submodule: Use cat instead of echo to avoid DOS line-endings
[git/dscho.git] / compat / nedmalloc / malloc.c.h
blobff7c2c4fd8642da754b1c85a9e177baf4ddc7136
1 /*
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 pre-2.8.4 Mon Nov 27 11:22:37 2006 (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!
13 * Quickstart
15 This library is all in one file to simplify the most common usage:
16 ftp it, compile it (-O3), and link it into another program. All of
17 the compile-time options default to reasonable values for use on
18 most platforms. You might later want to step through various
19 compile-time and dynamic tuning options.
21 For convenience, an include file for code using this malloc is at:
22 ftp://gee.cs.oswego.edu/pub/misc/malloc-2.8.4.h
23 You don't really need this .h file unless you call functions not
24 defined in your system include files. The .h file contains only the
25 excerpts from this file needed for using this malloc on ANSI C/C++
26 systems, so long as you haven't changed compile-time options about
27 naming and tuning parameters. If you do, then you can create your
28 own malloc.h that does include all settings by cutting at the point
29 indicated below. Note that you may already by default be using a C
30 library containing a malloc that is based on some version of this
31 malloc (for example in linux). You might still want to use the one
32 in this file to customize settings or to avoid overheads associated
33 with library versions.
35 * Vital statistics:
37 Supported pointer/size_t representation: 4 or 8 bytes
38 size_t MUST be an unsigned type of the same width as
39 pointers. (If you are using an ancient system that declares
40 size_t as a signed type, or need it to be a different width
41 than pointers, you can use a previous release of this malloc
42 (e.g. 2.7.2) supporting these.)
44 Alignment: 8 bytes (default)
45 This suffices for nearly all current machines and C compilers.
46 However, you can define MALLOC_ALIGNMENT to be wider than this
47 if necessary (up to 128bytes), at the expense of using more space.
49 Minimum overhead per allocated chunk: 4 or 8 bytes (if 4byte sizes)
50 8 or 16 bytes (if 8byte sizes)
51 Each malloced chunk has a hidden word of overhead holding size
52 and status information, and additional cross-check word
53 if FOOTERS is defined.
55 Minimum allocated size: 4-byte ptrs: 16 bytes (including overhead)
56 8-byte ptrs: 32 bytes (including overhead)
58 Even a request for zero bytes (i.e., malloc(0)) returns a
59 pointer to something of the minimum allocatable size.
60 The maximum overhead wastage (i.e., number of extra bytes
61 allocated than were requested in malloc) is less than or equal
62 to the minimum size, except for requests >= mmap_threshold that
63 are serviced via mmap(), where the worst case wastage is about
64 32 bytes plus the remainder from a system page (the minimal
65 mmap unit); typically 4096 or 8192 bytes.
67 Security: static-safe; optionally more or less
68 The "security" of malloc refers to the ability of malicious
69 code to accentuate the effects of errors (for example, freeing
70 space that is not currently malloc'ed or overwriting past the
71 ends of chunks) in code that calls malloc. This malloc
72 guarantees not to modify any memory locations below the base of
73 heap, i.e., static variables, even in the presence of usage
74 errors. The routines additionally detect most improper frees
75 and reallocs. All this holds as long as the static bookkeeping
76 for malloc itself is not corrupted by some other means. This
77 is only one aspect of security -- these checks do not, and
78 cannot, detect all possible programming errors.
80 If FOOTERS is defined nonzero, then each allocated chunk
81 carries an additional check word to verify that it was malloced
82 from its space. These check words are the same within each
83 execution of a program using malloc, but differ across
84 executions, so externally crafted fake chunks cannot be
85 freed. This improves security by rejecting frees/reallocs that
86 could corrupt heap memory, in addition to the checks preventing
87 writes to statics that are always on. This may further improve
88 security at the expense of time and space overhead. (Note that
89 FOOTERS may also be worth using with MSPACES.)
91 By default detected errors cause the program to abort (calling
92 "abort()"). You can override this to instead proceed past
93 errors by defining PROCEED_ON_ERROR. In this case, a bad free
94 has no effect, and a malloc that encounters a bad address
95 caused by user overwrites will ignore the bad address by
96 dropping pointers and indices to all known memory. This may
97 be appropriate for programs that should continue if at all
98 possible in the face of programming errors, although they may
99 run out of memory because dropped memory is never reclaimed.
101 If you don't like either of these options, you can define
102 CORRUPTION_ERROR_ACTION and USAGE_ERROR_ACTION to do anything
103 else. And if you are sure that your program using malloc has
104 no errors or vulnerabilities, you can define INSECURE to 1,
105 which might (or might not) provide a small performance improvement.
107 Thread-safety: NOT thread-safe unless USE_LOCKS defined
108 When USE_LOCKS is defined, each public call to malloc, free,
109 etc is surrounded with either a pthread mutex or a win32
110 spinlock (depending on WIN32). This is not especially fast, and
111 can be a major bottleneck. It is designed only to provide
112 minimal protection in concurrent environments, and to provide a
113 basis for extensions. If you are using malloc in a concurrent
114 program, consider instead using nedmalloc
115 (http://www.nedprod.com/programs/portable/nedmalloc/) or
116 ptmalloc (See http://www.malloc.de), which are derived
117 from versions of this malloc.
119 System requirements: Any combination of MORECORE and/or MMAP/MUNMAP
120 This malloc can use unix sbrk or any emulation (invoked using
121 the CALL_MORECORE macro) and/or mmap/munmap or any emulation
122 (invoked using CALL_MMAP/CALL_MUNMAP) to get and release system
123 memory. On most unix systems, it tends to work best if both
124 MORECORE and MMAP are enabled. On Win32, it uses emulations
125 based on VirtualAlloc. It also uses common C library functions
126 like memset.
128 Compliance: I believe it is compliant with the Single Unix Specification
129 (See http://www.unix.org). Also SVID/XPG, ANSI C, and probably
130 others as well.
132 * Overview of algorithms
134 This is not the fastest, most space-conserving, most portable, or
135 most tunable malloc ever written. However it is among the fastest
136 while also being among the most space-conserving, portable and
137 tunable. Consistent balance across these factors results in a good
138 general-purpose allocator for malloc-intensive programs.
140 In most ways, this malloc is a best-fit allocator. Generally, it
141 chooses the best-fitting existing chunk for a request, with ties
142 broken in approximately least-recently-used order. (This strategy
143 normally maintains low fragmentation.) However, for requests less
144 than 256bytes, it deviates from best-fit when there is not an
145 exactly fitting available chunk by preferring to use space adjacent
146 to that used for the previous small request, as well as by breaking
147 ties in approximately most-recently-used order. (These enhance
148 locality of series of small allocations.) And for very large requests
149 (>= 256Kb by default), it relies on system memory mapping
150 facilities, if supported. (This helps avoid carrying around and
151 possibly fragmenting memory used only for large chunks.)
153 All operations (except malloc_stats and mallinfo) have execution
154 times that are bounded by a constant factor of the number of bits in
155 a size_t, not counting any clearing in calloc or copying in realloc,
156 or actions surrounding MORECORE and MMAP that have times
157 proportional to the number of non-contiguous regions returned by
158 system allocation routines, which is often just 1. In real-time
159 applications, you can optionally suppress segment traversals using
160 NO_SEGMENT_TRAVERSAL, which assures bounded execution even when
161 system allocators return non-contiguous spaces, at the typical
162 expense of carrying around more memory and increased fragmentation.
164 The implementation is not very modular and seriously overuses
165 macros. Perhaps someday all C compilers will do as good a job
166 inlining modular code as can now be done by brute-force expansion,
167 but now, enough of them seem not to.
169 Some compilers issue a lot of warnings about code that is
170 dead/unreachable only on some platforms, and also about intentional
171 uses of negation on unsigned types. All known cases of each can be
172 ignored.
174 For a longer but out of date high-level description, see
175 http://gee.cs.oswego.edu/dl/html/malloc.html
177 * MSPACES
178 If MSPACES is defined, then in addition to malloc, free, etc.,
179 this file also defines mspace_malloc, mspace_free, etc. These
180 are versions of malloc routines that take an "mspace" argument
181 obtained using create_mspace, to control all internal bookkeeping.
182 If ONLY_MSPACES is defined, only these versions are compiled.
183 So if you would like to use this allocator for only some allocations,
184 and your system malloc for others, you can compile with
185 ONLY_MSPACES and then do something like...
186 static mspace mymspace = create_mspace(0,0); // for example
187 #define mymalloc(bytes) mspace_malloc(mymspace, bytes)
189 (Note: If you only need one instance of an mspace, you can instead
190 use "USE_DL_PREFIX" to relabel the global malloc.)
192 You can similarly create thread-local allocators by storing
193 mspaces as thread-locals. For example:
194 static __thread mspace tlms = 0;
195 void* tlmalloc(size_t bytes) {
196 if (tlms == 0) tlms = create_mspace(0, 0);
197 return mspace_malloc(tlms, bytes);
199 void tlfree(void* mem) { mspace_free(tlms, mem); }
201 Unless FOOTERS is defined, each mspace is completely independent.
202 You cannot allocate from one and free to another (although
203 conformance is only weakly checked, so usage errors are not always
204 caught). If FOOTERS is defined, then each chunk carries around a tag
205 indicating its originating mspace, and frees are directed to their
206 originating spaces.
208 ------------------------- Compile-time options ---------------------------
210 Be careful in setting #define values for numerical constants of type
211 size_t. On some systems, literal values are not automatically extended
212 to size_t precision unless they are explicitly casted. You can also
213 use the symbolic values MAX_SIZE_T, SIZE_T_ONE, etc below.
215 WIN32 default: defined if _WIN32 defined
216 Defining WIN32 sets up defaults for MS environment and compilers.
217 Otherwise defaults are for unix. Beware that there seem to be some
218 cases where this malloc might not be a pure drop-in replacement for
219 Win32 malloc: Random-looking failures from Win32 GDI API's (eg;
220 SetDIBits()) may be due to bugs in some video driver implementations
221 when pixel buffers are malloc()ed, and the region spans more than
222 one VirtualAlloc()ed region. Because dlmalloc uses a small (64Kb)
223 default granularity, pixel buffers may straddle virtual allocation
224 regions more often than when using the Microsoft allocator. You can
225 avoid this by using VirtualAlloc() and VirtualFree() for all pixel
226 buffers rather than using malloc(). If this is not possible,
227 recompile this malloc with a larger DEFAULT_GRANULARITY.
229 MALLOC_ALIGNMENT default: (size_t)8
230 Controls the minimum alignment for malloc'ed chunks. It must be a
231 power of two and at least 8, even on machines for which smaller
232 alignments would suffice. It may be defined as larger than this
233 though. Note however that code and data structures are optimized for
234 the case of 8-byte alignment.
236 MSPACES default: 0 (false)
237 If true, compile in support for independent allocation spaces.
238 This is only supported if HAVE_MMAP is true.
240 ONLY_MSPACES default: 0 (false)
241 If true, only compile in mspace versions, not regular versions.
243 USE_LOCKS default: 0 (false)
244 Causes each call to each public routine to be surrounded with
245 pthread or WIN32 mutex lock/unlock. (If set true, this can be
246 overridden on a per-mspace basis for mspace versions.) If set to a
247 non-zero value other than 1, locks are used, but their
248 implementation is left out, so lock functions must be supplied manually.
250 USE_SPIN_LOCKS default: 1 iff USE_LOCKS and on x86 using gcc or MSC
251 If true, uses custom spin locks for locking. This is currently
252 supported only for x86 platforms using gcc or recent MS compilers.
253 Otherwise, posix locks or win32 critical sections are used.
255 FOOTERS default: 0
256 If true, provide extra checking and dispatching by placing
257 information in the footers of allocated chunks. This adds
258 space and time overhead.
260 INSECURE default: 0
261 If true, omit checks for usage errors and heap space overwrites.
263 USE_DL_PREFIX default: NOT defined
264 Causes compiler to prefix all public routines with the string 'dl'.
265 This can be useful when you only want to use this malloc in one part
266 of a program, using your regular system malloc elsewhere.
268 ABORT default: defined as abort()
269 Defines how to abort on failed checks. On most systems, a failed
270 check cannot die with an "assert" or even print an informative
271 message, because the underlying print routines in turn call malloc,
272 which will fail again. Generally, the best policy is to simply call
273 abort(). It's not very useful to do more than this because many
274 errors due to overwriting will show up as address faults (null, odd
275 addresses etc) rather than malloc-triggered checks, so will also
276 abort. Also, most compilers know that abort() does not return, so
277 can better optimize code conditionally calling it.
279 PROCEED_ON_ERROR default: defined as 0 (false)
280 Controls whether detected bad addresses cause them to bypassed
281 rather than aborting. If set, detected bad arguments to free and
282 realloc are ignored. And all bookkeeping information is zeroed out
283 upon a detected overwrite of freed heap space, thus losing the
284 ability to ever return it from malloc again, but enabling the
285 application to proceed. If PROCEED_ON_ERROR is defined, the
286 static variable malloc_corruption_error_count is compiled in
287 and can be examined to see if errors have occurred. This option
288 generates slower code than the default abort policy.
290 DEBUG default: NOT defined
291 The DEBUG setting is mainly intended for people trying to modify
292 this code or diagnose problems when porting to new platforms.
293 However, it may also be able to better isolate user errors than just
294 using runtime checks. The assertions in the check routines spell
295 out in more detail the assumptions and invariants underlying the
296 algorithms. The checking is fairly extensive, and will slow down
297 execution noticeably. Calling malloc_stats or mallinfo with DEBUG
298 set will attempt to check every non-mmapped allocated and free chunk
299 in the course of computing the summaries.
301 ABORT_ON_ASSERT_FAILURE default: defined as 1 (true)
302 Debugging assertion failures can be nearly impossible if your
303 version of the assert macro causes malloc to be called, which will
304 lead to a cascade of further failures, blowing the runtime stack.
305 ABORT_ON_ASSERT_FAILURE cause assertions failures to call abort(),
306 which will usually make debugging easier.
308 MALLOC_FAILURE_ACTION default: sets errno to ENOMEM, or no-op on win32
309 The action to take before "return 0" when malloc fails to be able to
310 return memory because there is none available.
312 HAVE_MORECORE default: 1 (true) unless win32 or ONLY_MSPACES
313 True if this system supports sbrk or an emulation of it.
315 MORECORE default: sbrk
316 The name of the sbrk-style system routine to call to obtain more
317 memory. See below for guidance on writing custom MORECORE
318 functions. The type of the argument to sbrk/MORECORE varies across
319 systems. It cannot be size_t, because it supports negative
320 arguments, so it is normally the signed type of the same width as
321 size_t (sometimes declared as "intptr_t"). It doesn't much matter
322 though. Internally, we only call it with arguments less than half
323 the max value of a size_t, which should work across all reasonable
324 possibilities, although sometimes generating compiler warnings.
326 MORECORE_CONTIGUOUS default: 1 (true) if HAVE_MORECORE
327 If true, take advantage of fact that consecutive calls to MORECORE
328 with positive arguments always return contiguous increasing
329 addresses. This is true of unix sbrk. It does not hurt too much to
330 set it true anyway, since malloc copes with non-contiguities.
331 Setting it false when definitely non-contiguous saves time
332 and possibly wasted space it would take to discover this though.
334 MORECORE_CANNOT_TRIM default: NOT defined
335 True if MORECORE cannot release space back to the system when given
336 negative arguments. This is generally necessary only if you are
337 using a hand-crafted MORECORE function that cannot handle negative
338 arguments.
340 NO_SEGMENT_TRAVERSAL default: 0
341 If non-zero, suppresses traversals of memory segments
342 returned by either MORECORE or CALL_MMAP. This disables
343 merging of segments that are contiguous, and selectively
344 releasing them to the OS if unused, but bounds execution times.
346 HAVE_MMAP default: 1 (true)
347 True if this system supports mmap or an emulation of it. If so, and
348 HAVE_MORECORE is not true, MMAP is used for all system
349 allocation. If set and HAVE_MORECORE is true as well, MMAP is
350 primarily used to directly allocate very large blocks. It is also
351 used as a backup strategy in cases where MORECORE fails to provide
352 space from system. Note: A single call to MUNMAP is assumed to be
353 able to unmap memory that may have be allocated using multiple calls
354 to MMAP, so long as they are adjacent.
356 HAVE_MREMAP default: 1 on linux, else 0
357 If true realloc() uses mremap() to re-allocate large blocks and
358 extend or shrink allocation spaces.
360 MMAP_CLEARS default: 1 except on WINCE.
361 True if mmap clears memory so calloc doesn't need to. This is true
362 for standard unix mmap using /dev/zero and on WIN32 except for WINCE.
364 USE_BUILTIN_FFS default: 0 (i.e., not used)
365 Causes malloc to use the builtin ffs() function to compute indices.
366 Some compilers may recognize and intrinsify ffs to be faster than the
367 supplied C version. Also, the case of x86 using gcc is special-cased
368 to an asm instruction, so is already as fast as it can be, and so
369 this setting has no effect. Similarly for Win32 under recent MS compilers.
370 (On most x86s, the asm version is only slightly faster than the C version.)
372 malloc_getpagesize default: derive from system includes, or 4096.
373 The system page size. To the extent possible, this malloc manages
374 memory from the system in page-size units. This may be (and
375 usually is) a function rather than a constant. This is ignored
376 if WIN32, where page size is determined using getSystemInfo during
377 initialization.
379 USE_DEV_RANDOM default: 0 (i.e., not used)
380 Causes malloc to use /dev/random to initialize secure magic seed for
381 stamping footers. Otherwise, the current time is used.
383 NO_MALLINFO default: 0
384 If defined, don't compile "mallinfo". This can be a simple way
385 of dealing with mismatches between system declarations and
386 those in this file.
388 MALLINFO_FIELD_TYPE default: size_t
389 The type of the fields in the mallinfo struct. This was originally
390 defined as "int" in SVID etc, but is more usefully defined as
391 size_t. The value is used only if HAVE_USR_INCLUDE_MALLOC_H is not set
393 REALLOC_ZERO_BYTES_FREES default: not defined
394 This should be set if a call to realloc with zero bytes should
395 be the same as a call to free. Some people think it should. Otherwise,
396 since this malloc returns a unique pointer for malloc(0), so does
397 realloc(p, 0).
399 LACKS_UNISTD_H, LACKS_FCNTL_H, LACKS_SYS_PARAM_H, LACKS_SYS_MMAN_H
400 LACKS_STRINGS_H, LACKS_STRING_H, LACKS_SYS_TYPES_H, LACKS_ERRNO_H
401 LACKS_STDLIB_H default: NOT defined unless on WIN32
402 Define these if your system does not have these header files.
403 You might need to manually insert some of the declarations they provide.
405 DEFAULT_GRANULARITY default: page size if MORECORE_CONTIGUOUS,
406 system_info.dwAllocationGranularity in WIN32,
407 otherwise 64K.
408 Also settable using mallopt(M_GRANULARITY, x)
409 The unit for allocating and deallocating memory from the system. On
410 most systems with contiguous MORECORE, there is no reason to
411 make this more than a page. However, systems with MMAP tend to
412 either require or encourage larger granularities. You can increase
413 this value to prevent system allocation functions to be called so
414 often, especially if they are slow. The value must be at least one
415 page and must be a power of two. Setting to 0 causes initialization
416 to either page size or win32 region size. (Note: In previous
417 versions of malloc, the equivalent of this option was called
418 "TOP_PAD")
420 DEFAULT_TRIM_THRESHOLD default: 2MB
421 Also settable using mallopt(M_TRIM_THRESHOLD, x)
422 The maximum amount of unused top-most memory to keep before
423 releasing via malloc_trim in free(). Automatic trimming is mainly
424 useful in long-lived programs using contiguous MORECORE. Because
425 trimming via sbrk can be slow on some systems, and can sometimes be
426 wasteful (in cases where programs immediately afterward allocate
427 more large chunks) the value should be high enough so that your
428 overall system performance would improve by releasing this much
429 memory. As a rough guide, you might set to a value close to the
430 average size of a process (program) running on your system.
431 Releasing this much memory would allow such a process to run in
432 memory. Generally, it is worth tuning trim thresholds when a
433 program undergoes phases where several large chunks are allocated
434 and released in ways that can reuse each other's storage, perhaps
435 mixed with phases where there are no such chunks at all. The trim
436 value must be greater than page size to have any useful effect. To
437 disable trimming completely, you can set to MAX_SIZE_T. Note that the trick
438 some people use of mallocing a huge space and then freeing it at
439 program startup, in an attempt to reserve system memory, doesn't
440 have the intended effect under automatic trimming, since that memory
441 will immediately be returned to the system.
443 DEFAULT_MMAP_THRESHOLD default: 256K
444 Also settable using mallopt(M_MMAP_THRESHOLD, x)
445 The request size threshold for using MMAP to directly service a
446 request. Requests of at least this size that cannot be allocated
447 using already-existing space will be serviced via mmap. (If enough
448 normal freed space already exists it is used instead.) Using mmap
449 segregates relatively large chunks of memory so that they can be
450 individually obtained and released from the host system. A request
451 serviced through mmap is never reused by any other request (at least
452 not directly; the system may just so happen to remap successive
453 requests to the same locations). Segregating space in this way has
454 the benefits that: Mmapped space can always be individually released
455 back to the system, which helps keep the system level memory demands
456 of a long-lived program low. Also, mapped memory doesn't become
457 `locked' between other chunks, as can happen with normally allocated
458 chunks, which means that even trimming via malloc_trim would not
459 release them. However, it has the disadvantage that the space
460 cannot be reclaimed, consolidated, and then used to service later
461 requests, as happens with normal chunks. The advantages of mmap
462 nearly always outweigh disadvantages for "large" chunks, but the
463 value of "large" may vary across systems. The default is an
464 empirically derived value that works well in most systems. You can
465 disable mmap by setting to MAX_SIZE_T.
