]>
Commit | Line | Data |
---|---|---|
56137dbc | 1 | /* Malloc implementation for multiple threads without lock contention. |
d4697bc9 | 2 | Copyright (C) 1996-2014 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 |
59ba27a6 PE |
18 | License along with the GNU C Library; see the file COPYING.LIB. If |
19 | not, see <http://www.gnu.org/licenses/>. */ | |
f65fd747 | 20 | |
fa8d436c UD |
21 | /* |
22 | This is a version (aka ptmalloc2) of malloc/free/realloc written by | |
23 | Doug Lea and adapted to multiple threads/arenas by Wolfram Gloger. | |
24 | ||
da2d2fb6 UD |
25 | There have been substantial changesmade after the integration into |
26 | glibc in all parts of the code. Do not look for much commonality | |
27 | with the ptmalloc2 version. | |
28 | ||
fa8d436c | 29 | * Version ptmalloc2-20011215 |
fa8d436c UD |
30 | based on: |
31 | VERSION 2.7.0 Sun Mar 11 14:14:06 2001 Doug Lea (dl at gee) | |
f65fd747 | 32 | |
fa8d436c | 33 | * Quickstart |
f65fd747 | 34 | |
fa8d436c UD |
35 | In order to compile this implementation, a Makefile is provided with |
36 | the ptmalloc2 distribution, which has pre-defined targets for some | |
37 | popular systems (e.g. "make posix" for Posix threads). All that is | |
38 | typically required with regard to compiler flags is the selection of | |
39 | the thread package via defining one out of USE_PTHREADS, USE_THR or | |
40 | USE_SPROC. Check the thread-m.h file for what effects this has. | |
41 | Many/most systems will additionally require USE_TSD_DATA_HACK to be | |
42 | defined, so this is the default for "make posix". | |
f65fd747 UD |
43 | |
44 | * Why use this malloc? | |
45 | ||
46 | This is not the fastest, most space-conserving, most portable, or | |
47 | most tunable malloc ever written. However it is among the fastest | |
48 | while also being among the most space-conserving, portable and tunable. | |
49 | Consistent balance across these factors results in a good general-purpose | |
fa8d436c UD |
50 | allocator for malloc-intensive programs. |
51 | ||
52 | The main properties of the algorithms are: | |
53 | * For large (>= 512 bytes) requests, it is a pure best-fit allocator, | |
54 | with ties normally decided via FIFO (i.e. least recently used). | |
55 | * For small (<= 64 bytes by default) requests, it is a caching | |
56 | allocator, that maintains pools of quickly recycled chunks. | |
57 | * In between, and for combinations of large and small requests, it does | |
58 | the best it can trying to meet both goals at once. | |
59 | * For very large requests (>= 128KB by default), it relies on system | |
60 | memory mapping facilities, if supported. | |
61 | ||
62 | For a longer but slightly out of date high-level description, see | |
63 | http://gee.cs.oswego.edu/dl/html/malloc.html | |
64 | ||
65 | You may already by default be using a C library containing a malloc | |
66 | that is based on some version of this malloc (for example in | |
67 | linux). You might still want to use the one in this file in order to | |
68 | customize settings or to avoid overheads associated with library | |
69 | versions. | |
70 | ||
71 | * Contents, described in more detail in "description of public routines" below. | |
72 | ||
73 | Standard (ANSI/SVID/...) functions: | |
74 | malloc(size_t n); | |
75 | calloc(size_t n_elements, size_t element_size); | |
22a89187 UD |
76 | free(void* p); |
77 | realloc(void* p, size_t n); | |
fa8d436c UD |
78 | memalign(size_t alignment, size_t n); |
79 | valloc(size_t n); | |
80 | mallinfo() | |
81 | mallopt(int parameter_number, int parameter_value) | |
82 | ||
83 | Additional functions: | |
22a89187 UD |
84 | independent_calloc(size_t n_elements, size_t size, void* chunks[]); |
85 | independent_comalloc(size_t n_elements, size_t sizes[], void* chunks[]); | |
fa8d436c | 86 | pvalloc(size_t n); |
22a89187 | 87 | cfree(void* p); |
fa8d436c | 88 | malloc_trim(size_t pad); |
22a89187 | 89 | malloc_usable_size(void* p); |
fa8d436c | 90 | malloc_stats(); |
f65fd747 UD |
91 | |
92 | * Vital statistics: | |
93 | ||
fa8d436c | 94 | Supported pointer representation: 4 or 8 bytes |
a9177ff5 | 95 | Supported size_t representation: 4 or 8 bytes |
f65fd747 | 96 | Note that size_t is allowed to be 4 bytes even if pointers are 8. |
fa8d436c UD |
97 | You can adjust this by defining INTERNAL_SIZE_T |
98 | ||
99 | Alignment: 2 * sizeof(size_t) (default) | |
100 | (i.e., 8 byte alignment with 4byte size_t). This suffices for | |
101 | nearly all current machines and C compilers. However, you can | |
102 | define MALLOC_ALIGNMENT to be wider than this if necessary. | |
f65fd747 | 103 | |
fa8d436c UD |
104 | Minimum overhead per allocated chunk: 4 or 8 bytes |
105 | Each malloced chunk has a hidden word of overhead holding size | |
f65fd747 UD |
106 | and status information. |
107 | ||
108 | Minimum allocated size: 4-byte ptrs: 16 bytes (including 4 overhead) | |
72f90263 | 109 | 8-byte ptrs: 24/32 bytes (including, 4/8 overhead) |
f65fd747 UD |
110 | |
111 | When a chunk is freed, 12 (for 4byte ptrs) or 20 (for 8 byte | |
112 | ptrs but 4 byte size) or 24 (for 8/8) additional bytes are | |
fa8d436c UD |
113 | needed; 4 (8) for a trailing size field and 8 (16) bytes for |
114 | free list pointers. Thus, the minimum allocatable size is | |
115 | 16/24/32 bytes. | |
f65fd747 UD |
116 | |
117 | Even a request for zero bytes (i.e., malloc(0)) returns a | |
118 | pointer to something of the minimum allocatable size. | |
119 | ||
fa8d436c UD |
120 | The maximum overhead wastage (i.e., number of extra bytes |
121 | allocated than were requested in malloc) is less than or equal | |
122 | to the minimum size, except for requests >= mmap_threshold that | |
123 | are serviced via mmap(), where the worst case wastage is 2 * | |
124 | sizeof(size_t) bytes plus the remainder from a system page (the | |
125 | minimal mmap unit); typically 4096 or 8192 bytes. | |
f65fd747 | 126 | |
a9177ff5 | 127 | Maximum allocated size: 4-byte size_t: 2^32 minus about two pages |
72f90263 | 128 | 8-byte size_t: 2^64 minus about two pages |
fa8d436c UD |
129 | |
130 | It is assumed that (possibly signed) size_t values suffice to | |
f65fd747 UD |
131 | represent chunk sizes. `Possibly signed' is due to the fact |
132 | that `size_t' may be defined on a system as either a signed or | |
fa8d436c UD |
133 | an unsigned type. The ISO C standard says that it must be |
134 | unsigned, but a few systems are known not to adhere to this. | |
135 | Additionally, even when size_t is unsigned, sbrk (which is by | |
136 | default used to obtain memory from system) accepts signed | |
137 | arguments, and may not be able to handle size_t-wide arguments | |
138 | with negative sign bit. Generally, values that would | |
139 | appear as negative after accounting for overhead and alignment | |
140 | are supported only via mmap(), which does not have this | |
141 | limitation. | |
142 | ||
143 | Requests for sizes outside the allowed range will perform an optional | |
144 | failure action and then return null. (Requests may also | |
145 | also fail because a system is out of memory.) | |
146 | ||
22a89187 | 147 | Thread-safety: thread-safe |
fa8d436c UD |
148 | |
149 | Compliance: I believe it is compliant with the 1997 Single Unix Specification | |
2b0fba75 | 150 | Also SVID/XPG, ANSI C, and probably others as well. |
f65fd747 UD |
151 | |
152 | * Synopsis of compile-time options: | |
153 | ||
154 | People have reported using previous versions of this malloc on all | |
155 | versions of Unix, sometimes by tweaking some of the defines | |
22a89187 | 156 | below. It has been tested most extensively on Solaris and Linux. |
fa8d436c UD |
157 | People also report using it in stand-alone embedded systems. |
158 | ||
159 | The implementation is in straight, hand-tuned ANSI C. It is not | |
160 | at all modular. (Sorry!) It uses a lot of macros. To be at all | |
161 | usable, this code should be compiled using an optimizing compiler | |
162 | (for example gcc -O3) that can simplify expressions and control | |
163 | paths. (FAQ: some macros import variables as arguments rather than | |
164 | declare locals because people reported that some debuggers | |
165 | otherwise get confused.) | |
166 | ||
167 | OPTION DEFAULT VALUE | |
168 | ||
169 | Compilation Environment options: | |
170 | ||
2a26ef3a | 171 | HAVE_MREMAP 0 |
fa8d436c UD |
172 | |
173 | Changing default word sizes: | |
174 | ||
175 | INTERNAL_SIZE_T size_t | |
073f560e UD |
176 | MALLOC_ALIGNMENT MAX (2 * sizeof(INTERNAL_SIZE_T), |
177 | __alignof__ (long double)) | |
fa8d436c UD |
178 | |
179 | Configuration and functionality options: | |
180 | ||
fa8d436c UD |
181 | USE_PUBLIC_MALLOC_WRAPPERS NOT defined |
182 | USE_MALLOC_LOCK NOT defined | |
183 | MALLOC_DEBUG NOT defined | |
184 | REALLOC_ZERO_BYTES_FREES 1 | |
fa8d436c UD |
185 | TRIM_FASTBINS 0 |
186 | ||
187 | Options for customizing MORECORE: | |
188 | ||
189 | MORECORE sbrk | |
190 | MORECORE_FAILURE -1 | |
a9177ff5 | 191 | MORECORE_CONTIGUOUS 1 |
fa8d436c UD |
192 | MORECORE_CANNOT_TRIM NOT defined |
193 | MORECORE_CLEARS 1 | |
a9177ff5 | 194 | MMAP_AS_MORECORE_SIZE (1024 * 1024) |
fa8d436c UD |
195 | |
196 | Tuning options that are also dynamically changeable via mallopt: | |
197 | ||
425ce2ed | 198 | DEFAULT_MXFAST 64 (for 32bit), 128 (for 64bit) |
fa8d436c UD |
199 | DEFAULT_TRIM_THRESHOLD 128 * 1024 |
200 | DEFAULT_TOP_PAD 0 | |
201 | DEFAULT_MMAP_THRESHOLD 128 * 1024 | |
202 | DEFAULT_MMAP_MAX 65536 | |
203 | ||
204 | There are several other #defined constants and macros that you | |
205 | probably don't want to touch unless you are extending or adapting malloc. */ | |
f65fd747 UD |
206 | |
207 | /* | |
22a89187 | 208 | void* is the pointer type that malloc should say it returns |
f65fd747 UD |
209 | */ |
210 | ||
22a89187 UD |
211 | #ifndef void |
212 | #define void void | |
213 | #endif /*void*/ | |
f65fd747 | 214 | |
fa8d436c UD |
215 | #include <stddef.h> /* for size_t */ |
216 | #include <stdlib.h> /* for getenv(), abort() */ | |
2a26ef3a | 217 | #include <unistd.h> /* for __libc_enable_secure */ |
f65fd747 | 218 | |
3c6904fb | 219 | #include <malloc-machine.h> |
2a26ef3a | 220 | #include <malloc-sysdep.h> |
3c6904fb | 221 | |
425ce2ed | 222 | #include <atomic.h> |
eb96ffb0 | 223 | #include <_itoa.h> |
e404fb16 | 224 | #include <bits/wordsize.h> |
425ce2ed | 225 | #include <sys/sysinfo.h> |
c56da3a3 | 226 | |
02d46fc4 UD |
227 | #include <ldsodefs.h> |
228 | ||
fa8d436c | 229 | #include <unistd.h> |
fa8d436c | 230 | #include <stdio.h> /* needed for malloc_stats */ |
8e58439c | 231 | #include <errno.h> |
f65fd747 | 232 | |
66274218 AJ |
233 | #include <shlib-compat.h> |
234 | ||
5d78bb43 UD |
235 | /* For uintptr_t. */ |
236 | #include <stdint.h> | |
f65fd747 | 237 | |
3e030bd5 UD |
238 | /* For va_arg, va_start, va_end. */ |
239 | #include <stdarg.h> | |
240 | ||
070906ff RM |
241 | /* For MIN, MAX, powerof2. */ |
242 | #include <sys/param.h> | |
243 | ||
c0f62c56 | 244 | |
fa8d436c UD |
245 | /* |
246 | Debugging: | |
247 | ||
248 | Because freed chunks may be overwritten with bookkeeping fields, this | |
249 | malloc will often die when freed memory is overwritten by user | |
250 | programs. This can be very effective (albeit in an annoying way) | |
251 | in helping track down dangling pointers. | |
252 | ||
253 | If you compile with -DMALLOC_DEBUG, a number of assertion checks are | |
254 | enabled that will catch more memory errors. You probably won't be | |
255 | able to make much sense of the actual assertion errors, but they | |
256 | should help you locate incorrectly overwritten memory. The checking | |
257 | is fairly extensive, and will slow down execution | |
258 | noticeably. Calling malloc_stats or mallinfo with MALLOC_DEBUG set | |
259 | will attempt to check every non-mmapped allocated and free chunk in | |
260 | the course of computing the summmaries. (By nature, mmapped regions | |
261 | cannot be checked very much automatically.) | |
262 | ||
263 | Setting MALLOC_DEBUG may also be helpful if you are trying to modify | |
264 | this code. The assertions in the check routines spell out in more | |
265 | detail the assumptions and invariants underlying the algorithms. | |
266 | ||
267 | Setting MALLOC_DEBUG does NOT provide an automated mechanism for | |
268 | checking that all accesses to malloced memory stay within their | |
269 | bounds. However, there are several add-ons and adaptations of this | |
270 | or other mallocs available that do this. | |
f65fd747 UD |
271 | */ |
272 | ||
439bda32 WN |
273 | #ifndef MALLOC_DEBUG |
274 | #define MALLOC_DEBUG 0 | |
275 | #endif | |
276 | ||
72f90263 UD |
277 | #ifdef NDEBUG |
278 | # define assert(expr) ((void) 0) | |
279 | #else | |
280 | # define assert(expr) \ | |
281 | ((expr) \ | |
282 | ? ((void) 0) \ | |
283 | : __malloc_assert (__STRING (expr), __FILE__, __LINE__, __func__)) | |
284 | ||
285 | extern const char *__progname; | |
286 | ||
287 | static void | |
288 | __malloc_assert (const char *assertion, const char *file, unsigned int line, | |
289 | const char *function) | |
290 | { | |
291 | (void) __fxprintf (NULL, "%s%s%s:%u: %s%sAssertion `%s' failed.\n", | |
292 | __progname, __progname[0] ? ": " : "", | |
293 | file, line, | |
294 | function ? function : "", function ? ": " : "", | |
295 | assertion); | |
296 | fflush (stderr); | |
297 | abort (); | |
298 | } | |
299 | #endif | |
f65fd747 UD |
300 | |
301 | ||
302 | /* | |
303 | INTERNAL_SIZE_T is the word-size used for internal bookkeeping | |
fa8d436c UD |
304 | of chunk sizes. |
305 | ||
306 | The default version is the same as size_t. | |
307 | ||
308 | While not strictly necessary, it is best to define this as an | |
309 | unsigned type, even if size_t is a signed type. This may avoid some | |
310 | artificial size limitations on some systems. | |
311 | ||
312 | On a 64-bit machine, you may be able to reduce malloc overhead by | |
313 | defining INTERNAL_SIZE_T to be a 32 bit `unsigned int' at the | |
314 | expense of not being able to handle more than 2^32 of malloced | |
315 | space. If this limitation is acceptable, you are encouraged to set | |
316 | this unless you are on a platform requiring 16byte alignments. In | |
317 | this case the alignment requirements turn out to negate any | |
318 | potential advantages of decreasing size_t word size. | |
319 | ||
320 | Implementors: Beware of the possible combinations of: | |
321 | - INTERNAL_SIZE_T might be signed or unsigned, might be 32 or 64 bits, | |
322 | and might be the same width as int or as long | |
323 | - size_t might have different width and signedness as INTERNAL_SIZE_T | |
324 | - int and long might be 32 or 64 bits, and might be the same width | |
325 | To deal with this, most comparisons and difference computations | |
326 | among INTERNAL_SIZE_Ts should cast them to unsigned long, being | |
327 | aware of the fact that casting an unsigned int to a wider long does | |
328 | not sign-extend. (This also makes checking for negative numbers | |
329 | awkward.) Some of these casts result in harmless compiler warnings | |
330 | on some systems. | |
f65fd747 UD |
331 | */ |
332 | ||
333 | #ifndef INTERNAL_SIZE_T | |
334 | #define INTERNAL_SIZE_T size_t | |
335 | #endif | |
336 | ||
fa8d436c UD |
337 | /* The corresponding word size */ |
338 | #define SIZE_SZ (sizeof(INTERNAL_SIZE_T)) | |
339 | ||
340 | ||
341 | /* | |
342 | MALLOC_ALIGNMENT is the minimum alignment for malloc'ed chunks. | |
343 | It must be a power of two at least 2 * SIZE_SZ, even on machines | |
344 | for which smaller alignments would suffice. It may be defined as | |
345 | larger than this though. Note however that code and data structures | |
346 | are optimized for the case of 8-byte alignment. | |
347 | */ | |
348 | ||
349 | ||
350 | #ifndef MALLOC_ALIGNMENT | |
cbc00a03 RM |
351 | # if !SHLIB_COMPAT (libc, GLIBC_2_0, GLIBC_2_16) |
352 | /* This is the correct definition when there is no past ABI to constrain it. | |
353 | ||
354 | Among configurations with a past ABI constraint, it differs from | |
355 | 2*SIZE_SZ only on powerpc32. For the time being, changing this is | |
356 | causing more compatibility problems due to malloc_get_state and | |
357 | malloc_set_state than will returning blocks not adequately aligned for | |
358 | long double objects under -mlong-double-128. */ | |
359 | ||
6c8dbf00 OB |
360 | # define MALLOC_ALIGNMENT (2 *SIZE_SZ < __alignof__ (long double) \ |
361 | ? __alignof__ (long double) : 2 *SIZE_SZ) | |
cbc00a03 | 362 | # else |
6c8dbf00 | 363 | # define MALLOC_ALIGNMENT (2 *SIZE_SZ) |
cbc00a03 | 364 | # endif |
fa8d436c UD |
365 | #endif |
366 | ||
367 | /* The corresponding bit mask value */ | |
368 | #define MALLOC_ALIGN_MASK (MALLOC_ALIGNMENT - 1) | |
369 | ||
370 | ||
371 | ||
372 | /* | |
373 | REALLOC_ZERO_BYTES_FREES should be set if a call to | |
