunkptr q = align_as_chunk(s->base);
2754 mchunkptr lastq = 0;
2756 while (segment_holds(s, q) &&
2757 q != m->top && q->head != FENCEPOST_HEAD) {
2758 sum += chunksize(q);
2760 assert(!bin_find(m, q));
2761 do_check_inuse_chunk(m, q);
2764 assert(q == m->dv || bin_find(m, q));
2765 assert(lastq == 0 || cinuse(lastq)); /* Not 2 consecutive free */
2766 do_check_free_chunk(m, q);
2777 /* Check all properties of malloc_state. */
2778 static void do_check_malloc_state(mstate m) {
2782 for (i = 0; i < NSMALLBINS; ++i)
2783 do_check_smallbin(m, i);
2784 for (i = 0; i < NTREEBINS; ++i)
2785 do_check_treebin(m, i);
2787 if (m->dvsize != 0) { /* check dv chunk */
2788 do_check_any_chunk(m, m->dv);
2789 assert(m->dvsize == chunksize(m->dv));
2790 assert(m->dvsize >= MIN_CHUNK_SIZE);
2791 assert(bin_find(m, m->dv) == 0);
2794 if (m->top != 0) { /* check top chunk */
2795 do_check_top_chunk(m, m->top);
2796 assert(m->topsize == chunksize(m->top));
2797 assert(m->topsize > 0);
2798 assert(bin_find(m, m->top) == 0);
2801 total = traverse_and_check(m);
2802 assert(total <= m->footprint);
2803 assert(m->footprint <= m->max_footprint);
2807 /* ----------------------------- statistics ------------------------------ */
2810 static struct mallinfo internal_mallinfo(mstate m) {
2811 struct mallinfo nm = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
2812 if (!PREACTION(m)) {
2813 check_malloc_state(m);
2814 if (is_initialized(m)) {
2815 size_t nfree = SIZE_T_ONE; /* top always free */
2816 size_t mfree = m->topsize + TOP_FOOT_SIZE;
2818 msegmentptr s = &m->seg;
2820 mchunkptr q = align_as_chunk(s->base);
2821 while (segment_holds(s, q) &&
2822 q != m->top && q->head != FENCEPOST_HEAD) {
2823 size_t sz = chunksize(q);
2836 nm.hblkhd = m->footprint - sum;
2837 nm.usmblks = m->max_footprint;
2838 nm.uordblks = m->footprint - mfree;
2839 nm.fordblks = mfree;
2840 nm.keepcost = m->topsize;
2847 #endif /* !NO_MALLINFO */
2850 static void internal_malloc_stats(mstate m) {
2851 if (!PREACTION(m)) {
2855 check_malloc_state(m);
2856 if (is_initialized(m)) {
2857 msegmentptr s = &m->seg;
2858 maxfp = m->max_footprint;
2860 used = fp - (m->topsize + TOP_FOOT_SIZE);
2863 mchunkptr q = align_as_chunk(s->base);
2864 while (segment_holds(s, q) &&
2865 q != m->top && q->head != FENCEPOST_HEAD) {
2867 used -= chunksize(q);
2874 fprintf(stderr, "max system bytes = %10lu\n", (unsigned long)(maxfp));
2875 fprintf(stderr, "system bytes = %10lu\n", (unsigned long)(fp));
2876 fprintf(stderr, "in use bytes = %10lu\n", (unsigned long)(used));
2883 /* ----------------------- Operations on smallbins ----------------------- */
2886 Various forms of linking and unlinking are defined as macros. Even
2887 the ones for trees, which are very long but have very short typical
2888 paths. This is ugly but reduces reliance on inlining support of
2892 /* Link a free chunk into a smallbin */
2893 #define insert_small_chunk(M, P, S) {\
2894 bindex_t I = small_index(S);\
2895 mchunkptr B = smallbin_at(M, I);\
2897 assert(S >= MIN_CHUNK_SIZE);\
2898 if (!smallmap_is_marked(M, I))\
2899 mark_smallmap(M, I);\
2900 else if (RTCHECK(ok_address(M, B->fd)))\
2903 CORRUPTION_ERROR_ACTION(M);\
2911 /* Unlink a chunk from a smallbin */
2912 #define unlink_small_chunk(M, P, S) {\
2913 mchunkptr F = P->fd;\
2914 mchunkptr B = P->bk;\
2915 bindex_t I = small_index(S);\
2918 assert(chunksize(P) == small_index2size(I));\
2920 clear_smallmap(M, I);\
2921 else if (RTCHECK((F == smallbin_at(M,I) || ok_address(M, F)) &&\
2922 (B == smallbin_at(M,I) || ok_address(M, B)))) {\
2927 CORRUPTION_ERROR_ACTION(M);\
2931 /* Unlink the first chunk from a smallbin */
2932 #define unlink_first_small_chunk(M, B, P, I) {\
2933 mchunkptr F = P->fd;\
2936 assert(chunksize(P) == small_index2size(I));\
2938 clear_smallmap(M, I);\
2939 else if (RTCHECK(ok_address(M, F))) {\
2944 CORRUPTION_ERROR_ACTION(M);\
2948 /* Replace dv node, binning the old one */
2949 /* Used only when dvsize known to be small */
2950 #define replace_dv(M, P, S) {\
2951 size_t DVS = M->dvsize;\
2953 mchunkptr DV = M->dv;\
2954 assert(is_small(DVS));\
2955 insert_small_chunk(M, DV, DVS);\
2961 /* ------------------------- Operations on trees ------------------------- */
2963 /* Insert chunk into tree */
2964 #define insert_large_chunk(M, X, S) {\
2967 compute_tree_index(S, I);\
2968 H = treebin_at(M, I);\
2970 X->child[0] = X->child[1] = 0;\
2971 if (!treemap_is_marked(M, I)) {\
2972 mark_treemap(M, I);\
2974 X->parent = (tchunkptr)H;\
2979 size_t K = S << leftshift_for_tree_index(I);\
2981 if (chunksize(T) != S) {\
2982 tchunkptr* C = &(T->child[(K >> (SIZE_T_BITSIZE-SIZE_T_ONE)) & 1]);\
2986 else if (RTCHECK(ok_address(M, C))) {\
2993 CORRUPTION_ERROR_ACTION(M);\
2998 tchunkptr F = T->fd;\
2999 if (RTCHECK(ok_address(M, T) && ok_address(M, F))) {\
3007 CORRUPTION_ERROR_ACTION(M);\
3018 1. If x is a chained node, unlink it from its same-sized fd/bk links
3019 and choose its bk node as its replacement.
3020 2. If x was the last node of its size, but not a leaf node, it must
3021 be replaced with a leaf node (not merely one with an open left or
3022 right), to make sure that lefts and rights of descendents
3023 correspond properly to bit masks. We use the rightmost descendent
3024 of x. We could use any other leaf, but this is easy to locate and
3025 tends to counteract removal of leftmosts elsewhere, and so keeps
3026 paths shorter than minimally guaranteed. This doesn't loop much
3027 because on average a node in a tree is near the bottom.
3028 3. If x is the base of a chain (i.e., has parent links) relink
3029 x's parent and children to x's replacement (or null if none).
3032 #define unlink_large_chunk(M, X) {\
3033 tchunkptr XP = X->parent;\
3036 tchunkptr F = X->fd;\
3038 if (RTCHECK(ok_address(M, F))) {\
3043 CORRUPTION_ERROR_ACTION(M);\
3048 if (((R = *(RP = &(X->child[1]))) != 0) ||\
3049 ((R = *(RP = &(X->child[0]))) != 0)) {\
3051 while ((*(CP = &(R->child[1])) != 0) ||\
3052 (*(CP = &(R->child[0])) != 0)) {\
3055 if (RTCHECK(ok_address(M, RP)))\
3058 CORRUPTION_ERROR_ACTION(M);\
3063 tbinptr* H = treebin_at(M, X->index);\
3065 if ((*H = R) == 0) \
3066 clear_treemap(M, X->index);\
3068 else if (RTCHECK(ok_address(M, XP))) {\
3069 if (XP->child[0] == X) \
3075 CORRUPTION_ERROR_ACTION(M);\
3077 if (RTCHECK(ok_address(M, R))) {\
3080 if ((C0 = X->child[0]) != 0) {\
3081 if (RTCHECK(ok_address(M, C0))) {\
3086 CORRUPTION_ERROR_ACTION(M);\
3088 if ((C1 = X->child[1]) != 0) {\
3089 if (RTCHECK(ok_address(M, C1))) {\
3094 CORRUPTION_ERROR_ACTION(M);\
3098 CORRUPTION_ERROR_ACTION(M);\
3103 /* Relays to large vs small bin operations */
3105 #define insert_chunk(M, P, S)\
3106 if (is_small(S)) insert_small_chunk(M, P, S)\
3107 else { tchunkptr TP = (tchunkptr)(P); insert_large_chunk(M, TP, S); }
3109 #define unlink_chunk(M, P, S)\
3110 if (is_small(S)) unlink_small_chunk(M, P, S)\
3111 else { tchunkptr TP = (tchunkptr)(P); unlink_large_chunk(M, TP); }
3114 /* Relays to internal calls to malloc/free from realloc, memalign etc */
3117 #define internal_malloc(m, b) mspace_malloc(m, b)
3118 #define internal_free(m, mem) mspace_free(m,mem);
3119 #else /* ONLY_MSPACES */
3121 #define internal_malloc(m, b)\
3122 (m == gm)? dlmalloc(b) : mspace_malloc(m, b)
3123 #define internal_free(m, mem)\
3124 if (m == gm) dlfree(mem); else mspace_free(m,mem);
3126 #define internal_malloc(m, b) dlmalloc(b)
3127 #define internal_free(m, mem) dlfree(mem)
3128 #endif /* MSPACES */
3129 #endif /* ONLY_MSPACES */
3131 /* ----------------------- Direct-mmapping chunks ----------------------- */
3134 Directly mmapped chunks are set up with an offset to the start of
3135 the mmapped region stored in the prev_foot field of the chunk. This
3136 allows reconstruction of the required argument to MUNMAP when freed,
3137 and also allows adjustment of the returned chunk to meet alignment
3138 requirements (especially in memalign). There is also enough space
3139 allocated to hold a fake next chunk of size SIZE_T_SIZE to maintain
3140 the PINUSE bit so frees can be checked.
