]>
Commit | Line | Data |
---|---|---|
56137dbc | 1 | /* Malloc implementation for multiple threads without lock contention. |
7d013a64 | 2 | Copyright (C) 1996-2002,2003,2004,2005,2006 Free Software Foundation, Inc. |
f65fd747 | 3 | This file is part of the GNU C Library. |
fa8d436c UD |
4 | Contributed by Wolfram Gloger <wg@malloc.de> |
5 | and Doug Lea <dl@cs.oswego.edu>, 2001. | |
f65fd747 UD |
6 | |
7 | The GNU C Library is free software; you can redistribute it and/or | |
cc7375ce RM |
8 | modify it under the terms of the GNU Lesser General Public License as |
9 | published by the Free Software Foundation; either version 2.1 of the | |
fa8d436c | 10 | License, or (at your option) any later version. |
f65fd747 UD |
11 | |
12 | The GNU C Library is distributed in the hope that it will be useful, | |
13 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
cc7375ce | 15 | Lesser General Public License for more details. |
f65fd747 | 16 | |
cc7375ce | 17 | You should have received a copy of the GNU Lesser General Public |
fa8d436c UD |
18 | License along with the GNU C Library; see the file COPYING.LIB. If not, |
19 | write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
20 | Boston, MA 02111-1307, USA. */ | |
f65fd747 | 21 | |
fa8d436c UD |
22 | /* |
23 | This is a version (aka ptmalloc2) of malloc/free/realloc written by | |
24 | Doug Lea and adapted to multiple threads/arenas by Wolfram Gloger. | |
25 | ||
26 | * Version ptmalloc2-20011215 | |
fa8d436c UD |
27 | based on: |
28 | VERSION 2.7.0 Sun Mar 11 14:14:06 2001 Doug Lea (dl at gee) | |
f65fd747 | 29 | |
fa8d436c UD |
30 | Note: There may be an updated version of this malloc obtainable at |
31 | http://www.malloc.de/malloc/ptmalloc2.tar.gz | |
32 | Check before installing! | |
f65fd747 | 33 | |
fa8d436c | 34 | * Quickstart |
f65fd747 | 35 | |
fa8d436c UD |
36 | In order to compile this implementation, a Makefile is provided with |
37 | the ptmalloc2 distribution, which has pre-defined targets for some | |
38 | popular systems (e.g. "make posix" for Posix threads). All that is | |
39 | typically required with regard to compiler flags is the selection of | |
40 | the thread package via defining one out of USE_PTHREADS, USE_THR or | |
41 | USE_SPROC. Check the thread-m.h file for what effects this has. | |
42 | Many/most systems will additionally require USE_TSD_DATA_HACK to be | |
43 | defined, so this is the default for "make posix". | |
f65fd747 UD |
44 | |
45 | * Why use this malloc? | |
46 | ||
47 | This is not the fastest, most space-conserving, most portable, or | |
48 | most tunable malloc ever written. However it is among the fastest | |
49 | while also being among the most space-conserving, portable and tunable. | |
50 | Consistent balance across these factors results in a good general-purpose | |
fa8d436c UD |
51 | allocator for malloc-intensive programs. |
52 | ||
53 | The main properties of the algorithms are: | |
54 | * For large (>= 512 bytes) requests, it is a pure best-fit allocator, | |
55 | with ties normally decided via FIFO (i.e. least recently used). | |
56 | * For small (<= 64 bytes by default) requests, it is a caching | |
57 | allocator, that maintains pools of quickly recycled chunks. | |
58 | * In between, and for combinations of large and small requests, it does | |
59 | the best it can trying to meet both goals at once. | |
60 | * For very large requests (>= 128KB by default), it relies on system | |
61 | memory mapping facilities, if supported. | |
62 | ||
63 | For a longer but slightly out of date high-level description, see | |
64 | http://gee.cs.oswego.edu/dl/html/malloc.html | |
65 | ||
66 | You may already by default be using a C library containing a malloc | |
67 | that is based on some version of this malloc (for example in | |
68 | linux). You might still want to use the one in this file in order to | |
69 | customize settings or to avoid overheads associated with library | |
70 | versions. | |
71 | ||
72 | * Contents, described in more detail in "description of public routines" below. | |
73 | ||
74 | Standard (ANSI/SVID/...) functions: | |
75 | malloc(size_t n); | |
76 | calloc(size_t n_elements, size_t element_size); | |
77 | free(Void_t* p); | |
78 | realloc(Void_t* p, size_t n); | |
79 | memalign(size_t alignment, size_t n); | |
80 | valloc(size_t n); | |
81 | mallinfo() | |
82 | mallopt(int parameter_number, int parameter_value) | |
83 | ||
84 | Additional functions: | |
85 | independent_calloc(size_t n_elements, size_t size, Void_t* chunks[]); | |
86 | independent_comalloc(size_t n_elements, size_t sizes[], Void_t* chunks[]); | |
87 | pvalloc(size_t n); | |
88 | cfree(Void_t* p); | |
89 | malloc_trim(size_t pad); | |
90 | malloc_usable_size(Void_t* p); | |
91 | malloc_stats(); | |
f65fd747 UD |
92 | |
93 | * Vital statistics: | |
94 | ||
fa8d436c | 95 | Supported pointer representation: 4 or 8 bytes |
a9177ff5 | 96 | Supported size_t representation: 4 or 8 bytes |
f65fd747 | 97 | Note that size_t is allowed to be 4 bytes even if pointers are 8. |
fa8d436c UD |
98 | You can adjust this by defining INTERNAL_SIZE_T |
99 | ||
100 | Alignment: 2 * sizeof(size_t) (default) | |
101 | (i.e., 8 byte alignment with 4byte size_t). This suffices for | |
102 | nearly all current machines and C compilers. However, you can | |
103 | define MALLOC_ALIGNMENT to be wider than this if necessary. | |
f65fd747 | 104 | |
fa8d436c UD |
105 | Minimum overhead per allocated chunk: 4 or 8 bytes |
106 | Each malloced chunk has a hidden word of overhead holding size | |
f65fd747 UD |
107 | and status information. |
108 | ||
109 | Minimum allocated size: 4-byte ptrs: 16 bytes (including 4 overhead) | |
110 | 8-byte ptrs: 24/32 bytes (including, 4/8 overhead) | |
111 | ||
112 | When a chunk is freed, 12 (for 4byte ptrs) or 20 (for 8 byte | |
113 | ptrs but 4 byte size) or 24 (for 8/8) additional bytes are | |
fa8d436c UD |
114 | needed; 4 (8) for a trailing size field and 8 (16) bytes for |
115 | free list pointers. Thus, the minimum allocatable size is | |
116 | 16/24/32 bytes. | |
f65fd747 UD |
117 | |
118 | Even a request for zero bytes (i.e., malloc(0)) returns a | |
119 | pointer to something of the minimum allocatable size. | |
120 | ||
fa8d436c UD |
121 | The maximum overhead wastage (i.e., number of extra bytes |
122 | allocated than were requested in malloc) is less than or equal | |
123 | to the minimum size, except for requests >= mmap_threshold that | |
124 | are serviced via mmap(), where the worst case wastage is 2 * | |
125 | sizeof(size_t) bytes plus the remainder from a system page (the | |
126 | minimal mmap unit); typically 4096 or 8192 bytes. | |
f65fd747 | 127 | |
a9177ff5 | 128 | Maximum allocated size: 4-byte size_t: 2^32 minus about two pages |
fa8d436c UD |
129 | 8-byte size_t: 2^64 minus about two pages |
130 | ||
131 | It is assumed that (possibly signed) size_t values suffice to | |
f65fd747 UD |
132 | represent chunk sizes. `Possibly signed' is due to the fact |
133 | that `size_t' may be defined on a system as either a signed or | |
fa8d436c UD |
134 | an unsigned type. The ISO C standard says that it must be |
135 | unsigned, but a few systems are known not to adhere to this. | |
136 | Additionally, even when size_t is unsigned, sbrk (which is by | |
137 | default used to obtain memory from system) accepts signed | |
138 | arguments, and may not be able to handle size_t-wide arguments | |
139 | with negative sign bit. Generally, values that would | |
140 | appear as negative after accounting for overhead and alignment | |
141 | are supported only via mmap(), which does not have this | |
142 | limitation. | |
143 | ||
144 | Requests for sizes outside the allowed range will perform an optional | |
145 | failure action and then return null. (Requests may also | |
146 | also fail because a system is out of memory.) | |
147 | ||
148 | Thread-safety: thread-safe unless NO_THREADS is defined | |
149 | ||
150 | Compliance: I believe it is compliant with the 1997 Single Unix Specification | |
a9177ff5 | 151 | (See http://www.opennc.org). Also SVID/XPG, ANSI C, and probably |
fa8d436c | 152 | others as well. |
f65fd747 UD |
153 | |
154 | * Synopsis of compile-time options: | |
155 | ||
156 | People have reported using previous versions of this malloc on all | |
157 | versions of Unix, sometimes by tweaking some of the defines | |
158 | below. It has been tested most extensively on Solaris and | |
fa8d436c UD |
159 | Linux. It is also reported to work on WIN32 platforms. |
160 | People also report using it in stand-alone embedded systems. | |
161 | ||
162 | The implementation is in straight, hand-tuned ANSI C. It is not | |
163 | at all modular. (Sorry!) It uses a lot of macros. To be at all | |
164 | usable, this code should be compiled using an optimizing compiler | |
165 | (for example gcc -O3) that can simplify expressions and control | |
166 | paths. (FAQ: some macros import variables as arguments rather than | |
167 | declare locals because people reported that some debuggers | |
168 | otherwise get confused.) | |
169 | ||
170 | OPTION DEFAULT VALUE | |
171 | ||
172 | Compilation Environment options: | |
173 | ||
174 | __STD_C derived from C compiler defines | |
175 | WIN32 NOT defined | |
176 | HAVE_MEMCPY defined | |
177 | USE_MEMCPY 1 if HAVE_MEMCPY is defined | |
a9177ff5 | 178 | HAVE_MMAP defined as 1 |
fa8d436c UD |
179 | MMAP_CLEARS 1 |
180 | HAVE_MREMAP 0 unless linux defined | |
181 | USE_ARENAS the same as HAVE_MMAP | |
182 | malloc_getpagesize derived from system #includes, or 4096 if not | |
183 | HAVE_USR_INCLUDE_MALLOC_H NOT defined | |
184 | LACKS_UNISTD_H NOT defined unless WIN32 | |
185 | LACKS_SYS_PARAM_H NOT defined unless WIN32 | |
186 | LACKS_SYS_MMAN_H NOT defined unless WIN32 | |
187 | ||
188 | Changing default word sizes: | |
189 | ||
190 | INTERNAL_SIZE_T size_t | |
073f560e UD |
191 | MALLOC_ALIGNMENT MAX (2 * sizeof(INTERNAL_SIZE_T), |
192 | __alignof__ (long double)) | |
fa8d436c UD |
193 | |
194 | Configuration and functionality options: | |
195 | ||
196 | USE_DL_PREFIX NOT defined | |
197 | USE_PUBLIC_MALLOC_WRAPPERS NOT defined | |
198 | USE_MALLOC_LOCK NOT defined | |
199 | MALLOC_DEBUG NOT defined | |
200 | REALLOC_ZERO_BYTES_FREES 1 | |
201 | MALLOC_FAILURE_ACTION errno = ENOMEM, if __STD_C defined, else no-op | |
202 | TRIM_FASTBINS 0 | |
203 | ||
204 | Options for customizing MORECORE: | |
205 | ||
206 | MORECORE sbrk | |
207 | MORECORE_FAILURE -1 | |
a9177ff5 | 208 | MORECORE_CONTIGUOUS 1 |
fa8d436c UD |
209 | MORECORE_CANNOT_TRIM NOT defined |
210 | MORECORE_CLEARS 1 | |
a9177ff5 | 211 | MMAP_AS_MORECORE_SIZE (1024 * 1024) |
fa8d436c UD |
212 | |
213 | Tuning options that are also dynamically changeable via mallopt: | |
214 | ||
215 | DEFAULT_MXFAST 64 | |
216 | DEFAULT_TRIM_THRESHOLD 128 * 1024 | |
217 | DEFAULT_TOP_PAD 0 | |
218 | DEFAULT_MMAP_THRESHOLD 128 * 1024 | |
219 | DEFAULT_MMAP_MAX 65536 | |
220 | ||
221 | There are several other #defined constants and macros that you | |
222 | probably don't want to touch unless you are extending or adapting malloc. */ | |
f65fd747 UD |
223 | |
224 | /* | |
fa8d436c UD |
225 | __STD_C should be nonzero if using ANSI-standard C compiler, a C++ |
226 | compiler, or a C compiler sufficiently close to ANSI to get away | |
227 | with it. | |
f65fd747 UD |
228 | */ |
229 | ||
f65fd747 | 230 | #ifndef __STD_C |
fa8d436c | 231 | #if defined(__STDC__) || defined(__cplusplus) |
f65fd747 UD |
232 | #define __STD_C 1 |
233 | #else | |
234 | #define __STD_C 0 | |
a9177ff5 | 235 | #endif |
f65fd747 UD |
236 | #endif /*__STD_C*/ |
237 | ||
fa8d436c UD |
238 | |
239 | /* | |
240 | Void_t* is the pointer type that malloc should say it returns | |
241 | */ | |
242 | ||
f65fd747 | 243 | #ifndef Void_t |
fa8d436c | 244 | #if (__STD_C || defined(WIN32)) |
f65fd747 UD |
245 | #define Void_t void |
246 | #else | |
247 | #define Void_t char | |
248 | #endif | |
249 | #endif /*Void_t*/ | |
250 | ||
251 | #if __STD_C | |
fa8d436c UD |
252 | #include <stddef.h> /* for size_t */ |
253 | #include <stdlib.h> /* for getenv(), abort() */ | |
f65fd747 | 254 | #else |
fa8d436c | 255 | #include <sys/types.h> |
f65fd747 UD |
256 | #endif |
257 | ||
3c6904fb UD |
258 | #include <malloc-machine.h> |
259 | ||
c56da3a3 UD |
260 | #ifdef _LIBC |
261 | #include <stdio-common/_itoa.h> | |
e404fb16 | 262 | #include <bits/wordsize.h> |
c56da3a3 UD |
263 | #endif |
264 | ||
f65fd747 UD |
265 | #ifdef __cplusplus |
266 | extern "C" { | |
267 | #endif | |
268 | ||
fa8d436c | 269 | /* define LACKS_UNISTD_H if your system does not have a <unistd.h>. */ |
f65fd747 | 270 | |
fa8d436c | 271 | /* #define LACKS_UNISTD_H */ |
f65fd747 | 272 | |
fa8d436c UD |
273 | #ifndef LACKS_UNISTD_H |
274 | #include <unistd.h> | |
275 | #endif | |
f65fd747 | 276 | |
fa8d436c UD |
277 | /* define LACKS_SYS_PARAM_H if your system does not have a <sys/param.h>. */ |
278 | ||
279 | /* #define LACKS_SYS_PARAM_H */ | |
280 | ||
281 | ||
282 | #include <stdio.h> /* needed for malloc_stats */ | |
283 | #include <errno.h> /* needed for optional MALLOC_FAILURE_ACTION */ | |
f65fd747 | 284 | |
5d78bb43 UD |
285 | /* For uintptr_t. */ |
286 | #include <stdint.h> | |
f65fd747 | 287 | |
3e030bd5 UD |
288 | /* For va_arg, va_start, va_end. */ |
289 | #include <stdarg.h> | |
290 | ||
6bf4302e UD |
291 | /* For writev and struct iovec. */ |
292 | #include <sys/uio.h> | |
c0f62c56 | 293 | /* For syslog. */ |
54915e9e | 294 | #include <sys/syslog.h> |
6bf4302e | 295 | |
c0f62c56 UD |
296 | /* For various dynamic linking things. */ |
297 | #include <dlfcn.h> | |
298 | ||
299 | ||
fa8d436c UD |
300 | /* |
301 | Debugging: | |
302 | ||
303 | Because freed chunks may be overwritten with bookkeeping fields, this | |
304 | malloc will often die when freed memory is overwritten by user | |
305 | programs. This can be very effective (albeit in an annoying way) | |
306 | in helping track down dangling pointers. | |
307 | ||
308 | If you compile with -DMALLOC_DEBUG, a number of assertion checks are | |
309 | enabled that will catch more memory errors. You probably won't be | |
310 | able to make much sense of the actual assertion errors, but they | |
311 | should help you locate incorrectly overwritten memory. The checking | |
312 | is fairly extensive, and will slow down execution | |
313 | noticeably. Calling malloc_stats or mallinfo with MALLOC_DEBUG set | |
314 | will attempt to check every non-mmapped allocated and free chunk in | |
315 | the course of computing the summmaries. (By nature, mmapped regions | |
316 | cannot be checked very much automatically.) | |
317 | ||
318 | Setting MALLOC_DEBUG may also be helpful if you are trying to modify | |
319 | this code. The assertions in the check routines spell out in more | |
320 | detail the assumptions and invariants underlying the algorithms. | |
321 | ||
322 | Setting MALLOC_DEBUG does NOT provide an automated mechanism for | |
323 | checking that all accesses to malloced memory stay within their | |
324 | bounds. However, there are several add-ons and adaptations of this | |
325 | or other mallocs available that do this. | |
f65fd747 UD |
326 | */ |
327 | ||
328 | #if MALLOC_DEBUG | |
329 | #include <assert.h> | |
330 | #else | |
57449fa3 | 331 | #undef assert |
f65fd747 UD |
332 | #define assert(x) ((void)0) |
333 | #endif | |
334 | ||
335 | ||
336 | /* | |
337 | INTERNAL_SIZE_T is the word-size used for internal bookkeeping | |
fa8d436c UD |
338 | of chunk sizes. |
339 | ||
340 | The default version is the same as size_t. | |
341 | ||
342 | While not strictly necessary, it is best to define this as an | |
343 | unsigned type, even if size_t is a signed type. This may avoid some | |
344 | artificial size limitations on some systems. | |
345 | ||
346 | On a 64-bit machine, you may be able to reduce malloc overhead by | |
347 | defining INTERNAL_SIZE_T to be a 32 bit `unsigned int' at the | |
348 | expense of not being able to handle more than 2^32 of malloced | |
349 | space. If this limitation is acceptable, you are encouraged to set | |
350 | this unless you are on a platform requiring 16byte alignments. In | |
351 | this case the alignment requirements turn out to negate any | |
352 | potential advantages of decreasing size_t word size. | |
353 | ||
354 | Implementors: Beware of the possible combinations of: | |
355 | - INTERNAL_SIZE_T might be signed or unsigned, might be 32 or 64 bits, | |
356 | and might be the same width as int or as long | |
357 | - size_t might have different width and signedness as INTERNAL_SIZE_T | |
358 | - int and long might be 32 or 64 bits, and might be the same width | |
359 | To deal with this, most comparisons and difference computations | |
360 | among INTERNAL_SIZE_Ts should cast them to unsigned long, being | |
361 | aware of the fact that casting an unsigned int to a wider long does | |
362 | not sign-extend. (This also makes checking for negative numbers | |
363 | awkward.) Some of these casts result in harmless compiler warnings | |
364 | on some systems. | |
f65fd747 UD |
365 | */ |
366 | ||
367 | #ifndef INTERNAL_SIZE_T | |
368 | #define INTERNAL_SIZE_T size_t | |
369 | #endif | |
370 | ||
fa8d436c UD |
371 | /* The corresponding word size */ |
372 | #define SIZE_SZ (sizeof(INTERNAL_SIZE_T)) | |
373 | ||
374 | ||
375 | /* | |
376 | MALLOC_ALIGNMENT is the minimum alignment for malloc'ed chunks. | |
377 | It must be a power of two at least 2 * SIZE_SZ, even on machines | |
378 | for which smaller alignments would suffice. It may be defined as | |
379 | larger than this though. Note however that code and data structures | |
380 | are optimized for the case of 8-byte alignment. | |
381 | */ | |
382 | ||
383 | ||
384 | #ifndef MALLOC_ALIGNMENT | |
7d013a64 RM |
385 | /* XXX This is the correct definition. It differs from 2*SIZE_SZ only on |
386 | powerpc32. For the time being, changing this is causing more | |
387 | compatibility problems due to malloc_get_state/malloc_set_state than | |
388 | will returning blocks not adequately aligned for long double objects | |
16a10468 RM |
389 | under -mlong-double-128. |
390 | ||
073f560e UD |
391 | #define MALLOC_ALIGNMENT (2 * SIZE_SZ < __alignof__ (long double) \ |
392 | ? __alignof__ (long double) : 2 * SIZE_SZ) | |
7d013a64 RM |
393 | */ |
394 | #define MALLOC_ALIGNMENT (2 * SIZE_SZ) | |
fa8d436c UD |
395 | #endif |
396 | ||
397 | /* The corresponding bit mask value */ | |
398 | #define MALLOC_ALIGN_MASK (MALLOC_ALIGNMENT - 1) | |
399 | ||
400 | ||
401 | ||
402 | /* | |
403 | REALLOC_ZERO_BYTES_FREES should be set if a call to | |
404 | realloc with zero bytes should be the same as a call to free. | |
405 | This is required by the C standard. Otherwise, since this malloc | |
406 | returns a unique pointer for malloc(0), so does realloc(p, 0). | |
407 | */ | |
408 | ||
409 | #ifndef REALLOC_ZERO_BYTES_FREES | |
410 | #define REALLOC_ZERO_BYTES_FREES 1 | |
411 | #endif | |
412 | ||
413 | /* | |
414 | TRIM_FASTBINS controls whether free() of a very small chunk can | |
415 | immediately lead to trimming. Setting to true (1) can reduce memory | |
416 | footprint, but will almost always slow down programs that use a lot | |
417 | of small chunks. | |
418 | ||
419 | Define this only if you are willing to give up some speed to more | |
420 | aggressively reduce system-level memory footprint when releasing | |
421 | memory in programs that use many small chunks. You can get | |
422 | essentially the same effect by setting MXFAST to 0, but this can | |
423 | lead to even greater slowdowns in programs using many small chunks. | |
424 | TRIM_FASTBINS is an in-between compile-time option, that disables | |
425 | only those chunks bordering topmost memory from being placed in | |
426 | fastbins. | |
427 | */ | |
428 | ||
429 | #ifndef TRIM_FASTBINS | |
430 | #define TRIM_FASTBINS 0 | |
431 | #endif | |
432 | ||
433 | ||
f65fd747 | 434 | /* |
fa8d436c | 435 | USE_DL_PREFIX will prefix all public routines with the string 'dl'. |
a9177ff5 | 436 | This is necessary when you only want to use this malloc in one part |
fa8d436c UD |
437 | of a program, using your regular system malloc elsewhere. |
438 | */ | |
439 | ||
440 | /* #define USE_DL_PREFIX */ | |
441 | ||
442 | ||
a9177ff5 | 443 | /* |
fa8d436c UD |
444 | Two-phase name translation. |
445 | All of the actual routines are given mangled names. | |
446 | When wrappers are used, they become the public callable versions. | |
447 | When DL_PREFIX is used, the callable names are prefixed. | |
f65fd747 UD |
448 | */ |
449 | ||
fa8d436c UD |
450 | #ifdef USE_DL_PREFIX |
451 | #define public_cALLOc dlcalloc | |
452 | #define public_fREe dlfree | |
453 | #define public_cFREe dlcfree | |
454 | #define public_mALLOc dlmalloc | |
455 | #define public_mEMALIGn dlmemalign | |
456 | #define public_rEALLOc dlrealloc | |
457 | #define public_vALLOc dlvalloc | |
458 | #define public_pVALLOc dlpvalloc | |
459 | #define public_mALLINFo dlmallinfo | |
460 | #define public_mALLOPt dlmallopt | |
461 | #define public_mTRIm dlmalloc_trim | |
462 | #define public_mSTATs dlmalloc_stats | |
463 | #define public_mUSABLe dlmalloc_usable_size | |
464 | #define public_iCALLOc dlindependent_calloc | |
465 | #define public_iCOMALLOc dlindependent_comalloc | |
466 | #define public_gET_STATe dlget_state | |
467 | #define public_sET_STATe dlset_state | |
468 | #else /* USE_DL_PREFIX */ | |
469 | #ifdef _LIBC | |
470 | ||
471 | /* Special defines for the GNU C library. */ | |
472 | #define public_cALLOc __libc_calloc | |
473 | #define public_fREe __libc_free | |
474 | #define public_cFREe __libc_cfree | |
475 | #define public_mALLOc __libc_malloc | |
476 | #define public_mEMALIGn __libc_memalign | |
477 | #define public_rEALLOc __libc_realloc | |
478 | #define public_vALLOc __libc_valloc | |
479 | #define public_pVALLOc __libc_pvalloc | |
480 | #define public_mALLINFo __libc_mallinfo | |
481 | #define public_mALLOPt __libc_mallopt | |
482 | #define public_mTRIm __malloc_trim | |
483 | #define public_mSTATs __malloc_stats | |
484 | #define public_mUSABLe __malloc_usable_size | |
485 | #define public_iCALLOc __libc_independent_calloc | |
486 | #define public_iCOMALLOc __libc_independent_comalloc | |
487 | #define public_gET_STATe __malloc_get_state | |
488 | #define public_sET_STATe __malloc_set_state | |
489 | #define malloc_getpagesize __getpagesize() | |
490 | #define open __open | |
491 | #define mmap __mmap | |
492 | #define munmap __munmap | |
493 | #define mremap __mremap | |
494 | #define mprotect __mprotect | |
495 | #define MORECORE (*__morecore) | |
496 | #define MORECORE_FAILURE 0 | |
497 | ||
498 | Void_t * __default_morecore (ptrdiff_t); | |
499 | Void_t *(*__morecore)(ptrdiff_t) = __default_morecore; | |
f65fd747 | 500 | |
fa8d436c UD |
501 | #else /* !_LIBC */ |
502 | #define public_cALLOc calloc | |
503 | #define public_fREe free | |
504 | #define public_cFREe cfree | |
505 | #define public_mALLOc malloc | |
506 | #define public_mEMALIGn memalign | |
507 | #define public_rEALLOc realloc | |
508 | #define public_vALLOc valloc | |
509 | #define public_pVALLOc pvalloc | |
510 | #define public_mALLINFo mallinfo | |
511 | #define public_mALLOPt mallopt | |
512 | #define public_mTRIm malloc_trim | |
513 | #define public_mSTATs malloc_stats | |
514 | #define public_mUSABLe malloc_usable_size | |
515 | #define public_iCALLOc independent_calloc | |
516 | #define public_iCOMALLOc independent_comalloc | |
517 | #define public_gET_STATe malloc_get_state | |
518 | #define public_sET_STATe malloc_set_state | |
519 | #endif /* _LIBC */ | |
520 | #endif /* USE_DL_PREFIX */ | |
f65fd747 | 521 | |
d9af917d UD |
522 | #ifndef _LIBC |
523 | #define __builtin_expect(expr, val) (expr) | |
3ba06713 UD |
524 | |
525 | #define fwrite(buf, size, count, fp) _IO_fwrite (buf, size, count, fp) | |
d9af917d | 526 | #endif |
f65fd747 UD |
527 | |
528 | /* | |
529 | HAVE_MEMCPY should be defined if you are not otherwise using | |
530 | ANSI STD C, but still have memcpy and memset in your C library | |
531 | and want to use them in calloc and realloc. Otherwise simple | |
fa8d436c | 532 | macro versions are defined below. |
f65fd747 UD |
533 | |
534 | USE_MEMCPY should be defined as 1 if you actually want to | |
535 | have memset and memcpy called. People report that the macro | |
fa8d436c | 536 | versions are faster than libc versions on some systems. |
a9177ff5 | 537 | |
fa8d436c UD |
538 | Even if USE_MEMCPY is set to 1, loops to copy/clear small chunks |
539 | (of <= 36 bytes) are manually unrolled in realloc and calloc. | |
f65fd747 UD |
540 | */ |
541 | ||
fa8d436c | 542 | #define HAVE_MEMCPY |
f65fd747 UD |
543 | |
544 | #ifndef USE_MEMCPY | |
545 | #ifdef HAVE_MEMCPY | |
546 | #define USE_MEMCPY 1 | |
547 | #else | |
548 | #define USE_MEMCPY 0 | |
549 | #endif | |
550 | #endif | |
551 | ||
fa8d436c | 552 | |
f65fd747 UD |
553 | #if (__STD_C || defined(HAVE_MEMCPY)) |
554 | ||
c2afe833 RM |
555 | #ifdef _LIBC |
556 | # include <string.h> | |
557 | #else | |
fa8d436c UD |
558 | #ifdef WIN32 |
559 | /* On Win32 memset and memcpy are already declared in windows.h */ | |
560 | #else | |
f65fd747 UD |
561 | #if __STD_C |
562 | void* memset(void*, int, size_t); | |
563 | void* memcpy(void*, const void*, size_t); | |
564 | #else | |
565 | Void_t* memset(); | |
566 | Void_t* memcpy(); | |
fa8d436c | 567 | #endif |
f65fd747 UD |
568 | #endif |
569 | #endif | |
c2afe833 | 570 | #endif |
f65fd747 | 571 | |
fa8d436c UD |
572 | /* |
573 | MALLOC_FAILURE_ACTION is the action to take before "return 0" when | |
574 | malloc fails to be able to return memory, either because memory is | |
575 | exhausted or because of illegal arguments. | |
a9177ff5 RM |
576 | |
577 | By default, sets errno if running on STD_C platform, else does nothing. | |
fa8d436c | 578 | */ |
09f5e163 | 579 | |
fa8d436c UD |
580 | #ifndef MALLOC_FAILURE_ACTION |
581 | #if __STD_C | |
582 | #define MALLOC_FAILURE_ACTION \ | |
583 | errno = ENOMEM; | |
f65fd747 | 584 | |
fa8d436c UD |
585 | #else |
586 | #define MALLOC_FAILURE_ACTION | |
587 | #endif | |
588 | #endif | |
f65fd747 | 589 | |
fa8d436c UD |
590 | /* |
591 | MORECORE-related declarations. By default, rely on sbrk | |
592 | */ | |
09f5e163 | 593 | |
f65fd747 | 594 | |
fa8d436c UD |
595 | #ifdef LACKS_UNISTD_H |
596 | #if !defined(__FreeBSD__) && !defined(__OpenBSD__) && !defined(__NetBSD__) | |
597 | #if __STD_C | |
598 | extern Void_t* sbrk(ptrdiff_t); | |
599 | #else | |
600 | extern Void_t* sbrk(); | |
601 | #endif | |
602 | #endif | |
603 | #endif | |
f65fd747 | 604 | |
fa8d436c UD |
605 | /* |
606 | MORECORE is the name of the routine to call to obtain more memory | |
607 | from the system. See below for general guidance on writing | |
608 | alternative MORECORE functions, as well as a version for WIN32 and a | |
609 | sample version for pre-OSX macos. | |
610 | */ | |
f65fd747 | 611 | |
fa8d436c UD |
612 | #ifndef MORECORE |
613 | #define MORECORE sbrk | |
614 | #endif | |
f65fd747 | 615 | |
fa8d436c UD |
616 | /* |
617 | MORECORE_FAILURE is the value returned upon failure of MORECORE | |
618 | as well as mmap. Since it cannot be an otherwise valid memory address, | |
619 | and must reflect values of standard sys calls, you probably ought not | |
620 | try to redefine it. | |
621 | */ | |
09f5e163 | 622 | |
fa8d436c UD |
623 | #ifndef MORECORE_FAILURE |
624 | #define MORECORE_FAILURE (-1) | |
625 | #endif | |
626 | ||
627 | /* | |
628 | If MORECORE_CONTIGUOUS is true, take advantage of fact that | |
629 | consecutive calls to MORECORE with positive arguments always return | |
630 | contiguous increasing addresses. This is true of unix sbrk. Even | |
631 | if not defined, when regions happen to be contiguous, malloc will | |
632 | permit allocations spanning regions obtained from different | |
633 | calls. But defining this when applicable enables some stronger | |
634 | consistency checks and space efficiencies. | |
635 | */ | |
f65fd747 | 636 | |
fa8d436c UD |
637 | #ifndef MORECORE_CONTIGUOUS |
638 | #define MORECORE_CONTIGUOUS 1 | |
f65fd747 UD |
639 | #endif |
640 | ||
fa8d436c UD |
641 | /* |
642 | Define MORECORE_CANNOT_TRIM if your version of MORECORE | |
643 | cannot release space back to the system when given negative | |
644 | arguments. This is generally necessary only if you are using | |
645 | a hand-crafted MORECORE function that cannot handle negative arguments. | |
646 | */ | |
647 | ||
648 | /* #define MORECORE_CANNOT_TRIM */ | |
f65fd747 | 649 | |
fa8d436c UD |
650 | /* MORECORE_CLEARS (default 1) |
651 | The degree to which the routine mapped to MORECORE zeroes out | |
652 | memory: never (0), only for newly allocated space (1) or always | |
653 | (2). The distinction between (1) and (2) is necessary because on | |
654 | some systems, if the application first decrements and then | |
655 | increments the break value, the contents of the reallocated space | |
656 | are unspecified. | |
657 | */ | |
658 | ||
659 | #ifndef MORECORE_CLEARS | |
660 | #define MORECORE_CLEARS 1 | |
7cabd57c UD |
661 | #endif |
662 | ||
fa8d436c | 663 | |
f65fd747 | 664 | /* |
fa8d436c UD |
665 | Define HAVE_MMAP as true to optionally make malloc() use mmap() to |
666 | allocate very large blocks. These will be returned to the | |
667 | operating system immediately after a free(). Also, if mmap | |
668 | is available, it is used as a backup strategy in cases where | |
669 | MORECORE fails to provide space from system. | |
670 | ||
671 | This malloc is best tuned to work with mmap for large requests. | |
672 | If you do not have mmap, operations involving very large chunks (1MB | |
673 | or so) may be slower than you'd like. | |
f65fd747 UD |
674 | */ |
675 | ||
676 | #ifndef HAVE_MMAP | |
fa8d436c UD |
677 | #define HAVE_MMAP 1 |
678 | ||
a9177ff5 | 679 | /* |
fa8d436c UD |
680 | Standard unix mmap using /dev/zero clears memory so calloc doesn't |
681 | need to. | |
682 | */ | |
683 | ||
684 | #ifndef MMAP_CLEARS | |
685 | #define MMAP_CLEARS 1 | |
686 | #endif | |
687 | ||
688 | #else /* no mmap */ | |
689 | #ifndef MMAP_CLEARS | |
690 | #define MMAP_CLEARS 0 | |
691 | #endif | |
692 | #endif | |
693 | ||
694 | ||
a9177ff5 | 695 | /* |
fa8d436c UD |
696 | MMAP_AS_MORECORE_SIZE is the minimum mmap size argument to use if |
697 | sbrk fails, and mmap is used as a backup (which is done only if | |
698 | HAVE_MMAP). The value must be a multiple of page size. This | |
699 | backup strategy generally applies only when systems have "holes" in | |