467 MAX_RELEASE_CHECK_RATE default: 4095 unless not HAVE_MMAP
468 The number of consolidated frees between checks to release
469 unused segments when freeing. When using non-contiguous segments,
470 especially with multiple mspaces, checking only for topmost space
471 doesn't always suffice to trigger trimming. To compensate for this,
472 free() will, with a period of MAX_RELEASE_CHECK_RATE (or the
473 current number of segments, if greater) try to release unused
474 segments to the OS when freeing chunks that result in
475 consolidation. The best value for this parameter is a compromise
476 between slowing down frees with relatively costly checks that
477 rarely trigger versus holding on to unused memory. To effectively
478 disable, set to MAX_SIZE_T. This may lead to a very slight speed
479 improvement at the expense of carrying around more memory.
482 /* Version identifier to allow people to support multiple versions */
483 #ifndef DLMALLOC_VERSION
484 #define DLMALLOC_VERSION 20804
485 #endif /* DLMALLOC_VERSION */
487 #ifndef WIN32
488 #ifdef _WIN32
489 #define WIN32 1
490 #endif /* _WIN32 */
491 #ifdef _WIN32_WCE
492 #define LACKS_FCNTL_H
493 #define WIN32 1
494 #endif /* _WIN32_WCE */
495 #endif /* WIN32 */
496 #ifdef WIN32
497 #define WIN32_LEAN_AND_MEAN
498 #define _WIN32_WINNT 0x403
499 #include <windows.h>
500 #define HAVE_MMAP 1
501 #define HAVE_MORECORE 0
502 #define LACKS_UNISTD_H
503 #define LACKS_SYS_PARAM_H
504 #define LACKS_SYS_MMAN_H
505 #define LACKS_STRING_H
506 #define LACKS_STRINGS_H
507 #define LACKS_SYS_TYPES_H
508 #define LACKS_ERRNO_H
509 #ifndef MALLOC_FAILURE_ACTION
510 #define MALLOC_FAILURE_ACTION
511 #endif /* MALLOC_FAILURE_ACTION */
512 #ifdef _WIN32_WCE /* WINCE reportedly does not clear */
513 #define MMAP_CLEARS 0
514 #else
515 #define MMAP_CLEARS 1
516 #endif /* _WIN32_WCE */
517 #endif /* WIN32 */
519 #if defined(DARWIN) || defined(_DARWIN)
520 /* Mac OSX docs advise not to use sbrk; it seems better to use mmap */
521 #ifndef HAVE_MORECORE
522 #define HAVE_MORECORE 0
523 #define HAVE_MMAP 1
524 /* OSX allocators provide 16 byte alignment */
525 #ifndef MALLOC_ALIGNMENT
526 #define MALLOC_ALIGNMENT ((size_t)16U)
527 #endif
528 #endif /* HAVE_MORECORE */
529 #endif /* DARWIN */
531 #ifndef LACKS_SYS_TYPES_H
532 #include <sys/types.h> /* For size_t */
533 #endif /* LACKS_SYS_TYPES_H */
535 /* The maximum possible size_t value has all bits set */
536 #define MAX_SIZE_T (~(size_t)0)
538 #ifndef ONLY_MSPACES
539 #define ONLY_MSPACES 0 /* define to a value */
540 #else
541 #define ONLY_MSPACES 1
542 #endif /* ONLY_MSPACES */
543 #ifndef MSPACES
544 #if ONLY_MSPACES
545 #define MSPACES 1
546 #else /* ONLY_MSPACES */
547 #define MSPACES 0
548 #endif /* ONLY_MSPACES */
549 #endif /* MSPACES */
550 #ifndef MALLOC_ALIGNMENT
551 #define MALLOC_ALIGNMENT ((size_t)8U)
552 #endif /* MALLOC_ALIGNMENT */
553 #ifndef FOOTERS
554 #define FOOTERS 0
555 #endif /* FOOTERS */
556 #ifndef ABORT
557 #define ABORT abort()
558 #endif /* ABORT */
559 #ifndef ABORT_ON_ASSERT_FAILURE
560 #define ABORT_ON_ASSERT_FAILURE 1
561 #endif /* ABORT_ON_ASSERT_FAILURE */
562 #ifndef PROCEED_ON_ERROR
563 #define PROCEED_ON_ERROR 0
564 #endif /* PROCEED_ON_ERROR */
565 #ifndef USE_LOCKS
566 #define USE_LOCKS 0
567 #endif /* USE_LOCKS */
568 #ifndef USE_SPIN_LOCKS
569 #if USE_LOCKS && (defined(__GNUC__) && ((defined(__i386__) || defined(__x86_64__)))) || (defined(_MSC_VER) && _MSC_VER>=1310)
570 #define USE_SPIN_LOCKS 1
571 #else
572 #define USE_SPIN_LOCKS 0
573 #endif /* USE_LOCKS && ... */
574 #endif /* USE_SPIN_LOCKS */
575 #ifndef INSECURE
576 #define INSECURE 0
577 #endif /* INSECURE */
578 #ifndef HAVE_MMAP
579 #define HAVE_MMAP 1
580 #endif /* HAVE_MMAP */
581 #ifndef MMAP_CLEARS
582 #define MMAP_CLEARS 1
583 #endif /* MMAP_CLEARS */
584 #ifndef HAVE_MREMAP
585 #ifdef linux
586 #define HAVE_MREMAP 1
587 #else /* linux */
588 #define HAVE_MREMAP 0
589 #endif /* linux */
590 #endif /* HAVE_MREMAP */
591 #ifndef MALLOC_FAILURE_ACTION
592 #define MALLOC_FAILURE_ACTION errno = ENOMEM;
593 #endif /* MALLOC_FAILURE_ACTION */
594 #ifndef HAVE_MORECORE
595 #if ONLY_MSPACES
596 #define HAVE_MORECORE 0
597 #else /* ONLY_MSPACES */
598 #define HAVE_MORECORE 1
599 #endif /* ONLY_MSPACES */
600 #endif /* HAVE_MORECORE */
601 #if !HAVE_MORECORE
602 #define MORECORE_CONTIGUOUS 0
603 #else /* !HAVE_MORECORE */
604 #define MORECORE_DEFAULT sbrk
605 #ifndef MORECORE_CONTIGUOUS
606 #define MORECORE_CONTIGUOUS 1
607 #endif /* MORECORE_CONTIGUOUS */
608 #endif /* HAVE_MORECORE */
609 #ifndef DEFAULT_GRANULARITY
610 #if (MORECORE_CONTIGUOUS || defined(WIN32))
611 #define DEFAULT_GRANULARITY (0) /* 0 means to compute in init_mparams */
612 #else /* MORECORE_CONTIGUOUS */
613 #define DEFAULT_GRANULARITY ((size_t)64U * (size_t)1024U)
614 #endif /* MORECORE_CONTIGUOUS */
615 #endif /* DEFAULT_GRANULARITY */
616 #ifndef DEFAULT_TRIM_THRESHOLD
617 #ifndef MORECORE_CANNOT_TRIM
618 #define DEFAULT_TRIM_THRESHOLD ((size_t)2U * (size_t)1024U * (size_t)1024U)
619 #else /* MORECORE_CANNOT_TRIM */
620 #define DEFAULT_TRIM_THRESHOLD MAX_SIZE_T
621 #endif /* MORECORE_CANNOT_TRIM */
622 #endif /* DEFAULT_TRIM_THRESHOLD */
623 #ifndef DEFAULT_MMAP_THRESHOLD
624 #if HAVE_MMAP
625 #define DEFAULT_MMAP_THRESHOLD ((size_t)256U * (size_t)1024U)
626 #else /* HAVE_MMAP */
627 #define DEFAULT_MMAP_THRESHOLD MAX_SIZE_T
628 #endif /* HAVE_MMAP */
629 #endif /* DEFAULT_MMAP_THRESHOLD */
630 #ifndef MAX_RELEASE_CHECK_RATE
631 #if HAVE_MMAP
632 #define MAX_RELEASE_CHECK_RATE 4095
633 #else
634 #define MAX_RELEASE_CHECK_RATE MAX_SIZE_T
635 #endif /* HAVE_MMAP */
636 #endif /* MAX_RELEASE_CHECK_RATE */
637 #ifndef USE_BUILTIN_FFS
638 #define USE_BUILTIN_FFS 0
639 #endif /* USE_BUILTIN_FFS */
640 #ifndef USE_DEV_RANDOM
641 #define USE_DEV_RANDOM 0
642 #endif /* USE_DEV_RANDOM */
643 #ifndef NO_MALLINFO
644 #define NO_MALLINFO 0
645 #endif /* NO_MALLINFO */
646 #ifndef MALLINFO_FIELD_TYPE
647 #define MALLINFO_FIELD_TYPE size_t
648 #endif /* MALLINFO_FIELD_TYPE */
649 #ifndef NO_SEGMENT_TRAVERSAL
650 #define NO_SEGMENT_TRAVERSAL 0
651 #endif /* NO_SEGMENT_TRAVERSAL */
654 mallopt tuning options. SVID/XPG defines four standard parameter
655 numbers for mallopt, normally defined in malloc.h. None of these
656 are used in this malloc, so setting them has no effect. But this
657 malloc does support the following options.
660 #define M_TRIM_THRESHOLD (-1)
661 #define M_GRANULARITY (-2)
662 #define M_MMAP_THRESHOLD (-3)
664 /* ------------------------ Mallinfo declarations ------------------------ */
666 #if !NO_MALLINFO
668 This version of malloc supports the standard SVID/XPG mallinfo
669 routine that returns a struct containing usage properties and
670 statistics. It should work on any system that has a
671 /usr/include/malloc.h defining struct mallinfo. The main
672 declaration needed is the mallinfo struct that is returned (by-copy)
673 by mallinfo(). The malloinfo struct contains a bunch of fields that
674 are not even meaningful in this version of malloc. These fields are
675 are instead filled by mallinfo() with other numbers that might be of
676 interest.
678 HAVE_USR_INCLUDE_MALLOC_H should be set if you have a
679 /usr/include/malloc.h file that includes a declaration of struct
680 mallinfo. If so, it is included; else a compliant version is
681 declared below. These must be precisely the same for mallinfo() to
682 work. The original SVID version of this struct, defined on most
683 systems with mallinfo, declares all fields as ints. But some others
684 define as unsigned long. If your system defines the fields using a
685 type of different width than listed here, you MUST #include your
686 system version and #define HAVE_USR_INCLUDE_MALLOC_H.
689 /* #define HAVE_USR_INCLUDE_MALLOC_H */
691 #ifdef HAVE_USR_INCLUDE_MALLOC_H
692 #include "/usr/include/malloc.h"
693 #else /* HAVE_USR_INCLUDE_MALLOC_H */
694 #ifndef STRUCT_MALLINFO_DECLARED
695 #define STRUCT_MALLINFO_DECLARED 1
696 struct mallinfo {
697 MALLINFO_FIELD_TYPE arena; /* non-mmapped space allocated from system */
698 MALLINFO_FIELD_TYPE ordblks; /* number of free chunks */
699 MALLINFO_FIELD_TYPE smblks; /* always 0 */
700 MALLINFO_FIELD_TYPE hblks; /* always 0 */
701 MALLINFO_FIELD_TYPE hblkhd; /* space in mmapped regions */
702 MALLINFO_FIELD_TYPE usmblks; /* maximum total allocated space */
703 MALLINFO_FIELD_TYPE fsmblks; /* always 0 */
704 MALLINFO_FIELD_TYPE uordblks; /* total allocated space */
705 MALLINFO_FIELD_TYPE fordblks; /* total free space */
706 MALLINFO_FIELD_TYPE keepcost; /* releasable (via malloc_trim) space */
708 #endif /* STRUCT_MALLINFO_DECLARED */
709 #endif /* HAVE_USR_INCLUDE_MALLOC_H */
710 #endif /* NO_MALLINFO */
713 Try to persuade compilers to inline. The most critical functions for
714 inlining are defined as macros, so these aren't used for them.
717 #ifndef FORCEINLINE
718 #if defined(__GNUC__)
719 #define FORCEINLINE __inline __attribute__ ((always_inline))
720 #elif defined(_MSC_VER)
721 #define FORCEINLINE __forceinline
722 #endif
723 #endif
724 #ifndef NOINLINE
725 #if defined(__GNUC__)
726 #define NOINLINE __attribute__ ((noinline))
727 #elif defined(_MSC_VER)
728 #define NOINLINE __declspec(noinline)
729 #else
730 #define NOINLINE
731 #endif
732 #endif
734 #ifdef __cplusplus
735 extern "C" {
736 #ifndef FORCEINLINE
737 #define FORCEINLINE inline
738 #endif
739 #endif /* __cplusplus */
740 #ifndef FORCEINLINE
741 #define FORCEINLINE
742 #endif
744 #if !ONLY_MSPACES
746 /* ------------------- Declarations of public routines ------------------- */
748 #ifndef USE_DL_PREFIX
749 #define dlcalloc calloc
750 #define dlfree free
751 #define dlmalloc malloc
752 #define dlmemalign memalign
753 #define dlrealloc realloc
754 #define dlvalloc valloc
755 #define dlpvalloc pvalloc
756 #define dlmallinfo mallinfo
757 #define dlmallopt mallopt
758 #define dlmalloc_trim malloc_trim
759 #define dlmalloc_stats malloc_stats
760 #define dlmalloc_usable_size malloc_usable_size
761 #define dlmalloc_footprint malloc_footprint
762 #define dlmalloc_max_footprint malloc_max_footprint
763 #define dlindependent_calloc independent_calloc
764 #define dlindependent_comalloc independent_comalloc
765 #endif /* USE_DL_PREFIX */
769 malloc(size_t n)
770 Returns a pointer to a newly allocated chunk of at least n bytes, or
771 null if no space is available, in which case errno is set to ENOMEM
772 on ANSI C systems.
774 If n is zero, malloc returns a minimum-sized chunk. (The minimum
775 size is 16 bytes on most 32bit systems, and 32 bytes on 64bit
776 systems.) Note that size_t is an unsigned type, so calls with
777 arguments that would be negative if signed are interpreted as
778 requests for huge amounts of space, which will often fail. The
779 maximum supported value of n differs across systems, but is in all
780 cases less than the maximum representable value of a size_t.
782 void* dlmalloc(size_t);
785 free(void* p)
786 Releases the chunk of memory pointed to by p, that had been previously
787 allocated using malloc or a related routine such as realloc.
788 It has no effect if p is null. If p was not malloced or already
789 freed, free(p) will by default cause the current program to abort.
791 void dlfree(void*);
794 calloc(size_t n_elements, size_t element_size);
795 Returns a pointer to n_elements * element_size bytes, with all locations
796 set to zero.
798 void* dlcalloc(size_t, size_t);
801 realloc(void* p, size_t n)
802 Returns a pointer to a chunk of size n that contains the same data
803 as does chunk p up to the minimum of (n, p's size) bytes, or null
804 if no space is available.
806 The returned pointer may or may not be the same as p. The algorithm
807 prefers extending p in most cases when possible, otherwise it
808 employs the equivalent of a malloc-copy-free sequence.
810 If p is null, realloc is equivalent to malloc.
812 If space is not available, realloc returns null, errno is set (if on
813 ANSI) and p is NOT freed.
815 if n is for fewer bytes than already held by p, the newly unused
816 space is lopped off and freed if possible. realloc with a size
817 argument of zero (re)allocates a minimum-sized chunk.
819 The old unix realloc convention of allowing the last-free'd chunk
820 to be used as an argument to realloc is not supported.
823 void* dlrealloc(void*, size_t);
826 memalign(size_t alignment, size_t n);
827 Returns a pointer to a newly allocated chunk of n bytes, aligned
828 in accord with the alignment argument.
830 The alignment argument should be a power of two. If the argument is
831 not a power of two, the nearest greater power is used.
832 8-byte alignment is guaranteed by normal malloc calls, so don't
833 bother calling memalign with an argument of 8 or less.
835 Overreliance on memalign is a sure way to fragment space.
837 void* dlmemalign(size_t, size_t);
840 valloc(size_t n);
841 Equivalent to memalign(pagesize, n), where pagesize is the page
842 size of the system. If the pagesize is unknown, 4096 is used.
844 void* dlvalloc(size_t);
847 mallopt(int parameter_number, int parameter_value)
848 Sets tunable parameters The format is to provide a
849 (parameter-number, parameter-value) pair. mallopt then sets the
850 corresponding parameter to the argument value if it can (i.e., so
851 long as the value is meaningful), and returns 1 if successful else
852 0. To workaround the fact that mallopt is specified to use int,
853 not size_t parameters, the value -1 is specially treated as the
854 maximum unsigned size_t value.
856 SVID/XPG/ANSI defines four standard param numbers for mallopt,
857 normally defined in malloc.h. None of these are use in this malloc,
858 so setting them has no effect. But this malloc also supports other
859 options in mallopt. See below for details. Briefly, supported
860 parameters are as follows (listed defaults are for "typical"
861 configurations).
863 Symbol param # default allowed param values
864 M_TRIM_THRESHOLD -1 2*1024*1024 any (-1 disables)
865 M_GRANULARITY -2 page size any power of 2 >= page size
866 M_MMAP_THRESHOLD -3 256*1024 any (or 0 if no MMAP support)
868 int dlmallopt(int, int);
871 malloc_footprint();
872 Returns the number of bytes obtained from the system. The total
873 number of bytes allocated by malloc, realloc etc., is less than this
874 value. Unlike mallinfo, this function returns only a precomputed
875 result, so can be called frequently to monitor memory consumption.
876 Even if locks are otherwise defined, this function does not use them,
877 so results might not be up to date.
879 size_t dlmalloc_footprint(void);
882 malloc_max_footprint();
883 Returns the maximum number of bytes obtained from the system. This
884 value will be greater than current footprint if deallocated space
885 has been reclaimed by the system. The peak number of bytes allocated
886 by malloc, realloc etc., is less than this value. Unlike mallinfo,
887 this function returns only a precomputed result, so can be called
888 frequently to monitor memory consumption. Even if locks are
889 otherwise defined, this function does not use them, so results might
890 not be up to date.
892 size_t dlmalloc_max_footprint(void);
894 #if !NO_MALLINFO
896 mallinfo()
897 Returns (by copy) a struct containing various summary statistics:
899 arena: current total non-mmapped bytes allocated from system
900 ordblks: the number of free chunks
901 smblks: always zero.
902 hblks: current number of mmapped regions
903 hblkhd: total bytes held in mmapped regions
904 usmblks: the maximum total allocated space. This will be greater
905 than current total if trimming has occurred.
906 fsmblks: always zero
907 uordblks: current total allocated space (normal or mmapped)
908 fordblks: total free space
909 keepcost: the maximum number of bytes that could ideally be released
910 back to system via malloc_trim. ("ideally" means that
911 it ignores page restrictions etc.)
913 Because these fields are ints, but internal bookkeeping may
914 be kept as longs, the reported values may wrap around zero and
915 thus be inaccurate.
917 struct mallinfo dlmallinfo(void);
918 #endif /* NO_MALLINFO */
921 independent_calloc(size_t n_elements, size_t element_size, void* chunks[]);
923 independent_calloc is similar to calloc, but instead of returning a
924 single cleared space, it returns an array of pointers to n_elements
925 independent elements that can hold contents of size elem_size, each
926 of which starts out cleared, and can be independently freed,
927 realloc'ed etc. The elements are guaranteed to be adjacently
928 allocated (this is not guaranteed to occur with multiple callocs or
929 mallocs), which may also improve cache locality in some
930 applications.
932 The "chunks" argument is optional (i.e., may be null, which is
933 probably the most typical usage). If it is null, the returned array
934 is itself dynamically allocated and should also be freed when it is
935 no longer needed. Otherwise, the chunks array must be of at least
936 n_elements in length. It is filled in with the pointers to the
937 chunks.
939 In either case, independent_calloc returns this pointer array, or
940 null if the allocation failed. If n_elements is zero and "chunks"
941 is null, it returns a chunk representing an array with zero elements
942 (which should be freed if not wanted).
944 Each element must be individually freed when it is no longer
945 needed. If you'd like to instead be able to free all at once, you
946 should instead use regular calloc and assign pointers into this
947 space to represent elements. (In this case though, you cannot
948 independently free elements.)
950 independent_calloc simplifies and speeds up implementations of many
951 kinds of pools. It may also be useful when constructing large data
952 structures that initially have a fixed number of fixed-sized nodes,
953 but the number is not known at compile time, and some of the nodes
954 may later need to be freed. For example:
956 struct Node { int item; struct Node* next; };
958 struct Node* build_list() {
959 struct Node** pool;
960 int n = read_number_of_nodes_needed();
961 if (n <= 0) return 0;
962 pool = (struct Node**)(independent_calloc(n, sizeof(struct Node), 0);
963 if (pool == 0) die();
964 // organize into a linked list...