3143 /* Malloc using mmap */
3144 static void* mmap_alloc(mstate m, size_t nb) {
3145 size_t mmsize = granularity_align(nb + SIX_SIZE_T_SIZES + CHUNK_ALIGN_MASK);
3146 if (mmsize > nb) { /* Check for wrap around 0 */
3147 char* mm = (char*)(DIRECT_MMAP(mmsize));
3149 size_t offset = align_offset(chunk2mem(mm));
3150 size_t psize = mmsize - offset - MMAP_FOOT_PAD;
3151 mchunkptr p = (mchunkptr)(mm + offset);
3152 p->prev_foot = offset | IS_MMAPPED_BIT;
3153 (p)->head = (psize|CINUSE_BIT);
3154 mark_inuse_foot(m, p, psize);
3155 chunk_plus_offset(p, psize)->head = FENCEPOST_HEAD;
3156 chunk_plus_offset(p, psize+SIZE_T_SIZE)->head = 0;
3158 if (mm < m->least_addr)
3160 if ((m->footprint += mmsize) > m->max_footprint)
3161 m->max_footprint = m->footprint;
3162 assert(is_aligned(chunk2mem(p)));
3163 check_mmapped_chunk(m, p);
3164 return chunk2mem(p);
3172 /* Realloc using mmap */
3173 static mchunkptr mmap_resize(mstate m, mchunkptr oldp, size_t nb) {
3174 size_t oldsize = chunksize(oldp);
3175 if (is_small(nb)) /* Can't shrink mmap regions below small size */
3177 /* Keep old chunk if big enough but not too big */
3178 if (oldsize >= nb + SIZE_T_SIZE &&
3179 (oldsize - nb) <= (mparams.granularity << 1))
3182 size_t offset = oldp->prev_foot & ~IS_MMAPPED_BIT;
3183 size_t oldmmsize = oldsize + offset + MMAP_FOOT_PAD;
3184 size_t newmmsize = granularity_align(nb + SIX_SIZE_T_SIZES +
3186 char* cp = (char*)CALL_MREMAP((char*)oldp - offset,
3187 oldmmsize, newmmsize, 1);
3189 mchunkptr newp = (mchunkptr)(cp + offset);
3190 size_t psize = newmmsize - offset - MMAP_FOOT_PAD;
3191 newp->head = (psize|CINUSE_BIT);
3192 mark_inuse_foot(m, newp, psize);
3193 chunk_plus_offset(newp, psize)->head = FENCEPOST_HEAD;
3194 chunk_plus_offset(newp, psize+SIZE_T_SIZE)->head = 0;
3196 if (cp < m->least_addr)
3198 if ((m->footprint += newmmsize - oldmmsize) > m->max_footprint)
3199 m->max_footprint = m->footprint;
3200 check_mmapped_chunk(m, newp);
3209 /* -------------------------- mspace management -------------------------- */
3211 /* Initialize top chunk and its size */
3212 static void init_top(mstate m, mchunkptr p, size_t psize) {
3213 /* Ensure alignment */
3214 size_t offset = align_offset(chunk2mem(p));
3215 p = (mchunkptr)((char*)p + offset);
3220 p->head = psize | PINUSE_BIT;
3221 /* set size of fake trailing chunk holding overhead space only once */
3222 chunk_plus_offset(p, psize)->head = TOP_FOOT_SIZE;
3223 m->trim_check = mparams.trim_threshold; /* reset on each update */
3226 /* Initialize bins for a new mstate that is otherwise zeroed out */
3227 static void init_bins(mstate m) {
3228 /* Establish circular links for smallbins */
3230 for (i = 0; i < NSMALLBINS; ++i) {
3231 sbinptr bin = smallbin_at(m,i);
3232 bin->fd = bin->bk = bin;
3236 #if PROCEED_ON_ERROR
3238 /* default corruption action */
3239 static void reset_on_error(mstate m) {
3241 ++malloc_corruption_error_count;
3242 /* Reinitialize fields to forget about all memory */
3243 m->smallbins = m->treebins = 0;
3244 m->dvsize = m->topsize = 0;
3249 for (i = 0; i < NTREEBINS; ++i)
3250 *treebin_at(m, i) = 0;
3253 #endif /* PROCEED_ON_ERROR */
3255 /* Allocate chunk and prepend remainder with chunk in successor base. */
3256 static void* prepend_alloc(mstate m, char* newbase, char* oldbase,
3258 mchunkptr p = align_as_chunk(newbase);
3259 mchunkptr oldfirst = align_as_chunk(oldbase);
3260 size_t psize = (char*)oldfirst - (char*)p;
3261 mchunkptr q = chunk_plus_offset(p, nb);
3262 size_t qsize = psize - nb;
3263 set_size_and_pinuse_of_inuse_chunk(m, p, nb);
3265 assert((char*)oldfirst > (char*)q);
3266 assert(pinuse(oldfirst));
3267 assert(qsize >= MIN_CHUNK_SIZE);
3269 /* consolidate remainder with first chunk of old base */
3270 if (oldfirst == m->top) {
3271 size_t tsize = m->topsize += qsize;
3273 q->head = tsize | PINUSE_BIT;
3274 check_top_chunk(m, q);
3276 else if (oldfirst == m->dv) {
3277 size_t dsize = m->dvsize += qsize;
3279 set_size_and_pinuse_of_free_chunk(q, dsize);
3282 if (!cinuse(oldfirst)) {
3283 size_t nsize = chunksize(oldfirst);
3284 unlink_chunk(m, oldfirst, nsize);
3285 oldfirst = chunk_plus_offset(oldfirst, nsize);
3288 set_free_with_pinuse(q, qsize, oldfirst);
3289 insert_chunk(m, q, qsize);
3290 check_free_chunk(m, q);
3293 check_malloced_chunk(m, chunk2mem(p), nb);
3294 return chunk2mem(p);
3298 /* Add a segment to hold a new noncontiguous region */
3299 static void add_segment(mstate m, char* tbase, size_t tsize, flag_t mmapped) {
3300 /* Determine locations and sizes of segment, fenceposts, old top */
3301 char* old_top = (char*)m->top;
3302 msegmentptr oldsp = segment_holding(m, old_top);
3303 char* old_end = oldsp->base + oldsp->size;
3304 size_t ssize = pad_request(sizeof(struct malloc_segment));
3305 char* rawsp = old_end - (ssize + FOUR_SIZE_T_SIZES + CHUNK_ALIGN_MASK);
3306 size_t offset = align_offset(chunk2mem(rawsp));
3307 char* asp = rawsp + offset;
3308 char* csp = (asp < (old_top + MIN_CHUNK_SIZE))? old_top : asp;
3309 mchunkptr sp = (mchunkptr)csp;
3310 msegmentptr ss = (msegmentptr)(chunk2mem(sp));
3311 mchunkptr tnext = chunk_plus_offset(sp, ssize);
3312 mchunkptr p = tnext;
3315 /* reset top to new space */
3316 init_top(m, (mchunkptr)tbase, tsize - TOP_FOOT_SIZE);
3318 /* Set up segment record */
3319 assert(is_aligned(ss));
3320 set_size_and_pinuse_of_inuse_chunk(m, sp, ssize);
3321 *ss = m->seg; /* Push current record */
3322 m->seg.base = tbase;
3323 m->seg.size = tsize;
3324 m->seg.sflags = mmapped;
3327 /* Insert trailing fenceposts */
3329 mchunkptr nextp = chunk_plus_offset(p, SIZE_T_SIZE);
3330 p->head = FENCEPOST_HEAD;
3332 if ((char*)(&(nextp->head)) < old_end)
3337 assert(nfences >= 2);
3339 /* Insert the rest of old top into a bin as an ordinary free chunk */
3340 if (csp != old_top) {
3341 mchunkptr q = (mchunkptr)old_top;
3342 size_t psize = csp - old_top;
3343 mchunkptr tn = chunk_plus_offset(q, psize);
3344 set_free_with_pinuse(q, psize, tn);
3345 insert_chunk(m, q, psize);
3348 check_top_chunk(m, m->top);
3351 /* -------------------------- System allocation -------------------------- */
3353 /* Get memory from system using MORECORE or MMAP */
3354 static void* sys_alloc(mstate m, size_t nb) {
3355 char* tbase = CMFAIL;
3357 flag_t mmap_flag = 0;
3361 /* Directly map large chunks */
3362 if (use_mmap(m) && nb >= mparams.mmap_threshold) {
3363 void* mem = mmap_alloc(m, nb);
3369 Try getting memory in any of three ways (in most-preferred to
3370 least-preferred order):
3371 1. A call to MORECORE that can normally contiguously extend memory.