700 | address space, so sbrk cannot perform contiguous expansion, but | |
701 | there is still space available on system. On systems for which | |
702 | this is known to be useful (i.e. most linux kernels), this occurs | |
703 | only when programs allocate huge amounts of memory. Between this, | |
704 | and the fact that mmap regions tend to be limited, the size should | |
705 | be large, to avoid too many mmap calls and thus avoid running out | |
706 | of kernel resources. | |
707 | */ | |
708 | ||
709 | #ifndef MMAP_AS_MORECORE_SIZE | |
710 | #define MMAP_AS_MORECORE_SIZE (1024 * 1024) | |
f65fd747 UD |
711 | #endif |
712 | ||
713 | /* | |
714 | Define HAVE_MREMAP to make realloc() use mremap() to re-allocate | |
715 | large blocks. This is currently only possible on Linux with | |
716 | kernel versions newer than 1.3.77. | |
717 | */ | |
718 | ||
719 | #ifndef HAVE_MREMAP | |
fa8d436c UD |
720 | #ifdef linux |
721 | #define HAVE_MREMAP 1 | |
722 | #else | |
723 | #define HAVE_MREMAP 0 | |
f65fd747 UD |
724 | #endif |
725 | ||
fa8d436c UD |
726 | #endif /* HAVE_MMAP */ |
727 | ||
e9b3e3c5 UD |
728 | /* Define USE_ARENAS to enable support for multiple `arenas'. These |
729 | are allocated using mmap(), are necessary for threads and | |
730 | occasionally useful to overcome address space limitations affecting | |
731 | sbrk(). */ | |
732 | ||
733 | #ifndef USE_ARENAS | |
734 | #define USE_ARENAS HAVE_MMAP | |
735 | #endif | |
736 | ||
f65fd747 UD |
737 | |
738 | /* | |
fa8d436c UD |
739 | The system page size. To the extent possible, this malloc manages |
740 | memory from the system in page-size units. Note that this value is | |
741 | cached during initialization into a field of malloc_state. So even | |
742 | if malloc_getpagesize is a function, it is only called once. | |
743 | ||
744 | The following mechanics for getpagesize were adapted from bsd/gnu | |
745 | getpagesize.h. If none of the system-probes here apply, a value of | |
746 | 4096 is used, which should be OK: If they don't apply, then using | |
747 | the actual value probably doesn't impact performance. | |
f65fd747 UD |
748 | */ |
749 | ||
fa8d436c | 750 | |
f65fd747 | 751 | #ifndef malloc_getpagesize |
fa8d436c UD |
752 | |
753 | #ifndef LACKS_UNISTD_H | |
754 | # include <unistd.h> | |
755 | #endif | |
756 | ||
f65fd747 UD |
757 | # ifdef _SC_PAGESIZE /* some SVR4 systems omit an underscore */ |
758 | # ifndef _SC_PAGE_SIZE | |
759 | # define _SC_PAGE_SIZE _SC_PAGESIZE | |
760 | # endif | |
761 | # endif | |
fa8d436c | 762 | |
f65fd747 UD |
763 | # ifdef _SC_PAGE_SIZE |
764 | # define malloc_getpagesize sysconf(_SC_PAGE_SIZE) | |
765 | # else | |
766 | # if defined(BSD) || defined(DGUX) || defined(HAVE_GETPAGESIZE) | |
767 | extern size_t getpagesize(); | |
768 | # define malloc_getpagesize getpagesize() | |
769 | # else | |
fa8d436c | 770 | # ifdef WIN32 /* use supplied emulation of getpagesize */ |
a9177ff5 | 771 | # define malloc_getpagesize getpagesize() |
f65fd747 | 772 | # else |
fa8d436c UD |
773 | # ifndef LACKS_SYS_PARAM_H |
774 | # include <sys/param.h> | |
775 | # endif | |
776 | # ifdef EXEC_PAGESIZE | |
777 | # define malloc_getpagesize EXEC_PAGESIZE | |
f65fd747 | 778 | # else |
fa8d436c UD |
779 | # ifdef NBPG |
780 | # ifndef CLSIZE | |
781 | # define malloc_getpagesize NBPG | |
782 | # else | |
783 | # define malloc_getpagesize (NBPG * CLSIZE) | |
784 | # endif | |
f65fd747 | 785 | # else |
fa8d436c UD |
786 | # ifdef NBPC |
787 | # define malloc_getpagesize NBPC | |
f65fd747 | 788 | # else |
fa8d436c UD |
789 | # ifdef PAGESIZE |
790 | # define malloc_getpagesize PAGESIZE | |
791 | # else /* just guess */ | |
a9177ff5 | 792 | # define malloc_getpagesize (4096) |
fa8d436c | 793 | # endif |
f65fd747 UD |
794 | # endif |
795 | # endif | |
796 | # endif | |
797 | # endif | |
798 | # endif | |
799 | # endif | |
800 | #endif | |
801 | ||
f65fd747 | 802 | /* |
f65fd747 | 803 | This version of malloc supports the standard SVID/XPG mallinfo |
fa8d436c UD |
804 | routine that returns a struct containing usage properties and |
805 | statistics. It should work on any SVID/XPG compliant system that has | |
806 | a /usr/include/malloc.h defining struct mallinfo. (If you'd like to | |
807 | install such a thing yourself, cut out the preliminary declarations | |
808 | as described above and below and save them in a malloc.h file. But | |
809 | there's no compelling reason to bother to do this.) | |
f65fd747 UD |
810 | |
811 | The main declaration needed is the mallinfo struct that is returned | |
812 | (by-copy) by mallinfo(). The SVID/XPG malloinfo struct contains a | |
fa8d436c UD |
813 | bunch of fields that are not even meaningful in this version of |
814 | malloc. These fields are are instead filled by mallinfo() with | |
815 | other numbers that might be of interest. | |
f65fd747 UD |
816 | |
817 | HAVE_USR_INCLUDE_MALLOC_H should be set if you have a | |
818 | /usr/include/malloc.h file that includes a declaration of struct | |
819 | mallinfo. If so, it is included; else an SVID2/XPG2 compliant | |
820 | version is declared below. These must be precisely the same for | |
fa8d436c UD |
821 | mallinfo() to work. The original SVID version of this struct, |
822 | defined on most systems with mallinfo, declares all fields as | |
823 | ints. But some others define as unsigned long. If your system | |
824 | defines the fields using a type of different width than listed here, | |
825 | you must #include your system version and #define | |
826 | HAVE_USR_INCLUDE_MALLOC_H. | |
f65fd747 UD |
827 | */ |
828 | ||
829 | /* #define HAVE_USR_INCLUDE_MALLOC_H */ | |
830 | ||
fa8d436c UD |
831 | #ifdef HAVE_USR_INCLUDE_MALLOC_H |
832 | #include "/usr/include/malloc.h" | |
f65fd747 UD |
833 | #endif |
834 | ||
f65fd747 | 835 | |
fa8d436c | 836 | /* ---------- description of public routines ------------ */ |
f65fd747 UD |
837 | |
838 | /* | |
fa8d436c UD |
839 | malloc(size_t n) |
840 | Returns a pointer to a newly allocated chunk of at least n bytes, or null | |
841 | if no space is available. Additionally, on failure, errno is | |
842 | set to ENOMEM on ANSI C systems. | |
843 | ||
844 | If n is zero, malloc returns a minumum-sized chunk. (The minimum | |
845 | size is 16 bytes on most 32bit systems, and 24 or 32 bytes on 64bit | |
846 | systems.) On most systems, size_t is an unsigned type, so calls | |
847 | with negative arguments are interpreted as requests for huge amounts | |
848 | of space, which will often fail. The maximum supported value of n | |
849 | differs across systems, but is in all cases less than the maximum | |
850 | representable value of a size_t. | |
f65fd747 | 851 | */ |
fa8d436c UD |
852 | #if __STD_C |
853 | Void_t* public_mALLOc(size_t); | |
854 | #else | |
855 | Void_t* public_mALLOc(); | |
856 | #endif | |
aa420660 UD |
857 | #ifdef libc_hidden_proto |
858 | libc_hidden_proto (public_mALLOc) | |
859 | #endif | |
f65fd747 | 860 | |
fa8d436c UD |
861 | /* |
862 | free(Void_t* p) | |
863 | Releases the chunk of memory pointed to by p, that had been previously | |
864 | allocated using malloc or a related routine such as realloc. | |
865 | It has no effect if p is null. It can have arbitrary (i.e., bad!) | |
866 | effects if p has already been freed. | |
867 | ||
868 | Unless disabled (using mallopt), freeing very large spaces will | |
869 | when possible, automatically trigger operations that give | |
870 | back unused memory to the system, thus reducing program footprint. | |
871 | */ | |
872 | #if __STD_C | |
873 | void public_fREe(Void_t*); | |
874 | #else | |
875 | void public_fREe(); | |
876 | #endif | |
aa420660 UD |
877 | #ifdef libc_hidden_proto |
878 | libc_hidden_proto (public_fREe) | |
879 | #endif | |
f65fd747 | 880 | |
fa8d436c UD |
881 | /* |
882 | calloc(size_t n_elements, size_t element_size); | |
883 | Returns a pointer to n_elements * element_size bytes, with all locations | |
884 | set to zero. | |
885 | */ | |
886 | #if __STD_C | |
887 | Void_t* public_cALLOc(size_t, size_t); | |
888 | #else | |
889 | Void_t* public_cALLOc(); | |
f65fd747 UD |
890 | #endif |
891 | ||
892 | /* | |
fa8d436c UD |
893 | realloc(Void_t* p, size_t n) |
894 | Returns a pointer to a chunk of size n that contains the same data | |
895 | as does chunk p up to the minimum of (n, p's size) bytes, or null | |
a9177ff5 | 896 | if no space is available. |
f65fd747 | 897 | |
fa8d436c UD |
898 | The returned pointer may or may not be the same as p. The algorithm |
899 | prefers extending p when possible, otherwise it employs the | |
900 | equivalent of a malloc-copy-free sequence. | |
f65fd747 | 901 | |
a9177ff5 | 902 | If p is null, realloc is equivalent to malloc. |
f65fd747 | 903 | |
fa8d436c UD |
904 | If space is not available, realloc returns null, errno is set (if on |
905 | ANSI) and p is NOT freed. | |
f65fd747 | 906 | |
fa8d436c UD |
907 | if n is for fewer bytes than already held by p, the newly unused |
908 | space is lopped off and freed if possible. Unless the #define | |
909 | REALLOC_ZERO_BYTES_FREES is set, realloc with a size argument of | |
910 | zero (re)allocates a minimum-sized chunk. | |
f65fd747 | 911 | |
fa8d436c UD |
912 | Large chunks that were internally obtained via mmap will always |
913 | be reallocated using malloc-copy-free sequences unless | |
914 | the system supports MREMAP (currently only linux). | |
f65fd747 | 915 | |
fa8d436c UD |
916 | The old unix realloc convention of allowing the last-free'd chunk |
917 | to be used as an argument to realloc is not supported. | |
f65fd747 | 918 | */ |
fa8d436c UD |
919 | #if __STD_C |
920 | Void_t* public_rEALLOc(Void_t*, size_t); | |
921 | #else | |
922 | Void_t* public_rEALLOc(); | |
923 | #endif | |
aa420660 UD |
924 | #ifdef libc_hidden_proto |
925 | libc_hidden_proto (public_rEALLOc) | |
926 | #endif | |
f65fd747 | 927 | |
fa8d436c UD |
928 | /* |
929 | memalign(size_t alignment, size_t n); | |
930 | Returns a pointer to a newly allocated chunk of n bytes, aligned | |
931 | in accord with the alignment argument. | |
932 | ||
933 | The alignment argument should be a power of two. If the argument is | |
934 | not a power of two, the nearest greater power is used. | |
935 | 8-byte alignment is guaranteed by normal malloc calls, so don't | |
936 | bother calling memalign with an argument of 8 or less. | |
937 | ||
938 | Overreliance on memalign is a sure way to fragment space. | |
939 | */ | |
940 | #if __STD_C | |
941 | Void_t* public_mEMALIGn(size_t, size_t); | |
942 | #else | |
943 | Void_t* public_mEMALIGn(); | |
f65fd747 | 944 | #endif |
aa420660 UD |
945 | #ifdef libc_hidden_proto |
946 | libc_hidden_proto (public_mEMALIGn) | |
947 | #endif | |
f65fd747 UD |
948 | |
949 | /* | |
fa8d436c UD |
950 | valloc(size_t n); |
951 | Equivalent to memalign(pagesize, n), where pagesize is the page | |
952 | size of the system. If the pagesize is unknown, 4096 is used. | |
953 | */ | |
954 | #if __STD_C | |
955 | Void_t* public_vALLOc(size_t); | |
956 | #else | |
957 | Void_t* public_vALLOc(); | |
958 | #endif | |
959 | ||
f65fd747 | 960 | |
f65fd747 | 961 | |
fa8d436c UD |
962 | /* |
963 | mallopt(int parameter_number, int parameter_value) | |
964 | Sets tunable parameters The format is to provide a | |
965 | (parameter-number, parameter-value) pair. mallopt then sets the | |
966 | corresponding parameter to the argument value if it can (i.e., so | |
967 | long as the value is meaningful), and returns 1 if successful else | |
968 | 0. SVID/XPG/ANSI defines four standard param numbers for mallopt, | |
969 | normally defined in malloc.h. Only one of these (M_MXFAST) is used | |
970 | in this malloc. The others (M_NLBLKS, M_GRAIN, M_KEEP) don't apply, | |
971 | so setting them has no effect. But this malloc also supports four | |
972 | other options in mallopt. See below for details. Briefly, supported | |
973 | parameters are as follows (listed defaults are for "typical" | |
974 | configurations). | |
975 | ||
976 | Symbol param # default allowed param values | |
977 | M_MXFAST 1 64 0-80 (0 disables fastbins) | |
978 | M_TRIM_THRESHOLD -1 128*1024 any (-1U disables trimming) | |
a9177ff5 | 979 | M_TOP_PAD -2 0 any |
fa8d436c UD |
980 | M_MMAP_THRESHOLD -3 128*1024 any (or 0 if no MMAP support) |
981 | M_MMAP_MAX -4 65536 any (0 disables use of mmap) | |
982 | */ | |
983 | #if __STD_C | |
984 | int public_mALLOPt(int, int); | |
985 | #else | |
986 | int public_mALLOPt(); | |
987 | #endif | |
988 | ||
989 | ||
990 | /* | |
991 | mallinfo() | |
992 | Returns (by copy) a struct containing various summary statistics: | |
993 | ||
a9177ff5 RM |
994 | arena: current total non-mmapped bytes allocated from system |
995 | ordblks: the number of free chunks | |
fa8d436c UD |
996 | smblks: the number of fastbin blocks (i.e., small chunks that |
997 | have been freed but not use resused or consolidated) | |
a9177ff5 RM |
998 | hblks: current number of mmapped regions |
999 | hblkhd: total bytes held in mmapped regions | |
fa8d436c UD |
1000 | usmblks: the maximum total allocated space. This will be greater |
1001 | than current total if trimming has occurred. | |
a9177ff5 | 1002 | fsmblks: total bytes held in fastbin blocks |
fa8d436c | 1003 | uordblks: current total allocated space (normal or mmapped) |
a9177ff5 | 1004 | fordblks: total free space |
fa8d436c UD |
1005 | keepcost: the maximum number of bytes that could ideally be released |
1006 | back to system via malloc_trim. ("ideally" means that | |
1007 | it ignores page restrictions etc.) | |
1008 | ||
1009 | Because these fields are ints, but internal bookkeeping may | |
a9177ff5 | 1010 | be kept as longs, the reported values may wrap around zero and |
fa8d436c UD |
1011 | thus be inaccurate. |
1012 | */ | |
1013 | #if __STD_C | |
1014 | struct mallinfo public_mALLINFo(void); | |
1015 | #else | |
1016 | struct mallinfo public_mALLINFo(); | |
1017 | #endif | |
f65fd747 | 1018 | |
88764ae2 | 1019 | #ifndef _LIBC |
fa8d436c UD |
1020 | /* |
1021 | independent_calloc(size_t n_elements, size_t element_size, Void_t* chunks[]); | |
1022 | ||
1023 | independent_calloc is similar to calloc, but instead of returning a | |
1024 | single cleared space, it returns an array of pointers to n_elements | |
1025 | independent elements that can hold contents of size elem_size, each | |
1026 | of which starts out cleared, and can be independently freed, | |
1027 | realloc'ed etc. The elements are guaranteed to be adjacently | |
1028 | allocated (this is not guaranteed to occur with multiple callocs or | |
1029 | mallocs), which may also improve cache locality in some | |
1030 | applications. | |
1031 | ||
1032 | The "chunks" argument is optional (i.e., may be null, which is | |
1033 | probably the most typical usage). If it is null, the returned array | |
1034 | is itself dynamically allocated and should also be freed when it is | |
1035 | no longer needed. Otherwise, the chunks array must be of at least | |
1036 | n_elements in length. It is filled in with the pointers to the | |
1037 | chunks. | |
1038 | ||
1039 | In either case, independent_calloc returns this pointer array, or | |
1040 | null if the allocation failed. If n_elements is zero and "chunks" | |
1041 | is null, it returns a chunk representing an array with zero elements | |
1042 | (which should be freed if not wanted). | |
1043 | ||
1044 | Each element must be individually freed when it is no longer | |
1045 | needed. If you'd like to instead be able to free all at once, you | |
1046 | should instead use regular calloc and assign pointers into this | |
1047 | space to represent elements. (In this case though, you cannot | |
1048 | independently free elements.) | |
a9177ff5 | 1049 | |
fa8d436c UD |
1050 | independent_calloc simplifies and speeds up implementations of many |
1051 | kinds of pools. It may also be useful when constructing large data | |
1052 | structures that initially have a fixed number of fixed-sized nodes, | |
1053 | but the number is not known at compile time, and some of the nodes | |
1054 | may later need to be freed. For example: | |
1055 | ||
1056 | struct Node { int item; struct Node* next; }; | |
a9177ff5 | 1057 | |
fa8d436c UD |
1058 | struct Node* build_list() { |
1059 | struct Node** pool; | |
1060 | int n = read_number_of_nodes_needed(); | |
1061 | if (n <= 0) return 0; | |
1062 | pool = (struct Node**)(independent_calloc(n, sizeof(struct Node), 0); | |
a9177ff5 RM |
1063 | if (pool == 0) die(); |
1064 | // organize into a linked list... | |
fa8d436c | 1065 | struct Node* first = pool[0]; |
a9177ff5 | 1066 | for (i = 0; i < n-1; ++i) |
fa8d436c UD |
1067 | pool[i]->next = pool[i+1]; |
1068 | free(pool); // Can now free the array (or not, if it is needed later) | |
1069 | return first; | |
1070 | } | |
1071 | */ | |
1072 | #if __STD_C | |
1073 | Void_t** public_iCALLOc(size_t, size_t, Void_t**); | |
1074 | #else | |
1075 | Void_t** public_iCALLOc(); | |
1076 | #endif | |
f65fd747 | 1077 | |
fa8d436c UD |
1078 | /* |
1079 | independent_comalloc(size_t n_elements, size_t sizes[], Void_t* chunks[]); | |
1080 | ||
1081 | independent_comalloc allocates, all at once, a set of n_elements | |
1082 | chunks with sizes indicated in the "sizes" array. It returns | |
1083 | an array of pointers to these elements, each of which can be | |
1084 | independently freed, realloc'ed etc. The elements are guaranteed to | |
1085 | be adjacently allocated (this is not guaranteed to occur with | |
1086 | multiple callocs or mallocs), which may also improve cache locality | |
1087 | in some applications. | |
1088 | ||
1089 | The "chunks" argument is optional (i.e., may be null). If it is null | |
1090 | the returned array is itself dynamically allocated and should also | |
1091 | be freed when it is no longer needed. Otherwise, the chunks array | |
1092 | must be of at least n_elements in length. It is filled in with the | |
1093 | pointers to the chunks. | |
1094 | ||
1095 | In either case, independent_comalloc returns this pointer array, or | |
1096 | null if the allocation failed. If n_elements is zero and chunks is | |
1097 | null, it returns a chunk representing an array with zero elements | |
1098 | (which should be freed if not wanted). | |
a9177ff5 | 1099 | |
fa8d436c UD |
1100 | Each element must be individually freed when it is no longer |
1101 | needed. If you'd like to instead be able to free all at once, you | |
1102 | should instead use a single regular malloc, and assign pointers at | |
a9177ff5 | 1103 | particular offsets in the aggregate space. (In this case though, you |
fa8d436c UD |
1104 | cannot independently free elements.) |
1105 | ||
1106 | independent_comallac differs from independent_calloc in that each | |
1107 | element may have a different size, and also that it does not | |
1108 | automatically clear elements. | |
1109 | ||
1110 | independent_comalloc can be used to speed up allocation in cases | |
1111 | where several structs or objects must always be allocated at the | |
1112 | same time. For example: | |
1113 | ||
1114 | struct Head { ... } | |
1115 | struct Foot { ... } | |
1116 | ||
1117 | void send_message(char* msg) { | |
1118 | int msglen = strlen(msg); | |
1119 | size_t sizes[3] = { sizeof(struct Head), msglen, sizeof(struct Foot) }; | |
1120 | void* chunks[3]; | |
1121 | if (independent_comalloc(3, sizes, chunks) == 0) | |
1122 | die(); | |
1123 | struct Head* head = (struct Head*)(chunks[0]); | |
1124 | char* body = (char*)(chunks[1]); | |
1125 | struct Foot* foot = (struct Foot*)(chunks[2]); | |
1126 | // ... | |
1127 | } | |
f65fd747 | 1128 | |
fa8d436c UD |
1129 | In general though, independent_comalloc is worth using only for |
1130 | larger values of n_elements. For small values, you probably won't | |
1131 | detect enough difference from series of malloc calls to bother. | |
f65fd747 | 1132 | |
fa8d436c UD |
1133 | Overuse of independent_comalloc can increase overall memory usage, |
1134 | since it cannot reuse existing noncontiguous small chunks that | |
1135 | might be available for some of the elements. | |
f65fd747 | 1136 | */ |
fa8d436c UD |
1137 | #if __STD_C |
1138 | Void_t** public_iCOMALLOc(size_t, size_t*, Void_t**); | |
1139 | #else | |
1140 | Void_t** public_iCOMALLOc(); | |
1141 | #endif | |
f65fd747 | 1142 | |
88764ae2 UD |
1143 | #endif /* _LIBC */ |
1144 | ||
f65fd747 | 1145 | |
fa8d436c UD |
1146 | /* |
1147 | pvalloc(size_t n); | |
1148 | Equivalent to valloc(minimum-page-that-holds(n)), that is, | |
1149 | round up n to nearest pagesize. | |
1150 | */ | |
1151 | #if __STD_C | |
1152 | Void_t* public_pVALLOc(size_t); | |
1153 | #else | |
1154 | Void_t* public_pVALLOc(); | |
1155 | #endif | |
f65fd747 | 1156 | |
fa8d436c UD |
1157 | /* |
1158 | cfree(Void_t* p); | |
1159 | Equivalent to free(p). | |
1160 | ||
1161 | cfree is needed/defined on some systems that pair it with calloc, | |
a9177ff5 | 1162 | for odd historical reasons (such as: cfree is used in example |
fa8d436c UD |
1163 | code in the first edition of K&R). |
1164 | */ | |
1165 | #if __STD_C | |
1166 | void public_cFREe(Void_t*); | |
f65fd747 | 1167 | #else |
fa8d436c UD |
1168 | void public_cFREe(); |
1169 | #endif | |
1170 | ||
1171 | /* | |
1172 | malloc_trim(size_t pad); | |
1173 | ||
1174 | If possible, gives memory back to the system (via negative | |
1175 | arguments to sbrk) if there is unused memory at the `high' end of | |
1176 | the malloc pool. You can call this after freeing large blocks of | |
1177 | memory to potentially reduce the system-level memory requirements | |
1178 | of a program. However, it cannot guarantee to reduce memory. Under | |
1179 | some allocation patterns, some large free blocks of memory will be | |
1180 | locked between two used chunks, so they cannot be given back to | |
1181 | the system. | |
a9177ff5 | 1182 | |
fa8d436c UD |
1183 | The `pad' argument to malloc_trim represents the amount of free |
1184 | trailing space to leave untrimmed. If this argument is zero, | |
1185 | only the minimum amount of memory to maintain internal data | |
1186 | structures will be left (one page or less). Non-zero arguments | |
1187 | can be supplied to maintain enough trailing space to service | |
1188 | future expected allocations without having to re-obtain memory | |
1189 | from the system. | |
a9177ff5 | 1190 | |
fa8d436c UD |
1191 | Malloc_trim returns 1 if it actually released any memory, else 0. |
1192 | On systems that do not support "negative sbrks", it will always | |
1193 | rreturn 0. | |
1194 | */ | |
1195 | #if __STD_C | |
1196 | int public_mTRIm(size_t); | |
1197 | #else | |
1198 | int public_mTRIm(); | |
1199 | #endif | |
1200 | ||
1201 | /* | |
1202 | malloc_usable_size(Void_t* p); | |
1203 | ||
1204 | Returns the number of bytes you can actually use in | |
1205 | an allocated chunk, which may be more than you requested (although | |
1206 | often not) due to alignment and minimum size constraints. | |
1207 | You can use this many bytes without worrying about | |
1208 | overwriting other allocated objects. This is not a particularly great | |
1209 | programming practice. malloc_usable_size can be more useful in | |
1210 | debugging and assertions, for example: | |
1211 | ||
1212 | p = malloc(n); | |
1213 | assert(malloc_usable_size(p) >= 256); | |
1214 | ||
1215 | */ | |
1216 | #if __STD_C | |
1217 | size_t public_mUSABLe(Void_t*); | |
1218 | #else | |
1219 | size_t public_mUSABLe(); | |
f65fd747 | 1220 | #endif |
fa8d436c UD |
1221 | |
1222 | /* | |
1223 | malloc_stats(); | |
1224 | Prints on stderr the amount of space obtained from the system (both | |
1225 | via sbrk and mmap), the maximum amount (which may be more than | |
1226 | current if malloc_trim and/or munmap got called), and the current | |
1227 | number of bytes allocated via malloc (or realloc, etc) but not yet | |
1228 | freed. Note that this is the number of bytes allocated, not the | |
1229 | number requested. It will be larger than the number requested | |
1230 | because of alignment and bookkeeping overhead. Because it includes | |
1231 | alignment wastage as being in use, this figure may be greater than | |
1232 | zero even when no user-level chunks are allocated. | |
1233 | ||
1234 | The reported current and maximum system memory can be inaccurate if | |
1235 | a program makes other calls to system memory allocation functions | |
1236 | (normally sbrk) outside of malloc. | |
1237 | ||
1238 | malloc_stats prints only the most commonly interesting statistics. | |
1239 | More information can be obtained by calling mallinfo. | |
1240 | ||
1241 | */ | |
1242 | #if __STD_C | |
1243 | void public_mSTATs(void); | |
1244 | #else | |
1245 | void public_mSTATs(); | |
f65fd747 UD |
1246 | #endif |
1247 | ||
f7ddf3d3 UD |
1248 | /* |
1249 | malloc_get_state(void); | |
1250 | ||
1251 | Returns the state of all malloc variables in an opaque data | |
1252 | structure. | |
1253 | */ | |
1254 | #if __STD_C | |
1255 | Void_t* public_gET_STATe(void); | |
1256 | #else | |
1257 | Void_t* public_gET_STATe(); | |
1258 | #endif | |
1259 | ||
1260 | /* | |
1261 | malloc_set_state(Void_t* state); | |
1262 | ||
1263 | Restore the state of all malloc variables from data obtained with | |
1264 | malloc_get_state(). | |
1265 | */ | |
1266 | #if __STD_C | |
1267 | int public_sET_STATe(Void_t*); | |
1268 | #else | |
1269 | int public_sET_STATe(); | |
1270 | #endif | |
1271 | ||
1272 | #ifdef _LIBC | |
1273 | /* | |
1274 | posix_memalign(void **memptr, size_t alignment, size_t size); | |
1275 | ||
1276 | POSIX wrapper like memalign(), checking for validity of size. | |
1277 | */ | |
1278 | int __posix_memalign(void **, size_t, size_t); | |
1279 | #endif | |
1280 | ||
fa8d436c UD |
1281 | /* mallopt tuning options */ |
1282 | ||
f65fd747 | 1283 | /* |
fa8d436c UD |
1284 | M_MXFAST is the maximum request size used for "fastbins", special bins |
1285 | that hold returned chunks without consolidating their spaces. This | |
1286 | enables future requests for chunks of the same size to be handled | |
1287 | very quickly, but can increase fragmentation, and thus increase the | |
1288 | overall memory footprint of a program. | |
1289 | ||
1290 | This malloc manages fastbins very conservatively yet still | |
1291 | efficiently, so fragmentation is rarely a problem for values less | |
1292 | than or equal to the default. The maximum supported value of MXFAST | |
1293 | is 80. You wouldn't want it any higher than this anyway. Fastbins | |
1294 | are designed especially for use with many small structs, objects or | |
1295 | strings -- the default handles structs/objects/arrays with sizes up | |
1296 | to 8 4byte fields, or small strings representing words, tokens, | |
1297 | etc. Using fastbins for larger objects normally worsens | |
1298 | fragmentation without improving speed. | |
1299 | ||
1300 | M_MXFAST is set in REQUEST size units. It is internally used in | |
1301 | chunksize units, which adds padding and alignment. You can reduce | |
1302 | M_MXFAST to 0 to disable all use of fastbins. This causes the malloc | |
1303 | algorithm to be a closer approximation of fifo-best-fit in all cases, | |
1304 | not just for larger requests, but will generally cause it to be | |
1305 | slower. | |
f65fd747 UD |
1306 | */ |
1307 | ||
1308 | ||
fa8d436c UD |
1309 | /* M_MXFAST is a standard SVID/XPG tuning option, usually listed in malloc.h */ |
1310 | #ifndef M_MXFAST | |
a9177ff5 | 1311 | #define M_MXFAST 1 |
fa8d436c | 1312 | #endif |
f65fd747 | 1313 | |
fa8d436c UD |
1314 | #ifndef DEFAULT_MXFAST |
1315 | #define DEFAULT_MXFAST 64 | |
10dc2a90 UD |
1316 | #endif |
1317 | ||
10dc2a90 | 1318 | |
fa8d436c UD |
1319 | /* |
1320 | M_TRIM_THRESHOLD is the maximum amount of unused top-most memory | |
1321 | to keep before releasing via malloc_trim in free(). | |
1322 | ||
1323 | Automatic trimming is mainly useful in long-lived programs. | |
1324 | Because trimming via sbrk can be slow on some systems, and can | |
1325 | sometimes be wasteful (in cases where programs immediately | |
1326 | afterward allocate more large chunks) the value should be high | |
1327 | enough so that your overall system performance would improve by | |
1328 | releasing this much memory. | |
1329 | ||
1330 | The trim threshold and the mmap control parameters (see below) | |
1331 | can be traded off with one another. Trimming and mmapping are | |
1332 | two different ways of releasing unused memory back to the | |
1333 | system. Between these two, it is often possible to keep | |
1334 | system-level demands of a long-lived program down to a bare | |
1335 | minimum. For example, in one test suite of sessions measuring | |
1336 | the XF86 X server on Linux, using a trim threshold of 128K and a | |
1337 | mmap threshold of 192K led to near-minimal long term resource | |
1338 | consumption. | |
1339 | ||
1340 | If you are using this malloc in a long-lived program, it should | |
1341 | pay to experiment with these values. As a rough guide, you | |
1342 | might set to a value close to the average size of a process | |
1343 | (program) running on your system. Releasing this much memory | |
1344 | would allow such a process to run in memory. Generally, it's | |
1345 | worth it to tune for trimming rather tham memory mapping when a | |
1346 | program undergoes phases where several large chunks are | |
1347 | allocated and released in ways that can reuse each other's | |
1348 | storage, perhaps mixed with phases where there are no such | |
1349 | chunks at all. And in well-behaved long-lived programs, | |
1350 | controlling release of large blocks via trimming versus mapping | |
1351 | is usually faster. | |
1352 | ||
1353 | However, in most programs, these parameters serve mainly as | |
1354 | protection against the system-level effects of carrying around | |
1355 | massive amounts of unneeded memory. Since frequent calls to | |
1356 | sbrk, mmap, and munmap otherwise degrade performance, the default | |
1357 | parameters are set to relatively high values that serve only as | |
1358 | safeguards. | |
1359 | ||
1360 | The trim value It must be greater than page size to have any useful | |
a9177ff5 | 1361 | effect. To disable trimming completely, you can set to |
fa8d436c UD |
1362 | (unsigned long)(-1) |
1363 | ||
1364 | Trim settings interact with fastbin (MXFAST) settings: Unless | |
1365 | TRIM_FASTBINS is defined, automatic trimming never takes place upon | |
1366 | freeing a chunk with size less than or equal to MXFAST. Trimming is | |
1367 | instead delayed until subsequent freeing of larger chunks. However, | |