965 struct Node* first = pool[0];
966 for (i = 0; i < n-1; ++i)
967 pool[i]->next = pool[i+1];
968 free(pool); // Can now free the array (or not, if it is needed later)
969 return first;
972 void** dlindependent_calloc(size_t, size_t, void**);
975 independent_comalloc(size_t n_elements, size_t sizes[], void* chunks[]);
977 independent_comalloc allocates, all at once, a set of n_elements
978 chunks with sizes indicated in the "sizes" array. It returns
979 an array of pointers to these elements, each of which can be
980 independently freed, realloc'ed etc. The elements are guaranteed to
981 be adjacently allocated (this is not guaranteed to occur with
982 multiple callocs or mallocs), which may also improve cache locality
983 in some applications.
985 The "chunks" argument is optional (i.e., may be null). If it is null
986 the returned array is itself dynamically allocated and should also
987 be freed when it is no longer needed. Otherwise, the chunks array
988 must be of at least n_elements in length. It is filled in with the
989 pointers to the chunks.
991 In either case, independent_comalloc returns this pointer array, or
992 null if the allocation failed. If n_elements is zero and chunks is
993 null, it returns a chunk representing an array with zero elements
994 (which should be freed if not wanted).
996 Each element must be individually freed when it is no longer
997 needed. If you'd like to instead be able to free all at once, you
998 should instead use a single regular malloc, and assign pointers at
999 particular offsets in the aggregate space. (In this case though, you
1000 cannot independently free elements.)
1002 independent_comallac differs from independent_calloc in that each
1003 element may have a different size, and also that it does not
1004 automatically clear elements.
1006 independent_comalloc can be used to speed up allocation in cases
1007 where several structs or objects must always be allocated at the
1008 same time. For example:
1010 struct Head { ... }
1011 struct Foot { ... }
1013 void send_message(char* msg) {
1014 int msglen = strlen(msg);
1015 size_t sizes[3] = { sizeof(struct Head), msglen, sizeof(struct Foot) };
1016 void* chunks[3];
1017 if (independent_comalloc(3, sizes, chunks) == 0)
1018 die();
1019 struct Head* head = (struct Head*)(chunks[0]);
1020 char* body = (char*)(chunks[1]);
1021 struct Foot* foot = (struct Foot*)(chunks[2]);
1022 // ...
1025 In general though, independent_comalloc is worth using only for
1026 larger values of n_elements. For small values, you probably won't
1027 detect enough difference from series of malloc calls to bother.
1029 Overuse of independent_comalloc can increase overall memory usage,
1030 since it cannot reuse existing noncontiguous small chunks that
1031 might be available for some of the elements.
1033 void** dlindependent_comalloc(size_t, size_t*, void**);
1037 pvalloc(size_t n);
1038 Equivalent to valloc(minimum-page-that-holds(n)), that is,
1039 round up n to nearest pagesize.
1041 void* dlpvalloc(size_t);
1044 malloc_trim(size_t pad);
1046 If possible, gives memory back to the system (via negative arguments
1047 to sbrk) if there is unused memory at the `high' end of the malloc
1048 pool or in unused MMAP segments. You can call this after freeing
1049 large blocks of memory to potentially reduce the system-level memory
1050 requirements of a program. However, it cannot guarantee to reduce
1051 memory. Under some allocation patterns, some large free blocks of
1052 memory will be locked between two used chunks, so they cannot be
1053 given back to the system.
1055 The `pad' argument to malloc_trim represents the amount of free
1056 trailing space to leave untrimmed. If this argument is zero, only
1057 the minimum amount of memory to maintain internal data structures
1058 will be left. Non-zero arguments can be supplied to maintain enough
1059 trailing space to service future expected allocations without having
1060 to re-obtain memory from the system.
1062 Malloc_trim returns 1 if it actually released any memory, else 0.
1064 int dlmalloc_trim(size_t);
1067 malloc_stats();
1068 Prints on stderr the amount of space obtained from the system (both
1069 via sbrk and mmap), the maximum amount (which may be more than
1070 current if malloc_trim and/or munmap got called), and the current
1071 number of bytes allocated via malloc (or realloc, etc) but not yet
1072 freed. Note that this is the number of bytes allocated, not the
1073 number requested. It will be larger than the number requested
1074 because of alignment and bookkeeping overhead. Because it includes
1075 alignment wastage as being in use, this figure may be greater than
1076 zero even when no user-level chunks are allocated.
1078 The reported current and maximum system memory can be inaccurate if
1079 a program makes other calls to system memory allocation functions
1080 (normally sbrk) outside of malloc.
1082 malloc_stats prints only the most commonly interesting statistics.
1083 More information can be obtained by calling mallinfo.
1085 void dlmalloc_stats(void);
1087 #endif /* ONLY_MSPACES */
1090 malloc_usable_size(void* p);
1092 Returns the number of bytes you can actually use in
1093 an allocated chunk, which may be more than you requested (although
1094 often not) due to alignment and minimum size constraints.
1095 You can use this many bytes without worrying about
1096 overwriting other allocated objects. This is not a particularly great
1097 programming practice. malloc_usable_size can be more useful in
1098 debugging and assertions, for example:
1100 p = malloc(n);
1101 assert(malloc_usable_size(p) >= 256);
1103 size_t dlmalloc_usable_size(void*);
1106 #if MSPACES
1109 mspace is an opaque type representing an independent
1110 region of space that supports mspace_malloc, etc.
1112 typedef void* mspace;
1115 create_mspace creates and returns a new independent space with the
1116 given initial capacity, or, if 0, the default granularity size. It
1117 returns null if there is no system memory available to create the
1118 space. If argument locked is non-zero, the space uses a separate
1119 lock to control access. The capacity of the space will grow
1120 dynamically as needed to service mspace_malloc requests. You can
1121 control the sizes of incremental increases of this space by
1122 compiling with a different DEFAULT_GRANULARITY or dynamically
1123 setting with mallopt(M_GRANULARITY, value).
1125 mspace create_mspace(size_t capacity, int locked);
1128 destroy_mspace destroys the given space, and attempts to return all
1129 of its memory back to the system, returning the total number of
1130 bytes freed. After destruction, the results of access to all memory
1131 used by the space become undefined.
1133 size_t destroy_mspace(mspace msp);
1136 create_mspace_with_base uses the memory supplied as the initial base
1137 of a new mspace. Part (less than 128*sizeof(size_t) bytes) of this
1138 space is used for bookkeeping, so the capacity must be at least this
1139 large. (Otherwise 0 is returned.) When this initial space is
1140 exhausted, additional memory will be obtained from the system.
1141 Destroying this space will deallocate all additionally allocated
1142 space (if possible) but not the initial base.
1144 mspace create_mspace_with_base(void* base, size_t capacity, int locked);
1147 mspace_mmap_large_chunks controls whether requests for large chunks
1148 are allocated in their own mmapped regions, separate from others in
1149 this mspace. By default this is enabled, which reduces
1150 fragmentation. However, such chunks are not necessarily released to
1151 the system upon destroy_mspace. Disabling by setting to false may
1152 increase fragmentation, but avoids leakage when relying on
1153 destroy_mspace to release all memory allocated using this space.
1155 int mspace_mmap_large_chunks(mspace msp, int enable);
1159 mspace_malloc behaves as malloc, but operates within
1160 the given space.
1162 void* mspace_malloc(mspace msp, size_t bytes);
1165 mspace_free behaves as free, but operates within
1166 the given space.
1168 If compiled with FOOTERS==1, mspace_free is not actually needed.
1169 free may be called instead of mspace_free because freed chunks from
1170 any space are handled by their originating spaces.
1172 void mspace_free(mspace msp, void* mem);
1175 mspace_realloc behaves as realloc, but operates within
1176 the given space.
1178 If compiled with FOOTERS==1, mspace_realloc is not actually
1179 needed. realloc may be called instead of mspace_realloc because
1180 realloced chunks from any space are handled by their originating
1181 spaces.
1183 void* mspace_realloc(mspace msp, void* mem, size_t newsize);
1186 mspace_calloc behaves as calloc, but operates within
1187 the given space.
1189 void* mspace_calloc(mspace msp, size_t n_elements, size_t elem_size);
1192 mspace_memalign behaves as memalign, but operates within
1193 the given space.
1195 void* mspace_memalign(mspace msp, size_t alignment, size_t bytes);
1198 mspace_independent_calloc behaves as independent_calloc, but
1199 operates within the given space.
1201 void** mspace_independent_calloc(mspace msp, size_t n_elements,
1202 size_t elem_size, void* chunks[]);
1205 mspace_independent_comalloc behaves as independent_comalloc, but
1206 operates within the given space.
1208 void** mspace_independent_comalloc(mspace msp, size_t n_elements,
1209 size_t sizes[], void* chunks[]);
1212 mspace_footprint() returns the number of bytes obtained from the
1213 system for this space.
1215 size_t mspace_footprint(mspace msp);
1218 mspace_max_footprint() returns the peak number of bytes obtained from the
1219 system for this space.
1221 size_t mspace_max_footprint(mspace msp);
1224 #if !NO_MALLINFO
1226 mspace_mallinfo behaves as mallinfo, but reports properties of
1227 the given space.
1229 struct mallinfo mspace_mallinfo(mspace msp);
1230 #endif /* NO_MALLINFO */
1233 malloc_usable_size(void* p) behaves the same as malloc_usable_size;
1235 size_t mspace_usable_size(void* mem);
1238 mspace_malloc_stats behaves as malloc_stats, but reports
1239 properties of the given space.
1241 void mspace_malloc_stats(mspace msp);
1244 mspace_trim behaves as malloc_trim, but
1245 operates within the given space.
1247 int mspace_trim(mspace msp, size_t pad);
1250 An alias for mallopt.
1252 int mspace_mallopt(int, int);
1254 #endif /* MSPACES */
1256 #ifdef __cplusplus
1257 }; /* end of extern "C" */
1258 #endif /* __cplusplus */
1261 ========================================================================
1262 To make a fully customizable malloc.h header file, cut everything
1263 above this line, put into file malloc.h, edit to suit, and #include it
1264 on the next line, as well as in programs that use this malloc.
1265 ========================================================================
1268 /* #include "malloc.h" */
1270 /*------------------------------ internal #includes ---------------------- */
1272 #ifdef WIN32
1273 #ifndef __GNUC__
1274 #pragma warning( disable : 4146 ) /* no "unsigned" warnings */
1275 #endif
1276 #endif /* WIN32 */
1278 #include <stdio.h> /* for printing in malloc_stats */
1280 #ifndef LACKS_ERRNO_H
1281 #include <errno.h> /* for MALLOC_FAILURE_ACTION */
1282 #endif /* LACKS_ERRNO_H */
1283 #if FOOTERS
1284 #include <time.h> /* for magic initialization */
1285 #endif /* FOOTERS */
1286 #ifndef LACKS_STDLIB_H
1287 #include <stdlib.h> /* for abort() */
1288 #endif /* LACKS_STDLIB_H */
1289 #ifdef DEBUG
1290 #if ABORT_ON_ASSERT_FAILURE
1291 #define assert(x) if(!(x)) ABORT
1292 #else /* ABORT_ON_ASSERT_FAILURE */
1293 #include <assert.h>
1294 #endif /* ABORT_ON_ASSERT_FAILURE */
1295 #else /* DEBUG */
1296 #ifndef assert
1297 #define assert(x)
1298 #endif
1299 #define DEBUG 0
1300 #endif /* DEBUG */
1301 #ifndef LACKS_STRING_H
1302 #include <string.h> /* for memset etc */
1303 #endif /* LACKS_STRING_H */
1304 #if USE_BUILTIN_FFS
1305 #ifndef LACKS_STRINGS_H
1306 #include <strings.h> /* for ffs */
1307 #endif /* LACKS_STRINGS_H */
1308 #endif /* USE_BUILTIN_FFS */
1309 #if HAVE_MMAP
1310 #ifndef LACKS_SYS_MMAN_H
1311 #include <sys/mman.h> /* for mmap */
1312 #endif /* LACKS_SYS_MMAN_H */
1313 #ifndef LACKS_FCNTL_H
1314 #include <fcntl.h>
1315 #endif /* LACKS_FCNTL_H */
1316 #endif /* HAVE_MMAP */
1317 #ifndef LACKS_UNISTD_H
1318 #include <unistd.h> /* for sbrk, sysconf */
1319 #else /* LACKS_UNISTD_H */
1320 #if !defined(__FreeBSD__) && !defined(__OpenBSD__) && !defined(__NetBSD__)
1321 extern void* sbrk(ptrdiff_t);
1322 #endif /* FreeBSD etc */
1323 #endif /* LACKS_UNISTD_H */
1325 /* Declarations for locking */
1326 #if USE_LOCKS
1327 #ifndef WIN32
1328 #include <pthread.h>
1329 #if defined (__SVR4) && defined (__sun) /* solaris */
1330 #include <thread.h>
1331 #endif /* solaris */
1332 #else
1333 #ifndef _M_AMD64
1334 /* These are already defined on AMD64 builds */
1335 #ifdef __cplusplus
1336 extern "C" {
1337 #endif /* __cplusplus */
1338 #ifndef __MINGW32__
1339 LONG __cdecl _InterlockedCompareExchange(LONG volatile *Dest, LONG Exchange, LONG Comp);
1340 LONG __cdecl _InterlockedExchange(LONG volatile *Target, LONG Value);
1341 #endif
1342 #ifdef __cplusplus
1344 #endif /* __cplusplus */
1345 #endif /* _M_AMD64 */
1346 #ifndef __MINGW32__
1347 #pragma intrinsic (_InterlockedCompareExchange)
1348 #pragma intrinsic (_InterlockedExchange)
1349 #else
1350 /* --[ start GCC compatibility ]----------------------------------------------
1351 * Compatibility <intrin_x86.h> header for GCC -- GCC equivalents of intrinsic
1352 * Microsoft Visual C++ functions. Originally developed for the ReactOS
1353 * (<http://www.reactos.org/>) and TinyKrnl (<http://www.tinykrnl.org/>)
1354 * projects.
1356 * Copyright (c) 2006 KJK::Hyperion <hackbunny@reactos.com>
1358 * Permission is hereby granted, free of charge, to any person obtaining a
1359 * copy of this software and associated documentation files (the "Software"),
1360 * to deal in the Software without restriction, including without limitation
1361 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
1362 * and/or sell copies of the Software, and to permit persons to whom the
1363 * Software is furnished to do so, subject to the following conditions:
1365 * The above copyright notice and this permission notice shall be included in
1366 * all copies or substantial portions of the Software.
1368 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
1369 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
1370 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
1371 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
1372 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
1373 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
1374 * DEALINGS IN THE SOFTWARE.
1377 /*** Atomic operations ***/
1378 #if (__GNUC__ * 10000 + __GNUC_MINOR__ * 100 + __GNUC_PATCHLEVEL__) > 40100
1379 #define _ReadWriteBarrier() __sync_synchronize()
1380 #else
1381 static __inline__ __attribute__((always_inline)) long __sync_lock_test_and_set(volatile long * const Target, const long Value)
1383 long res;
1384 __asm__ __volatile__("xchg%z0 %2, %0" : "=g" (*(Target)), "=r" (res) : "1" (Value));
1385 return res;
1387 static void __inline__ __attribute__((always_inline)) _MemoryBarrier(void)
1389 __asm__ __volatile__("" : : : "memory");
1391 #define _ReadWriteBarrier() _MemoryBarrier()
1392 #endif
1393 /* BUGBUG: GCC only supports full barriers */
1394 static __inline__ __attribute__((always_inline)) long _InterlockedExchange(volatile long * const Target, const long Value)
1396 /* NOTE: __sync_lock_test_and_set would be an acquire barrier, so we force a full barrier */
1397 _ReadWriteBarrier();
1398 return __sync_lock_test_and_set(Target, Value);
1400 /* --[ end GCC compatibility ]---------------------------------------------- */
1401 #endif
1402 #define interlockedcompareexchange _InterlockedCompareExchange
1403 #define interlockedexchange _InterlockedExchange
1404 #endif /* Win32 */
1405 #endif /* USE_LOCKS */
1407 /* Declarations for bit scanning on win32 */
1408 #if defined(_MSC_VER) && _MSC_VER>=1300
1409 #ifndef BitScanForward /* Try to avoid pulling in WinNT.h */
1410 #ifdef __cplusplus
1411 extern "C" {
1412 #endif /* __cplusplus */
1413 unsigned char _BitScanForward(unsigned long *index, unsigned long mask);
1414 unsigned char _BitScanReverse(unsigned long *index, unsigned long mask);
1415 #ifdef __cplusplus
1417 #endif /* __cplusplus */
1419 #define BitScanForward _BitScanForward
1420 #define BitScanReverse _BitScanReverse
1421 #pragma intrinsic(_BitScanForward)
1422 #pragma intrinsic(_BitScanReverse)
1423 #endif /* BitScanForward */
1424 #endif /* defined(_MSC_VER) && _MSC_VER>=1300 */
1426 #ifndef WIN32
1427 #ifndef malloc_getpagesize
1428 # ifdef _SC_PAGESIZE /* some SVR4 systems omit an underscore */
1429 # ifndef _SC_PAGE_SIZE
1430 # define _SC_PAGE_SIZE _SC_PAGESIZE
1431 # endif
1432 # endif
1433 # ifdef _SC_PAGE_SIZE
1434 # define malloc_getpagesize sysconf(_SC_PAGE_SIZE)
1435 # else
1436 # if defined(BSD) || defined(DGUX) || defined(HAVE_GETPAGESIZE)
1437 extern size_t getpagesize();
1438 # define malloc_getpagesize getpagesize()
1439 # else
1440 # ifdef WIN32 /* use supplied emulation of getpagesize */
1441 # define malloc_getpagesize getpagesize()
1442 # else
1443 # ifndef LACKS_SYS_PARAM_H
1444 # include <sys/param.h>
1445 # endif
1446 # ifdef EXEC_PAGESIZE
1447 # define malloc_getpagesize EXEC_PAGESIZE
1448 # else
1449 # ifdef NBPG
1450 # ifndef CLSIZE
1451 # define malloc_getpagesize NBPG
1452 # else
1453 # define malloc_getpagesize (NBPG * CLSIZE)
1454 # endif
1455 # else
1456 # ifdef NBPC
1457 # define malloc_getpagesize NBPC
1458 # else
1459 # ifdef PAGESIZE
1460 # define malloc_getpagesize PAGESIZE
1461 # else /* just guess */
1462 # define malloc_getpagesize ((size_t)4096U)
1463 # endif
1464 # endif
1465 # endif
1466 # endif
1467 # endif
1468 # endif
1469 # endif
1470 #endif
1471 #endif
1475 /* ------------------- size_t and alignment properties -------------------- */
1477 /* The byte and bit size of a size_t */
1478 #define SIZE_T_SIZE (sizeof(size_t))
1479 #define SIZE_T_BITSIZE (sizeof(size_t) << 3)
1481 /* Some constants coerced to size_t */
1482 /* Annoying but necessary to avoid errors on some platforms */
1483 #define SIZE_T_ZERO ((size_t)0)
1484 #define SIZE_T_ONE ((size_t)1)
1485 #define SIZE_T_TWO ((size_t)2)
1486 #define SIZE_T_FOUR ((size_t)4)
1487 #define TWO_SIZE_T_SIZES (SIZE_T_SIZE<<1)
1488 #define FOUR_SIZE_T_SIZES (SIZE_T_SIZE<<2)
1489 #define SIX_SIZE_T_SIZES (FOUR_SIZE_T_SIZES+TWO_SIZE_T_SIZES)
1490 #define HALF_MAX_SIZE_T (MAX_SIZE_T / 2U)
1492 /* The bit mask value corresponding to MALLOC_ALIGNMENT */
1493 #define CHUNK_ALIGN_MASK (MALLOC_ALIGNMENT - SIZE_T_ONE)
1495 /* True if address a has acceptable alignment */
1496 #define is_aligned(A) (((size_t)((A)) & (CHUNK_ALIGN_MASK)) == 0)
1498 /* the number of bytes to offset an address to align it */
1499 #define align_offset(A)\
1500 ((((size_t)(A) & CHUNK_ALIGN_MASK) == 0)? 0 :\
1501 ((MALLOC_ALIGNMENT - ((size_t)(A) & CHUNK_ALIGN_MASK)) & CHUNK_ALIGN_MASK))
1503 /* -------------------------- MMAP preliminaries ------------------------- */
1506 If HAVE_MORECORE or HAVE_MMAP are false, we just define calls and
1507 checks to fail so compiler optimizer can delete code rather than
1508 using so many "#if"s.
1512 /* MORECORE and MMAP must return MFAIL on failure */
1513 #define MFAIL ((void*)(MAX_SIZE_T))
1514 #define CMFAIL ((char*)(MFAIL)) /* defined for convenience */
1516 #if HAVE_MMAP
1518 #ifndef WIN32
1519 #define MUNMAP_DEFAULT(a, s) munmap((a), (s))
1520 #define MMAP_PROT (PROT_READ|PROT_WRITE)
1521 #if !defined(MAP_ANONYMOUS) && defined(MAP_ANON)
1522 #define MAP_ANONYMOUS MAP_ANON
1523 #endif /* MAP_ANON */
1524 #ifdef MAP_ANONYMOUS
1525 #define MMAP_FLAGS (MAP_PRIVATE|MAP_ANONYMOUS)
1526 #define MMAP_DEFAULT(s) mmap(0, (s), MMAP_PROT, MMAP_FLAGS, -1, 0)
1527 #else /* MAP_ANONYMOUS */
1529 Nearly all versions of mmap support MAP_ANONYMOUS, so the following
1530 is unlikely to be needed, but is supplied just in case.