3372 (disabled if not MORECORE_CONTIGUOUS or not HAVE_MORECORE or
3373 or main space is mmapped or a previous contiguous call failed)
3374 2. A call to MMAP new space (disabled if not HAVE_MMAP).
3375 Note that under the default settings, if MORECORE is unable to
3376 fulfill a request, and HAVE_MMAP is true, then mmap is
3377 used as a noncontiguous system allocator. This is a useful backup
3378 strategy for systems with holes in address spaces -- in this case
3379 sbrk cannot contiguously expand the heap, but mmap may be able to
3381 3. A call to MORECORE that cannot usually contiguously extend memory.
3382 (disabled if not HAVE_MORECORE)
3385 if (MORECORE_CONTIGUOUS && !use_noncontiguous(m)) {
3387 msegmentptr ss = (m->top == 0)? 0 : segment_holding(m, (char*)m->top);
3389 ACQUIRE_MORECORE_LOCK();
3391 if (ss == 0) { /* First time through or recovery */
3392 char* base = (char*)CALL_MORECORE(0);
3393 if (base != CMFAIL) {
3394 asize = granularity_align(nb + TOP_FOOT_SIZE + SIZE_T_ONE);
3395 /* Adjust to end on a page boundary */
3396 if (!is_page_aligned(base))
3397 asize += (page_align((size_t)base) - (size_t)base);
3398 /* Can't call MORECORE if size is negative when treated as signed */
3399 if (asize < HALF_MAX_SIZE_T &&
3400 (br = (char*)(CALL_MORECORE(asize))) == base) {
3407 /* Subtract out existing available top space from MORECORE request. */
3408 asize = granularity_align(nb - m->topsize + TOP_FOOT_SIZE + SIZE_T_ONE);
3409 /* Use mem here only if it did continuously extend old space */
3410 if (asize < HALF_MAX_SIZE_T &&
3411 (br = (char*)(CALL_MORECORE(asize))) == ss->base+ss->size) {
3417 if (tbase == CMFAIL) { /* Cope with partial failure */
3418 if (br != CMFAIL) { /* Try to use/extend the space we did get */
3419 if (asize < HALF_MAX_SIZE_T &&
3420 asize < nb + TOP_FOOT_SIZE + SIZE_T_ONE) {
3421 size_t esize = granularity_align(nb + TOP_FOOT_SIZE + SIZE_T_ONE - asize);
3422 if (esize < HALF_MAX_SIZE_T) {
3423 char* end = (char*)CALL_MORECORE(esize);
3426 else { /* Can't use; try to release */
3427 CALL_MORECORE(-asize);
3433 if (br != CMFAIL) { /* Use the space we did get */
3438 disable_contiguous(m); /* Don't try contiguous path in the future */
3441 RELEASE_MORECORE_LOCK();
3444 if (HAVE_MMAP && tbase == CMFAIL) { /* Try MMAP */
3445 size_t req = nb + TOP_FOOT_SIZE + SIZE_T_ONE;
3446 size_t rsize = granularity_align(req);
3447 if (rsize > nb) { /* Fail if wraps around zero */
3448 char* mp = (char*)(CALL_MMAP(rsize));
3452 mmap_flag = IS_MMAPPED_BIT;
3457 if (HAVE_MORECORE && tbase == CMFAIL) { /* Try noncontiguous MORECORE */
3458 size_t asize = granularity_align(nb + TOP_FOOT_SIZE + SIZE_T_ONE);
3459 if (asize < HALF_MAX_SIZE_T) {
3462 ACQUIRE_MORECORE_LOCK();
3463 br = (char*)(CALL_MORECORE(asize));
3464 end = (char*)(CALL_MORECORE(0));
3465 RELEASE_MORECORE_LOCK();
3466 if (br != CMFAIL && end != CMFAIL && br < end) {
3467 size_t ssize = end - br;
3468 if (ssize > nb + TOP_FOOT_SIZE) {
3476 if (tbase != CMFAIL) {
3478 if ((m->footprint += tsize) > m->max_footprint)
3479 m->max_footprint = m->footprint;
3481 if (!is_initialized(m)) { /* first-time initialization */
3482 m->seg.base = m->least_addr = tbase;
3483 m->seg.size = tsize;
3484 m->seg.sflags = mmap_flag;
3485 m->magic = mparams.magic;
3488 init_top(m, (mchunkptr)tbase, tsize - TOP_FOOT_SIZE);
3490 /* Offset top by embedded malloc_state */
3491 mchunkptr mn = next_chunk(mem2chunk(m));
3492 init_top(m, mn, (size_t)((tbase + tsize) - (char*)mn) -TOP_FOOT_SIZE);
3497 /* Try to merge with an existing segment */
3498 msegmentptr sp = &m->seg;
3499 while (sp != 0 && tbase != sp->base + sp->size)
3502 !is_extern_segment(sp) &&
3503 (sp->sflags & IS_MMAPPED_BIT) == mmap_flag &&
3504 segment_holds(sp, m->top)) { /* append */
3506 init_top(m, m->top, m->topsize + tsize);
3509 if (tbase < m->least_addr)
3510 m->least_addr = tbase;
3512 while (sp != 0 && sp->base != tbase + tsize)
3515 !is_extern_segment(sp) &&
3516 (sp->sflags & IS_MMAPPED_BIT) == mmap_flag) {
3517 char* oldbase = sp->base;
3520 return prepend_alloc(m, tbase, oldbase, nb);
3523 add_segment(m, tbase, tsize, mmap_flag);
3527 if (nb < m->topsize) { /* Allocate from new or extended top space */
3528 size_t rsize = m->topsize -= nb;
3529 mchunkptr p = m->top;
3530 mchunkptr r = m->top = chunk_plus_offset(p, nb);
3531 r->head = rsize | PINUSE_BIT;
3532 set_size_and_pinuse_of_inuse_chunk(m, p, nb);
3533 check_top_chunk(m, m->top);
3534 check_malloced_chunk(m, chunk2mem(p), nb);
3535 return chunk2mem(p);
3539 MALLOC_FAILURE_ACTION;
3543 /* ----------------------- system deallocation -------------------------- */
3545 /* Unmap and unlink any mmapped segments that don't contain used chunks */
3546 static size_t release_unused_segments(mstate m) {
3547 size_t released = 0;
3548 msegmentptr pred = &m->seg;
3549 msegmentptr sp = pred->next;
3551 char* base = sp->base;
3552 size_t size = sp->size;
3553 msegmentptr next = sp->next;
3554 if (is_mmapped_segment(sp) && !is_extern_segment(sp)) {
3555 mchunkptr p = align_as_chunk(base);
3556 size_t psize = chunksize(p);
3557 /* Can unmap if first chunk holds entire segment and not pinned */
3558 if (!cinuse(p) && (char*)p + psize >= base + size - TOP_FOOT_SIZE) {
3559 tchunkptr tp = (tchunkptr)p;
3560 assert(segment_holds(sp, (char*)sp));
3566 unlink_large_chunk(m, tp);
3568 if (CALL_MUNMAP(base, size) == 0) {
3570 m->footprint -= size;
3571 /* unlink obsoleted record */
3575 else { /* back out if cannot unmap */
3576 insert_large_chunk(m, tp, psize);
3586 static int sys_trim(mstate m, size_t pad) {
3587 size_t released = 0;
3588 if (pad < MAX_REQUEST && is_initialized(m)) {
3589 pad += TOP_FOOT_SIZE; /* ensure enough room for segment overhead */
3591 if (m->topsize > pad) {
3592 /* Shrink top space in granularity-size units, keeping at least one */
3593 size_t unit = mparams.granularity;
3594 size_t extra = ((m->topsize - pad + (unit - SIZE_T_ONE)) / unit -
3596 msegmentptr sp = segment_holding(m, (char*)m->top);
3598 if (!is_extern_segment(sp)) {
3599 if (is_mmapped_segment(sp)) {
3601 sp->size >= extra &&
3602 !has_segment_link(m, sp)) { /* can't shrink if pinned */
3603 size_t newsize = sp->size - extra;
3604 /* Prefer mremap, fall back to munmap */
3605 if ((CALL_MREMAP(sp->base, sp->size, newsize, 0) != MFAIL) ||
3606 (CALL_MUNMAP(sp->base + newsize, extra) == 0)) {
3611 else if (HAVE_MORECORE) {
3612 if (extra >= HALF_MAX_SIZE_T) /* Avoid wrapping negative */
3613 extra = (HALF_MAX_SIZE_T) + SIZE_T_ONE - unit;
3614 ACQUIRE_MORECORE_LOCK();
3616 /* Make sure end of memory is where we last set it. */
3617 char* old_br = (char*)(CALL_MORECORE(0));
3618 if (old_br == sp->base + sp->size) {
3619 char* rel_br = (char*)(CALL_MORECORE(-extra));
3620 char* new_br = (char*)(CALL_MORECORE(0));
3621 if (rel_br != CMFAIL && new_br < old_br)
3622 released = old_br - new_br;
3625 RELEASE_MORECORE_LOCK();
3629 if (released != 0) {
3630 sp->size -= released;
3631 m->footprint -= released;
3632 init_top(m, m->top, m->topsize - released);
3633 check_top_chunk(m, m->top);