1368 | you can still force an attempted trim by calling malloc_trim. | |
1369 | ||
1370 | Also, trimming is not generally possible in cases where | |
1371 | the main arena is obtained via mmap. | |
1372 | ||
1373 | Note that the trick some people use of mallocing a huge space and | |
1374 | then freeing it at program startup, in an attempt to reserve system | |
1375 | memory, doesn't have the intended effect under automatic trimming, | |
1376 | since that memory will immediately be returned to the system. | |
1377 | */ | |
1378 | ||
1379 | #define M_TRIM_THRESHOLD -1 | |
1380 | ||
1381 | #ifndef DEFAULT_TRIM_THRESHOLD | |
1382 | #define DEFAULT_TRIM_THRESHOLD (128 * 1024) | |
1383 | #endif | |
1384 | ||
1385 | /* | |
1386 | M_TOP_PAD is the amount of extra `padding' space to allocate or | |
1387 | retain whenever sbrk is called. It is used in two ways internally: | |
1388 | ||
1389 | * When sbrk is called to extend the top of the arena to satisfy | |
1390 | a new malloc request, this much padding is added to the sbrk | |
1391 | request. | |
1392 | ||
1393 | * When malloc_trim is called automatically from free(), | |
1394 | it is used as the `pad' argument. | |
1395 | ||
1396 | In both cases, the actual amount of padding is rounded | |
1397 | so that the end of the arena is always a system page boundary. | |
1398 | ||
1399 | The main reason for using padding is to avoid calling sbrk so | |
1400 | often. Having even a small pad greatly reduces the likelihood | |
1401 | that nearly every malloc request during program start-up (or | |
1402 | after trimming) will invoke sbrk, which needlessly wastes | |
1403 | time. | |
1404 | ||
1405 | Automatic rounding-up to page-size units is normally sufficient | |
1406 | to avoid measurable overhead, so the default is 0. However, in | |
1407 | systems where sbrk is relatively slow, it can pay to increase | |
1408 | this value, at the expense of carrying around more memory than | |
1409 | the program needs. | |
1410 | */ | |
10dc2a90 | 1411 | |
fa8d436c | 1412 | #define M_TOP_PAD -2 |
10dc2a90 | 1413 | |
fa8d436c UD |
1414 | #ifndef DEFAULT_TOP_PAD |
1415 | #define DEFAULT_TOP_PAD (0) | |
1416 | #endif | |
f65fd747 | 1417 | |
1d05c2fb UD |
1418 | /* |
1419 | MMAP_THRESHOLD_MAX and _MIN are the bounds on the dynamically | |
1420 | adjusted MMAP_THRESHOLD. | |
1421 | */ | |
1422 | ||
1423 | #ifndef DEFAULT_MMAP_THRESHOLD_MIN | |
1424 | #define DEFAULT_MMAP_THRESHOLD_MIN (128 * 1024) | |
1425 | #endif | |
1426 | ||
1427 | #ifndef DEFAULT_MMAP_THRESHOLD_MAX | |
e404fb16 UD |
1428 | /* For 32-bit platforms we cannot increase the maximum mmap |
1429 | threshold much because it is also the minimum value for the | |
bd2c2341 UD |
1430 | maximum heap size and its alignment. Going above 512k (i.e., 1M |
1431 | for new heaps) wastes too much address space. */ | |
e404fb16 | 1432 | # if __WORDSIZE == 32 |
bd2c2341 | 1433 | # define DEFAULT_MMAP_THRESHOLD_MAX (512 * 1024) |
e404fb16 | 1434 | # else |
bd2c2341 | 1435 | # define DEFAULT_MMAP_THRESHOLD_MAX (4 * 1024 * 1024 * sizeof(long)) |
e404fb16 | 1436 | # endif |
1d05c2fb UD |
1437 | #endif |
1438 | ||
fa8d436c UD |
1439 | /* |
1440 | M_MMAP_THRESHOLD is the request size threshold for using mmap() | |
1441 | to service a request. Requests of at least this size that cannot | |
1442 | be allocated using already-existing space will be serviced via mmap. | |
1443 | (If enough normal freed space already exists it is used instead.) | |
1444 | ||
1445 | Using mmap segregates relatively large chunks of memory so that | |
1446 | they can be individually obtained and released from the host | |
1447 | system. A request serviced through mmap is never reused by any | |
1448 | other request (at least not directly; the system may just so | |
1449 | happen to remap successive requests to the same locations). | |
1450 | ||
1451 | Segregating space in this way has the benefits that: | |
1452 | ||
a9177ff5 RM |
1453 | 1. Mmapped space can ALWAYS be individually released back |
1454 | to the system, which helps keep the system level memory | |
1455 | demands of a long-lived program low. | |
fa8d436c UD |
1456 | 2. Mapped memory can never become `locked' between |
1457 | other chunks, as can happen with normally allocated chunks, which | |
1458 | means that even trimming via malloc_trim would not release them. | |
1459 | 3. On some systems with "holes" in address spaces, mmap can obtain | |
1460 | memory that sbrk cannot. | |
1461 | ||
1462 | However, it has the disadvantages that: | |
1463 | ||
1464 | 1. The space cannot be reclaimed, consolidated, and then | |
1465 | used to service later requests, as happens with normal chunks. | |
1466 | 2. It can lead to more wastage because of mmap page alignment | |
1467 | requirements | |
1468 | 3. It causes malloc performance to be more dependent on host | |
1469 | system memory management support routines which may vary in | |
1470 | implementation quality and may impose arbitrary | |
1471 | limitations. Generally, servicing a request via normal | |
1472 | malloc steps is faster than going through a system's mmap. | |
1473 | ||
1474 | The advantages of mmap nearly always outweigh disadvantages for | |
1475 | "large" chunks, but the value of "large" varies across systems. The | |
1476 | default is an empirically derived value that works well in most | |
1477 | systems. | |
1d05c2fb UD |
1478 | |
1479 | ||
1480 | Update in 2006: | |
1481 | The above was written in 2001. Since then the world has changed a lot. | |
1482 | Memory got bigger. Applications got bigger. The virtual address space | |
1483 | layout in 32 bit linux changed. | |
1484 | ||
1485 | In the new situation, brk() and mmap space is shared and there are no | |
1486 | artificial limits on brk size imposed by the kernel. What is more, | |
1487 | applications have started using transient allocations larger than the | |
1488 | 128Kb as was imagined in 2001. | |
1489 | ||
1490 | The price for mmap is also high now; each time glibc mmaps from the | |
1491 | kernel, the kernel is forced to zero out the memory it gives to the | |
1492 | application. Zeroing memory is expensive and eats a lot of cache and | |
1493 | memory bandwidth. This has nothing to do with the efficiency of the | |
1494 | virtual memory system, by doing mmap the kernel just has no choice but | |
1495 | to zero. | |
1496 | ||
1497 | In 2001, the kernel had a maximum size for brk() which was about 800 | |
1498 | megabytes on 32 bit x86, at that point brk() would hit the first | |
1499 | mmaped shared libaries and couldn't expand anymore. With current 2.6 | |
1500 | kernels, the VA space layout is different and brk() and mmap | |
1501 | both can span the entire heap at will. | |
1502 | ||
1503 | Rather than using a static threshold for the brk/mmap tradeoff, | |
1504 | we are now using a simple dynamic one. The goal is still to avoid | |
1505 | fragmentation. The old goals we kept are | |
1506 | 1) try to get the long lived large allocations to use mmap() | |
1507 | 2) really large allocations should always use mmap() | |
1508 | and we're adding now: | |
1509 | 3) transient allocations should use brk() to avoid forcing the kernel | |
1510 | having to zero memory over and over again | |
1511 | ||
1512 | The implementation works with a sliding threshold, which is by default | |
1513 | limited to go between 128Kb and 32Mb (64Mb for 64 bitmachines) and starts | |
1514 | out at 128Kb as per the 2001 default. | |
1515 | ||
1516 | This allows us to satisfy requirement 1) under the assumption that long | |
1517 | lived allocations are made early in the process' lifespan, before it has | |
1518 | started doing dynamic allocations of the same size (which will | |
1519 | increase the threshold). | |
1520 | ||
1521 | The upperbound on the threshold satisfies requirement 2) | |
1522 | ||
1523 | The threshold goes up in value when the application frees memory that was | |
1524 | allocated with the mmap allocator. The idea is that once the application | |
1525 | starts freeing memory of a certain size, it's highly probable that this is | |
1526 | a size the application uses for transient allocations. This estimator | |
1527 | is there to satisfy the new third requirement. | |
1528 | ||
f65fd747 UD |
1529 | */ |
1530 | ||
fa8d436c | 1531 | #define M_MMAP_THRESHOLD -3 |
f65fd747 | 1532 | |
fa8d436c | 1533 | #ifndef DEFAULT_MMAP_THRESHOLD |
1d05c2fb | 1534 | #define DEFAULT_MMAP_THRESHOLD DEFAULT_MMAP_THRESHOLD_MIN |
fa8d436c UD |
1535 | #endif |
1536 | ||
1537 | /* | |
1538 | M_MMAP_MAX is the maximum number of requests to simultaneously | |
1539 | service using mmap. This parameter exists because | |
1540 | some systems have a limited number of internal tables for | |
1541 | use by mmap, and using more than a few of them may degrade | |
1542 | performance. | |
1543 | ||
1544 | The default is set to a value that serves only as a safeguard. | |
1545 | Setting to 0 disables use of mmap for servicing large requests. If | |
1546 | HAVE_MMAP is not set, the default value is 0, and attempts to set it | |
1547 | to non-zero values in mallopt will fail. | |
1548 | */ | |
f65fd747 | 1549 | |
fa8d436c UD |
1550 | #define M_MMAP_MAX -4 |
1551 | ||
1552 | #ifndef DEFAULT_MMAP_MAX | |
1553 | #if HAVE_MMAP | |
1554 | #define DEFAULT_MMAP_MAX (65536) | |
1555 | #else | |
1556 | #define DEFAULT_MMAP_MAX (0) | |
1557 | #endif | |
f65fd747 UD |
1558 | #endif |
1559 | ||
fa8d436c | 1560 | #ifdef __cplusplus |
3c6904fb | 1561 | } /* end of extern "C" */ |
fa8d436c | 1562 | #endif |
f65fd747 | 1563 | |
100351c3 | 1564 | #include <malloc.h> |
f65fd747 | 1565 | |
fa8d436c UD |
1566 | #ifndef BOUNDED_N |
1567 | #define BOUNDED_N(ptr, sz) (ptr) | |
1568 | #endif | |
1569 | #ifndef RETURN_ADDRESS | |
1570 | #define RETURN_ADDRESS(X_) (NULL) | |
9ae6fc54 | 1571 | #endif |
431c33c0 UD |
1572 | |
1573 | /* On some platforms we can compile internal, not exported functions better. | |
1574 | Let the environment provide a macro and define it to be empty if it | |
1575 | is not available. */ | |
1576 | #ifndef internal_function | |
1577 | # define internal_function | |
1578 | #endif | |
1579 | ||
fa8d436c UD |
1580 | /* Forward declarations. */ |
1581 | struct malloc_chunk; | |
1582 | typedef struct malloc_chunk* mchunkptr; | |
431c33c0 | 1583 | |
fa8d436c | 1584 | /* Internal routines. */ |
f65fd747 | 1585 | |
fa8d436c | 1586 | #if __STD_C |
f65fd747 | 1587 | |
f1c5213d RM |
1588 | Void_t* _int_malloc(mstate, size_t); |
1589 | void _int_free(mstate, Void_t*); | |
1590 | Void_t* _int_realloc(mstate, Void_t*, size_t); | |
1591 | Void_t* _int_memalign(mstate, size_t, size_t); | |
1592 | Void_t* _int_valloc(mstate, size_t); | |
fa8d436c UD |
1593 | static Void_t* _int_pvalloc(mstate, size_t); |
1594 | /*static Void_t* cALLOc(size_t, size_t);*/ | |
88764ae2 | 1595 | #ifndef _LIBC |
fa8d436c UD |
1596 | static Void_t** _int_icalloc(mstate, size_t, size_t, Void_t**); |
1597 | static Void_t** _int_icomalloc(mstate, size_t, size_t*, Void_t**); | |
88764ae2 | 1598 | #endif |
fa8d436c UD |
1599 | static int mTRIm(size_t); |
1600 | static size_t mUSABLe(Void_t*); | |
1601 | static void mSTATs(void); | |
1602 | static int mALLOPt(int, int); | |
1603 | static struct mallinfo mALLINFo(mstate); | |
6bf4302e | 1604 | static void malloc_printerr(int action, const char *str, void *ptr); |
fa8d436c UD |
1605 | |
1606 | static Void_t* internal_function mem2mem_check(Void_t *p, size_t sz); | |
1607 | static int internal_function top_check(void); | |
1608 | static void internal_function munmap_chunk(mchunkptr p); | |
a9177ff5 | 1609 | #if HAVE_MREMAP |
fa8d436c | 1610 | static mchunkptr internal_function mremap_chunk(mchunkptr p, size_t new_size); |
a9177ff5 | 1611 | #endif |
fa8d436c UD |
1612 | |
1613 | static Void_t* malloc_check(size_t sz, const Void_t *caller); | |
1614 | static void free_check(Void_t* mem, const Void_t *caller); | |
1615 | static Void_t* realloc_check(Void_t* oldmem, size_t bytes, | |
1616 | const Void_t *caller); | |
1617 | static Void_t* memalign_check(size_t alignment, size_t bytes, | |
1618 | const Void_t *caller); | |
1619 | #ifndef NO_THREADS | |
fde89ad0 | 1620 | # ifdef _LIBC |
32c075e1 | 1621 | # if USE___THREAD || (defined USE_TLS && !defined SHARED) |
fde89ad0 RM |
1622 | /* These routines are never needed in this configuration. */ |
1623 | # define NO_STARTER | |
1624 | # endif | |
1625 | # endif | |
1626 | # ifdef NO_STARTER | |
1627 | # undef NO_STARTER | |
1628 | # else | |
fa8d436c | 1629 | static Void_t* malloc_starter(size_t sz, const Void_t *caller); |
fde89ad0 | 1630 | static Void_t* memalign_starter(size_t aln, size_t sz, const Void_t *caller); |
fa8d436c | 1631 | static void free_starter(Void_t* mem, const Void_t *caller); |
fde89ad0 | 1632 | # endif |
fa8d436c UD |
1633 | static Void_t* malloc_atfork(size_t sz, const Void_t *caller); |
1634 | static void free_atfork(Void_t* mem, const Void_t *caller); | |
1635 | #endif | |
f65fd747 | 1636 | |
fa8d436c | 1637 | #else |
f65fd747 | 1638 | |
fa8d436c UD |
1639 | Void_t* _int_malloc(); |
1640 | void _int_free(); | |
1641 | Void_t* _int_realloc(); | |
1642 | Void_t* _int_memalign(); | |
1643 | Void_t* _int_valloc(); | |
1644 | Void_t* _int_pvalloc(); | |
1645 | /*static Void_t* cALLOc();*/ | |
1646 | static Void_t** _int_icalloc(); | |
1647 | static Void_t** _int_icomalloc(); | |
1648 | static int mTRIm(); | |
1649 | static size_t mUSABLe(); | |
1650 | static void mSTATs(); | |
1651 | static int mALLOPt(); | |
1652 | static struct mallinfo mALLINFo(); | |
f65fd747 | 1653 | |
fa8d436c | 1654 | #endif |
f65fd747 | 1655 | |
f65fd747 | 1656 | |
f65fd747 | 1657 | |
f65fd747 | 1658 | |
fa8d436c | 1659 | /* ------------- Optional versions of memcopy ---------------- */ |
f65fd747 | 1660 | |
a1648746 | 1661 | |
fa8d436c | 1662 | #if USE_MEMCPY |
a1648746 | 1663 | |
a9177ff5 | 1664 | /* |
fa8d436c UD |
1665 | Note: memcpy is ONLY invoked with non-overlapping regions, |
1666 | so the (usually slower) memmove is not needed. | |
1667 | */ | |
a1648746 | 1668 | |
fa8d436c UD |
1669 | #define MALLOC_COPY(dest, src, nbytes) memcpy(dest, src, nbytes) |
1670 | #define MALLOC_ZERO(dest, nbytes) memset(dest, 0, nbytes) | |
f65fd747 | 1671 | |
fa8d436c | 1672 | #else /* !USE_MEMCPY */ |
f65fd747 | 1673 | |
fa8d436c | 1674 | /* Use Duff's device for good zeroing/copying performance. */ |
f65fd747 | 1675 | |
fa8d436c UD |
1676 | #define MALLOC_ZERO(charp, nbytes) \ |
1677 | do { \ | |
1678 | INTERNAL_SIZE_T* mzp = (INTERNAL_SIZE_T*)(charp); \ | |
1679 | unsigned long mctmp = (nbytes)/sizeof(INTERNAL_SIZE_T); \ | |
1680 | long mcn; \ | |
1681 | if (mctmp < 8) mcn = 0; else { mcn = (mctmp-1)/8; mctmp %= 8; } \ | |
1682 | switch (mctmp) { \ | |
1683 | case 0: for(;;) { *mzp++ = 0; \ | |
1684 | case 7: *mzp++ = 0; \ | |
1685 | case 6: *mzp++ = 0; \ | |
1686 | case 5: *mzp++ = 0; \ | |
1687 | case 4: *mzp++ = 0; \ | |
1688 | case 3: *mzp++ = 0; \ | |
1689 | case 2: *mzp++ = 0; \ | |
1690 | case 1: *mzp++ = 0; if(mcn <= 0) break; mcn--; } \ | |
1691 | } \ | |
1692 | } while(0) | |
f65fd747 | 1693 | |
fa8d436c UD |
1694 | #define MALLOC_COPY(dest,src,nbytes) \ |
1695 | do { \ | |
1696 | INTERNAL_SIZE_T* mcsrc = (INTERNAL_SIZE_T*) src; \ | |
1697 | INTERNAL_SIZE_T* mcdst = (INTERNAL_SIZE_T*) dest; \ | |
1698 | unsigned long mctmp = (nbytes)/sizeof(INTERNAL_SIZE_T); \ | |
1699 | long mcn; \ | |
1700 | if (mctmp < 8) mcn = 0; else { mcn = (mctmp-1)/8; mctmp %= 8; } \ | |
1701 | switch (mctmp) { \ | |
1702 | case 0: for(;;) { *mcdst++ = *mcsrc++; \ | |
1703 | case 7: *mcdst++ = *mcsrc++; \ | |
1704 | case 6: *mcdst++ = *mcsrc++; \ | |
1705 | case 5: *mcdst++ = *mcsrc++; \ | |
1706 | case 4: *mcdst++ = *mcsrc++; \ | |
1707 | case 3: *mcdst++ = *mcsrc++; \ | |
1708 | case 2: *mcdst++ = *mcsrc++; \ | |
1709 | case 1: *mcdst++ = *mcsrc++; if(mcn <= 0) break; mcn--; } \ | |
1710 | } \ | |
1711 | } while(0) | |
f65fd747 | 1712 | |
f65fd747 UD |
1713 | #endif |
1714 | ||
fa8d436c | 1715 | /* ------------------ MMAP support ------------------ */ |
f65fd747 | 1716 | |
f65fd747 | 1717 | |
fa8d436c | 1718 | #if HAVE_MMAP |
f65fd747 | 1719 | |
fa8d436c UD |
1720 | #include <fcntl.h> |
1721 | #ifndef LACKS_SYS_MMAN_H | |
1722 | #include <sys/mman.h> | |
1723 | #endif | |
f65fd747 | 1724 | |
fa8d436c UD |
1725 | #if !defined(MAP_ANONYMOUS) && defined(MAP_ANON) |
1726 | # define MAP_ANONYMOUS MAP_ANON | |
1727 | #endif | |
1728 | #if !defined(MAP_FAILED) | |
1729 | # define MAP_FAILED ((char*)-1) | |
1730 | #endif | |
f65fd747 | 1731 | |
fa8d436c UD |
1732 | #ifndef MAP_NORESERVE |
1733 | # ifdef MAP_AUTORESRV | |
1734 | # define MAP_NORESERVE MAP_AUTORESRV | |
1735 | # else | |
1736 | # define MAP_NORESERVE 0 | |
1737 | # endif | |
f65fd747 UD |
1738 | #endif |
1739 | ||
a9177ff5 RM |
1740 | /* |
1741 | Nearly all versions of mmap support MAP_ANONYMOUS, | |
fa8d436c UD |
1742 | so the following is unlikely to be needed, but is |
1743 | supplied just in case. | |
1744 | */ | |
f65fd747 | 1745 | |
fa8d436c | 1746 | #ifndef MAP_ANONYMOUS |
f65fd747 | 1747 | |
fa8d436c | 1748 | static int dev_zero_fd = -1; /* Cached file descriptor for /dev/zero. */ |
2f6d1f1b | 1749 | |
fa8d436c UD |
1750 | #define MMAP(addr, size, prot, flags) ((dev_zero_fd < 0) ? \ |
1751 | (dev_zero_fd = open("/dev/zero", O_RDWR), \ | |
1752 | mmap((addr), (size), (prot), (flags), dev_zero_fd, 0)) : \ | |
1753 | mmap((addr), (size), (prot), (flags), dev_zero_fd, 0)) | |
f65fd747 | 1754 | |
fa8d436c | 1755 | #else |
f65fd747 | 1756 | |
fa8d436c UD |
1757 | #define MMAP(addr, size, prot, flags) \ |
1758 | (mmap((addr), (size), (prot), (flags)|MAP_ANONYMOUS, -1, 0)) | |
f65fd747 | 1759 | |
e9b3e3c5 | 1760 | #endif |
f65fd747 UD |
1761 | |
1762 | ||
fa8d436c UD |
1763 | #endif /* HAVE_MMAP */ |
1764 | ||
1765 | ||
f65fd747 | 1766 | /* |
fa8d436c | 1767 | ----------------------- Chunk representations ----------------------- |
f65fd747 UD |
1768 | */ |
1769 | ||
1770 | ||
fa8d436c UD |
1771 | /* |
1772 | This struct declaration is misleading (but accurate and necessary). | |
1773 | It declares a "view" into memory allowing access to necessary | |
1774 | fields at known offsets from a given base. See explanation below. | |
1775 | */ | |
1776 | ||
1777 | struct malloc_chunk { | |
1778 | ||
1779 | INTERNAL_SIZE_T prev_size; /* Size of previous chunk (if free). */ | |
1780 | INTERNAL_SIZE_T size; /* Size in bytes, including overhead. */ | |
1781 | ||
1782 | struct malloc_chunk* fd; /* double links -- used only if free. */ | |
f65fd747 UD |
1783 | struct malloc_chunk* bk; |
1784 | }; | |
1785 | ||
f65fd747 UD |
1786 | |
1787 | /* | |
f65fd747 UD |
1788 | malloc_chunk details: |
1789 | ||
1790 | (The following includes lightly edited explanations by Colin Plumb.) | |
1791 | ||
1792 | Chunks of memory are maintained using a `boundary tag' method as | |
1793 | described in e.g., Knuth or Standish. (See the paper by Paul | |
1794 | Wilson ftp://ftp.cs.utexas.edu/pub/garbage/allocsrv.ps for a | |
1795 | survey of such techniques.) Sizes of free chunks are stored both | |
1796 | in the front of each chunk and at the end. This makes | |
1797 | consolidating fragmented chunks into bigger chunks very fast. The | |
1798 | size fields also hold bits representing whether chunks are free or | |
1799 | in use. | |
1800 | ||
1801 | An allocated chunk looks like this: | |
1802 | ||
1803 | ||
1804 | chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
1805 | | Size of previous chunk, if allocated | | | |
1806 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
8088488d | 1807 | | Size of chunk, in bytes |M|P| |
f65fd747 UD |
1808 | mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
1809 | | User data starts here... . | |
1810 | . . | |
9ea9af19 | 1811 | . (malloc_usable_size() bytes) . |
f65fd747 UD |
1812 | . | |
1813 | nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
1814 | | Size of chunk | | |
1815 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
1816 | ||
1817 | ||
1818 | Where "chunk" is the front of the chunk for the purpose of most of | |
1819 | the malloc code, but "mem" is the pointer that is returned to the | |
1820 | user. "Nextchunk" is the beginning of the next contiguous chunk. | |
1821 | ||
fa8d436c | 1822 | Chunks always begin on even word boundries, so the mem portion |
f65fd747 | 1823 | (which is returned to the user) is also on an even word boundary, and |
fa8d436c | 1824 | thus at least double-word aligned. |
f65fd747 UD |
1825 | |
1826 | Free chunks are stored in circular doubly-linked lists, and look like this: | |
1827 | ||
1828 | chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
1829 | | Size of previous chunk | | |
1830 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
1831 | `head:' | Size of chunk, in bytes |P| | |
1832 | mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
1833 | | Forward pointer to next chunk in list | | |
1834 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
1835 | | Back pointer to previous chunk in list | | |
1836 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
1837 | | Unused space (may be 0 bytes long) . | |
1838 | . . | |
1839 | . | | |
1840 | nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
1841 | `foot:' | Size of chunk, in bytes | | |
1842 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
1843 | ||
1844 | The P (PREV_INUSE) bit, stored in the unused low-order bit of the | |
1845 | chunk size (which is always a multiple of two words), is an in-use | |
1846 | bit for the *previous* chunk. If that bit is *clear*, then the | |
1847 | word before the current chunk size contains the previous chunk | |
1848 | size, and can be used to find the front of the previous chunk. | |
fa8d436c UD |
1849 | The very first chunk allocated always has this bit set, |
1850 | preventing access to non-existent (or non-owned) memory. If | |
1851 | prev_inuse is set for any given chunk, then you CANNOT determine | |
1852 | the size of the previous chunk, and might even get a memory | |
1853 | addressing fault when trying to do so. | |
f65fd747 UD |
1854 | |
1855 | Note that the `foot' of the current chunk is actually represented | |
fa8d436c UD |
1856 | as the prev_size of the NEXT chunk. This makes it easier to |
1857 | deal with alignments etc but can be very confusing when trying | |
1858 | to extend or adapt this code. | |
f65fd747 UD |
1859 | |
1860 | The two exceptions to all this are | |
1861 | ||
fa8d436c UD |
1862 | 1. The special chunk `top' doesn't bother using the |
1863 | trailing size field since there is no next contiguous chunk | |
1864 | that would have to index off it. After initialization, `top' | |
1865 | is forced to always exist. If it would become less than | |
1866 | MINSIZE bytes long, it is replenished. | |
f65fd747 UD |
1867 | |
1868 | 2. Chunks allocated via mmap, which have the second-lowest-order | |
8088488d | 1869 | bit M (IS_MMAPPED) set in their size fields. Because they are |
fa8d436c | 1870 | allocated one-by-one, each must contain its own trailing size field. |
f65fd747 UD |
1871 | |
1872 | */ | |
1873 | ||
1874 | /* | |
fa8d436c UD |
1875 | ---------- Size and alignment checks and conversions ---------- |
1876 | */ | |
f65fd747 | 1877 | |
fa8d436c | 1878 | /* conversion from malloc headers to user pointers, and back */ |
f65fd747 | 1879 | |
fa8d436c UD |
1880 | #define chunk2mem(p) ((Void_t*)((char*)(p) + 2*SIZE_SZ)) |
1881 | #define mem2chunk(mem) ((mchunkptr)((char*)(mem) - 2*SIZE_SZ)) | |
f65fd747 | 1882 | |
fa8d436c UD |
1883 | /* The smallest possible chunk */ |
1884 | #define MIN_CHUNK_SIZE (sizeof(struct malloc_chunk)) | |
f65fd747 | 1885 | |
fa8d436c | 1886 | /* The smallest size we can malloc is an aligned minimal chunk */ |
f65fd747 | 1887 | |
fa8d436c UD |
1888 | #define MINSIZE \ |
1889 | (unsigned long)(((MIN_CHUNK_SIZE+MALLOC_ALIGN_MASK) & ~MALLOC_ALIGN_MASK)) | |
f65fd747 | 1890 | |
fa8d436c | 1891 | /* Check if m has acceptable alignment */ |
f65fd747 | 1892 | |
073f560e UD |
1893 | #define aligned_OK(m) (((unsigned long)(m) & MALLOC_ALIGN_MASK) == 0) |
1894 | ||
1895 | #define misaligned_chunk(p) \ | |
1896 | ((uintptr_t)(MALLOC_ALIGNMENT == 2 * SIZE_SZ ? (p) : chunk2mem (p)) \ | |
1897 | & MALLOC_ALIGN_MASK) | |
f65fd747 | 1898 | |
f65fd747 | 1899 | |
a9177ff5 | 1900 | /* |
fa8d436c UD |
1901 | Check if a request is so large that it would wrap around zero when |
1902 | padded and aligned. To simplify some other code, the bound is made | |
1903 | low enough so that adding MINSIZE will also not wrap around zero. | |
1904 | */ | |
f65fd747 | 1905 | |
fa8d436c UD |
1906 | #define REQUEST_OUT_OF_RANGE(req) \ |
1907 | ((unsigned long)(req) >= \ | |
a9177ff5 | 1908 | (unsigned long)(INTERNAL_SIZE_T)(-2 * MINSIZE)) |
f65fd747 | 1909 | |
fa8d436c | 1910 | /* pad request bytes into a usable size -- internal version */ |
f65fd747 | 1911 | |
fa8d436c UD |
1912 | #define request2size(req) \ |
1913 | (((req) + SIZE_SZ + MALLOC_ALIGN_MASK < MINSIZE) ? \ | |
1914 | MINSIZE : \ | |
1915 | ((req) + SIZE_SZ + MALLOC_ALIGN_MASK) & ~MALLOC_ALIGN_MASK) | |
f65fd747 | 1916 | |
fa8d436c | 1917 | /* Same, except also perform argument check */ |
f65fd747 | 1918 | |
fa8d436c UD |
1919 | #define checked_request2size(req, sz) \ |
1920 | if (REQUEST_OUT_OF_RANGE(req)) { \ | |
1921 | MALLOC_FAILURE_ACTION; \ | |
1922 | return 0; \ | |
1923 | } \ | |
a9177ff5 | 1924 | (sz) = request2size(req); |
f65fd747 UD |
1925 | |
1926 | /* | |
fa8d436c | 1927 | --------------- Physical chunk operations --------------- |
f65fd747 UD |
1928 | */ |
1929 | ||
10dc2a90 | 1930 | |
fa8d436c UD |
1931 | /* size field is or'ed with PREV_INUSE when previous adjacent chunk in use */ |
1932 | #define PREV_INUSE 0x1 | |
f65fd747 | 1933 | |
fa8d436c UD |
1934 | /* extract inuse bit of previous chunk */ |
1935 | #define prev_inuse(p) ((p)->size & PREV_INUSE) | |
f65fd747 | 1936 | |
f65fd747 | 1937 | |
fa8d436c UD |
1938 | /* size field is or'ed with IS_MMAPPED if the chunk was obtained with mmap() */ |
1939 | #define IS_MMAPPED 0x2 | |
f65fd747 | 1940 | |
fa8d436c UD |
1941 | /* check for mmap()'ed chunk */ |
1942 | #define chunk_is_mmapped(p) ((p)->size & IS_MMAPPED) | |
f65fd747 | 1943 | |
f65fd747 | 1944 | |
fa8d436c UD |
1945 | /* size field is or'ed with NON_MAIN_ARENA if the chunk was obtained |
1946 | from a non-main arena. This is only set immediately before handing | |
1947 | the chunk to the user, if necessary. */ | |
1948 | #define NON_MAIN_ARENA 0x4 | |
f65fd747 | 1949 | |
fa8d436c UD |
1950 | /* check for chunk from non-main arena */ |
1951 | #define chunk_non_main_arena(p) ((p)->size & NON_MAIN_ARENA) | |
f65fd747 UD |
1952 | |
1953 | ||
a9177ff5 RM |
1954 | /* |
1955 | Bits to mask off when extracting size | |
f65fd747 | 1956 | |
fa8d436c UD |
1957 | Note: IS_MMAPPED is intentionally not masked off from size field in |
1958 | macros for which mmapped chunks should never be seen. This should | |
1959 | cause helpful core dumps to occur if it is tried by accident by | |
1960 | people extending or adapting this malloc. | |
f65fd747 | 1961 | */ |
fa8d436c | 1962 | #define SIZE_BITS (PREV_INUSE|IS_MMAPPED|NON_MAIN_ARENA) |
f65fd747 | 1963 | |
fa8d436c UD |
1964 | /* Get size, ignoring use bits */ |
1965 | #define chunksize(p) ((p)->size & ~(SIZE_BITS)) | |
f65fd747 | 1966 | |
f65fd747 | 1967 | |
fa8d436c UD |
1968 | /* Ptr to next physical malloc_chunk. */ |
1969 | #define next_chunk(p) ((mchunkptr)( ((char*)(p)) + ((p)->size & ~SIZE_BITS) )) | |
f65fd747 | 1970 | |
fa8d436c UD |
1971 | /* Ptr to previous physical malloc_chunk */ |
1972 | #define prev_chunk(p) ((mchunkptr)( ((char*)(p)) - ((p)->prev_size) )) | |
f65fd747 | 1973 | |
fa8d436c UD |
1974 | /* Treat space at ptr + offset as a chunk */ |
1975 | #define chunk_at_offset(p, s) ((mchunkptr)(((char*)(p)) + (s))) | |
1976 | ||
1977 | /* extract p's inuse bit */ | |
1978 | #define inuse(p)\ | |
1979 | ((((mchunkptr)(((char*)(p))+((p)->size & ~SIZE_BITS)))->size) & PREV_INUSE) | |
f65fd747 | 1980 | |
fa8d436c UD |
1981 | /* set/clear chunk as being inuse without otherwise disturbing */ |
1982 | #define set_inuse(p)\ | |
1983 | ((mchunkptr)(((char*)(p)) + ((p)->size & ~SIZE_BITS)))->size |= PREV_INUSE | |
f65fd747 | 1984 | |
fa8d436c UD |
1985 | #define clear_inuse(p)\ |
1986 | ((mchunkptr)(((char*)(p)) + ((p)->size & ~SIZE_BITS)))->size &= ~(PREV_INUSE) | |
f65fd747 UD |
1987 | |
1988 | ||
fa8d436c UD |
1989 | /* check/set/clear inuse bits in known places */ |
1990 | #define inuse_bit_at_offset(p, s)\ | |
1991 | (((mchunkptr)(((char*)(p)) + (s)))->size & PREV_INUSE) | |
f65fd747 | 1992 | |
fa8d436c UD |
1993 | #define set_inuse_bit_at_offset(p, s)\ |
1994 | (((mchunkptr)(((char*)(p)) + (s)))->size |= PREV_INUSE) | |
f65fd747 | 1995 | |
fa8d436c UD |
1996 | #define clear_inuse_bit_at_offset(p, s)\ |
1997 | (((mchunkptr)(((char*)(p)) + (s)))->size &= ~(PREV_INUSE)) | |
f65fd747 | 1998 | |
f65fd747 | 1999 | |
fa8d436c UD |
2000 | /* Set size at head, without disturbing its use bit */ |
2001 | #define set_head_size(p, s) ((p)->size = (((p)->size & SIZE_BITS) | (s))) | |
f65fd747 | 2002 | |
fa8d436c UD |
2003 | /* Set size/use field */ |
2004 | #define set_head(p, s) ((p)->size = (s)) | |
f65fd747 | 2005 | |
fa8d436c UD |
2006 | /* Set size at footer (only when chunk is not in use) */ |
2007 | #define set_foot(p, s) (((mchunkptr)((char*)(p) + (s)))->prev_size = (s)) | |
f65fd747 UD |
2008 | |
2009 | ||
fa8d436c UD |
2010 | /* |
2011 | -------------------- Internal data structures -------------------- | |
2012 | ||
2013 | All internal state is held in an instance of malloc_state defined | |
2014 | below. There are no other static variables, except in two optional | |
a9177ff5 RM |
2015 | cases: |
2016 | * If USE_MALLOC_LOCK is defined, the mALLOC_MUTEx declared above. | |
fa8d436c UD |
2017 | * If HAVE_MMAP is true, but mmap doesn't support |
2018 | MAP_ANONYMOUS, a dummy file descriptor for mmap. | |
2019 | ||
2020 | Beware of lots of tricks that minimize the total bookkeeping space | |
2021 | requirements. The result is a little over 1K bytes (for 4byte | |
2022 | pointers and size_t.) | |
2023 | */ | |
f65fd747 UD |
2024 | |
2025 | /* | |
fa8d436c UD |
2026 | Bins |
2027 | ||
2028 | An array of bin headers for free chunks. Each bin is doubly | |
2029 | linked. The bins are approximately proportionally (log) spaced. | |
2030 | There are a lot of these bins (128). This may look excessive, but | |
2031 | works very well in practice. Most bins hold sizes that are | |
2032 | unusual as malloc request sizes, but are more usual for fragments | |