374 | realloc with zero bytes should be the same as a call to free. | |
375 | This is required by the C standard. Otherwise, since this malloc | |
376 | returns a unique pointer for malloc(0), so does realloc(p, 0). | |
377 | */ | |
378 | ||
379 | #ifndef REALLOC_ZERO_BYTES_FREES | |
380 | #define REALLOC_ZERO_BYTES_FREES 1 | |
381 | #endif | |
382 | ||
383 | /* | |
384 | TRIM_FASTBINS controls whether free() of a very small chunk can | |
385 | immediately lead to trimming. Setting to true (1) can reduce memory | |
386 | footprint, but will almost always slow down programs that use a lot | |
387 | of small chunks. | |
388 | ||
389 | Define this only if you are willing to give up some speed to more | |
390 | aggressively reduce system-level memory footprint when releasing | |
391 | memory in programs that use many small chunks. You can get | |
392 | essentially the same effect by setting MXFAST to 0, but this can | |
393 | lead to even greater slowdowns in programs using many small chunks. | |
394 | TRIM_FASTBINS is an in-between compile-time option, that disables | |
395 | only those chunks bordering topmost memory from being placed in | |
396 | fastbins. | |
397 | */ | |
398 | ||
399 | #ifndef TRIM_FASTBINS | |
400 | #define TRIM_FASTBINS 0 | |
401 | #endif | |
402 | ||
403 | ||
3b49edc0 | 404 | /* Definition for getting more memory from the OS. */ |
fa8d436c UD |
405 | #define MORECORE (*__morecore) |
406 | #define MORECORE_FAILURE 0 | |
22a89187 UD |
407 | void * __default_morecore (ptrdiff_t); |
408 | void *(*__morecore)(ptrdiff_t) = __default_morecore; | |
f65fd747 | 409 | |
f65fd747 | 410 | |
22a89187 | 411 | #include <string.h> |
f65fd747 | 412 | |
fa8d436c UD |
413 | /* |
414 | MORECORE-related declarations. By default, rely on sbrk | |
415 | */ | |
09f5e163 | 416 | |
f65fd747 | 417 | |
fa8d436c UD |
418 | /* |
419 | MORECORE is the name of the routine to call to obtain more memory | |
420 | from the system. See below for general guidance on writing | |
421 | alternative MORECORE functions, as well as a version for WIN32 and a | |
422 | sample version for pre-OSX macos. | |
423 | */ | |
f65fd747 | 424 | |
fa8d436c UD |
425 | #ifndef MORECORE |
426 | #define MORECORE sbrk | |
427 | #endif | |
f65fd747 | 428 | |
fa8d436c UD |
429 | /* |
430 | MORECORE_FAILURE is the value returned upon failure of MORECORE | |
431 | as well as mmap. Since it cannot be an otherwise valid memory address, | |
432 | and must reflect values of standard sys calls, you probably ought not | |
433 | try to redefine it. | |
434 | */ | |
09f5e163 | 435 | |
fa8d436c UD |
436 | #ifndef MORECORE_FAILURE |
437 | #define MORECORE_FAILURE (-1) | |
438 | #endif | |
439 | ||
440 | /* | |
441 | If MORECORE_CONTIGUOUS is true, take advantage of fact that | |
442 | consecutive calls to MORECORE with positive arguments always return | |
443 | contiguous increasing addresses. This is true of unix sbrk. Even | |
444 | if not defined, when regions happen to be contiguous, malloc will | |
445 | permit allocations spanning regions obtained from different | |
446 | calls. But defining this when applicable enables some stronger | |
447 | consistency checks and space efficiencies. | |
448 | */ | |
f65fd747 | 449 | |
fa8d436c UD |
450 | #ifndef MORECORE_CONTIGUOUS |
451 | #define MORECORE_CONTIGUOUS 1 | |
f65fd747 UD |
452 | #endif |
453 | ||
fa8d436c UD |
454 | /* |
455 | Define MORECORE_CANNOT_TRIM if your version of MORECORE | |
456 | cannot release space back to the system when given negative | |
457 | arguments. This is generally necessary only if you are using | |
458 | a hand-crafted MORECORE function that cannot handle negative arguments. | |
459 | */ | |
460 | ||
461 | /* #define MORECORE_CANNOT_TRIM */ | |
f65fd747 | 462 | |
fa8d436c UD |
463 | /* MORECORE_CLEARS (default 1) |
464 | The degree to which the routine mapped to MORECORE zeroes out | |
465 | memory: never (0), only for newly allocated space (1) or always | |
466 | (2). The distinction between (1) and (2) is necessary because on | |
467 | some systems, if the application first decrements and then | |
468 | increments the break value, the contents of the reallocated space | |
469 | are unspecified. | |
6c8dbf00 | 470 | */ |
fa8d436c UD |
471 | |
472 | #ifndef MORECORE_CLEARS | |
6c8dbf00 | 473 | # define MORECORE_CLEARS 1 |
7cabd57c UD |
474 | #endif |
475 | ||
fa8d436c | 476 | |
a9177ff5 | 477 | /* |
fa8d436c | 478 | MMAP_AS_MORECORE_SIZE is the minimum mmap size argument to use if |
22a89187 UD |
479 | sbrk fails, and mmap is used as a backup. The value must be a |
480 | multiple of page size. This backup strategy generally applies only | |
481 | when systems have "holes" in address space, so sbrk cannot perform | |
482 | contiguous expansion, but there is still space available on system. | |
483 | On systems for which this is known to be useful (i.e. most linux | |
484 | kernels), this occurs only when programs allocate huge amounts of | |
485 | memory. Between this, and the fact that mmap regions tend to be | |
486 | limited, the size should be large, to avoid too many mmap calls and | |
487 | thus avoid running out of kernel resources. */ | |
fa8d436c UD |
488 | |
489 | #ifndef MMAP_AS_MORECORE_SIZE | |
490 | #define MMAP_AS_MORECORE_SIZE (1024 * 1024) | |
f65fd747 UD |
491 | #endif |
492 | ||
493 | /* | |
494 | Define HAVE_MREMAP to make realloc() use mremap() to re-allocate | |
2a26ef3a | 495 | large blocks. |
f65fd747 UD |
496 | */ |
497 | ||
498 | #ifndef HAVE_MREMAP | |
fa8d436c | 499 | #define HAVE_MREMAP 0 |
f65fd747 UD |
500 | #endif |
501 | ||
f65fd747 | 502 | |
f65fd747 | 503 | /* |
f65fd747 | 504 | This version of malloc supports the standard SVID/XPG mallinfo |
fa8d436c UD |
505 | routine that returns a struct containing usage properties and |
506 | statistics. It should work on any SVID/XPG compliant system that has | |
507 | a /usr/include/malloc.h defining struct mallinfo. (If you'd like to | |
508 | install such a thing yourself, cut out the preliminary declarations | |
509 | as described above and below and save them in a malloc.h file. But | |
510 | there's no compelling reason to bother to do this.) | |
f65fd747 UD |
511 | |
512 | The main declaration needed is the mallinfo struct that is returned | |
513 | (by-copy) by mallinfo(). The SVID/XPG malloinfo struct contains a | |
fa8d436c UD |
514 | bunch of fields that are not even meaningful in this version of |
515 | malloc. These fields are are instead filled by mallinfo() with | |
516 | other numbers that might be of interest. | |
f65fd747 UD |
517 | */ |
518 | ||
f65fd747 | 519 | |
fa8d436c | 520 | /* ---------- description of public routines ------------ */ |
f65fd747 UD |
521 | |
522 | /* | |
fa8d436c UD |
523 | malloc(size_t n) |
524 | Returns a pointer to a newly allocated chunk of at least n bytes, or null | |
525 | if no space is available. Additionally, on failure, errno is | |
526 | set to ENOMEM on ANSI C systems. | |
527 | ||
528 | If n is zero, malloc returns a minumum-sized chunk. (The minimum | |
529 | size is 16 bytes on most 32bit systems, and 24 or 32 bytes on 64bit | |
530 | systems.) On most systems, size_t is an unsigned type, so calls | |
531 | with negative arguments are interpreted as requests for huge amounts | |
532 | of space, which will often fail. The maximum supported value of n | |
533 | differs across systems, but is in all cases less than the maximum | |
534 | representable value of a size_t. | |
f65fd747 | 535 | */ |
3b49edc0 UD |
536 | void* __libc_malloc(size_t); |
537 | libc_hidden_proto (__libc_malloc) | |
f65fd747 | 538 | |
fa8d436c | 539 | /* |
22a89187 | 540 | free(void* p) |
fa8d436c UD |
541 | Releases the chunk of memory pointed to by p, that had been previously |
542 | allocated using malloc or a related routine such as realloc. | |
543 | It has no effect if p is null. It can have arbitrary (i.e., bad!) | |
544 | effects if p has already been freed. | |
545 | ||
546 | Unless disabled (using mallopt), freeing very large spaces will | |
547 | when possible, automatically trigger operations that give | |
548 | back unused memory to the system, thus reducing program footprint. | |
549 | */ | |
3b49edc0 UD |
550 | void __libc_free(void*); |
551 | libc_hidden_proto (__libc_free) | |
f65fd747 | 552 | |
fa8d436c UD |
553 | /* |
554 | calloc(size_t n_elements, size_t element_size); | |
555 | Returns a pointer to n_elements * element_size bytes, with all locations | |
556 | set to zero. | |
557 | */ | |
3b49edc0 | 558 | void* __libc_calloc(size_t, size_t); |
f65fd747 UD |
559 | |
560 | /* | |
22a89187 | 561 | realloc(void* p, size_t n) |
fa8d436c UD |
562 | Returns a pointer to a chunk of size n that contains the same data |
563 | as does chunk p up to the minimum of (n, p's size) bytes, or null | |
a9177ff5 | 564 | if no space is available. |
f65fd747 | 565 | |
fa8d436c UD |
566 | The returned pointer may or may not be the same as p. The algorithm |
567 | prefers extending p when possible, otherwise it employs the | |
568 | equivalent of a malloc-copy-free sequence. | |
f65fd747 | 569 | |
a9177ff5 | 570 | If p is null, realloc is equivalent to malloc. |
f65fd747 | 571 | |
fa8d436c UD |
572 | If space is not available, realloc returns null, errno is set (if on |
573 | ANSI) and p is NOT freed. | |
f65fd747 | 574 | |
fa8d436c UD |
575 | if n is for fewer bytes than already held by p, the newly unused |
576 | space is lopped off and freed if possible. Unless the #define | |
577 | REALLOC_ZERO_BYTES_FREES is set, realloc with a size argument of | |
578 | zero (re)allocates a minimum-sized chunk. | |
f65fd747 | 579 | |
fa8d436c UD |
580 | Large chunks that were internally obtained via mmap will always |
581 | be reallocated using malloc-copy-free sequences unless | |
582 | the system supports MREMAP (currently only linux). | |
f65fd747 | 583 | |
fa8d436c UD |
584 | The old unix realloc convention of allowing the last-free'd chunk |
585 | to be used as an argument to realloc is not supported. | |
f65fd747 | 586 | */ |
3b49edc0 UD |
587 | void* __libc_realloc(void*, size_t); |
588 | libc_hidden_proto (__libc_realloc) | |
f65fd747 | 589 | |
fa8d436c UD |
590 | /* |
591 | memalign(size_t alignment, size_t n); | |
592 | Returns a pointer to a newly allocated chunk of n bytes, aligned | |
593 | in accord with the alignment argument. | |
594 | ||
595 | The alignment argument should be a power of two. If the argument is | |
596 | not a power of two, the nearest greater power is used. | |
597 | 8-byte alignment is guaranteed by normal malloc calls, so don't | |
598 | bother calling memalign with an argument of 8 or less. | |
599 | ||
600 | Overreliance on memalign is a sure way to fragment space. | |
601 | */ | |
3b49edc0 UD |
602 | void* __libc_memalign(size_t, size_t); |
603 | libc_hidden_proto (__libc_memalign) | |
f65fd747 UD |
604 | |
605 | /* | |
fa8d436c UD |
606 | valloc(size_t n); |
607 | Equivalent to memalign(pagesize, n), where pagesize is the page | |
608 | size of the system. If the pagesize is unknown, 4096 is used. | |
609 | */ | |
3b49edc0 | 610 | void* __libc_valloc(size_t); |
fa8d436c | 611 | |
f65fd747 | 612 | |
f65fd747 | 613 | |
fa8d436c UD |
614 | /* |
615 | mallopt(int parameter_number, int parameter_value) | |
616 | Sets tunable parameters The format is to provide a | |
617 | (parameter-number, parameter-value) pair. mallopt then sets the | |
618 | corresponding parameter to the argument value if it can (i.e., so | |
619 | long as the value is meaningful), and returns 1 if successful else | |
620 | 0. SVID/XPG/ANSI defines four standard param numbers for mallopt, | |
621 | normally defined in malloc.h. Only one of these (M_MXFAST) is used | |
622 | in this malloc. The others (M_NLBLKS, M_GRAIN, M_KEEP) don't apply, | |
623 | so setting them has no effect. But this malloc also supports four | |
624 | other options in mallopt. See below for details. Briefly, supported | |
625 | parameters are as follows (listed defaults are for "typical" | |
626 | configurations). | |
627 | ||
628 | Symbol param # default allowed param values | |
629 | M_MXFAST 1 64 0-80 (0 disables fastbins) | |
630 | M_TRIM_THRESHOLD -1 128*1024 any (-1U disables trimming) | |
a9177ff5 | 631 | M_TOP_PAD -2 0 any |
fa8d436c UD |
632 | M_MMAP_THRESHOLD -3 128*1024 any (or 0 if no MMAP support) |
633 | M_MMAP_MAX -4 65536 any (0 disables use of mmap) | |
634 | */ | |
3b49edc0 UD |
635 | int __libc_mallopt(int, int); |
636 | libc_hidden_proto (__libc_mallopt) | |
fa8d436c UD |
637 | |
638 | ||
639 | /* | |
640 | mallinfo() | |
641 | Returns (by copy) a struct containing various summary statistics: | |
642 | ||
a9177ff5 RM |
643 | arena: current total non-mmapped bytes allocated from system |
644 | ordblks: the number of free chunks | |
fa8d436c | 645 | smblks: the number of fastbin blocks (i.e., small chunks that |
72f90263 | 646 | have been freed but not use resused or consolidated) |
a9177ff5 RM |
647 | hblks: current number of mmapped regions |
648 | hblkhd: total bytes held in mmapped regions | |
fa8d436c | 649 | usmblks: the maximum total allocated space. This will be greater |
72f90263 | 650 | than current total if trimming has occurred. |
a9177ff5 | 651 | fsmblks: total bytes held in fastbin blocks |
fa8d436c | 652 | uordblks: current total allocated space (normal or mmapped) |
a9177ff5 | 653 | fordblks: total free space |
fa8d436c | 654 | keepcost: the maximum number of bytes that could ideally be released |
72f90263 UD |
655 | back to system via malloc_trim. ("ideally" means that |
656 | it ignores page restrictions etc.) | |
fa8d436c UD |
657 | |
658 | Because these fields are ints, but internal bookkeeping may | |
a9177ff5 | 659 | be kept as longs, the reported values may wrap around zero and |
fa8d436c UD |
660 | thus be inaccurate. |
661 | */ | |
3b49edc0 | 662 | struct mallinfo __libc_mallinfo(void); |
88764ae2 | 663 | |
f65fd747 | 664 | |
fa8d436c UD |
665 | /* |
666 | pvalloc(size_t n); | |
667 | Equivalent to valloc(minimum-page-that-holds(n)), that is, | |
668 | round up n to nearest pagesize. | |
669 | */ | |
3b49edc0 | 670 | void* __libc_pvalloc(size_t); |
fa8d436c UD |
671 | |
672 | /* | |
673 | malloc_trim(size_t pad); | |
674 | ||
675 | If possible, gives memory back to the system (via negative | |
676 | arguments to sbrk) if there is unused memory at the `high' end of | |
677 | the malloc pool. You can call this after freeing large blocks of | |
678 | memory to potentially reduce the system-level memory requirements | |
679 | of a program. However, it cannot guarantee to reduce memory. Under | |
680 | some allocation patterns, some large free blocks of memory will be | |
681 | locked between two used chunks, so they cannot be given back to | |
682 | the system. | |
a9177ff5 | 683 | |
fa8d436c UD |
684 | The `pad' argument to malloc_trim represents the amount of free |
685 | trailing space to leave untrimmed. If this argument is zero, | |
686 | only the minimum amount of memory to maintain internal data | |
687 | structures will be left (one page or less). Non-zero arguments | |
688 | can be supplied to maintain enough trailing space to service | |
689 | future expected allocations without having to re-obtain memory | |
690 | from the system. | |
a9177ff5 | 691 | |
fa8d436c UD |
692 | Malloc_trim returns 1 if it actually released any memory, else 0. |
693 | On systems that do not support "negative sbrks", it will always | |
c958a6a4 | 694 | return 0. |
fa8d436c | 695 | */ |
3b49edc0 | 696 | int __malloc_trim(size_t); |
fa8d436c UD |
697 | |
698 | /* | |
22a89187 | 699 | malloc_usable_size(void* p); |
fa8d436c UD |
700 | |
701 | Returns the number of bytes you can actually use in | |
702 | an allocated chunk, which may be more than you requested (although | |
703 | often not) due to alignment and minimum size constraints. | |
704 | You can use this many bytes without worrying about | |
705 | overwriting other allocated objects. This is not a particularly great | |
706 | programming practice. malloc_usable_size can be more useful in | |
707 | debugging and assertions, for example: | |
708 | ||
709 | p = malloc(n); | |
710 | assert(malloc_usable_size(p) >= 256); | |
711 | ||
712 | */ | |
3b49edc0 | 713 | size_t __malloc_usable_size(void*); |
fa8d436c UD |
714 | |
715 | /* | |
716 | malloc_stats(); | |
717 | Prints on stderr the amount of space obtained from the system (both | |
718 | via sbrk and mmap), the maximum amount (which may be more than | |
719 | current if malloc_trim and/or munmap got called), and the current | |
720 | number of bytes allocated via malloc (or realloc, etc) but not yet | |
721 | freed. Note that this is the number of bytes allocated, not the | |
722 | number requested. It will be larger than the number requested | |
723 | because of alignment and bookkeeping overhead. Because it includes | |