3637 /* Unmap any unused mmapped segments */
3639 released += release_unused_segments(m);
3641 /* On failure, disable autotrim to avoid repeated failed future calls */
3643 m->trim_check = MAX_SIZE_T;
3646 return (released != 0)? 1 : 0;
3649 /* ---------------------------- malloc support --------------------------- */
3651 /* allocate a large request from the best fitting chunk in a treebin */
3652 static void* tmalloc_large(mstate m, size_t nb) {
3654 size_t rsize = -nb; /* Unsigned negation */
3657 compute_tree_index(nb, idx);
3659 if ((t = *treebin_at(m, idx)) != 0) {
3660 /* Traverse tree for this bin looking for node with size == nb */
3661 size_t sizebits = nb << leftshift_for_tree_index(idx);
3662 tchunkptr rst = 0; /* The deepest untaken right subtree */
3665 size_t trem = chunksize(t) - nb;
3668 if ((rsize = trem) == 0)
3672 t = t->child[(sizebits >> (SIZE_T_BITSIZE-SIZE_T_ONE)) & 1];
3673 if (rt != 0 && rt != t)
3676 t = rst; /* set t to least subtree holding sizes > nb */
3683 if (t == 0 && v == 0) { /* set t to root of next non-empty treebin */
3684 binmap_t leftbits = left_bits(idx2bit(idx)) & m->treemap;
3685 if (leftbits != 0) {
3687 binmap_t leastbit = least_bit(leftbits);
3688 compute_bit2idx(leastbit, i);
3689 t = *treebin_at(m, i);
3693 while (t != 0) { /* find smallest of tree or subtree */
3694 size_t trem = chunksize(t) - nb;
3699 t = leftmost_child(t);
3702 /* If dv is a better fit, return 0 so malloc will use it */
3703 if (v != 0 && rsize < (size_t)(m->dvsize - nb)) {
3704 if (RTCHECK(ok_address(m, v))) { /* split */
3705 mchunkptr r = chunk_plus_offset(v, nb);
3706 assert(chunksize(v) == rsize + nb);
3707 if (RTCHECK(ok_next(v, r))) {
3708 unlink_large_chunk(m, v);
3709 if (rsize < MIN_CHUNK_SIZE)
3710 set_inuse_and_pinuse(m, v, (rsize + nb));
3712 set_size_and_pinuse_of_inuse_chunk(m, v, nb);
3713 set_size_and_pinuse_of_free_chunk(r, rsize);
3714 insert_chunk(m, r, rsize);
3716 return chunk2mem(v);
3719 CORRUPTION_ERROR_ACTION(m);
3724 /* allocate a small request from the best fitting chunk in a treebin */
3725 static void* tmalloc_small(mstate m, size_t nb) {
3729 binmap_t leastbit = least_bit(m->treemap);
3730 compute_bit2idx(leastbit, i);
3732 v = t = *treebin_at(m, i);
3733 rsize = chunksize(t) - nb;
3735 while ((t = leftmost_child(t)) != 0) {
3736 size_t trem = chunksize(t) - nb;
3743 if (RTCHECK(ok_address(m, v))) {
3744 mchunkptr r = chunk_plus_offset(v, nb);
3745 assert(chunksize(v) == rsize + nb);
3746 if (RTCHECK(ok_next(v, r))) {
3747 unlink_large_chunk(m, v);
3748 if (rsize < MIN_CHUNK_SIZE)
3749 set_inuse_and_pinuse(m, v, (rsize + nb));
3751 set_size_and_pinuse_of_inuse_chunk(m, v, nb);
3752 set_size_and_pinuse_of_free_chunk(r, rsize);
3753 replace_dv(m, r, rsize);
3755 return chunk2mem(v);
3759 CORRUPTION_ERROR_ACTION(m);
3763 /* --------------------------- realloc support --------------------------- */
3767 static void* internal_realloc(mstate m, void* oldmem, size_t bytes) {
3768 if (bytes >= MAX_REQUEST) {
3769 MALLOC_FAILURE_ACTION;
3772 if (!PREACTION(m)) {
3773 mchunkptr oldp = mem2chunk(oldmem);
3774 size_t oldsize = chunksize(oldp);
3775 mchunkptr next = chunk_plus_offset(oldp, oldsize);
3779 /* Try to either shrink or extend into top. Else malloc-copy-free */
3781 if (RTCHECK(ok_address(m, oldp) && ok_cinuse(oldp) &&
3782 ok_next(oldp, next) && ok_pinuse(next))) {
3783 size_t nb = request2size(bytes);
3784 if (is_mmapped(oldp))
3785 newp = mmap_resize(m, oldp, nb);
3786 else if (oldsize >= nb) { /* already big enough */
3787 size_t rsize = oldsize - nb;
3789 if (rsize >= MIN_CHUNK_SIZE) {
3790 mchunkptr remainder = chunk_plus_offset(newp, nb);
3791 set_inuse(m, newp, nb);
3792 set_inuse(m, remainder, rsize);
3793 extra = chunk2mem(remainder);
3796 else if (next == m->top && oldsize + m->topsize > nb) {
3797 /* Expand into top */
3798 size_t newsize = oldsize + m->topsize;
3799 size_t newtopsize = newsize - nb;
3800 mchunkptr newtop = chunk_plus_offset(oldp, nb);
3801 set_inuse(m, oldp, nb);
3802 newtop->head = newtopsize |PINUSE_BIT;
3804 m->topsize = newtopsize;
3809 USAGE_ERROR_ACTION(m, oldmem);
3818 internal_free(m, extra);
3820 check_inuse_chunk(m, newp);
3821 return chunk2mem(newp);
3824 void* newmem = internal_malloc(m, bytes);
3826 size_t oc = oldsize - overhead_for(oldp);
3827 memcpy(newmem, oldmem, (oc < bytes)? oc : bytes);
3828 internal_free(m, oldmem);
3838 /* --------------------------- memalign support -------------------------- */
3840 static void* internal_memalign(mstate m, size_t alignment, size_t bytes) {
3841 if (alignment <= MALLOC_ALIGNMENT) /* Can just use malloc */
3842 return internal_malloc(m, bytes);
3843 if (alignment < MIN_CHUNK_SIZE) /* must be at least a minimum chunk size */
3844 alignment = MIN_CHUNK_SIZE;
3845 if ((alignment & (alignment-SIZE_T_ONE)) != 0) {/* Ensure a power of 2 */
3846 size_t a = MALLOC_ALIGNMENT << 1;
3847 while (a < alignment) a <<= 1;
3851 if (bytes >= MAX_REQUEST - alignment) {
3852 if (m != 0) { /* Test isn't needed but avoids compiler warning */
3853 MALLOC_FAILURE_ACTION;
3857 size_t nb = request2size(bytes);
3858 size_t req = nb + alignment + MIN_CHUNK_SIZE - CHUNK_OVERHEAD;
3859 char* mem = (char*)internal_malloc(m, req);
3863 mchunkptr p = mem2chunk(mem);
3865 if (PREACTION(m)) return 0;
3866 if ((((size_t)(mem)) % alignment) != 0) { /* misaligned */
3868 Find an aligned spot inside chunk. Since we need to give
3869 back leading space in a chunk of at least MIN_CHUNK_SIZE, if
3870 the first calculation places us at a spot with less than
3871 MIN_CHUNK_SIZE leader, we can move to the next aligned spot.
3872 We've allocated enough total room so that this is always
3875 char* br = (char*)mem2chunk((size_t)(((size_t)(mem +
3879 char* pos = ((size_t)(br - (char*)(p)) >= MIN_CHUNK_SIZE)?
3881 mchunkptr newp = (mchunkptr)pos;
3882 size_t leadsize = pos - (char*)(p);
3883 size_t newsize = chunksize(p) - leadsize;
3885 if (is_mmapped(p)) { /* For mmapped chunks, just adjust offset */
3886 newp->prev_foot = p->prev_foot + leadsize;
3887 newp->head = (newsize|CINUSE_BIT);
3889 else { /* Otherwise, give back leader, use the rest */
3890 set_inuse(m, newp, newsize);
3891 set_inuse(m, p, leadsize);
3892 leader = chunk2mem(p);
3897 /* Give back spare room at the end */
3898 if (!is_mmapped(p)) {
3899 size_t size = chunksize(p);
3900 if (size > nb + MIN_CHUNK_SIZE) {
3901 size_t remainder_size = size - nb;
3902 mchunkptr remainder = chunk_plus_offset(p, nb);
3903 set_inuse(m, p, nb);
3904 set_inuse(m, remainder, remainder_size);
3905 trailer = chunk2mem(remainder);
3909 assert (chunksize(p) >= nb);
3910 assert((((size_t)(chunk2mem(p))) % alignment) == 0);
3911 check_inuse_chunk(m, p);
3914 internal_free(m, leader);
3917 internal_free(m, trailer);
3919 return chunk2mem(p);
3927 /* ------------------------ comalloc/coalloc support --------------------- */
3929 static void** ialloc(mstate m,
3935 This provides common support for independent_X routines, handling
3936 all of the combinations that can result.