2033 | and consolidated sets of chunks, which is what these bins hold, so | |
2034 | they can be found quickly. All procedures maintain the invariant | |
2035 | that no consolidated chunk physically borders another one, so each | |
2036 | chunk in a list is known to be preceeded and followed by either | |
2037 | inuse chunks or the ends of memory. | |
2038 | ||
2039 | Chunks in bins are kept in size order, with ties going to the | |
2040 | approximately least recently used chunk. Ordering isn't needed | |
2041 | for the small bins, which all contain the same-sized chunks, but | |
2042 | facilitates best-fit allocation for larger chunks. These lists | |
2043 | are just sequential. Keeping them in order almost never requires | |
2044 | enough traversal to warrant using fancier ordered data | |
a9177ff5 | 2045 | structures. |
fa8d436c UD |
2046 | |
2047 | Chunks of the same size are linked with the most | |
2048 | recently freed at the front, and allocations are taken from the | |
2049 | back. This results in LRU (FIFO) allocation order, which tends | |
2050 | to give each chunk an equal opportunity to be consolidated with | |
2051 | adjacent freed chunks, resulting in larger free chunks and less | |
2052 | fragmentation. | |
2053 | ||
2054 | To simplify use in double-linked lists, each bin header acts | |
2055 | as a malloc_chunk. This avoids special-casing for headers. | |
2056 | But to conserve space and improve locality, we allocate | |
2057 | only the fd/bk pointers of bins, and then use repositioning tricks | |
a9177ff5 | 2058 | to treat these as the fields of a malloc_chunk*. |
f65fd747 UD |
2059 | */ |
2060 | ||
fa8d436c | 2061 | typedef struct malloc_chunk* mbinptr; |
f65fd747 | 2062 | |
fa8d436c | 2063 | /* addressing -- note that bin_at(0) does not exist */ |
41999a1a UD |
2064 | #define bin_at(m, i) \ |
2065 | (mbinptr) (((char *) &((m)->bins[((i) - 1) * 2])) \ | |
2066 | - offsetof (struct malloc_chunk, fd)) | |
f65fd747 | 2067 | |
fa8d436c UD |
2068 | /* analog of ++bin */ |
2069 | #define next_bin(b) ((mbinptr)((char*)(b) + (sizeof(mchunkptr)<<1))) | |
f65fd747 | 2070 | |
fa8d436c UD |
2071 | /* Reminders about list directionality within bins */ |
2072 | #define first(b) ((b)->fd) | |
2073 | #define last(b) ((b)->bk) | |
f65fd747 | 2074 | |
fa8d436c UD |
2075 | /* Take a chunk off a bin list */ |
2076 | #define unlink(P, BK, FD) { \ | |
2077 | FD = P->fd; \ | |
2078 | BK = P->bk; \ | |
3e030bd5 | 2079 | if (__builtin_expect (FD->bk != P || BK->fd != P, 0)) \ |
6bf4302e UD |
2080 | malloc_printerr (check_action, "corrupted double-linked list", P); \ |
2081 | else { \ | |
2082 | FD->bk = BK; \ | |
2083 | BK->fd = FD; \ | |
2084 | } \ | |
fa8d436c | 2085 | } |
f65fd747 | 2086 | |
fa8d436c UD |
2087 | /* |
2088 | Indexing | |
2089 | ||
2090 | Bins for sizes < 512 bytes contain chunks of all the same size, spaced | |
2091 | 8 bytes apart. Larger bins are approximately logarithmically spaced: | |
f65fd747 | 2092 | |
fa8d436c UD |
2093 | 64 bins of size 8 |
2094 | 32 bins of size 64 | |
2095 | 16 bins of size 512 | |
2096 | 8 bins of size 4096 | |
2097 | 4 bins of size 32768 | |
2098 | 2 bins of size 262144 | |
2099 | 1 bin of size what's left | |
f65fd747 | 2100 | |
fa8d436c UD |
2101 | There is actually a little bit of slop in the numbers in bin_index |
2102 | for the sake of speed. This makes no difference elsewhere. | |
f65fd747 | 2103 | |
fa8d436c UD |
2104 | The bins top out around 1MB because we expect to service large |
2105 | requests via mmap. | |
2106 | */ | |
f65fd747 | 2107 | |
fa8d436c UD |
2108 | #define NBINS 128 |
2109 | #define NSMALLBINS 64 | |
2110 | #define SMALLBIN_WIDTH 8 | |
2111 | #define MIN_LARGE_SIZE 512 | |
f65fd747 | 2112 | |
fa8d436c UD |
2113 | #define in_smallbin_range(sz) \ |
2114 | ((unsigned long)(sz) < (unsigned long)MIN_LARGE_SIZE) | |
f65fd747 | 2115 | |
fa8d436c | 2116 | #define smallbin_index(sz) (((unsigned)(sz)) >> 3) |
f65fd747 | 2117 | |
fa8d436c UD |
2118 | #define largebin_index(sz) \ |
2119 | (((((unsigned long)(sz)) >> 6) <= 32)? 56 + (((unsigned long)(sz)) >> 6): \ | |
2120 | ((((unsigned long)(sz)) >> 9) <= 20)? 91 + (((unsigned long)(sz)) >> 9): \ | |
2121 | ((((unsigned long)(sz)) >> 12) <= 10)? 110 + (((unsigned long)(sz)) >> 12): \ | |
2122 | ((((unsigned long)(sz)) >> 15) <= 4)? 119 + (((unsigned long)(sz)) >> 15): \ | |
2123 | ((((unsigned long)(sz)) >> 18) <= 2)? 124 + (((unsigned long)(sz)) >> 18): \ | |
2124 | 126) | |
f65fd747 | 2125 | |
fa8d436c UD |
2126 | #define bin_index(sz) \ |
2127 | ((in_smallbin_range(sz)) ? smallbin_index(sz) : largebin_index(sz)) | |
f65fd747 | 2128 | |
f65fd747 UD |
2129 | |
2130 | /* | |
fa8d436c UD |
2131 | Unsorted chunks |
2132 | ||
2133 | All remainders from chunk splits, as well as all returned chunks, | |
2134 | are first placed in the "unsorted" bin. They are then placed | |
2135 | in regular bins after malloc gives them ONE chance to be used before | |
2136 | binning. So, basically, the unsorted_chunks list acts as a queue, | |
2137 | with chunks being placed on it in free (and malloc_consolidate), | |
2138 | and taken off (to be either used or placed in bins) in malloc. | |
2139 | ||
2140 | The NON_MAIN_ARENA flag is never set for unsorted chunks, so it | |
2141 | does not have to be taken into account in size comparisons. | |
f65fd747 UD |
2142 | */ |
2143 | ||
fa8d436c UD |
2144 | /* The otherwise unindexable 1-bin is used to hold unsorted chunks. */ |
2145 | #define unsorted_chunks(M) (bin_at(M, 1)) | |
f65fd747 | 2146 | |
fa8d436c UD |
2147 | /* |
2148 | Top | |
2149 | ||
2150 | The top-most available chunk (i.e., the one bordering the end of | |
2151 | available memory) is treated specially. It is never included in | |
2152 | any bin, is used only if no other chunk is available, and is | |
2153 | released back to the system if it is very large (see | |
2154 | M_TRIM_THRESHOLD). Because top initially | |
2155 | points to its own bin with initial zero size, thus forcing | |
2156 | extension on the first malloc request, we avoid having any special | |
2157 | code in malloc to check whether it even exists yet. But we still | |
2158 | need to do so when getting memory from system, so we make | |
2159 | initial_top treat the bin as a legal but unusable chunk during the | |
2160 | interval between initialization and the first call to | |
2161 | sYSMALLOc. (This is somewhat delicate, since it relies on | |
2162 | the 2 preceding words to be zero during this interval as well.) | |
2163 | */ | |
f65fd747 | 2164 | |
fa8d436c UD |
2165 | /* Conveniently, the unsorted bin can be used as dummy top on first call */ |
2166 | #define initial_top(M) (unsorted_chunks(M)) | |
f65fd747 | 2167 | |
fa8d436c UD |
2168 | /* |
2169 | Binmap | |
f65fd747 | 2170 | |
fa8d436c UD |
2171 | To help compensate for the large number of bins, a one-level index |
2172 | structure is used for bin-by-bin searching. `binmap' is a | |
2173 | bitvector recording whether bins are definitely empty so they can | |
2174 | be skipped over during during traversals. The bits are NOT always | |
2175 | cleared as soon as bins are empty, but instead only | |
2176 | when they are noticed to be empty during traversal in malloc. | |
2177 | */ | |
f65fd747 | 2178 | |
fa8d436c UD |
2179 | /* Conservatively use 32 bits per map word, even if on 64bit system */ |
2180 | #define BINMAPSHIFT 5 | |
2181 | #define BITSPERMAP (1U << BINMAPSHIFT) | |
2182 | #define BINMAPSIZE (NBINS / BITSPERMAP) | |
f65fd747 | 2183 | |
fa8d436c UD |
2184 | #define idx2block(i) ((i) >> BINMAPSHIFT) |
2185 | #define idx2bit(i) ((1U << ((i) & ((1U << BINMAPSHIFT)-1)))) | |
f65fd747 | 2186 | |
fa8d436c UD |
2187 | #define mark_bin(m,i) ((m)->binmap[idx2block(i)] |= idx2bit(i)) |
2188 | #define unmark_bin(m,i) ((m)->binmap[idx2block(i)] &= ~(idx2bit(i))) | |
2189 | #define get_binmap(m,i) ((m)->binmap[idx2block(i)] & idx2bit(i)) | |
f65fd747 | 2190 | |
fa8d436c UD |
2191 | /* |
2192 | Fastbins | |
2193 | ||
2194 | An array of lists holding recently freed small chunks. Fastbins | |
2195 | are not doubly linked. It is faster to single-link them, and | |
2196 | since chunks are never removed from the middles of these lists, | |
2197 | double linking is not necessary. Also, unlike regular bins, they | |
2198 | are not even processed in FIFO order (they use faster LIFO) since | |
2199 | ordering doesn't much matter in the transient contexts in which | |
2200 | fastbins are normally used. | |
2201 | ||
2202 | Chunks in fastbins keep their inuse bit set, so they cannot | |
2203 | be consolidated with other free chunks. malloc_consolidate | |
2204 | releases all chunks in fastbins and consolidates them with | |
a9177ff5 | 2205 | other free chunks. |
fa8d436c | 2206 | */ |
f65fd747 | 2207 | |
fa8d436c | 2208 | typedef struct malloc_chunk* mfastbinptr; |
f65fd747 | 2209 | |
fa8d436c UD |
2210 | /* offset 2 to use otherwise unindexable first 2 bins */ |
2211 | #define fastbin_index(sz) ((((unsigned int)(sz)) >> 3) - 2) | |
f65fd747 | 2212 | |
fa8d436c UD |
2213 | /* The maximum fastbin request size we support */ |
2214 | #define MAX_FAST_SIZE 80 | |
f65fd747 | 2215 | |
fa8d436c | 2216 | #define NFASTBINS (fastbin_index(request2size(MAX_FAST_SIZE))+1) |
f65fd747 UD |
2217 | |
2218 | /* | |
fa8d436c UD |
2219 | FASTBIN_CONSOLIDATION_THRESHOLD is the size of a chunk in free() |
2220 | that triggers automatic consolidation of possibly-surrounding | |
2221 | fastbin chunks. This is a heuristic, so the exact value should not | |
2222 | matter too much. It is defined at half the default trim threshold as a | |
2223 | compromise heuristic to only attempt consolidation if it is likely | |
2224 | to lead to trimming. However, it is not dynamically tunable, since | |
a9177ff5 | 2225 | consolidation reduces fragmentation surrounding large chunks even |
fa8d436c | 2226 | if trimming is not used. |
f65fd747 UD |
2227 | */ |
2228 | ||
fa8d436c | 2229 | #define FASTBIN_CONSOLIDATION_THRESHOLD (65536UL) |
f65fd747 UD |
2230 | |
2231 | /* | |
a9177ff5 | 2232 | Since the lowest 2 bits in max_fast don't matter in size comparisons, |
fa8d436c | 2233 | they are used as flags. |
f65fd747 UD |
2234 | */ |
2235 | ||
fa8d436c UD |
2236 | /* |
2237 | FASTCHUNKS_BIT held in max_fast indicates that there are probably | |
2238 | some fastbin chunks. It is set true on entering a chunk into any | |
2239 | fastbin, and cleared only in malloc_consolidate. | |
f65fd747 | 2240 | |
fa8d436c UD |
2241 | The truth value is inverted so that have_fastchunks will be true |
2242 | upon startup (since statics are zero-filled), simplifying | |
2243 | initialization checks. | |
2244 | */ | |
f65fd747 | 2245 | |
fa8d436c | 2246 | #define FASTCHUNKS_BIT (1U) |
f65fd747 | 2247 | |
9bf248c6 UD |
2248 | #define have_fastchunks(M) (((M)->flags & FASTCHUNKS_BIT) == 0) |
2249 | #define clear_fastchunks(M) ((M)->flags |= FASTCHUNKS_BIT) | |
2250 | #define set_fastchunks(M) ((M)->flags &= ~FASTCHUNKS_BIT) | |
f65fd747 UD |
2251 | |
2252 | /* | |
fa8d436c UD |
2253 | NONCONTIGUOUS_BIT indicates that MORECORE does not return contiguous |
2254 | regions. Otherwise, contiguity is exploited in merging together, | |
2255 | when possible, results from consecutive MORECORE calls. | |
f65fd747 | 2256 | |
fa8d436c UD |
2257 | The initial value comes from MORECORE_CONTIGUOUS, but is |
2258 | changed dynamically if mmap is ever used as an sbrk substitute. | |
f65fd747 UD |
2259 | */ |
2260 | ||
fa8d436c | 2261 | #define NONCONTIGUOUS_BIT (2U) |
f65fd747 | 2262 | |
9bf248c6 UD |
2263 | #define contiguous(M) (((M)->flags & NONCONTIGUOUS_BIT) == 0) |
2264 | #define noncontiguous(M) (((M)->flags & NONCONTIGUOUS_BIT) != 0) | |
2265 | #define set_noncontiguous(M) ((M)->flags |= NONCONTIGUOUS_BIT) | |
2266 | #define set_contiguous(M) ((M)->flags &= ~NONCONTIGUOUS_BIT) | |
f65fd747 | 2267 | |
a9177ff5 RM |
2268 | /* |
2269 | Set value of max_fast. | |
fa8d436c UD |
2270 | Use impossibly small value if 0. |
2271 | Precondition: there are no existing fastbin chunks. | |
2272 | Setting the value clears fastchunk bit but preserves noncontiguous bit. | |
f65fd747 UD |
2273 | */ |
2274 | ||
9bf248c6 UD |
2275 | #define set_max_fast(s) \ |
2276 | global_max_fast = ((s) == 0)? SMALLBIN_WIDTH: request2size(s) | |
2277 | #define get_max_fast() global_max_fast | |
f65fd747 | 2278 | |
f65fd747 UD |
2279 | |
2280 | /* | |
fa8d436c | 2281 | ----------- Internal state representation and initialization ----------- |
f65fd747 UD |
2282 | */ |
2283 | ||
fa8d436c UD |
2284 | struct malloc_state { |
2285 | /* Serialize access. */ | |
2286 | mutex_t mutex; | |
9bf248c6 UD |
2287 | |
2288 | /* Flags (formerly in max_fast). */ | |
2289 | int flags; | |
f65fd747 | 2290 | |
4f27c496 | 2291 | #if THREAD_STATS |
fa8d436c UD |
2292 | /* Statistics for locking. Only used if THREAD_STATS is defined. */ |
2293 | long stat_lock_direct, stat_lock_loop, stat_lock_wait; | |
4f27c496 | 2294 | #endif |
f65fd747 | 2295 | |
fa8d436c UD |
2296 | /* Fastbins */ |
2297 | mfastbinptr fastbins[NFASTBINS]; | |
f65fd747 | 2298 | |
fa8d436c UD |
2299 | /* Base of the topmost chunk -- not otherwise kept in a bin */ |
2300 | mchunkptr top; | |
f65fd747 | 2301 | |
fa8d436c UD |
2302 | /* The remainder from the most recent split of a small request */ |
2303 | mchunkptr last_remainder; | |
f65fd747 | 2304 | |
fa8d436c | 2305 | /* Normal bins packed as described above */ |
41999a1a | 2306 | mchunkptr bins[NBINS * 2 - 2]; |
f65fd747 | 2307 | |
fa8d436c UD |
2308 | /* Bitmap of bins */ |
2309 | unsigned int binmap[BINMAPSIZE]; | |
f65fd747 | 2310 | |
fa8d436c UD |
2311 | /* Linked list */ |
2312 | struct malloc_state *next; | |
f65fd747 | 2313 | |
fa8d436c UD |
2314 | /* Memory allocated from the system in this arena. */ |
2315 | INTERNAL_SIZE_T system_mem; | |
2316 | INTERNAL_SIZE_T max_system_mem; | |
2317 | }; | |
f65fd747 | 2318 | |
fa8d436c UD |
2319 | struct malloc_par { |
2320 | /* Tunable parameters */ | |
2321 | unsigned long trim_threshold; | |
2322 | INTERNAL_SIZE_T top_pad; | |
2323 | INTERNAL_SIZE_T mmap_threshold; | |
2324 | ||
2325 | /* Memory map support */ | |
2326 | int n_mmaps; | |
2327 | int n_mmaps_max; | |
2328 | int max_n_mmaps; | |
1d05c2fb UD |
2329 | /* the mmap_threshold is dynamic, until the user sets |
2330 | it manually, at which point we need to disable any | |
2331 | dynamic behavior. */ | |
2332 | int no_dyn_threshold; | |
fa8d436c UD |
2333 | |
2334 | /* Cache malloc_getpagesize */ | |
a9177ff5 | 2335 | unsigned int pagesize; |
fa8d436c UD |
2336 | |
2337 | /* Statistics */ | |
2338 | INTERNAL_SIZE_T mmapped_mem; | |
2339 | /*INTERNAL_SIZE_T sbrked_mem;*/ | |
2340 | /*INTERNAL_SIZE_T max_sbrked_mem;*/ | |
2341 | INTERNAL_SIZE_T max_mmapped_mem; | |
2342 | INTERNAL_SIZE_T max_total_mem; /* only kept for NO_THREADS */ | |
2343 | ||
2344 | /* First address handed out by MORECORE/sbrk. */ | |
2345 | char* sbrk_base; | |
2346 | }; | |
f65fd747 | 2347 | |
fa8d436c UD |
2348 | /* There are several instances of this struct ("arenas") in this |
2349 | malloc. If you are adapting this malloc in a way that does NOT use | |
2350 | a static or mmapped malloc_state, you MUST explicitly zero-fill it | |
2351 | before using. This malloc relies on the property that malloc_state | |
2352 | is initialized to all zeroes (as is true of C statics). */ | |
f65fd747 | 2353 | |
fa8d436c | 2354 | static struct malloc_state main_arena; |
f65fd747 | 2355 | |
fa8d436c | 2356 | /* There is only one instance of the malloc parameters. */ |
f65fd747 | 2357 | |
fa8d436c | 2358 | static struct malloc_par mp_; |
f65fd747 | 2359 | |
9bf248c6 UD |
2360 | |
2361 | /* Maximum size of memory handled in fastbins. */ | |
2362 | static INTERNAL_SIZE_T global_max_fast; | |
2363 | ||
fa8d436c UD |
2364 | /* |
2365 | Initialize a malloc_state struct. | |
f65fd747 | 2366 | |
fa8d436c UD |
2367 | This is called only from within malloc_consolidate, which needs |
2368 | be called in the same contexts anyway. It is never called directly | |
2369 | outside of malloc_consolidate because some optimizing compilers try | |
2370 | to inline it at all call points, which turns out not to be an | |
2371 | optimization at all. (Inlining it in malloc_consolidate is fine though.) | |
2372 | */ | |
f65fd747 | 2373 | |
fa8d436c UD |
2374 | #if __STD_C |
2375 | static void malloc_init_state(mstate av) | |
2376 | #else | |
2377 | static void malloc_init_state(av) mstate av; | |
2378 | #endif | |
2379 | { | |
2380 | int i; | |
2381 | mbinptr bin; | |
a9177ff5 | 2382 | |
fa8d436c | 2383 | /* Establish circular links for normal bins */ |
a9177ff5 | 2384 | for (i = 1; i < NBINS; ++i) { |
fa8d436c UD |
2385 | bin = bin_at(av,i); |
2386 | bin->fd = bin->bk = bin; | |
2387 | } | |
f65fd747 | 2388 | |
fa8d436c UD |
2389 | #if MORECORE_CONTIGUOUS |
2390 | if (av != &main_arena) | |
2391 | #endif | |
2392 | set_noncontiguous(av); | |
9bf248c6 UD |
2393 | if (av == &main_arena) |
2394 | set_max_fast(DEFAULT_MXFAST); | |
2395 | av->flags |= FASTCHUNKS_BIT; | |
f65fd747 | 2396 | |
fa8d436c UD |
2397 | av->top = initial_top(av); |
2398 | } | |
e9b3e3c5 | 2399 | |
a9177ff5 | 2400 | /* |
fa8d436c UD |
2401 | Other internal utilities operating on mstates |
2402 | */ | |
f65fd747 | 2403 | |
fa8d436c UD |
2404 | #if __STD_C |
2405 | static Void_t* sYSMALLOc(INTERNAL_SIZE_T, mstate); | |
2406 | static int sYSTRIm(size_t, mstate); | |
2407 | static void malloc_consolidate(mstate); | |
88764ae2 | 2408 | #ifndef _LIBC |
fa8d436c | 2409 | static Void_t** iALLOc(mstate, size_t, size_t*, int, Void_t**); |
88764ae2 | 2410 | #endif |
831372e7 | 2411 | #else |
fa8d436c UD |
2412 | static Void_t* sYSMALLOc(); |
2413 | static int sYSTRIm(); | |
2414 | static void malloc_consolidate(); | |
2415 | static Void_t** iALLOc(); | |
831372e7 | 2416 | #endif |
7e3be507 | 2417 | |
404d4cef RM |
2418 | |
2419 | /* -------------- Early definitions for debugging hooks ---------------- */ | |
2420 | ||
2421 | /* Define and initialize the hook variables. These weak definitions must | |
2422 | appear before any use of the variables in a function (arena.c uses one). */ | |
2423 | #ifndef weak_variable | |
2424 | #ifndef _LIBC | |
2425 | #define weak_variable /**/ | |
2426 | #else | |
2427 | /* In GNU libc we want the hook variables to be weak definitions to | |
2428 | avoid a problem with Emacs. */ | |
2429 | #define weak_variable weak_function | |
2430 | #endif | |
2431 | #endif | |
2432 | ||
2433 | /* Forward declarations. */ | |
2434 | static Void_t* malloc_hook_ini __MALLOC_P ((size_t sz, | |
2435 | const __malloc_ptr_t caller)); | |
2436 | static Void_t* realloc_hook_ini __MALLOC_P ((Void_t* ptr, size_t sz, | |
2437 | const __malloc_ptr_t caller)); | |
2438 | static Void_t* memalign_hook_ini __MALLOC_P ((size_t alignment, size_t sz, | |
2439 | const __malloc_ptr_t caller)); | |
2440 | ||
06d6611a UD |
2441 | void weak_variable (*__malloc_initialize_hook) (void) = NULL; |
2442 | void weak_variable (*__free_hook) (__malloc_ptr_t __ptr, | |
2443 | const __malloc_ptr_t) = NULL; | |
404d4cef | 2444 | __malloc_ptr_t weak_variable (*__malloc_hook) |
06d6611a | 2445 | (size_t __size, const __malloc_ptr_t) = malloc_hook_ini; |
404d4cef | 2446 | __malloc_ptr_t weak_variable (*__realloc_hook) |
06d6611a | 2447 | (__malloc_ptr_t __ptr, size_t __size, const __malloc_ptr_t) |
404d4cef RM |
2448 | = realloc_hook_ini; |
2449 | __malloc_ptr_t weak_variable (*__memalign_hook) | |
06d6611a | 2450 | (size_t __alignment, size_t __size, const __malloc_ptr_t) |
404d4cef | 2451 | = memalign_hook_ini; |
06d6611a | 2452 | void weak_variable (*__after_morecore_hook) (void) = NULL; |
404d4cef RM |
2453 | |
2454 | ||
3e030bd5 UD |
2455 | /* ---------------- Error behavior ------------------------------------ */ |
2456 | ||
2457 | #ifndef DEFAULT_CHECK_ACTION | |
2458 | #define DEFAULT_CHECK_ACTION 3 | |
2459 | #endif | |
2460 | ||
2461 | static int check_action = DEFAULT_CHECK_ACTION; | |
2462 | ||
2463 | ||
854278df UD |
2464 | /* ------------------ Testing support ----------------------------------*/ |
2465 | ||
2466 | static int perturb_byte; | |
2467 | ||
2468 | #define alloc_perturb(p, n) memset (p, (perturb_byte ^ 0xff) & 0xff, n) | |
2469 | #define free_perturb(p, n) memset (p, perturb_byte & 0xff, n) | |
2470 | ||
2471 | ||
fa8d436c UD |
2472 | /* ------------------- Support for multiple arenas -------------------- */ |
2473 | #include "arena.c" | |
f65fd747 | 2474 | |
fa8d436c UD |
2475 | /* |
2476 | Debugging support | |
f65fd747 | 2477 | |
fa8d436c UD |
2478 | These routines make a number of assertions about the states |
2479 | of data structures that should be true at all times. If any | |
2480 | are not true, it's very likely that a user program has somehow | |
2481 | trashed memory. (It's also possible that there is a coding error | |
2482 | in malloc. In which case, please report it!) | |
2483 | */ | |
ee74a442 | 2484 | |
fa8d436c | 2485 | #if ! MALLOC_DEBUG |
d8f00d46 | 2486 | |
fa8d436c UD |
2487 | #define check_chunk(A,P) |
2488 | #define check_free_chunk(A,P) | |
2489 | #define check_inuse_chunk(A,P) | |
2490 | #define check_remalloced_chunk(A,P,N) | |
2491 | #define check_malloced_chunk(A,P,N) | |
2492 | #define check_malloc_state(A) | |
d8f00d46 | 2493 | |
fa8d436c | 2494 | #else |
ca34d7a7 | 2495 | |
fa8d436c UD |
2496 | #define check_chunk(A,P) do_check_chunk(A,P) |
2497 | #define check_free_chunk(A,P) do_check_free_chunk(A,P) | |
2498 | #define check_inuse_chunk(A,P) do_check_inuse_chunk(A,P) | |
2499 | #define check_remalloced_chunk(A,P,N) do_check_remalloced_chunk(A,P,N) | |
2500 | #define check_malloced_chunk(A,P,N) do_check_malloced_chunk(A,P,N) | |
2501 | #define check_malloc_state(A) do_check_malloc_state(A) | |
ca34d7a7 | 2502 | |
fa8d436c UD |
2503 | /* |
2504 | Properties of all chunks | |
2505 | */ | |
ca34d7a7 | 2506 | |
fa8d436c UD |
2507 | #if __STD_C |
2508 | static void do_check_chunk(mstate av, mchunkptr p) | |
2509 | #else | |
2510 | static void do_check_chunk(av, p) mstate av; mchunkptr p; | |
ca34d7a7 | 2511 | #endif |
ca34d7a7 | 2512 | { |
fa8d436c UD |
2513 | unsigned long sz = chunksize(p); |
2514 | /* min and max possible addresses assuming contiguous allocation */ | |
2515 | char* max_address = (char*)(av->top) + chunksize(av->top); | |
2516 | char* min_address = max_address - av->system_mem; | |
2517 | ||
2518 | if (!chunk_is_mmapped(p)) { | |
a9177ff5 | 2519 | |
fa8d436c UD |
2520 | /* Has legal address ... */ |
2521 | if (p != av->top) { | |
2522 | if (contiguous(av)) { | |
2523 | assert(((char*)p) >= min_address); | |
2524 | assert(((char*)p + sz) <= ((char*)(av->top))); | |
2525 | } | |
2526 | } | |
2527 | else { | |
2528 | /* top size is always at least MINSIZE */ | |
2529 | assert((unsigned long)(sz) >= MINSIZE); | |
2530 | /* top predecessor always marked inuse */ | |
2531 | assert(prev_inuse(p)); | |
2532 | } | |
a9177ff5 | 2533 | |
ca34d7a7 | 2534 | } |
fa8d436c UD |
2535 | else { |
2536 | #if HAVE_MMAP | |
2537 | /* address is outside main heap */ | |
2538 | if (contiguous(av) && av->top != initial_top(av)) { | |
2539 | assert(((char*)p) < min_address || ((char*)p) > max_address); | |
2540 | } | |
2541 | /* chunk is page-aligned */ | |
2542 | assert(((p->prev_size + sz) & (mp_.pagesize-1)) == 0); | |
2543 | /* mem is aligned */ | |
2544 | assert(aligned_OK(chunk2mem(p))); | |
2545 | #else | |
2546 | /* force an appropriate assert violation if debug set */ | |
2547 | assert(!chunk_is_mmapped(p)); | |
eb406346 | 2548 | #endif |
eb406346 | 2549 | } |
eb406346 UD |
2550 | } |
2551 | ||
fa8d436c UD |
2552 | /* |
2553 | Properties of free chunks | |
2554 | */ | |
ee74a442 | 2555 | |
fa8d436c UD |
2556 | #if __STD_C |
2557 | static void do_check_free_chunk(mstate av, mchunkptr p) | |
2558 | #else | |
2559 | static void do_check_free_chunk(av, p) mstate av; mchunkptr p; | |
10dc2a90 | 2560 | #endif |
67c94753 | 2561 | { |
fa8d436c UD |
2562 | INTERNAL_SIZE_T sz = p->size & ~(PREV_INUSE|NON_MAIN_ARENA); |
2563 | mchunkptr next = chunk_at_offset(p, sz); | |
67c94753 | 2564 | |
fa8d436c | 2565 | do_check_chunk(av, p); |
67c94753 | 2566 | |
fa8d436c UD |
2567 | /* Chunk must claim to be free ... */ |
2568 | assert(!inuse(p)); | |
2569 | assert (!chunk_is_mmapped(p)); | |
67c94753 | 2570 | |
fa8d436c UD |
2571 | /* Unless a special marker, must have OK fields */ |
2572 | if ((unsigned long)(sz) >= MINSIZE) | |
2573 | { | |
2574 | assert((sz & MALLOC_ALIGN_MASK) == 0); | |
2575 | assert(aligned_OK(chunk2mem(p))); | |
2576 | /* ... matching footer field */ | |
2577 | assert(next->prev_size == sz); | |
2578 | /* ... and is fully consolidated */ | |
2579 | assert(prev_inuse(p)); | |
2580 | assert (next == av->top || inuse(next)); | |
2581 | ||
2582 | /* ... and has minimally sane links */ | |
2583 | assert(p->fd->bk == p); | |
2584 | assert(p->bk->fd == p); | |
2585 | } | |
2586 | else /* markers are always of size SIZE_SZ */ | |
2587 | assert(sz == SIZE_SZ); | |
67c94753 | 2588 | } |
67c94753 | 2589 | |
fa8d436c UD |
2590 | /* |
2591 | Properties of inuse chunks | |
2592 | */ | |
2593 | ||
2594 | #if __STD_C | |
2595 | static void do_check_inuse_chunk(mstate av, mchunkptr p) | |
f65fd747 | 2596 | #else |
fa8d436c | 2597 | static void do_check_inuse_chunk(av, p) mstate av; mchunkptr p; |
f65fd747 UD |
2598 | #endif |
2599 | { | |
fa8d436c | 2600 | mchunkptr next; |
f65fd747 | 2601 | |
fa8d436c | 2602 | do_check_chunk(av, p); |
f65fd747 | 2603 | |
fa8d436c UD |
2604 | if (chunk_is_mmapped(p)) |
2605 | return; /* mmapped chunks have no next/prev */ | |
ca34d7a7 | 2606 | |
fa8d436c UD |
2607 | /* Check whether it claims to be in use ... */ |
2608 | assert(inuse(p)); | |
10dc2a90 | 2609 | |
fa8d436c | 2610 | next = next_chunk(p); |
10dc2a90 | 2611 | |
fa8d436c UD |
2612 | /* ... and is surrounded by OK chunks. |
2613 | Since more things can be checked with free chunks than inuse ones, | |
2614 | if an inuse chunk borders them and debug is on, it's worth doing them. | |
2615 | */ | |
2616 | if (!prev_inuse(p)) { | |
2617 | /* Note that we cannot even look at prev unless it is not inuse */ | |
2618 | mchunkptr prv = prev_chunk(p); | |
2619 | assert(next_chunk(prv) == p); | |
2620 | do_check_free_chunk(av, prv); | |
2621 | } | |
2622 | ||
2623 | if (next == av->top) { | |
2624 | assert(prev_inuse(next)); | |
2625 | assert(chunksize(next) >= MINSIZE); | |
2626 | } | |
2627 | else if (!inuse(next)) | |
2628 | do_check_free_chunk(av, next); | |
10dc2a90 UD |
2629 | } |
2630 | ||
fa8d436c UD |
2631 | /* |
2632 | Properties of chunks recycled from fastbins | |
2633 | */ | |
2634 | ||
10dc2a90 | 2635 | #if __STD_C |
fa8d436c | 2636 | static void do_check_remalloced_chunk(mstate av, mchunkptr p, INTERNAL_SIZE_T s) |
10dc2a90 | 2637 | #else |
fa8d436c UD |
2638 | static void do_check_remalloced_chunk(av, p, s) |
2639 | mstate av; mchunkptr p; INTERNAL_SIZE_T s; | |
a2b08ee5 | 2640 | #endif |
10dc2a90 | 2641 | { |
fa8d436c UD |
2642 | INTERNAL_SIZE_T sz = p->size & ~(PREV_INUSE|NON_MAIN_ARENA); |
2643 | ||
2644 | if (!chunk_is_mmapped(p)) { | |
2645 | assert(av == arena_for_chunk(p)); | |
2646 | if (chunk_non_main_arena(p)) | |
2647 | assert(av != &main_arena); | |
2648 | else | |
2649 | assert(av == &main_arena); | |
2650 | } | |
2651 | ||
2652 | do_check_inuse_chunk(av, p); | |
2653 | ||
2654 | /* Legal size ... */ | |
2655 | assert((sz & MALLOC_ALIGN_MASK) == 0); | |
2656 | assert((unsigned long)(sz) >= MINSIZE); | |
2657 | /* ... and alignment */ | |
2658 | assert(aligned_OK(chunk2mem(p))); | |
2659 | /* chunk is less than MINSIZE more than request */ | |
2660 | assert((long)(sz) - (long)(s) >= 0); | |
2661 | assert((long)(sz) - (long)(s + MINSIZE) < 0); | |
10dc2a90 UD |
2662 | } |
2663 | ||
fa8d436c UD |
2664 | /* |
2665 | Properties of nonrecycled chunks at the point they are malloced | |
2666 | */ | |
2667 | ||
10dc2a90 | 2668 | #if __STD_C |
fa8d436c | 2669 | static void do_check_malloced_chunk(mstate av, mchunkptr p, INTERNAL_SIZE_T s) |
10dc2a90 | 2670 | #else |
fa8d436c UD |
2671 | static void do_check_malloced_chunk(av, p, s) |
2672 | mstate av; mchunkptr p; INTERNAL_SIZE_T s; | |
a2b08ee5 | 2673 | #endif |
10dc2a90 | 2674 | { |
fa8d436c UD |
2675 | /* same as recycled case ... */ |
2676 | do_check_remalloced_chunk(av, p, s); | |
10dc2a90 | 2677 | |
fa8d436c UD |
2678 | /* |
2679 | ... plus, must obey implementation invariant that prev_inuse is | |
2680 | always true of any allocated chunk; i.e., that each allocated | |
2681 | chunk borders either a previously allocated and still in-use | |
2682 | chunk, or the base of its memory arena. This is ensured | |
2683 | by making all allocations from the the `lowest' part of any found | |
2684 | chunk. This does not necessarily hold however for chunks | |
2685 | recycled via fastbins. | |
2686 | */ | |
10dc2a90 | 2687 | |
fa8d436c UD |
2688 | assert(prev_inuse(p)); |
2689 | } | |
10dc2a90 | 2690 | |
f65fd747 | 2691 | |
fa8d436c UD |
2692 | /* |
2693 | Properties of malloc_state. | |
f65fd747 | 2694 | |
fa8d436c UD |
2695 | This may be useful for debugging malloc, as well as detecting user |
2696 | programmer errors that somehow write into malloc_state. | |
f65fd747 | 2697 | |
fa8d436c UD |
2698 | If you are extending or experimenting with this malloc, you can |
2699 | probably figure out how to hack this routine to print out or | |
2700 | display chunk addresses, sizes, bins, and other instrumentation. | |
2701 | */ | |
f65fd747 | 2702 | |
fa8d436c UD |
2703 | static void do_check_malloc_state(mstate av) |
2704 | { | |
2705 | int i; | |
2706 | mchunkptr p; | |
2707 | mchunkptr q; | |
2708 | mbinptr b; | |
2709 | unsigned int binbit; | |
2710 | int empty; | |
2711 | unsigned int idx; | |
2712 | INTERNAL_SIZE_T size; | |
2713 | unsigned long total = 0; | |
2714 | int max_fast_bin; | |
f65fd747 | 2715 | |
fa8d436c UD |
2716 | /* internal size_t must be no wider than pointer type */ |
2717 | assert(sizeof(INTERNAL_SIZE_T) <= sizeof(char*)); | |
f65fd747 | 2718 | |
fa8d436c UD |
2719 | /* alignment is a power of 2 */ |
2720 | assert((MALLOC_ALIGNMENT & (MALLOC_ALIGNMENT-1)) == 0); | |
f65fd747 | 2721 | |
fa8d436c UD |
2722 | /* cannot run remaining checks until fully initialized */ |
2723 | if (av->top == 0 || av->top == initial_top(av)) | |
2724 | return; | |
f65fd747 | 2725 | |
fa8d436c UD |
2726 | /* pagesize is a power of 2 */ |
2727 | assert((mp_.pagesize & (mp_.pagesize-1)) == 0); | |
f65fd747 | 2728 | |
fa8d436c UD |
2729 | /* A contiguous main_arena is consistent with sbrk_base. */ |