1532 #define MMAP_FLAGS (MAP_PRIVATE)
1533 static int dev_zero_fd = -1; /* Cached file descriptor for /dev/zero. */
1534 #define MMAP_DEFAULT(s) ((dev_zero_fd < 0) ? \
1535 (dev_zero_fd = open("/dev/zero", O_RDWR), \
1536 mmap(0, (s), MMAP_PROT, MMAP_FLAGS, dev_zero_fd, 0)) : \
1537 mmap(0, (s), MMAP_PROT, MMAP_FLAGS, dev_zero_fd, 0))
1538 #endif /* MAP_ANONYMOUS */
1540 #define DIRECT_MMAP_DEFAULT(s) MMAP_DEFAULT(s)
1542 #else /* WIN32 */
1544 /* Win32 MMAP via VirtualAlloc */
1545 static FORCEINLINE void* win32mmap(size_t size) {
1546 void* ptr = VirtualAlloc(0, size, MEM_RESERVE|MEM_COMMIT, PAGE_READWRITE);
1547 return (ptr != 0)? ptr: MFAIL;
1550 /* For direct MMAP, use MEM_TOP_DOWN to minimize interference */
1551 static FORCEINLINE void* win32direct_mmap(size_t size) {
1552 void* ptr = VirtualAlloc(0, size, MEM_RESERVE|MEM_COMMIT|MEM_TOP_DOWN,
1553 PAGE_READWRITE);
1554 return (ptr != 0)? ptr: MFAIL;
1557 /* This function supports releasing coalesed segments */
1558 static FORCEINLINE int win32munmap(void* ptr, size_t size) {
1559 MEMORY_BASIC_INFORMATION minfo;
1560 char* cptr = (char*)ptr;
1561 while (size) {
1562 if (VirtualQuery(cptr, &minfo, sizeof(minfo)) == 0)
1563 return -1;
1564 if (minfo.BaseAddress != cptr || minfo.AllocationBase != cptr ||
1565 minfo.State != MEM_COMMIT || minfo.RegionSize > size)
1566 return -1;
1567 if (VirtualFree(cptr, 0, MEM_RELEASE) == 0)
1568 return -1;
1569 cptr += minfo.RegionSize;
1570 size -= minfo.RegionSize;
1572 return 0;
1575 #define MMAP_DEFAULT(s) win32mmap(s)
1576 #define MUNMAP_DEFAULT(a, s) win32munmap((a), (s))
1577 #define DIRECT_MMAP_DEFAULT(s) win32direct_mmap(s)
1578 #endif /* WIN32 */
1579 #endif /* HAVE_MMAP */
1581 #if HAVE_MREMAP
1582 #ifndef WIN32
1583 #define MREMAP_DEFAULT(addr, osz, nsz, mv) mremap((addr), (osz), (nsz), (mv))
1584 #endif /* WIN32 */
1585 #endif /* HAVE_MREMAP */
1589 * Define CALL_MORECORE
1591 #if HAVE_MORECORE
1592 #ifdef MORECORE
1593 #define CALL_MORECORE(S) MORECORE(S)
1594 #else /* MORECORE */
1595 #define CALL_MORECORE(S) MORECORE_DEFAULT(S)
1596 #endif /* MORECORE */
1597 #else /* HAVE_MORECORE */
1598 #define CALL_MORECORE(S) MFAIL
1599 #endif /* HAVE_MORECORE */
1602 * Define CALL_MMAP/CALL_MUNMAP/CALL_DIRECT_MMAP
1604 #if HAVE_MMAP
1605 #define IS_MMAPPED_BIT (SIZE_T_ONE)
1606 #define USE_MMAP_BIT (SIZE_T_ONE)
1608 #ifdef MMAP
1609 #define CALL_MMAP(s) MMAP(s)
1610 #else /* MMAP */
1611 #define CALL_MMAP(s) MMAP_DEFAULT(s)
1612 #endif /* MMAP */
1613 #ifdef MUNMAP
1614 #define CALL_MUNMAP(a, s) MUNMAP((a), (s))
1615 #else /* MUNMAP */
1616 #define CALL_MUNMAP(a, s) MUNMAP_DEFAULT((a), (s))
1617 #endif /* MUNMAP */
1618 #ifdef DIRECT_MMAP
1619 #define CALL_DIRECT_MMAP(s) DIRECT_MMAP(s)
1620 #else /* DIRECT_MMAP */
1621 #define CALL_DIRECT_MMAP(s) DIRECT_MMAP_DEFAULT(s)
1622 #endif /* DIRECT_MMAP */
1623 #else /* HAVE_MMAP */
1624 #define IS_MMAPPED_BIT (SIZE_T_ZERO)
1625 #define USE_MMAP_BIT (SIZE_T_ZERO)
1627 #define MMAP(s) MFAIL
1628 #define MUNMAP(a, s) (-1)
1629 #define DIRECT_MMAP(s) MFAIL
1630 #define CALL_DIRECT_MMAP(s) DIRECT_MMAP(s)
1631 #define CALL_MMAP(s) MMAP(s)
1632 #define CALL_MUNMAP(a, s) MUNMAP((a), (s))
1633 #endif /* HAVE_MMAP */
1636 * Define CALL_MREMAP
1638 #if HAVE_MMAP && HAVE_MREMAP
1639 #ifdef MREMAP
1640 #define CALL_MREMAP(addr, osz, nsz, mv) MREMAP((addr), (osz), (nsz), (mv))
1641 #else /* MREMAP */
1642 #define CALL_MREMAP(addr, osz, nsz, mv) MREMAP_DEFAULT((addr), (osz), (nsz), (mv))
1643 #endif /* MREMAP */
1644 #else /* HAVE_MMAP && HAVE_MREMAP */
1645 #define CALL_MREMAP(addr, osz, nsz, mv) MFAIL
1646 #endif /* HAVE_MMAP && HAVE_MREMAP */
1648 /* mstate bit set if continguous morecore disabled or failed */
1649 #define USE_NONCONTIGUOUS_BIT (4U)
1651 /* segment bit set in create_mspace_with_base */
1652 #define EXTERN_BIT (8U)
1655 /* --------------------------- Lock preliminaries ------------------------ */
1658 When locks are defined, there is one global lock, plus
1659 one per-mspace lock.
1661 The global lock_ensures that mparams.magic and other unique
1662 mparams values are initialized only once. It also protects
1663 sequences of calls to MORECORE. In many cases sys_alloc requires
1664 two calls, that should not be interleaved with calls by other
1665 threads. This does not protect against direct calls to MORECORE
1666 by other threads not using this lock, so there is still code to
1667 cope the best we can on interference.
1669 Per-mspace locks surround calls to malloc, free, etc. To enable use
1670 in layered extensions, per-mspace locks are reentrant.
1672 Because lock-protected regions generally have bounded times, it is
1673 OK to use the supplied simple spinlocks in the custom versions for
1674 x86.
1676 If USE_LOCKS is > 1, the definitions of lock routines here are
1677 bypassed, in which case you will need to define at least
1678 INITIAL_LOCK, ACQUIRE_LOCK, RELEASE_LOCK and possibly TRY_LOCK
1679 (which is not used in this malloc, but commonly needed in
1680 extensions.)
1683 #if USE_LOCKS == 1
1685 #if USE_SPIN_LOCKS
1686 #ifndef WIN32
1688 /* Custom pthread-style spin locks on x86 and x64 for gcc */
1689 struct pthread_mlock_t {
1690 volatile unsigned int l;
1691 volatile unsigned int c;
1692 volatile pthread_t threadid;
1694 #define MLOCK_T struct pthread_mlock_t
1695 #define CURRENT_THREAD pthread_self()
1696 #define INITIAL_LOCK(sl) (memset(sl, 0, sizeof(MLOCK_T)), 0)
1697 #define ACQUIRE_LOCK(sl) pthread_acquire_lock(sl)
1698 #define RELEASE_LOCK(sl) pthread_release_lock(sl)
1699 #define TRY_LOCK(sl) pthread_try_lock(sl)
1700 #define SPINS_PER_YIELD 63
1702 static MLOCK_T malloc_global_mutex = { 0, 0, 0};
1704 static FORCEINLINE int pthread_acquire_lock (MLOCK_T *sl) {
1705 int spins = 0;
1706 volatile unsigned int* lp = &sl->l;
1707 for (;;) {
1708 if (*lp != 0) {
1709 if (sl->threadid == CURRENT_THREAD) {
1710 ++sl->c;
1711 return 0;
1714 else {
1715 /* place args to cmpxchgl in locals to evade oddities in some gccs */
1716 int cmp = 0;
1717 int val = 1;
1718 int ret;
1719 __asm__ __volatile__ ("lock; cmpxchgl %1, %2"
1720 : "=a" (ret)
1721 : "r" (val), "m" (*(lp)), "0"(cmp)
1722 : "memory", "cc");
1723 if (!ret) {
1724 assert(!sl->threadid);
1725 sl->c = 1;
1726 sl->threadid = CURRENT_THREAD;
1727 return 0;
1729 if ((++spins & SPINS_PER_YIELD) == 0) {
1730 #if defined (__SVR4) && defined (__sun) /* solaris */
1731 thr_yield();
1732 #else
1733 #if defined(__linux__) || defined(__FreeBSD__) || defined(__APPLE__)
1734 sched_yield();
1735 #else /* no-op yield on unknown systems */
1737 #endif /* __linux__ || __FreeBSD__ || __APPLE__ */
1738 #endif /* solaris */
1744 static FORCEINLINE void pthread_release_lock (MLOCK_T *sl) {
1745 assert(sl->l != 0);
1746 assert(sl->threadid == CURRENT_THREAD);
1747 if (--sl->c == 0) {
1748 sl->threadid = 0;
1749 volatile unsigned int* lp = &sl->l;
1750 int prev = 0;
1751 int ret;
1752 __asm__ __volatile__ ("lock; xchgl %0, %1"
1753 : "=r" (ret)
1754 : "m" (*(lp)), "0"(prev)
1755 : "memory");
1759 static FORCEINLINE int pthread_try_lock (MLOCK_T *sl) {
1760 volatile unsigned int* lp = &sl->l;
1761 if (*lp != 0) {
1762 if (sl->threadid == CURRENT_THREAD) {
1763 ++sl->c;
1764 return 1;
1767 else {
1768 int cmp = 0;
1769 int val = 1;
1770 int ret;
1771 __asm__ __volatile__ ("lock; cmpxchgl %1, %2"
1772 : "=a" (ret)
1773 : "r" (val), "m" (*(lp)), "0"(cmp)
1774 : "memory", "cc");
1775 if (!ret) {
1776 assert(!sl->threadid);
1777 sl->c = 1;
1778 sl->threadid = CURRENT_THREAD;
1779 return 1;
1782 return 0;
1786 #else /* WIN32 */
1787 /* Custom win32-style spin locks on x86 and x64 for MSC */
1788 struct win32_mlock_t
1790 volatile long l;
1791 volatile unsigned int c;
1792 volatile long threadid;
1795 #define MLOCK_T struct win32_mlock_t
1796 #define CURRENT_THREAD win32_getcurrentthreadid()
1797 #define INITIAL_LOCK(sl) (memset(sl, 0, sizeof(MLOCK_T)), 0)
1798 #define ACQUIRE_LOCK(sl) win32_acquire_lock(sl)
1799 #define RELEASE_LOCK(sl) win32_release_lock(sl)
1800 #define TRY_LOCK(sl) win32_try_lock(sl)
1801 #define SPINS_PER_YIELD 63
1803 static MLOCK_T malloc_global_mutex = { 0, 0, 0};
1805 static FORCEINLINE long win32_getcurrentthreadid() {
1806 #ifdef _MSC_VER
1807 #if defined(_M_IX86)
1808 long *threadstruct=(long *)__readfsdword(0x18);
1809 long threadid=threadstruct[0x24/sizeof(long)];
1810 return threadid;
1811 #elif defined(_M_X64)
1812 /* todo */
1813 return GetCurrentThreadId();
1814 #else
1815 return GetCurrentThreadId();
1816 #endif
1817 #else
1818 return GetCurrentThreadId();
1819 #endif
1822 static FORCEINLINE int win32_acquire_lock (MLOCK_T *sl) {
1823 int spins = 0;
1824 for (;;) {
1825 if (sl->l != 0) {
1826 if (sl->threadid == CURRENT_THREAD) {
1827 ++sl->c;
1828 return 0;
1831 else {
1832 if (!interlockedexchange(&sl->l, 1)) {
1833 assert(!sl->threadid);
1834 sl->c=CURRENT_THREAD;
1835 sl->threadid = CURRENT_THREAD;
1836 sl->c = 1;
1837 return 0;
1840 if ((++spins & SPINS_PER_YIELD) == 0)
1841 SleepEx(0, FALSE);
1845 static FORCEINLINE void win32_release_lock (MLOCK_T *sl) {
1846 assert(sl->threadid == CURRENT_THREAD);
1847 assert(sl->l != 0);
1848 if (--sl->c == 0) {
1849 sl->threadid = 0;
1850 interlockedexchange (&sl->l, 0);
1854 static FORCEINLINE int win32_try_lock (MLOCK_T *sl) {
1855 if(sl->l != 0) {
1856 if (sl->threadid == CURRENT_THREAD) {
1857 ++sl->c;
1858 return 1;
1861 else {
1862 if (!interlockedexchange(&sl->l, 1)){
1863 assert(!sl->threadid);
1864 sl->threadid = CURRENT_THREAD;
1865 sl->c = 1;
1866 return 1;
1869 return 0;
1872 #endif /* WIN32 */
1873 #else /* USE_SPIN_LOCKS */
1875 #ifndef WIN32
1876 /* pthreads-based locks */
1878 #define MLOCK_T pthread_mutex_t
1879 #define CURRENT_THREAD pthread_self()
1880 #define INITIAL_LOCK(sl) pthread_init_lock(sl)
1881 #define ACQUIRE_LOCK(sl) pthread_mutex_lock(sl)
1882 #define RELEASE_LOCK(sl) pthread_mutex_unlock(sl)
1883 #define TRY_LOCK(sl) (!pthread_mutex_trylock(sl))
1885 static MLOCK_T malloc_global_mutex = PTHREAD_MUTEX_INITIALIZER;
1887 /* Cope with old-style linux recursive lock initialization by adding */
1888 /* skipped internal declaration from pthread.h */
1889 #ifdef linux
1890 #ifndef PTHREAD_MUTEX_RECURSIVE
1891 extern int pthread_mutexattr_setkind_np __P ((pthread_mutexattr_t *__attr,
1892 int __kind));
1893 #define PTHREAD_MUTEX_RECURSIVE PTHREAD_MUTEX_RECURSIVE_NP
1894 #define pthread_mutexattr_settype(x,y) pthread_mutexattr_setkind_np(x,y)
1895 #endif
1896 #endif
1898 static int pthread_init_lock (MLOCK_T *sl) {
1899 pthread_mutexattr_t attr;
1900 if (pthread_mutexattr_init(&attr)) return 1;
1901 if (pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_RECURSIVE)) return 1;
1902 if (pthread_mutex_init(sl, &attr)) return 1;
1903 if (pthread_mutexattr_destroy(&attr)) return 1;
1904 return 0;
1907 #else /* WIN32 */
1908 /* Win32 critical sections */
1909 #define MLOCK_T CRITICAL_SECTION
1910 #define CURRENT_THREAD GetCurrentThreadId()
1911 #define INITIAL_LOCK(s) (!InitializeCriticalSectionAndSpinCount((s), 0x80000000|4000))
1912 #define ACQUIRE_LOCK(s) (EnterCriticalSection(s), 0)
1913 #define RELEASE_LOCK(s) LeaveCriticalSection(s)
1914 #define TRY_LOCK(s) TryEnterCriticalSection(s)
1915 #define NEED_GLOBAL_LOCK_INIT
1917 static MLOCK_T malloc_global_mutex;
1918 static volatile long malloc_global_mutex_status;
1920 /* Use spin loop to initialize global lock */
1921 static void init_malloc_global_mutex() {
1922 for (;;) {
1923 long stat = malloc_global_mutex_status;
1924 if (stat > 0)
1925 return;
1926 /* transition to < 0 while initializing, then to > 0) */
1927 if (stat == 0 &&
1928 interlockedcompareexchange(&malloc_global_mutex_status, -1, 0) == 0) {
1929 InitializeCriticalSection(&malloc_global_mutex);
1930 interlockedexchange(&malloc_global_mutex_status,1);
1931 return;
1933 SleepEx(0, FALSE);
1937 #endif /* WIN32 */
1938 #endif /* USE_SPIN_LOCKS */
1939 #endif /* USE_LOCKS == 1 */
1941 /* ----------------------- User-defined locks ------------------------ */
1943 #if USE_LOCKS > 1
1944 /* Define your own lock implementation here */
1945 /* #define INITIAL_LOCK(sl) ... */
1946 /* #define ACQUIRE_LOCK(sl) ... */
1947 /* #define RELEASE_LOCK(sl) ... */
1948 /* #define TRY_LOCK(sl) ... */
1949 /* static MLOCK_T malloc_global_mutex = ... */
1950 #endif /* USE_LOCKS > 1 */
1952 /* ----------------------- Lock-based state ------------------------ */
1954 #if USE_LOCKS
1955 #define USE_LOCK_BIT (2U)
1956 #else /* USE_LOCKS */
1957 #define USE_LOCK_BIT (0U)
1958 #define INITIAL_LOCK(l)
1959 #endif /* USE_LOCKS */
1961 #if USE_LOCKS
1962 #define ACQUIRE_MALLOC_GLOBAL_LOCK() ACQUIRE_LOCK(&malloc_global_mutex);
1963 #define RELEASE_MALLOC_GLOBAL_LOCK() RELEASE_LOCK(&malloc_global_mutex);
1964 #else /* USE_LOCKS */
1965 #define ACQUIRE_MALLOC_GLOBAL_LOCK()
1966 #define RELEASE_MALLOC_GLOBAL_LOCK()
1967 #endif /* USE_LOCKS */
1970 /* ----------------------- Chunk representations ------------------------ */
1973 (The following includes lightly edited explanations by Colin Plumb.)
1975 The malloc_chunk declaration below is misleading (but accurate and
1976 necessary). It declares a "view" into memory allowing access to
1977 necessary fields at known offsets from a given base.
1979 Chunks of memory are maintained using a `boundary tag' method as
1980 originally described by Knuth. (See the paper by Paul Wilson
1981 ftp://ftp.cs.utexas.edu/pub/garbage/allocsrv.ps for a survey of such
1982 techniques.) Sizes of free chunks are stored both in the front of
1983 each chunk and at the end. This makes consolidating fragmented
1984 chunks into bigger chunks fast. The head fields also hold bits
1985 representing whether chunks are free or in use.
1987 Here are some pictures to make it clearer. They are "exploded" to
1988 show that the state of a chunk can be thought of as extending from
1989 the high 31 bits of the head field of its header through the
1990 prev_foot and PINUSE_BIT bit of the following chunk header.
1992 A chunk that's in use looks like:
1994 chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1995 | Size of previous chunk (if P = 0) |
1996 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1997 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |P|
1998 | Size of this chunk 1| +-+
1999 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2001 +- -+
2003 +- -+
2005 +- size - sizeof(size_t) available payload bytes -+
2007 chunk-> +- -+
2009 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2010 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |1|
2011 | Size of next chunk (may or may not be in use) | +-+
2012 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2014 And if it's free, it looks like this:
2016 chunk-> +- -+
2017 | User payload (must be in use, or we would have merged!) |
2018 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2019 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |P|
2020 | Size of this chunk 0| +-+
2021 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2022 | Next pointer |
2023 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2024 | Prev pointer |
2025 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2027 +- size - sizeof(struct chunk) unused bytes -+
2029 chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2030 | Size of this chunk |
2031 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2032 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |0|
2033 | Size of next chunk (must be in use, or we would have merged)| +-+
2034 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2036 +- User payload -+
2038 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2041 Note that since we always merge adjacent free chunks, the chunks
2042 adjacent to a free chunk must be in use.
2044 Given a pointer to a chunk (which can be derived trivially from the
2045 payload pointer) we can, in O(1) time, find out whether the adjacent
2046 chunks are free, and if so, unlink them from the lists that they
2047 are on and merge them with the current chunk.
2049 Chunks always begin on even word boundaries, so the mem portion
2050 (which is returned to the user) is also on an even word boundary, and
2051 thus at least double-word aligned.
2053 The P (PINUSE_BIT) bit, stored in the unused low-order bit of the
2054 chunk size (which is always a multiple of two words), is an in-use
2055 bit for the *previous* chunk. If that bit is *clear*, then the
2056 word before the current chunk size contains the previous chunk
2057 size, and can be used to find the front of the previous chunk.
2058 The very first chunk allocated always has this bit set, preventing
2059 access to non-existent (or non-owned) memory. If pinuse is set for
2060 any given chunk, then you CANNOT determine the size of the
2061 previous chunk, and might even get a memory addressing fault when
2062 trying to do so.