3939 bit 0 set if all elements are same size (using sizes[0])
3940 bit 1 set if elements should be zeroed
3943 size_t element_size; /* chunksize of each element, if all same */
3944 size_t contents_size; /* total size of elements */
3945 size_t array_size; /* request size of pointer array */
3946 void* mem; /* malloced aggregate space */
3947 mchunkptr p; /* corresponding chunk */
3948 size_t remainder_size; /* remaining bytes while splitting */
3949 void** marray; /* either "chunks" or malloced ptr array */
3950 mchunkptr array_chunk; /* chunk for malloced ptr array */
3951 flag_t was_enabled; /* to disable mmap */
3955 /* compute array length, if needed */
3957 if (n_elements == 0)
3958 return chunks; /* nothing to do */
3963 /* if empty req, must still return chunk representing empty array */
3964 if (n_elements == 0)
3965 return (void**)internal_malloc(m, 0);
3967 array_size = request2size(n_elements * (sizeof(void*)));
3970 /* compute total element size */
3971 if (opts & 0x1) { /* all-same-size */
3972 element_size = request2size(*sizes);
3973 contents_size = n_elements * element_size;
3975 else { /* add up all the sizes */
3978 for (i = 0; i != n_elements; ++i)
3979 contents_size += request2size(sizes[i]);
3982 size = contents_size + array_size;
3985 Allocate the aggregate chunk. First disable direct-mmapping so
3986 malloc won't use it, since we would not be able to later
3987 free/realloc space internal to a segregated mmap region.
3989 was_enabled = use_mmap(m);
3991 mem = internal_malloc(m, size - CHUNK_OVERHEAD);
3997 if (PREACTION(m)) return 0;
3999 remainder_size = chunksize(p);
4001 assert(!is_mmapped(p));
4003 if (opts & 0x2) { /* optionally clear the elements */
4004 memset((size_t*)mem, 0, remainder_size - SIZE_T_SIZE - array_size);
4007 /* If not provided, allocate the pointer array as final part of chunk */
4009 size_t array_chunk_size;
4010 array_chunk = chunk_plus_offset(p, contents_size);
4011 array_chunk_size = remainder_size - contents_size;
4012 marray = (void**) (chunk2mem(array_chunk));
4013 set_size_and_pinuse_of_inuse_chunk(m, array_chunk, array_chunk_size);
4014 remainder_size = contents_size;
4017 /* split out elements */
4018 for (i = 0; ; ++i) {
4019 marray[i] = chunk2mem(p);
4020 if (i != n_elements-1) {
4021 if (element_size != 0)
4022 size = element_size;
4024 size = request2size(sizes[i]);
4025 remainder_size -= size;
4026 set_size_and_pinuse_of_inuse_chunk(m, p, size);
4027 p = chunk_plus_offset(p, size);
4029 else { /* the final element absorbs any overallocation slop */
4030 set_size_and_pinuse_of_inuse_chunk(m, p, remainder_size);
4036 if (marray != chunks) {
4037 /* final element must have exactly exhausted chunk */
4038 if (element_size != 0) {
4039 assert(remainder_size == element_size);
4042 assert(remainder_size == request2size(sizes[i]));
4044 check_inuse_chunk(m, mem2chunk(marray));
4046 for (i = 0; i != n_elements; ++i)
4047 check_inuse_chunk(m, mem2chunk(marray[i]));
4057 /* -------------------------- public routines ---------------------------- */
4061 void* dlmalloc(size_t bytes) {
4064 If a small request (< 256 bytes minus per-chunk overhead):
4065 1. If one exists, use a remainderless chunk in associated smallbin.
4066 (Remainderless means that there are too few excess bytes to
4067 represent as a chunk.)
4068 2. If it is big enough, use the dv chunk, which is normally the
4069 chunk adjacent to the one used for the most recent small request.
4070 3. If one exists, split the smallest available chunk in a bin,
4071 saving remainder in dv.
4072 4. If it is big enough, use the top chunk.
4073 5. If available, get memory from system and use it
4074 Otherwise, for a large request:
4075 1. Find the smallest available binned chunk that fits, and use it
4076 if it is better fitting than dv chunk, splitting if necessary.
4077 2. If better fitting than any binned chunk, use the dv chunk.
4078 3. If it is big enough, use the top chunk.
4079 4. If request size >= mmap threshold, try to directly mmap this chunk.
4080 5. If available, get memory from system and use it
4082 The ugly goto's here ensure that postaction occurs along all paths.
4085 if (!PREACTION(gm)) {
4088 if (bytes <= MAX_SMALL_REQUEST) {
4091 nb = (bytes < MIN_REQUEST)? MIN_CHUNK_SIZE : pad_request(bytes);
4092 idx = small_index(nb);
4093 smallbits = gm->smallmap >> idx;
4095 if ((smallbits & 0x3U) != 0) { /* Remainderless fit to a smallbin. */
4097 idx += ~smallbits & 1; /* Uses next bin if idx empty */
4098 b = smallbin_at(gm, idx);
4100 assert(chunksize(p) == small_index2size(idx));
4101 unlink_first_small_chunk(gm, b, p, idx);
4102 set_inuse_and_pinuse(gm, p, small_index2size(idx));
4104 check_malloced_chunk(gm, mem, nb);
4108 else if (nb > gm->dvsize) {
4109 if (smallbits != 0) { /* Use chunk in next nonempty smallbin */
4113 binmap_t leftbits = (smallbits << idx) & left_bits(idx2bit(idx));
4114 binmap_t leastbit = least_bit(leftbits);
4115 compute_bit2idx(leastbit, i);
4116 b = smallbin_at(gm, i);
4118 assert(chunksize(p) == small_index2size(i));
4119 unlink_first_small_chunk(gm, b, p, i);
4120 rsize = small_index2size(i) - nb;
4121 /* Fit here cannot be remainderless if 4byte sizes */
4122 if (SIZE_T_SIZE != 4 && rsize < MIN_CHUNK_SIZE)
4123 set_inuse_and_pinuse(gm, p, small_index2size(i));
4125 set_size_and_pinuse_of_inuse_chunk(gm, p, nb);
4126 r = chunk_plus_offset(p, nb);
4127 set_size_and_pinuse_of_free_chunk(r, rsize);
4128 replace_dv(gm, r, rsize);
4131 check_malloced_chunk(gm, mem, nb);
4135 else if (gm->treemap != 0 && (mem = tmalloc_small(gm, nb)) != 0) {
4136 check_malloced_chunk(gm, mem, nb);
4141 else if (bytes >= MAX_REQUEST)
4142 nb = MAX_SIZE_T; /* Too big to allocate. Force failure (in sys alloc) */
4144 nb = pad_request(bytes);
4145 if (gm->treemap != 0 && (mem = tmalloc_large(gm, nb)) != 0) {
4146 check_malloced_chunk(gm, mem, nb);
4151 if (nb <= gm->dvsize) {
4152 size_t rsize = gm->dvsize - nb;
4153 mchunkptr p = gm->dv;
4154 if (rsize >= MIN_CHUNK_SIZE) { /* split dv */
4155 mchunkptr r = gm->dv = chunk_plus_offset(p, nb);
4157 set_size_and_pinuse_of_free_chunk(r, rsize);
4158 set_size_and_pinuse_of_inuse_chunk(gm, p, nb);
4160 else { /* exhaust dv */
4161 size_t dvs = gm->dvsize;
4164 set_inuse_and_pinuse(gm, p, dvs);
4167 check_malloced_chunk(gm, mem, nb);
4171 else if (nb < gm->topsize) { /* Split top */
4172 size_t rsize = gm->topsize -= nb;
4173 mchunkptr p = gm->top;
4174 mchunkptr r = gm->top = chunk_plus_offset(p, nb);
4175 r->head = rsize | PINUSE_BIT;
4176 set_size_and_pinuse_of_inuse_chunk(gm, p, nb);
4178 check_top_chunk(gm, gm->top);
4179 check_malloced_chunk(gm, mem, nb);
4183 mem = sys_alloc(gm, nb);
4193 void dlfree(void* mem) {
4195 Consolidate freed chunks with preceeding or succeeding bordering
4196 free chunks, if they exist, and then place in a bin. Intermixed
4197 with special cases for top, dv, mmapped chunks, and usage errors.