2730 | if (av == &main_arena && contiguous(av)) | |
2731 | assert((char*)mp_.sbrk_base + av->system_mem == | |
2732 | (char*)av->top + chunksize(av->top)); | |
2733 | ||
2734 | /* properties of fastbins */ | |
2735 | ||
2736 | /* max_fast is in allowed range */ | |
9bf248c6 | 2737 | assert((get_max_fast () & ~1) <= request2size(MAX_FAST_SIZE)); |
fa8d436c | 2738 | |
9bf248c6 | 2739 | max_fast_bin = fastbin_index(get_max_fast ()); |
fa8d436c UD |
2740 | |
2741 | for (i = 0; i < NFASTBINS; ++i) { | |
2742 | p = av->fastbins[i]; | |
2743 | ||
2744 | /* all bins past max_fast are empty */ | |
32c075e1 | 2745 | if (i > max_fast_bin) |
fa8d436c UD |
2746 | assert(p == 0); |
2747 | ||
2748 | while (p != 0) { | |
2749 | /* each chunk claims to be inuse */ | |
2750 | do_check_inuse_chunk(av, p); | |
2751 | total += chunksize(p); | |
2752 | /* chunk belongs in this bin */ | |
2753 | assert(fastbin_index(chunksize(p)) == i); | |
2754 | p = p->fd; | |
2755 | } | |
2756 | } | |
2757 | ||
2758 | if (total != 0) | |
2759 | assert(have_fastchunks(av)); | |
2760 | else if (!have_fastchunks(av)) | |
2761 | assert(total == 0); | |
2762 | ||
2763 | /* check normal bins */ | |
2764 | for (i = 1; i < NBINS; ++i) { | |
2765 | b = bin_at(av,i); | |
2766 | ||
2767 | /* binmap is accurate (except for bin 1 == unsorted_chunks) */ | |
2768 | if (i >= 2) { | |
2769 | binbit = get_binmap(av,i); | |
2770 | empty = last(b) == b; | |
2771 | if (!binbit) | |
2772 | assert(empty); | |
2773 | else if (!empty) | |
2774 | assert(binbit); | |
2775 | } | |
2776 | ||
2777 | for (p = last(b); p != b; p = p->bk) { | |
2778 | /* each chunk claims to be free */ | |
2779 | do_check_free_chunk(av, p); | |
2780 | size = chunksize(p); | |
2781 | total += size; | |
2782 | if (i >= 2) { | |
2783 | /* chunk belongs in bin */ | |
2784 | idx = bin_index(size); | |
2785 | assert(idx == i); | |
2786 | /* lists are sorted */ | |
a9177ff5 | 2787 | assert(p->bk == b || |
fa8d436c UD |
2788 | (unsigned long)chunksize(p->bk) >= (unsigned long)chunksize(p)); |
2789 | } | |
2790 | /* chunk is followed by a legal chain of inuse chunks */ | |
2791 | for (q = next_chunk(p); | |
a9177ff5 | 2792 | (q != av->top && inuse(q) && |
fa8d436c UD |
2793 | (unsigned long)(chunksize(q)) >= MINSIZE); |
2794 | q = next_chunk(q)) | |
2795 | do_check_inuse_chunk(av, q); | |
2796 | } | |
2797 | } | |
f65fd747 | 2798 | |
fa8d436c UD |
2799 | /* top chunk is OK */ |
2800 | check_chunk(av, av->top); | |
2801 | ||
2802 | /* sanity checks for statistics */ | |
2803 | ||
2804 | #ifdef NO_THREADS | |
2805 | assert(total <= (unsigned long)(mp_.max_total_mem)); | |
2806 | assert(mp_.n_mmaps >= 0); | |
f65fd747 | 2807 | #endif |
fa8d436c UD |
2808 | assert(mp_.n_mmaps <= mp_.n_mmaps_max); |
2809 | assert(mp_.n_mmaps <= mp_.max_n_mmaps); | |
2810 | ||
2811 | assert((unsigned long)(av->system_mem) <= | |
2812 | (unsigned long)(av->max_system_mem)); | |
f65fd747 | 2813 | |
fa8d436c UD |
2814 | assert((unsigned long)(mp_.mmapped_mem) <= |
2815 | (unsigned long)(mp_.max_mmapped_mem)); | |
2816 | ||
2817 | #ifdef NO_THREADS | |
2818 | assert((unsigned long)(mp_.max_total_mem) >= | |
2819 | (unsigned long)(mp_.mmapped_mem) + (unsigned long)(av->system_mem)); | |
dfd2257a | 2820 | #endif |
fa8d436c UD |
2821 | } |
2822 | #endif | |
2823 | ||
2824 | ||
2825 | /* ----------------- Support for debugging hooks -------------------- */ | |
2826 | #include "hooks.c" | |
2827 | ||
2828 | ||
2829 | /* ----------- Routines dealing with system allocation -------------- */ | |
2830 | ||
2831 | /* | |
2832 | sysmalloc handles malloc cases requiring more memory from the system. | |
2833 | On entry, it is assumed that av->top does not have enough | |
2834 | space to service request for nb bytes, thus requiring that av->top | |
2835 | be extended or replaced. | |
2836 | */ | |
2837 | ||
f65fd747 | 2838 | #if __STD_C |
fa8d436c | 2839 | static Void_t* sYSMALLOc(INTERNAL_SIZE_T nb, mstate av) |
f65fd747 | 2840 | #else |
fa8d436c | 2841 | static Void_t* sYSMALLOc(nb, av) INTERNAL_SIZE_T nb; mstate av; |
f65fd747 UD |
2842 | #endif |
2843 | { | |
fa8d436c UD |
2844 | mchunkptr old_top; /* incoming value of av->top */ |
2845 | INTERNAL_SIZE_T old_size; /* its size */ | |
2846 | char* old_end; /* its end address */ | |
f65fd747 | 2847 | |
fa8d436c UD |
2848 | long size; /* arg to first MORECORE or mmap call */ |
2849 | char* brk; /* return value from MORECORE */ | |
f65fd747 | 2850 | |
fa8d436c UD |
2851 | long correction; /* arg to 2nd MORECORE call */ |
2852 | char* snd_brk; /* 2nd return val */ | |
f65fd747 | 2853 | |
fa8d436c UD |
2854 | INTERNAL_SIZE_T front_misalign; /* unusable bytes at front of new space */ |
2855 | INTERNAL_SIZE_T end_misalign; /* partial page left at end of new space */ | |
2856 | char* aligned_brk; /* aligned offset into brk */ | |
f65fd747 | 2857 | |
fa8d436c UD |
2858 | mchunkptr p; /* the allocated/returned chunk */ |
2859 | mchunkptr remainder; /* remainder from allocation */ | |
2860 | unsigned long remainder_size; /* its size */ | |
2861 | ||
2862 | unsigned long sum; /* for updating stats */ | |
2863 | ||
2864 | size_t pagemask = mp_.pagesize - 1; | |
7463d5cb | 2865 | bool tried_mmap = false; |
fa8d436c UD |
2866 | |
2867 | ||
2868 | #if HAVE_MMAP | |
2869 | ||
2870 | /* | |
2871 | If have mmap, and the request size meets the mmap threshold, and | |
2872 | the system supports mmap, and there are few enough currently | |
2873 | allocated mmapped regions, try to directly map this request | |
2874 | rather than expanding top. | |
2875 | */ | |
f65fd747 | 2876 | |
fa8d436c UD |
2877 | if ((unsigned long)(nb) >= (unsigned long)(mp_.mmap_threshold) && |
2878 | (mp_.n_mmaps < mp_.n_mmaps_max)) { | |
f65fd747 | 2879 | |
fa8d436c UD |
2880 | char* mm; /* return value from mmap call*/ |
2881 | ||
e404fb16 | 2882 | try_mmap: |
fa8d436c UD |
2883 | /* |
2884 | Round up size to nearest page. For mmapped chunks, the overhead | |
2885 | is one SIZE_SZ unit larger than for normal chunks, because there | |
2886 | is no following chunk whose prev_size field could be used. | |
2887 | */ | |
2888 | size = (nb + SIZE_SZ + MALLOC_ALIGN_MASK + pagemask) & ~pagemask; | |
7463d5cb | 2889 | tried_mmap = true; |
fa8d436c UD |
2890 | |
2891 | /* Don't try if size wraps around 0 */ | |
2892 | if ((unsigned long)(size) > (unsigned long)(nb)) { | |
2893 | ||
2894 | mm = (char*)(MMAP(0, size, PROT_READ|PROT_WRITE, MAP_PRIVATE)); | |
a9177ff5 | 2895 | |
fa8d436c | 2896 | if (mm != MAP_FAILED) { |
a9177ff5 | 2897 | |
fa8d436c UD |
2898 | /* |
2899 | The offset to the start of the mmapped region is stored | |
2900 | in the prev_size field of the chunk. This allows us to adjust | |
a9177ff5 | 2901 | returned start address to meet alignment requirements here |
fa8d436c UD |
2902 | and in memalign(), and still be able to compute proper |
2903 | address argument for later munmap in free() and realloc(). | |
2904 | */ | |
a9177ff5 | 2905 | |
fa8d436c UD |
2906 | front_misalign = (INTERNAL_SIZE_T)chunk2mem(mm) & MALLOC_ALIGN_MASK; |
2907 | if (front_misalign > 0) { | |
2908 | correction = MALLOC_ALIGNMENT - front_misalign; | |
2909 | p = (mchunkptr)(mm + correction); | |
2910 | p->prev_size = correction; | |
2911 | set_head(p, (size - correction) |IS_MMAPPED); | |
2912 | } | |
32c075e1 JJ |
2913 | else { |
2914 | p = (mchunkptr)mm; | |
2915 | set_head(p, size|IS_MMAPPED); | |
2916 | } | |
a9177ff5 | 2917 | |
fa8d436c | 2918 | /* update statistics */ |
a9177ff5 RM |
2919 | |
2920 | if (++mp_.n_mmaps > mp_.max_n_mmaps) | |
fa8d436c | 2921 | mp_.max_n_mmaps = mp_.n_mmaps; |
a9177ff5 | 2922 | |
fa8d436c | 2923 | sum = mp_.mmapped_mem += size; |
a9177ff5 | 2924 | if (sum > (unsigned long)(mp_.max_mmapped_mem)) |
fa8d436c | 2925 | mp_.max_mmapped_mem = sum; |
8a4b65b4 | 2926 | #ifdef NO_THREADS |
fa8d436c | 2927 | sum += av->system_mem; |
a9177ff5 | 2928 | if (sum > (unsigned long)(mp_.max_total_mem)) |
fa8d436c | 2929 | mp_.max_total_mem = sum; |
8a4b65b4 | 2930 | #endif |
fa8d436c UD |
2931 | |
2932 | check_chunk(av, p); | |
a9177ff5 | 2933 | |
fa8d436c UD |
2934 | return chunk2mem(p); |
2935 | } | |
2936 | } | |
2937 | } | |
2938 | #endif | |
2939 | ||
2940 | /* Record incoming configuration of top */ | |
2941 | ||
2942 | old_top = av->top; | |
2943 | old_size = chunksize(old_top); | |
2944 | old_end = (char*)(chunk_at_offset(old_top, old_size)); | |
2945 | ||
a9177ff5 | 2946 | brk = snd_brk = (char*)(MORECORE_FAILURE); |
fa8d436c | 2947 | |
a9177ff5 | 2948 | /* |
fa8d436c UD |
2949 | If not the first time through, we require old_size to be |
2950 | at least MINSIZE and to have prev_inuse set. | |
2951 | */ | |
2952 | ||
a9177ff5 | 2953 | assert((old_top == initial_top(av) && old_size == 0) || |
fa8d436c UD |
2954 | ((unsigned long) (old_size) >= MINSIZE && |
2955 | prev_inuse(old_top) && | |
2956 | ((unsigned long)old_end & pagemask) == 0)); | |
2957 | ||
2958 | /* Precondition: not enough current space to satisfy nb request */ | |
2959 | assert((unsigned long)(old_size) < (unsigned long)(nb + MINSIZE)); | |
2960 | ||
2961 | /* Precondition: all fastbins are consolidated */ | |
2962 | assert(!have_fastchunks(av)); | |
2963 | ||
2964 | ||
2965 | if (av != &main_arena) { | |
2966 | ||
2967 | heap_info *old_heap, *heap; | |
2968 | size_t old_heap_size; | |
2969 | ||
2970 | /* First try to extend the current heap. */ | |
2971 | old_heap = heap_for_ptr(old_top); | |
2972 | old_heap_size = old_heap->size; | |
469615bd UD |
2973 | if ((long) (MINSIZE + nb - old_size) > 0 |
2974 | && grow_heap(old_heap, MINSIZE + nb - old_size) == 0) { | |
fa8d436c UD |
2975 | av->system_mem += old_heap->size - old_heap_size; |
2976 | arena_mem += old_heap->size - old_heap_size; | |
2977 | #if 0 | |
2978 | if(mmapped_mem + arena_mem + sbrked_mem > max_total_mem) | |
2979 | max_total_mem = mmapped_mem + arena_mem + sbrked_mem; | |
2980 | #endif | |
2981 | set_head(old_top, (((char *)old_heap + old_heap->size) - (char *)old_top) | |
2982 | | PREV_INUSE); | |
e6ac0e78 UD |
2983 | } |
2984 | else if ((heap = new_heap(nb + (MINSIZE + sizeof(*heap)), mp_.top_pad))) { | |
2985 | /* Use a newly allocated heap. */ | |
2986 | heap->ar_ptr = av; | |
2987 | heap->prev = old_heap; | |
2988 | av->system_mem += heap->size; | |
2989 | arena_mem += heap->size; | |
fa8d436c | 2990 | #if 0 |
e6ac0e78 UD |
2991 | if((unsigned long)(mmapped_mem + arena_mem + sbrked_mem) > max_total_mem) |
2992 | max_total_mem = mmapped_mem + arena_mem + sbrked_mem; | |
fa8d436c | 2993 | #endif |
fa8d436c UD |
2994 | /* Set up the new top. */ |
2995 | top(av) = chunk_at_offset(heap, sizeof(*heap)); | |
2996 | set_head(top(av), (heap->size - sizeof(*heap)) | PREV_INUSE); | |
2997 | ||
2998 | /* Setup fencepost and free the old top chunk. */ | |
2999 | /* The fencepost takes at least MINSIZE bytes, because it might | |
3000 | become the top chunk again later. Note that a footer is set | |
3001 | up, too, although the chunk is marked in use. */ | |
3002 | old_size -= MINSIZE; | |
3003 | set_head(chunk_at_offset(old_top, old_size + 2*SIZE_SZ), 0|PREV_INUSE); | |
3004 | if (old_size >= MINSIZE) { | |
3005 | set_head(chunk_at_offset(old_top, old_size), (2*SIZE_SZ)|PREV_INUSE); | |
3006 | set_foot(chunk_at_offset(old_top, old_size), (2*SIZE_SZ)); | |
3007 | set_head(old_top, old_size|PREV_INUSE|NON_MAIN_ARENA); | |
3008 | _int_free(av, chunk2mem(old_top)); | |
3009 | } else { | |
3010 | set_head(old_top, (old_size + 2*SIZE_SZ)|PREV_INUSE); | |
3011 | set_foot(old_top, (old_size + 2*SIZE_SZ)); | |
3012 | } | |
3013 | } | |
7463d5cb | 3014 | else if (!tried_mmap) |
e404fb16 UD |
3015 | /* We can at least try to use to mmap memory. */ |
3016 | goto try_mmap; | |
fa8d436c UD |
3017 | |
3018 | } else { /* av == main_arena */ | |
3019 | ||
3020 | ||
3021 | /* Request enough space for nb + pad + overhead */ | |
3022 | ||
3023 | size = nb + mp_.top_pad + MINSIZE; | |
3024 | ||
3025 | /* | |
3026 | If contiguous, we can subtract out existing space that we hope to | |
3027 | combine with new space. We add it back later only if | |
3028 | we don't actually get contiguous space. | |
3029 | */ | |
3030 | ||
3031 | if (contiguous(av)) | |
3032 | size -= old_size; | |
3033 | ||
3034 | /* | |
3035 | Round to a multiple of page size. | |
3036 | If MORECORE is not contiguous, this ensures that we only call it | |
3037 | with whole-page arguments. And if MORECORE is contiguous and | |
3038 | this is not first time through, this preserves page-alignment of | |
3039 | previous calls. Otherwise, we correct to page-align below. | |
3040 | */ | |
3041 | ||
3042 | size = (size + pagemask) & ~pagemask; | |
3043 | ||
3044 | /* | |
3045 | Don't try to call MORECORE if argument is so big as to appear | |
3046 | negative. Note that since mmap takes size_t arg, it may succeed | |
3047 | below even if we cannot call MORECORE. | |
3048 | */ | |
3049 | ||
a9177ff5 | 3050 | if (size > 0) |
fa8d436c UD |
3051 | brk = (char*)(MORECORE(size)); |
3052 | ||
3053 | if (brk != (char*)(MORECORE_FAILURE)) { | |
3054 | /* Call the `morecore' hook if necessary. */ | |
3055 | if (__after_morecore_hook) | |
3056 | (*__after_morecore_hook) (); | |
3057 | } else { | |
3058 | /* | |
3059 | If have mmap, try using it as a backup when MORECORE fails or | |
3060 | cannot be used. This is worth doing on systems that have "holes" in | |
3061 | address space, so sbrk cannot extend to give contiguous space, but | |
3062 | space is available elsewhere. Note that we ignore mmap max count | |
3063 | and threshold limits, since the space will not be used as a | |
3064 | segregated mmap region. | |
3065 | */ | |
3066 | ||
3067 | #if HAVE_MMAP | |
3068 | /* Cannot merge with old top, so add its size back in */ | |
3069 | if (contiguous(av)) | |
3070 | size = (size + old_size + pagemask) & ~pagemask; | |
3071 | ||
3072 | /* If we are relying on mmap as backup, then use larger units */ | |
3073 | if ((unsigned long)(size) < (unsigned long)(MMAP_AS_MORECORE_SIZE)) | |
3074 | size = MMAP_AS_MORECORE_SIZE; | |
3075 | ||
3076 | /* Don't try if size wraps around 0 */ | |
3077 | if ((unsigned long)(size) > (unsigned long)(nb)) { | |
3078 | ||
75bfdfc7 | 3079 | char *mbrk = (char*)(MMAP(0, size, PROT_READ|PROT_WRITE, MAP_PRIVATE)); |
a9177ff5 | 3080 | |
75bfdfc7 | 3081 | if (mbrk != MAP_FAILED) { |
a9177ff5 | 3082 | |
fa8d436c | 3083 | /* We do not need, and cannot use, another sbrk call to find end */ |
75bfdfc7 | 3084 | brk = mbrk; |
fa8d436c | 3085 | snd_brk = brk + size; |
a9177ff5 RM |
3086 | |
3087 | /* | |
3088 | Record that we no longer have a contiguous sbrk region. | |
fa8d436c UD |
3089 | After the first time mmap is used as backup, we do not |
3090 | ever rely on contiguous space since this could incorrectly | |
3091 | bridge regions. | |
3092 | */ | |
3093 | set_noncontiguous(av); | |
3094 | } | |
3095 | } | |
3096 | #endif | |
3097 | } | |
3098 | ||
3099 | if (brk != (char*)(MORECORE_FAILURE)) { | |
3100 | if (mp_.sbrk_base == 0) | |
3101 | mp_.sbrk_base = brk; | |
3102 | av->system_mem += size; | |
3103 | ||
3104 | /* | |
3105 | If MORECORE extends previous space, we can likewise extend top size. | |
3106 | */ | |
a9177ff5 | 3107 | |
fa8d436c UD |
3108 | if (brk == old_end && snd_brk == (char*)(MORECORE_FAILURE)) |
3109 | set_head(old_top, (size + old_size) | PREV_INUSE); | |
3110 | ||
886d5973 | 3111 | else if (contiguous(av) && old_size && brk < old_end) { |
fa8d436c UD |
3112 | /* Oops! Someone else killed our space.. Can't touch anything. */ |
3113 | assert(0); | |
3114 | } | |
3115 | ||
3116 | /* | |
3117 | Otherwise, make adjustments: | |
a9177ff5 | 3118 | |
fa8d436c UD |
3119 | * If the first time through or noncontiguous, we need to call sbrk |
3120 | just to find out where the end of memory lies. | |
3121 | ||
3122 | * We need to ensure that all returned chunks from malloc will meet | |
3123 | MALLOC_ALIGNMENT | |
3124 | ||
3125 | * If there was an intervening foreign sbrk, we need to adjust sbrk | |
3126 | request size to account for fact that we will not be able to | |
3127 | combine new space with existing space in old_top. | |
3128 | ||
3129 | * Almost all systems internally allocate whole pages at a time, in | |
3130 | which case we might as well use the whole last page of request. | |
3131 | So we allocate enough more memory to hit a page boundary now, | |
3132 | which in turn causes future contiguous calls to page-align. | |
3133 | */ | |
a9177ff5 | 3134 | |
fa8d436c | 3135 | else { |
fa8d436c UD |
3136 | front_misalign = 0; |
3137 | end_misalign = 0; | |
3138 | correction = 0; | |
3139 | aligned_brk = brk; | |
a9177ff5 | 3140 | |
fa8d436c | 3141 | /* handle contiguous cases */ |
a9177ff5 RM |
3142 | if (contiguous(av)) { |
3143 | ||
0cb71e02 UD |
3144 | /* Count foreign sbrk as system_mem. */ |
3145 | if (old_size) | |
3146 | av->system_mem += brk - old_end; | |
3147 | ||
fa8d436c UD |
3148 | /* Guarantee alignment of first new chunk made from this space */ |
3149 | ||
3150 | front_misalign = (INTERNAL_SIZE_T)chunk2mem(brk) & MALLOC_ALIGN_MASK; | |
3151 | if (front_misalign > 0) { | |
3152 | ||
3153 | /* | |
3154 | Skip over some bytes to arrive at an aligned position. | |
3155 | We don't need to specially mark these wasted front bytes. | |
3156 | They will never be accessed anyway because | |
3157 | prev_inuse of av->top (and any chunk created from its start) | |
3158 | is always true after initialization. | |
3159 | */ | |
3160 | ||
3161 | correction = MALLOC_ALIGNMENT - front_misalign; | |
3162 | aligned_brk += correction; | |
3163 | } | |
a9177ff5 | 3164 | |
fa8d436c UD |
3165 | /* |
3166 | If this isn't adjacent to existing space, then we will not | |
3167 | be able to merge with old_top space, so must add to 2nd request. | |
3168 | */ | |
a9177ff5 | 3169 | |
fa8d436c | 3170 | correction += old_size; |
a9177ff5 | 3171 | |
fa8d436c UD |
3172 | /* Extend the end address to hit a page boundary */ |
3173 | end_misalign = (INTERNAL_SIZE_T)(brk + size + correction); | |
3174 | correction += ((end_misalign + pagemask) & ~pagemask) - end_misalign; | |
a9177ff5 | 3175 | |
fa8d436c UD |
3176 | assert(correction >= 0); |
3177 | snd_brk = (char*)(MORECORE(correction)); | |
a9177ff5 | 3178 | |
fa8d436c UD |
3179 | /* |
3180 | If can't allocate correction, try to at least find out current | |
3181 | brk. It might be enough to proceed without failing. | |
a9177ff5 | 3182 | |
fa8d436c UD |
3183 | Note that if second sbrk did NOT fail, we assume that space |
3184 | is contiguous with first sbrk. This is a safe assumption unless | |
3185 | program is multithreaded but doesn't use locks and a foreign sbrk | |
3186 | occurred between our first and second calls. | |
3187 | */ | |
a9177ff5 | 3188 | |
fa8d436c UD |
3189 | if (snd_brk == (char*)(MORECORE_FAILURE)) { |
3190 | correction = 0; | |
3191 | snd_brk = (char*)(MORECORE(0)); | |
3192 | } else | |
3193 | /* Call the `morecore' hook if necessary. */ | |
3194 | if (__after_morecore_hook) | |
3195 | (*__after_morecore_hook) (); | |
3196 | } | |
a9177ff5 | 3197 | |
fa8d436c | 3198 | /* handle non-contiguous cases */ |
a9177ff5 | 3199 | else { |
fa8d436c UD |
3200 | /* MORECORE/mmap must correctly align */ |
3201 | assert(((unsigned long)chunk2mem(brk) & MALLOC_ALIGN_MASK) == 0); | |
a9177ff5 | 3202 | |
fa8d436c UD |
3203 | /* Find out current end of memory */ |
3204 | if (snd_brk == (char*)(MORECORE_FAILURE)) { | |
3205 | snd_brk = (char*)(MORECORE(0)); | |
3206 | } | |
3207 | } | |
a9177ff5 | 3208 | |
fa8d436c UD |
3209 | /* Adjust top based on results of second sbrk */ |
3210 | if (snd_brk != (char*)(MORECORE_FAILURE)) { | |
3211 | av->top = (mchunkptr)aligned_brk; | |
3212 | set_head(av->top, (snd_brk - aligned_brk + correction) | PREV_INUSE); | |
3213 | av->system_mem += correction; | |
a9177ff5 | 3214 | |
fa8d436c UD |
3215 | /* |
3216 | If not the first time through, we either have a | |
3217 | gap due to foreign sbrk or a non-contiguous region. Insert a | |
3218 | double fencepost at old_top to prevent consolidation with space | |
3219 | we don't own. These fenceposts are artificial chunks that are | |
3220 | marked as inuse and are in any case too small to use. We need | |
3221 | two to make sizes and alignments work out. | |
3222 | */ | |
a9177ff5 | 3223 | |
fa8d436c | 3224 | if (old_size != 0) { |
a9177ff5 | 3225 | /* |
fa8d436c UD |
3226 | Shrink old_top to insert fenceposts, keeping size a |
3227 | multiple of MALLOC_ALIGNMENT. We know there is at least | |
3228 | enough space in old_top to do this. | |
3229 | */ | |
3230 | old_size = (old_size - 4*SIZE_SZ) & ~MALLOC_ALIGN_MASK; | |
3231 | set_head(old_top, old_size | PREV_INUSE); | |
a9177ff5 | 3232 | |
fa8d436c UD |
3233 | /* |
3234 | Note that the following assignments completely overwrite | |
3235 | old_top when old_size was previously MINSIZE. This is | |
3236 | intentional. We need the fencepost, even if old_top otherwise gets | |
3237 | lost. | |
3238 | */ | |
3239 | chunk_at_offset(old_top, old_size )->size = | |
3240 | (2*SIZE_SZ)|PREV_INUSE; | |
3241 | ||
3242 | chunk_at_offset(old_top, old_size + 2*SIZE_SZ)->size = | |
3243 | (2*SIZE_SZ)|PREV_INUSE; | |
3244 | ||
3245 | /* If possible, release the rest. */ | |
3246 | if (old_size >= MINSIZE) { | |
3247 | _int_free(av, chunk2mem(old_top)); | |
3248 | } | |
3249 | ||
3250 | } | |
3251 | } | |
3252 | } | |
a9177ff5 | 3253 | |
fa8d436c UD |
3254 | /* Update statistics */ |
3255 | #ifdef NO_THREADS | |
3256 | sum = av->system_mem + mp_.mmapped_mem; | |
3257 | if (sum > (unsigned long)(mp_.max_total_mem)) | |
3258 | mp_.max_total_mem = sum; | |
3259 | #endif | |
3260 | ||
3261 | } | |
3262 | ||
3263 | } /* if (av != &main_arena) */ | |
3264 | ||
3265 | if ((unsigned long)av->system_mem > (unsigned long)(av->max_system_mem)) | |
3266 | av->max_system_mem = av->system_mem; | |
3267 | check_malloc_state(av); | |
a9177ff5 | 3268 | |
fa8d436c UD |
3269 | /* finally, do the allocation */ |
3270 | p = av->top; | |
3271 | size = chunksize(p); | |
3272 | ||
3273 | /* check that one of the above allocation paths succeeded */ | |
3274 | if ((unsigned long)(size) >= (unsigned long)(nb + MINSIZE)) { | |
3275 | remainder_size = size - nb; | |
3276 | remainder = chunk_at_offset(p, nb); | |
3277 | av->top = remainder; | |
3278 | set_head(p, nb | PREV_INUSE | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
3279 | set_head(remainder, remainder_size | PREV_INUSE); | |
3280 | check_malloced_chunk(av, p, nb); | |
3281 | return chunk2mem(p); | |
3282 | } | |
3283 | ||
3284 | /* catch all failure paths */ | |
3285 | MALLOC_FAILURE_ACTION; | |
3286 | return 0; | |
3287 | } | |
3288 | ||
3289 | ||
3290 | /* | |
3291 | sYSTRIm is an inverse of sorts to sYSMALLOc. It gives memory back | |
3292 | to the system (via negative arguments to sbrk) if there is unused | |
3293 | memory at the `high' end of the malloc pool. It is called | |
3294 | automatically by free() when top space exceeds the trim | |
3295 | threshold. It is also called by the public malloc_trim routine. It | |
3296 | returns 1 if it actually released any memory, else 0. | |
3297 | */ | |
3298 | ||
3299 | #if __STD_C | |
3300 | static int sYSTRIm(size_t pad, mstate av) | |
3301 | #else | |
3302 | static int sYSTRIm(pad, av) size_t pad; mstate av; | |
3303 | #endif | |
3304 | { | |
3305 | long top_size; /* Amount of top-most memory */ | |
3306 | long extra; /* Amount to release */ | |
3307 | long released; /* Amount actually released */ | |
3308 | char* current_brk; /* address returned by pre-check sbrk call */ | |
3309 | char* new_brk; /* address returned by post-check sbrk call */ | |
3310 | size_t pagesz; | |
3311 | ||
3312 | pagesz = mp_.pagesize; | |
3313 | top_size = chunksize(av->top); | |
a9177ff5 | 3314 | |
fa8d436c UD |
3315 | /* Release in pagesize units, keeping at least one page */ |
3316 | extra = ((top_size - pad - MINSIZE + (pagesz-1)) / pagesz - 1) * pagesz; | |
a9177ff5 | 3317 | |
fa8d436c | 3318 | if (extra > 0) { |
a9177ff5 | 3319 | |
fa8d436c UD |
3320 | /* |
3321 | Only proceed if end of memory is where we last set it. | |
3322 | This avoids problems if there were foreign sbrk calls. | |
3323 | */ | |
3324 | current_brk = (char*)(MORECORE(0)); | |
3325 | if (current_brk == (char*)(av->top) + top_size) { | |
a9177ff5 | 3326 | |
fa8d436c UD |
3327 | /* |
3328 | Attempt to release memory. We ignore MORECORE return value, | |
3329 | and instead call again to find out where new end of memory is. | |
3330 | This avoids problems if first call releases less than we asked, | |
3331 | of if failure somehow altered brk value. (We could still | |
3332 | encounter problems if it altered brk in some very bad way, | |
3333 | but the only thing we can do is adjust anyway, which will cause | |
3334 | some downstream failure.) | |
3335 | */ | |
a9177ff5 | 3336 | |
fa8d436c UD |
3337 | MORECORE(-extra); |
3338 | /* Call the `morecore' hook if necessary. */ | |
3339 | if (__after_morecore_hook) | |
3340 | (*__after_morecore_hook) (); | |
3341 | new_brk = (char*)(MORECORE(0)); | |
a9177ff5 | 3342 | |
fa8d436c UD |
3343 | if (new_brk != (char*)MORECORE_FAILURE) { |
3344 | released = (long)(current_brk - new_brk); | |
a9177ff5 | 3345 | |
fa8d436c UD |
3346 | if (released != 0) { |
3347 | /* Success. Adjust top. */ | |
3348 | av->system_mem -= released; | |
3349 | set_head(av->top, (top_size - released) | PREV_INUSE); | |
3350 | check_malloc_state(av); | |
3351 | return 1; | |
3352 | } | |
3353 | } | |
3354 | } | |
3355 | } | |
3356 | return 0; | |
f65fd747 UD |
3357 | } |
3358 | ||
fa8d436c UD |
3359 | #ifdef HAVE_MMAP |
3360 | ||
431c33c0 UD |
3361 | static void |
3362 | internal_function | |
f65fd747 | 3363 | #if __STD_C |
431c33c0 | 3364 | munmap_chunk(mchunkptr p) |
f65fd747 | 3365 | #else |
431c33c0 | 3366 | munmap_chunk(p) mchunkptr p; |
f65fd747 UD |
3367 | #endif |
3368 | { | |
3369 | INTERNAL_SIZE_T size = chunksize(p); | |
f65fd747 UD |
3370 | |
3371 | assert (chunk_is_mmapped(p)); | |
fa8d436c UD |
3372 | #if 0 |
3373 | assert(! ((char*)p >= mp_.sbrk_base && (char*)p < mp_.sbrk_base + mp_.sbrked_mem)); | |
3374 | assert((mp_.n_mmaps > 0)); | |
3375 | #endif | |
8e635611 UD |
3376 | |
3377 | uintptr_t block = (uintptr_t) p - p->prev_size; | |
3378 | size_t total_size = p->prev_size + size; | |
3379 | /* Unfortunately we have to do the compilers job by hand here. Normally | |
3380 | we would test BLOCK and TOTAL-SIZE separately for compliance with the | |
3381 | page size. But gcc does not recognize the optimization possibility | |
3382 | (in the moment at least) so we combine the two values into one before | |
3383 | the bit test. */ | |
3384 | if (__builtin_expect (((block | total_size) & (mp_.pagesize - 1)) != 0, 0)) | |
3385 | { | |
3386 | malloc_printerr (check_action, "munmap_chunk(): invalid pointer", | |
3387 | chunk2mem (p)); | |
3388 | return; | |
3389 | } | |
f65fd747 | 3390 | |
fa8d436c | 3391 | mp_.n_mmaps--; |
8e635611 | 3392 | mp_.mmapped_mem -= total_size; |
f65fd747 | 3393 | |
2182b1ea | 3394 | int ret __attribute__ ((unused)) = munmap((char *)block, total_size); |
f65fd747 UD |
3395 | |
3396 | /* munmap returns non-zero on failure */ | |
3397 | assert(ret == 0); | |
3398 | } | |
3399 | ||
3400 | #if HAVE_MREMAP | |
3401 | ||
431c33c0 UD |
3402 | static mchunkptr |
3403 | internal_function | |
f65fd747 | 3404 | #if __STD_C |
431c33c0 | 3405 | mremap_chunk(mchunkptr p, size_t new_size) |
f65fd747 | 3406 | #else |
431c33c0 | 3407 | mremap_chunk(p, new_size) mchunkptr p; size_t new_size; |
f65fd747 UD |
3408 | #endif |
3409 | { | |
fa8d436c | 3410 | size_t page_mask = mp_.pagesize - 1; |
f65fd747 UD |
3411 | INTERNAL_SIZE_T offset = p->prev_size; |
3412 | INTERNAL_SIZE_T size = chunksize(p); | |
3413 | char *cp; | |
3414 | ||
3415 | assert (chunk_is_mmapped(p)); | |
fa8d436c UD |
3416 | #if 0 |
3417 | assert(! ((char*)p >= mp_.sbrk_base && (char*)p < mp_.sbrk_base + mp_.sbrked_mem)); | |
3418 | assert((mp_.n_mmaps > 0)); | |
3419 | #endif | |
3420 | assert(((size + offset) & (mp_.pagesize-1)) == 0); | |
f65fd747 UD |
3421 | |
3422 | /* Note the extra SIZE_SZ overhead as in mmap_chunk(). */ | |
3423 | new_size = (new_size + offset + SIZE_SZ + page_mask) & ~page_mask; | |
3424 | ||
3425 | cp = (char *)mremap((char *)p - offset, size + offset, new_size, | |
3426 | MREMAP_MAYMOVE); | |
3427 | ||
431c33c0 | 3428 | if (cp == MAP_FAILED) return 0; |
f65fd747 UD |
3429 | |
3430 | p = (mchunkptr)(cp + offset); | |
3431 | ||
3432 | assert(aligned_OK(chunk2mem(p))); | |
3433 | ||
3434 | assert((p->prev_size == offset)); | |
3435 | set_head(p, (new_size - offset)|IS_MMAPPED); | |
3436 | ||
fa8d436c UD |
3437 | mp_.mmapped_mem -= size + offset; |
3438 | mp_.mmapped_mem += new_size; | |
3439 | if ((unsigned long)mp_.mmapped_mem > (unsigned long)mp_.max_mmapped_mem) | |
3440 | mp_.max_mmapped_mem = mp_.mmapped_mem; | |
8a4b65b4 | 3441 | #ifdef NO_THREADS |
fa8d436c UD |
3442 | if ((unsigned long)(mp_.mmapped_mem + arena_mem + main_arena.system_mem) > |
3443 | mp_.max_total_mem) | |
3444 | mp_.max_total_mem = mp_.mmapped_mem + arena_mem + main_arena.system_mem; | |
8a4b65b4 | 3445 | #endif |
f65fd747 UD |
3446 | return p; |
3447 | } | |
3448 | ||
3449 | #endif /* HAVE_MREMAP */ | |
3450 | ||
3451 | #endif /* HAVE_MMAP */ | |
3452 | ||
fa8d436c | 3453 | /*------------------------ Public wrappers. --------------------------------*/ |
f65fd747 | 3454 | |
fa8d436c UD |
3455 | Void_t* |
3456 | public_mALLOc(size_t bytes) | |
3457 | { | |
3458 | mstate ar_ptr; | |
3459 | Void_t *victim; | |
f65fd747 | 3460 | |
06d6611a | 3461 | __malloc_ptr_t (*hook) (size_t, __const __malloc_ptr_t) = __malloc_hook; |
fa8d436c UD |
3462 | if (hook != NULL) |
3463 | return (*hook)(bytes, RETURN_ADDRESS (0)); | |
f65fd747 | 3464 | |
fa8d436c UD |
3465 | arena_get(ar_ptr, bytes); |
3466 | if(!ar_ptr) | |
f65fd747 | 3467 | return 0; |
fa8d436c UD |
3468 | victim = _int_malloc(ar_ptr, bytes); |
3469 | if(!victim) { | |
3470 | /* Maybe the failure is due to running out of mmapped areas. */ | |
3471 | if(ar_ptr != &main_arena) { | |
3472 | (void)mutex_unlock(&ar_ptr->mutex); | |
3473 | (void)mutex_lock(&main_arena.mutex); | |
3474 | victim = _int_malloc(&main_arena, bytes); | |
3475 | (void)mutex_unlock(&main_arena.mutex); | |
3476 | } else { | |
3477 | #if USE_ARENAS | |
3478 | /* ... or sbrk() has failed and there is still a chance to mmap() */ | |
3479 | ar_ptr = arena_get2(ar_ptr->next ? ar_ptr : 0, bytes); | |
3480 | (void)mutex_unlock(&main_arena.mutex); | |
3481 | if(ar_ptr) { | |
3482 | victim = _int_malloc(ar_ptr, bytes); | |
3483 | (void)mutex_unlock(&ar_ptr->mutex); | |