724 | alignment wastage as being in use, this figure may be greater than | |
725 | zero even when no user-level chunks are allocated. | |
726 | ||
727 | The reported current and maximum system memory can be inaccurate if | |
728 | a program makes other calls to system memory allocation functions | |
729 | (normally sbrk) outside of malloc. | |
730 | ||
731 | malloc_stats prints only the most commonly interesting statistics. | |
732 | More information can be obtained by calling mallinfo. | |
733 | ||
734 | */ | |
3b49edc0 | 735 | void __malloc_stats(void); |
f65fd747 | 736 | |
f7ddf3d3 UD |
737 | /* |
738 | malloc_get_state(void); | |
739 | ||
740 | Returns the state of all malloc variables in an opaque data | |
741 | structure. | |
742 | */ | |
3b49edc0 | 743 | void* __malloc_get_state(void); |
f7ddf3d3 UD |
744 | |
745 | /* | |
22a89187 | 746 | malloc_set_state(void* state); |
f7ddf3d3 UD |
747 | |
748 | Restore the state of all malloc variables from data obtained with | |
749 | malloc_get_state(). | |
750 | */ | |
3b49edc0 | 751 | int __malloc_set_state(void*); |
f7ddf3d3 | 752 | |
f7ddf3d3 UD |
753 | /* |
754 | posix_memalign(void **memptr, size_t alignment, size_t size); | |
755 | ||
756 | POSIX wrapper like memalign(), checking for validity of size. | |
757 | */ | |
758 | int __posix_memalign(void **, size_t, size_t); | |
f7ddf3d3 | 759 | |
fa8d436c UD |
760 | /* mallopt tuning options */ |
761 | ||
f65fd747 | 762 | /* |
fa8d436c UD |
763 | M_MXFAST is the maximum request size used for "fastbins", special bins |
764 | that hold returned chunks without consolidating their spaces. This | |
765 | enables future requests for chunks of the same size to be handled | |
766 | very quickly, but can increase fragmentation, and thus increase the | |
767 | overall memory footprint of a program. | |
768 | ||
769 | This malloc manages fastbins very conservatively yet still | |
770 | efficiently, so fragmentation is rarely a problem for values less | |
771 | than or equal to the default. The maximum supported value of MXFAST | |
772 | is 80. You wouldn't want it any higher than this anyway. Fastbins | |
773 | are designed especially for use with many small structs, objects or | |
774 | strings -- the default handles structs/objects/arrays with sizes up | |
775 | to 8 4byte fields, or small strings representing words, tokens, | |
776 | etc. Using fastbins for larger objects normally worsens | |
777 | fragmentation without improving speed. | |
778 | ||
779 | M_MXFAST is set in REQUEST size units. It is internally used in | |
780 | chunksize units, which adds padding and alignment. You can reduce | |
781 | M_MXFAST to 0 to disable all use of fastbins. This causes the malloc | |
782 | algorithm to be a closer approximation of fifo-best-fit in all cases, | |
783 | not just for larger requests, but will generally cause it to be | |
784 | slower. | |
f65fd747 UD |
785 | */ |
786 | ||
787 | ||
fa8d436c UD |
788 | /* M_MXFAST is a standard SVID/XPG tuning option, usually listed in malloc.h */ |
789 | #ifndef M_MXFAST | |
a9177ff5 | 790 | #define M_MXFAST 1 |
fa8d436c | 791 | #endif |
f65fd747 | 792 | |
fa8d436c | 793 | #ifndef DEFAULT_MXFAST |
425ce2ed | 794 | #define DEFAULT_MXFAST (64 * SIZE_SZ / 4) |
10dc2a90 UD |
795 | #endif |
796 | ||
10dc2a90 | 797 | |
fa8d436c UD |
798 | /* |
799 | M_TRIM_THRESHOLD is the maximum amount of unused top-most memory | |
800 | to keep before releasing via malloc_trim in free(). | |
801 | ||
802 | Automatic trimming is mainly useful in long-lived programs. | |
803 | Because trimming via sbrk can be slow on some systems, and can | |
804 | sometimes be wasteful (in cases where programs immediately | |
805 | afterward allocate more large chunks) the value should be high | |
806 | enough so that your overall system performance would improve by | |
807 | releasing this much memory. | |
808 | ||
809 | The trim threshold and the mmap control parameters (see below) | |
810 | can be traded off with one another. Trimming and mmapping are | |
811 | two different ways of releasing unused memory back to the | |
812 | system. Between these two, it is often possible to keep | |
813 | system-level demands of a long-lived program down to a bare | |
814 | minimum. For example, in one test suite of sessions measuring | |
815 | the XF86 X server on Linux, using a trim threshold of 128K and a | |
816 | mmap threshold of 192K led to near-minimal long term resource | |
817 | consumption. | |
818 | ||
819 | If you are using this malloc in a long-lived program, it should | |
820 | pay to experiment with these values. As a rough guide, you | |
821 | might set to a value close to the average size of a process | |
822 | (program) running on your system. Releasing this much memory | |
823 | would allow such a process to run in memory. Generally, it's | |
824 | worth it to tune for trimming rather tham memory mapping when a | |
825 | program undergoes phases where several large chunks are | |
826 | allocated and released in ways that can reuse each other's | |
827 | storage, perhaps mixed with phases where there are no such | |
828 | chunks at all. And in well-behaved long-lived programs, | |
829 | controlling release of large blocks via trimming versus mapping | |
830 | is usually faster. | |
831 | ||
832 | However, in most programs, these parameters serve mainly as | |
833 | protection against the system-level effects of carrying around | |
834 | massive amounts of unneeded memory. Since frequent calls to | |
835 | sbrk, mmap, and munmap otherwise degrade performance, the default | |
836 | parameters are set to relatively high values that serve only as | |
837 | safeguards. | |
838 | ||
839 | The trim value It must be greater than page size to have any useful | |
a9177ff5 | 840 | effect. To disable trimming completely, you can set to |
fa8d436c UD |
841 | (unsigned long)(-1) |
842 | ||
843 | Trim settings interact with fastbin (MXFAST) settings: Unless | |
844 | TRIM_FASTBINS is defined, automatic trimming never takes place upon | |
845 | freeing a chunk with size less than or equal to MXFAST. Trimming is | |
846 | instead delayed until subsequent freeing of larger chunks. However, | |
847 | you can still force an attempted trim by calling malloc_trim. | |
848 | ||
849 | Also, trimming is not generally possible in cases where | |
850 | the main arena is obtained via mmap. | |
851 | ||
852 | Note that the trick some people use of mallocing a huge space and | |
853 | then freeing it at program startup, in an attempt to reserve system | |
854 | memory, doesn't have the intended effect under automatic trimming, | |
855 | since that memory will immediately be returned to the system. | |
856 | */ | |
857 | ||
858 | #define M_TRIM_THRESHOLD -1 | |
859 | ||
860 | #ifndef DEFAULT_TRIM_THRESHOLD | |
861 | #define DEFAULT_TRIM_THRESHOLD (128 * 1024) | |
862 | #endif | |
863 | ||
864 | /* | |
865 | M_TOP_PAD is the amount of extra `padding' space to allocate or | |
866 | retain whenever sbrk is called. It is used in two ways internally: | |
867 | ||
868 | * When sbrk is called to extend the top of the arena to satisfy | |
869 | a new malloc request, this much padding is added to the sbrk | |
870 | request. | |
871 | ||
872 | * When malloc_trim is called automatically from free(), | |
873 | it is used as the `pad' argument. | |
874 | ||
875 | In both cases, the actual amount of padding is rounded | |
876 | so that the end of the arena is always a system page boundary. | |
877 | ||
878 | The main reason for using padding is to avoid calling sbrk so | |
879 | often. Having even a small pad greatly reduces the likelihood | |
880 | that nearly every malloc request during program start-up (or | |
881 | after trimming) will invoke sbrk, which needlessly wastes | |
882 | time. | |
883 | ||
884 | Automatic rounding-up to page-size units is normally sufficient | |
885 | to avoid measurable overhead, so the default is 0. However, in | |
886 | systems where sbrk is relatively slow, it can pay to increase | |
887 | this value, at the expense of carrying around more memory than | |
888 | the program needs. | |
889 | */ | |
10dc2a90 | 890 | |
fa8d436c | 891 | #define M_TOP_PAD -2 |
10dc2a90 | 892 | |
fa8d436c UD |
893 | #ifndef DEFAULT_TOP_PAD |
894 | #define DEFAULT_TOP_PAD (0) | |
895 | #endif | |
f65fd747 | 896 | |
1d05c2fb UD |
897 | /* |
898 | MMAP_THRESHOLD_MAX and _MIN are the bounds on the dynamically | |
899 | adjusted MMAP_THRESHOLD. | |
900 | */ | |
901 | ||
902 | #ifndef DEFAULT_MMAP_THRESHOLD_MIN | |
903 | #define DEFAULT_MMAP_THRESHOLD_MIN (128 * 1024) | |
904 | #endif | |
905 | ||
906 | #ifndef DEFAULT_MMAP_THRESHOLD_MAX | |
e404fb16 UD |
907 | /* For 32-bit platforms we cannot increase the maximum mmap |
908 | threshold much because it is also the minimum value for the | |
bd2c2341 UD |
909 | maximum heap size and its alignment. Going above 512k (i.e., 1M |
910 | for new heaps) wastes too much address space. */ | |
e404fb16 | 911 | # if __WORDSIZE == 32 |
bd2c2341 | 912 | # define DEFAULT_MMAP_THRESHOLD_MAX (512 * 1024) |
e404fb16 | 913 | # else |
bd2c2341 | 914 | # define DEFAULT_MMAP_THRESHOLD_MAX (4 * 1024 * 1024 * sizeof(long)) |
e404fb16 | 915 | # endif |
1d05c2fb UD |
916 | #endif |
917 | ||
fa8d436c UD |
918 | /* |
919 | M_MMAP_THRESHOLD is the request size threshold for using mmap() | |
920 | to service a request. Requests of at least this size that cannot | |
921 | be allocated using already-existing space will be serviced via mmap. | |
922 | (If enough normal freed space already exists it is used instead.) | |
923 | ||
924 | Using mmap segregates relatively large chunks of memory so that | |
925 | they can be individually obtained and released from the host | |
926 | system. A request serviced through mmap is never reused by any | |
927 | other request (at least not directly; the system may just so | |
928 | happen to remap successive requests to the same locations). | |
929 | ||
930 | Segregating space in this way has the benefits that: | |
931 | ||
a9177ff5 RM |
932 | 1. Mmapped space can ALWAYS be individually released back |
933 | to the system, which helps keep the system level memory | |
934 | demands of a long-lived program low. | |
fa8d436c UD |
935 | 2. Mapped memory can never become `locked' between |
936 | other chunks, as can happen with normally allocated chunks, which | |
937 | means that even trimming via malloc_trim would not release them. | |
938 | 3. On some systems with "holes" in address spaces, mmap can obtain | |
939 | memory that sbrk cannot. | |
940 | ||
941 | However, it has the disadvantages that: | |
942 | ||
943 | 1. The space cannot be reclaimed, consolidated, and then | |
944 | used to service later requests, as happens with normal chunks. | |
945 | 2. It can lead to more wastage because of mmap page alignment | |
946 | requirements | |
947 | 3. It causes malloc performance to be more dependent on host | |
948 | system memory management support routines which may vary in | |
949 | implementation quality and may impose arbitrary | |
950 | limitations. Generally, servicing a request via normal | |
951 | malloc steps is faster than going through a system's mmap. | |
952 | ||
953 | The advantages of mmap nearly always outweigh disadvantages for | |
954 | "large" chunks, but the value of "large" varies across systems. The | |
955 | default is an empirically derived value that works well in most | |
956 | systems. | |
1d05c2fb UD |
957 | |
958 | ||
959 | Update in 2006: | |
960 | The above was written in 2001. Since then the world has changed a lot. | |
961 | Memory got bigger. Applications got bigger. The virtual address space | |
962 | layout in 32 bit linux changed. | |
963 | ||
964 | In the new situation, brk() and mmap space is shared and there are no | |
965 | artificial limits on brk size imposed by the kernel. What is more, | |
966 | applications have started using transient allocations larger than the | |
967 | 128Kb as was imagined in 2001. | |
968 | ||
969 | The price for mmap is also high now; each time glibc mmaps from the | |
970 | kernel, the kernel is forced to zero out the memory it gives to the | |
971 | application. Zeroing memory is expensive and eats a lot of cache and | |
972 | memory bandwidth. This has nothing to do with the efficiency of the | |
973 | virtual memory system, by doing mmap the kernel just has no choice but | |
974 | to zero. | |
975 | ||
976 | In 2001, the kernel had a maximum size for brk() which was about 800 | |
977 | megabytes on 32 bit x86, at that point brk() would hit the first | |
978 | mmaped shared libaries and couldn't expand anymore. With current 2.6 | |
979 | kernels, the VA space layout is different and brk() and mmap | |
980 | both can span the entire heap at will. | |
981 | ||
982 | Rather than using a static threshold for the brk/mmap tradeoff, | |
983 | we are now using a simple dynamic one. The goal is still to avoid | |
984 | fragmentation. The old goals we kept are | |
985 | 1) try to get the long lived large allocations to use mmap() | |
986 | 2) really large allocations should always use mmap() | |
987 | and we're adding now: | |
988 | 3) transient allocations should use brk() to avoid forcing the kernel | |
989 | having to zero memory over and over again | |
990 | ||
991 | The implementation works with a sliding threshold, which is by default | |
992 | limited to go between 128Kb and 32Mb (64Mb for 64 bitmachines) and starts | |
993 | out at 128Kb as per the 2001 default. | |
994 | ||
995 | This allows us to satisfy requirement 1) under the assumption that long | |
996 | lived allocations are made early in the process' lifespan, before it has | |
997 | started doing dynamic allocations of the same size (which will | |
998 | increase the threshold). | |
999 | ||
1000 | The upperbound on the threshold satisfies requirement 2) | |
1001 | ||
1002 | The threshold goes up in value when the application frees memory that was | |
1003 | allocated with the mmap allocator. The idea is that once the application | |
1004 | starts freeing memory of a certain size, it's highly probable that this is | |
1005 | a size the application uses for transient allocations. This estimator | |
1006 | is there to satisfy the new third requirement. | |
1007 | ||
f65fd747 UD |
1008 | */ |
1009 | ||
fa8d436c | 1010 | #define M_MMAP_THRESHOLD -3 |
f65fd747 | 1011 | |
fa8d436c | 1012 | #ifndef DEFAULT_MMAP_THRESHOLD |
1d05c2fb | 1013 | #define DEFAULT_MMAP_THRESHOLD DEFAULT_MMAP_THRESHOLD_MIN |
fa8d436c UD |
1014 | #endif |
1015 | ||
1016 | /* | |
1017 | M_MMAP_MAX is the maximum number of requests to simultaneously | |
1018 | service using mmap. This parameter exists because | |
1019 | some systems have a limited number of internal tables for | |
1020 | use by mmap, and using more than a few of them may degrade | |
1021 | performance. | |
1022 | ||
1023 | The default is set to a value that serves only as a safeguard. | |
22a89187 | 1024 | Setting to 0 disables use of mmap for servicing large requests. |
fa8d436c | 1025 | */ |
f65fd747 | 1026 | |
fa8d436c UD |
1027 | #define M_MMAP_MAX -4 |
1028 | ||
1029 | #ifndef DEFAULT_MMAP_MAX | |
fa8d436c | 1030 | #define DEFAULT_MMAP_MAX (65536) |
f65fd747 UD |
1031 | #endif |
1032 | ||
100351c3 | 1033 | #include <malloc.h> |
f65fd747 | 1034 | |
fa8d436c UD |
1035 | #ifndef RETURN_ADDRESS |
1036 | #define RETURN_ADDRESS(X_) (NULL) | |
9ae6fc54 | 1037 | #endif |
431c33c0 UD |
1038 | |
1039 | /* On some platforms we can compile internal, not exported functions better. | |
1040 | Let the environment provide a macro and define it to be empty if it | |
1041 | is not available. */ | |
1042 | #ifndef internal_function | |
1043 | # define internal_function | |
1044 | #endif | |
1045 | ||
fa8d436c UD |
1046 | /* Forward declarations. */ |
1047 | struct malloc_chunk; | |
1048 | typedef struct malloc_chunk* mchunkptr; | |
431c33c0 | 1049 | |
fa8d436c | 1050 | /* Internal routines. */ |
f65fd747 | 1051 | |
22a89187 | 1052 | static void* _int_malloc(mstate, size_t); |
425ce2ed | 1053 | static void _int_free(mstate, mchunkptr, int); |
22a89187 | 1054 | static void* _int_realloc(mstate, mchunkptr, INTERNAL_SIZE_T, |
6e4b2107 | 1055 | INTERNAL_SIZE_T); |
22a89187 | 1056 | static void* _int_memalign(mstate, size_t, size_t); |
10ad46bc OB |
1057 | static void* _mid_memalign(size_t, size_t, void *); |
1058 | ||
6bf4302e | 1059 | static void malloc_printerr(int action, const char *str, void *ptr); |
fa8d436c | 1060 | |
22a89187 | 1061 | static void* internal_function mem2mem_check(void *p, size_t sz); |
fa8d436c UD |
1062 | static int internal_function top_check(void); |
1063 | static void internal_function munmap_chunk(mchunkptr p); | |
a9177ff5 | 1064 | #if HAVE_MREMAP |
fa8d436c | 1065 | static mchunkptr internal_function mremap_chunk(mchunkptr p, size_t new_size); |
a9177ff5 | 1066 | #endif |