2064 The C (CINUSE_BIT) bit, stored in the unused second-lowest bit of
2065 the chunk size redundantly records whether the current chunk is
2066 inuse. This redundancy enables usage checks within free and realloc,
2067 and reduces indirection when freeing and consolidating chunks.
2069 Each freshly allocated chunk must have both cinuse and pinuse set.
2070 That is, each allocated chunk borders either a previously allocated
2071 and still in-use chunk, or the base of its memory arena. This is
2072 ensured by making all allocations from the `lowest' part of any
2073 found chunk. Further, no free chunk physically borders another one,
2074 so each free chunk is known to be preceded and followed by either
2075 inuse chunks or the ends of memory.
2077 Note that the `foot' of the current chunk is actually represented
2078 as the prev_foot of the NEXT chunk. This makes it easier to
2079 deal with alignments etc but can be very confusing when trying
2080 to extend or adapt this code.
2082 The exceptions to all this are
2084 1. The special chunk `top' is the top-most available chunk (i.e.,
2085 the one bordering the end of available memory). It is treated
2086 specially. Top is never included in any bin, is used only if
2087 no other chunk is available, and is released back to the
2088 system if it is very large (see M_TRIM_THRESHOLD). In effect,
2089 the top chunk is treated as larger (and thus less well
2090 fitting) than any other available chunk. The top chunk
2091 doesn't update its trailing size field since there is no next
2092 contiguous chunk that would have to index off it. However,
2093 space is still allocated for it (TOP_FOOT_SIZE) to enable
2094 separation or merging when space is extended.
2096 3. Chunks allocated via mmap, which have the lowest-order bit
2097 (IS_MMAPPED_BIT) set in their prev_foot fields, and do not set
2098 PINUSE_BIT in their head fields. Because they are allocated
2099 one-by-one, each must carry its own prev_foot field, which is
2100 also used to hold the offset this chunk has within its mmapped
2101 region, which is needed to preserve alignment. Each mmapped
2102 chunk is trailed by the first two fields of a fake next-chunk
2103 for sake of usage checks.
2107 struct malloc_chunk {
2108 size_t prev_foot; /* Size of previous chunk (if free). */
2109 size_t head; /* Size and inuse bits. */
2110 struct malloc_chunk* fd; /* double links -- used only if free. */
2111 struct malloc_chunk* bk;
2114 typedef struct malloc_chunk mchunk;
2115 typedef struct malloc_chunk* mchunkptr;
2116 typedef struct malloc_chunk* sbinptr; /* The type of bins of chunks */
2117 typedef unsigned int bindex_t; /* Described below */
2118 typedef unsigned int binmap_t; /* Described below */
2119 typedef unsigned int flag_t; /* The type of various bit flag sets */
2121 /* ------------------- Chunks sizes and alignments ----------------------- */
2123 #define MCHUNK_SIZE (sizeof(mchunk))
2125 #if FOOTERS
2126 #define CHUNK_OVERHEAD (TWO_SIZE_T_SIZES)
2127 #else /* FOOTERS */
2128 #define CHUNK_OVERHEAD (SIZE_T_SIZE)
2129 #endif /* FOOTERS */
2131 /* MMapped chunks need a second word of overhead ... */
2132 #define MMAP_CHUNK_OVERHEAD (TWO_SIZE_T_SIZES)
2133 /* ... and additional padding for fake next-chunk at foot */
2134 #define MMAP_FOOT_PAD (FOUR_SIZE_T_SIZES)
2136 /* The smallest size we can malloc is an aligned minimal chunk */
2137 #define MIN_CHUNK_SIZE\
2138 ((MCHUNK_SIZE + CHUNK_ALIGN_MASK) & ~CHUNK_ALIGN_MASK)
2140 /* conversion from malloc headers to user pointers, and back */
2141 #define chunk2mem(p) ((void*)((char*)(p) + TWO_SIZE_T_SIZES))
2142 #define mem2chunk(mem) ((mchunkptr)((char*)(mem) - TWO_SIZE_T_SIZES))
2143 /* chunk associated with aligned address A */
2144 #define align_as_chunk(A) (mchunkptr)((A) + align_offset(chunk2mem(A)))
2146 /* Bounds on request (not chunk) sizes. */
2147 #define MAX_REQUEST ((-MIN_CHUNK_SIZE) << 2)
2148 #define MIN_REQUEST (MIN_CHUNK_SIZE - CHUNK_OVERHEAD - SIZE_T_ONE)
2150 /* pad request bytes into a usable size */
2151 #define pad_request(req) \
2152 (((req) + CHUNK_OVERHEAD + CHUNK_ALIGN_MASK) & ~CHUNK_ALIGN_MASK)
2154 /* pad request, checking for minimum (but not maximum) */
2155 #define request2size(req) \
2156 (((req) < MIN_REQUEST)? MIN_CHUNK_SIZE : pad_request(req))
2159 /* ------------------ Operations on head and foot fields ----------------- */
2162 The head field of a chunk is or'ed with PINUSE_BIT when previous
2163 adjacent chunk in use, and or'ed with CINUSE_BIT if this chunk is in
2164 use. If the chunk was obtained with mmap, the prev_foot field has
2165 IS_MMAPPED_BIT set, otherwise holding the offset of the base of the
2166 mmapped region to the base of the chunk.
2168 FLAG4_BIT is not used by this malloc, but might be useful in extensions.
2171 #define PINUSE_BIT (SIZE_T_ONE)
2172 #define CINUSE_BIT (SIZE_T_TWO)
2173 #define FLAG4_BIT (SIZE_T_FOUR)
2174 #define INUSE_BITS (PINUSE_BIT|CINUSE_BIT)
2175 #define FLAG_BITS (PINUSE_BIT|CINUSE_BIT|FLAG4_BIT)
2177 /* Head value for fenceposts */
2178 #define FENCEPOST_HEAD (INUSE_BITS|SIZE_T_SIZE)
2180 /* extraction of fields from head words */
2181 #define cinuse(p) ((p)->head & CINUSE_BIT)
2182 #define pinuse(p) ((p)->head & PINUSE_BIT)
2183 #define chunksize(p) ((p)->head & ~(FLAG_BITS))
2185 #define clear_pinuse(p) ((p)->head &= ~PINUSE_BIT)
2186 #define clear_cinuse(p) ((p)->head &= ~CINUSE_BIT)
2188 /* Treat space at ptr +/- offset as a chunk */
2189 #define chunk_plus_offset(p, s) ((mchunkptr)(((char*)(p)) + (s)))
2190 #define chunk_minus_offset(p, s) ((mchunkptr)(((char*)(p)) - (s)))
2192 /* Ptr to next or previous physical malloc_chunk. */
2193 #define next_chunk(p) ((mchunkptr)( ((char*)(p)) + ((p)->head & ~FLAG_BITS)))
2194 #define prev_chunk(p) ((mchunkptr)( ((char*)(p)) - ((p)->prev_foot) ))
2196 /* extract next chunk's pinuse bit */
2197 #define next_pinuse(p) ((next_chunk(p)->head) & PINUSE_BIT)
2199 /* Get/set size at footer */
2200 #define get_foot(p, s) (((mchunkptr)((char*)(p) + (s)))->prev_foot)
2201 #define set_foot(p, s) (((mchunkptr)((char*)(p) + (s)))->prev_foot = (s))
2203 /* Set size, pinuse bit, and foot */
2204 #define set_size_and_pinuse_of_free_chunk(p, s)\
2205 ((p)->head = (s|PINUSE_BIT), set_foot(p, s))
2207 /* Set size, pinuse bit, foot, and clear next pinuse */
2208 #define set_free_with_pinuse(p, s, n)\
2209 (clear_pinuse(n), set_size_and_pinuse_of_free_chunk(p, s))
2211 #define is_mmapped(p)\
2212 (!((p)->head & PINUSE_BIT) && ((p)->prev_foot & IS_MMAPPED_BIT))
2214 /* Get the internal overhead associated with chunk p */
2215 #define overhead_for(p)\
2216 (is_mmapped(p)? MMAP_CHUNK_OVERHEAD : CHUNK_OVERHEAD)
2218 /* Return true if malloced space is not necessarily cleared */
2219 #if MMAP_CLEARS
2220 #define calloc_must_clear(p) (!is_mmapped(p))
2221 #else /* MMAP_CLEARS */
2222 #define calloc_must_clear(p) (1)
2223 #endif /* MMAP_CLEARS */
2225 /* ---------------------- Overlaid data structures ----------------------- */
2228 When chunks are not in use, they are treated as nodes of either
2229 lists or trees.
2231 "Small" chunks are stored in circular doubly-linked lists, and look
2232 like this:
2234 chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2235 | Size of previous chunk |
2236 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2237 `head:' | Size of chunk, in bytes |P|
2238 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2239 | Forward pointer to next chunk in list |
2240 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2241 | Back pointer to previous chunk in list |
2242 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2243 | Unused space (may be 0 bytes long) .
2246 nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2247 `foot:' | Size of chunk, in bytes |
2248 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2250 Larger chunks are kept in a form of bitwise digital trees (aka
2251 tries) keyed on chunksizes. Because malloc_tree_chunks are only for
2252 free chunks greater than 256 bytes, their size doesn't impose any
2253 constraints on user chunk sizes. Each node looks like:
2255 chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2256 | Size of previous chunk |
2257 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2258 `head:' | Size of chunk, in bytes |P|
2259 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2260 | Forward pointer to next chunk of same size |
2261 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2262 | Back pointer to previous chunk of same size |
2263 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2264 | Pointer to left child (child[0]) |
2265 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2266 | Pointer to right child (child[1]) |
2267 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2268 | Pointer to parent |
2269 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2270 | bin index of this chunk |
2271 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2272 | Unused space .
2274 nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2275 `foot:' | Size of chunk, in bytes |
2276 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2278 Each tree holding treenodes is a tree of unique chunk sizes. Chunks
2279 of the same size are arranged in a circularly-linked list, with only
2280 the oldest chunk (the next to be used, in our FIFO ordering)
2281 actually in the tree. (Tree members are distinguished by a non-null
2282 parent pointer.) If a chunk with the same size as an existing node
2283 is inserted, it is linked off the existing node using pointers that
2284 work in the same way as fd/bk pointers of small chunks.
2286 Each tree contains a power of 2 sized range of chunk sizes (the
2287 smallest is 0x100 <= x < 0x180), which is divided in half at each
2288 tree level, with the chunks in the smaller half of the range (0x100
2289 <= x < 0x140 for the top nose) in the left subtree and the larger
2290 half (0x140 <= x < 0x180) in the right subtree. This is, of course,
2291 done by inspecting individual bits.
2293 Using these rules, each node's left subtree contains all smaller
2294 sizes than its right subtree. However, the node at the root of each
2295 subtree has no particular ordering relationship to either. (The
2296 dividing line between the subtree sizes is based on trie relation.)
2297 If we remove the last chunk of a given size from the interior of the
2298 tree, we need to replace it with a leaf node. The tree ordering
2299 rules permit a node to be replaced by any leaf below it.
2301 The smallest chunk in a tree (a common operation in a best-fit
2302 allocator) can be found by walking a path to the leftmost leaf in
2303 the tree. Unlike a usual binary tree, where we follow left child
2304 pointers until we reach a null, here we follow the right child
2305 pointer any time the left one is null, until we reach a leaf with
2306 both child pointers null. The smallest chunk in the tree will be
2307 somewhere along that path.
2309 The worst case number of steps to add, find, or remove a node is
2310 bounded by the number of bits differentiating chunks within
2311 bins. Under current bin calculations, this ranges from 6 up to 21
2312 (for 32 bit sizes) or up to 53 (for 64 bit sizes). The typical case
2313 is of course much better.
2316 struct malloc_tree_chunk {
2317 /* The first four fields must be compatible with malloc_chunk */
2318 size_t prev_foot;
2319 size_t head;
2320 struct malloc_tree_chunk* fd;
2321 struct malloc_tree_chunk* bk;
2323 struct malloc_tree_chunk* child[2];
2324 struct malloc_tree_chunk* parent;
2325 bindex_t index;
2328 typedef struct malloc_tree_chunk tchunk;
2329 typedef struct malloc_tree_chunk* tchunkptr;
2330 typedef struct malloc_tree_chunk* tbinptr; /* The type of bins of trees */
2332 /* A little helper macro for trees */
2333 #define leftmost_child(t) ((t)->child[0] != 0? (t)->child[0] : (t)->child[1])
2335 /* ----------------------------- Segments -------------------------------- */
2338 Each malloc space may include non-contiguous segments, held in a
2339 list headed by an embedded malloc_segment record representing the
2340 top-most space. Segments also include flags holding properties of
2341 the space. Large chunks that are directly allocated by mmap are not
2342 included in this list. They are instead independently created and
2343 destroyed without otherwise keeping track of them.
2345 Segment management mainly comes into play for spaces allocated by
2346 MMAP. Any call to MMAP might or might not return memory that is
2347 adjacent to an existing segment. MORECORE normally contiguously
2348 extends the current space, so this space is almost always adjacent,
2349 which is simpler and faster to deal with. (This is why MORECORE is
2350 used preferentially to MMAP when both are available -- see
2351 sys_alloc.) When allocating using MMAP, we don't use any of the
2352 hinting mechanisms (inconsistently) supported in various
2353 implementations of unix mmap, or distinguish reserving from
2354 committing memory. Instead, we just ask for space, and exploit
2355 contiguity when we get it. It is probably possible to do
2356 better than this on some systems, but no general scheme seems
2357 to be significantly better.
2359 Management entails a simpler variant of the consolidation scheme
2360 used for chunks to reduce fragmentation -- new adjacent memory is
2361 normally prepended or appended to an existing segment. However,
2362 there are limitations compared to chunk consolidation that mostly
2363 reflect the fact that segment processing is relatively infrequent
2364 (occurring only when getting memory from system) and that we
2365 don't expect to have huge numbers of segments:
2367 * Segments are not indexed, so traversal requires linear scans. (It
2368 would be possible to index these, but is not worth the extra
2369 overhead and complexity for most programs on most platforms.)
2370 * New segments are only appended to old ones when holding top-most
2371 memory; if they cannot be prepended to others, they are held in
2372 different segments.
2374 Except for the top-most segment of an mstate, each segment record
2375 is kept at the tail of its segment. Segments are added by pushing
2376 segment records onto the list headed by &mstate.seg for the
2377 containing mstate.
2379 Segment flags control allocation/merge/deallocation policies:
2380 * If EXTERN_BIT set, then we did not allocate this segment,
2381 and so should not try to deallocate or merge with others.
2382 (This currently holds only for the initial segment passed
2383 into create_mspace_with_base.)
2384 * If IS_MMAPPED_BIT set, the segment may be merged with
2385 other surrounding mmapped segments and trimmed/de-allocated
2386 using munmap.
2387 * If neither bit is set, then the segment was obtained using
2388 MORECORE so can be merged with surrounding MORECORE'd segments
2389 and deallocated/trimmed using MORECORE with negative arguments.
2392 struct malloc_segment {
2393 char* base; /* base address */
2394 size_t size; /* allocated size */
2395 struct malloc_segment* next; /* ptr to next segment */
2396 flag_t sflags; /* mmap and extern flag */
2399 #define is_mmapped_segment(S) ((S)->sflags & IS_MMAPPED_BIT)
2400 #define is_extern_segment(S) ((S)->sflags & EXTERN_BIT)
2402 typedef struct malloc_segment msegment;
2403 typedef struct malloc_segment* msegmentptr;
2405 /* ---------------------------- malloc_state ----------------------------- */
2408 A malloc_state holds all of the bookkeeping for a space.
2409 The main fields are:
2412 The topmost chunk of the currently active segment. Its size is
2413 cached in topsize. The actual size of topmost space is
2414 topsize+TOP_FOOT_SIZE, which includes space reserved for adding
2415 fenceposts and segment records if necessary when getting more
2416 space from the system. The size at which to autotrim top is
2417 cached from mparams in trim_check, except that it is disabled if
2418 an autotrim fails.
2420 Designated victim (dv)
2421 This is the preferred chunk for servicing small requests that
2422 don't have exact fits. It is normally the chunk split off most
2423 recently to service another small request. Its size is cached in
2424 dvsize. The link fields of this chunk are not maintained since it
2425 is not kept in a bin.
2427 SmallBins
2428 An array of bin headers for free chunks. These bins hold chunks
2429 with sizes less than MIN_LARGE_SIZE bytes. Each bin contains
2430 chunks of all the same size, spaced 8 bytes apart. To simplify
2431 use in double-linked lists, each bin header acts as a malloc_chunk
2432 pointing to the real first node, if it exists (else pointing to
2433 itself). This avoids special-casing for headers. But to avoid
2434 waste, we allocate only the fd/bk pointers of bins, and then use
2435 repositioning tricks to treat these as the fields of a chunk.
2437 TreeBins
2438 Treebins are pointers to the roots of trees holding a range of
2439 sizes. There are 2 equally spaced treebins for each power of two
2440 from TREE_SHIFT to TREE_SHIFT+16. The last bin holds anything
2441 larger.
2443 Bin maps
2444 There is one bit map for small bins ("smallmap") and one for
2445 treebins ("treemap). Each bin sets its bit when non-empty, and
2446 clears the bit when empty. Bit operations are then used to avoid
2447 bin-by-bin searching -- nearly all "search" is done without ever
2448 looking at bins that won't be selected. The bit maps
2449 conservatively use 32 bits per map word, even if on 64bit system.
2450 For a good description of some of the bit-based techniques used
2451 here, see Henry S. Warren Jr's book "Hacker's Delight" (and
2452 supplement at http://hackersdelight.org/). Many of these are
2453 intended to reduce the branchiness of paths through malloc etc, as
2454 well as to reduce the number of memory locations read or written.
2456 Segments
2457 A list of segments headed by an embedded malloc_segment record
2458 representing the initial space.
2460 Address check support
2461 The least_addr field is the least address ever obtained from
2462 MORECORE or MMAP. Attempted frees and reallocs of any address less
2463 than this are trapped (unless INSECURE is defined).
2465 Magic tag
2466 A cross-check field that should always hold same value as mparams.magic.
2468 Flags
2469 Bits recording whether to use MMAP, locks, or contiguous MORECORE
2471 Statistics
2472 Each space keeps track of current and maximum system memory
2473 obtained via MORECORE or MMAP.
2475 Trim support
2476 Fields holding the amount of unused topmost memory that should trigger
2477 timming, and a counter to force periodic scanning to release unused
2478 non-topmost segments.
2480 Locking
2481 If USE_LOCKS is defined, the "mutex" lock is acquired and released
2482 around every public call using this mspace.
2484 Extension support
2485 A void* pointer and a size_t field that can be used to help implement
2486 extensions to this malloc.
2489 /* Bin types, widths and sizes */
2490 #define NSMALLBINS (32U)
2491 #define NTREEBINS (32U)
2492 #define SMALLBIN_SHIFT (3U)
2493 #define SMALLBIN_WIDTH (SIZE_T_ONE << SMALLBIN_SHIFT)
2494 #define TREEBIN_SHIFT (8U)
2495 #define MIN_LARGE_SIZE (SIZE_T_ONE << TREEBIN_SHIFT)
2496 #define MAX_SMALL_SIZE (MIN_LARGE_SIZE - SIZE_T_ONE)
2497 #define MAX_SMALL_REQUEST (MAX_SMALL_SIZE - CHUNK_ALIGN_MASK - CHUNK_OVERHEAD)
2499 struct malloc_state {
2500 binmap_t smallmap;
2501 binmap_t treemap;
2502 size_t dvsize;
2503 size_t topsize;
2504 char* least_addr;
2505 mchunkptr dv;
2506 mchunkptr top;
2507 size_t trim_check;
2508 size_t release_checks;
2509 size_t magic;
2510 mchunkptr smallbins[(NSMALLBINS+1)*2];
2511 tbinptr treebins[NTREEBINS];
2512 size_t footprint;
2513 size_t max_footprint;
2514 flag_t mflags;
2515 #if USE_LOCKS
2516 MLOCK_T mutex; /* locate lock among fields that rarely change */
2517 #endif /* USE_LOCKS */
2518 msegment seg;
2519 void* extp; /* Unused but available for extensions */
2520 size_t exts;
2523 typedef struct malloc_state* mstate;
2525 /* ------------- Global malloc_state and malloc_params ------------------- */
2528 malloc_params holds global properties, including those that can be
2529 dynamically set using mallopt. There is a single instance, mparams,
2530 initialized in init_mparams. Note that the non-zeroness of "magic"
2531 also serves as an initialization flag.