3484 | } | |
3485 | #endif | |
60f0e64b | 3486 | } |
fa8d436c UD |
3487 | } else |
3488 | (void)mutex_unlock(&ar_ptr->mutex); | |
3489 | assert(!victim || chunk_is_mmapped(mem2chunk(victim)) || | |
3490 | ar_ptr == arena_for_chunk(mem2chunk(victim))); | |
3491 | return victim; | |
f65fd747 | 3492 | } |
aa420660 UD |
3493 | #ifdef libc_hidden_def |
3494 | libc_hidden_def(public_mALLOc) | |
3495 | #endif | |
f65fd747 | 3496 | |
fa8d436c UD |
3497 | void |
3498 | public_fREe(Void_t* mem) | |
f65fd747 | 3499 | { |
fa8d436c UD |
3500 | mstate ar_ptr; |
3501 | mchunkptr p; /* chunk corresponding to mem */ | |
3502 | ||
06d6611a | 3503 | void (*hook) (__malloc_ptr_t, __const __malloc_ptr_t) = __free_hook; |
fa8d436c UD |
3504 | if (hook != NULL) { |
3505 | (*hook)(mem, RETURN_ADDRESS (0)); | |
3506 | return; | |
f65fd747 | 3507 | } |
f65fd747 | 3508 | |
fa8d436c UD |
3509 | if (mem == 0) /* free(0) has no effect */ |
3510 | return; | |
f65fd747 | 3511 | |
fa8d436c | 3512 | p = mem2chunk(mem); |
f65fd747 | 3513 | |
fa8d436c UD |
3514 | #if HAVE_MMAP |
3515 | if (chunk_is_mmapped(p)) /* release mmapped memory. */ | |
3516 | { | |
1d05c2fb UD |
3517 | /* see if the dynamic brk/mmap threshold needs adjusting */ |
3518 | if (!mp_.no_dyn_threshold | |
3519 | && p->size > mp_.mmap_threshold | |
3520 | && p->size <= DEFAULT_MMAP_THRESHOLD_MAX) | |
3521 | { | |
3522 | mp_.mmap_threshold = chunksize (p); | |
3523 | mp_.trim_threshold = 2 * mp_.mmap_threshold; | |
3524 | } | |
fa8d436c UD |
3525 | munmap_chunk(p); |
3526 | return; | |
8a4b65b4 | 3527 | } |
f65fd747 | 3528 | #endif |
f65fd747 | 3529 | |
fa8d436c UD |
3530 | ar_ptr = arena_for_chunk(p); |
3531 | #if THREAD_STATS | |
3532 | if(!mutex_trylock(&ar_ptr->mutex)) | |
3533 | ++(ar_ptr->stat_lock_direct); | |
3534 | else { | |
3535 | (void)mutex_lock(&ar_ptr->mutex); | |
3536 | ++(ar_ptr->stat_lock_wait); | |
f65fd747 | 3537 | } |
f65fd747 | 3538 | #else |
fa8d436c | 3539 | (void)mutex_lock(&ar_ptr->mutex); |
f65fd747 | 3540 | #endif |
fa8d436c UD |
3541 | _int_free(ar_ptr, mem); |
3542 | (void)mutex_unlock(&ar_ptr->mutex); | |
f65fd747 | 3543 | } |
aa420660 UD |
3544 | #ifdef libc_hidden_def |
3545 | libc_hidden_def (public_fREe) | |
3546 | #endif | |
f65fd747 | 3547 | |
fa8d436c UD |
3548 | Void_t* |
3549 | public_rEALLOc(Void_t* oldmem, size_t bytes) | |
f65fd747 | 3550 | { |
fa8d436c UD |
3551 | mstate ar_ptr; |
3552 | INTERNAL_SIZE_T nb; /* padded request size */ | |
f65fd747 | 3553 | |
fa8d436c UD |
3554 | mchunkptr oldp; /* chunk corresponding to oldmem */ |
3555 | INTERNAL_SIZE_T oldsize; /* its size */ | |
8a4b65b4 | 3556 | |
fa8d436c | 3557 | Void_t* newp; /* chunk to return */ |
f65fd747 | 3558 | |
06d6611a | 3559 | __malloc_ptr_t (*hook) (__malloc_ptr_t, size_t, __const __malloc_ptr_t) = |
fa8d436c UD |
3560 | __realloc_hook; |
3561 | if (hook != NULL) | |
3562 | return (*hook)(oldmem, bytes, RETURN_ADDRESS (0)); | |
f65fd747 | 3563 | |
fa8d436c UD |
3564 | #if REALLOC_ZERO_BYTES_FREES |
3565 | if (bytes == 0 && oldmem != NULL) { public_fREe(oldmem); return 0; } | |
f65fd747 | 3566 | #endif |
f65fd747 | 3567 | |
fa8d436c UD |
3568 | /* realloc of null is supposed to be same as malloc */ |
3569 | if (oldmem == 0) return public_mALLOc(bytes); | |
f65fd747 | 3570 | |
fa8d436c UD |
3571 | oldp = mem2chunk(oldmem); |
3572 | oldsize = chunksize(oldp); | |
f65fd747 | 3573 | |
dc165f7b UD |
3574 | /* Little security check which won't hurt performance: the |
3575 | allocator never wrapps around at the end of the address space. | |
3576 | Therefore we can exclude some size values which might appear | |
3577 | here by accident or by "design" from some intruder. */ | |
3578 | if (__builtin_expect ((uintptr_t) oldp > (uintptr_t) -oldsize, 0) | |
073f560e | 3579 | || __builtin_expect (misaligned_chunk (oldp), 0)) |
dc165f7b UD |
3580 | { |
3581 | malloc_printerr (check_action, "realloc(): invalid pointer", oldmem); | |
3582 | return NULL; | |
3583 | } | |
3584 | ||
fa8d436c | 3585 | checked_request2size(bytes, nb); |
f65fd747 | 3586 | |
fa8d436c UD |
3587 | #if HAVE_MMAP |
3588 | if (chunk_is_mmapped(oldp)) | |
3589 | { | |
3590 | Void_t* newmem; | |
f65fd747 | 3591 | |
fa8d436c UD |
3592 | #if HAVE_MREMAP |
3593 | newp = mremap_chunk(oldp, nb); | |
3594 | if(newp) return chunk2mem(newp); | |
f65fd747 | 3595 | #endif |
fa8d436c UD |
3596 | /* Note the extra SIZE_SZ overhead. */ |
3597 | if(oldsize - SIZE_SZ >= nb) return oldmem; /* do nothing */ | |
3598 | /* Must alloc, copy, free. */ | |
3599 | newmem = public_mALLOc(bytes); | |
3600 | if (newmem == 0) return 0; /* propagate failure */ | |
3601 | MALLOC_COPY(newmem, oldmem, oldsize - 2*SIZE_SZ); | |
3602 | munmap_chunk(oldp); | |
3603 | return newmem; | |
3604 | } | |
dfd2257a | 3605 | #endif |
fa8d436c UD |
3606 | |
3607 | ar_ptr = arena_for_chunk(oldp); | |
3608 | #if THREAD_STATS | |
3609 | if(!mutex_trylock(&ar_ptr->mutex)) | |
3610 | ++(ar_ptr->stat_lock_direct); | |
3611 | else { | |
3612 | (void)mutex_lock(&ar_ptr->mutex); | |
3613 | ++(ar_ptr->stat_lock_wait); | |
3614 | } | |
f65fd747 | 3615 | #else |
fa8d436c | 3616 | (void)mutex_lock(&ar_ptr->mutex); |
f65fd747 | 3617 | #endif |
f65fd747 | 3618 | |
fa8d436c UD |
3619 | #ifndef NO_THREADS |
3620 | /* As in malloc(), remember this arena for the next allocation. */ | |
3621 | tsd_setspecific(arena_key, (Void_t *)ar_ptr); | |
f65fd747 UD |
3622 | #endif |
3623 | ||
fa8d436c | 3624 | newp = _int_realloc(ar_ptr, oldmem, bytes); |
f65fd747 | 3625 | |
fa8d436c UD |
3626 | (void)mutex_unlock(&ar_ptr->mutex); |
3627 | assert(!newp || chunk_is_mmapped(mem2chunk(newp)) || | |
3628 | ar_ptr == arena_for_chunk(mem2chunk(newp))); | |
07014fca UD |
3629 | |
3630 | if (newp == NULL) | |
3631 | { | |
3632 | /* Try harder to allocate memory in other arenas. */ | |
3633 | newp = public_mALLOc(bytes); | |
3634 | if (newp != NULL) | |
3635 | { | |
3636 | MALLOC_COPY (newp, oldmem, oldsize - 2 * SIZE_SZ); | |
3637 | #if THREAD_STATS | |
3638 | if(!mutex_trylock(&ar_ptr->mutex)) | |
3639 | ++(ar_ptr->stat_lock_direct); | |
3640 | else { | |
3641 | (void)mutex_lock(&ar_ptr->mutex); | |
3642 | ++(ar_ptr->stat_lock_wait); | |
3643 | } | |
3644 | #else | |
3645 | (void)mutex_lock(&ar_ptr->mutex); | |
3646 | #endif | |
3647 | _int_free(ar_ptr, oldmem); | |
3648 | (void)mutex_unlock(&ar_ptr->mutex); | |
3649 | } | |
3650 | } | |
3651 | ||
fa8d436c UD |
3652 | return newp; |
3653 | } | |
aa420660 UD |
3654 | #ifdef libc_hidden_def |
3655 | libc_hidden_def (public_rEALLOc) | |
3656 | #endif | |
f65fd747 | 3657 | |
fa8d436c UD |
3658 | Void_t* |
3659 | public_mEMALIGn(size_t alignment, size_t bytes) | |
3660 | { | |
3661 | mstate ar_ptr; | |
3662 | Void_t *p; | |
f65fd747 | 3663 | |
fa8d436c UD |
3664 | __malloc_ptr_t (*hook) __MALLOC_PMT ((size_t, size_t, |
3665 | __const __malloc_ptr_t)) = | |
3666 | __memalign_hook; | |
3667 | if (hook != NULL) | |
3668 | return (*hook)(alignment, bytes, RETURN_ADDRESS (0)); | |
f65fd747 | 3669 | |
fa8d436c UD |
3670 | /* If need less alignment than we give anyway, just relay to malloc */ |
3671 | if (alignment <= MALLOC_ALIGNMENT) return public_mALLOc(bytes); | |
1228ed5c | 3672 | |
fa8d436c UD |
3673 | /* Otherwise, ensure that it is at least a minimum chunk size */ |
3674 | if (alignment < MINSIZE) alignment = MINSIZE; | |
f65fd747 | 3675 | |
fa8d436c UD |
3676 | arena_get(ar_ptr, bytes + alignment + MINSIZE); |
3677 | if(!ar_ptr) | |
3678 | return 0; | |
3679 | p = _int_memalign(ar_ptr, alignment, bytes); | |
3680 | (void)mutex_unlock(&ar_ptr->mutex); | |
3681 | if(!p) { | |
3682 | /* Maybe the failure is due to running out of mmapped areas. */ | |
3683 | if(ar_ptr != &main_arena) { | |
3684 | (void)mutex_lock(&main_arena.mutex); | |
3685 | p = _int_memalign(&main_arena, alignment, bytes); | |
3686 | (void)mutex_unlock(&main_arena.mutex); | |
f65fd747 | 3687 | } else { |
e9b3e3c5 | 3688 | #if USE_ARENAS |
fa8d436c UD |
3689 | /* ... or sbrk() has failed and there is still a chance to mmap() */ |
3690 | ar_ptr = arena_get2(ar_ptr->next ? ar_ptr : 0, bytes); | |
3691 | if(ar_ptr) { | |
3692 | p = _int_memalign(ar_ptr, alignment, bytes); | |
3693 | (void)mutex_unlock(&ar_ptr->mutex); | |
3694 | } | |
e9b3e3c5 | 3695 | #endif |
f65fd747 UD |
3696 | } |
3697 | } | |
fa8d436c UD |
3698 | assert(!p || chunk_is_mmapped(mem2chunk(p)) || |
3699 | ar_ptr == arena_for_chunk(mem2chunk(p))); | |
3700 | return p; | |
f65fd747 | 3701 | } |
aa420660 UD |
3702 | #ifdef libc_hidden_def |
3703 | libc_hidden_def (public_mEMALIGn) | |
3704 | #endif | |
f65fd747 | 3705 | |
fa8d436c UD |
3706 | Void_t* |
3707 | public_vALLOc(size_t bytes) | |
3708 | { | |
3709 | mstate ar_ptr; | |
3710 | Void_t *p; | |
f65fd747 | 3711 | |
fa8d436c UD |
3712 | if(__malloc_initialized < 0) |
3713 | ptmalloc_init (); | |
8088488d UD |
3714 | |
3715 | __malloc_ptr_t (*hook) __MALLOC_PMT ((size_t, size_t, | |
3716 | __const __malloc_ptr_t)) = | |
3717 | __memalign_hook; | |
3718 | if (hook != NULL) | |
3719 | return (*hook)(mp_.pagesize, bytes, RETURN_ADDRESS (0)); | |
3720 | ||
fa8d436c UD |
3721 | arena_get(ar_ptr, bytes + mp_.pagesize + MINSIZE); |
3722 | if(!ar_ptr) | |
3723 | return 0; | |
3724 | p = _int_valloc(ar_ptr, bytes); | |
3725 | (void)mutex_unlock(&ar_ptr->mutex); | |
3726 | return p; | |
3727 | } | |
f65fd747 | 3728 | |
fa8d436c UD |
3729 | Void_t* |
3730 | public_pVALLOc(size_t bytes) | |
3731 | { | |
3732 | mstate ar_ptr; | |
3733 | Void_t *p; | |
f65fd747 | 3734 | |
fa8d436c UD |
3735 | if(__malloc_initialized < 0) |
3736 | ptmalloc_init (); | |
8088488d UD |
3737 | |
3738 | __malloc_ptr_t (*hook) __MALLOC_PMT ((size_t, size_t, | |
3739 | __const __malloc_ptr_t)) = | |
3740 | __memalign_hook; | |
3741 | if (hook != NULL) | |
3742 | return (*hook)(mp_.pagesize, | |
3743 | (bytes + mp_.pagesize - 1) & ~(mp_.pagesize - 1), | |
3744 | RETURN_ADDRESS (0)); | |
3745 | ||
fa8d436c UD |
3746 | arena_get(ar_ptr, bytes + 2*mp_.pagesize + MINSIZE); |
3747 | p = _int_pvalloc(ar_ptr, bytes); | |
3748 | (void)mutex_unlock(&ar_ptr->mutex); | |
3749 | return p; | |
3750 | } | |
f65fd747 | 3751 | |
fa8d436c UD |
3752 | Void_t* |
3753 | public_cALLOc(size_t n, size_t elem_size) | |
f65fd747 | 3754 | { |
fa8d436c UD |
3755 | mstate av; |
3756 | mchunkptr oldtop, p; | |
0950889b | 3757 | INTERNAL_SIZE_T bytes, sz, csz, oldtopsize; |
fa8d436c UD |
3758 | Void_t* mem; |
3759 | unsigned long clearsize; | |
3760 | unsigned long nclears; | |
3761 | INTERNAL_SIZE_T* d; | |
6c6bb055 | 3762 | __malloc_ptr_t (*hook) __MALLOC_PMT ((size_t, __const __malloc_ptr_t)) = |
fa8d436c | 3763 | __malloc_hook; |
0950889b UD |
3764 | |
3765 | /* size_t is unsigned so the behavior on overflow is defined. */ | |
3766 | bytes = n * elem_size; | |
d9af917d UD |
3767 | #define HALF_INTERNAL_SIZE_T \ |
3768 | (((INTERNAL_SIZE_T) 1) << (8 * sizeof (INTERNAL_SIZE_T) / 2)) | |
3769 | if (__builtin_expect ((n | elem_size) >= HALF_INTERNAL_SIZE_T, 0)) { | |
0be405c2 | 3770 | if (elem_size != 0 && bytes / elem_size != n) { |
d9af917d UD |
3771 | MALLOC_FAILURE_ACTION; |
3772 | return 0; | |
3773 | } | |
0950889b UD |
3774 | } |
3775 | ||
6c6bb055 | 3776 | if (hook != NULL) { |
0950889b | 3777 | sz = bytes; |
fa8d436c UD |
3778 | mem = (*hook)(sz, RETURN_ADDRESS (0)); |
3779 | if(mem == 0) | |
3780 | return 0; | |
3781 | #ifdef HAVE_MEMCPY | |
3782 | return memset(mem, 0, sz); | |
a2b08ee5 | 3783 | #else |
fa8d436c UD |
3784 | while(sz > 0) ((char*)mem)[--sz] = 0; /* rather inefficient */ |
3785 | return mem; | |
a2b08ee5 | 3786 | #endif |
10dc2a90 | 3787 | } |
10dc2a90 | 3788 | |
0950889b | 3789 | sz = bytes; |
fa8d436c UD |
3790 | |
3791 | arena_get(av, sz); | |
3792 | if(!av) | |
f65fd747 | 3793 | return 0; |
fa8d436c UD |
3794 | |
3795 | /* Check if we hand out the top chunk, in which case there may be no | |
3796 | need to clear. */ | |
3797 | #if MORECORE_CLEARS | |
3798 | oldtop = top(av); | |
3799 | oldtopsize = chunksize(top(av)); | |
3800 | #if MORECORE_CLEARS < 2 | |
3801 | /* Only newly allocated memory is guaranteed to be cleared. */ | |
3802 | if (av == &main_arena && | |
3803 | oldtopsize < mp_.sbrk_base + av->max_system_mem - (char *)oldtop) | |
3804 | oldtopsize = (mp_.sbrk_base + av->max_system_mem - (char *)oldtop); | |
3805 | #endif | |
3806 | #endif | |
3807 | mem = _int_malloc(av, sz); | |
3808 | ||
3809 | /* Only clearing follows, so we can unlock early. */ | |
3810 | (void)mutex_unlock(&av->mutex); | |
3811 | ||
3812 | assert(!mem || chunk_is_mmapped(mem2chunk(mem)) || | |
3813 | av == arena_for_chunk(mem2chunk(mem))); | |
3814 | ||
3815 | if (mem == 0) { | |
7799b7b3 | 3816 | /* Maybe the failure is due to running out of mmapped areas. */ |
fa8d436c | 3817 | if(av != &main_arena) { |
7799b7b3 | 3818 | (void)mutex_lock(&main_arena.mutex); |
fa8d436c | 3819 | mem = _int_malloc(&main_arena, sz); |
7799b7b3 | 3820 | (void)mutex_unlock(&main_arena.mutex); |
e9b3e3c5 UD |
3821 | } else { |
3822 | #if USE_ARENAS | |
3823 | /* ... or sbrk() has failed and there is still a chance to mmap() */ | |
fa8d436c UD |
3824 | (void)mutex_lock(&main_arena.mutex); |
3825 | av = arena_get2(av->next ? av : 0, sz); | |
e9b3e3c5 | 3826 | (void)mutex_unlock(&main_arena.mutex); |
fa8d436c UD |
3827 | if(av) { |
3828 | mem = _int_malloc(av, sz); | |
3829 | (void)mutex_unlock(&av->mutex); | |
e9b3e3c5 UD |
3830 | } |
3831 | #endif | |
7799b7b3 | 3832 | } |
fa8d436c UD |
3833 | if (mem == 0) return 0; |
3834 | } | |
3835 | p = mem2chunk(mem); | |
f65fd747 | 3836 | |
fa8d436c UD |
3837 | /* Two optional cases in which clearing not necessary */ |
3838 | #if HAVE_MMAP | |
9ea9af19 UD |
3839 | if (chunk_is_mmapped (p)) |
3840 | { | |
3841 | if (__builtin_expect (perturb_byte, 0)) | |
3842 | MALLOC_ZERO (mem, sz); | |
3843 | return mem; | |
3844 | } | |
f65fd747 | 3845 | #endif |
f65fd747 | 3846 | |
fa8d436c | 3847 | csz = chunksize(p); |
f65fd747 | 3848 | |
fa8d436c | 3849 | #if MORECORE_CLEARS |
56137dbc | 3850 | if (perturb_byte == 0 && (p == oldtop && csz > oldtopsize)) { |
fa8d436c UD |
3851 | /* clear only the bytes from non-freshly-sbrked memory */ |
3852 | csz = oldtopsize; | |
f65fd747 | 3853 | } |
fa8d436c | 3854 | #endif |
f65fd747 | 3855 | |
fa8d436c UD |
3856 | /* Unroll clear of <= 36 bytes (72 if 8byte sizes). We know that |
3857 | contents have an odd number of INTERNAL_SIZE_T-sized words; | |
3858 | minimally 3. */ | |
3859 | d = (INTERNAL_SIZE_T*)mem; | |
3860 | clearsize = csz - SIZE_SZ; | |
3861 | nclears = clearsize / sizeof(INTERNAL_SIZE_T); | |
3862 | assert(nclears >= 3); | |
f65fd747 | 3863 | |
fa8d436c UD |
3864 | if (nclears > 9) |
3865 | MALLOC_ZERO(d, clearsize); | |
f65fd747 | 3866 | |
fa8d436c UD |
3867 | else { |
3868 | *(d+0) = 0; | |
3869 | *(d+1) = 0; | |
3870 | *(d+2) = 0; | |
3871 | if (nclears > 4) { | |
3872 | *(d+3) = 0; | |
3873 | *(d+4) = 0; | |
3874 | if (nclears > 6) { | |
3875 | *(d+5) = 0; | |
3876 | *(d+6) = 0; | |
3877 | if (nclears > 8) { | |
3878 | *(d+7) = 0; | |
3879 | *(d+8) = 0; | |
3880 | } | |
f65fd747 UD |
3881 | } |
3882 | } | |
f65fd747 UD |
3883 | } |
3884 | ||
fa8d436c UD |
3885 | return mem; |
3886 | } | |
f65fd747 | 3887 | |
88764ae2 UD |
3888 | #ifndef _LIBC |
3889 | ||
fa8d436c UD |
3890 | Void_t** |
3891 | public_iCALLOc(size_t n, size_t elem_size, Void_t** chunks) | |
3892 | { | |
3893 | mstate ar_ptr; | |
3894 | Void_t** m; | |
f65fd747 | 3895 | |
fa8d436c UD |
3896 | arena_get(ar_ptr, n*elem_size); |
3897 | if(!ar_ptr) | |
3898 | return 0; | |
f65fd747 | 3899 | |
fa8d436c UD |
3900 | m = _int_icalloc(ar_ptr, n, elem_size, chunks); |
3901 | (void)mutex_unlock(&ar_ptr->mutex); | |
3902 | return m; | |
3903 | } | |
f65fd747 | 3904 | |
fa8d436c UD |
3905 | Void_t** |
3906 | public_iCOMALLOc(size_t n, size_t sizes[], Void_t** chunks) | |
3907 | { | |
3908 | mstate ar_ptr; | |
3909 | Void_t** m; | |
f65fd747 | 3910 | |
fa8d436c UD |
3911 | arena_get(ar_ptr, 0); |
3912 | if(!ar_ptr) | |
3913 | return 0; | |
f65fd747 | 3914 | |
fa8d436c UD |
3915 | m = _int_icomalloc(ar_ptr, n, sizes, chunks); |
3916 | (void)mutex_unlock(&ar_ptr->mutex); | |
3917 | return m; | |
3918 | } | |
f65fd747 | 3919 | |
fa8d436c UD |
3920 | void |
3921 | public_cFREe(Void_t* m) | |
3922 | { | |
3923 | public_fREe(m); | |
3924 | } | |
f65fd747 | 3925 | |
fa8d436c | 3926 | #endif /* _LIBC */ |
f65fd747 | 3927 | |
fa8d436c UD |
3928 | int |
3929 | public_mTRIm(size_t s) | |
3930 | { | |
3931 | int result; | |
f65fd747 | 3932 | |
88764ae2 UD |
3933 | if(__malloc_initialized < 0) |
3934 | ptmalloc_init (); | |
fa8d436c UD |
3935 | (void)mutex_lock(&main_arena.mutex); |
3936 | result = mTRIm(s); | |
3937 | (void)mutex_unlock(&main_arena.mutex); | |
3938 | return result; | |
3939 | } | |
f65fd747 | 3940 | |
fa8d436c UD |
3941 | size_t |
3942 | public_mUSABLe(Void_t* m) | |
3943 | { | |
3944 | size_t result; | |
f65fd747 | 3945 | |
fa8d436c UD |
3946 | result = mUSABLe(m); |
3947 | return result; | |
3948 | } | |
f65fd747 | 3949 | |
fa8d436c UD |
3950 | void |
3951 | public_mSTATs() | |
3952 | { | |
3953 | mSTATs(); | |
3954 | } | |
f65fd747 | 3955 | |
fa8d436c UD |
3956 | struct mallinfo public_mALLINFo() |
3957 | { | |
3958 | struct mallinfo m; | |
f65fd747 | 3959 | |
6a00759b UD |
3960 | if(__malloc_initialized < 0) |
3961 | ptmalloc_init (); | |
fa8d436c UD |
3962 | (void)mutex_lock(&main_arena.mutex); |
3963 | m = mALLINFo(&main_arena); | |
3964 | (void)mutex_unlock(&main_arena.mutex); | |
3965 | return m; | |
f65fd747 UD |
3966 | } |
3967 | ||
fa8d436c UD |
3968 | int |
3969 | public_mALLOPt(int p, int v) | |
3970 | { | |
3971 | int result; | |
3972 | result = mALLOPt(p, v); | |
3973 | return result; | |
3974 | } | |
f65fd747 UD |
3975 | |
3976 | /* | |
fa8d436c | 3977 | ------------------------------ malloc ------------------------------ |
f65fd747 UD |
3978 | */ |
3979 | ||
f1c5213d | 3980 | Void_t* |
fa8d436c | 3981 | _int_malloc(mstate av, size_t bytes) |
f65fd747 | 3982 | { |
fa8d436c UD |
3983 | INTERNAL_SIZE_T nb; /* normalized request size */ |
3984 | unsigned int idx; /* associated bin index */ | |
3985 | mbinptr bin; /* associated bin */ | |
3986 | mfastbinptr* fb; /* associated fastbin */ | |
f65fd747 | 3987 | |
fa8d436c UD |
3988 | mchunkptr victim; /* inspected/selected chunk */ |
3989 | INTERNAL_SIZE_T size; /* its size */ | |
3990 | int victim_index; /* its bin index */ | |
f65fd747 | 3991 | |
fa8d436c UD |
3992 | mchunkptr remainder; /* remainder from a split */ |
3993 | unsigned long remainder_size; /* its size */ | |
8a4b65b4 | 3994 | |
fa8d436c UD |
3995 | unsigned int block; /* bit map traverser */ |
3996 | unsigned int bit; /* bit map traverser */ | |
3997 | unsigned int map; /* current word of binmap */ | |
8a4b65b4 | 3998 | |
fa8d436c UD |
3999 | mchunkptr fwd; /* misc temp for linking */ |
4000 | mchunkptr bck; /* misc temp for linking */ | |
8a4b65b4 | 4001 | |
fa8d436c UD |
4002 | /* |
4003 | Convert request size to internal form by adding SIZE_SZ bytes | |
4004 | overhead plus possibly more to obtain necessary alignment and/or | |
4005 | to obtain a size of at least MINSIZE, the smallest allocatable | |
4006 | size. Also, checked_request2size traps (returning 0) request sizes | |
4007 | that are so large that they wrap around zero when padded and | |
4008 | aligned. | |
4009 | */ | |
f65fd747 | 4010 | |
fa8d436c | 4011 | checked_request2size(bytes, nb); |
f65fd747 | 4012 | |
fa8d436c UD |
4013 | /* |
4014 | If the size qualifies as a fastbin, first check corresponding bin. | |
4015 | This code is safe to execute even if av is not yet initialized, so we | |
4016 | can try it without checking, which saves some time on this fast path. | |
4017 | */ | |
f65fd747 | 4018 | |
9bf248c6 | 4019 | if ((unsigned long)(nb) <= (unsigned long)(get_max_fast ())) { |
6cce6540 UD |
4020 | long int idx = fastbin_index(nb); |
4021 | fb = &(av->fastbins[idx]); | |
fa8d436c | 4022 | if ( (victim = *fb) != 0) { |
6cce6540 UD |
4023 | if (__builtin_expect (fastbin_index (chunksize (victim)) != idx, 0)) |
4024 | malloc_printerr (check_action, "malloc(): memory corruption (fast)", | |
4025 | chunk2mem (victim)); | |
fa8d436c UD |
4026 | *fb = victim->fd; |
4027 | check_remalloced_chunk(av, victim, nb); | |
854278df UD |
4028 | void *p = chunk2mem(victim); |
4029 | if (__builtin_expect (perturb_byte, 0)) | |
4030 | alloc_perturb (p, bytes); | |
4031 | return p; | |
fa8d436c | 4032 | } |
f65fd747 UD |
4033 | } |
4034 | ||
fa8d436c UD |
4035 | /* |
4036 | If a small request, check regular bin. Since these "smallbins" | |
4037 | hold one size each, no searching within bins is necessary. | |
4038 | (For a large request, we need to wait until unsorted chunks are | |
4039 | processed to find best fit. But for small ones, fits are exact | |
4040 | anyway, so we can check now, which is faster.) | |
4041 | */ | |
f65fd747 | 4042 | |
fa8d436c UD |
4043 | if (in_smallbin_range(nb)) { |
4044 | idx = smallbin_index(nb); | |
4045 | bin = bin_at(av,idx); | |
7799b7b3 | 4046 | |
fa8d436c UD |
4047 | if ( (victim = last(bin)) != bin) { |
4048 | if (victim == 0) /* initialization check */ | |
4049 | malloc_consolidate(av); | |
4050 | else { | |
4051 | bck = victim->bk; | |
4052 | set_inuse_bit_at_offset(victim, nb); | |
4053 | bin->bk = bck; | |
4054 | bck->fd = bin; | |
4055 | ||
4056 | if (av != &main_arena) | |
4057 | victim->size |= NON_MAIN_ARENA; | |
4058 | check_malloced_chunk(av, victim, nb); | |
854278df UD |
4059 | void *p = chunk2mem(victim); |
4060 | if (__builtin_expect (perturb_byte, 0)) | |
4061 | alloc_perturb (p, bytes); | |
4062 | return p; | |
fa8d436c UD |
4063 | } |
4064 | } | |
f65fd747 UD |
4065 | } |
4066 | ||
a9177ff5 | 4067 | /* |
fa8d436c UD |
4068 | If this is a large request, consolidate fastbins before continuing. |
4069 | While it might look excessive to kill all fastbins before | |
4070 | even seeing if there is space available, this avoids | |
4071 | fragmentation problems normally associated with fastbins. | |
4072 | Also, in practice, programs tend to have runs of either small or | |
a9177ff5 | 4073 | large requests, but less often mixtures, so consolidation is not |
fa8d436c UD |
4074 | invoked all that often in most programs. And the programs that |
4075 | it is called frequently in otherwise tend to fragment. | |
4076 | */ | |
7799b7b3 | 4077 | |
fa8d436c UD |
4078 | else { |
4079 | idx = largebin_index(nb); | |
a9177ff5 | 4080 | if (have_fastchunks(av)) |
fa8d436c | 4081 | malloc_consolidate(av); |
7799b7b3 | 4082 | } |
f65fd747 | 4083 | |
fa8d436c UD |
4084 | /* |
4085 | Process recently freed or remaindered chunks, taking one only if | |
4086 | it is exact fit, or, if this a small request, the chunk is remainder from | |
4087 | the most recent non-exact fit. Place other traversed chunks in | |
4088 | bins. Note that this step is the only place in any routine where | |
4089 | chunks are placed in bins. | |
4090 | ||
4091 | The outer loop here is needed because we might not realize until | |
4092 | near the end of malloc that we should have consolidated, so must | |
4093 | do so and retry. This happens at most once, and only when we would | |
4094 | otherwise need to expand memory to service a "small" request. | |
4095 | */ | |
a9177ff5 RM |
4096 | |
4097 | for(;;) { | |
4098 | ||
72320021 | 4099 | int iters = 0; |
32c075e1 | 4100 | bool any_larger = false; |
fa8d436c UD |
4101 | while ( (victim = unsorted_chunks(av)->bk) != unsorted_chunks(av)) { |
4102 | bck = victim->bk; | |
6cce6540 UD |
4103 | if (__builtin_expect (victim->size <= 2 * SIZE_SZ, 0) |
4104 | || __builtin_expect (victim->size > av->system_mem, 0)) | |
4105 | malloc_printerr (check_action, "malloc(): memory corruption", | |
4106 | chunk2mem (victim)); | |
fa8d436c UD |
4107 | size = chunksize(victim); |
4108 | ||
a9177ff5 | 4109 | /* |
fa8d436c UD |
4110 | If a small request, try to use last remainder if it is the |
4111 | only chunk in unsorted bin. This helps promote locality for | |
4112 | runs of consecutive small requests. This is the only | |
4113 | exception to best-fit, and applies only when there is | |
4114 | no exact fit for a small chunk. | |
4115 | */ | |
4116 | ||
a9177ff5 | 4117 | if (in_smallbin_range(nb) && |
fa8d436c UD |
4118 | bck == unsorted_chunks(av) && |
4119 | victim == av->last_remainder && | |
4120 | (unsigned long)(size) > (unsigned long)(nb + MINSIZE)) { | |
4121 | ||
4122 | /* split and reattach remainder */ | |
4123 | remainder_size = size - nb; | |
4124 | remainder = chunk_at_offset(victim, nb); | |
4125 | unsorted_chunks(av)->bk = unsorted_chunks(av)->fd = remainder; | |
a9177ff5 | 4126 | av->last_remainder = remainder; |
fa8d436c | 4127 | remainder->bk = remainder->fd = unsorted_chunks(av); |
a9177ff5 | 4128 | |
fa8d436c UD |
4129 | set_head(victim, nb | PREV_INUSE | |
4130 | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
4131 | set_head(remainder, remainder_size | PREV_INUSE); | |
4132 | set_foot(remainder, remainder_size); | |
a9177ff5 | 4133 | |
fa8d436c | 4134 | check_malloced_chunk(av, victim, nb); |
854278df UD |
4135 | void *p = chunk2mem(victim); |
4136 | if (__builtin_expect (perturb_byte, 0)) | |
4137 | alloc_perturb (p, bytes); | |
4138 | return p; | |
fa8d436c | 4139 | } |
f65fd747 | 4140 | |
fa8d436c UD |
4141 | /* remove from unsorted list */ |
4142 | unsorted_chunks(av)->bk = bck; | |
4143 | bck->fd = unsorted_chunks(av); | |
a9177ff5 | 4144 | |
fa8d436c | 4145 | /* Take now instead of binning if exact fit */ |
a9177ff5 | 4146 | |
fa8d436c UD |
4147 | if (size == nb) { |
4148 | set_inuse_bit_at_offset(victim, size); | |
4149 | if (av != &main_arena) | |
4150 | victim->size |= NON_MAIN_ARENA; | |
4151 | check_malloced_chunk(av, victim, nb); | |
854278df UD |
4152 | void *p = chunk2mem(victim); |
4153 | if (__builtin_expect (perturb_byte, 0)) | |
4154 | alloc_perturb (p, bytes); | |
4155 | return p; | |
fa8d436c | 4156 | } |
a9177ff5 | 4157 | |
fa8d436c | 4158 | /* place chunk in bin */ |
a9177ff5 | 4159 | |
fa8d436c UD |
4160 | if (in_smallbin_range(size)) { |
4161 | victim_index = smallbin_index(size); | |
4162 | bck = bin_at(av, victim_index); | |
4163 | fwd = bck->fd; | |
4164 | } | |
4165 | else { | |
4166 | victim_index = largebin_index(size); | |
4167 | bck = bin_at(av, victim_index); | |
4168 | fwd = bck->fd; | |
4169 | ||
4170 | /* maintain large bins in sorted order */ | |
4171 | if (fwd != bck) { | |
4172 | /* Or with inuse bit to speed comparisons */ | |
4173 | size |= PREV_INUSE; | |
4174 | /* if smaller than smallest, bypass loop below */ | |
4175 | assert((bck->bk->size & NON_MAIN_ARENA) == 0); | |
4176 | if ((unsigned long)(size) <= (unsigned long)(bck->bk->size)) { | |
4177 | fwd = bck; | |
4178 | bck = bck->bk; | |
4179 | } | |
4180 | else { | |
4181 | assert((fwd->size & NON_MAIN_ARENA) == 0); | |
4182 | while ((unsigned long)(size) < (unsigned long)(fwd->size)) { | |
4183 | fwd = fwd->fd; | |
4184 | assert((fwd->size & NON_MAIN_ARENA) == 0); | |
4185 | } | |
4186 | bck = fwd->bk; | |
4187 | } | |
4188 | } | |
4189 | } | |
a9177ff5 | 4190 | |
fa8d436c UD |
4191 | mark_bin(av, victim_index); |
4192 | victim->bk = bck; | |
4193 | victim->fd = fwd; | |
4194 | fwd->bk = victim; | |
4195 | bck->fd = victim; | |
3997b7c4 | 4196 | |
32c075e1 JJ |
4197 | if (size >= nb + MINSIZE) |
4198 | any_larger = true; | |
3997b7c4 UD |
4199 | #define MAX_ITERS 10000 |
4200 | if (++iters >= MAX_ITERS) | |
4201 | break; | |
fa8d436c | 4202 | } |
a9177ff5 | 4203 | |
fa8d436c UD |
4204 | /* |
4205 | If a large request, scan through the chunks of current bin in | |
4206 | sorted order to find smallest that fits. This is the only step | |
4207 | where an unbounded number of chunks might be scanned without doing | |
4208 | anything useful with them. However the lists tend to be short. | |
4209 | */ | |
a9177ff5 | 4210 | |
fa8d436c UD |
4211 | if (!in_smallbin_range(nb)) { |
4212 | bin = bin_at(av, idx); | |
f65fd747 | 4213 | |
fa8d436c UD |
4214 | /* skip scan if empty or largest chunk is too small */ |
4215 | if ((victim = last(bin)) != bin && | |
4216 | (unsigned long)(first(bin)->size) >= (unsigned long)(nb)) { | |
f65fd747 | 4217 | |
a9177ff5 | 4218 | while (((unsigned long)(size = chunksize(victim)) < |
fa8d436c UD |
4219 | (unsigned long)(nb))) |
4220 | victim = victim->bk; | |
f65fd747 | 4221 | |
fa8d436c UD |
4222 | remainder_size = size - nb; |
4223 | unlink(victim, bck, fwd); | |
a9177ff5 | 4224 | |
fa8d436c UD |
4225 | /* Exhaust */ |
4226 | if (remainder_size < MINSIZE) { | |
4227 | set_inuse_bit_at_offset(victim, size); | |
4228 | if (av != &main_arena) | |
4229 | victim->size |= NON_MAIN_ARENA; | |
fa8d436c UD |
4230 | } |
4231 | /* Split */ | |
4232 | else { | |
4233 | remainder = chunk_at_offset(victim, nb); | |
b80770b2 UD |
4234 | /* We cannot assume the unsorted list is empty and therefore |
4235 | have to perform a complete insert here. */ | |
4236 | bck = unsorted_chunks(av); | |
4237 | fwd = bck->fd; | |
4238 | remainder->bk = bck; | |
4239 | remainder->fd = fwd; | |
4240 | bck->fd = remainder; | |
4241 | fwd->bk = remainder; | |
fa8d436c UD |
4242 | set_head(victim, nb | PREV_INUSE | |
4243 | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
4244 | set_head(remainder, remainder_size | PREV_INUSE); | |
4245 | set_foot(remainder, remainder_size); | |
a9177ff5 | 4246 | } |
854278df UD |
4247 | check_malloced_chunk(av, victim, nb); |
4248 | void *p = chunk2mem(victim); | |
4249 | if (__builtin_expect (perturb_byte, 0)) | |
4250 | alloc_perturb (p, bytes); | |
4251 | return p; | |
fa8d436c | 4252 | } |
a9177ff5 | 4253 | } |
f65fd747 | 4254 | |
fa8d436c UD |
4255 | /* |
4256 | Search for a chunk by scanning bins, starting with next largest | |
4257 | bin. This search is strictly by best-fit; i.e., the smallest | |
4258 | (with ties going to approximately the least recently used) chunk | |
4259 | that fits is selected. | |
a9177ff5 | 4260 | |
fa8d436c UD |
4261 | The bitmap avoids needing to check that most blocks are nonempty. |
4262 | The particular case of skipping all bins during warm-up phases | |
4263 | when no chunks have been returned yet is faster than it might look. | |
4264 | */ | |
a9177ff5 | 4265 | |
fa8d436c UD |
4266 | ++idx; |
4267 | bin = bin_at(av,idx); | |
4268 | block = idx2block(idx); | |
4269 | map = av->binmap[block]; | |
4270 | bit = idx2bit(idx); | |
a9177ff5 | 4271 | |
fa8d436c UD |
4272 | for (;;) { |
4273 | ||
4274 | /* Skip rest of block if there are no more set bits in this block. */ | |
4275 | if (bit > map || bit == 0) { | |
4276 | do { | |
4277 | if (++block >= BINMAPSIZE) /* out of bins */ | |