fa8d436c | 1067 | |
22a89187 UD |
1068 | static void* malloc_check(size_t sz, const void *caller); |
1069 | static void free_check(void* mem, const void *caller); | |
1070 | static void* realloc_check(void* oldmem, size_t bytes, | |
1071 | const void *caller); | |
1072 | static void* memalign_check(size_t alignment, size_t bytes, | |
1073 | const void *caller); | |
750c1f2a | 1074 | #ifndef NO_THREADS |
22a89187 UD |
1075 | static void* malloc_atfork(size_t sz, const void *caller); |
1076 | static void free_atfork(void* mem, const void *caller); | |
750c1f2a | 1077 | #endif |
f65fd747 | 1078 | |
fa8d436c | 1079 | /* ------------------ MMAP support ------------------ */ |
f65fd747 | 1080 | |
f65fd747 | 1081 | |
fa8d436c | 1082 | #include <fcntl.h> |
fa8d436c | 1083 | #include <sys/mman.h> |
f65fd747 | 1084 | |
fa8d436c UD |
1085 | #if !defined(MAP_ANONYMOUS) && defined(MAP_ANON) |
1086 | # define MAP_ANONYMOUS MAP_ANON | |
1087 | #endif | |
f65fd747 | 1088 | |
fa8d436c | 1089 | #ifndef MAP_NORESERVE |
3b49edc0 | 1090 | # define MAP_NORESERVE 0 |
f65fd747 UD |
1091 | #endif |
1092 | ||
fa8d436c | 1093 | #define MMAP(addr, size, prot, flags) \ |
3b49edc0 | 1094 | __mmap((addr), (size), (prot), (flags)|MAP_ANONYMOUS|MAP_PRIVATE, -1, 0) |
f65fd747 | 1095 | |
f65fd747 UD |
1096 | |
1097 | /* | |
fa8d436c | 1098 | ----------------------- Chunk representations ----------------------- |
f65fd747 UD |
1099 | */ |
1100 | ||
1101 | ||
fa8d436c UD |
1102 | /* |
1103 | This struct declaration is misleading (but accurate and necessary). | |
1104 | It declares a "view" into memory allowing access to necessary | |
1105 | fields at known offsets from a given base. See explanation below. | |
1106 | */ | |
1107 | ||
1108 | struct malloc_chunk { | |
1109 | ||
1110 | INTERNAL_SIZE_T prev_size; /* Size of previous chunk (if free). */ | |
1111 | INTERNAL_SIZE_T size; /* Size in bytes, including overhead. */ | |
1112 | ||
1113 | struct malloc_chunk* fd; /* double links -- used only if free. */ | |
f65fd747 | 1114 | struct malloc_chunk* bk; |
7ecfbd38 UD |
1115 | |
1116 | /* Only used for large blocks: pointer to next larger size. */ | |
1117 | struct malloc_chunk* fd_nextsize; /* double links -- used only if free. */ | |
1118 | struct malloc_chunk* bk_nextsize; | |
f65fd747 UD |
1119 | }; |
1120 | ||
f65fd747 UD |
1121 | |
1122 | /* | |
f65fd747 UD |
1123 | malloc_chunk details: |
1124 | ||
1125 | (The following includes lightly edited explanations by Colin Plumb.) | |
1126 | ||
1127 | Chunks of memory are maintained using a `boundary tag' method as | |
1128 | described in e.g., Knuth or Standish. (See the paper by Paul | |
1129 | Wilson ftp://ftp.cs.utexas.edu/pub/garbage/allocsrv.ps for a | |
1130 | survey of such techniques.) Sizes of free chunks are stored both | |
1131 | in the front of each chunk and at the end. This makes | |
1132 | consolidating fragmented chunks into bigger chunks very fast. The | |
1133 | size fields also hold bits representing whether chunks are free or | |
1134 | in use. | |
1135 | ||
1136 | An allocated chunk looks like this: | |
1137 | ||
1138 | ||
1139 | chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
72f90263 UD |
1140 | | Size of previous chunk, if allocated | | |
1141 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
1142 | | Size of chunk, in bytes |M|P| | |
f65fd747 | 1143 | mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
72f90263 UD |
1144 | | User data starts here... . |
1145 | . . | |
1146 | . (malloc_usable_size() bytes) . | |
1147 | . | | |
f65fd747 | 1148 | nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
72f90263 UD |
1149 | | Size of chunk | |
1150 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
f65fd747 UD |
1151 | |
1152 | ||
1153 | Where "chunk" is the front of the chunk for the purpose of most of | |
1154 | the malloc code, but "mem" is the pointer that is returned to the | |
1155 | user. "Nextchunk" is the beginning of the next contiguous chunk. | |
1156 | ||
6f65e668 | 1157 | Chunks always begin on even word boundaries, so the mem portion |
f65fd747 | 1158 | (which is returned to the user) is also on an even word boundary, and |
fa8d436c | 1159 | thus at least double-word aligned. |
f65fd747 UD |
1160 | |
1161 | Free chunks are stored in circular doubly-linked lists, and look like this: | |
1162 | ||
1163 | chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
72f90263 UD |
1164 | | Size of previous chunk | |
1165 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
f65fd747 UD |
1166 | `head:' | Size of chunk, in bytes |P| |
1167 | mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
72f90263 UD |
1168 | | Forward pointer to next chunk in list | |
1169 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
1170 | | Back pointer to previous chunk in list | | |
1171 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
1172 | | Unused space (may be 0 bytes long) . | |
1173 | . . | |
1174 | . | | |
f65fd747 UD |
1175 | nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
1176 | `foot:' | Size of chunk, in bytes | | |
72f90263 | 1177 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
f65fd747 UD |
1178 | |
1179 | The P (PREV_INUSE) bit, stored in the unused low-order bit of the | |
1180 | chunk size (which is always a multiple of two words), is an in-use | |
1181 | bit for the *previous* chunk. If that bit is *clear*, then the | |
1182 | word before the current chunk size contains the previous chunk | |
1183 | size, and can be used to find the front of the previous chunk. | |
fa8d436c UD |
1184 | The very first chunk allocated always has this bit set, |
1185 | preventing access to non-existent (or non-owned) memory. If | |
1186 | prev_inuse is set for any given chunk, then you CANNOT determine | |
1187 | the size of the previous chunk, and might even get a memory | |
1188 | addressing fault when trying to do so. | |
f65fd747 UD |
1189 | |
1190 | Note that the `foot' of the current chunk is actually represented | |
fa8d436c UD |
1191 | as the prev_size of the NEXT chunk. This makes it easier to |
1192 | deal with alignments etc but can be very confusing when trying | |
1193 | to extend or adapt this code. | |
f65fd747 UD |
1194 | |
1195 | The two exceptions to all this are | |
1196 | ||
fa8d436c | 1197 | 1. The special chunk `top' doesn't bother using the |
72f90263 UD |
1198 | trailing size field since there is no next contiguous chunk |
1199 | that would have to index off it. After initialization, `top' | |
1200 | is forced to always exist. If it would become less than | |
1201 | MINSIZE bytes long, it is replenished. | |
f65fd747 UD |
1202 | |
1203 | 2. Chunks allocated via mmap, which have the second-lowest-order | |
72f90263 UD |
1204 | bit M (IS_MMAPPED) set in their size fields. Because they are |
1205 | allocated one-by-one, each must contain its own trailing size field. | |
f65fd747 UD |
1206 | |
1207 | */ | |
1208 | ||
1209 | /* | |
fa8d436c UD |
1210 | ---------- Size and alignment checks and conversions ---------- |
1211 | */ | |
f65fd747 | 1212 | |
fa8d436c | 1213 | /* conversion from malloc headers to user pointers, and back */ |
f65fd747 | 1214 | |
22a89187 | 1215 | #define chunk2mem(p) ((void*)((char*)(p) + 2*SIZE_SZ)) |
fa8d436c | 1216 | #define mem2chunk(mem) ((mchunkptr)((char*)(mem) - 2*SIZE_SZ)) |
f65fd747 | 1217 | |
fa8d436c | 1218 | /* The smallest possible chunk */ |
7ecfbd38 | 1219 | #define MIN_CHUNK_SIZE (offsetof(struct malloc_chunk, fd_nextsize)) |
f65fd747 | 1220 | |
fa8d436c | 1221 | /* The smallest size we can malloc is an aligned minimal chunk */ |
f65fd747 | 1222 | |
fa8d436c UD |
1223 | #define MINSIZE \ |
1224 | (unsigned long)(((MIN_CHUNK_SIZE+MALLOC_ALIGN_MASK) & ~MALLOC_ALIGN_MASK)) | |
f65fd747 | 1225 | |
fa8d436c | 1226 | /* Check if m has acceptable alignment */ |
f65fd747 | 1227 | |
073f560e UD |
1228 | #define aligned_OK(m) (((unsigned long)(m) & MALLOC_ALIGN_MASK) == 0) |
1229 | ||
1230 | #define misaligned_chunk(p) \ | |
1231 | ((uintptr_t)(MALLOC_ALIGNMENT == 2 * SIZE_SZ ? (p) : chunk2mem (p)) \ | |
1232 | & MALLOC_ALIGN_MASK) | |
f65fd747 | 1233 | |
f65fd747 | 1234 | |
a9177ff5 | 1235 | /* |
fa8d436c UD |
1236 | Check if a request is so large that it would wrap around zero when |
1237 | padded and aligned. To simplify some other code, the bound is made | |
1238 | low enough so that adding MINSIZE will also not wrap around zero. | |
6c8dbf00 | 1239 | */ |
f65fd747 | 1240 | |
fa8d436c | 1241 | #define REQUEST_OUT_OF_RANGE(req) \ |
6c8dbf00 OB |
1242 | ((unsigned long) (req) >= \ |
1243 | (unsigned long) (INTERNAL_SIZE_T) (-2 * MINSIZE)) | |
f65fd747 | 1244 | |
fa8d436c | 1245 | /* pad request bytes into a usable size -- internal version */ |
f65fd747 | 1246 | |
fa8d436c UD |
1247 | #define request2size(req) \ |
1248 | (((req) + SIZE_SZ + MALLOC_ALIGN_MASK < MINSIZE) ? \ | |
1249 | MINSIZE : \ | |
1250 | ((req) + SIZE_SZ + MALLOC_ALIGN_MASK) & ~MALLOC_ALIGN_MASK) | |
f65fd747 | 1251 | |
fa8d436c | 1252 | /* Same, except also perform argument check */ |
f65fd747 | 1253 | |
fa8d436c | 1254 | #define checked_request2size(req, sz) \ |
6c8dbf00 OB |
1255 | if (REQUEST_OUT_OF_RANGE (req)) { \ |
1256 | __set_errno (ENOMEM); \ | |
1257 | return 0; \ | |
1258 | } \ | |
1259 | (sz) = request2size (req); | |
f65fd747 UD |
1260 | |
1261 | /* | |
6c8dbf00 OB |
1262 | --------------- Physical chunk operations --------------- |
1263 | */ | |
f65fd747 | 1264 | |
10dc2a90 | 1265 | |
fa8d436c UD |
1266 | /* size field is or'ed with PREV_INUSE when previous adjacent chunk in use */ |
1267 | #define PREV_INUSE 0x1 | |
f65fd747 | 1268 | |
fa8d436c UD |
1269 | /* extract inuse bit of previous chunk */ |
1270 | #define prev_inuse(p) ((p)->size & PREV_INUSE) | |
f65fd747 | 1271 | |
f65fd747 | 1272 | |
fa8d436c UD |
1273 | /* size field is or'ed with IS_MMAPPED if the chunk was obtained with mmap() */ |
1274 | #define IS_MMAPPED 0x2 | |
f65fd747 | 1275 | |
fa8d436c UD |
1276 | /* check for mmap()'ed chunk */ |
1277 | #define chunk_is_mmapped(p) ((p)->size & IS_MMAPPED) | |
f65fd747 | 1278 | |
f65fd747 | 1279 | |
fa8d436c UD |
1280 | /* size field is or'ed with NON_MAIN_ARENA if the chunk was obtained |
1281 | from a non-main arena. This is only set immediately before handing | |
1282 | the chunk to the user, if necessary. */ | |
1283 | #define NON_MAIN_ARENA 0x4 | |
f65fd747 | 1284 | |
fa8d436c UD |
1285 | /* check for chunk from non-main arena */ |
1286 | #define chunk_non_main_arena(p) ((p)->size & NON_MAIN_ARENA) | |
f65fd747 UD |
1287 | |
1288 | ||
a9177ff5 | 1289 | /* |
6c8dbf00 | 1290 | Bits to mask off when extracting size |
f65fd747 | 1291 | |
6c8dbf00 OB |
1292 | Note: IS_MMAPPED is intentionally not masked off from size field in |
1293 | macros for which mmapped chunks should never be seen. This should | |
1294 | cause helpful core dumps to occur if it is tried by accident by | |
1295 | people extending or adapting this malloc. | |
1296 | */ | |
1297 | #define SIZE_BITS (PREV_INUSE | IS_MMAPPED | NON_MAIN_ARENA) | |
f65fd747 | 1298 | |
fa8d436c UD |
1299 | /* Get size, ignoring use bits */ |
1300 | #define chunksize(p) ((p)->size & ~(SIZE_BITS)) | |
f65fd747 | 1301 | |
f65fd747 | 1302 | |
fa8d436c | 1303 | /* Ptr to next physical malloc_chunk. */ |
6c8dbf00 | 1304 | #define next_chunk(p) ((mchunkptr) (((char *) (p)) + ((p)->size & ~SIZE_BITS))) |
f65fd747 | 1305 | |
fa8d436c | 1306 | /* Ptr to previous physical malloc_chunk */ |
6c8dbf00 | 1307 | #define prev_chunk(p) ((mchunkptr) (((char *) (p)) - ((p)->prev_size))) |
f65fd747 | 1308 | |
fa8d436c | 1309 | /* Treat space at ptr + offset as a chunk */ |
6c8dbf00 | 1310 | #define chunk_at_offset(p, s) ((mchunkptr) (((char *) (p)) + (s))) |
fa8d436c UD |
1311 | |
1312 | /* extract p's inuse bit */ | |
6c8dbf00 OB |
1313 | #define inuse(p) \ |
1314 | ((((mchunkptr) (((char *) (p)) + ((p)->size & ~SIZE_BITS)))->size) & PREV_INUSE) | |
f65fd747 | 1315 | |
fa8d436c | 1316 | /* set/clear chunk as being inuse without otherwise disturbing */ |
6c8dbf00 OB |
1317 | #define set_inuse(p) \ |
1318 | ((mchunkptr) (((char *) (p)) + ((p)->size & ~SIZE_BITS)))->size |= PREV_INUSE | |
f65fd747 | 1319 | |
6c8dbf00 OB |
1320 | #define clear_inuse(p) \ |
1321 | ((mchunkptr) (((char *) (p)) + ((p)->size & ~SIZE_BITS)))->size &= ~(PREV_INUSE) | |
f65fd747 UD |
1322 | |
1323 | ||
fa8d436c | 1324 | /* check/set/clear inuse bits in known places */ |
6c8dbf00 OB |
1325 | #define inuse_bit_at_offset(p, s) \ |
1326 | (((mchunkptr) (((char *) (p)) + (s)))->size & PREV_INUSE) | |
f65fd747 | 1327 | |
6c8dbf00 OB |
1328 | #define set_inuse_bit_at_offset(p, s) \ |
1329 | (((mchunkptr) (((char *) (p)) + (s)))->size |= PREV_INUSE) | |
f65fd747 | 1330 | |
6c8dbf00 OB |
1331 | #define clear_inuse_bit_at_offset(p, s) \ |
1332 | (((mchunkptr) (((char *) (p)) + (s)))->size &= ~(PREV_INUSE)) | |
f65fd747 | 1333 | |
f65fd747 | 1334 | |
fa8d436c UD |
1335 | /* Set size at head, without disturbing its use bit */ |
1336 | #define set_head_size(p, s) ((p)->size = (((p)->size & SIZE_BITS) | (s))) | |
f65fd747 | 1337 | |
fa8d436c UD |
1338 | /* Set size/use field */ |
1339 | #define set_head(p, s) ((p)->size = (s)) | |
f65fd747 | 1340 | |
fa8d436c | 1341 | /* Set size at footer (only when chunk is not in use) */ |
6c8dbf00 | 1342 | #define set_foot(p, s) (((mchunkptr) ((char *) (p) + (s)))->prev_size = (s)) |
f65fd747 UD |
1343 | |
1344 | ||
fa8d436c | 1345 | /* |
6c8dbf00 | 1346 | -------------------- Internal data structures -------------------- |
fa8d436c UD |
1347 | |
1348 | All internal state is held in an instance of malloc_state defined | |
1349 | below. There are no other static variables, except in two optional | |
a9177ff5 | 1350 | cases: |
6c8dbf00 OB |
1351 | * If USE_MALLOC_LOCK is defined, the mALLOC_MUTEx declared above. |
1352 | * If mmap doesn't support MAP_ANONYMOUS, a dummy file descriptor | |
22a89187 | 1353 | for mmap. |
fa8d436c UD |
1354 | |
1355 | Beware of lots of tricks that minimize the total bookkeeping space | |
1356 | requirements. The result is a little over 1K bytes (for 4byte | |
1357 | pointers and size_t.) | |
6c8dbf00 | 1358 | */ |
f65fd747 UD |
1359 | |
1360 | /* | |
6c8dbf00 | 1361 | Bins |
fa8d436c UD |
1362 | |
1363 | An array of bin headers for free chunks. Each bin is doubly | |
1364 | linked. The bins are approximately proportionally (log) spaced. | |
1365 | There are a lot of these bins (128). This may look excessive, but | |
1366 | works very well in practice. Most bins hold sizes that are | |
1367 | unusual as malloc request sizes, but are more usual for fragments | |
1368 | and consolidated sets of chunks, which is what these bins hold, so | |
1369 | they can be found quickly. All procedures maintain the invariant | |
1370 | that no consolidated chunk physically borders another one, so each | |
1371 | chunk in a list is known to be preceeded and followed by either | |
1372 | inuse chunks or the ends of memory. | |
1373 | ||
1374 | Chunks in bins are kept in size order, with ties going to the | |
1375 | approximately least recently used chunk. Ordering isn't needed | |
1376 | for the small bins, which all contain the same-sized chunks, but | |
1377 | facilitates best-fit allocation for larger chunks. These lists | |
1378 | are just sequential. Keeping them in order almost never requires | |
1379 | enough traversal to warrant using fancier ordered data | |
a9177ff5 | 1380 | structures. |
fa8d436c UD |
1381 | |
1382 | Chunks of the same size are linked with the most | |
1383 | recently freed at the front, and allocations are taken from the | |
1384 | back. This results in LRU (FIFO) allocation order, which tends | |
1385 | to give each chunk an equal opportunity to be consolidated with | |
1386 | adjacent freed chunks, resulting in larger free chunks and less | |
1387 | fragmentation. | |
1388 | ||
1389 | To simplify use in double-linked lists, each bin header acts | |
1390 | as a malloc_chunk. This avoids special-casing for headers. | |
1391 | But to conserve space and improve locality, we allocate | |
1392 | only the fd/bk pointers of bins, and then use repositioning tricks | |
a9177ff5 | 1393 | to treat these as the fields of a malloc_chunk*. |
6c8dbf00 | 1394 | */ |
f65fd747 | 1395 | |
6c8dbf00 | 1396 | typedef struct malloc_chunk *mbinptr; |
f65fd747 | 1397 | |
fa8d436c | 1398 | /* addressing -- note that bin_at(0) does not exist */ |
41999a1a UD |
1399 | #define bin_at(m, i) \ |
1400 | (mbinptr) (((char *) &((m)->bins[((i) - 1) * 2])) \ | |
6c8dbf00 | 1401 | - offsetof (struct malloc_chunk, fd)) |
f65fd747 | 1402 | |
fa8d436c | 1403 | /* analog of ++bin */ |
6c8dbf00 | 1404 | #define next_bin(b) ((mbinptr) ((char *) (b) + (sizeof (mchunkptr) << 1))) |
f65fd747 | 1405 | |
fa8d436c UD |
1406 | /* Reminders about list directionality within bins */ |
1407 | #define first(b) ((b)->fd) | |