4201 mchunkptr p = mem2chunk(mem);
4203 mstate fm = get_mstate_for(p);
4204 if (!ok_magic(fm)) {
4205 USAGE_ERROR_ACTION(fm, p);
4210 #endif /* FOOTERS */
4211 if (!PREACTION(fm)) {
4212 check_inuse_chunk(fm, p);
4213 if (RTCHECK(ok_address(fm, p) && ok_cinuse(p))) {
4214 size_t psize = chunksize(p);
4215 mchunkptr next = chunk_plus_offset(p, psize);
4217 size_t prevsize = p->prev_foot;
4218 if ((prevsize & IS_MMAPPED_BIT) != 0) {
4219 prevsize &= ~IS_MMAPPED_BIT;
4220 psize += prevsize + MMAP_FOOT_PAD;
4221 if (CALL_MUNMAP((char*)p - prevsize, psize) == 0)
4222 fm->footprint -= psize;
4226 mchunkptr prev = chunk_minus_offset(p, prevsize);
4229 if (RTCHECK(ok_address(fm, prev))) { /* consolidate backward */
4231 unlink_chunk(fm, p, prevsize);
4233 else if ((next->head & INUSE_BITS) == INUSE_BITS) {
4235 set_free_with_pinuse(p, psize, next);
4244 if (RTCHECK(ok_next(p, next) && ok_pinuse(next))) {
4245 if (!cinuse(next)) { /* consolidate forward */
4246 if (next == fm->top) {
4247 size_t tsize = fm->topsize += psize;
4249 p->head = tsize | PINUSE_BIT;
4254 if (should_trim(fm, tsize))
4258 else if (next == fm->dv) {
4259 size_t dsize = fm->dvsize += psize;
4261 set_size_and_pinuse_of_free_chunk(p, dsize);
4265 size_t nsize = chunksize(next);
4267 unlink_chunk(fm, next, nsize);
4268 set_size_and_pinuse_of_free_chunk(p, psize);
4276 set_free_with_pinuse(p, psize, next);
4277 insert_chunk(fm, p, psize);
4278 check_free_chunk(fm, p);
4283 USAGE_ERROR_ACTION(fm, p);
4290 #endif /* FOOTERS */
4295 void* dlcalloc(size_t n_elements, size_t elem_size) {
4298 if (n_elements != 0) {
4299 req = n_elements * elem_size;
4300 if (((n_elements | elem_size) & ~(size_t)0xffff) &&
4301 (req / n_elements != elem_size))
4302 req = MAX_SIZE_T; /* force downstream failure on overflow */
4304 mem = dlmalloc(req);
4305 if (mem != 0 && calloc_must_clear(mem2chunk(mem)))
4306 memset(mem, 0, req);
4310 void* dlrealloc(void* oldmem, size_t bytes) {
4312 return dlmalloc(bytes);
4313 #ifdef REALLOC_ZERO_BYTES_FREES
4318 #endif /* REALLOC_ZERO_BYTES_FREES */
4323 mstate m = get_mstate_for(mem2chunk(oldmem));
4325 USAGE_ERROR_ACTION(m, oldmem);
4328 #endif /* FOOTERS */
4329 return internal_realloc(m, oldmem, bytes);
4335 void* dlmemalign(size_t alignment, size_t bytes) {
4336 return internal_memalign(gm, alignment, bytes);
4341 void** dlindependent_calloc(size_t n_elements, size_t elem_size,
4343 size_t sz = elem_size; /* serves as 1-element array */
4344 return ialloc(gm, n_elements, &sz, 3, chunks);
4347 void** dlindependent_comalloc(size_t n_elements, size_t sizes[],
4349 return ialloc(gm, n_elements, sizes, 0, chunks);
4352 void* dlvalloc(size_t bytes) {
4355 pagesz = mparams.page_size;
4356 return dlmemalign(pagesz, bytes);
4359 void* dlpvalloc(size_t bytes) {
4362 pagesz = mparams.page_size;
4363 return dlmemalign(pagesz, (bytes + pagesz - SIZE_T_ONE) & ~(pagesz - SIZE_T_ONE));
4366 int dlmalloc_trim(size_t pad) {
4368 if (!PREACTION(gm)) {
4369 result = sys_trim(gm, pad);
4375 size_t dlmalloc_footprint(void) {
4376 return gm->footprint;
4379 size_t dlmalloc_max_footprint(void) {
4380 return gm->max_footprint;
4384 struct mallinfo dlmallinfo(void) {
4385 return internal_mallinfo(gm);
4387 #endif /* NO_MALLINFO */
4389 void dlmalloc_stats() {
4390 internal_malloc_stats(gm);
4393 size_t dlmalloc_usable_size(void* mem) {
4395 mchunkptr p = mem2chunk(mem);
4397 return chunksize(p) - overhead_for(p);
4402 int dlmallopt(int param_number, int value) {
4403 return change_mparam(param_number, value);
4408 #endif /* !ONLY_MSPACES */
4410 /* ----------------------------- user mspaces ---------------------------- */
4414 static mstate init_user_mstate(char* tbase, size_t tsize) {
4415 size_t msize = pad_request(sizeof(struct malloc_state));
4417 mchunkptr msp = align_as_chunk(tbase);
4418 mstate m = (mstate)(chunk2mem(msp));
4419 memset(m, 0, msize);
4420 INITIAL_LOCK(&m->mutex);
4421 msp->head = (msize|PINUSE_BIT|CINUSE_BIT);
4422 m->seg.base = m->least_addr = tbase;
4423 m->seg.size = m->footprint = m->max_footprint = tsize;
4424 m->magic = mparams.magic;
4425 m->mflags = mparams.default_mflags;
4426 disable_contiguous(m);
4428 mn = next_chunk(mem2chunk(m));
4429 init_top(m, mn, (size_t)((tbase + tsize) - (char*)mn) - TOP_FOOT_SIZE);
4430 check_top_chunk(m, m->top);
4434 mspace create_mspace(size_t capacity, int locked) {
4436 size_t msize = pad_request(sizeof(struct malloc_state));
4437 init_mparams(); /* Ensure pagesize etc initialized */
4439 if (capacity < (size_t) -(msize + TOP_FOOT_SIZE + mparams.page_size)) {
4440 size_t rs = ((capacity == 0)? mparams.granularity :
4441 (capacity + TOP_FOOT_SIZE + msize));
4442 size_t tsize = granularity_align(rs);
4443 char* tbase = (char*)(CALL_MMAP(tsize));
4444 if (tbase != CMFAIL) {
4445 m = init_user_mstate(tbase, tsize);
4446 m->seg.sflags = IS_MMAPPED_BIT;
4447 set_lock(m, locked);
4453 mspace create_mspace_with_base(void* base, size_t capacity, int locked) {
4455 size_t msize = pad_request(sizeof(struct malloc_state));
4456 init_mparams(); /* Ensure pagesize etc initialized */
4458 if (capacity > msize + TOP_FOOT_SIZE &&
4459 capacity < (size_t) -(msize + TOP_FOOT_SIZE + mparams.page_size)) {
4460 m = init_user_mstate((char*)base, capacity);
4461 m->seg.sflags = EXTERN_BIT;
4462 set_lock(m, locked);
4467 size_t destroy_mspace(mspace msp) {
4469 mstate ms = (mstate)msp;
4471 msegmentptr sp = &ms->seg;
4473 char* base = sp->base;
4474 size_t size = sp->size;
4475 flag_t flag = sp->sflags;
4477 if ((flag & IS_MMAPPED_BIT) && !(flag & EXTERN_BIT) &&
4478 CALL_MUNMAP(base, size) == 0)
4483 USAGE_ERROR_ACTION(ms,ms);
4489 mspace versions of routines are near-clones of the global
4490 versions. This is not so nice but better than the alternatives.
4493 void* mspace_malloc(mspace msp, size_t bytes) {
4494 mstate ms = (mstate)msp;
4495 if (!ok_magic(ms)) {
4496 USAGE_ERROR_ACTION(ms,ms);
4499 if (!PREACTION(ms)) {
4502 if (bytes <= MAX_SMALL_REQUEST) {
4505 nb = (bytes < MIN_REQUEST)? MIN_CHUNK_SIZE : pad_request(bytes);
4506 idx = small_index(nb);
4507 smallbits = ms->smallmap >> idx;
4509 if ((smallbits & 0x3U) != 0) { /* Remainderless fit to a smallbin. */
4511 idx += ~smallbits & 1; /* Uses next bin if idx empty */
4512 b = smallbin_at(ms, idx);
4514 assert(chunksize(p) == small_index2size(idx));
4515 unlink_first_small_chunk(ms, b, p, idx);
4516 set_inuse_and_pinuse(ms, p, small_index2size(idx));
4518 check_malloced_chunk(ms, mem, nb);
4522 else if (nb > ms->dvsize) {
4523 if (smallbits != 0) { /* Use chunk in next nonempty smallbin */
4527 binmap_t leftbits = (smallbits << idx) & left_bits(idx2bit(idx));
4528 binmap_t leastbit = least_bit(leftbits);
4529 compute_bit2idx(leastbit, i);
4530 b = smallbin_at(ms, i);
4532 assert(chunksize(p) == small_index2size(i));
4533 unlink_first_small_chunk(ms, b, p, i);
4534 rsize = small_index2size(i) - nb;
4535 /* Fit here cannot be remainderless if 4byte sizes */
4536 if (SIZE_T_SIZE != 4 && rsize < MIN_CHUNK_SIZE)
4537 set_inuse_and_pinuse(ms, p, small_index2size(i));
4539 set_size_and_pinuse_of_inuse_chunk(ms, p, nb);
4540 r = chunk_plus_offset(p, nb);
4541 set_size_and_pinuse_of_free_chunk(r, rsize);
4542 replace_dv(ms, r, rsize);
4545 check_malloced_chunk(ms, mem, nb);
4549 else if (ms->treemap != 0 && (mem = tmalloc_small(ms, nb)) != 0) {
4550 check_malloced_chunk(ms, mem, nb);
4555 else if (bytes >= MAX_REQUEST)
4556 nb = MAX_SIZE_T; /* Too big to allocate. Force failure (in sys alloc) */
4558 nb = pad_request(bytes);
4559 if (ms->treemap != 0 && (mem = tmalloc_large(ms, nb)) != 0) {
4560 check_malloced_chunk(ms, mem, nb);
4565 if (nb <= ms->dvsize) {
4566 size_t rsize = ms->dvsize - nb;
4567 mchunkptr p = ms->dv;
4568 if (rsize >= MIN_CHUNK_SIZE) { /* split dv */
4569 mchunkptr r = ms->dv = chunk_plus_offset(p, nb);
4571 set_size_and_pinuse_of_free_chunk(r, rsize);
4572 set_size_and_pinuse_of_inuse_chunk(ms, p, nb);
4574 else { /* exhaust dv */
4575 size_t dvs = ms->dvsize;
4578 set_inuse_and_pinuse(ms, p, dvs);