2534 struct malloc_params {
2535 volatile size_t magic;
2536 size_t page_size;
2537 size_t granularity;
2538 size_t mmap_threshold;
2539 size_t trim_threshold;
2540 flag_t default_mflags;
2543 static struct malloc_params mparams;
2545 /* Ensure mparams initialized */
2546 #define ensure_initialization() ((void)(mparams.magic != 0 || init_mparams()))
2548 #if !ONLY_MSPACES
2550 /* The global malloc_state used for all non-"mspace" calls */
2551 static struct malloc_state _gm_;
2552 #define gm (&_gm_)
2553 #define is_global(M) ((M) == &_gm_)
2555 #endif /* !ONLY_MSPACES */
2557 #define is_initialized(M) ((M)->top != 0)
2559 /* -------------------------- system alloc setup ------------------------- */
2561 /* Operations on mflags */
2563 #define use_lock(M) ((M)->mflags & USE_LOCK_BIT)
2564 #define enable_lock(M) ((M)->mflags |= USE_LOCK_BIT)
2565 #define disable_lock(M) ((M)->mflags &= ~USE_LOCK_BIT)
2567 #define use_mmap(M) ((M)->mflags & USE_MMAP_BIT)
2568 #define enable_mmap(M) ((M)->mflags |= USE_MMAP_BIT)
2569 #define disable_mmap(M) ((M)->mflags &= ~USE_MMAP_BIT)
2571 #define use_noncontiguous(M) ((M)->mflags & USE_NONCONTIGUOUS_BIT)
2572 #define disable_contiguous(M) ((M)->mflags |= USE_NONCONTIGUOUS_BIT)
2574 #define set_lock(M,L)\
2575 ((M)->mflags = (L)?\
2576 ((M)->mflags | USE_LOCK_BIT) :\
2577 ((M)->mflags & ~USE_LOCK_BIT))
2579 /* page-align a size */
2580 #define page_align(S)\
2581 (((S) + (mparams.page_size - SIZE_T_ONE)) & ~(mparams.page_size - SIZE_T_ONE))
2583 /* granularity-align a size */
2584 #define granularity_align(S)\
2585 (((S) + (mparams.granularity - SIZE_T_ONE))\
2586 & ~(mparams.granularity - SIZE_T_ONE))
2589 /* For mmap, use granularity alignment on windows, else page-align */
2590 #ifdef WIN32
2591 #define mmap_align(S) granularity_align(S)
2592 #else
2593 #define mmap_align(S) page_align(S)
2594 #endif
2596 /* For sys_alloc, enough padding to ensure can malloc request on success */
2597 #define SYS_ALLOC_PADDING (TOP_FOOT_SIZE + MALLOC_ALIGNMENT)
2599 #define is_page_aligned(S)\
2600 (((size_t)(S) & (mparams.page_size - SIZE_T_ONE)) == 0)
2601 #define is_granularity_aligned(S)\
2602 (((size_t)(S) & (mparams.granularity - SIZE_T_ONE)) == 0)
2604 /* True if segment S holds address A */
2605 #define segment_holds(S, A)\
2606 ((char*)(A) >= S->base && (char*)(A) < S->base + S->size)
2608 /* Return segment holding given address */
2609 static msegmentptr segment_holding(mstate m, char* addr) {
2610 msegmentptr sp = &m->seg;
2611 for (;;) {
2612 if (addr >= sp->base && addr < sp->base + sp->size)
2613 return sp;
2614 if ((sp = sp->next) == 0)
2615 return 0;
2619 /* Return true if segment contains a segment link */
2620 static int has_segment_link(mstate m, msegmentptr ss) {
2621 msegmentptr sp = &m->seg;
2622 for (;;) {
2623 if ((char*)sp >= ss->base && (char*)sp < ss->base + ss->size)
2624 return 1;
2625 if ((sp = sp->next) == 0)
2626 return 0;
2630 #ifndef MORECORE_CANNOT_TRIM
2631 #define should_trim(M,s) ((s) > (M)->trim_check)
2632 #else /* MORECORE_CANNOT_TRIM */
2633 #define should_trim(M,s) (0)
2634 #endif /* MORECORE_CANNOT_TRIM */
2637 TOP_FOOT_SIZE is padding at the end of a segment, including space
2638 that may be needed to place segment records and fenceposts when new
2639 noncontiguous segments are added.
2641 #define TOP_FOOT_SIZE\
2642 (align_offset(chunk2mem(0))+pad_request(sizeof(struct malloc_segment))+MIN_CHUNK_SIZE)
2645 /* ------------------------------- Hooks -------------------------------- */
2648 PREACTION should be defined to return 0 on success, and nonzero on
2649 failure. If you are not using locking, you can redefine these to do
2650 anything you like.
2653 #if USE_LOCKS
2655 #define PREACTION(M) ((use_lock(M))? ACQUIRE_LOCK(&(M)->mutex) : 0)
2656 #define POSTACTION(M) { if (use_lock(M)) RELEASE_LOCK(&(M)->mutex); }
2657 #else /* USE_LOCKS */
2659 #ifndef PREACTION
2660 #define PREACTION(M) (0)
2661 #endif /* PREACTION */
2663 #ifndef POSTACTION
2664 #define POSTACTION(M)
2665 #endif /* POSTACTION */
2667 #endif /* USE_LOCKS */
2670 CORRUPTION_ERROR_ACTION is triggered upon detected bad addresses.
2671 USAGE_ERROR_ACTION is triggered on detected bad frees and
2672 reallocs. The argument p is an address that might have triggered the
2673 fault. It is ignored by the two predefined actions, but might be
2674 useful in custom actions that try to help diagnose errors.
2677 #if PROCEED_ON_ERROR
2679 /* A count of the number of corruption errors causing resets */
2680 int malloc_corruption_error_count;
2682 /* default corruption action */
2683 static void reset_on_error(mstate m);
2685 #define CORRUPTION_ERROR_ACTION(m) reset_on_error(m)
2686 #define USAGE_ERROR_ACTION(m, p)
2688 #else /* PROCEED_ON_ERROR */
2690 #ifndef CORRUPTION_ERROR_ACTION
2691 #define CORRUPTION_ERROR_ACTION(m) ABORT
2692 #endif /* CORRUPTION_ERROR_ACTION */
2694 #ifndef USAGE_ERROR_ACTION
2695 #define USAGE_ERROR_ACTION(m,p) ABORT
2696 #endif /* USAGE_ERROR_ACTION */
2698 #endif /* PROCEED_ON_ERROR */
2700 /* -------------------------- Debugging setup ---------------------------- */
2702 #if ! DEBUG
2704 #define check_free_chunk(M,P)
2705 #define check_inuse_chunk(M,P)
2706 #define check_malloced_chunk(M,P,N)
2707 #define check_mmapped_chunk(M,P)
2708 #define check_malloc_state(M)
2709 #define check_top_chunk(M,P)
2711 #else /* DEBUG */
2712 #define check_free_chunk(M,P) do_check_free_chunk(M,P)
2713 #define check_inuse_chunk(M,P) do_check_inuse_chunk(M,P)
2714 #define check_top_chunk(M,P) do_check_top_chunk(M,P)
2715 #define check_malloced_chunk(M,P,N) do_check_malloced_chunk(M,P,N)
2716 #define check_mmapped_chunk(M,P) do_check_mmapped_chunk(M,P)
2717 #define check_malloc_state(M) do_check_malloc_state(M)
2719 static void do_check_any_chunk(mstate m, mchunkptr p);
2720 static void do_check_top_chunk(mstate m, mchunkptr p);
2721 static void do_check_mmapped_chunk(mstate m, mchunkptr p);
2722 static void do_check_inuse_chunk(mstate m, mchunkptr p);
2723 static void do_check_free_chunk(mstate m, mchunkptr p);
2724 static void do_check_malloced_chunk(mstate m, void* mem, size_t s);
2725 static void do_check_tree(mstate m, tchunkptr t);
2726 static void do_check_treebin(mstate m, bindex_t i);
2727 static void do_check_smallbin(mstate m, bindex_t i);
2728 static void do_check_malloc_state(mstate m);
2729 static int bin_find(mstate m, mchunkptr x);
2730 static size_t traverse_and_check(mstate m);
2731 #endif /* DEBUG */
2733 /* ---------------------------- Indexing Bins ---------------------------- */
2735 #define is_small(s) (((s) >> SMALLBIN_SHIFT) < NSMALLBINS)
2736 #define small_index(s) ((s) >> SMALLBIN_SHIFT)
2737 #define small_index2size(i) ((i) << SMALLBIN_SHIFT)
2738 #define MIN_SMALL_INDEX (small_index(MIN_CHUNK_SIZE))
2740 /* addressing by index. See above about smallbin repositioning */
2741 #define smallbin_at(M, i) ((sbinptr)((char*)&((M)->smallbins[(i)<<1])))
2742 #define treebin_at(M,i) (&((M)->treebins[i]))
2744 /* assign tree index for size S to variable I. Use x86 asm if possible */
2745 #if defined(__GNUC__) && (defined(__i386__) || defined(__x86_64__))
2746 #define compute_tree_index(S, I)\
2748 unsigned int X = S >> TREEBIN_SHIFT;\
2749 if (X == 0)\
2750 I = 0;\
2751 else if (X > 0xFFFF)\
2752 I = NTREEBINS-1;\
2753 else {\
2754 unsigned int K;\
2755 __asm__("bsrl\t%1, %0\n\t" : "=r" (K) : "rm" (X));\
2756 I = (bindex_t)((K << 1) + ((S >> (K + (TREEBIN_SHIFT-1)) & 1)));\
2760 #elif defined (__INTEL_COMPILER)
2761 #define compute_tree_index(S, I)\
2763 size_t X = S >> TREEBIN_SHIFT;\
2764 if (X == 0)\
2765 I = 0;\
2766 else if (X > 0xFFFF)\
2767 I = NTREEBINS-1;\
2768 else {\
2769 unsigned int K = _bit_scan_reverse (X); \
2770 I = (bindex_t)((K << 1) + ((S >> (K + (TREEBIN_SHIFT-1)) & 1)));\
2774 #elif defined(_MSC_VER) && _MSC_VER>=1300
2775 #define compute_tree_index(S, I)\
2777 size_t X = S >> TREEBIN_SHIFT;\
2778 if (X == 0)\
2779 I = 0;\
2780 else if (X > 0xFFFF)\
2781 I = NTREEBINS-1;\
2782 else {\
2783 unsigned int K;\
2784 _BitScanReverse((DWORD *) &K, X);\
2785 I = (bindex_t)((K << 1) + ((S >> (K + (TREEBIN_SHIFT-1)) & 1)));\
2789 #else /* GNUC */
2790 #define compute_tree_index(S, I)\
2792 size_t X = S >> TREEBIN_SHIFT;\
2793 if (X == 0)\
2794 I = 0;\
2795 else if (X > 0xFFFF)\
2796 I = NTREEBINS-1;\
2797 else {\
2798 unsigned int Y = (unsigned int)X;\
2799 unsigned int N = ((Y - 0x100) >> 16) & 8;\
2800 unsigned int K = (((Y <<= N) - 0x1000) >> 16) & 4;\
2801 N += K;\
2802 N += K = (((Y <<= K) - 0x4000) >> 16) & 2;\
2803 K = 14 - N + ((Y <<= K) >> 15);\
2804 I = (K << 1) + ((S >> (K + (TREEBIN_SHIFT-1)) & 1));\
2807 #endif /* GNUC */
2809 /* Bit representing maximum resolved size in a treebin at i */
2810 #define bit_for_tree_index(i) \
2811 (i == NTREEBINS-1)? (SIZE_T_BITSIZE-1) : (((i) >> 1) + TREEBIN_SHIFT - 2)
2813 /* Shift placing maximum resolved bit in a treebin at i as sign bit */
2814 #define leftshift_for_tree_index(i) \
2815 ((i == NTREEBINS-1)? 0 : \
2816 ((SIZE_T_BITSIZE-SIZE_T_ONE) - (((i) >> 1) + TREEBIN_SHIFT - 2)))
2818 /* The size of the smallest chunk held in bin with index i */
2819 #define minsize_for_tree_index(i) \
2820 ((SIZE_T_ONE << (((i) >> 1) + TREEBIN_SHIFT)) | \
2821 (((size_t)((i) & SIZE_T_ONE)) << (((i) >> 1) + TREEBIN_SHIFT - 1)))
2824 /* ------------------------ Operations on bin maps ----------------------- */
2826 /* bit corresponding to given index */
2827 #define idx2bit(i) ((binmap_t)(1) << (i))
2829 /* Mark/Clear bits with given index */
2830 #define mark_smallmap(M,i) ((M)->smallmap |= idx2bit(i))
2831 #define clear_smallmap(M,i) ((M)->smallmap &= ~idx2bit(i))
2832 #define smallmap_is_marked(M,i) ((M)->smallmap & idx2bit(i))
2834 #define mark_treemap(M,i) ((M)->treemap |= idx2bit(i))
2835 #define clear_treemap(M,i) ((M)->treemap &= ~idx2bit(i))
2836 #define treemap_is_marked(M,i) ((M)->treemap & idx2bit(i))
2838 /* isolate the least set bit of a bitmap */
2839 #define least_bit(x) ((x) & -(x))
2841 /* mask with all bits to left of least bit of x on */
2842 #define left_bits(x) ((x<<1) | -(x<<1))
2844 /* mask with all bits to left of or equal to least bit of x on */
2845 #define same_or_left_bits(x) ((x) | -(x))
2847 /* index corresponding to given bit. Use x86 asm if possible */
2849 #if defined(__GNUC__) && (defined(__i386__) || defined(__x86_64__))
2850 #define compute_bit2idx(X, I)\
2852 unsigned int J;\
2853 __asm__("bsfl\t%1, %0\n\t" : "=r" (J) : "rm" (X));\
2854 I = (bindex_t)J;\
2857 #elif defined (__INTEL_COMPILER)
2858 #define compute_bit2idx(X, I)\
2860 unsigned int J;\
2861 J = _bit_scan_forward (X); \
2862 I = (bindex_t)J;\
2865 #elif defined(_MSC_VER) && _MSC_VER>=1300
2866 #define compute_bit2idx(X, I)\
2868 unsigned int J;\
2869 _BitScanForward((DWORD *) &J, X);\
2870 I = (bindex_t)J;\
2873 #elif USE_BUILTIN_FFS
2874 #define compute_bit2idx(X, I) I = ffs(X)-1
2876 #else
2877 #define compute_bit2idx(X, I)\
2879 unsigned int Y = X - 1;\
2880 unsigned int K = Y >> (16-4) & 16;\
2881 unsigned int N = K; Y >>= K;\
2882 N += K = Y >> (8-3) & 8; Y >>= K;\
2883 N += K = Y >> (4-2) & 4; Y >>= K;\
2884 N += K = Y >> (2-1) & 2; Y >>= K;\
2885 N += K = Y >> (1-0) & 1; Y >>= K;\
2886 I = (bindex_t)(N + Y);\
2888 #endif /* GNUC */
2891 /* ----------------------- Runtime Check Support ------------------------- */
2894 For security, the main invariant is that malloc/free/etc never
2895 writes to a static address other than malloc_state, unless static
2896 malloc_state itself has been corrupted, which cannot occur via
2897 malloc (because of these checks). In essence this means that we
2898 believe all pointers, sizes, maps etc held in malloc_state, but
2899 check all of those linked or offsetted from other embedded data
2900 structures. These checks are interspersed with main code in a way
2901 that tends to minimize their run-time cost.
2903 When FOOTERS is defined, in addition to range checking, we also
2904 verify footer fields of inuse chunks, which can be used guarantee
2905 that the mstate controlling malloc/free is intact. This is a
2906 streamlined version of the approach described by William Robertson
2907 et al in "Run-time Detection of Heap-based Overflows" LISA'03
2908 http://www.usenix.org/events/lisa03/tech/robertson.html The footer
2909 of an inuse chunk holds the xor of its mstate and a random seed,
2910 that is checked upon calls to free() and realloc(). This is
2911 (probablistically) unguessable from outside the program, but can be
2912 computed by any code successfully malloc'ing any chunk, so does not
2913 itself provide protection against code that has already broken
2914 security through some other means. Unlike Robertson et al, we
2915 always dynamically check addresses of all offset chunks (previous,
2916 next, etc). This turns out to be cheaper than relying on hashes.
2919 #if !INSECURE
2920 /* Check if address a is at least as high as any from MORECORE or MMAP */
2921 #define ok_address(M, a) ((char*)(a) >= (M)->least_addr)
2922 /* Check if address of next chunk n is higher than base chunk p */
2923 #define ok_next(p, n) ((char*)(p) < (char*)(n))
2924 /* Check if p has its cinuse bit on */
2925 #define ok_cinuse(p) cinuse(p)
2926 /* Check if p has its pinuse bit on */
2927 #define ok_pinuse(p) pinuse(p)
2929 #else /* !INSECURE */
2930 #define ok_address(M, a) (1)
2931 #define ok_next(b, n) (1)
2932 #define ok_cinuse(p) (1)
2933 #define ok_pinuse(p) (1)
2934 #endif /* !INSECURE */
2936 #if (FOOTERS && !INSECURE)
2937 /* Check if (alleged) mstate m has expected magic field */
2938 #define ok_magic(M) ((M)->magic == mparams.magic)
2939 #else /* (FOOTERS && !INSECURE) */
2940 #define ok_magic(M) (1)
2941 #endif /* (FOOTERS && !INSECURE) */
2944 /* In gcc, use __builtin_expect to minimize impact of checks */
2945 #if !INSECURE
2946 #if defined(__GNUC__) && __GNUC__ >= 3
2947 #define RTCHECK(e) __builtin_expect(e, 1)
2948 #else /* GNUC */
2949 #define RTCHECK(e) (e)
2950 #endif /* GNUC */
2951 #else /* !INSECURE */
2952 #define RTCHECK(e) (1)
2953 #endif /* !INSECURE */
2955 /* macros to set up inuse chunks with or without footers */
2957 #if !FOOTERS
2959 #define mark_inuse_foot(M,p,s)
2961 /* Set cinuse bit and pinuse bit of next chunk */
2962 #define set_inuse(M,p,s)\
2963 ((p)->head = (((p)->head & PINUSE_BIT)|s|CINUSE_BIT),\
2964 ((mchunkptr)(((char*)(p)) + (s)))->head |= PINUSE_BIT)
2966 /* Set cinuse and pinuse of this chunk and pinuse of next chunk */
2967 #define set_inuse_and_pinuse(M,p,s)\
2968 ((p)->head = (s|PINUSE_BIT|CINUSE_BIT),\
2969 ((mchunkptr)(((char*)(p)) + (s)))->head |= PINUSE_BIT)
2971 /* Set size, cinuse and pinuse bit of this chunk */
2972 #define set_size_and_pinuse_of_inuse_chunk(M, p, s)\
2973 ((p)->head = (s|PINUSE_BIT|CINUSE_BIT))
2975 #else /* FOOTERS */
2977 /* Set foot of inuse chunk to be xor of mstate and seed */
2978 #define mark_inuse_foot(M,p,s)\
2979 (((mchunkptr)((char*)(p) + (s)))->prev_foot = ((size_t)(M) ^ mparams.magic))
2981 #define get_mstate_for(p)\
2982 ((mstate)(((mchunkptr)((char*)(p) +\
2983 (chunksize(p))))->prev_foot ^ mparams.magic))
2985 #define set_inuse(M,p,s)\
2986 ((p)->head = (((p)->head & PINUSE_BIT)|s|CINUSE_BIT),\
2987 (((mchunkptr)(((char*)(p)) + (s)))->head |= PINUSE_BIT), \
2988 mark_inuse_foot(M,p,s))
2990 #define set_inuse_and_pinuse(M,p,s)\
2991 ((p)->head = (s|PINUSE_BIT|CINUSE_BIT),\
2992 (((mchunkptr)(((char*)(p)) + (s)))->head |= PINUSE_BIT),\
2993 mark_inuse_foot(M,p,s))
2995 #define set_size_and_pinuse_of_inuse_chunk(M, p, s)\
2996 ((p)->head = (s|PINUSE_BIT|CINUSE_BIT),\
2997 mark_inuse_foot(M, p, s))
2999 #endif /* !FOOTERS */
3001 /* ---------------------------- setting mparams -------------------------- */
3003 /* Initialize mparams */
3004 static int init_mparams(void) {
3005 #ifdef NEED_GLOBAL_LOCK_INIT
3006 if (malloc_global_mutex_status <= 0)
3007 init_malloc_global_mutex();
3008 #endif
3010 ACQUIRE_MALLOC_GLOBAL_LOCK();
3011 if (mparams.magic == 0) {
3012 size_t magic;
3013 size_t psize;
3014 size_t gsize;
3016 #ifndef WIN32
3017 psize = malloc_getpagesize;
3018 gsize = ((DEFAULT_GRANULARITY != 0)? DEFAULT_GRANULARITY : psize);
3019 #else /* WIN32 */
3021 SYSTEM_INFO system_info;
3022 GetSystemInfo(&system_info);
3023 psize = system_info.dwPageSize;
3024 gsize = ((DEFAULT_GRANULARITY != 0)?
3025 DEFAULT_GRANULARITY : system_info.dwAllocationGranularity);
3027 #endif /* WIN32 */
3029 /* Sanity-check configuration:
3030 size_t must be unsigned and as wide as pointer type.
3031 ints must be at least 4 bytes.
3032 alignment must be at least 8.