1408 | #define last(b) ((b)->bk) | |
f65fd747 | 1409 | |
fa8d436c UD |
1410 | /* Take a chunk off a bin list */ |
1411 | #define unlink(P, BK, FD) { \ | |
6c8dbf00 OB |
1412 | FD = P->fd; \ |
1413 | BK = P->bk; \ | |
1414 | if (__builtin_expect (FD->bk != P || BK->fd != P, 0)) \ | |
1415 | malloc_printerr (check_action, "corrupted double-linked list", P); \ | |
1416 | else { \ | |
1417 | FD->bk = BK; \ | |
1418 | BK->fd = FD; \ | |
1419 | if (!in_smallbin_range (P->size) \ | |
1420 | && __builtin_expect (P->fd_nextsize != NULL, 0)) { \ | |
1421 | assert (P->fd_nextsize->bk_nextsize == P); \ | |
1422 | assert (P->bk_nextsize->fd_nextsize == P); \ | |
1423 | if (FD->fd_nextsize == NULL) { \ | |
1424 | if (P->fd_nextsize == P) \ | |
1425 | FD->fd_nextsize = FD->bk_nextsize = FD; \ | |
1426 | else { \ | |
1427 | FD->fd_nextsize = P->fd_nextsize; \ | |
1428 | FD->bk_nextsize = P->bk_nextsize; \ | |
1429 | P->fd_nextsize->bk_nextsize = FD; \ | |
1430 | P->bk_nextsize->fd_nextsize = FD; \ | |
1431 | } \ | |
1432 | } else { \ | |
1433 | P->fd_nextsize->bk_nextsize = P->bk_nextsize; \ | |
1434 | P->bk_nextsize->fd_nextsize = P->fd_nextsize; \ | |
1435 | } \ | |
1436 | } \ | |
1437 | } \ | |
fa8d436c | 1438 | } |
f65fd747 | 1439 | |
fa8d436c | 1440 | /* |
6c8dbf00 | 1441 | Indexing |
fa8d436c UD |
1442 | |
1443 | Bins for sizes < 512 bytes contain chunks of all the same size, spaced | |
1444 | 8 bytes apart. Larger bins are approximately logarithmically spaced: | |
f65fd747 | 1445 | |
fa8d436c UD |
1446 | 64 bins of size 8 |
1447 | 32 bins of size 64 | |
1448 | 16 bins of size 512 | |
1449 | 8 bins of size 4096 | |
1450 | 4 bins of size 32768 | |
1451 | 2 bins of size 262144 | |
1452 | 1 bin of size what's left | |
f65fd747 | 1453 | |
fa8d436c UD |
1454 | There is actually a little bit of slop in the numbers in bin_index |
1455 | for the sake of speed. This makes no difference elsewhere. | |
f65fd747 | 1456 | |
fa8d436c UD |
1457 | The bins top out around 1MB because we expect to service large |
1458 | requests via mmap. | |
b5a2bbe6 L |
1459 | |
1460 | Bin 0 does not exist. Bin 1 is the unordered list; if that would be | |
1461 | a valid chunk size the small bins are bumped up one. | |
6c8dbf00 | 1462 | */ |
f65fd747 | 1463 | |
fa8d436c UD |
1464 | #define NBINS 128 |
1465 | #define NSMALLBINS 64 | |
1d47e92f | 1466 | #define SMALLBIN_WIDTH MALLOC_ALIGNMENT |
b5a2bbe6 L |
1467 | #define SMALLBIN_CORRECTION (MALLOC_ALIGNMENT > 2 * SIZE_SZ) |
1468 | #define MIN_LARGE_SIZE ((NSMALLBINS - SMALLBIN_CORRECTION) * SMALLBIN_WIDTH) | |
f65fd747 | 1469 | |
fa8d436c | 1470 | #define in_smallbin_range(sz) \ |
6c8dbf00 | 1471 | ((unsigned long) (sz) < (unsigned long) MIN_LARGE_SIZE) |
f65fd747 | 1472 | |
1d47e92f | 1473 | #define smallbin_index(sz) \ |
6c8dbf00 | 1474 | ((SMALLBIN_WIDTH == 16 ? (((unsigned) (sz)) >> 4) : (((unsigned) (sz)) >> 3))\ |
b5a2bbe6 | 1475 | + SMALLBIN_CORRECTION) |
f65fd747 | 1476 | |
1d47e92f | 1477 | #define largebin_index_32(sz) \ |
6c8dbf00 OB |
1478 | (((((unsigned long) (sz)) >> 6) <= 38) ? 56 + (((unsigned long) (sz)) >> 6) :\ |
1479 | ((((unsigned long) (sz)) >> 9) <= 20) ? 91 + (((unsigned long) (sz)) >> 9) :\ | |
1480 | ((((unsigned long) (sz)) >> 12) <= 10) ? 110 + (((unsigned long) (sz)) >> 12) :\ | |
1481 | ((((unsigned long) (sz)) >> 15) <= 4) ? 119 + (((unsigned long) (sz)) >> 15) :\ | |
1482 | ((((unsigned long) (sz)) >> 18) <= 2) ? 124 + (((unsigned long) (sz)) >> 18) :\ | |
1483 | 126) | |
f65fd747 | 1484 | |
b5a2bbe6 | 1485 | #define largebin_index_32_big(sz) \ |
6c8dbf00 OB |
1486 | (((((unsigned long) (sz)) >> 6) <= 45) ? 49 + (((unsigned long) (sz)) >> 6) :\ |
1487 | ((((unsigned long) (sz)) >> 9) <= 20) ? 91 + (((unsigned long) (sz)) >> 9) :\ | |
1488 | ((((unsigned long) (sz)) >> 12) <= 10) ? 110 + (((unsigned long) (sz)) >> 12) :\ | |
1489 | ((((unsigned long) (sz)) >> 15) <= 4) ? 119 + (((unsigned long) (sz)) >> 15) :\ | |
1490 | ((((unsigned long) (sz)) >> 18) <= 2) ? 124 + (((unsigned long) (sz)) >> 18) :\ | |
1491 | 126) | |
b5a2bbe6 | 1492 | |
1d47e92f UD |
1493 | // XXX It remains to be seen whether it is good to keep the widths of |
1494 | // XXX the buckets the same or whether it should be scaled by a factor | |
1495 | // XXX of two as well. | |
1496 | #define largebin_index_64(sz) \ | |
6c8dbf00 OB |
1497 | (((((unsigned long) (sz)) >> 6) <= 48) ? 48 + (((unsigned long) (sz)) >> 6) :\ |
1498 | ((((unsigned long) (sz)) >> 9) <= 20) ? 91 + (((unsigned long) (sz)) >> 9) :\ | |
1499 | ((((unsigned long) (sz)) >> 12) <= 10) ? 110 + (((unsigned long) (sz)) >> 12) :\ | |
1500 | ((((unsigned long) (sz)) >> 15) <= 4) ? 119 + (((unsigned long) (sz)) >> 15) :\ | |
1501 | ((((unsigned long) (sz)) >> 18) <= 2) ? 124 + (((unsigned long) (sz)) >> 18) :\ | |
1502 | 126) | |
1d47e92f UD |
1503 | |
1504 | #define largebin_index(sz) \ | |
b5a2bbe6 L |
1505 | (SIZE_SZ == 8 ? largebin_index_64 (sz) \ |
1506 | : MALLOC_ALIGNMENT == 16 ? largebin_index_32_big (sz) \ | |
1507 | : largebin_index_32 (sz)) | |
1d47e92f | 1508 | |
fa8d436c | 1509 | #define bin_index(sz) \ |
6c8dbf00 | 1510 | ((in_smallbin_range (sz)) ? smallbin_index (sz) : largebin_index (sz)) |
f65fd747 | 1511 | |
f65fd747 UD |
1512 | |
1513 | /* | |
6c8dbf00 | 1514 | Unsorted chunks |
fa8d436c UD |
1515 | |
1516 | All remainders from chunk splits, as well as all returned chunks, | |
1517 | are first placed in the "unsorted" bin. They are then placed | |
1518 | in regular bins after malloc gives them ONE chance to be used before | |
1519 | binning. So, basically, the unsorted_chunks list acts as a queue, | |
1520 | with chunks being placed on it in free (and malloc_consolidate), | |
1521 | and taken off (to be either used or placed in bins) in malloc. | |
1522 | ||
1523 | The NON_MAIN_ARENA flag is never set for unsorted chunks, so it | |
1524 | does not have to be taken into account in size comparisons. | |
6c8dbf00 | 1525 | */ |
f65fd747 | 1526 | |
fa8d436c | 1527 | /* The otherwise unindexable 1-bin is used to hold unsorted chunks. */ |
6c8dbf00 | 1528 | #define unsorted_chunks(M) (bin_at (M, 1)) |
f65fd747 | 1529 | |
fa8d436c | 1530 | /* |
6c8dbf00 | 1531 | Top |
fa8d436c UD |
1532 | |
1533 | The top-most available chunk (i.e., the one bordering the end of | |
1534 | available memory) is treated specially. It is never included in | |
1535 | any bin, is used only if no other chunk is available, and is | |
1536 | released back to the system if it is very large (see | |
1537 | M_TRIM_THRESHOLD). Because top initially | |
1538 | points to its own bin with initial zero size, thus forcing | |
1539 | extension on the first malloc request, we avoid having any special | |
1540 | code in malloc to check whether it even exists yet. But we still | |
1541 | need to do so when getting memory from system, so we make | |
1542 | initial_top treat the bin as a legal but unusable chunk during the | |
1543 | interval between initialization and the first call to | |
3b49edc0 | 1544 | sysmalloc. (This is somewhat delicate, since it relies on |
fa8d436c | 1545 | the 2 preceding words to be zero during this interval as well.) |
6c8dbf00 | 1546 | */ |
f65fd747 | 1547 | |
fa8d436c | 1548 | /* Conveniently, the unsorted bin can be used as dummy top on first call */ |
6c8dbf00 | 1549 | #define initial_top(M) (unsorted_chunks (M)) |
f65fd747 | 1550 | |
fa8d436c | 1551 | /* |
6c8dbf00 | 1552 | Binmap |
f65fd747 | 1553 | |
fa8d436c UD |
1554 | To help compensate for the large number of bins, a one-level index |
1555 | structure is used for bin-by-bin searching. `binmap' is a | |
1556 | bitvector recording whether bins are definitely empty so they can | |
1557 | be skipped over during during traversals. The bits are NOT always | |
1558 | cleared as soon as bins are empty, but instead only | |
1559 | when they are noticed to be empty during traversal in malloc. | |
6c8dbf00 | 1560 | */ |
f65fd747 | 1561 | |
fa8d436c UD |
1562 | /* Conservatively use 32 bits per map word, even if on 64bit system */ |
1563 | #define BINMAPSHIFT 5 | |
1564 | #define BITSPERMAP (1U << BINMAPSHIFT) | |
1565 | #define BINMAPSIZE (NBINS / BITSPERMAP) | |
f65fd747 | 1566 | |
fa8d436c | 1567 | #define idx2block(i) ((i) >> BINMAPSHIFT) |
6c8dbf00 | 1568 | #define idx2bit(i) ((1U << ((i) & ((1U << BINMAPSHIFT) - 1)))) |
f65fd747 | 1569 | |
6c8dbf00 OB |
1570 | #define mark_bin(m, i) ((m)->binmap[idx2block (i)] |= idx2bit (i)) |
1571 | #define unmark_bin(m, i) ((m)->binmap[idx2block (i)] &= ~(idx2bit (i))) | |
1572 | #define get_binmap(m, i) ((m)->binmap[idx2block (i)] & idx2bit (i)) | |
f65fd747 | 1573 | |
fa8d436c | 1574 | /* |
6c8dbf00 | 1575 | Fastbins |
fa8d436c UD |
1576 | |
1577 | An array of lists holding recently freed small chunks. Fastbins | |
1578 | are not doubly linked. It is faster to single-link them, and | |
1579 | since chunks are never removed from the middles of these lists, | |
1580 | double linking is not necessary. Also, unlike regular bins, they | |
1581 | are not even processed in FIFO order (they use faster LIFO) since | |
1582 | ordering doesn't much matter in the transient contexts in which | |
1583 | fastbins are normally used. | |
1584 | ||
1585 | Chunks in fastbins keep their inuse bit set, so they cannot | |
1586 | be consolidated with other free chunks. malloc_consolidate | |
1587 | releases all chunks in fastbins and consolidates them with | |
a9177ff5 | 1588 | other free chunks. |
6c8dbf00 | 1589 | */ |
f65fd747 | 1590 | |
6c8dbf00 | 1591 | typedef struct malloc_chunk *mfastbinptr; |
425ce2ed | 1592 | #define fastbin(ar_ptr, idx) ((ar_ptr)->fastbinsY[idx]) |
f65fd747 | 1593 | |
fa8d436c | 1594 | /* offset 2 to use otherwise unindexable first 2 bins */ |
425ce2ed | 1595 | #define fastbin_index(sz) \ |
6c8dbf00 | 1596 | ((((unsigned int) (sz)) >> (SIZE_SZ == 8 ? 4 : 3)) - 2) |
425ce2ed | 1597 | |
f65fd747 | 1598 | |
fa8d436c | 1599 | /* The maximum fastbin request size we support */ |
425ce2ed | 1600 | #define MAX_FAST_SIZE (80 * SIZE_SZ / 4) |
f65fd747 | 1601 | |
6c8dbf00 | 1602 | #define NFASTBINS (fastbin_index (request2size (MAX_FAST_SIZE)) + 1) |
f65fd747 UD |
1603 | |
1604 | /* | |
6c8dbf00 OB |
1605 | FASTBIN_CONSOLIDATION_THRESHOLD is the size of a chunk in free() |
1606 | that triggers automatic consolidation of possibly-surrounding | |
1607 | fastbin chunks. This is a heuristic, so the exact value should not | |
1608 | matter too much. It is defined at half the default trim threshold as a | |
1609 | compromise heuristic to only attempt consolidation if it is likely | |
1610 | to lead to trimming. However, it is not dynamically tunable, since | |
1611 | consolidation reduces fragmentation surrounding large chunks even | |
1612 | if trimming is not used. | |
1613 | */ | |
f65fd747 | 1614 | |
fa8d436c | 1615 | #define FASTBIN_CONSOLIDATION_THRESHOLD (65536UL) |
f65fd747 UD |
1616 | |
1617 | /* | |
6c8dbf00 OB |
1618 | Since the lowest 2 bits in max_fast don't matter in size comparisons, |
1619 | they are used as flags. | |
1620 | */ | |
f65fd747 | 1621 | |
fa8d436c | 1622 | /* |
6c8dbf00 OB |
1623 | FASTCHUNKS_BIT held in max_fast indicates that there are probably |
1624 | some fastbin chunks. It is set true on entering a chunk into any | |
1625 | fastbin, and cleared only in malloc_consolidate. | |
f65fd747 | 1626 | |
6c8dbf00 OB |
1627 | The truth value is inverted so that have_fastchunks will be true |
1628 | upon startup (since statics are zero-filled), simplifying | |
1629 | initialization checks. | |
1630 | */ | |
f65fd747 | 1631 | |
fa8d436c | 1632 | #define FASTCHUNKS_BIT (1U) |
f65fd747 | 1633 | |
6c8dbf00 | 1634 | #define have_fastchunks(M) (((M)->flags & FASTCHUNKS_BIT) == 0) |
425ce2ed UD |
1635 | #define clear_fastchunks(M) catomic_or (&(M)->flags, FASTCHUNKS_BIT) |
1636 | #define set_fastchunks(M) catomic_and (&(M)->flags, ~FASTCHUNKS_BIT) | |
f65fd747 UD |
1637 | |
1638 | /* | |
6c8dbf00 OB |
1639 | NONCONTIGUOUS_BIT indicates that MORECORE does not return contiguous |
1640 | regions. Otherwise, contiguity is exploited in merging together, | |
1641 | when possible, results from consecutive MORECORE calls. | |
f65fd747 | 1642 | |
6c8dbf00 OB |
1643 | The initial value comes from MORECORE_CONTIGUOUS, but is |
1644 | changed dynamically if mmap is ever used as an sbrk substitute. | |
1645 | */ | |
f65fd747 | 1646 | |
fa8d436c | 1647 | #define NONCONTIGUOUS_BIT (2U) |
f65fd747 | 1648 | |
6c8dbf00 OB |
1649 | #define contiguous(M) (((M)->flags & NONCONTIGUOUS_BIT) == 0) |
1650 | #define noncontiguous(M) (((M)->flags & NONCONTIGUOUS_BIT) != 0) | |
1651 | #define set_noncontiguous(M) ((M)->flags |= NONCONTIGUOUS_BIT) | |
9bf248c6 | 1652 | #define set_contiguous(M) ((M)->flags &= ~NONCONTIGUOUS_BIT) |
f65fd747 | 1653 | |
a9177ff5 RM |
1654 | /* |
1655 | Set value of max_fast. | |
fa8d436c UD |
1656 | Use impossibly small value if 0. |
1657 | Precondition: there are no existing fastbin chunks. | |
1658 | Setting the value clears fastchunk bit but preserves noncontiguous bit. | |
6c8dbf00 | 1659 | */ |
f65fd747 | 1660 | |
9bf248c6 | 1661 | #define set_max_fast(s) \ |
991eda1e | 1662 | global_max_fast = (((s) == 0) \ |
6c8dbf00 | 1663 | ? SMALLBIN_WIDTH : ((s + SIZE_SZ) & ~MALLOC_ALIGN_MASK)) |
9bf248c6 | 1664 | #define get_max_fast() global_max_fast |
f65fd747 | 1665 | |
f65fd747 UD |
1666 | |
1667 | /* | |
fa8d436c | 1668 | ----------- Internal state representation and initialization ----------- |
6c8dbf00 | 1669 | */ |
f65fd747 | 1670 | |
6c8dbf00 OB |
1671 | struct malloc_state |
1672 | { | |
fa8d436c UD |
1673 | /* Serialize access. */ |
1674 | mutex_t mutex; | |
9bf248c6 UD |
1675 | |
1676 | /* Flags (formerly in max_fast). */ | |
1677 | int flags; | |
f65fd747 | 1678 | |
fa8d436c | 1679 | /* Fastbins */ |
6c8dbf00 | 1680 | mfastbinptr fastbinsY[NFASTBINS]; |
f65fd747 | 1681 | |
fa8d436c | 1682 | /* Base of the topmost chunk -- not otherwise kept in a bin */ |
6c8dbf00 | 1683 | mchunkptr top; |
f65fd747 | 1684 | |
fa8d436c | 1685 | /* The remainder from the most recent split of a small request */ |
6c8dbf00 | 1686 | mchunkptr last_remainder; |
f65fd747 | 1687 | |
fa8d436c | 1688 | /* Normal bins packed as described above */ |
6c8dbf00 | 1689 | mchunkptr bins[NBINS * 2 - 2]; |
f65fd747 | 1690 | |
fa8d436c | 1691 | /* Bitmap of bins */ |
6c8dbf00 | 1692 | unsigned int binmap[BINMAPSIZE]; |
f65fd747 | 1693 | |
fa8d436c UD |
1694 | /* Linked list */ |
1695 | struct malloc_state *next; | |
f65fd747 | 1696 | |
425ce2ed UD |
1697 | /* Linked list for free arenas. */ |
1698 | struct malloc_state *next_free; | |
425ce2ed | 1699 | |
fa8d436c UD |
1700 | /* Memory allocated from the system in this arena. */ |
1701 | INTERNAL_SIZE_T system_mem; | |
1702 | INTERNAL_SIZE_T max_system_mem; | |
1703 | }; | |
f65fd747 | 1704 | |
6c8dbf00 OB |
1705 | struct malloc_par |
1706 | { | |
fa8d436c | 1707 | /* Tunable parameters */ |
6c8dbf00 OB |
1708 | unsigned long trim_threshold; |
1709 | INTERNAL_SIZE_T top_pad; | |
1710 | INTERNAL_SIZE_T mmap_threshold; | |
1711 | INTERNAL_SIZE_T arena_test; | |
1712 | INTERNAL_SIZE_T arena_max; | |
fa8d436c UD |
1713 | |
1714 | /* Memory map support */ | |
6c8dbf00 OB |
1715 | int n_mmaps; |
1716 | int n_mmaps_max; | |
1717 | int max_n_mmaps; | |
1d05c2fb UD |
1718 | /* the mmap_threshold is dynamic, until the user sets |
1719 | it manually, at which point we need to disable any | |
1720 | dynamic behavior. */ | |
6c8dbf00 | 1721 | int no_dyn_threshold; |
fa8d436c | 1722 | |
fa8d436c | 1723 | /* Statistics */ |
6c8dbf00 | 1724 | INTERNAL_SIZE_T mmapped_mem; |
fa8d436c UD |
1725 | /*INTERNAL_SIZE_T sbrked_mem;*/ |
1726 | /*INTERNAL_SIZE_T max_sbrked_mem;*/ | |
6c8dbf00 OB |
1727 | INTERNAL_SIZE_T max_mmapped_mem; |
1728 | INTERNAL_SIZE_T max_total_mem; /* only kept for NO_THREADS */ | |
fa8d436c UD |
1729 | |
1730 | /* First address handed out by MORECORE/sbrk. */ | |
6c8dbf00 | 1731 | char *sbrk_base; |
fa8d436c | 1732 | }; |
f65fd747 | 1733 | |
fa8d436c UD |
1734 | /* There are several instances of this struct ("arenas") in this |
1735 | malloc. If you are adapting this malloc in a way that does NOT use | |
1736 | a static or mmapped malloc_state, you MUST explicitly zero-fill it | |
1737 | before using. This malloc relies on the property that malloc_state | |
1738 | is initialized to all zeroes (as is true of C statics). */ | |
f65fd747 | 1739 | |
02d46fc4 | 1740 | static struct malloc_state main_arena = |
6c8dbf00 OB |
1741 | { |
1742 | .mutex = MUTEX_INITIALIZER, | |
1743 | .next = &main_arena | |
1744 | }; | |
f65fd747 | 1745 | |
fa8d436c | 1746 | /* There is only one instance of the malloc parameters. */ |
f65fd747 | 1747 | |
02d46fc4 | 1748 | static struct malloc_par mp_ = |
6c8dbf00 OB |
1749 | { |
1750 | .top_pad = DEFAULT_TOP_PAD, | |
1751 | .n_mmaps_max = DEFAULT_MMAP_MAX, | |
1752 | .mmap_threshold = DEFAULT_MMAP_THRESHOLD, | |
1753 | .trim_threshold = DEFAULT_TRIM_THRESHOLD, | |