4278 | goto use_top; | |
4279 | } while ( (map = av->binmap[block]) == 0); | |
4280 | ||
4281 | bin = bin_at(av, (block << BINMAPSHIFT)); | |
4282 | bit = 1; | |
4283 | } | |
a9177ff5 | 4284 | |
fa8d436c UD |
4285 | /* Advance to bin with set bit. There must be one. */ |
4286 | while ((bit & map) == 0) { | |
4287 | bin = next_bin(bin); | |
4288 | bit <<= 1; | |
4289 | assert(bit != 0); | |
4290 | } | |
a9177ff5 | 4291 | |
fa8d436c UD |
4292 | /* Inspect the bin. It is likely to be non-empty */ |
4293 | victim = last(bin); | |
a9177ff5 | 4294 | |
fa8d436c UD |
4295 | /* If a false alarm (empty bin), clear the bit. */ |
4296 | if (victim == bin) { | |
4297 | av->binmap[block] = map &= ~bit; /* Write through */ | |
4298 | bin = next_bin(bin); | |
4299 | bit <<= 1; | |
4300 | } | |
a9177ff5 | 4301 | |
fa8d436c UD |
4302 | else { |
4303 | size = chunksize(victim); | |
4304 | ||
4305 | /* We know the first chunk in this bin is big enough to use. */ | |
4306 | assert((unsigned long)(size) >= (unsigned long)(nb)); | |
4307 | ||
4308 | remainder_size = size - nb; | |
a9177ff5 | 4309 | |
fa8d436c UD |
4310 | /* unlink */ |
4311 | bck = victim->bk; | |
4312 | bin->bk = bck; | |
4313 | bck->fd = bin; | |
a9177ff5 | 4314 | |
fa8d436c UD |
4315 | /* Exhaust */ |
4316 | if (remainder_size < MINSIZE) { | |
4317 | set_inuse_bit_at_offset(victim, size); | |
4318 | if (av != &main_arena) | |
4319 | victim->size |= NON_MAIN_ARENA; | |
fa8d436c | 4320 | } |
a9177ff5 | 4321 | |
fa8d436c UD |
4322 | /* Split */ |
4323 | else { | |
4324 | remainder = chunk_at_offset(victim, nb); | |
a9177ff5 | 4325 | |
41999a1a UD |
4326 | /* We cannot assume the unsorted list is empty and therefore |
4327 | have to perform a complete insert here. */ | |
4328 | bck = unsorted_chunks(av); | |
4329 | fwd = bck->fd; | |
4330 | remainder->bk = bck; | |
4331 | remainder->fd = fwd; | |
4332 | bck->fd = remainder; | |
4333 | fwd->bk = remainder; | |
4334 | ||
fa8d436c | 4335 | /* advertise as last remainder */ |
a9177ff5 RM |
4336 | if (in_smallbin_range(nb)) |
4337 | av->last_remainder = remainder; | |
4338 | ||
fa8d436c UD |
4339 | set_head(victim, nb | PREV_INUSE | |
4340 | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
4341 | set_head(remainder, remainder_size | PREV_INUSE); | |
4342 | set_foot(remainder, remainder_size); | |
fa8d436c | 4343 | } |
854278df UD |
4344 | check_malloced_chunk(av, victim, nb); |
4345 | void *p = chunk2mem(victim); | |
4346 | if (__builtin_expect (perturb_byte, 0)) | |
4347 | alloc_perturb (p, bytes); | |
4348 | return p; | |
fa8d436c UD |
4349 | } |
4350 | } | |
f65fd747 | 4351 | |
a9177ff5 | 4352 | use_top: |
fa8d436c UD |
4353 | /* |
4354 | If large enough, split off the chunk bordering the end of memory | |
4355 | (held in av->top). Note that this is in accord with the best-fit | |
4356 | search rule. In effect, av->top is treated as larger (and thus | |
4357 | less well fitting) than any other available chunk since it can | |
4358 | be extended to be as large as necessary (up to system | |
4359 | limitations). | |
4360 | ||
4361 | We require that av->top always exists (i.e., has size >= | |
4362 | MINSIZE) after initialization, so if it would otherwise be | |
4363 | exhuasted by current request, it is replenished. (The main | |
4364 | reason for ensuring it exists is that we may need MINSIZE space | |
4365 | to put in fenceposts in sysmalloc.) | |
4366 | */ | |
f65fd747 | 4367 | |
fa8d436c UD |
4368 | victim = av->top; |
4369 | size = chunksize(victim); | |
a9177ff5 | 4370 | |
fa8d436c UD |
4371 | if ((unsigned long)(size) >= (unsigned long)(nb + MINSIZE)) { |
4372 | remainder_size = size - nb; | |
4373 | remainder = chunk_at_offset(victim, nb); | |
4374 | av->top = remainder; | |
4375 | set_head(victim, nb | PREV_INUSE | | |
4376 | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
4377 | set_head(remainder, remainder_size | PREV_INUSE); | |
f65fd747 | 4378 | |
fa8d436c | 4379 | check_malloced_chunk(av, victim, nb); |
854278df UD |
4380 | void *p = chunk2mem(victim); |
4381 | if (__builtin_expect (perturb_byte, 0)) | |
4382 | alloc_perturb (p, bytes); | |
4383 | return p; | |
fa8d436c | 4384 | } |
f65fd747 | 4385 | |
fa8d436c UD |
4386 | /* |
4387 | If there is space available in fastbins, consolidate and retry, | |
4388 | to possibly avoid expanding memory. This can occur only if nb is | |
4389 | in smallbin range so we didn't consolidate upon entry. | |
4390 | */ | |
f65fd747 | 4391 | |
fa8d436c UD |
4392 | else if (have_fastchunks(av)) { |
4393 | assert(in_smallbin_range(nb)); | |
4394 | malloc_consolidate(av); | |
4395 | idx = smallbin_index(nb); /* restore original bin index */ | |
4396 | } | |
f65fd747 | 4397 | |
a9177ff5 RM |
4398 | /* |
4399 | Otherwise, relay to handle system-dependent cases | |
fa8d436c | 4400 | */ |
854278df UD |
4401 | else { |
4402 | void *p = sYSMALLOc(nb, av); | |
4403 | if (__builtin_expect (perturb_byte, 0)) | |
4404 | alloc_perturb (p, bytes); | |
4405 | return p; | |
4406 | } | |
fa8d436c UD |
4407 | } |
4408 | } | |
f65fd747 | 4409 | |
fa8d436c UD |
4410 | /* |
4411 | ------------------------------ free ------------------------------ | |
f65fd747 UD |
4412 | */ |
4413 | ||
f1c5213d | 4414 | void |
fa8d436c | 4415 | _int_free(mstate av, Void_t* mem) |
f65fd747 | 4416 | { |
fa8d436c UD |
4417 | mchunkptr p; /* chunk corresponding to mem */ |
4418 | INTERNAL_SIZE_T size; /* its size */ | |
4419 | mfastbinptr* fb; /* associated fastbin */ | |
4420 | mchunkptr nextchunk; /* next contiguous chunk */ | |
4421 | INTERNAL_SIZE_T nextsize; /* its size */ | |
4422 | int nextinuse; /* true if nextchunk is used */ | |
4423 | INTERNAL_SIZE_T prevsize; /* size of previous contiguous chunk */ | |
4424 | mchunkptr bck; /* misc temp for linking */ | |
4425 | mchunkptr fwd; /* misc temp for linking */ | |
4426 | ||
37fa1953 | 4427 | const char *errstr = NULL; |
f65fd747 | 4428 | |
37fa1953 UD |
4429 | p = mem2chunk(mem); |
4430 | size = chunksize(p); | |
f65fd747 | 4431 | |
37fa1953 UD |
4432 | /* Little security check which won't hurt performance: the |
4433 | allocator never wrapps around at the end of the address space. | |
4434 | Therefore we can exclude some size values which might appear | |
4435 | here by accident or by "design" from some intruder. */ | |
dc165f7b | 4436 | if (__builtin_expect ((uintptr_t) p > (uintptr_t) -size, 0) |
073f560e | 4437 | || __builtin_expect (misaligned_chunk (p), 0)) |
37fa1953 UD |
4438 | { |
4439 | errstr = "free(): invalid pointer"; | |
4440 | errout: | |
4441 | malloc_printerr (check_action, errstr, mem); | |
4442 | return; | |
fa8d436c | 4443 | } |
bf589066 UD |
4444 | /* We know that each chunk is at least MINSIZE bytes in size. */ |
4445 | if (__builtin_expect (size < MINSIZE, 0)) | |
4446 | { | |
4447 | errstr = "free(): invalid size"; | |
4448 | goto errout; | |
4449 | } | |
f65fd747 | 4450 | |
37fa1953 | 4451 | check_inuse_chunk(av, p); |
f65fd747 | 4452 | |
37fa1953 UD |
4453 | /* |
4454 | If eligible, place chunk on a fastbin so it can be found | |
4455 | and used quickly in malloc. | |
4456 | */ | |
6bf4302e | 4457 | |
9bf248c6 | 4458 | if ((unsigned long)(size) <= (unsigned long)(get_max_fast ()) |
6bf4302e | 4459 | |
37fa1953 UD |
4460 | #if TRIM_FASTBINS |
4461 | /* | |
4462 | If TRIM_FASTBINS set, don't place chunks | |
4463 | bordering top into fastbins | |
4464 | */ | |
4465 | && (chunk_at_offset(p, size) != av->top) | |
4466 | #endif | |
4467 | ) { | |
fa8d436c | 4468 | |
893e6098 UD |
4469 | if (__builtin_expect (chunk_at_offset (p, size)->size <= 2 * SIZE_SZ, 0) |
4470 | || __builtin_expect (chunksize (chunk_at_offset (p, size)) | |
4471 | >= av->system_mem, 0)) | |
4472 | { | |
76761b63 | 4473 | errstr = "free(): invalid next size (fast)"; |
893e6098 UD |
4474 | goto errout; |
4475 | } | |
4476 | ||
37fa1953 UD |
4477 | set_fastchunks(av); |
4478 | fb = &(av->fastbins[fastbin_index(size)]); | |
4479 | /* Another simple check: make sure the top of the bin is not the | |
4480 | record we are going to add (i.e., double free). */ | |
4481 | if (__builtin_expect (*fb == p, 0)) | |
4482 | { | |
4483 | errstr = "double free or corruption (fasttop)"; | |
4484 | goto errout; | |
fa8d436c | 4485 | } |
854278df UD |
4486 | |
4487 | if (__builtin_expect (perturb_byte, 0)) | |
4488 | free_perturb (mem, size - SIZE_SZ); | |
4489 | ||
37fa1953 UD |
4490 | p->fd = *fb; |
4491 | *fb = p; | |
4492 | } | |
f65fd747 | 4493 | |
37fa1953 UD |
4494 | /* |
4495 | Consolidate other non-mmapped chunks as they arrive. | |
4496 | */ | |
fa8d436c | 4497 | |
37fa1953 UD |
4498 | else if (!chunk_is_mmapped(p)) { |
4499 | nextchunk = chunk_at_offset(p, size); | |
fa8d436c | 4500 | |
37fa1953 UD |
4501 | /* Lightweight tests: check whether the block is already the |
4502 | top block. */ | |
4503 | if (__builtin_expect (p == av->top, 0)) | |
4504 | { | |
4505 | errstr = "double free or corruption (top)"; | |
4506 | goto errout; | |
4507 | } | |
4508 | /* Or whether the next chunk is beyond the boundaries of the arena. */ | |
4509 | if (__builtin_expect (contiguous (av) | |
4510 | && (char *) nextchunk | |
4511 | >= ((char *) av->top + chunksize(av->top)), 0)) | |
4512 | { | |
4513 | errstr = "double free or corruption (out)"; | |
4514 | goto errout; | |
4515 | } | |
4516 | /* Or whether the block is actually not marked used. */ | |
4517 | if (__builtin_expect (!prev_inuse(nextchunk), 0)) | |
4518 | { | |
4519 | errstr = "double free or corruption (!prev)"; | |
4520 | goto errout; | |
4521 | } | |
fa8d436c | 4522 | |
37fa1953 | 4523 | nextsize = chunksize(nextchunk); |
893e6098 UD |
4524 | if (__builtin_expect (nextchunk->size <= 2 * SIZE_SZ, 0) |
4525 | || __builtin_expect (nextsize >= av->system_mem, 0)) | |
4526 | { | |
76761b63 | 4527 | errstr = "free(): invalid next size (normal)"; |
893e6098 UD |
4528 | goto errout; |
4529 | } | |
fa8d436c | 4530 | |
854278df UD |
4531 | if (__builtin_expect (perturb_byte, 0)) |
4532 | free_perturb (mem, size - SIZE_SZ); | |
4533 | ||
37fa1953 UD |
4534 | /* consolidate backward */ |
4535 | if (!prev_inuse(p)) { | |
4536 | prevsize = p->prev_size; | |
4537 | size += prevsize; | |
4538 | p = chunk_at_offset(p, -((long) prevsize)); | |
4539 | unlink(p, bck, fwd); | |
4540 | } | |
a9177ff5 | 4541 | |
37fa1953 UD |
4542 | if (nextchunk != av->top) { |
4543 | /* get and clear inuse bit */ | |
4544 | nextinuse = inuse_bit_at_offset(nextchunk, nextsize); | |
4545 | ||
4546 | /* consolidate forward */ | |
4547 | if (!nextinuse) { | |
4548 | unlink(nextchunk, bck, fwd); | |
4549 | size += nextsize; | |
4550 | } else | |
4551 | clear_inuse_bit_at_offset(nextchunk, 0); | |
10dc2a90 | 4552 | |
fa8d436c | 4553 | /* |
37fa1953 UD |
4554 | Place the chunk in unsorted chunk list. Chunks are |
4555 | not placed into regular bins until after they have | |
4556 | been given one chance to be used in malloc. | |
fa8d436c | 4557 | */ |
f65fd747 | 4558 | |
37fa1953 UD |
4559 | bck = unsorted_chunks(av); |
4560 | fwd = bck->fd; | |
4561 | p->bk = bck; | |
4562 | p->fd = fwd; | |
4563 | bck->fd = p; | |
4564 | fwd->bk = p; | |
8a4b65b4 | 4565 | |
37fa1953 UD |
4566 | set_head(p, size | PREV_INUSE); |
4567 | set_foot(p, size); | |
4568 | ||
4569 | check_free_chunk(av, p); | |
4570 | } | |
4571 | ||
4572 | /* | |
4573 | If the chunk borders the current high end of memory, | |
4574 | consolidate into top | |
4575 | */ | |
4576 | ||
4577 | else { | |
4578 | size += nextsize; | |
4579 | set_head(p, size | PREV_INUSE); | |
4580 | av->top = p; | |
4581 | check_chunk(av, p); | |
4582 | } | |
4583 | ||
4584 | /* | |
4585 | If freeing a large space, consolidate possibly-surrounding | |
4586 | chunks. Then, if the total unused topmost memory exceeds trim | |
4587 | threshold, ask malloc_trim to reduce top. | |
4588 | ||
4589 | Unless max_fast is 0, we don't know if there are fastbins | |
4590 | bordering top, so we cannot tell for sure whether threshold | |
4591 | has been reached unless fastbins are consolidated. But we | |
4592 | don't want to consolidate on each free. As a compromise, | |
4593 | consolidation is performed if FASTBIN_CONSOLIDATION_THRESHOLD | |
4594 | is reached. | |
4595 | */ | |
fa8d436c | 4596 | |
37fa1953 UD |
4597 | if ((unsigned long)(size) >= FASTBIN_CONSOLIDATION_THRESHOLD) { |
4598 | if (have_fastchunks(av)) | |
4599 | malloc_consolidate(av); | |
fa8d436c | 4600 | |
37fa1953 | 4601 | if (av == &main_arena) { |
a9177ff5 | 4602 | #ifndef MORECORE_CANNOT_TRIM |
37fa1953 UD |
4603 | if ((unsigned long)(chunksize(av->top)) >= |
4604 | (unsigned long)(mp_.trim_threshold)) | |
4605 | sYSTRIm(mp_.top_pad, av); | |
fa8d436c | 4606 | #endif |
37fa1953 UD |
4607 | } else { |
4608 | /* Always try heap_trim(), even if the top chunk is not | |
4609 | large, because the corresponding heap might go away. */ | |
4610 | heap_info *heap = heap_for_ptr(top(av)); | |
fa8d436c | 4611 | |
37fa1953 UD |
4612 | assert(heap->ar_ptr == av); |
4613 | heap_trim(heap, mp_.top_pad); | |
fa8d436c | 4614 | } |
fa8d436c | 4615 | } |
10dc2a90 | 4616 | |
37fa1953 UD |
4617 | } |
4618 | /* | |
4619 | If the chunk was allocated via mmap, release via munmap(). Note | |
4620 | that if HAVE_MMAP is false but chunk_is_mmapped is true, then | |
4621 | user must have overwritten memory. There's nothing we can do to | |
4622 | catch this error unless MALLOC_DEBUG is set, in which case | |
4623 | check_inuse_chunk (above) will have triggered error. | |
4624 | */ | |
4625 | ||
4626 | else { | |
fa8d436c | 4627 | #if HAVE_MMAP |
c120d94d | 4628 | munmap_chunk (p); |
fa8d436c | 4629 | #endif |
fa8d436c | 4630 | } |
10dc2a90 UD |
4631 | } |
4632 | ||
fa8d436c UD |
4633 | /* |
4634 | ------------------------- malloc_consolidate ------------------------- | |
4635 | ||
4636 | malloc_consolidate is a specialized version of free() that tears | |
4637 | down chunks held in fastbins. Free itself cannot be used for this | |
4638 | purpose since, among other things, it might place chunks back onto | |
4639 | fastbins. So, instead, we need to use a minor variant of the same | |
4640 | code. | |
a9177ff5 | 4641 | |
fa8d436c UD |
4642 | Also, because this routine needs to be called the first time through |
4643 | malloc anyway, it turns out to be the perfect place to trigger | |
4644 | initialization code. | |
4645 | */ | |
4646 | ||
10dc2a90 | 4647 | #if __STD_C |
fa8d436c | 4648 | static void malloc_consolidate(mstate av) |
10dc2a90 | 4649 | #else |
fa8d436c | 4650 | static void malloc_consolidate(av) mstate av; |
10dc2a90 UD |
4651 | #endif |
4652 | { | |
fa8d436c UD |
4653 | mfastbinptr* fb; /* current fastbin being consolidated */ |
4654 | mfastbinptr* maxfb; /* last fastbin (for loop control) */ | |
4655 | mchunkptr p; /* current chunk being consolidated */ | |
4656 | mchunkptr nextp; /* next chunk to consolidate */ | |
4657 | mchunkptr unsorted_bin; /* bin header */ | |
4658 | mchunkptr first_unsorted; /* chunk to link to */ | |
4659 | ||
4660 | /* These have same use as in free() */ | |
4661 | mchunkptr nextchunk; | |
4662 | INTERNAL_SIZE_T size; | |
4663 | INTERNAL_SIZE_T nextsize; | |
4664 | INTERNAL_SIZE_T prevsize; | |
4665 | int nextinuse; | |
4666 | mchunkptr bck; | |
4667 | mchunkptr fwd; | |
10dc2a90 | 4668 | |
fa8d436c UD |
4669 | /* |
4670 | If max_fast is 0, we know that av hasn't | |
4671 | yet been initialized, in which case do so below | |
4672 | */ | |
10dc2a90 | 4673 | |
9bf248c6 | 4674 | if (get_max_fast () != 0) { |
fa8d436c | 4675 | clear_fastchunks(av); |
10dc2a90 | 4676 | |
fa8d436c | 4677 | unsorted_bin = unsorted_chunks(av); |
10dc2a90 | 4678 | |
fa8d436c UD |
4679 | /* |
4680 | Remove each chunk from fast bin and consolidate it, placing it | |
4681 | then in unsorted bin. Among other reasons for doing this, | |
4682 | placing in unsorted bin avoids needing to calculate actual bins | |
4683 | until malloc is sure that chunks aren't immediately going to be | |
4684 | reused anyway. | |
4685 | */ | |
a9177ff5 | 4686 | |
9bf248c6 | 4687 | maxfb = &(av->fastbins[fastbin_index(get_max_fast ())]); |
fa8d436c UD |
4688 | fb = &(av->fastbins[0]); |
4689 | do { | |
4690 | if ( (p = *fb) != 0) { | |
4691 | *fb = 0; | |
a9177ff5 | 4692 | |
fa8d436c UD |
4693 | do { |
4694 | check_inuse_chunk(av, p); | |
4695 | nextp = p->fd; | |
a9177ff5 | 4696 | |
fa8d436c UD |
4697 | /* Slightly streamlined version of consolidation code in free() */ |
4698 | size = p->size & ~(PREV_INUSE|NON_MAIN_ARENA); | |
4699 | nextchunk = chunk_at_offset(p, size); | |
4700 | nextsize = chunksize(nextchunk); | |
a9177ff5 | 4701 | |
fa8d436c UD |
4702 | if (!prev_inuse(p)) { |
4703 | prevsize = p->prev_size; | |
4704 | size += prevsize; | |
4705 | p = chunk_at_offset(p, -((long) prevsize)); | |
4706 | unlink(p, bck, fwd); | |
4707 | } | |
a9177ff5 | 4708 | |
fa8d436c UD |
4709 | if (nextchunk != av->top) { |
4710 | nextinuse = inuse_bit_at_offset(nextchunk, nextsize); | |
a9177ff5 | 4711 | |
fa8d436c UD |
4712 | if (!nextinuse) { |
4713 | size += nextsize; | |
4714 | unlink(nextchunk, bck, fwd); | |
4715 | } else | |
4716 | clear_inuse_bit_at_offset(nextchunk, 0); | |
a9177ff5 | 4717 | |
fa8d436c UD |
4718 | first_unsorted = unsorted_bin->fd; |
4719 | unsorted_bin->fd = p; | |
4720 | first_unsorted->bk = p; | |
a9177ff5 | 4721 | |
fa8d436c UD |
4722 | set_head(p, size | PREV_INUSE); |
4723 | p->bk = unsorted_bin; | |
4724 | p->fd = first_unsorted; | |
4725 | set_foot(p, size); | |
4726 | } | |
a9177ff5 | 4727 | |
fa8d436c UD |
4728 | else { |
4729 | size += nextsize; | |
4730 | set_head(p, size | PREV_INUSE); | |
4731 | av->top = p; | |
4732 | } | |
a9177ff5 | 4733 | |
fa8d436c | 4734 | } while ( (p = nextp) != 0); |
a9177ff5 | 4735 | |
fa8d436c UD |
4736 | } |
4737 | } while (fb++ != maxfb); | |
4738 | } | |
4739 | else { | |
4740 | malloc_init_state(av); | |
4741 | check_malloc_state(av); | |
4742 | } | |
4743 | } | |
10dc2a90 | 4744 | |
fa8d436c UD |
4745 | /* |
4746 | ------------------------------ realloc ------------------------------ | |
4747 | */ | |
f65fd747 | 4748 | |
f1c5213d | 4749 | Void_t* |
fa8d436c UD |
4750 | _int_realloc(mstate av, Void_t* oldmem, size_t bytes) |
4751 | { | |
4752 | INTERNAL_SIZE_T nb; /* padded request size */ | |
f65fd747 | 4753 | |
fa8d436c UD |
4754 | mchunkptr oldp; /* chunk corresponding to oldmem */ |
4755 | INTERNAL_SIZE_T oldsize; /* its size */ | |
f65fd747 | 4756 | |
fa8d436c UD |
4757 | mchunkptr newp; /* chunk to return */ |
4758 | INTERNAL_SIZE_T newsize; /* its size */ | |
4759 | Void_t* newmem; /* corresponding user mem */ | |
f65fd747 | 4760 | |
fa8d436c | 4761 | mchunkptr next; /* next contiguous chunk after oldp */ |
f65fd747 | 4762 | |
fa8d436c UD |
4763 | mchunkptr remainder; /* extra space at end of newp */ |
4764 | unsigned long remainder_size; /* its size */ | |
f65fd747 | 4765 | |
fa8d436c UD |
4766 | mchunkptr bck; /* misc temp for linking */ |
4767 | mchunkptr fwd; /* misc temp for linking */ | |
2ed5fd9a | 4768 | |
fa8d436c UD |
4769 | unsigned long copysize; /* bytes to copy */ |
4770 | unsigned int ncopies; /* INTERNAL_SIZE_T words to copy */ | |
a9177ff5 | 4771 | INTERNAL_SIZE_T* s; /* copy source */ |
fa8d436c | 4772 | INTERNAL_SIZE_T* d; /* copy destination */ |
f65fd747 | 4773 | |
76761b63 | 4774 | const char *errstr = NULL; |
f65fd747 | 4775 | |
f65fd747 | 4776 | |
fa8d436c | 4777 | checked_request2size(bytes, nb); |
f65fd747 | 4778 | |
fa8d436c UD |
4779 | oldp = mem2chunk(oldmem); |
4780 | oldsize = chunksize(oldp); | |
f65fd747 | 4781 | |
76761b63 | 4782 | /* Simple tests for old block integrity. */ |
073f560e | 4783 | if (__builtin_expect (misaligned_chunk (oldp), 0)) |
76761b63 UD |
4784 | { |
4785 | errstr = "realloc(): invalid pointer"; | |
4786 | errout: | |
4787 | malloc_printerr (check_action, errstr, oldmem); | |
4788 | return NULL; | |
4789 | } | |
4790 | if (__builtin_expect (oldp->size <= 2 * SIZE_SZ, 0) | |
4791 | || __builtin_expect (oldsize >= av->system_mem, 0)) | |
4792 | { | |
4b04154d | 4793 | errstr = "realloc(): invalid old size"; |
76761b63 UD |
4794 | goto errout; |
4795 | } | |
4796 | ||
fa8d436c | 4797 | check_inuse_chunk(av, oldp); |
f65fd747 | 4798 | |
fa8d436c | 4799 | if (!chunk_is_mmapped(oldp)) { |
f65fd747 | 4800 | |
76761b63 UD |
4801 | next = chunk_at_offset(oldp, oldsize); |
4802 | INTERNAL_SIZE_T nextsize = chunksize(next); | |
4803 | if (__builtin_expect (next->size <= 2 * SIZE_SZ, 0) | |
4804 | || __builtin_expect (nextsize >= av->system_mem, 0)) | |
4805 | { | |
4806 | errstr = "realloc(): invalid next size"; | |
4807 | goto errout; | |
4808 | } | |
4809 | ||
fa8d436c UD |
4810 | if ((unsigned long)(oldsize) >= (unsigned long)(nb)) { |
4811 | /* already big enough; split below */ | |
4812 | newp = oldp; | |
4813 | newsize = oldsize; | |
7799b7b3 | 4814 | } |
f65fd747 | 4815 | |
fa8d436c | 4816 | else { |
fa8d436c UD |
4817 | /* Try to expand forward into top */ |
4818 | if (next == av->top && | |
76761b63 | 4819 | (unsigned long)(newsize = oldsize + nextsize) >= |
fa8d436c UD |
4820 | (unsigned long)(nb + MINSIZE)) { |
4821 | set_head_size(oldp, nb | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
4822 | av->top = chunk_at_offset(oldp, nb); | |
4823 | set_head(av->top, (newsize - nb) | PREV_INUSE); | |
4824 | check_inuse_chunk(av, oldp); | |
4825 | return chunk2mem(oldp); | |
4826 | } | |
a9177ff5 | 4827 | |
fa8d436c | 4828 | /* Try to expand forward into next chunk; split off remainder below */ |
a9177ff5 | 4829 | else if (next != av->top && |
fa8d436c | 4830 | !inuse(next) && |
76761b63 | 4831 | (unsigned long)(newsize = oldsize + nextsize) >= |
fa8d436c UD |
4832 | (unsigned long)(nb)) { |
4833 | newp = oldp; | |
4834 | unlink(next, bck, fwd); | |
4835 | } | |
f65fd747 | 4836 | |
fa8d436c UD |
4837 | /* allocate, copy, free */ |
4838 | else { | |
4839 | newmem = _int_malloc(av, nb - MALLOC_ALIGN_MASK); | |
4840 | if (newmem == 0) | |
4841 | return 0; /* propagate failure */ | |
a9177ff5 | 4842 | |
fa8d436c UD |
4843 | newp = mem2chunk(newmem); |
4844 | newsize = chunksize(newp); | |
a9177ff5 | 4845 | |
fa8d436c UD |
4846 | /* |
4847 | Avoid copy if newp is next chunk after oldp. | |
4848 | */ | |
4849 | if (newp == next) { | |
4850 | newsize += oldsize; | |
4851 | newp = oldp; | |
4852 | } | |
4853 | else { | |
4854 | /* | |
4855 | Unroll copy of <= 36 bytes (72 if 8byte sizes) | |
4856 | We know that contents have an odd number of | |
4857 | INTERNAL_SIZE_T-sized words; minimally 3. | |
4858 | */ | |
a9177ff5 | 4859 | |
fa8d436c UD |
4860 | copysize = oldsize - SIZE_SZ; |
4861 | s = (INTERNAL_SIZE_T*)(oldmem); | |
4862 | d = (INTERNAL_SIZE_T*)(newmem); | |
4863 | ncopies = copysize / sizeof(INTERNAL_SIZE_T); | |
4864 | assert(ncopies >= 3); | |
a9177ff5 | 4865 | |
fa8d436c UD |
4866 | if (ncopies > 9) |
4867 | MALLOC_COPY(d, s, copysize); | |
a9177ff5 | 4868 | |
fa8d436c UD |
4869 | else { |
4870 | *(d+0) = *(s+0); | |
4871 | *(d+1) = *(s+1); | |
4872 | *(d+2) = *(s+2); | |
4873 | if (ncopies > 4) { | |
4874 | *(d+3) = *(s+3); | |
4875 | *(d+4) = *(s+4); | |
4876 | if (ncopies > 6) { | |
4877 | *(d+5) = *(s+5); | |
4878 | *(d+6) = *(s+6); | |
4879 | if (ncopies > 8) { | |
4880 | *(d+7) = *(s+7); | |
4881 | *(d+8) = *(s+8); | |
4882 | } | |
4883 | } | |
4884 | } | |
4885 | } | |
a9177ff5 | 4886 | |
fa8d436c UD |
4887 | _int_free(av, oldmem); |
4888 | check_inuse_chunk(av, newp); | |
4889 | return chunk2mem(newp); | |
4890 | } | |
4891 | } | |
f65fd747 UD |
4892 | } |
4893 | ||
fa8d436c | 4894 | /* If possible, free extra space in old or extended chunk */ |
f65fd747 | 4895 | |
fa8d436c | 4896 | assert((unsigned long)(newsize) >= (unsigned long)(nb)); |
f65fd747 | 4897 | |
f65fd747 | 4898 | remainder_size = newsize - nb; |
f65fd747 | 4899 | |
fa8d436c UD |
4900 | if (remainder_size < MINSIZE) { /* not enough extra to split off */ |
4901 | set_head_size(newp, newsize | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
4902 | set_inuse_bit_at_offset(newp, newsize); | |
4903 | } | |
4904 | else { /* split remainder */ | |
4905 | remainder = chunk_at_offset(newp, nb); | |
4906 | set_head_size(newp, nb | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
4907 | set_head(remainder, remainder_size | PREV_INUSE | | |
4908 | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
4909 | /* Mark remainder as inuse so free() won't complain */ | |
4910 | set_inuse_bit_at_offset(remainder, remainder_size); | |
a9177ff5 | 4911 | _int_free(av, chunk2mem(remainder)); |
fa8d436c | 4912 | } |
f65fd747 | 4913 | |
fa8d436c UD |
4914 | check_inuse_chunk(av, newp); |
4915 | return chunk2mem(newp); | |
4916 | } | |
f65fd747 | 4917 | |
fa8d436c UD |
4918 | /* |
4919 | Handle mmap cases | |
4920 | */ | |
f65fd747 | 4921 | |
fa8d436c UD |
4922 | else { |
4923 | #if HAVE_MMAP | |
f65fd747 | 4924 | |
fa8d436c UD |
4925 | #if HAVE_MREMAP |
4926 | INTERNAL_SIZE_T offset = oldp->prev_size; | |
4927 | size_t pagemask = mp_.pagesize - 1; | |
4928 | char *cp; | |
4929 | unsigned long sum; | |
a9177ff5 | 4930 | |
fa8d436c UD |
4931 | /* Note the extra SIZE_SZ overhead */ |
4932 | newsize = (nb + offset + SIZE_SZ + pagemask) & ~pagemask; | |
4933 | ||
4934 | /* don't need to remap if still within same page */ | |
a9177ff5 | 4935 | if (oldsize == newsize - offset) |
fa8d436c UD |
4936 | return oldmem; |
4937 | ||
4938 | cp = (char*)mremap((char*)oldp - offset, oldsize + offset, newsize, 1); | |
a9177ff5 | 4939 | |
fa8d436c UD |
4940 | if (cp != MAP_FAILED) { |
4941 | ||
4942 | newp = (mchunkptr)(cp + offset); | |
4943 | set_head(newp, (newsize - offset)|IS_MMAPPED); | |
a9177ff5 | 4944 | |
fa8d436c UD |
4945 | assert(aligned_OK(chunk2mem(newp))); |
4946 | assert((newp->prev_size == offset)); | |
a9177ff5 | 4947 | |
fa8d436c UD |
4948 | /* update statistics */ |
4949 | sum = mp_.mmapped_mem += newsize - oldsize; | |
a9177ff5 | 4950 | if (sum > (unsigned long)(mp_.max_mmapped_mem)) |
fa8d436c UD |
4951 | mp_.max_mmapped_mem = sum; |
4952 | #ifdef NO_THREADS | |
4953 | sum += main_arena.system_mem; | |
a9177ff5 | 4954 | if (sum > (unsigned long)(mp_.max_total_mem)) |
fa8d436c UD |
4955 | mp_.max_total_mem = sum; |
4956 | #endif | |
a9177ff5 | 4957 | |
fa8d436c UD |
4958 | return chunk2mem(newp); |
4959 | } | |
f65fd747 | 4960 | #endif |
10dc2a90 | 4961 | |
fa8d436c | 4962 | /* Note the extra SIZE_SZ overhead. */ |
a9177ff5 | 4963 | if ((unsigned long)(oldsize) >= (unsigned long)(nb + SIZE_SZ)) |
fa8d436c UD |
4964 | newmem = oldmem; /* do nothing */ |
4965 | else { | |
4966 | /* Must alloc, copy, free. */ | |
4967 | newmem = _int_malloc(av, nb - MALLOC_ALIGN_MASK); | |
4968 | if (newmem != 0) { | |
4969 | MALLOC_COPY(newmem, oldmem, oldsize - 2*SIZE_SZ); | |
4970 | _int_free(av, oldmem); | |
4971 | } | |
4972 | } | |
4973 | return newmem; | |
10dc2a90 | 4974 | |
a9177ff5 | 4975 | #else |
fa8d436c UD |
4976 | /* If !HAVE_MMAP, but chunk_is_mmapped, user must have overwritten mem */ |
4977 | check_malloc_state(av); | |
4978 | MALLOC_FAILURE_ACTION; | |
4979 | return 0; | |
a2b08ee5 | 4980 | #endif |
10dc2a90 | 4981 | } |
fa8d436c UD |
4982 | } |
4983 | ||
4984 | /* | |
4985 | ------------------------------ memalign ------------------------------ | |
4986 | */ | |
4987 | ||
f1c5213d | 4988 | Void_t* |
fa8d436c UD |
4989 | _int_memalign(mstate av, size_t alignment, size_t bytes) |
4990 | { | |
4991 | INTERNAL_SIZE_T nb; /* padded request size */ | |
4992 | char* m; /* memory returned by malloc call */ | |
4993 | mchunkptr p; /* corresponding chunk */ | |
4994 | char* brk; /* alignment point within p */ | |
4995 | mchunkptr newp; /* chunk to return */ | |
4996 | INTERNAL_SIZE_T newsize; /* its size */ | |
4997 | INTERNAL_SIZE_T leadsize; /* leading space before alignment point */ | |
4998 | mchunkptr remainder; /* spare room at end to split off */ | |
4999 | unsigned long remainder_size; /* its size */ | |
5000 | INTERNAL_SIZE_T size; | |
f65fd747 UD |
5001 | |
5002 | /* If need less alignment than we give anyway, just relay to malloc */ | |
5003 | ||
fa8d436c | 5004 | if (alignment <= MALLOC_ALIGNMENT) return _int_malloc(av, bytes); |
f65fd747 UD |
5005 | |
5006 | /* Otherwise, ensure that it is at least a minimum chunk size */ | |
5007 | ||
5008 | if (alignment < MINSIZE) alignment = MINSIZE; | |
5009 | ||
fa8d436c UD |
5010 | /* Make sure alignment is power of 2 (in case MINSIZE is not). */ |
5011 | if ((alignment & (alignment - 1)) != 0) { | |
5012 | size_t a = MALLOC_ALIGNMENT * 2; | |
5013 | while ((unsigned long)a < (unsigned long)alignment) a <<= 1; | |
5014 | alignment = a; | |
7799b7b3 | 5015 | } |
f65fd747 | 5016 | |
fa8d436c UD |
5017 | checked_request2size(bytes, nb); |
5018 | ||
5019 | /* | |
5020 | Strategy: find a spot within that chunk that meets the alignment | |
5021 | request, and then possibly free the leading and trailing space. | |
5022 | */ | |
5023 | ||
5024 | ||
5025 | /* Call malloc with worst case padding to hit alignment. */ | |
5026 | ||
5027 | m = (char*)(_int_malloc(av, nb + alignment + MINSIZE)); | |
5028 | ||
5029 | if (m == 0) return 0; /* propagate failure */ | |
5030 | ||
5031 | p = mem2chunk(m); | |
5032 | ||
5033 | if ((((unsigned long)(m)) % alignment) != 0) { /* misaligned */ | |
5034 | ||
f65fd747 | 5035 | /* |
fa8d436c UD |
5036 | Find an aligned spot inside chunk. Since we need to give back |
5037 | leading space in a chunk of at least MINSIZE, if the first | |
5038 | calculation places us at a spot with less than MINSIZE leader, | |
5039 | we can move to the next aligned spot -- we've allocated enough | |
5040 | total room so that this is always possible. | |
f65fd747 UD |
5041 | */ |
5042 | ||
fa8d436c UD |
5043 | brk = (char*)mem2chunk(((unsigned long)(m + alignment - 1)) & |
5044 | -((signed long) alignment)); | |
5045 | if ((unsigned long)(brk - (char*)(p)) < MINSIZE) | |
5046 | brk += alignment; | |
f65fd747 | 5047 | |
fa8d436c | 5048 | newp = (mchunkptr)brk; |
f65fd747 UD |
5049 | leadsize = brk - (char*)(p); |
5050 | newsize = chunksize(p) - leadsize; | |
5051 | ||
fa8d436c UD |
5052 | /* For mmapped chunks, just adjust offset */ |
5053 | if (chunk_is_mmapped(p)) { | |
f65fd747 UD |
5054 | newp->prev_size = p->prev_size + leadsize; |
5055 | set_head(newp, newsize|IS_MMAPPED); | |
fa8d436c | 5056 | return chunk2mem(newp); |
f65fd747 | 5057 | } |
f65fd747 | 5058 | |
fa8d436c UD |
5059 | /* Otherwise, give back leader, use the rest */ |
5060 | set_head(newp, newsize | PREV_INUSE | | |
5061 | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
f65fd747 | 5062 | set_inuse_bit_at_offset(newp, newsize); |
fa8d436c UD |