3033 Alignment, min chunk size, and page size must all be powers of 2.
3035 if ((sizeof(size_t) != sizeof(char*)) ||
3036 (MAX_SIZE_T < MIN_CHUNK_SIZE) ||
3037 (sizeof(int) < 4) ||
3038 (MALLOC_ALIGNMENT < (size_t)8U) ||
3039 ((MALLOC_ALIGNMENT & (MALLOC_ALIGNMENT-SIZE_T_ONE)) != 0) ||
3040 ((MCHUNK_SIZE & (MCHUNK_SIZE-SIZE_T_ONE)) != 0) ||
3041 ((gsize & (gsize-SIZE_T_ONE)) != 0) ||
3042 ((psize & (psize-SIZE_T_ONE)) != 0))
3043 ABORT;
3045 mparams.granularity = gsize;
3046 mparams.page_size = psize;
3047 mparams.mmap_threshold = DEFAULT_MMAP_THRESHOLD;
3048 mparams.trim_threshold = DEFAULT_TRIM_THRESHOLD;
3049 #if MORECORE_CONTIGUOUS
3050 mparams.default_mflags = USE_LOCK_BIT|USE_MMAP_BIT;
3051 #else /* MORECORE_CONTIGUOUS */
3052 mparams.default_mflags = USE_LOCK_BIT|USE_MMAP_BIT|USE_NONCONTIGUOUS_BIT;
3053 #endif /* MORECORE_CONTIGUOUS */
3055 #if !ONLY_MSPACES
3056 /* Set up lock for main malloc area */
3057 gm->mflags = mparams.default_mflags;
3058 INITIAL_LOCK(&gm->mutex);
3059 #endif
3061 #if (FOOTERS && !INSECURE)
3063 #if USE_DEV_RANDOM
3064 int fd;
3065 unsigned char buf[sizeof(size_t)];
3066 /* Try to use /dev/urandom, else fall back on using time */
3067 if ((fd = open("/dev/urandom", O_RDONLY)) >= 0 &&
3068 read(fd, buf, sizeof(buf)) == sizeof(buf)) {
3069 magic = *((size_t *) buf);
3070 close(fd);
3072 else
3073 #endif /* USE_DEV_RANDOM */
3074 #ifdef WIN32
3075 magic = (size_t)(GetTickCount() ^ (size_t)0x55555555U);
3076 #else
3077 magic = (size_t)(time(0) ^ (size_t)0x55555555U);
3078 #endif
3079 magic |= (size_t)8U; /* ensure nonzero */
3080 magic &= ~(size_t)7U; /* improve chances of fault for bad values */
3082 #else /* (FOOTERS && !INSECURE) */
3083 magic = (size_t)0x58585858U;
3084 #endif /* (FOOTERS && !INSECURE) */
3086 mparams.magic = magic;
3089 RELEASE_MALLOC_GLOBAL_LOCK();
3090 return 1;
3093 /* support for mallopt */
3094 static int change_mparam(int param_number, int value) {
3095 size_t val = (value == -1)? MAX_SIZE_T : (size_t)value;
3096 ensure_initialization();
3097 switch(param_number) {
3098 case M_TRIM_THRESHOLD:
3099 mparams.trim_threshold = val;
3100 return 1;
3101 case M_GRANULARITY:
3102 if (val >= mparams.page_size && ((val & (val-1)) == 0)) {
3103 mparams.granularity = val;
3104 return 1;
3106 else
3107 return 0;
3108 case M_MMAP_THRESHOLD:
3109 mparams.mmap_threshold = val;
3110 return 1;
3111 default:
3112 return 0;
3116 #if DEBUG
3117 /* ------------------------- Debugging Support --------------------------- */
3119 /* Check properties of any chunk, whether free, inuse, mmapped etc */
3120 static void do_check_any_chunk(mstate m, mchunkptr p) {
3121 assert((is_aligned(chunk2mem(p))) || (p->head == FENCEPOST_HEAD));
3122 assert(ok_address(m, p));
3125 /* Check properties of top chunk */
3126 static void do_check_top_chunk(mstate m, mchunkptr p) {
3127 msegmentptr sp = segment_holding(m, (char*)p);
3128 size_t sz = p->head & ~INUSE_BITS; /* third-lowest bit can be set! */
3129 assert(sp != 0);
3130 assert((is_aligned(chunk2mem(p))) || (p->head == FENCEPOST_HEAD));
3131 assert(ok_address(m, p));
3132 assert(sz == m->topsize);
3133 assert(sz > 0);
3134 assert(sz == ((sp->base + sp->size) - (char*)p) - TOP_FOOT_SIZE);
3135 assert(pinuse(p));
3136 assert(!pinuse(chunk_plus_offset(p, sz)));
3139 /* Check properties of (inuse) mmapped chunks */
3140 static void do_check_mmapped_chunk(mstate m, mchunkptr p) {
3141 size_t sz = chunksize(p);
3142 size_t len = (sz + (p->prev_foot & ~IS_MMAPPED_BIT) + MMAP_FOOT_PAD);
3143 assert(is_mmapped(p));
3144 assert(use_mmap(m));
3145 assert((is_aligned(chunk2mem(p))) || (p->head == FENCEPOST_HEAD));
3146 assert(ok_address(m, p));
3147 assert(!is_small(sz));
3148 assert((len & (mparams.page_size-SIZE_T_ONE)) == 0);
3149 assert(chunk_plus_offset(p, sz)->head == FENCEPOST_HEAD);
3150 assert(chunk_plus_offset(p, sz+SIZE_T_SIZE)->head == 0);
3153 /* Check properties of inuse chunks */
3154 static void do_check_inuse_chunk(mstate m, mchunkptr p) {
3155 do_check_any_chunk(m, p);
3156 assert(cinuse(p));
3157 assert(next_pinuse(p));
3158 /* If not pinuse and not mmapped, previous chunk has OK offset */
3159 assert(is_mmapped(p) || pinuse(p) || next_chunk(prev_chunk(p)) == p);
3160 if (is_mmapped(p))
3161 do_check_mmapped_chunk(m, p);
3164 /* Check properties of free chunks */
3165 static void do_check_free_chunk(mstate m, mchunkptr p) {
3166 size_t sz = chunksize(p);
3167 mchunkptr next = chunk_plus_offset(p, sz);
3168 do_check_any_chunk(m, p);
3169 assert(!cinuse(p));
3170 assert(!next_pinuse(p));
3171 assert (!is_mmapped(p));
3172 if (p != m->dv && p != m->top) {
3173 if (sz >= MIN_CHUNK_SIZE) {
3174 assert((sz & CHUNK_ALIGN_MASK) == 0);
3175 assert(is_aligned(chunk2mem(p)));
3176 assert(next->prev_foot == sz);
3177 assert(pinuse(p));
3178 assert (next == m->top || cinuse(next));
3179 assert(p->fd->bk == p);
3180 assert(p->bk->fd == p);
3182 else /* markers are always of size SIZE_T_SIZE */
3183 assert(sz == SIZE_T_SIZE);
3187 /* Check properties of malloced chunks at the point they are malloced */
3188 static void do_check_malloced_chunk(mstate m, void* mem, size_t s) {
3189 if (mem != 0) {
3190 mchunkptr p = mem2chunk(mem);
3191 size_t sz = p->head & ~(PINUSE_BIT|CINUSE_BIT);
3192 do_check_inuse_chunk(m, p);
3193 assert((sz & CHUNK_ALIGN_MASK) == 0);
3194 assert(sz >= MIN_CHUNK_SIZE);
3195 assert(sz >= s);
3196 /* unless mmapped, size is less than MIN_CHUNK_SIZE more than request */
3197 assert(is_mmapped(p) || sz < (s + MIN_CHUNK_SIZE));
3201 /* Check a tree and its subtrees. */
3202 static void do_check_tree(mstate m, tchunkptr t) {
3203 tchunkptr head = 0;
3204 tchunkptr u = t;
3205 bindex_t tindex = t->index;
3206 size_t tsize = chunksize(t);
3207 bindex_t idx;
3208 compute_tree_index(tsize, idx);
3209 assert(tindex == idx);
3210 assert(tsize >= MIN_LARGE_SIZE);
3211 assert(tsize >= minsize_for_tree_index(idx));
3212 assert((idx == NTREEBINS-1) || (tsize < minsize_for_tree_index((idx+1))));
3214 do { /* traverse through chain of same-sized nodes */
3215 do_check_any_chunk(m, ((mchunkptr)u));
3216 assert(u->index == tindex);
3217 assert(chunksize(u) == tsize);
3218 assert(!cinuse(u));
3219 assert(!next_pinuse(u));
3220 assert(u->fd->bk == u);
3221 assert(u->bk->fd == u);
3222 if (u->parent == 0) {
3223 assert(u->child[0] == 0);
3224 assert(u->child[1] == 0);
3226 else {
3227 assert(head == 0); /* only one node on chain has parent */
3228 head = u;
3229 assert(u->parent != u);
3230 assert (u->parent->child[0] == u ||
3231 u->parent->child[1] == u ||
3232 *((tbinptr*)(u->parent)) == u);
3233 if (u->child[0] != 0) {
3234 assert(u->child[0]->parent == u);
3235 assert(u->child[0] != u);
3236 do_check_tree(m, u->child[0]);
3238 if (u->child[1] != 0) {
3239 assert(u->child[1]->parent == u);
3240 assert(u->child[1] != u);
3241 do_check_tree(m, u->child[1]);
3243 if (u->child[0] != 0 && u->child[1] != 0) {
3244 assert(chunksize(u->child[0]) < chunksize(u->child[1]));
3247 u = u->fd;
3248 } while (u != t);
3249 assert(head != 0);
3252 /* Check all the chunks in a treebin. */
3253 static void do_check_treebin(mstate m, bindex_t i) {
3254 tbinptr* tb = treebin_at(m, i);
3255 tchunkptr t = *tb;
3256 int empty = (m->treemap & (1U << i)) == 0;
3257 if (t == 0)
3258 assert(empty);
3259 if (!empty)
3260 do_check_tree(m, t);
3263 /* Check all the chunks in a smallbin. */
3264 static void do_check_smallbin(mstate m, bindex_t i) {
3265 sbinptr b = smallbin_at(m, i);
3266 mchunkptr p = b->bk;
3267 unsigned int empty = (m->smallmap & (1U << i)) == 0;
3268 if (p == b)
3269 assert(empty);
3270 if (!empty) {
3271 for (; p != b; p = p->bk) {
3272 size_t size = chunksize(p);
3273 mchunkptr q;
3274 /* each chunk claims to be free */
3275 do_check_free_chunk(m, p);
3276 /* chunk belongs in bin */
3277 assert(small_index(size) == i);
3278 assert(p->bk == b || chunksize(p->bk) == chunksize(p));
3279 /* chunk is followed by an inuse chunk */
3280 q = next_chunk(p);
3281 if (q->head != FENCEPOST_HEAD)
3282 do_check_inuse_chunk(m, q);
3287 /* Find x in a bin. Used in other check functions. */
3288 static int bin_find(mstate m, mchunkptr x) {
3289 size_t size = chunksize(x);
3290 if (is_small(size)) {
3291 bindex_t sidx = small_index(size);
3292 sbinptr b = smallbin_at(m, sidx);
3293 if (smallmap_is_marked(m, sidx)) {
3294 mchunkptr p = b;
3295 do {
3296 if (p == x)
3297 return 1;
3298 } while ((p = p->fd) != b);
3301 else {
3302 bindex_t tidx;
3303 compute_tree_index(size, tidx);
3304 if (treemap_is_marked(m, tidx)) {
3305 tchunkptr t = *treebin_at(m, tidx);
3306 size_t sizebits = size << leftshift_for_tree_index(tidx);
3307 while (t != 0 && chunksize(t) != size) {
3308 t = t->child[(sizebits >> (SIZE_T_BITSIZE-SIZE_T_ONE)) & 1];
3309 sizebits <<= 1;
3311 if (t != 0) {
3312 tchunkptr u = t;
3313 do {
3314 if (u == (tchunkptr)x)
3315 return 1;
3316 } while ((u = u->fd) != t);
3320 return 0;
3323 /* Traverse each chunk and check it; return total */
3324 static size_t traverse_and_check(mstate m) {
3325 size_t sum = 0;
3326 if (is_initialized(m)) {
3327 msegmentptr s = &m->seg;
3328 sum += m->topsize + TOP_FOOT_SIZE;
3329 while (s != 0) {
3330 mchunkptr q = align_as_chunk(s->base);
3331 mchunkptr lastq = 0;
3332 assert(pinuse(q));
3333 while (segment_holds(s, q) &&
3334 q != m->top && q->head != FENCEPOST_HEAD) {
3335 sum += chunksize(q);
3336 if (cinuse(q)) {
3337 assert(!bin_find(m, q));
3338 do_check_inuse_chunk(m, q);
3340 else {
3341 assert(q == m->dv || bin_find(m, q));
3342 assert(lastq == 0 || cinuse(lastq)); /* Not 2 consecutive free */
3343 do_check_free_chunk(m, q);
3345 lastq = q;
3346 q = next_chunk(q);
3348 s = s->next;
3351 return sum;
3354 /* Check all properties of malloc_state. */
3355 static void do_check_malloc_state(mstate m) {
3356 bindex_t i;
3357 size_t total;
3358 /* check bins */
3359 for (i = 0; i < NSMALLBINS; ++i)
3360 do_check_smallbin(m, i);
3361 for (i = 0; i < NTREEBINS; ++i)
3362 do_check_treebin(m, i);
3364 if (m->dvsize != 0) { /* check dv chunk */
3365 do_check_any_chunk(m, m->dv);
3366 assert(m->dvsize == chunksize(m->dv));
3367 assert(m->dvsize >= MIN_CHUNK_SIZE);
3368 assert(bin_find(m, m->dv) == 0);
3371 if (m->top != 0) { /* check top chunk */
3372 do_check_top_chunk(m, m->top);
3373 /*assert(m->topsize == chunksize(m->top)); redundant */
3374 assert(m->topsize > 0);
3375 assert(bin_find(m, m->top) == 0);
3378 total = traverse_and_check(m);
3379 assert(total <= m->footprint);
3380 assert(m->footprint <= m->max_footprint);
3382 #endif /* DEBUG */
3384 /* ----------------------------- statistics ------------------------------ */
3386 #if !NO_MALLINFO
3387 static struct mallinfo internal_mallinfo(mstate m) {
3388 struct mallinfo nm = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
3389 ensure_initialization();
3390 if (!PREACTION(m)) {
3391 check_malloc_state(m);
3392 if (is_initialized(m)) {
3393 size_t nfree = SIZE_T_ONE; /* top always free */
3394 size_t mfree = m->topsize + TOP_FOOT_SIZE;
3395 size_t sum = mfree;
3396 msegmentptr s = &m->seg;
3397 while (s != 0) {
3398 mchunkptr q = align_as_chunk(s->base);
3399 while (segment_holds(s, q) &&
3400 q != m->top && q->head != FENCEPOST_HEAD) {
3401 size_t sz = chunksize(q);
3402 sum += sz;
3403 if (!cinuse(q)) {
3404 mfree += sz;
3405 ++nfree;
3407 q = next_chunk(q);
3409 s = s->next;
3412 nm.arena = sum;
3413 nm.ordblks = nfree;
3414 nm.hblkhd = m->footprint - sum;
3415 nm.usmblks = m->max_footprint;
3416 nm.uordblks = m->footprint - mfree;
3417 nm.fordblks = mfree;
3418 nm.keepcost = m->topsize;
3421 POSTACTION(m);
3423 return nm;
3425 #endif /* !NO_MALLINFO */
3427 static void internal_malloc_stats(mstate m) {
3428 ensure_initialization();
3429 if (!PREACTION(m)) {
3430 size_t maxfp = 0;
3431 size_t fp = 0;
3432 size_t used = 0;
3433 check_malloc_state(m);
3434 if (is_initialized(m)) {
3435 msegmentptr s = &m->seg;
3436 maxfp = m->max_footprint;
3437 fp = m->footprint;
3438 used = fp - (m->topsize + TOP_FOOT_SIZE);
3440 while (s != 0) {
3441 mchunkptr q = align_as_chunk(s->base);
3442 while (segment_holds(s, q) &&
3443 q != m->top && q->head != FENCEPOST_HEAD) {
3444 if (!cinuse(q))
3445 used -= chunksize(q);
3446 q = next_chunk(q);
3448 s = s->next;
3452 fprintf(stderr, "max system bytes = %10lu\n", (unsigned long)(maxfp));
3453 fprintf(stderr, "system bytes = %10lu\n", (unsigned long)(fp));
3454 fprintf(stderr, "in use bytes = %10lu\n", (unsigned long)(used));
3456 POSTACTION(m);
3460 /* ----------------------- Operations on smallbins ----------------------- */
3463 Various forms of linking and unlinking are defined as macros. Even
3464 the ones for trees, which are very long but have very short typical
3465 paths. This is ugly but reduces reliance on inlining support of
3466 compilers.
3469 /* Link a free chunk into a smallbin */
3470 #define insert_small_chunk(M, P, S) {\
3471 bindex_t I = small_index(S);\
3472 mchunkptr B = smallbin_at(M, I);\
3473 mchunkptr F = B;\
3474 assert(S >= MIN_CHUNK_SIZE);\
3475 if (!smallmap_is_marked(M, I))\
3476 mark_smallmap(M, I);\
3477 else if (RTCHECK(ok_address(M, B->fd)))\
3478 F = B->fd;\
3479 else {\
3480 CORRUPTION_ERROR_ACTION(M);\
3482 B->fd = P;\
3483 F->bk = P;\
3484 P->fd = F;\
3485 P->bk = B;\
3488 /* Unlink a chunk from a smallbin */
3489 #define unlink_small_chunk(M, P, S) {\
3490 mchunkptr F = P->fd;\
3491 mchunkptr B = P->bk;\
3492 bindex_t I = small_index(S);\
3493 assert(P != B);\
3494 assert(P != F);\
3495 assert(chunksize(P) == small_index2size(I));\
3496 if (F == B)\
3497 clear_smallmap(M, I);\
3498 else if (RTCHECK((F == smallbin_at(M,I) || ok_address(M, F)) &&\
3499 (B == smallbin_at(M,I) || ok_address(M, B)))) {\
3500 F->bk = B;\
3501 B->fd = F;\
3503 else {\
3504 CORRUPTION_ERROR_ACTION(M);\
3508 /* Unlink the first chunk from a smallbin */
3509 #define unlink_first_small_chunk(M, B, P, I) {\
3510 mchunkptr F = P->fd;\
3511 assert(P != B);\
3512 assert(P != F);\
3513 assert(chunksize(P) == small_index2size(I));\
3514 if (B == F)\
3515 clear_smallmap(M, I);\
3516 else if (RTCHECK(ok_address(M, F))) {\
3517 B->fd = F;\
3518 F->bk = B;\
3520 else {\
3521 CORRUPTION_ERROR_ACTION(M);\
3527 /* Replace dv node, binning the old one */
3528 /* Used only when dvsize known to be small */
3529 #define replace_dv(M, P, S) {\
3530 size_t DVS = M->dvsize;\
3531 if (DVS != 0) {\
3532 mchunkptr DV = M->dv;\
3533 assert(is_small(DVS));\
3534 insert_small_chunk(M, DV, DVS);\
3536 M->dvsize = S;\
3537 M->dv = P;\
3540 /* ------------------------- Operations on trees ------------------------- */
3542 /* Insert chunk into tree */
3543 #define insert_large_chunk(M, X, S) {\
3544 tbinptr* H;\
3545 bindex_t I;\
3546 compute_tree_index(S, I);\
3547 H = treebin_at(M, I);\
3548 X->index = I;\
3549 X->child[0] = X->child[1] = 0;\
3550 if (!treemap_is_marked(M, I)) {\
3551 mark_treemap(M, I);\
3552 *H = X;\
3553 X->parent = (tchunkptr)H;\
3554 X->fd = X->bk = X;\
3556 else {\
3557 tchunkptr T = *H;\
3558 size_t K = S << leftshift_for_tree_index(I);\
3559 for (;;) {\
3560 if (chunksize(T) != S) {\
3561 tchunkptr* C = &(T->child[(K >> (SIZE_T_BITSIZE-SIZE_T_ONE)) & 1]);\
3562 K <<= 1;\
3563 if (*C != 0)\
3564 T = *C;\
3565 else if (RTCHECK(ok_address(M, C))) {\
3566 *C = X;\
3567 X->parent = T;\
3568 X->fd = X->bk = X;\
3569 break;\
3571 else {\
3572 CORRUPTION_ERROR_ACTION(M);\
3573 break;\
3576 else {\
3577 tchunkptr F = T->fd;\
3578 if (RTCHECK(ok_address(M, T) && ok_address(M, F))) {\
3579 T->fd = F->bk = X;\
3580 X->fd = F;\
3581 X->bk = T;\
3582 X->parent = 0;\
3583 break;\
3585 else {\
3586 CORRUPTION_ERROR_ACTION(M);\
3587 break;\
3595 Unlink steps:
3597 1. If x is a chained node, unlink it from its same-sized fd/bk links
3598 and choose its bk node as its replacement.
3599 2. If x was the last node of its size, but not a leaf node, it must
3600 be replaced with a leaf node (not merely one with an open left or
3601 right), to make sure that lefts and rights of descendents
3602 correspond properly to bit masks. We use the rightmost descendent
3603 of x. We could use any other leaf, but this is easy to locate and
3604 tends to counteract removal of leftmosts elsewhere, and so keeps
3605 paths shorter than minimally guaranteed. This doesn't loop much
3606 because on average a node in a tree is near the bottom.
3607 3. If x is the base of a chain (i.e., has parent links) relink
3608 x's parent and children to x's replacement (or null if none).