1754 | #define NARENAS_FROM_NCORES(n) ((n) * (sizeof (long) == 4 ? 2 : 8)) | |
1755 | .arena_test = NARENAS_FROM_NCORES (1) | |
1756 | }; | |
f65fd747 | 1757 | |
9bf248c6 | 1758 | |
425ce2ed UD |
1759 | /* Non public mallopt parameters. */ |
1760 | #define M_ARENA_TEST -7 | |
1761 | #define M_ARENA_MAX -8 | |
425ce2ed UD |
1762 | |
1763 | ||
9bf248c6 UD |
1764 | /* Maximum size of memory handled in fastbins. */ |
1765 | static INTERNAL_SIZE_T global_max_fast; | |
1766 | ||
fa8d436c | 1767 | /* |
6c8dbf00 | 1768 | Initialize a malloc_state struct. |
f65fd747 | 1769 | |
6c8dbf00 OB |
1770 | This is called only from within malloc_consolidate, which needs |
1771 | be called in the same contexts anyway. It is never called directly | |
1772 | outside of malloc_consolidate because some optimizing compilers try | |
1773 | to inline it at all call points, which turns out not to be an | |
1774 | optimization at all. (Inlining it in malloc_consolidate is fine though.) | |
1775 | */ | |
f65fd747 | 1776 | |
6c8dbf00 OB |
1777 | static void |
1778 | malloc_init_state (mstate av) | |
fa8d436c | 1779 | { |
6c8dbf00 | 1780 | int i; |
fa8d436c | 1781 | mbinptr bin; |
a9177ff5 | 1782 | |
fa8d436c | 1783 | /* Establish circular links for normal bins */ |
6c8dbf00 OB |
1784 | for (i = 1; i < NBINS; ++i) |
1785 | { | |
1786 | bin = bin_at (av, i); | |
1787 | bin->fd = bin->bk = bin; | |
1788 | } | |
f65fd747 | 1789 | |
fa8d436c UD |
1790 | #if MORECORE_CONTIGUOUS |
1791 | if (av != &main_arena) | |
1792 | #endif | |
6c8dbf00 | 1793 | set_noncontiguous (av); |
9bf248c6 | 1794 | if (av == &main_arena) |
6c8dbf00 | 1795 | set_max_fast (DEFAULT_MXFAST); |
9bf248c6 | 1796 | av->flags |= FASTCHUNKS_BIT; |
f65fd747 | 1797 | |
6c8dbf00 | 1798 | av->top = initial_top (av); |
fa8d436c | 1799 | } |
e9b3e3c5 | 1800 | |
a9177ff5 | 1801 | /* |
fa8d436c | 1802 | Other internal utilities operating on mstates |
6c8dbf00 | 1803 | */ |
f65fd747 | 1804 | |
6c8dbf00 OB |
1805 | static void *sysmalloc (INTERNAL_SIZE_T, mstate); |
1806 | static int systrim (size_t, mstate); | |
1807 | static void malloc_consolidate (mstate); | |
7e3be507 | 1808 | |
404d4cef RM |
1809 | |
1810 | /* -------------- Early definitions for debugging hooks ---------------- */ | |
1811 | ||
1812 | /* Define and initialize the hook variables. These weak definitions must | |
1813 | appear before any use of the variables in a function (arena.c uses one). */ | |
1814 | #ifndef weak_variable | |
404d4cef RM |
1815 | /* In GNU libc we want the hook variables to be weak definitions to |
1816 | avoid a problem with Emacs. */ | |
22a89187 | 1817 | # define weak_variable weak_function |
404d4cef RM |
1818 | #endif |
1819 | ||
1820 | /* Forward declarations. */ | |
6c8dbf00 OB |
1821 | static void *malloc_hook_ini (size_t sz, |
1822 | const void *caller) __THROW; | |
1823 | static void *realloc_hook_ini (void *ptr, size_t sz, | |
1824 | const void *caller) __THROW; | |
1825 | static void *memalign_hook_ini (size_t alignment, size_t sz, | |
1826 | const void *caller) __THROW; | |
404d4cef | 1827 | |
06d6611a | 1828 | void weak_variable (*__malloc_initialize_hook) (void) = NULL; |
a222d91a | 1829 | void weak_variable (*__free_hook) (void *__ptr, |
6c8dbf00 | 1830 | const void *) = NULL; |
a222d91a | 1831 | void *weak_variable (*__malloc_hook) |
6c8dbf00 | 1832 | (size_t __size, const void *) = malloc_hook_ini; |
a222d91a | 1833 | void *weak_variable (*__realloc_hook) |
6c8dbf00 OB |
1834 | (void *__ptr, size_t __size, const void *) |
1835 | = realloc_hook_ini; | |
a222d91a | 1836 | void *weak_variable (*__memalign_hook) |
6c8dbf00 OB |
1837 | (size_t __alignment, size_t __size, const void *) |
1838 | = memalign_hook_ini; | |
06d6611a | 1839 | void weak_variable (*__after_morecore_hook) (void) = NULL; |
404d4cef RM |
1840 | |
1841 | ||
3e030bd5 UD |
1842 | /* ---------------- Error behavior ------------------------------------ */ |
1843 | ||
1844 | #ifndef DEFAULT_CHECK_ACTION | |
6c8dbf00 | 1845 | # define DEFAULT_CHECK_ACTION 3 |
3e030bd5 UD |
1846 | #endif |
1847 | ||
1848 | static int check_action = DEFAULT_CHECK_ACTION; | |
1849 | ||
1850 | ||
854278df UD |
1851 | /* ------------------ Testing support ----------------------------------*/ |
1852 | ||
1853 | static int perturb_byte; | |
1854 | ||
e8349efd OB |
1855 | static inline void |
1856 | alloc_perturb (char *p, size_t n) | |
1857 | { | |
1858 | if (__glibc_unlikely (perturb_byte)) | |
1859 | memset (p, perturb_byte ^ 0xff, n); | |
1860 | } | |
1861 | ||
1862 | static inline void | |
1863 | free_perturb (char *p, size_t n) | |
1864 | { | |
1865 | if (__glibc_unlikely (perturb_byte)) | |
1866 | memset (p, perturb_byte, n); | |
1867 | } | |
1868 | ||
854278df UD |
1869 | |
1870 | ||
3ea5be54 AO |
1871 | #include <stap-probe.h> |
1872 | ||
fa8d436c UD |
1873 | /* ------------------- Support for multiple arenas -------------------- */ |
1874 | #include "arena.c" | |
f65fd747 | 1875 | |
fa8d436c | 1876 | /* |
6c8dbf00 | 1877 | Debugging support |
f65fd747 | 1878 | |
6c8dbf00 OB |
1879 | These routines make a number of assertions about the states |
1880 | of data structures that should be true at all times. If any | |
1881 | are not true, it's very likely that a user program has somehow | |
1882 | trashed memory. (It's also possible that there is a coding error | |
1883 | in malloc. In which case, please report it!) | |
1884 | */ | |
ee74a442 | 1885 | |
6c8dbf00 | 1886 | #if !MALLOC_DEBUG |
d8f00d46 | 1887 | |
6c8dbf00 OB |
1888 | # define check_chunk(A, P) |
1889 | # define check_free_chunk(A, P) | |
1890 | # define check_inuse_chunk(A, P) | |
1891 | # define check_remalloced_chunk(A, P, N) | |
1892 | # define check_malloced_chunk(A, P, N) | |
1893 | # define check_malloc_state(A) | |
d8f00d46 | 1894 | |
fa8d436c | 1895 | #else |
ca34d7a7 | 1896 | |
6c8dbf00 OB |
1897 | # define check_chunk(A, P) do_check_chunk (A, P) |
1898 | # define check_free_chunk(A, P) do_check_free_chunk (A, P) | |
1899 | # define check_inuse_chunk(A, P) do_check_inuse_chunk (A, P) | |
1900 | # define check_remalloced_chunk(A, P, N) do_check_remalloced_chunk (A, P, N) | |
1901 | # define check_malloced_chunk(A, P, N) do_check_malloced_chunk (A, P, N) | |
1902 | # define check_malloc_state(A) do_check_malloc_state (A) | |
ca34d7a7 | 1903 | |
fa8d436c | 1904 | /* |
6c8dbf00 OB |
1905 | Properties of all chunks |
1906 | */ | |
ca34d7a7 | 1907 | |
6c8dbf00 OB |
1908 | static void |
1909 | do_check_chunk (mstate av, mchunkptr p) | |
ca34d7a7 | 1910 | { |
6c8dbf00 | 1911 | unsigned long sz = chunksize (p); |
fa8d436c | 1912 | /* min and max possible addresses assuming contiguous allocation */ |
6c8dbf00 OB |
1913 | char *max_address = (char *) (av->top) + chunksize (av->top); |
1914 | char *min_address = max_address - av->system_mem; | |
fa8d436c | 1915 | |
6c8dbf00 OB |
1916 | if (!chunk_is_mmapped (p)) |
1917 | { | |
1918 | /* Has legal address ... */ | |
1919 | if (p != av->top) | |
1920 | { | |
1921 | if (contiguous (av)) | |
1922 | { | |
1923 | assert (((char *) p) >= min_address); | |
1924 | assert (((char *) p + sz) <= ((char *) (av->top))); | |
1925 | } | |
1926 | } | |
1927 | else | |
1928 | { | |
1929 | /* top size is always at least MINSIZE */ | |
1930 | assert ((unsigned long) (sz) >= MINSIZE); | |
1931 | /* top predecessor always marked inuse */ | |
1932 | assert (prev_inuse (p)); | |
1933 | } | |
fa8d436c | 1934 | } |
6c8dbf00 OB |
1935 | else |
1936 | { | |
1937 | /* address is outside main heap */ | |
1938 | if (contiguous (av) && av->top != initial_top (av)) | |
1939 | { | |
1940 | assert (((char *) p) < min_address || ((char *) p) >= max_address); | |
1941 | } | |
1942 | /* chunk is page-aligned */ | |
1943 | assert (((p->prev_size + sz) & (GLRO (dl_pagesize) - 1)) == 0); | |
1944 | /* mem is aligned */ | |
1945 | assert (aligned_OK (chunk2mem (p))); | |
fa8d436c | 1946 | } |
eb406346 UD |
1947 | } |
1948 | ||
fa8d436c | 1949 | /* |
6c8dbf00 OB |
1950 | Properties of free chunks |
1951 | */ | |
ee74a442 | 1952 | |
6c8dbf00 OB |
1953 | static void |
1954 | do_check_free_chunk (mstate av, mchunkptr p) | |
67c94753 | 1955 | { |
6c8dbf00 OB |
1956 | INTERNAL_SIZE_T sz = p->size & ~(PREV_INUSE | NON_MAIN_ARENA); |
1957 | mchunkptr next = chunk_at_offset (p, sz); | |
67c94753 | 1958 | |
6c8dbf00 | 1959 | do_check_chunk (av, p); |
67c94753 | 1960 | |
fa8d436c | 1961 | /* Chunk must claim to be free ... */ |
6c8dbf00 OB |
1962 | assert (!inuse (p)); |
1963 | assert (!chunk_is_mmapped (p)); | |
67c94753 | 1964 | |
fa8d436c | 1965 | /* Unless a special marker, must have OK fields */ |
6c8dbf00 OB |
1966 | if ((unsigned long) (sz) >= MINSIZE) |
1967 | { | |
1968 | assert ((sz & MALLOC_ALIGN_MASK) == 0); | |
1969 | assert (aligned_OK (chunk2mem (p))); | |
1970 | /* ... matching footer field */ | |
1971 | assert (next->prev_size == sz); | |
1972 | /* ... and is fully consolidated */ | |
1973 | assert (prev_inuse (p)); | |
1974 | assert (next == av->top || inuse (next)); | |
1975 | ||
1976 | /* ... and has minimally sane links */ | |
1977 | assert (p->fd->bk == p); | |
1978 | assert (p->bk->fd == p); | |
1979 | } | |
fa8d436c | 1980 | else /* markers are always of size SIZE_SZ */ |
6c8dbf00 | 1981 | assert (sz == SIZE_SZ); |
67c94753 | 1982 | } |
67c94753 | 1983 | |
fa8d436c | 1984 | /* |
6c8dbf00 OB |
1985 | Properties of inuse chunks |
1986 | */ | |
fa8d436c | 1987 | |
6c8dbf00 OB |
1988 | static void |
1989 | do_check_inuse_chunk (mstate av, mchunkptr p) | |
f65fd747 | 1990 | { |
fa8d436c | 1991 | mchunkptr next; |
f65fd747 | 1992 | |
6c8dbf00 | 1993 | do_check_chunk (av, p); |
f65fd747 | 1994 | |
6c8dbf00 | 1995 | if (chunk_is_mmapped (p)) |
fa8d436c | 1996 | return; /* mmapped chunks have no next/prev */ |
ca34d7a7 | 1997 | |
fa8d436c | 1998 | /* Check whether it claims to be in use ... */ |
6c8dbf00 | 1999 | assert (inuse (p)); |
10dc2a90 | 2000 | |
6c8dbf00 | 2001 | next = next_chunk (p); |
10dc2a90 | 2002 | |
fa8d436c | 2003 | /* ... and is surrounded by OK chunks. |
6c8dbf00 OB |
2004 | Since more things can be checked with free chunks than inuse ones, |
2005 | if an inuse chunk borders them and debug is on, it's worth doing them. | |
2006 | */ | |
2007 | if (!prev_inuse (p)) | |
2008 | { | |
2009 | /* Note that we cannot even look at prev unless it is not inuse */ | |
2010 | mchunkptr prv = prev_chunk (p); | |
2011 | assert (next_chunk (prv) == p); | |
2012 | do_check_free_chunk (av, prv); | |
2013 | } | |
fa8d436c | 2014 | |
6c8dbf00 OB |
2015 | if (next == av->top) |
2016 | { | |
2017 | assert (prev_inuse (next)); | |
2018 | assert (chunksize (next) >= MINSIZE); | |
2019 | } | |
2020 | else if (!inuse (next)) | |
2021 | do_check_free_chunk (av, next); | |
10dc2a90 UD |
2022 | } |
2023 | ||
fa8d436c | 2024 | /* |
6c8dbf00 OB |
2025 | Properties of chunks recycled from fastbins |
2026 | */ | |
fa8d436c | 2027 | |
6c8dbf00 OB |
2028 | static void |
2029 | do_check_remalloced_chunk (mstate av, mchunkptr p, INTERNAL_SIZE_T s) | |
10dc2a90 | 2030 | { |
6c8dbf00 | 2031 | INTERNAL_SIZE_T sz = p->size & ~(PREV_INUSE | NON_MAIN_ARENA); |
fa8d436c | 2032 | |
6c8dbf00 OB |
2033 | if (!chunk_is_mmapped (p)) |
2034 | { | |
2035 | assert (av == arena_for_chunk (p)); | |
2036 | if (chunk_non_main_arena (p)) | |
2037 | assert (av != &main_arena); | |
2038 | else | |
2039 | assert (av == &main_arena); | |
2040 | } | |
fa8d436c | 2041 | |
6c8dbf00 | 2042 | do_check_inuse_chunk (av, p); |
fa8d436c UD |
2043 | |
2044 | /* Legal size ... */ | |
6c8dbf00 OB |
2045 | assert ((sz & MALLOC_ALIGN_MASK) == 0); |
2046 | assert ((unsigned long) (sz) >= MINSIZE); | |
fa8d436c | 2047 | /* ... and alignment */ |
6c8dbf00 | 2048 | assert (aligned_OK (chunk2mem (p))); |
fa8d436c | 2049 | /* chunk is less than MINSIZE more than request */ |
6c8dbf00 OB |
2050 | assert ((long) (sz) - (long) (s) >= 0); |
2051 | assert ((long) (sz) - (long) (s + MINSIZE) < 0); | |
10dc2a90 UD |
2052 | } |
2053 | ||
fa8d436c | 2054 | /* |
6c8dbf00 OB |
2055 | Properties of nonrecycled chunks at the point they are malloced |
2056 | */ | |
fa8d436c | 2057 | |
6c8dbf00 OB |
2058 | static void |
2059 | do_check_malloced_chunk (mstate av, mchunkptr p, INTERNAL_SIZE_T s) | |
10dc2a90 | 2060 | { |
fa8d436c | 2061 | /* same as recycled case ... */ |
6c8dbf00 | 2062 | do_check_remalloced_chunk (av, p, s); |
10dc2a90 | 2063 | |
fa8d436c | 2064 | /* |
6c8dbf00 OB |
2065 | ... plus, must obey implementation invariant that prev_inuse is |
2066 | always true of any allocated chunk; i.e., that each allocated | |
2067 | chunk borders either a previously allocated and still in-use | |
2068 | chunk, or the base of its memory arena. This is ensured | |
2069 | by making all allocations from the `lowest' part of any found | |
2070 | chunk. This does not necessarily hold however for chunks | |
2071 | recycled via fastbins. | |
2072 | */ | |
2073 | ||
2074 | assert (prev_inuse (p)); | |
fa8d436c | 2075 | } |
10dc2a90 | 2076 | |
f65fd747 | 2077 | |
fa8d436c | 2078 | /* |
6c8dbf00 | 2079 | Properties of malloc_state. |
f65fd747 | 2080 | |
6c8dbf00 OB |
2081 | This may be useful for debugging malloc, as well as detecting user |
2082 | programmer errors that somehow write into malloc_state. | |
f65fd747 | 2083 | |
6c8dbf00 OB |
2084 | If you are extending or experimenting with this malloc, you can |
2085 | probably figure out how to hack this routine to print out or | |
2086 | display chunk addresses, sizes, bins, and other instrumentation. | |
2087 | */ | |
f65fd747 | 2088 | |
6c8dbf00 OB |
2089 | static void |
2090 | do_check_malloc_state (mstate av) | |
fa8d436c UD |
2091 | { |
2092 | int i; | |
2093 | mchunkptr p; | |
2094 | mchunkptr q; | |
2095 | mbinptr b; | |
fa8d436c UD |
2096 | unsigned int idx; |
2097 | INTERNAL_SIZE_T size; | |
2098 | unsigned long total = 0; | |
2099 | int max_fast_bin; | |
f65fd747 | 2100 | |
fa8d436c | 2101 | /* internal size_t must be no wider than pointer type */ |
6c8dbf00 | 2102 | assert (sizeof (INTERNAL_SIZE_T) <= sizeof (char *)); |
f65fd747 | 2103 | |
fa8d436c | 2104 | /* alignment is a power of 2 */ |
6c8dbf00 | 2105 | assert ((MALLOC_ALIGNMENT & (MALLOC_ALIGNMENT - 1)) == 0); |
f65fd747 | 2106 | |
fa8d436c | 2107 | /* cannot run remaining checks until fully initialized */ |
6c8dbf00 | 2108 | if (av->top == 0 || av->top == initial_top (av)) |
fa8d436c | 2109 | return; |
f65fd747 | 2110 | |
fa8d436c | 2111 | /* pagesize is a power of 2 */ |
6c8dbf00 | 2112 | assert ((GLRO (dl_pagesize) & (GLRO (dl_pagesize) - 1)) == 0); |
f65fd747 | 2113 | |
fa8d436c | 2114 | /* A contiguous main_arena is consistent with sbrk_base. */ |
6c8dbf00 OB |
2115 | if (av == &main_arena && contiguous (av)) |
2116 | assert ((char *) mp_.sbrk_base + av->system_mem == | |
2117 | (char *) av->top + chunksize (av->top)); | |
fa8d436c UD |
2118 | |
2119 | /* properties of fastbins */ | |
2120 | ||
2121 | /* max_fast is in allowed range */ | |
6c8dbf00 OB |
2122 | assert ((get_max_fast () & ~1) <= request2size (MAX_FAST_SIZE)); |
2123 | ||
2124 | max_fast_bin = fastbin_index (get_max_fast ()); | |
2125 | ||
2126 | for (i = 0; i < NFASTBINS; ++i) | |
2127 | { | |
2128 | p = fastbin (av, i); | |
2129 | ||
2130 | /* The following test can only be performed for the main arena. | |
2131 | While mallopt calls malloc_consolidate to get rid of all fast | |
2132 | bins (especially those larger than the new maximum) this does | |
2133 | only happen for the main arena. Trying to do this for any | |
2134 | other arena would mean those arenas have to be locked and | |
2135 | malloc_consolidate be called for them. This is excessive. And | |
2136 | even if this is acceptable to somebody it still cannot solve | |
2137 | the problem completely since if the arena is locked a | |
2138 | concurrent malloc call might create a new arena which then | |
2139 | could use the newly invalid fast bins. */ | |
2140 | ||
2141 | /* all bins past max_fast are empty */ | |
2142 | if (av == &main_arena && i > max_fast_bin) | |
2143 | assert (p == 0); | |
2144 | ||
2145 | while (p != 0) | |
2146 | { | |
2147 | /* each chunk claims to be inuse */ | |
2148 | do_check_inuse_chunk (av, p); | |
2149 | total += chunksize (p); | |
2150 | /* chunk belongs in this bin */ | |
2151 | assert (fastbin_index (chunksize (p)) == i); | |
2152 | p = p->fd; | |
2153 | } | |
fa8d436c | 2154 | } |
fa8d436c UD |
2155 | |
2156 | if (total != 0) | |
6c8dbf00 OB |
2157 | assert (have_fastchunks (av)); |
2158 | else if (!have_fastchunks (av)) | |
2159 | assert (total == 0); | |
fa8d436c UD |
2160 | |
2161 | /* check normal bins */ | |
6c8dbf00 OB |
2162 | for (i = 1; i < NBINS; ++i) |
2163 | { | |
2164 | b = bin_at (av, i); | |
2165 | ||
2166 | /* binmap is accurate (except for bin 1 == unsorted_chunks) */ | |
2167 | if (i >= 2) | |
2168 | { | |
2169 | unsigned int binbit = get_binmap (av, i); | |
2170 | int empty = last (b) == b; | |