4581 check_malloced_chunk(ms, mem, nb);
4585 else if (nb < ms->topsize) { /* Split top */
4586 size_t rsize = ms->topsize -= nb;
4587 mchunkptr p = ms->top;
4588 mchunkptr r = ms->top = chunk_plus_offset(p, nb);
4589 r->head = rsize | PINUSE_BIT;
4590 set_size_and_pinuse_of_inuse_chunk(ms, p, nb);
4592 check_top_chunk(ms, ms->top);
4593 check_malloced_chunk(ms, mem, nb);
4597 mem = sys_alloc(ms, nb);
4607 void mspace_free(mspace msp, void* mem) {
4609 mchunkptr p = mem2chunk(mem);
4611 mstate fm = get_mstate_for(p);
4613 mstate fm = (mstate)msp;
4614 #endif /* FOOTERS */
4615 if (!ok_magic(fm)) {
4616 USAGE_ERROR_ACTION(fm, p);
4619 if (!PREACTION(fm)) {
4620 check_inuse_chunk(fm, p);
4621 if (RTCHECK(ok_address(fm, p) && ok_cinuse(p))) {
4622 size_t psize = chunksize(p);
4623 mchunkptr next = chunk_plus_offset(p, psize);
4625 size_t prevsize = p->prev_foot;
4626 if ((prevsize & IS_MMAPPED_BIT) != 0) {
4627 prevsize &= ~IS_MMAPPED_BIT;
4628 psize += prevsize + MMAP_FOOT_PAD;
4629 if (CALL_MUNMAP((char*)p - prevsize, psize) == 0)
4630 fm->footprint -= psize;
4634 mchunkptr prev = chunk_minus_offset(p, prevsize);
4637 if (RTCHECK(ok_address(fm, prev))) { /* consolidate backward */
4639 unlink_chunk(fm, p, prevsize);
4641 else if ((next->head & INUSE_BITS) == INUSE_BITS) {
4643 set_free_with_pinuse(p, psize, next);
4652 if (RTCHECK(ok_next(p, next) && ok_pinuse(next))) {
4653 if (!cinuse(next)) { /* consolidate forward */
4654 if (next == fm->top) {
4655 size_t tsize = fm->topsize += psize;
4657 p->head = tsize | PINUSE_BIT;
4662 if (should_trim(fm, tsize))
4666 else if (next == fm->dv) {
4667 size_t dsize = fm->dvsize += psize;
4669 set_size_and_pinuse_of_free_chunk(p, dsize);
4673 size_t nsize = chunksize(next);
4675 unlink_chunk(fm, next, nsize);
4676 set_size_and_pinuse_of_free_chunk(p, psize);
4684 set_free_with_pinuse(p, psize, next);
4685 insert_chunk(fm, p, psize);
4686 check_free_chunk(fm, p);
4691 USAGE_ERROR_ACTION(fm, p);
4698 void* mspace_calloc(mspace msp, size_t n_elements, size_t elem_size) {
4701 mstate ms = (mstate)msp;
4702 if (!ok_magic(ms)) {
4703 USAGE_ERROR_ACTION(ms,ms);
4706 if (n_elements != 0) {
4707 req = n_elements * elem_size;
4708 if (((n_elements | elem_size) & ~(size_t)0xffff) &&
4709 (req / n_elements != elem_size))
4710 req = MAX_SIZE_T; /* force downstream failure on overflow */
4712 mem = internal_malloc(ms, req);
4713 if (mem != 0 && calloc_must_clear(mem2chunk(mem)))
4714 memset(mem, 0, req);
4718 void* mspace_realloc(mspace msp, void* oldmem, size_t bytes) {
4720 return mspace_malloc(msp, bytes);
4721 #ifdef REALLOC_ZERO_BYTES_FREES
4723 mspace_free(msp, oldmem);
4726 #endif /* REALLOC_ZERO_BYTES_FREES */
4729 mchunkptr p = mem2chunk(oldmem);
4730 mstate ms = get_mstate_for(p);
4732 mstate ms = (mstate)msp;
4733 #endif /* FOOTERS */
4734 if (!ok_magic(ms)) {
4735 USAGE_ERROR_ACTION(ms,ms);
4738 return internal_realloc(ms, oldmem, bytes);
4742 void* mspace_memalign(mspace msp, size_t alignment, size_t bytes) {
4743 mstate ms = (mstate)msp;
4744 if (!ok_magic(ms)) {
4745 USAGE_ERROR_ACTION(ms,ms);
4748 return internal_memalign(ms, alignment, bytes);
4751 void** mspace_independent_calloc(mspace msp, size_t n_elements,
4752 size_t elem_size, void* chunks[]) {
4753 size_t sz = elem_size; /* serves as 1-element array */
4754 mstate ms = (mstate)msp;
4755 if (!ok_magic(ms)) {
4756 USAGE_ERROR_ACTION(ms,ms);
4759 return ialloc(ms, n_elements, &sz, 3, chunks);
4762 void** mspace_independent_comalloc(mspace msp, size_t n_elements,
4763 size_t sizes[], void* chunks[]) {
4764 mstate ms = (mstate)msp;
4765 if (!ok_magic(ms)) {
4766 USAGE_ERROR_ACTION(ms,ms);
4769 return ialloc(ms, n_elements, sizes, 0, chunks);
4772 int mspace_trim(mspace msp, size_t pad) {
4774 mstate ms = (mstate)msp;
4776 if (!PREACTION(ms)) {
4777 result = sys_trim(ms, pad);
4782 USAGE_ERROR_ACTION(ms,ms);
4787 void mspace_malloc_stats(mspace msp) {
4788 mstate ms = (mstate)msp;
4790 internal_malloc_stats(ms);
4793 USAGE_ERROR_ACTION(ms,ms);
4797 size_t mspace_footprint(mspace msp) {
4799 mstate ms = (mstate)msp;
4801 result = ms->footprint;
4803 USAGE_ERROR_ACTION(ms,ms);
4808 size_t mspace_max_footprint(mspace msp) {
4810 mstate ms = (mstate)msp;
4812 result = ms->max_footprint;
4814 USAGE_ERROR_ACTION(ms,ms);
4820 struct mallinfo mspace_mallinfo(mspace msp) {
4821 mstate ms = (mstate)msp;
4822 if (!ok_magic(ms)) {
4823 USAGE_ERROR_ACTION(ms,ms);
4825 return internal_mallinfo(ms);
4827 #endif /* NO_MALLINFO */
4829 int mspace_mallopt(int param_number, int value) {
4830 return change_mparam(param_number, value);
4833 #endif /* MSPACES */
4835 /* -------------------- Alternative MORECORE functions ------------------- */
4838 Guidelines for creating a custom version of MORECORE:
4840 * For best performance, MORECORE should allocate in multiples of pagesize.
4841 * MORECORE may allocate more memory than requested. (Or even less,
4842 but this will usually result in a malloc failure.)
4843 * MORECORE must not allocate memory when given argument zero, but
4844 instead return one past the end address of memory from previous
4846 * For best performance, consecutive calls to MORECORE with positive
4847 arguments should return increasing addresses, indicating that
4848 space has been contiguously extended.
4849 * Even though consecutive calls to MORECORE need not return contiguous
4850 addresses, it must be OK for malloc'ed chunks to span multiple
4851 regions in those cases where they do happen to be contiguous.
4852 * MORECORE need not handle negative arguments -- it may instead
4853 just return MFAIL when given negative arguments.
4854 Negative arguments are always multiples of pagesize. MORECORE
4855 must not misinterpret negative args as large positive unsigned
4856 args. You can suppress all such calls from even occurring by defining
4857 MORECORE_CANNOT_TRIM,
4859 As an example alternative MORECORE, here is a custom allocator
4860 kindly contributed for pre-OSX macOS. It uses virtually but not
4861 necessarily physically contiguous non-paged memory (locked in,
4862 present and won't get swapped out). You can use it by uncommenting
4863 this section, adding some #includes, and setting up the appropriate
4866 #define MORECORE osMoreCore
4868 There is also a shutdown routine that should somehow be called for
4869 cleanup upon program exit.
4871 #define MAX_POOL_ENTRIES 100
4872 #define MINIMUM_MORECORE_SIZE (64 * 1024U)
4873 static int next_os_pool;
4874 void *our_os_pools[MAX_POOL_ENTRIES];
4876 void *osMoreCore(int size)
4879 static void *sbrk_top = 0;
4883 if (size < MINIMUM_MORECORE_SIZE)
4884 size = MINIMUM_MORECORE_SIZE;
4885 if (CurrentExecutionLevel() == kTaskLevel)
4886 ptr = PoolAllocateResident(size + RM_PAGE_SIZE, 0);
4889 return (void *) MFAIL;
4891 // save ptrs so they can be freed during cleanup
4892 our_os_pools[next_os_pool] = ptr;
4894 ptr = (void *) ((((size_t) ptr) + RM_PAGE_MASK) & ~RM_PAGE_MASK);
4895 sbrk_top = (char *) ptr + size;
4900 // we don't currently support shrink behavior
4901 return (void *) MFAIL;
4909 // cleanup any allocated memory pools
4910 // called as last thing before shutting down driver
4912 void osCleanupMem(void)
4916 for (ptr = our_os_pools; ptr < &our_os_pools[MAX_POOL_ENTRIES]; ptr++)
4919 PoolDeallocate(*ptr);
4927 /* -----------------------------------------------------------------------
4929 V2.8.3 Thu Sep 22 11:16:32 2005 Doug Lea (dl at gee)
4930 * Add max_footprint functions
4931 * Ensure all appropriate literals are size_t
4932 * Fix conditional compilation problem for some #define settings
4933 * Avoid concatenating segments with the one provided
4934 in create_mspace_with_base
4935 * Rename some variables to avoid compiler shadowing warnings
4936 * Use explicit lock initialization.
4937 * Better handling of sbrk interference.
4938 * Simplify and fix segment insertion, trimming and mspace_destroy
4939 * Reinstate REALLOC_ZERO_BYTES_FREES option from 2.7.x
4940 * Thanks especially to Dennis Flanagan for help on these.