3611 #define unlink_large_chunk(M, X) {\
3612 tchunkptr XP = X->parent;\
3613 tchunkptr R;\
3614 if (X->bk != X) {\
3615 tchunkptr F = X->fd;\
3616 R = X->bk;\
3617 if (RTCHECK(ok_address(M, F))) {\
3618 F->bk = R;\
3619 R->fd = F;\
3621 else {\
3622 CORRUPTION_ERROR_ACTION(M);\
3625 else {\
3626 tchunkptr* RP;\
3627 if (((R = *(RP = &(X->child[1]))) != 0) ||\
3628 ((R = *(RP = &(X->child[0]))) != 0)) {\
3629 tchunkptr* CP;\
3630 while ((*(CP = &(R->child[1])) != 0) ||\
3631 (*(CP = &(R->child[0])) != 0)) {\
3632 R = *(RP = CP);\
3634 if (RTCHECK(ok_address(M, RP)))\
3635 *RP = 0;\
3636 else {\
3637 CORRUPTION_ERROR_ACTION(M);\
3641 if (XP != 0) {\
3642 tbinptr* H = treebin_at(M, X->index);\
3643 if (X == *H) {\
3644 if ((*H = R) == 0) \
3645 clear_treemap(M, X->index);\
3647 else if (RTCHECK(ok_address(M, XP))) {\
3648 if (XP->child[0] == X) \
3649 XP->child[0] = R;\
3650 else \
3651 XP->child[1] = R;\
3653 else\
3654 CORRUPTION_ERROR_ACTION(M);\
3655 if (R != 0) {\
3656 if (RTCHECK(ok_address(M, R))) {\
3657 tchunkptr C0, C1;\
3658 R->parent = XP;\
3659 if ((C0 = X->child[0]) != 0) {\
3660 if (RTCHECK(ok_address(M, C0))) {\
3661 R->child[0] = C0;\
3662 C0->parent = R;\
3664 else\
3665 CORRUPTION_ERROR_ACTION(M);\
3667 if ((C1 = X->child[1]) != 0) {\
3668 if (RTCHECK(ok_address(M, C1))) {\
3669 R->child[1] = C1;\
3670 C1->parent = R;\
3672 else\
3673 CORRUPTION_ERROR_ACTION(M);\
3676 else\
3677 CORRUPTION_ERROR_ACTION(M);\
3682 /* Relays to large vs small bin operations */
3684 #define insert_chunk(M, P, S)\
3685 if (is_small(S)) insert_small_chunk(M, P, S)\
3686 else { tchunkptr TP = (tchunkptr)(P); insert_large_chunk(M, TP, S); }
3688 #define unlink_chunk(M, P, S)\
3689 if (is_small(S)) unlink_small_chunk(M, P, S)\
3690 else { tchunkptr TP = (tchunkptr)(P); unlink_large_chunk(M, TP); }
3693 /* Relays to internal calls to malloc/free from realloc, memalign etc */
3695 #if ONLY_MSPACES
3696 #define internal_malloc(m, b) mspace_malloc(m, b)
3697 #define internal_free(m, mem) mspace_free(m,mem);
3698 #else /* ONLY_MSPACES */
3699 #if MSPACES
3700 #define internal_malloc(m, b)\
3701 (m == gm)? dlmalloc(b) : mspace_malloc(m, b)
3702 #define internal_free(m, mem)\
3703 if (m == gm) dlfree(mem); else mspace_free(m,mem);
3704 #else /* MSPACES */
3705 #define internal_malloc(m, b) dlmalloc(b)
3706 #define internal_free(m, mem) dlfree(mem)
3707 #endif /* MSPACES */
3708 #endif /* ONLY_MSPACES */
3710 /* ----------------------- Direct-mmapping chunks ----------------------- */
3713 Directly mmapped chunks are set up with an offset to the start of
3714 the mmapped region stored in the prev_foot field of the chunk. This
3715 allows reconstruction of the required argument to MUNMAP when freed,
3716 and also allows adjustment of the returned chunk to meet alignment
3717 requirements (especially in memalign). There is also enough space
3718 allocated to hold a fake next chunk of size SIZE_T_SIZE to maintain
3719 the PINUSE bit so frees can be checked.
3722 /* Malloc using mmap */
3723 static void* mmap_alloc(mstate m, size_t nb) {
3724 size_t mmsize = mmap_align(nb + SIX_SIZE_T_SIZES + CHUNK_ALIGN_MASK);
3725 if (mmsize > nb) { /* Check for wrap around 0 */
3726 char* mm = (char*)(CALL_DIRECT_MMAP(mmsize));
3727 if (mm != CMFAIL) {
3728 size_t offset = align_offset(chunk2mem(mm));
3729 size_t psize = mmsize - offset - MMAP_FOOT_PAD;
3730 mchunkptr p = (mchunkptr)(mm + offset);
3731 p->prev_foot = offset | IS_MMAPPED_BIT;
3732 (p)->head = (psize|CINUSE_BIT);
3733 mark_inuse_foot(m, p, psize);
3734 chunk_plus_offset(p, psize)->head = FENCEPOST_HEAD;
3735 chunk_plus_offset(p, psize+SIZE_T_SIZE)->head = 0;
3737 if (mm < m->least_addr)
3738 m->least_addr = mm;
3739 if ((m->footprint += mmsize) > m->max_footprint)
3740 m->max_footprint = m->footprint;
3741 assert(is_aligned(chunk2mem(p)));
3742 check_mmapped_chunk(m, p);
3743 return chunk2mem(p);
3746 return 0;
3749 /* Realloc using mmap */
3750 static mchunkptr mmap_resize(mstate m, mchunkptr oldp, size_t nb) {
3751 size_t oldsize = chunksize(oldp);
3752 if (is_small(nb)) /* Can't shrink mmap regions below small size */
3753 return 0;
3754 /* Keep old chunk if big enough but not too big */
3755 if (oldsize >= nb + SIZE_T_SIZE &&
3756 (oldsize - nb) <= (mparams.granularity << 1))
3757 return oldp;
3758 else {
3759 size_t offset = oldp->prev_foot & ~IS_MMAPPED_BIT;
3760 size_t oldmmsize = oldsize + offset + MMAP_FOOT_PAD;
3761 size_t newmmsize = mmap_align(nb + SIX_SIZE_T_SIZES + CHUNK_ALIGN_MASK);
3762 char* cp = (char*)CALL_MREMAP((char*)oldp - offset,
3763 oldmmsize, newmmsize, 1);
3764 if (cp != CMFAIL) {
3765 mchunkptr newp = (mchunkptr)(cp + offset);
3766 size_t psize = newmmsize - offset - MMAP_FOOT_PAD;
3767 newp->head = (psize|CINUSE_BIT);
3768 mark_inuse_foot(m, newp, psize);
3769 chunk_plus_offset(newp, psize)->head = FENCEPOST_HEAD;
3770 chunk_plus_offset(newp, psize+SIZE_T_SIZE)->head = 0;
3772 if (cp < m->least_addr)
3773 m->least_addr = cp;
3774 if ((m->footprint += newmmsize - oldmmsize) > m->max_footprint)
3775 m->max_footprint = m->footprint;
3776 check_mmapped_chunk(m, newp);
3777 return newp;
3780 return 0;
3783 /* -------------------------- mspace management -------------------------- */
3785 /* Initialize top chunk and its size */
3786 static void init_top(mstate m, mchunkptr p, size_t psize) {
3787 /* Ensure alignment */
3788 size_t offset = align_offset(chunk2mem(p));
3789 p = (mchunkptr)((char*)p + offset);
3790 psize -= offset;
3792 m->top = p;
3793 m->topsize = psize;
3794 p->head = psize | PINUSE_BIT;
3795 /* set size of fake trailing chunk holding overhead space only once */
3796 chunk_plus_offset(p, psize)->head = TOP_FOOT_SIZE;
3797 m->trim_check = mparams.trim_threshold; /* reset on each update */
3800 /* Initialize bins for a new mstate that is otherwise zeroed out */
3801 static void init_bins(mstate m) {
3802 /* Establish circular links for smallbins */
3803 bindex_t i;
3804 for (i = 0; i < NSMALLBINS; ++i) {
3805 sbinptr bin = smallbin_at(m,i);
3806 bin->fd = bin->bk = bin;
3810 #if PROCEED_ON_ERROR
3812 /* default corruption action */
3813 static void reset_on_error(mstate m) {
3814 int i;
3815 ++malloc_corruption_error_count;
3816 /* Reinitialize fields to forget about all memory */
3817 m->smallbins = m->treebins = 0;
3818 m->dvsize = m->topsize = 0;
3819 m->seg.base = 0;
3820 m->seg.size = 0;
3821 m->seg.next = 0;
3822 m->top = m->dv = 0;
3823 for (i = 0; i < NTREEBINS; ++i)
3824 *treebin_at(m, i) = 0;
3825 init_bins(m);
3827 #endif /* PROCEED_ON_ERROR */
3829 /* Allocate chunk and prepend remainder with chunk in successor base. */
3830 static void* prepend_alloc(mstate m, char* newbase, char* oldbase,
3831 size_t nb) {
3832 mchunkptr p = align_as_chunk(newbase);
3833 mchunkptr oldfirst = align_as_chunk(oldbase);
3834 size_t psize = (char*)oldfirst - (char*)p;
3835 mchunkptr q = chunk_plus_offset(p, nb);
3836 size_t qsize = psize - nb;
3837 set_size_and_pinuse_of_inuse_chunk(m, p, nb);
3839 assert((char*)oldfirst > (char*)q);
3840 assert(pinuse(oldfirst));
3841 assert(qsize >= MIN_CHUNK_SIZE);
3843 /* consolidate remainder with first chunk of old base */
3844 if (oldfirst == m->top) {
3845 size_t tsize = m->topsize += qsize;
3846 m->top = q;
3847 q->head = tsize | PINUSE_BIT;
3848 check_top_chunk(m, q);
3850 else if (oldfirst == m->dv) {
3851 size_t dsize = m->dvsize += qsize;
3852 m->dv = q;
3853 set_size_and_pinuse_of_free_chunk(q, dsize);
3855 else {
3856 if (!cinuse(oldfirst)) {
3857 size_t nsize = chunksize(oldfirst);
3858 unlink_chunk(m, oldfirst, nsize);
3859 oldfirst = chunk_plus_offset(oldfirst, nsize);
3860 qsize += nsize;
3862 set_free_with_pinuse(q, qsize, oldfirst);
3863 insert_chunk(m, q, qsize);
3864 check_free_chunk(m, q);
3867 check_malloced_chunk(m, chunk2mem(p), nb);
3868 return chunk2mem(p);
3871 /* Add a segment to hold a new noncontiguous region */
3872 static void add_segment(mstate m, char* tbase, size_t tsize, flag_t mmapped) {
3873 /* Determine locations and sizes of segment, fenceposts, old top */
3874 char* old_top = (char*)m->top;
3875 msegmentptr oldsp = segment_holding(m, old_top);
3876 char* old_end = oldsp->base + oldsp->size;
3877 size_t ssize = pad_request(sizeof(struct malloc_segment));
3878 char* rawsp = old_end - (ssize + FOUR_SIZE_T_SIZES + CHUNK_ALIGN_MASK);
3879 size_t offset = align_offset(chunk2mem(rawsp));
3880 char* asp = rawsp + offset;
3881 char* csp = (asp < (old_top + MIN_CHUNK_SIZE))? old_top : asp;
3882 mchunkptr sp = (mchunkptr)csp;
3883 msegmentptr ss = (msegmentptr)(chunk2mem(sp));
3884 mchunkptr tnext = chunk_plus_offset(sp, ssize);
3885 mchunkptr p = tnext;
3886 int nfences = 0;
3888 /* reset top to new space */
3889 init_top(m, (mchunkptr)tbase, tsize - TOP_FOOT_SIZE);
3891 /* Set up segment record */
3892 assert(is_aligned(ss));
3893 set_size_and_pinuse_of_inuse_chunk(m, sp, ssize);
3894 *ss = m->seg; /* Push current record */
3895 m->seg.base = tbase;
3896 m->seg.size = tsize;
3897 m->seg.sflags = mmapped;
3898 m->seg.next = ss;
3900 /* Insert trailing fenceposts */
3901 for (;;) {
3902 mchunkptr nextp = chunk_plus_offset(p, SIZE_T_SIZE);
3903 p->head = FENCEPOST_HEAD;
3904 ++nfences;
3905 if ((char*)(&(nextp->head)) < old_end)
3906 p = nextp;
3907 else
3908 break;
3910 assert(nfences >= 2);
3912 /* Insert the rest of old top into a bin as an ordinary free chunk */
3913 if (csp != old_top) {
3914 mchunkptr q = (mchunkptr)old_top;
3915 size_t psize = csp - old_top;
3916 mchunkptr tn = chunk_plus_offset(q, psize);
3917 set_free_with_pinuse(q, psize, tn);
3918 insert_chunk(m, q, psize);
3921 check_top_chunk(m, m->top);
3924 /* -------------------------- System allocation -------------------------- */
3926 /* Get memory from system using MORECORE or MMAP */
3927 static void* sys_alloc(mstate m, size_t nb) {
3928 char* tbase = CMFAIL;
3929 size_t tsize = 0;
3930 flag_t mmap_flag = 0;
3932 ensure_initialization();
3934 /* Directly map large chunks */
3935 if (use_mmap(m) && nb >= mparams.mmap_threshold) {
3936 void* mem = mmap_alloc(m, nb);
3937 if (mem != 0)
3938 return mem;
3942 Try getting memory in any of three ways (in most-preferred to
3943 least-preferred order):
3944 1. A call to MORECORE that can normally contiguously extend memory.
3945 (disabled if not MORECORE_CONTIGUOUS or not HAVE_MORECORE or
3946 main space is mmapped or a previous contiguous call failed)
3947 2. A call to MMAP new space (disabled if not HAVE_MMAP).
3948 Note that under the default settings, if MORECORE is unable to
3949 fulfill a request, and HAVE_MMAP is true, then mmap is
3950 used as a noncontiguous system allocator. This is a useful backup
3951 strategy for systems with holes in address spaces -- in this case
3952 sbrk cannot contiguously expand the heap, but mmap may be able to
3953 find space.
3954 3. A call to MORECORE that cannot usually contiguously extend memory.
3955 (disabled if not HAVE_MORECORE)
3957 In all cases, we need to request enough bytes from system to ensure
3958 we can malloc nb bytes upon success, so pad with enough space for
3959 top_foot, plus alignment-pad to make sure we don't lose bytes if
3960 not on boundary, and round this up to a granularity unit.
3963 if (MORECORE_CONTIGUOUS && !use_noncontiguous(m)) {
3964 char* br = CMFAIL;
3965 msegmentptr ss = (m->top == 0)? 0 : segment_holding(m, (char*)m->top);
3966 size_t asize = 0;
3967 ACQUIRE_MALLOC_GLOBAL_LOCK();
3969 if (ss == 0) { /* First time through or recovery */
3970 char* base = (char*)CALL_MORECORE(0);
3971 if (base != CMFAIL) {
3972 asize = granularity_align(nb + SYS_ALLOC_PADDING);
3973 /* Adjust to end on a page boundary */
3974 if (!is_page_aligned(base))
3975 asize += (page_align((size_t)base) - (size_t)base);
3976 /* Can't call MORECORE if size is negative when treated as signed */
3977 if (asize < HALF_MAX_SIZE_T &&
3978 (br = (char*)(CALL_MORECORE(asize))) == base) {
3979 tbase = base;
3980 tsize = asize;
3984 else {
3985 /* Subtract out existing available top space from MORECORE request. */
3986 asize = granularity_align(nb - m->topsize + SYS_ALLOC_PADDING);
3987 /* Use mem here only if it did continuously extend old space */
3988 if (asize < HALF_MAX_SIZE_T &&
3989 (br = (char*)(CALL_MORECORE(asize))) == ss->base+ss->size) {
3990 tbase = br;
3991 tsize = asize;
3995 if (tbase == CMFAIL) { /* Cope with partial failure */
3996 if (br != CMFAIL) { /* Try to use/extend the space we did get */
3997 if (asize < HALF_MAX_SIZE_T &&
3998 asize < nb + SYS_ALLOC_PADDING) {
3999 size_t esize = granularity_align(nb + SYS_ALLOC_PADDING - asize);
4000 if (esize < HALF_MAX_SIZE_T) {
4001 char* end = (char*)CALL_MORECORE(esize);
4002 if (end != CMFAIL)
4003 asize += esize;
4004 else { /* Can't use; try to release */
4005 (void) CALL_MORECORE(-asize);
4006 br = CMFAIL;
4011 if (br != CMFAIL) { /* Use the space we did get */
4012 tbase = br;
4013 tsize = asize;
4015 else
4016 disable_contiguous(m); /* Don't try contiguous path in the future */
4019 RELEASE_MALLOC_GLOBAL_LOCK();
4022 if (HAVE_MMAP && tbase == CMFAIL) { /* Try MMAP */
4023 size_t rsize = granularity_align(nb + SYS_ALLOC_PADDING);
4024 if (rsize > nb) { /* Fail if wraps around zero */
4025 char* mp = (char*)(CALL_MMAP(rsize));
4026 if (mp != CMFAIL) {
4027 tbase = mp;
4028 tsize = rsize;
4029 mmap_flag = IS_MMAPPED_BIT;
4034 if (HAVE_MORECORE && tbase == CMFAIL) { /* Try noncontiguous MORECORE */
4035 size_t asize = granularity_align(nb + SYS_ALLOC_PADDING);
4036 if (asize < HALF_MAX_SIZE_T) {
4037 char* br = CMFAIL;
4038 char* end = CMFAIL;
4039 ACQUIRE_MALLOC_GLOBAL_LOCK();
4040 br = (char*)(CALL_MORECORE(asize));
4041 end = (char*)(CALL_MORECORE(0));
4042 RELEASE_MALLOC_GLOBAL_LOCK();
4043 if (br != CMFAIL && end != CMFAIL && br < end) {
4044 size_t ssize = end - br;
4045 if (ssize > nb + TOP_FOOT_SIZE) {
4046 tbase = br;
4047 tsize = ssize;
4053 if (tbase != CMFAIL) {
4055 if ((m->footprint += tsize) > m->max_footprint)
4056 m->max_footprint = m->footprint;
4058 if (!is_initialized(m)) { /* first-time initialization */
4059 m->seg.base = m->least_addr = tbase;
4060 m->seg.size = tsize;
4061 m->seg.sflags = mmap_flag;
4062 m->magic = mparams.magic;
4063 m->release_checks = MAX_RELEASE_CHECK_RATE;
4064 init_bins(m);
4065 #if !ONLY_MSPACES
4066 if (is_global(m))
4067 init_top(m, (mchunkptr)tbase, tsize - TOP_FOOT_SIZE);
4068 else
4069 #endif
4071 /* Offset top by embedded malloc_state */
4072 mchunkptr mn = next_chunk(mem2chunk(m));
4073 init_top(m, mn, (size_t)((tbase + tsize) - (char*)mn) -TOP_FOOT_SIZE);
4077 else {
4078 /* Try to merge with an existing segment */
4079 msegmentptr sp = &m->seg;
4080 /* Only consider most recent segment if traversal suppressed */
4081 while (sp != 0 && tbase != sp->base + sp->size)
4082 sp = (NO_SEGMENT_TRAVERSAL) ? 0 : sp->next;
4083 if (sp != 0 &&
4084 !is_extern_segment(sp) &&
4085 (sp->sflags & IS_MMAPPED_BIT) == mmap_flag &&
4086 segment_holds(sp, m->top)) { /* append */
4087 sp->size += tsize;
4088 init_top(m, m->top, m->topsize + tsize);
4090 else {
4091 if (tbase < m->least_addr)
4092 m->least_addr = tbase;
4093 sp = &m->seg;
4094 while (sp != 0 && sp->base != tbase + tsize)
4095 sp = (NO_SEGMENT_TRAVERSAL) ? 0 : sp->next;
4096 if (sp != 0 &&
4097 !is_extern_segment(sp) &&
4098 (sp->sflags & IS_MMAPPED_BIT) == mmap_flag) {
4099 char* oldbase = sp->base;
4100 sp->base = tbase;
4101 sp->size += tsize;
4102 return prepend_alloc(m, tbase, oldbase, nb);
4104 else
4105 add_segment(m, tbase, tsize, mmap_flag);
4109 if (nb < m->topsize) { /* Allocate from new or extended top space */
4110 size_t rsize = m->topsize -= nb;
4111 mchunkptr p = m->top;
4112 mchunkptr r = m->top = chunk_plus_offset(p, nb);
4113 r->head = rsize | PINUSE_BIT;
4114 set_size_and_pinuse_of_inuse_chunk(m, p, nb);
4115 check_top_chunk(m, m->top);
4116 check_malloced_chunk(m, chunk2mem(p), nb);
4117 return chunk2mem(p);
4121 MALLOC_FAILURE_ACTION;
4122 return 0;
4125 /* ----------------------- system deallocation -------------------------- */
4127 /* Unmap and unlink any mmapped segments that don't contain used chunks */
4128 static size_t release_unused_segments(mstate m) {
4129 size_t released = 0;
4130 int nsegs = 0;
4131 msegmentptr pred = &m->seg;
4132 msegmentptr sp = pred->next;
4133 while (sp != 0) {
4134 char* base = sp->base;
4135 size_t size = sp->size;
4136 msegmentptr next = sp->next;
4137 ++nsegs;
4138 if (is_mmapped_segment(sp) && !is_extern_segment(sp)) {
4139 mchunkptr p = align_as_chunk(base);
4140 size_t psize = chunksize(p);
4141 /* Can unmap if first chunk holds entire segment and not pinned */