2171 | if (!binbit) | |
2172 | assert (empty); | |
2173 | else if (!empty) | |
2174 | assert (binbit); | |
2175 | } | |
2176 | ||
2177 | for (p = last (b); p != b; p = p->bk) | |
2178 | { | |
2179 | /* each chunk claims to be free */ | |
2180 | do_check_free_chunk (av, p); | |
2181 | size = chunksize (p); | |
2182 | total += size; | |
2183 | if (i >= 2) | |
2184 | { | |
2185 | /* chunk belongs in bin */ | |
2186 | idx = bin_index (size); | |
2187 | assert (idx == i); | |
2188 | /* lists are sorted */ | |
2189 | assert (p->bk == b || | |
2190 | (unsigned long) chunksize (p->bk) >= (unsigned long) chunksize (p)); | |
2191 | ||
2192 | if (!in_smallbin_range (size)) | |
2193 | { | |
2194 | if (p->fd_nextsize != NULL) | |
2195 | { | |
2196 | if (p->fd_nextsize == p) | |
2197 | assert (p->bk_nextsize == p); | |
2198 | else | |
2199 | { | |
2200 | if (p->fd_nextsize == first (b)) | |
2201 | assert (chunksize (p) < chunksize (p->fd_nextsize)); | |
2202 | else | |
2203 | assert (chunksize (p) > chunksize (p->fd_nextsize)); | |
2204 | ||
2205 | if (p == first (b)) | |
2206 | assert (chunksize (p) > chunksize (p->bk_nextsize)); | |
2207 | else | |
2208 | assert (chunksize (p) < chunksize (p->bk_nextsize)); | |
2209 | } | |
2210 | } | |
2211 | else | |
2212 | assert (p->bk_nextsize == NULL); | |
2213 | } | |
2214 | } | |
2215 | else if (!in_smallbin_range (size)) | |
2216 | assert (p->fd_nextsize == NULL && p->bk_nextsize == NULL); | |
2217 | /* chunk is followed by a legal chain of inuse chunks */ | |
2218 | for (q = next_chunk (p); | |
2219 | (q != av->top && inuse (q) && | |
2220 | (unsigned long) (chunksize (q)) >= MINSIZE); | |
2221 | q = next_chunk (q)) | |
2222 | do_check_inuse_chunk (av, q); | |
2223 | } | |
fa8d436c | 2224 | } |
f65fd747 | 2225 | |
fa8d436c | 2226 | /* top chunk is OK */ |
6c8dbf00 | 2227 | check_chunk (av, av->top); |
fa8d436c UD |
2228 | } |
2229 | #endif | |
2230 | ||
2231 | ||
2232 | /* ----------------- Support for debugging hooks -------------------- */ | |
2233 | #include "hooks.c" | |
2234 | ||
2235 | ||
2236 | /* ----------- Routines dealing with system allocation -------------- */ | |
2237 | ||
2238 | /* | |
6c8dbf00 OB |
2239 | sysmalloc handles malloc cases requiring more memory from the system. |
2240 | On entry, it is assumed that av->top does not have enough | |
2241 | space to service request for nb bytes, thus requiring that av->top | |
2242 | be extended or replaced. | |
2243 | */ | |
fa8d436c | 2244 | |
6c8dbf00 OB |
2245 | static void * |
2246 | sysmalloc (INTERNAL_SIZE_T nb, mstate av) | |
f65fd747 | 2247 | { |
6c8dbf00 | 2248 | mchunkptr old_top; /* incoming value of av->top */ |
fa8d436c | 2249 | INTERNAL_SIZE_T old_size; /* its size */ |
6c8dbf00 | 2250 | char *old_end; /* its end address */ |
f65fd747 | 2251 | |
6c8dbf00 OB |
2252 | long size; /* arg to first MORECORE or mmap call */ |
2253 | char *brk; /* return value from MORECORE */ | |
f65fd747 | 2254 | |
6c8dbf00 OB |
2255 | long correction; /* arg to 2nd MORECORE call */ |
2256 | char *snd_brk; /* 2nd return val */ | |
f65fd747 | 2257 | |
fa8d436c UD |
2258 | INTERNAL_SIZE_T front_misalign; /* unusable bytes at front of new space */ |
2259 | INTERNAL_SIZE_T end_misalign; /* partial page left at end of new space */ | |
6c8dbf00 | 2260 | char *aligned_brk; /* aligned offset into brk */ |
f65fd747 | 2261 | |
6c8dbf00 OB |
2262 | mchunkptr p; /* the allocated/returned chunk */ |
2263 | mchunkptr remainder; /* remainder from allocation */ | |
2264 | unsigned long remainder_size; /* its size */ | |
fa8d436c | 2265 | |
fa8d436c | 2266 | |
6c8dbf00 OB |
2267 | size_t pagemask = GLRO (dl_pagesize) - 1; |
2268 | bool tried_mmap = false; | |
fa8d436c UD |
2269 | |
2270 | ||
fa8d436c | 2271 | /* |
6c8dbf00 OB |
2272 | If have mmap, and the request size meets the mmap threshold, and |
2273 | the system supports mmap, and there are few enough currently | |
2274 | allocated mmapped regions, try to directly map this request | |
2275 | rather than expanding top. | |
2276 | */ | |
2277 | ||
2278 | if ((unsigned long) (nb) >= (unsigned long) (mp_.mmap_threshold) && | |
2279 | (mp_.n_mmaps < mp_.n_mmaps_max)) | |
2280 | { | |
2281 | char *mm; /* return value from mmap call*/ | |
a9177ff5 | 2282 | |
6c8dbf00 OB |
2283 | try_mmap: |
2284 | /* | |
2285 | Round up size to nearest page. For mmapped chunks, the overhead | |
2286 | is one SIZE_SZ unit larger than for normal chunks, because there | |
2287 | is no following chunk whose prev_size field could be used. | |
2288 | ||
2289 | See the front_misalign handling below, for glibc there is no | |
2290 | need for further alignments unless we have have high alignment. | |
2291 | */ | |
2292 | if (MALLOC_ALIGNMENT == 2 * SIZE_SZ) | |
2293 | size = (nb + SIZE_SZ + pagemask) & ~pagemask; | |
2294 | else | |
2295 | size = (nb + SIZE_SZ + MALLOC_ALIGN_MASK + pagemask) & ~pagemask; | |
2296 | tried_mmap = true; | |
2297 | ||
2298 | /* Don't try if size wraps around 0 */ | |
2299 | if ((unsigned long) (size) > (unsigned long) (nb)) | |
2300 | { | |
2301 | mm = (char *) (MMAP (0, size, PROT_READ | PROT_WRITE, 0)); | |
2302 | ||
2303 | if (mm != MAP_FAILED) | |
2304 | { | |
2305 | /* | |
2306 | The offset to the start of the mmapped region is stored | |
2307 | in the prev_size field of the chunk. This allows us to adjust | |
2308 | returned start address to meet alignment requirements here | |
2309 | and in memalign(), and still be able to compute proper | |
2310 | address argument for later munmap in free() and realloc(). | |
2311 | */ | |
2312 | ||
2313 | if (MALLOC_ALIGNMENT == 2 * SIZE_SZ) | |
2314 | { | |
2315 | /* For glibc, chunk2mem increases the address by 2*SIZE_SZ and | |
2316 | MALLOC_ALIGN_MASK is 2*SIZE_SZ-1. Each mmap'ed area is page | |
2317 | aligned and therefore definitely MALLOC_ALIGN_MASK-aligned. */ | |
2318 | assert (((INTERNAL_SIZE_T) chunk2mem (mm) & MALLOC_ALIGN_MASK) == 0); | |
2319 | front_misalign = 0; | |
2320 | } | |
2321 | else | |
2322 | front_misalign = (INTERNAL_SIZE_T) chunk2mem (mm) & MALLOC_ALIGN_MASK; | |
2323 | if (front_misalign > 0) | |
2324 | { | |
2325 | correction = MALLOC_ALIGNMENT - front_misalign; | |
2326 | p = (mchunkptr) (mm + correction); | |
2327 | p->prev_size = correction; | |
2328 | set_head (p, (size - correction) | IS_MMAPPED); | |
2329 | } | |
2330 | else | |
2331 | { | |
2332 | p = (mchunkptr) mm; | |
2333 | set_head (p, size | IS_MMAPPED); | |
2334 | } | |
2335 | ||
2336 | /* update statistics */ | |
2337 | ||
2338 | int new = atomic_exchange_and_add (&mp_.n_mmaps, 1) + 1; | |
2339 | atomic_max (&mp_.max_n_mmaps, new); | |
2340 | ||
2341 | unsigned long sum; | |
2342 | sum = atomic_exchange_and_add (&mp_.mmapped_mem, size) + size; | |
2343 | atomic_max (&mp_.max_mmapped_mem, sum); | |
2344 | ||
2345 | check_chunk (av, p); | |
2346 | ||
2347 | return chunk2mem (p); | |
2348 | } | |
2349 | } | |
fa8d436c | 2350 | } |
fa8d436c UD |
2351 | |
2352 | /* Record incoming configuration of top */ | |
2353 | ||
6c8dbf00 OB |
2354 | old_top = av->top; |
2355 | old_size = chunksize (old_top); | |
2356 | old_end = (char *) (chunk_at_offset (old_top, old_size)); | |
fa8d436c | 2357 | |
6c8dbf00 | 2358 | brk = snd_brk = (char *) (MORECORE_FAILURE); |
fa8d436c | 2359 | |
a9177ff5 | 2360 | /* |
fa8d436c UD |
2361 | If not the first time through, we require old_size to be |
2362 | at least MINSIZE and to have prev_inuse set. | |
6c8dbf00 | 2363 | */ |
fa8d436c | 2364 | |
6c8dbf00 OB |
2365 | assert ((old_top == initial_top (av) && old_size == 0) || |
2366 | ((unsigned long) (old_size) >= MINSIZE && | |
2367 | prev_inuse (old_top) && | |
2368 | ((unsigned long) old_end & pagemask) == 0)); | |
fa8d436c UD |
2369 | |
2370 | /* Precondition: not enough current space to satisfy nb request */ | |
6c8dbf00 | 2371 | assert ((unsigned long) (old_size) < (unsigned long) (nb + MINSIZE)); |
a9177ff5 | 2372 | |
72f90263 | 2373 | |
6c8dbf00 OB |
2374 | if (av != &main_arena) |
2375 | { | |
2376 | heap_info *old_heap, *heap; | |
2377 | size_t old_heap_size; | |
2378 | ||
2379 | /* First try to extend the current heap. */ | |
2380 | old_heap = heap_for_ptr (old_top); | |
2381 | old_heap_size = old_heap->size; | |
2382 | if ((long) (MINSIZE + nb - old_size) > 0 | |
2383 | && grow_heap (old_heap, MINSIZE + nb - old_size) == 0) | |
2384 | { | |
2385 | av->system_mem += old_heap->size - old_heap_size; | |
2386 | arena_mem += old_heap->size - old_heap_size; | |
2387 | set_head (old_top, (((char *) old_heap + old_heap->size) - (char *) old_top) | |
2388 | | PREV_INUSE); | |
2389 | } | |
2390 | else if ((heap = new_heap (nb + (MINSIZE + sizeof (*heap)), mp_.top_pad))) | |
2391 | { | |
2392 | /* Use a newly allocated heap. */ | |
2393 | heap->ar_ptr = av; | |
2394 | heap->prev = old_heap; | |
2395 | av->system_mem += heap->size; | |
2396 | arena_mem += heap->size; | |
2397 | /* Set up the new top. */ | |
2398 | top (av) = chunk_at_offset (heap, sizeof (*heap)); | |
2399 | set_head (top (av), (heap->size - sizeof (*heap)) | PREV_INUSE); | |
2400 | ||
2401 | /* Setup fencepost and free the old top chunk with a multiple of | |
2402 | MALLOC_ALIGNMENT in size. */ | |
2403 | /* The fencepost takes at least MINSIZE bytes, because it might | |
2404 | become the top chunk again later. Note that a footer is set | |
2405 | up, too, although the chunk is marked in use. */ | |
2406 | old_size = (old_size - MINSIZE) & ~MALLOC_ALIGN_MASK; | |
2407 | set_head (chunk_at_offset (old_top, old_size + 2 * SIZE_SZ), 0 | PREV_INUSE); | |
2408 | if (old_size >= MINSIZE) | |
2409 | { | |
2410 | set_head (chunk_at_offset (old_top, old_size), (2 * SIZE_SZ) | PREV_INUSE); | |
2411 | set_foot (chunk_at_offset (old_top, old_size), (2 * SIZE_SZ)); | |
2412 | set_head (old_top, old_size | PREV_INUSE | NON_MAIN_ARENA); | |
2413 | _int_free (av, old_top, 1); | |
2414 | } | |
2415 | else | |
2416 | { | |
2417 | set_head (old_top, (old_size + 2 * SIZE_SZ) | PREV_INUSE); | |
2418 | set_foot (old_top, (old_size + 2 * SIZE_SZ)); | |
2419 | } | |
2420 | } | |
2421 | else if (!tried_mmap) | |
2422 | /* We can at least try to use to mmap memory. */ | |
2423 | goto try_mmap; | |
fa8d436c | 2424 | } |
6c8dbf00 | 2425 | else /* av == main_arena */ |
fa8d436c | 2426 | |
fa8d436c | 2427 | |
6c8dbf00 OB |
2428 | { /* Request enough space for nb + pad + overhead */ |
2429 | size = nb + mp_.top_pad + MINSIZE; | |
a9177ff5 | 2430 | |
6c8dbf00 OB |
2431 | /* |
2432 | If contiguous, we can subtract out existing space that we hope to | |
2433 | combine with new space. We add it back later only if | |
2434 | we don't actually get contiguous space. | |
2435 | */ | |
a9177ff5 | 2436 | |
6c8dbf00 OB |
2437 | if (contiguous (av)) |
2438 | size -= old_size; | |
fa8d436c | 2439 | |
6c8dbf00 OB |
2440 | /* |
2441 | Round to a multiple of page size. | |
2442 | If MORECORE is not contiguous, this ensures that we only call it | |
2443 | with whole-page arguments. And if MORECORE is contiguous and | |
2444 | this is not first time through, this preserves page-alignment of | |
2445 | previous calls. Otherwise, we correct to page-align below. | |
2446 | */ | |
fa8d436c | 2447 | |
6c8dbf00 | 2448 | size = (size + pagemask) & ~pagemask; |
fa8d436c | 2449 | |
6c8dbf00 OB |
2450 | /* |
2451 | Don't try to call MORECORE if argument is so big as to appear | |
2452 | negative. Note that since mmap takes size_t arg, it may succeed | |
2453 | below even if we cannot call MORECORE. | |
2454 | */ | |
2455 | ||
2456 | if (size > 0) | |
2457 | { | |
2458 | brk = (char *) (MORECORE (size)); | |
2459 | LIBC_PROBE (memory_sbrk_more, 2, brk, size); | |
2460 | } | |
2461 | ||
2462 | if (brk != (char *) (MORECORE_FAILURE)) | |
2463 | { | |
2464 | /* Call the `morecore' hook if necessary. */ | |
2465 | void (*hook) (void) = atomic_forced_read (__after_morecore_hook); | |
2466 | if (__builtin_expect (hook != NULL, 0)) | |
2467 | (*hook)(); | |
2468 | } | |
2469 | else | |
2470 | { | |
2471 | /* | |
2472 | If have mmap, try using it as a backup when MORECORE fails or | |
2473 | cannot be used. This is worth doing on systems that have "holes" in | |
2474 | address space, so sbrk cannot extend to give contiguous space, but | |
2475 | space is available elsewhere. Note that we ignore mmap max count | |
2476 | and threshold limits, since the space will not be used as a | |
2477 | segregated mmap region. | |
2478 | */ | |
2479 | ||
2480 | /* Cannot merge with old top, so add its size back in */ | |
2481 | if (contiguous (av)) | |
2482 | size = (size + old_size + pagemask) & ~pagemask; | |
2483 | ||
2484 | /* If we are relying on mmap as backup, then use larger units */ | |
2485 | if ((unsigned long) (size) < (unsigned long) (MMAP_AS_MORECORE_SIZE)) | |
2486 | size = MMAP_AS_MORECORE_SIZE; | |
2487 | ||
2488 | /* Don't try if size wraps around 0 */ | |
2489 | if ((unsigned long) (size) > (unsigned long) (nb)) | |
2490 | { | |
2491 | char *mbrk = (char *) (MMAP (0, size, PROT_READ | PROT_WRITE, 0)); | |
2492 | ||
2493 | if (mbrk != MAP_FAILED) | |
2494 | { | |
2495 | /* We do not need, and cannot use, another sbrk call to find end */ | |
2496 | brk = mbrk; | |
2497 | snd_brk = brk + size; | |
2498 | ||
2499 | /* | |
2500 | Record that we no longer have a contiguous sbrk region. | |
2501 | After the first time mmap is used as backup, we do not | |
2502 | ever rely on contiguous space since this could incorrectly | |
2503 | bridge regions. | |
2504 | */ | |
2505 | set_noncontiguous (av); | |
2506 | } | |
2507 | } | |
2508 | } | |
2509 | ||
2510 | if (brk != (char *) (MORECORE_FAILURE)) | |
2511 | { | |
2512 | if (mp_.sbrk_base == 0) | |
2513 | mp_.sbrk_base = brk; | |
2514 | av->system_mem += size; | |
2515 | ||
2516 | /* | |
2517 | If MORECORE extends previous space, we can likewise extend top size. | |
2518 | */ | |
2519 | ||
2520 | if (brk == old_end && snd_brk == (char *) (MORECORE_FAILURE)) | |
2521 | set_head (old_top, (size + old_size) | PREV_INUSE); | |
2522 | ||
2523 | else if (contiguous (av) && old_size && brk < old_end) | |
2524 | { | |
2525 | /* Oops! Someone else killed our space.. Can't touch anything. */ | |
2526 | malloc_printerr (3, "break adjusted to free malloc space", brk); | |
2527 | } | |
2528 | ||
2529 | /* | |
2530 | Otherwise, make adjustments: | |
2531 | ||
2532 | * If the first time through or noncontiguous, we need to call sbrk | |
2533 | just to find out where the end of memory lies. | |
2534 | ||
2535 | * We need to ensure that all returned chunks from malloc will meet | |
2536 | MALLOC_ALIGNMENT | |
2537 | ||
2538 | * If there was an intervening foreign sbrk, we need to adjust sbrk | |
2539 | request size to account for fact that we will not be able to | |
2540 | combine new space with existing space in old_top. | |
2541 | ||
2542 | * Almost all systems internally allocate whole pages at a time, in | |
2543 | which case we might as well use the whole last page of request. | |
2544 | So we allocate enough more memory to hit a page boundary now, | |
2545 | which in turn causes future contiguous calls to page-align. | |
2546 | */ | |
2547 | ||
2548 | else | |
2549 | { | |
2550 | front_misalign = 0; | |
2551 | end_misalign = 0; | |
2552 | correction = 0; | |
2553 | aligned_brk = brk; | |
2554 | ||
2555 | /* handle contiguous cases */ | |
2556 | if (contiguous (av)) | |
2557 | { | |
2558 | /* Count foreign sbrk as system_mem. */ | |
2559 | if (old_size) | |
2560 | av->system_mem += brk - old_end; | |
2561 | ||
2562 | /* Guarantee alignment of first new chunk made from this space */ | |
2563 | ||
2564 | front_misalign = (INTERNAL_SIZE_T) chunk2mem (brk) & MALLOC_ALIGN_MASK; | |
2565 | if (front_misalign > 0) | |
2566 | { | |
2567 | /* | |
2568 | Skip over some bytes to arrive at an aligned position. | |
2569 | We don't need to specially mark these wasted front bytes. | |
2570 | They will never be accessed anyway because | |
2571 | prev_inuse of av->top (and any chunk created from its start) | |
2572 | is always true after initialization. | |
2573 | */ | |
2574 | ||
2575 | correction = MALLOC_ALIGNMENT - front_misalign; | |
2576 | aligned_brk += correction; | |
2577 | } | |
2578 | ||
2579 | /* | |
2580 | If this isn't adjacent to existing space, then we will not | |
2581 | be able to merge with old_top space, so must add to 2nd request. | |
2582 | */ | |
2583 | ||
2584 | correction += old_size; | |
2585 | ||
2586 | /* Extend the end address to hit a page boundary */ | |
2587 | end_misalign = (INTERNAL_SIZE_T) (brk + size + correction); | |
2588 | correction += ((end_misalign + pagemask) & ~pagemask) - end_misalign; | |
2589 | ||
2590 | assert (correction >= 0); | |
2591 | snd_brk = (char *) (MORECORE (correction)); | |
2592 | ||
2593 | /* | |
2594 | If can't allocate correction, try to at least find out current | |
2595 | brk. It might be enough to proceed without failing. | |
2596 | ||