4942 V2.8.2 Sun Jun 12 16:01:10 2005 Doug Lea (dl at gee)
4943 * Fix memalign brace error.
4945 V2.8.1 Wed Jun 8 16:11:46 2005 Doug Lea (dl at gee)
4946 * Fix improper #endif nesting in C++
4947 * Add explicit casts needed for C++
4949 V2.8.0 Mon May 30 14:09:02 2005 Doug Lea (dl at gee)
4950 * Use trees for large bins
4952 * Use segments to unify sbrk-based and mmap-based system allocation,
4953 removing need for emulation on most platforms without sbrk.
4954 * Default safety checks
4955 * Optional footer checks. Thanks to William Robertson for the idea.
4956 * Internal code refactoring
4957 * Incorporate suggestions and platform-specific changes.
4958 Thanks to Dennis Flanagan, Colin Plumb, Niall Douglas,
4959 Aaron Bachmann, Emery Berger, and others.
4960 * Speed up non-fastbin processing enough to remove fastbins.
4961 * Remove useless cfree() to avoid conflicts with other apps.
4962 * Remove internal memcpy, memset. Compilers handle builtins better.
4963 * Remove some options that no one ever used and rename others.
4965 V2.7.2 Sat Aug 17 09:07:30 2002 Doug Lea (dl at gee)
4966 * Fix malloc_state bitmap array misdeclaration
4968 V2.7.1 Thu Jul 25 10:58:03 2002 Doug Lea (dl at gee)
4969 * Allow tuning of FIRST_SORTED_BIN_SIZE
4970 * Use PTR_UINT as type for all ptr->int casts. Thanks to John Belmonte.
4971 * Better detection and support for non-contiguousness of MORECORE.
4972 Thanks to Andreas Mueller, Conal Walsh, and Wolfram Gloger
4973 * Bypass most of malloc if no frees. Thanks To Emery Berger.
4974 * Fix freeing of old top non-contiguous chunk im sysmalloc.
4975 * Raised default trim and map thresholds to 256K.
4976 * Fix mmap-related #defines. Thanks to Lubos Lunak.
4977 * Fix copy macros; added LACKS_FCNTL_H. Thanks to Neal Walfield.
4978 * Branch-free bin calculation
4979 * Default trim and mmap thresholds now 256K.
4981 V2.7.0 Sun Mar 11 14:14:06 2001 Doug Lea (dl at gee)
4982 * Introduce independent_comalloc and independent_calloc.
4983 Thanks to Michael Pachos for motivation and help.
4984 * Make optional .h file available
4985 * Allow > 2GB requests on 32bit systems.
4986 * new WIN32 sbrk, mmap, munmap, lock code from <Walter@GeNeSys-e.de>.
4987 Thanks also to Andreas Mueller <a.mueller at paradatec.de>,
4989 * Allow override of MALLOC_ALIGNMENT (Thanks to Ruud Waij for
4991 * memalign: check alignment arg
4992 * realloc: don't try to shift chunks backwards, since this
4993 leads to more fragmentation in some programs and doesn't
4994 seem to help in any others.
4995 * Collect all cases in malloc requiring system memory into sysmalloc
4996 * Use mmap as backup to sbrk
4997 * Place all internal state in malloc_state
4998 * Introduce fastbins (although similar to 2.5.1)
4999 * Many minor tunings and cosmetic improvements
5000 * Introduce USE_PUBLIC_MALLOC_WRAPPERS, USE_MALLOC_LOCK
5001 * Introduce MALLOC_FAILURE_ACTION, MORECORE_CONTIGUOUS
5002 Thanks to Tony E. Bennett <tbennett@nvidia.com> and others.
5003 * Include errno.h to support default failure action.
5005 V2.6.6 Sun Dec 5 07:42:19 1999 Doug Lea (dl at gee)
5006 * return null for negative arguments
5007 * Added Several WIN32 cleanups from Martin C. Fong <mcfong at yahoo.com>
5008 * Add 'LACKS_SYS_PARAM_H' for those systems without 'sys/param.h'
5009 (e.g. WIN32 platforms)
5010 * Cleanup header file inclusion for WIN32 platforms
5011 * Cleanup code to avoid Microsoft Visual C++ compiler complaints
5012 * Add 'USE_DL_PREFIX' to quickly allow co-existence with existing
5013 memory allocation routines
5014 * Set 'malloc_getpagesize' for WIN32 platforms (needs more work)
5015 * Use 'assert' rather than 'ASSERT' in WIN32 code to conform to
5016 usage of 'assert' in non-WIN32 code
5017 * Improve WIN32 'sbrk()' emulation's 'findRegion()' routine to
5019 * Always call 'fREe()' rather than 'free()'
5021 V2.6.5 Wed Jun 17 15:57:31 1998 Doug Lea (dl at gee)
5022 * Fixed ordering problem with boundary-stamping
5024 V2.6.3 Sun May 19 08:17:58 1996 Doug Lea (dl at gee)
5025 * Added pvalloc, as recommended by H.J. Liu
5026 * Added 64bit pointer support mainly from Wolfram Gloger
5027 * Added anonymously donated WIN32 sbrk emulation
5028 * Malloc, calloc, getpagesize: add optimizations from Raymond Nijssen
5029 * malloc_extend_top: fix mask error that caused wastage after
5031 * Add linux mremap support code from HJ Liu
5033 V2.6.2 Tue Dec 5 06:52:55 1995 Doug Lea (dl at gee)
5034 * Integrated most documentation with the code.
5035 * Add support for mmap, with help from
5036 Wolfram Gloger (Gloger@lrz.uni-muenchen.de).
5037 * Use last_remainder in more cases.
5038 * Pack bins using idea from colin@nyx10.cs.du.edu
5039 * Use ordered bins instead of best-fit threshhold
5040 * Eliminate block-local decls to simplify tracing and debugging.
5041 * Support another case of realloc via move into top
5042 * Fix error occuring when initial sbrk_base not word-aligned.
5043 * Rely on page size for units instead of SBRK_UNIT to
5044 avoid surprises about sbrk alignment conventions.
5045 * Add mallinfo, mallopt. Thanks to Raymond Nijssen
5046 (raymond@es.ele.tue.nl) for the suggestion.
5047 * Add `pad' argument to malloc_trim and top_pad mallopt parameter.
5048 * More precautions for cases where other routines call sbrk,
5049 courtesy of Wolfram Gloger (Gloger@lrz.uni-muenchen.de).
5050 * Added macros etc., allowing use in linux libc from
5051 H.J. Lu (hjl@gnu.ai.mit.edu)
5052 * Inverted this history list
5054 V2.6.1 Sat Dec 2 14:10:57 1995 Doug Lea (dl at gee)
5055 * Re-tuned and fixed to behave more nicely with V2.6.0 changes.
5056 * Removed all preallocation code since under current scheme
5057 the work required to undo bad preallocations exceeds
5058 the work saved in good cases for most test programs.
5059 * No longer use return list or unconsolidated bins since
5060 no scheme using them consistently outperforms those that don't
5061 given above changes.
5062 * Use best fit for very large chunks to prevent some worst-cases.
5063 * Added some support for debugging
5065 V2.6.0 Sat Nov 4 07:05:23 1995 Doug Lea (dl at gee)
5066 * Removed footers when chunks are in use. Thanks to
5067 Paul Wilson (wilson@cs.texas.edu) for the suggestion.
5069 V2.5.4 Wed Nov 1 07:54:51 1995 Doug Lea (dl at gee)
5070 * Added malloc_trim, with help from Wolfram Gloger
5071 (wmglo@Dent.MED.Uni-Muenchen.DE).
5073 V2.5.3 Tue Apr 26 10:16:01 1994 Doug Lea (dl at g)
5075 V2.5.2 Tue Apr 5 16:20:40 1994 Doug Lea (dl at g)
5076 * realloc: try to expand in both directions
5077 * malloc: swap order of clean-bin strategy;
5078 * realloc: only conditionally expand backwards
5079 * Try not to scavenge used bins
5080 * Use bin counts as a guide to preallocation
5081 * Occasionally bin return list chunks in first scan
5082 * Add a few optimizations from colin@nyx10.cs.du.edu
5084 V2.5.1 Sat Aug 14 15:40:43 1993 Doug Lea (dl at g)
5085 * faster bin computation & slightly different binning
5086 * merged all consolidations to one part of malloc proper
5087 (eliminating old malloc_find_space & malloc_clean_bin)
5088 * Scan 2 returns chunks (not just 1)
5089 * Propagate failure in realloc if malloc returns 0
5090 * Add stuff to allow compilation on non-ANSI compilers
5091 from kpv@research.att.com
5093 V2.5 Sat Aug 7 07:41:59 1993 Doug Lea (dl at g.oswego.edu)
5094 * removed potential for odd address access in prev_chunk
5095 * removed dependency on getpagesize.h
5096 * misc cosmetics and a bit more internal documentation
5097 * anticosmetics: mangled names in macros to evade debugger strangeness
5098 * tested on sparc, hp-700, dec-mips, rs6000
5099 with gcc & native cc (hp, dec only) allowing
5100 Detlefs & Zorn comparison study (in SIGPLAN Notices.)
5102 Trial version Fri Aug 28 13:14:29 1992 Doug Lea (dl at g.oswego.edu)
5103 * Based loosely on libg++-1.2X malloc. (It retains some of the overall
5104 structure of old version, but most details differ.)