5063 | set_head_size(p, leadsize | (av != &main_arena ? NON_MAIN_ARENA : 0)); |
5064 | _int_free(av, chunk2mem(p)); | |
f65fd747 UD |
5065 | p = newp; |
5066 | ||
fa8d436c UD |
5067 | assert (newsize >= nb && |
5068 | (((unsigned long)(chunk2mem(p))) % alignment) == 0); | |
f65fd747 UD |
5069 | } |
5070 | ||
5071 | /* Also give back spare room at the end */ | |
fa8d436c UD |
5072 | if (!chunk_is_mmapped(p)) { |
5073 | size = chunksize(p); | |
5074 | if ((unsigned long)(size) > (unsigned long)(nb + MINSIZE)) { | |
5075 | remainder_size = size - nb; | |
5076 | remainder = chunk_at_offset(p, nb); | |
5077 | set_head(remainder, remainder_size | PREV_INUSE | | |
5078 | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
5079 | set_head_size(p, nb); | |
5080 | _int_free(av, chunk2mem(remainder)); | |
5081 | } | |
f65fd747 UD |
5082 | } |
5083 | ||
fa8d436c UD |
5084 | check_inuse_chunk(av, p); |
5085 | return chunk2mem(p); | |
f65fd747 UD |
5086 | } |
5087 | ||
fa8d436c UD |
5088 | #if 0 |
5089 | /* | |
5090 | ------------------------------ calloc ------------------------------ | |
5091 | */ | |
5092 | ||
5093 | #if __STD_C | |
5094 | Void_t* cALLOc(size_t n_elements, size_t elem_size) | |
5095 | #else | |
5096 | Void_t* cALLOc(n_elements, elem_size) size_t n_elements; size_t elem_size; | |
5097 | #endif | |
5098 | { | |
5099 | mchunkptr p; | |
5100 | unsigned long clearsize; | |
5101 | unsigned long nclears; | |
5102 | INTERNAL_SIZE_T* d; | |
5103 | ||
5104 | Void_t* mem = mALLOc(n_elements * elem_size); | |
5105 | ||
5106 | if (mem != 0) { | |
5107 | p = mem2chunk(mem); | |
5108 | ||
5109 | #if MMAP_CLEARS | |
5110 | if (!chunk_is_mmapped(p)) /* don't need to clear mmapped space */ | |
5111 | #endif | |
a9177ff5 | 5112 | { |
fa8d436c UD |
5113 | /* |
5114 | Unroll clear of <= 36 bytes (72 if 8byte sizes) | |
5115 | We know that contents have an odd number of | |
5116 | INTERNAL_SIZE_T-sized words; minimally 3. | |
5117 | */ | |
5118 | ||
5119 | d = (INTERNAL_SIZE_T*)mem; | |
5120 | clearsize = chunksize(p) - SIZE_SZ; | |
5121 | nclears = clearsize / sizeof(INTERNAL_SIZE_T); | |
5122 | assert(nclears >= 3); | |
f65fd747 | 5123 | |
fa8d436c UD |
5124 | if (nclears > 9) |
5125 | MALLOC_ZERO(d, clearsize); | |
5126 | ||
5127 | else { | |
5128 | *(d+0) = 0; | |
5129 | *(d+1) = 0; | |
5130 | *(d+2) = 0; | |
5131 | if (nclears > 4) { | |
5132 | *(d+3) = 0; | |
5133 | *(d+4) = 0; | |
5134 | if (nclears > 6) { | |
5135 | *(d+5) = 0; | |
5136 | *(d+6) = 0; | |
5137 | if (nclears > 8) { | |
5138 | *(d+7) = 0; | |
5139 | *(d+8) = 0; | |
5140 | } | |
5141 | } | |
5142 | } | |
5143 | } | |
5144 | } | |
5145 | } | |
5146 | return mem; | |
5147 | } | |
5148 | #endif /* 0 */ | |
f65fd747 | 5149 | |
88764ae2 | 5150 | #ifndef _LIBC |
f65fd747 | 5151 | /* |
fa8d436c | 5152 | ------------------------- independent_calloc ------------------------- |
f65fd747 UD |
5153 | */ |
5154 | ||
f1c5213d | 5155 | Void_t** |
f65fd747 | 5156 | #if __STD_C |
fa8d436c | 5157 | _int_icalloc(mstate av, size_t n_elements, size_t elem_size, Void_t* chunks[]) |
f65fd747 | 5158 | #else |
fa8d436c UD |
5159 | _int_icalloc(av, n_elements, elem_size, chunks) |
5160 | mstate av; size_t n_elements; size_t elem_size; Void_t* chunks[]; | |
f65fd747 UD |
5161 | #endif |
5162 | { | |
fa8d436c UD |
5163 | size_t sz = elem_size; /* serves as 1-element array */ |
5164 | /* opts arg of 3 means all elements are same size, and should be cleared */ | |
5165 | return iALLOc(av, n_elements, &sz, 3, chunks); | |
f65fd747 UD |
5166 | } |
5167 | ||
5168 | /* | |
fa8d436c | 5169 | ------------------------- independent_comalloc ------------------------- |
f65fd747 UD |
5170 | */ |
5171 | ||
f1c5213d | 5172 | Void_t** |
f65fd747 | 5173 | #if __STD_C |
fa8d436c | 5174 | _int_icomalloc(mstate av, size_t n_elements, size_t sizes[], Void_t* chunks[]) |
f65fd747 | 5175 | #else |
fa8d436c UD |
5176 | _int_icomalloc(av, n_elements, sizes, chunks) |
5177 | mstate av; size_t n_elements; size_t sizes[]; Void_t* chunks[]; | |
f65fd747 UD |
5178 | #endif |
5179 | { | |
fa8d436c | 5180 | return iALLOc(av, n_elements, sizes, 0, chunks); |
f65fd747 UD |
5181 | } |
5182 | ||
f65fd747 | 5183 | |
fa8d436c UD |
5184 | /* |
5185 | ------------------------------ ialloc ------------------------------ | |
5186 | ialloc provides common support for independent_X routines, handling all of | |
5187 | the combinations that can result. | |
f65fd747 | 5188 | |
fa8d436c UD |
5189 | The opts arg has: |
5190 | bit 0 set if all elements are same size (using sizes[0]) | |
5191 | bit 1 set if elements should be zeroed | |
f65fd747 UD |
5192 | */ |
5193 | ||
fa8d436c UD |
5194 | |
5195 | static Void_t** | |
f65fd747 | 5196 | #if __STD_C |
fa8d436c | 5197 | iALLOc(mstate av, size_t n_elements, size_t* sizes, int opts, Void_t* chunks[]) |
f65fd747 | 5198 | #else |
fa8d436c UD |
5199 | iALLOc(av, n_elements, sizes, opts, chunks) |
5200 | mstate av; size_t n_elements; size_t* sizes; int opts; Void_t* chunks[]; | |
f65fd747 UD |
5201 | #endif |
5202 | { | |
fa8d436c UD |
5203 | INTERNAL_SIZE_T element_size; /* chunksize of each element, if all same */ |
5204 | INTERNAL_SIZE_T contents_size; /* total size of elements */ | |
5205 | INTERNAL_SIZE_T array_size; /* request size of pointer array */ | |
5206 | Void_t* mem; /* malloced aggregate space */ | |
5207 | mchunkptr p; /* corresponding chunk */ | |
5208 | INTERNAL_SIZE_T remainder_size; /* remaining bytes while splitting */ | |
5209 | Void_t** marray; /* either "chunks" or malloced ptr array */ | |
5210 | mchunkptr array_chunk; /* chunk for malloced ptr array */ | |
5211 | int mmx; /* to disable mmap */ | |
a9177ff5 | 5212 | INTERNAL_SIZE_T size; |
fa8d436c UD |
5213 | INTERNAL_SIZE_T size_flags; |
5214 | size_t i; | |
5215 | ||
5216 | /* Ensure initialization/consolidation */ | |
5217 | if (have_fastchunks(av)) malloc_consolidate(av); | |
5218 | ||
5219 | /* compute array length, if needed */ | |
5220 | if (chunks != 0) { | |
5221 | if (n_elements == 0) | |
5222 | return chunks; /* nothing to do */ | |
5223 | marray = chunks; | |
5224 | array_size = 0; | |
5225 | } | |
5226 | else { | |
5227 | /* if empty req, must still return chunk representing empty array */ | |
a9177ff5 | 5228 | if (n_elements == 0) |
fa8d436c UD |
5229 | return (Void_t**) _int_malloc(av, 0); |
5230 | marray = 0; | |
5231 | array_size = request2size(n_elements * (sizeof(Void_t*))); | |
5232 | } | |
f65fd747 | 5233 | |
fa8d436c UD |
5234 | /* compute total element size */ |
5235 | if (opts & 0x1) { /* all-same-size */ | |
5236 | element_size = request2size(*sizes); | |
5237 | contents_size = n_elements * element_size; | |
5238 | } | |
5239 | else { /* add up all the sizes */ | |
5240 | element_size = 0; | |
5241 | contents_size = 0; | |
a9177ff5 RM |
5242 | for (i = 0; i != n_elements; ++i) |
5243 | contents_size += request2size(sizes[i]); | |
10dc2a90 | 5244 | } |
f65fd747 | 5245 | |
fa8d436c UD |
5246 | /* subtract out alignment bytes from total to minimize overallocation */ |
5247 | size = contents_size + array_size - MALLOC_ALIGN_MASK; | |
a9177ff5 RM |
5248 | |
5249 | /* | |
fa8d436c UD |
5250 | Allocate the aggregate chunk. |
5251 | But first disable mmap so malloc won't use it, since | |
5252 | we would not be able to later free/realloc space internal | |
5253 | to a segregated mmap region. | |
5254 | */ | |
5255 | mmx = mp_.n_mmaps_max; /* disable mmap */ | |
5256 | mp_.n_mmaps_max = 0; | |
5257 | mem = _int_malloc(av, size); | |
5258 | mp_.n_mmaps_max = mmx; /* reset mmap */ | |
a9177ff5 | 5259 | if (mem == 0) |
f65fd747 UD |
5260 | return 0; |
5261 | ||
fa8d436c | 5262 | p = mem2chunk(mem); |
a9177ff5 | 5263 | assert(!chunk_is_mmapped(p)); |
fa8d436c | 5264 | remainder_size = chunksize(p); |
f65fd747 | 5265 | |
fa8d436c UD |
5266 | if (opts & 0x2) { /* optionally clear the elements */ |
5267 | MALLOC_ZERO(mem, remainder_size - SIZE_SZ - array_size); | |
7799b7b3 | 5268 | } |
f65fd747 | 5269 | |
fa8d436c | 5270 | size_flags = PREV_INUSE | (av != &main_arena ? NON_MAIN_ARENA : 0); |
f65fd747 | 5271 | |
fa8d436c UD |
5272 | /* If not provided, allocate the pointer array as final part of chunk */ |
5273 | if (marray == 0) { | |
5274 | array_chunk = chunk_at_offset(p, contents_size); | |
5275 | marray = (Void_t**) (chunk2mem(array_chunk)); | |
5276 | set_head(array_chunk, (remainder_size - contents_size) | size_flags); | |
5277 | remainder_size = contents_size; | |
5278 | } | |
f65fd747 | 5279 | |
fa8d436c UD |
5280 | /* split out elements */ |
5281 | for (i = 0; ; ++i) { | |
5282 | marray[i] = chunk2mem(p); | |
5283 | if (i != n_elements-1) { | |
a9177ff5 | 5284 | if (element_size != 0) |
fa8d436c UD |
5285 | size = element_size; |
5286 | else | |
a9177ff5 | 5287 | size = request2size(sizes[i]); |
fa8d436c UD |
5288 | remainder_size -= size; |
5289 | set_head(p, size | size_flags); | |
5290 | p = chunk_at_offset(p, size); | |
5291 | } | |
5292 | else { /* the final element absorbs any overallocation slop */ | |
5293 | set_head(p, remainder_size | size_flags); | |
5294 | break; | |
5295 | } | |
5296 | } | |
f65fd747 | 5297 | |
fa8d436c UD |
5298 | #if MALLOC_DEBUG |
5299 | if (marray != chunks) { | |
5300 | /* final element must have exactly exhausted chunk */ | |
a9177ff5 | 5301 | if (element_size != 0) |
fa8d436c UD |
5302 | assert(remainder_size == element_size); |
5303 | else | |
5304 | assert(remainder_size == request2size(sizes[i])); | |
5305 | check_inuse_chunk(av, mem2chunk(marray)); | |
7799b7b3 | 5306 | } |
fa8d436c UD |
5307 | |
5308 | for (i = 0; i != n_elements; ++i) | |
5309 | check_inuse_chunk(av, mem2chunk(marray[i])); | |
f65fd747 UD |
5310 | #endif |
5311 | ||
fa8d436c | 5312 | return marray; |
f65fd747 | 5313 | } |
88764ae2 | 5314 | #endif /* _LIBC */ |
f65fd747 | 5315 | |
f65fd747 | 5316 | |
fa8d436c UD |
5317 | /* |
5318 | ------------------------------ valloc ------------------------------ | |
f65fd747 UD |
5319 | */ |
5320 | ||
f1c5213d | 5321 | Void_t* |
f65fd747 | 5322 | #if __STD_C |
fa8d436c | 5323 | _int_valloc(mstate av, size_t bytes) |
f65fd747 | 5324 | #else |
fa8d436c | 5325 | _int_valloc(av, bytes) mstate av; size_t bytes; |
f65fd747 UD |
5326 | #endif |
5327 | { | |
fa8d436c UD |
5328 | /* Ensure initialization/consolidation */ |
5329 | if (have_fastchunks(av)) malloc_consolidate(av); | |
5330 | return _int_memalign(av, mp_.pagesize, bytes); | |
f65fd747 | 5331 | } |
f65fd747 UD |
5332 | |
5333 | /* | |
fa8d436c | 5334 | ------------------------------ pvalloc ------------------------------ |
f65fd747 UD |
5335 | */ |
5336 | ||
fa8d436c | 5337 | |
f1c5213d | 5338 | Void_t* |
f65fd747 | 5339 | #if __STD_C |
fa8d436c | 5340 | _int_pvalloc(mstate av, size_t bytes) |
f65fd747 | 5341 | #else |
fa8d436c | 5342 | _int_pvalloc(av, bytes) mstate av, size_t bytes; |
f65fd747 UD |
5343 | #endif |
5344 | { | |
fa8d436c | 5345 | size_t pagesz; |
f65fd747 | 5346 | |
fa8d436c UD |
5347 | /* Ensure initialization/consolidation */ |
5348 | if (have_fastchunks(av)) malloc_consolidate(av); | |
5349 | pagesz = mp_.pagesize; | |
5350 | return _int_memalign(av, pagesz, (bytes + pagesz - 1) & ~(pagesz - 1)); | |
f65fd747 | 5351 | } |
a9177ff5 | 5352 | |
f65fd747 | 5353 | |
fa8d436c UD |
5354 | /* |
5355 | ------------------------------ malloc_trim ------------------------------ | |
5356 | */ | |
8a4b65b4 | 5357 | |
f65fd747 | 5358 | #if __STD_C |
fa8d436c | 5359 | int mTRIm(size_t pad) |
f65fd747 | 5360 | #else |
fa8d436c | 5361 | int mTRIm(pad) size_t pad; |
f65fd747 UD |
5362 | #endif |
5363 | { | |
fa8d436c | 5364 | mstate av = &main_arena; /* already locked */ |
f65fd747 | 5365 | |
fa8d436c UD |
5366 | /* Ensure initialization/consolidation */ |
5367 | malloc_consolidate(av); | |
8a4b65b4 | 5368 | |
a9177ff5 | 5369 | #ifndef MORECORE_CANNOT_TRIM |
fa8d436c | 5370 | return sYSTRIm(pad, av); |
8a4b65b4 | 5371 | #else |
fa8d436c | 5372 | return 0; |
f65fd747 | 5373 | #endif |
f65fd747 UD |
5374 | } |
5375 | ||
f65fd747 UD |
5376 | |
5377 | /* | |
fa8d436c | 5378 | ------------------------- malloc_usable_size ------------------------- |
f65fd747 UD |
5379 | */ |
5380 | ||
5381 | #if __STD_C | |
fa8d436c | 5382 | size_t mUSABLe(Void_t* mem) |
f65fd747 | 5383 | #else |
fa8d436c | 5384 | size_t mUSABLe(mem) Void_t* mem; |
f65fd747 UD |
5385 | #endif |
5386 | { | |
5387 | mchunkptr p; | |
fa8d436c | 5388 | if (mem != 0) { |
f65fd747 | 5389 | p = mem2chunk(mem); |
fa8d436c UD |
5390 | if (chunk_is_mmapped(p)) |
5391 | return chunksize(p) - 2*SIZE_SZ; | |
5392 | else if (inuse(p)) | |
f65fd747 | 5393 | return chunksize(p) - SIZE_SZ; |
f65fd747 | 5394 | } |
fa8d436c | 5395 | return 0; |
f65fd747 UD |
5396 | } |
5397 | ||
fa8d436c UD |
5398 | /* |
5399 | ------------------------------ mallinfo ------------------------------ | |
5400 | */ | |
f65fd747 | 5401 | |
fa8d436c | 5402 | struct mallinfo mALLINFo(mstate av) |
f65fd747 | 5403 | { |
fa8d436c | 5404 | struct mallinfo mi; |
6dd67bd5 | 5405 | size_t i; |
f65fd747 UD |
5406 | mbinptr b; |
5407 | mchunkptr p; | |
f65fd747 | 5408 | INTERNAL_SIZE_T avail; |
fa8d436c UD |
5409 | INTERNAL_SIZE_T fastavail; |
5410 | int nblocks; | |
5411 | int nfastblocks; | |
f65fd747 | 5412 | |
fa8d436c UD |
5413 | /* Ensure initialization */ |
5414 | if (av->top == 0) malloc_consolidate(av); | |
8a4b65b4 | 5415 | |
fa8d436c | 5416 | check_malloc_state(av); |
8a4b65b4 | 5417 | |
fa8d436c UD |
5418 | /* Account for top */ |
5419 | avail = chunksize(av->top); | |
5420 | nblocks = 1; /* top always exists */ | |
f65fd747 | 5421 | |
fa8d436c UD |
5422 | /* traverse fastbins */ |
5423 | nfastblocks = 0; | |
5424 | fastavail = 0; | |
5425 | ||
5426 | for (i = 0; i < NFASTBINS; ++i) { | |
5427 | for (p = av->fastbins[i]; p != 0; p = p->fd) { | |
5428 | ++nfastblocks; | |
5429 | fastavail += chunksize(p); | |
5430 | } | |
5431 | } | |
5432 | ||
5433 | avail += fastavail; | |
f65fd747 | 5434 | |
fa8d436c UD |
5435 | /* traverse regular bins */ |
5436 | for (i = 1; i < NBINS; ++i) { | |
5437 | b = bin_at(av, i); | |
5438 | for (p = last(b); p != b; p = p->bk) { | |
5439 | ++nblocks; | |
5440 | avail += chunksize(p); | |
5441 | } | |
5442 | } | |
f65fd747 | 5443 | |
fa8d436c UD |
5444 | mi.smblks = nfastblocks; |
5445 | mi.ordblks = nblocks; | |
5446 | mi.fordblks = avail; | |
5447 | mi.uordblks = av->system_mem - avail; | |
5448 | mi.arena = av->system_mem; | |
5449 | mi.hblks = mp_.n_mmaps; | |
5450 | mi.hblkhd = mp_.mmapped_mem; | |
5451 | mi.fsmblks = fastavail; | |
5452 | mi.keepcost = chunksize(av->top); | |
5453 | mi.usmblks = mp_.max_total_mem; | |
5454 | return mi; | |
5455 | } | |
f65fd747 | 5456 | |
fa8d436c UD |
5457 | /* |
5458 | ------------------------------ malloc_stats ------------------------------ | |
f65fd747 UD |
5459 | */ |
5460 | ||
fa8d436c | 5461 | void mSTATs() |
f65fd747 | 5462 | { |
8a4b65b4 | 5463 | int i; |
fa8d436c | 5464 | mstate ar_ptr; |
8a4b65b4 | 5465 | struct mallinfo mi; |
fa8d436c | 5466 | unsigned int in_use_b = mp_.mmapped_mem, system_b = in_use_b; |
8a4b65b4 UD |
5467 | #if THREAD_STATS |
5468 | long stat_lock_direct = 0, stat_lock_loop = 0, stat_lock_wait = 0; | |
5469 | #endif | |
5470 | ||
a234e27d UD |
5471 | if(__malloc_initialized < 0) |
5472 | ptmalloc_init (); | |
8dab36a1 UD |
5473 | #ifdef _LIBC |
5474 | _IO_flockfile (stderr); | |
5475 | int old_flags2 = ((_IO_FILE *) stderr)->_flags2; | |
5476 | ((_IO_FILE *) stderr)->_flags2 |= _IO_FLAGS2_NOTCANCEL; | |
5477 | #endif | |
fa8d436c UD |
5478 | for (i=0, ar_ptr = &main_arena;; i++) { |
5479 | (void)mutex_lock(&ar_ptr->mutex); | |
5480 | mi = mALLINFo(ar_ptr); | |
8a4b65b4 UD |
5481 | fprintf(stderr, "Arena %d:\n", i); |
5482 | fprintf(stderr, "system bytes = %10u\n", (unsigned int)mi.arena); | |
5483 | fprintf(stderr, "in use bytes = %10u\n", (unsigned int)mi.uordblks); | |
fa8d436c UD |
5484 | #if MALLOC_DEBUG > 1 |
5485 | if (i > 0) | |
5486 | dump_heap(heap_for_ptr(top(ar_ptr))); | |
5487 | #endif | |
8a4b65b4 UD |
5488 | system_b += mi.arena; |
5489 | in_use_b += mi.uordblks; | |
5490 | #if THREAD_STATS | |
5491 | stat_lock_direct += ar_ptr->stat_lock_direct; | |
5492 | stat_lock_loop += ar_ptr->stat_lock_loop; | |
5493 | stat_lock_wait += ar_ptr->stat_lock_wait; | |
5494 | #endif | |
fa8d436c | 5495 | (void)mutex_unlock(&ar_ptr->mutex); |
7e3be507 UD |
5496 | ar_ptr = ar_ptr->next; |
5497 | if(ar_ptr == &main_arena) break; | |
8a4b65b4 | 5498 | } |
7799b7b3 | 5499 | #if HAVE_MMAP |
8a4b65b4 | 5500 | fprintf(stderr, "Total (incl. mmap):\n"); |
7799b7b3 UD |
5501 | #else |
5502 | fprintf(stderr, "Total:\n"); | |
5503 | #endif | |
8a4b65b4 UD |
5504 | fprintf(stderr, "system bytes = %10u\n", system_b); |
5505 | fprintf(stderr, "in use bytes = %10u\n", in_use_b); | |
5506 | #ifdef NO_THREADS | |
fa8d436c | 5507 | fprintf(stderr, "max system bytes = %10u\n", (unsigned int)mp_.max_total_mem); |
8a4b65b4 | 5508 | #endif |
f65fd747 | 5509 | #if HAVE_MMAP |
fa8d436c UD |
5510 | fprintf(stderr, "max mmap regions = %10u\n", (unsigned int)mp_.max_n_mmaps); |
5511 | fprintf(stderr, "max mmap bytes = %10lu\n", | |
5512 | (unsigned long)mp_.max_mmapped_mem); | |
f65fd747 UD |
5513 | #endif |
5514 | #if THREAD_STATS | |
8a4b65b4 | 5515 | fprintf(stderr, "heaps created = %10d\n", stat_n_heaps); |
f65fd747 UD |
5516 | fprintf(stderr, "locked directly = %10ld\n", stat_lock_direct); |
5517 | fprintf(stderr, "locked in loop = %10ld\n", stat_lock_loop); | |
8a4b65b4 UD |
5518 | fprintf(stderr, "locked waiting = %10ld\n", stat_lock_wait); |
5519 | fprintf(stderr, "locked total = %10ld\n", | |
5520 | stat_lock_direct + stat_lock_loop + stat_lock_wait); | |
f65fd747 | 5521 | #endif |
8dab36a1 UD |
5522 | #ifdef _LIBC |
5523 | ((_IO_FILE *) stderr)->_flags2 |= old_flags2; | |
5524 | _IO_funlockfile (stderr); | |
5525 | #endif | |
f65fd747 UD |
5526 | } |
5527 | ||
f65fd747 UD |
5528 | |
5529 | /* | |
fa8d436c | 5530 | ------------------------------ mallopt ------------------------------ |
f65fd747 UD |
5531 | */ |
5532 | ||
5533 | #if __STD_C | |
5534 | int mALLOPt(int param_number, int value) | |
5535 | #else | |
5536 | int mALLOPt(param_number, value) int param_number; int value; | |
5537 | #endif | |
5538 | { | |
fa8d436c UD |
5539 | mstate av = &main_arena; |
5540 | int res = 1; | |
f65fd747 | 5541 | |
0cb71e02 UD |
5542 | if(__malloc_initialized < 0) |
5543 | ptmalloc_init (); | |
fa8d436c UD |
5544 | (void)mutex_lock(&av->mutex); |
5545 | /* Ensure initialization/consolidation */ | |
5546 | malloc_consolidate(av); | |
2f6d1f1b | 5547 | |
fa8d436c UD |
5548 | switch(param_number) { |
5549 | case M_MXFAST: | |
5550 | if (value >= 0 && value <= MAX_FAST_SIZE) { | |
9bf248c6 | 5551 | set_max_fast(value); |
fa8d436c UD |
5552 | } |
5553 | else | |
5554 | res = 0; | |
5555 | break; | |
2f6d1f1b | 5556 | |
fa8d436c UD |
5557 | case M_TRIM_THRESHOLD: |
5558 | mp_.trim_threshold = value; | |
1d05c2fb | 5559 | mp_.no_dyn_threshold = 1; |
fa8d436c | 5560 | break; |
2f6d1f1b | 5561 | |
fa8d436c UD |
5562 | case M_TOP_PAD: |
5563 | mp_.top_pad = value; | |
1d05c2fb | 5564 | mp_.no_dyn_threshold = 1; |
fa8d436c | 5565 | break; |
2f6d1f1b | 5566 | |
fa8d436c UD |
5567 | case M_MMAP_THRESHOLD: |
5568 | #if USE_ARENAS | |
5569 | /* Forbid setting the threshold too high. */ | |
5570 | if((unsigned long)value > HEAP_MAX_SIZE/2) | |
5571 | res = 0; | |
5572 | else | |
2f6d1f1b | 5573 | #endif |
fa8d436c | 5574 | mp_.mmap_threshold = value; |
1d05c2fb | 5575 | mp_.no_dyn_threshold = 1; |
fa8d436c | 5576 | break; |
2f6d1f1b | 5577 | |
fa8d436c UD |
5578 | case M_MMAP_MAX: |
5579 | #if !HAVE_MMAP | |
5580 | if (value != 0) | |
5581 | res = 0; | |
5582 | else | |
9a51759b | 5583 | #endif |
fa8d436c | 5584 | mp_.n_mmaps_max = value; |
1d05c2fb | 5585 | mp_.no_dyn_threshold = 1; |
fa8d436c | 5586 | break; |
10dc2a90 | 5587 | |
fa8d436c UD |
5588 | case M_CHECK_ACTION: |
5589 | check_action = value; | |
5590 | break; | |
854278df UD |
5591 | |
5592 | case M_PERTURB: | |
5593 | perturb_byte = value; | |
5594 | break; | |
b22fc5f5 | 5595 | } |
fa8d436c UD |
5596 | (void)mutex_unlock(&av->mutex); |
5597 | return res; | |
b22fc5f5 UD |
5598 | } |
5599 | ||
10dc2a90 | 5600 | |
a9177ff5 | 5601 | /* |
fa8d436c UD |
5602 | -------------------- Alternative MORECORE functions -------------------- |
5603 | */ | |
10dc2a90 | 5604 | |
b22fc5f5 | 5605 | |
fa8d436c UD |
5606 | /* |
5607 | General Requirements for MORECORE. | |
b22fc5f5 | 5608 | |
fa8d436c | 5609 | The MORECORE function must have the following properties: |
b22fc5f5 | 5610 | |
fa8d436c | 5611 | If MORECORE_CONTIGUOUS is false: |
10dc2a90 | 5612 | |
fa8d436c UD |
5613 | * MORECORE must allocate in multiples of pagesize. It will |
5614 | only be called with arguments that are multiples of pagesize. | |
10dc2a90 | 5615 | |
a9177ff5 | 5616 | * MORECORE(0) must return an address that is at least |
fa8d436c | 5617 | MALLOC_ALIGNMENT aligned. (Page-aligning always suffices.) |
10dc2a90 | 5618 | |
fa8d436c | 5619 | else (i.e. If MORECORE_CONTIGUOUS is true): |
10dc2a90 | 5620 | |
fa8d436c UD |
5621 | * Consecutive calls to MORECORE with positive arguments |
5622 | return increasing addresses, indicating that space has been | |
5623 | contiguously extended. | |
10dc2a90 | 5624 | |
fa8d436c UD |
5625 | * MORECORE need not allocate in multiples of pagesize. |
5626 | Calls to MORECORE need not have args of multiples of pagesize. | |
10dc2a90 | 5627 | |
fa8d436c | 5628 | * MORECORE need not page-align. |
10dc2a90 | 5629 | |
fa8d436c | 5630 | In either case: |
10dc2a90 | 5631 | |
fa8d436c UD |
5632 | * MORECORE may allocate more memory than requested. (Or even less, |
5633 | but this will generally result in a malloc failure.) | |
10dc2a90 | 5634 | |
fa8d436c UD |
5635 | * MORECORE must not allocate memory when given argument zero, but |
5636 | instead return one past the end address of memory from previous | |
5637 | nonzero call. This malloc does NOT call MORECORE(0) | |
5638 | until at least one call with positive arguments is made, so | |
5639 | the initial value returned is not important. | |
10dc2a90 | 5640 | |
fa8d436c UD |
5641 | * Even though consecutive calls to MORECORE need not return contiguous |
5642 | addresses, it must be OK for malloc'ed chunks to span multiple | |
5643 | regions in those cases where they do happen to be contiguous. | |
10dc2a90 | 5644 | |
fa8d436c UD |
5645 | * MORECORE need not handle negative arguments -- it may instead |
5646 | just return MORECORE_FAILURE when given negative arguments. | |
5647 | Negative arguments are always multiples of pagesize. MORECORE | |
5648 | must not misinterpret negative args as large positive unsigned | |
5649 | args. You can suppress all such calls from even occurring by defining | |
5650 | MORECORE_CANNOT_TRIM, | |
10dc2a90 | 5651 | |
fa8d436c UD |
5652 | There is some variation across systems about the type of the |
5653 | argument to sbrk/MORECORE. If size_t is unsigned, then it cannot | |
5654 | actually be size_t, because sbrk supports negative args, so it is | |
5655 | normally the signed type of the same width as size_t (sometimes | |
5656 | declared as "intptr_t", and sometimes "ptrdiff_t"). It doesn't much | |
5657 | matter though. Internally, we use "long" as arguments, which should | |
5658 | work across all reasonable possibilities. | |
ee74a442 | 5659 | |
fa8d436c UD |
5660 | Additionally, if MORECORE ever returns failure for a positive |
5661 | request, and HAVE_MMAP is true, then mmap is used as a noncontiguous | |
5662 | system allocator. This is a useful backup strategy for systems with | |
5663 | holes in address spaces -- in this case sbrk cannot contiguously | |
5664 | expand the heap, but mmap may be able to map noncontiguous space. | |
7e3be507 | 5665 | |
fa8d436c UD |
5666 | If you'd like mmap to ALWAYS be used, you can define MORECORE to be |
5667 | a function that always returns MORECORE_FAILURE. | |
2e65ca2b | 5668 | |
fa8d436c UD |
5669 | If you are using this malloc with something other than sbrk (or its |
5670 | emulation) to supply memory regions, you probably want to set | |
5671 | MORECORE_CONTIGUOUS as false. As an example, here is a custom | |
5672 | allocator kindly contributed for pre-OSX macOS. It uses virtually | |
5673 | but not necessarily physically contiguous non-paged memory (locked | |
5674 | in, present and won't get swapped out). You can use it by | |
5675 | uncommenting this section, adding some #includes, and setting up the | |
5676 | appropriate defines above: | |
7e3be507 | 5677 | |
fa8d436c UD |
5678 | #define MORECORE osMoreCore |
5679 | #define MORECORE_CONTIGUOUS 0 | |
7e3be507 | 5680 | |
fa8d436c UD |
5681 | There is also a shutdown routine that should somehow be called for |
5682 | cleanup upon program exit. | |
7e3be507 | 5683 | |
fa8d436c UD |
5684 | #define MAX_POOL_ENTRIES 100 |
5685 | #define MINIMUM_MORECORE_SIZE (64 * 1024) | |
5686 | static int next_os_pool; | |
5687 | void *our_os_pools[MAX_POOL_ENTRIES]; | |
7e3be507 | 5688 | |
fa8d436c UD |
5689 | void *osMoreCore(int size) |
5690 | { | |
5691 | void *ptr = 0; | |
5692 | static void *sbrk_top = 0; | |
ca34d7a7 | 5693 | |
fa8d436c UD |
5694 | if (size > 0) |
5695 | { | |
5696 | if (size < MINIMUM_MORECORE_SIZE) | |
5697 | size = MINIMUM_MORECORE_SIZE; | |
5698 | if (CurrentExecutionLevel() == kTaskLevel) | |
5699 | ptr = PoolAllocateResident(size + RM_PAGE_SIZE, 0); | |
5700 | if (ptr == 0) | |
5701 | { | |
5702 | return (void *) MORECORE_FAILURE; | |
5703 | } | |
5704 | // save ptrs so they can be freed during cleanup | |
5705 | our_os_pools[next_os_pool] = ptr; | |
5706 | next_os_pool++; | |
5707 | ptr = (void *) ((((unsigned long) ptr) + RM_PAGE_MASK) & ~RM_PAGE_MASK); | |
5708 | sbrk_top = (char *) ptr + size; | |
5709 | return ptr; | |
5710 | } | |
5711 | else if (size < 0) | |
5712 | { | |
5713 | // we don't currently support shrink behavior | |
5714 | return (void *) MORECORE_FAILURE; | |
5715 | } | |
5716 | else | |
5717 | { | |
5718 | return sbrk_top; | |
431c33c0 | 5719 | } |
ca34d7a7 | 5720 | } |
ca34d7a7 | 5721 | |
fa8d436c UD |
5722 | // cleanup any allocated memory pools |
5723 | // called as last thing before shutting down driver | |
ca34d7a7 | 5724 | |
fa8d436c | 5725 | void osCleanupMem(void) |
ca34d7a7 | 5726 | { |
fa8d436c | 5727 | void **ptr; |
ca34d7a7 | 5728 | |
fa8d436c UD |
5729 | for (ptr = our_os_pools; ptr < &our_os_pools[MAX_POOL_ENTRIES]; ptr++) |
5730 | if (*ptr) | |
5731 | { | |
5732 | PoolDeallocate(*ptr); | |
5733 | *ptr = 0; | |
5734 | } | |
5735 | } | |
ee74a442 | 5736 | |
fa8d436c | 5737 | */ |
f65fd747 | 5738 | |
7e3be507 | 5739 | |
3e030bd5 UD |
5740 | /* Helper code. */ |
5741 | ||
ae7f5313 UD |
5742 | extern char **__libc_argv attribute_hidden; |
5743 | ||
3e030bd5 | 5744 | static void |
6bf4302e | 5745 | malloc_printerr(int action, const char *str, void *ptr) |
3e030bd5 | 5746 | { |
553cc5f9 UD |
5747 | if ((action & 5) == 5) |
5748 | __libc_message (action & 2, "%s\n", str); | |
5749 | else if (action & 1) | |
3e030bd5 | 5750 | { |
a9055cab | 5751 | char buf[2 * sizeof (uintptr_t) + 1]; |
3e030bd5 | 5752 | |
a9055cab UD |
5753 | buf[sizeof (buf) - 1] = '\0'; |
5754 | char *cp = _itoa_word ((uintptr_t) ptr, &buf[sizeof (buf) - 1], 16, 0); | |
5755 | while (cp > buf) | |
5756 | *--cp = '0'; | |
5757 | ||
5758 | __libc_message (action & 2, | |
553cc5f9 | 5759 | "*** glibc detected *** %s: %s: 0x%s ***\n", |
ae7f5313 | 5760 | __libc_argv[0] ?: "<unknown>", str, cp); |
3e030bd5 | 5761 | } |
a9055cab | 5762 | else if (action & 2) |
3e030bd5 UD |
5763 | abort (); |
5764 | } | |
5765 | ||
7e3be507 | 5766 | #ifdef _LIBC |
b2bffca2 | 5767 | # include <sys/param.h> |
fa8d436c | 5768 | |
a204dbb2 UD |
5769 | /* We need a wrapper function for one of the additions of POSIX. */ |
5770 | int | |
5771 | __posix_memalign (void **memptr, size_t alignment, size_t size) | |
5772 | { | |
5773 | void *mem; | |
e796f92f UD |
5774 | __malloc_ptr_t (*hook) __MALLOC_PMT ((size_t, size_t, |
5775 | __const __malloc_ptr_t)) = | |
5776 | __memalign_hook; | |
a204dbb2 UD |
5777 | |
5778 | /* Test whether the SIZE argument is valid. It must be a power of | |
5779 | two multiple of sizeof (void *). */ | |
de02bd05 UD |
5780 | if (alignment % sizeof (void *) != 0 |
5781 | || !powerof2 (alignment / sizeof (void *)) != 0 | |
5782 | || alignment == 0) | |
a204dbb2 UD |
5783 | return EINVAL; |
5784 | ||
e796f92f UD |
5785 | /* Call the hook here, so that caller is posix_memalign's caller |
5786 | and not posix_memalign itself. */ | |
5787 | if (hook != NULL) | |
5788 | mem = (*hook)(alignment, size, RETURN_ADDRESS (0)); | |
5789 | else | |
aa420660 | 5790 | mem = public_mEMALIGn (alignment, size); |
a204dbb2 | 5791 | |
fa8d436c UD |
5792 | if (mem != NULL) { |
5793 | *memptr = mem; | |
5794 | return 0; | |
5795 | } | |
a204dbb2 UD |
5796 | |
5797 | return ENOMEM; | |
5798 | } | |
5799 | weak_alias (__posix_memalign, posix_memalign) | |
5800 | ||
eba19d2b UD |
5801 | strong_alias (__libc_calloc, __calloc) weak_alias (__libc_calloc, calloc) |
5802 | strong_alias (__libc_free, __cfree) weak_alias (__libc_free, cfree) | |
5803 | strong_alias (__libc_free, __free) strong_alias (__libc_free, free) | |
5804 | strong_alias (__libc_malloc, __malloc) strong_alias (__libc_malloc, malloc) | |
5805 | strong_alias (__libc_memalign, __memalign) | |
5806 | weak_alias (__libc_memalign, memalign) | |
5807 | strong_alias (__libc_realloc, __realloc) strong_alias (__libc_realloc, realloc) | |
5808 | strong_alias (__libc_valloc, __valloc) weak_alias (__libc_valloc, valloc) | |
5809 | strong_alias (__libc_pvalloc, __pvalloc) weak_alias (__libc_pvalloc, pvalloc) | |
5810 | strong_alias (__libc_mallinfo, __mallinfo) | |
5811 | weak_alias (__libc_mallinfo, mallinfo) | |
5812 | strong_alias (__libc_mallopt, __mallopt) weak_alias (__libc_mallopt, mallopt) | |
7e3be507 UD |
5813 | |
5814 | weak_alias (__malloc_stats, malloc_stats) | |
5815 | weak_alias (__malloc_usable_size, malloc_usable_size) | |
5816 | weak_alias (__malloc_trim, malloc_trim) | |
2f6d1f1b UD |
5817 | weak_alias (__malloc_get_state, malloc_get_state) |
5818 | weak_alias (__malloc_set_state, malloc_set_state) | |
7e3be507 | 5819 | |
fa8d436c | 5820 | #endif /* _LIBC */ |
f65fd747 | 5821 | |
fa8d436c | 5822 | /* ------------------------------------------------------------ |
f65fd747 UD |
5823 | History: |
5824 | ||
fa8d436c | 5825 | [see ftp://g.oswego.edu/pub/misc/malloc.c for the history of dlmalloc] |
f65fd747 UD |
5826 | |
5827 | */ | |
fa8d436c UD |
5828 | /* |
5829 | * Local variables: | |
5830 | * c-basic-offset: 2 | |
5831 | * End: | |
5832 | */ |