2597 | Note that if second sbrk did NOT fail, we assume that space | |
2598 | is contiguous with first sbrk. This is a safe assumption unless | |
2599 | program is multithreaded but doesn't use locks and a foreign sbrk | |
2600 | occurred between our first and second calls. | |
2601 | */ | |
2602 | ||
2603 | if (snd_brk == (char *) (MORECORE_FAILURE)) | |
2604 | { | |
2605 | correction = 0; | |
2606 | snd_brk = (char *) (MORECORE (0)); | |
2607 | } | |
2608 | else | |
2609 | { | |
2610 | /* Call the `morecore' hook if necessary. */ | |
2611 | void (*hook) (void) = atomic_forced_read (__after_morecore_hook); | |
2612 | if (__builtin_expect (hook != NULL, 0)) | |
2613 | (*hook)(); | |
2614 | } | |
2615 | } | |
2616 | ||
2617 | /* handle non-contiguous cases */ | |
2618 | else | |
2619 | { | |
2620 | if (MALLOC_ALIGNMENT == 2 * SIZE_SZ) | |
2621 | /* MORECORE/mmap must correctly align */ | |
2622 | assert (((unsigned long) chunk2mem (brk) & MALLOC_ALIGN_MASK) == 0); | |
2623 | else | |
2624 | { | |
2625 | front_misalign = (INTERNAL_SIZE_T) chunk2mem (brk) & MALLOC_ALIGN_MASK; | |
2626 | if (front_misalign > 0) | |
2627 | { | |
2628 | /* | |
2629 | Skip over some bytes to arrive at an aligned position. | |
2630 | We don't need to specially mark these wasted front bytes. | |
2631 | They will never be accessed anyway because | |
2632 | prev_inuse of av->top (and any chunk created from its start) | |
2633 | is always true after initialization. | |
2634 | */ | |
2635 | ||
2636 | aligned_brk += MALLOC_ALIGNMENT - front_misalign; | |
2637 | } | |
2638 | } | |
2639 | ||
2640 | /* Find out current end of memory */ | |
2641 | if (snd_brk == (char *) (MORECORE_FAILURE)) | |
2642 | { | |
2643 | snd_brk = (char *) (MORECORE (0)); | |
2644 | } | |
2645 | } | |
2646 | ||
2647 | /* Adjust top based on results of second sbrk */ | |
2648 | if (snd_brk != (char *) (MORECORE_FAILURE)) | |
2649 | { | |
2650 | av->top = (mchunkptr) aligned_brk; | |
2651 | set_head (av->top, (snd_brk - aligned_brk + correction) | PREV_INUSE); | |
2652 | av->system_mem += correction; | |
2653 | ||
2654 | /* | |
2655 | If not the first time through, we either have a | |
2656 | gap due to foreign sbrk or a non-contiguous region. Insert a | |
2657 | double fencepost at old_top to prevent consolidation with space | |
2658 | we don't own. These fenceposts are artificial chunks that are | |
2659 | marked as inuse and are in any case too small to use. We need | |
2660 | two to make sizes and alignments work out. | |
2661 | */ | |
2662 | ||
2663 | if (old_size != 0) | |
2664 | { | |
2665 | /* | |
2666 | Shrink old_top to insert fenceposts, keeping size a | |
2667 | multiple of MALLOC_ALIGNMENT. We know there is at least | |
2668 | enough space in old_top to do this. | |
2669 | */ | |
2670 | old_size = (old_size - 4 * SIZE_SZ) & ~MALLOC_ALIGN_MASK; | |
2671 | set_head (old_top, old_size | PREV_INUSE); | |
2672 | ||
2673 | /* | |
2674 | Note that the following assignments completely overwrite | |
2675 | old_top when old_size was previously MINSIZE. This is | |
2676 | intentional. We need the fencepost, even if old_top otherwise gets | |
2677 | lost. | |
2678 | */ | |
2679 | chunk_at_offset (old_top, old_size)->size = | |
2680 | (2 * SIZE_SZ) | PREV_INUSE; | |
2681 | ||
2682 | chunk_at_offset (old_top, old_size + 2 * SIZE_SZ)->size = | |
2683 | (2 * SIZE_SZ) | PREV_INUSE; | |
2684 | ||
2685 | /* If possible, release the rest. */ | |
2686 | if (old_size >= MINSIZE) | |
2687 | { | |
2688 | _int_free (av, old_top, 1); | |
2689 | } | |
2690 | } | |
2691 | } | |
2692 | } | |
2693 | } | |
2694 | } /* if (av != &main_arena) */ | |
2695 | ||
2696 | if ((unsigned long) av->system_mem > (unsigned long) (av->max_system_mem)) | |
fa8d436c | 2697 | av->max_system_mem = av->system_mem; |
6c8dbf00 | 2698 | check_malloc_state (av); |
a9177ff5 | 2699 | |
fa8d436c UD |
2700 | /* finally, do the allocation */ |
2701 | p = av->top; | |
6c8dbf00 | 2702 | size = chunksize (p); |
fa8d436c UD |
2703 | |
2704 | /* check that one of the above allocation paths succeeded */ | |
6c8dbf00 OB |
2705 | if ((unsigned long) (size) >= (unsigned long) (nb + MINSIZE)) |
2706 | { | |
2707 | remainder_size = size - nb; | |
2708 | remainder = chunk_at_offset (p, nb); | |
2709 | av->top = remainder; | |
2710 | set_head (p, nb | PREV_INUSE | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
2711 | set_head (remainder, remainder_size | PREV_INUSE); | |
2712 | check_malloced_chunk (av, p, nb); | |
2713 | return chunk2mem (p); | |
2714 | } | |
fa8d436c UD |
2715 | |
2716 | /* catch all failure paths */ | |
8e58439c | 2717 | __set_errno (ENOMEM); |
fa8d436c UD |
2718 | return 0; |
2719 | } | |
2720 | ||
2721 | ||
2722 | /* | |
6c8dbf00 OB |
2723 | systrim is an inverse of sorts to sysmalloc. It gives memory back |
2724 | to the system (via negative arguments to sbrk) if there is unused | |
2725 | memory at the `high' end of the malloc pool. It is called | |
2726 | automatically by free() when top space exceeds the trim | |
2727 | threshold. It is also called by the public malloc_trim routine. It | |
2728 | returns 1 if it actually released any memory, else 0. | |
2729 | */ | |
fa8d436c | 2730 | |
6c8dbf00 OB |
2731 | static int |
2732 | systrim (size_t pad, mstate av) | |
fa8d436c | 2733 | { |
6c8dbf00 OB |
2734 | long top_size; /* Amount of top-most memory */ |
2735 | long extra; /* Amount to release */ | |
2736 | long released; /* Amount actually released */ | |
2737 | char *current_brk; /* address returned by pre-check sbrk call */ | |
2738 | char *new_brk; /* address returned by post-check sbrk call */ | |
fa8d436c | 2739 | size_t pagesz; |
6c8dbf00 | 2740 | long top_area; |
fa8d436c | 2741 | |
6c8dbf00 OB |
2742 | pagesz = GLRO (dl_pagesize); |
2743 | top_size = chunksize (av->top); | |
a9177ff5 | 2744 | |
4b5b548c FS |
2745 | top_area = top_size - MINSIZE - 1; |
2746 | if (top_area <= pad) | |
2747 | return 0; | |
2748 | ||
fa8d436c | 2749 | /* Release in pagesize units, keeping at least one page */ |
4b5b548c | 2750 | extra = (top_area - pad) & ~(pagesz - 1); |
a9177ff5 | 2751 | |
51a7380b WN |
2752 | if (extra == 0) |
2753 | return 0; | |
2754 | ||
4b5b548c | 2755 | /* |
6c8dbf00 OB |
2756 | Only proceed if end of memory is where we last set it. |
2757 | This avoids problems if there were foreign sbrk calls. | |
2758 | */ | |
2759 | current_brk = (char *) (MORECORE (0)); | |
2760 | if (current_brk == (char *) (av->top) + top_size) | |
2761 | { | |
2762 | /* | |
2763 | Attempt to release memory. We ignore MORECORE return value, | |
2764 | and instead call again to find out where new end of memory is. | |
2765 | This avoids problems if first call releases less than we asked, | |
2766 | of if failure somehow altered brk value. (We could still | |
2767 | encounter problems if it altered brk in some very bad way, | |
2768 | but the only thing we can do is adjust anyway, which will cause | |
2769 | some downstream failure.) | |
2770 | */ | |
2771 | ||
2772 | MORECORE (-extra); | |
2773 | /* Call the `morecore' hook if necessary. */ | |
2774 | void (*hook) (void) = atomic_forced_read (__after_morecore_hook); | |
2775 | if (__builtin_expect (hook != NULL, 0)) | |
2776 | (*hook)(); | |
2777 | new_brk = (char *) (MORECORE (0)); | |
2778 | ||
2779 | LIBC_PROBE (memory_sbrk_less, 2, new_brk, extra); | |
2780 | ||
2781 | if (new_brk != (char *) MORECORE_FAILURE) | |
2782 | { | |
2783 | released = (long) (current_brk - new_brk); | |
2784 | ||
2785 | if (released != 0) | |
2786 | { | |
2787 | /* Success. Adjust top. */ | |
2788 | av->system_mem -= released; | |
2789 | set_head (av->top, (top_size - released) | PREV_INUSE); | |
2790 | check_malloc_state (av); | |
2791 | return 1; | |
2792 | } | |
2793 | } | |
fa8d436c | 2794 | } |
fa8d436c | 2795 | return 0; |
f65fd747 UD |
2796 | } |
2797 | ||
431c33c0 UD |
2798 | static void |
2799 | internal_function | |
6c8dbf00 | 2800 | munmap_chunk (mchunkptr p) |
f65fd747 | 2801 | { |
6c8dbf00 | 2802 | INTERNAL_SIZE_T size = chunksize (p); |
f65fd747 | 2803 | |
6c8dbf00 | 2804 | assert (chunk_is_mmapped (p)); |
8e635611 UD |
2805 | |
2806 | uintptr_t block = (uintptr_t) p - p->prev_size; | |
2807 | size_t total_size = p->prev_size + size; | |
2808 | /* Unfortunately we have to do the compilers job by hand here. Normally | |
2809 | we would test BLOCK and TOTAL-SIZE separately for compliance with the | |
2810 | page size. But gcc does not recognize the optimization possibility | |
2811 | (in the moment at least) so we combine the two values into one before | |
2812 | the bit test. */ | |
6c8dbf00 | 2813 | if (__builtin_expect (((block | total_size) & (GLRO (dl_pagesize) - 1)) != 0, 0)) |
8e635611 UD |
2814 | { |
2815 | malloc_printerr (check_action, "munmap_chunk(): invalid pointer", | |
6c8dbf00 | 2816 | chunk2mem (p)); |
8e635611 UD |
2817 | return; |
2818 | } | |
f65fd747 | 2819 | |
c6e4925d OB |
2820 | atomic_decrement (&mp_.n_mmaps); |
2821 | atomic_add (&mp_.mmapped_mem, -total_size); | |
f65fd747 | 2822 | |
6ef76f3b UD |
2823 | /* If munmap failed the process virtual memory address space is in a |
2824 | bad shape. Just leave the block hanging around, the process will | |
2825 | terminate shortly anyway since not much can be done. */ | |
6c8dbf00 | 2826 | __munmap ((char *) block, total_size); |
f65fd747 UD |
2827 | } |
2828 | ||
2829 | #if HAVE_MREMAP | |
2830 | ||
431c33c0 UD |
2831 | static mchunkptr |
2832 | internal_function | |
6c8dbf00 | 2833 | mremap_chunk (mchunkptr p, size_t new_size) |
f65fd747 | 2834 | { |
6c8dbf00 | 2835 | size_t page_mask = GLRO (dl_pagesize) - 1; |
f65fd747 | 2836 | INTERNAL_SIZE_T offset = p->prev_size; |
6c8dbf00 | 2837 | INTERNAL_SIZE_T size = chunksize (p); |
f65fd747 UD |
2838 | char *cp; |
2839 | ||
6c8dbf00 OB |
2840 | assert (chunk_is_mmapped (p)); |
2841 | assert (((size + offset) & (GLRO (dl_pagesize) - 1)) == 0); | |
f65fd747 UD |
2842 | |
2843 | /* Note the extra SIZE_SZ overhead as in mmap_chunk(). */ | |
2844 | new_size = (new_size + offset + SIZE_SZ + page_mask) & ~page_mask; | |
2845 | ||
68f3802d UD |
2846 | /* No need to remap if the number of pages does not change. */ |
2847 | if (size + offset == new_size) | |
2848 | return p; | |
2849 | ||
6c8dbf00 OB |
2850 | cp = (char *) __mremap ((char *) p - offset, size + offset, new_size, |
2851 | MREMAP_MAYMOVE); | |
f65fd747 | 2852 | |
6c8dbf00 OB |
2853 | if (cp == MAP_FAILED) |
2854 | return 0; | |
f65fd747 | 2855 | |
6c8dbf00 | 2856 | p = (mchunkptr) (cp + offset); |
f65fd747 | 2857 | |
6c8dbf00 | 2858 | assert (aligned_OK (chunk2mem (p))); |
f65fd747 | 2859 | |
6c8dbf00 OB |
2860 | assert ((p->prev_size == offset)); |
2861 | set_head (p, (new_size - offset) | IS_MMAPPED); | |
f65fd747 | 2862 | |
c6e4925d OB |
2863 | INTERNAL_SIZE_T new; |
2864 | new = atomic_exchange_and_add (&mp_.mmapped_mem, new_size - size - offset) | |
6c8dbf00 | 2865 | + new_size - size - offset; |
c6e4925d | 2866 | atomic_max (&mp_.max_mmapped_mem, new); |
f65fd747 UD |
2867 | return p; |
2868 | } | |
f65fd747 UD |
2869 | #endif /* HAVE_MREMAP */ |
2870 | ||
fa8d436c | 2871 | /*------------------------ Public wrappers. --------------------------------*/ |
f65fd747 | 2872 | |
6c8dbf00 OB |
2873 | void * |
2874 | __libc_malloc (size_t bytes) | |
fa8d436c UD |
2875 | { |
2876 | mstate ar_ptr; | |
22a89187 | 2877 | void *victim; |
f65fd747 | 2878 | |
a222d91a | 2879 | void *(*hook) (size_t, const void *) |
f3eeb3fc | 2880 | = atomic_forced_read (__malloc_hook); |
bfacf1af | 2881 | if (__builtin_expect (hook != NULL, 0)) |
fa8d436c | 2882 | return (*hook)(bytes, RETURN_ADDRESS (0)); |
f65fd747 | 2883 | |
6c8dbf00 | 2884 | arena_lookup (ar_ptr); |
425ce2ed | 2885 | |
6c8dbf00 OB |
2886 | arena_lock (ar_ptr, bytes); |
2887 | if (!ar_ptr) | |
f65fd747 | 2888 | return 0; |
6c8dbf00 OB |
2889 | |
2890 | victim = _int_malloc (ar_ptr, bytes); | |
2891 | if (!victim) | |
2892 | { | |
2893 | LIBC_PROBE (memory_malloc_retry, 1, bytes); | |
2894 | ar_ptr = arena_get_retry (ar_ptr, bytes); | |
2895 | if (__builtin_expect (ar_ptr != NULL, 1)) | |
2896 | { | |
2897 | victim = _int_malloc (ar_ptr, bytes); | |
2898 | (void) mutex_unlock (&ar_ptr->mutex); | |
2899 | } | |
60f0e64b | 2900 | } |
6c8dbf00 OB |
2901 | else |
2902 | (void) mutex_unlock (&ar_ptr->mutex); | |
2903 | assert (!victim || chunk_is_mmapped (mem2chunk (victim)) || | |
2904 | ar_ptr == arena_for_chunk (mem2chunk (victim))); | |
fa8d436c | 2905 | return victim; |
f65fd747 | 2906 | } |
6c8dbf00 | 2907 | libc_hidden_def (__libc_malloc) |
f65fd747 | 2908 | |
fa8d436c | 2909 | void |
6c8dbf00 | 2910 | __libc_free (void *mem) |
f65fd747 | 2911 | { |
fa8d436c UD |
2912 | mstate ar_ptr; |
2913 | mchunkptr p; /* chunk corresponding to mem */ | |
2914 | ||
a222d91a | 2915 | void (*hook) (void *, const void *) |
f3eeb3fc | 2916 | = atomic_forced_read (__free_hook); |
6c8dbf00 OB |
2917 | if (__builtin_expect (hook != NULL, 0)) |
2918 | { | |
2919 | (*hook)(mem, RETURN_ADDRESS (0)); | |
2920 | return; | |
2921 | } | |
f65fd747 | 2922 | |
fa8d436c UD |
2923 | if (mem == 0) /* free(0) has no effect */ |
2924 | return; | |
f65fd747 | 2925 | |
6c8dbf00 | 2926 | p = mem2chunk (mem); |
f65fd747 | 2927 | |
6c8dbf00 OB |
2928 | if (chunk_is_mmapped (p)) /* release mmapped memory. */ |
2929 | { | |
2930 | /* see if the dynamic brk/mmap threshold needs adjusting */ | |
2931 | if (!mp_.no_dyn_threshold | |
2932 | && p->size > mp_.mmap_threshold | |
2933 | && p->size <= DEFAULT_MMAP_THRESHOLD_MAX) | |
2934 | { | |
2935 | mp_.mmap_threshold = chunksize (p); | |
2936 | mp_.trim_threshold = 2 * mp_.mmap_threshold; | |
2937 | LIBC_PROBE (memory_mallopt_free_dyn_thresholds, 2, | |
2938 | mp_.mmap_threshold, mp_.trim_threshold); | |
2939 | } | |
2940 | munmap_chunk (p); | |
2941 | return; | |
2942 | } | |
f65fd747 | 2943 | |
6c8dbf00 OB |
2944 | ar_ptr = arena_for_chunk (p); |
2945 | _int_free (ar_ptr, p, 0); | |
f65fd747 | 2946 | } |
3b49edc0 | 2947 | libc_hidden_def (__libc_free) |
f65fd747 | 2948 | |
6c8dbf00 OB |
2949 | void * |
2950 | __libc_realloc (void *oldmem, size_t bytes) | |
f65fd747 | 2951 | { |
fa8d436c | 2952 | mstate ar_ptr; |
6c8dbf00 | 2953 | INTERNAL_SIZE_T nb; /* padded request size */ |
f65fd747 | 2954 | |
6c8dbf00 | 2955 | void *newp; /* chunk to return */ |
f65fd747 | 2956 | |
a222d91a | 2957 | void *(*hook) (void *, size_t, const void *) = |
f3eeb3fc | 2958 | atomic_forced_read (__realloc_hook); |
bfacf1af | 2959 | if (__builtin_expect (hook != NULL, 0)) |
fa8d436c | 2960 | return (*hook)(oldmem, bytes, RETURN_ADDRESS (0)); |
f65fd747 | 2961 | |
fa8d436c | 2962 | #if REALLOC_ZERO_BYTES_FREES |
6c8dbf00 OB |
2963 | if (bytes == 0 && oldmem != NULL) |
2964 | { | |
2965 | __libc_free (oldmem); return 0; | |
2966 | } | |
f65fd747 | 2967 | #endif |
f65fd747 | 2968 | |
fa8d436c | 2969 | /* realloc of null is supposed to be same as malloc */ |
6c8dbf00 OB |
2970 | if (oldmem == 0) |
2971 | return __libc_malloc (bytes); | |
f65fd747 | 2972 | |
78ac92ad | 2973 | /* chunk corresponding to oldmem */ |
6c8dbf00 | 2974 | const mchunkptr oldp = mem2chunk (oldmem); |
78ac92ad | 2975 | /* its size */ |
6c8dbf00 | 2976 | const INTERNAL_SIZE_T oldsize = chunksize (oldp); |
f65fd747 | 2977 | |
dc165f7b UD |
2978 | /* Little security check which won't hurt performance: the |
2979 | allocator never wrapps around at the end of the address space. | |
2980 | Therefore we can exclude some size values which might appear | |
2981 | here by accident or by "design" from some intruder. */ | |
2982 | if (__builtin_expect ((uintptr_t) oldp > (uintptr_t) -oldsize, 0) | |
073f560e | 2983 | || __builtin_expect (misaligned_chunk (oldp), 0)) |
dc165f7b UD |
2984 | { |
2985 | malloc_printerr (check_action, "realloc(): invalid pointer", oldmem); | |
2986 | return NULL; | |
2987 | } | |
2988 | ||
6c8dbf00 | 2989 | checked_request2size (bytes, nb); |
f65fd747 | 2990 | |
6c8dbf00 OB |
2991 | if (chunk_is_mmapped (oldp)) |
2992 | { | |
2993 | void *newmem; | |
f65fd747 | 2994 | |
fa8d436c | 2995 | #if HAVE_MREMAP |
6c8dbf00 OB |
2996 | newp = mremap_chunk (oldp, nb); |
2997 | if (newp) | |
2998 | return chunk2mem (newp); | |
f65fd747 | 2999 | #endif |
6c8dbf00 OB |
3000 | /* Note the extra SIZE_SZ overhead. */ |
3001 | if (oldsize - SIZE_SZ >= nb) | |
3002 | return oldmem; /* do nothing */ | |
3003 | ||
3004 | /* Must alloc, copy, free. */ | |
3005 | newmem = __libc_malloc (bytes); | |
3006 | if (newmem == 0) | |
3007 | return 0; /* propagate failure */ | |
fa8d436c | 3008 | |
6c8dbf00 OB |
3009 | memcpy (newmem, oldmem, oldsize - 2 * SIZE_SZ); |
3010 | munmap_chunk (oldp); | |
3011 | return newmem; | |
3012 | } | |
3013 | ||
3014 | ar_ptr = arena_for_chunk (oldp); | |
6c8dbf00 | 3015 | (void) mutex_lock (&ar_ptr->mutex); |
f65fd747 | 3016 | |
f65fd747 | 3017 | |
6c8dbf00 | 3018 | newp = _int_realloc (ar_ptr, oldp, oldsize, nb); |
f65fd747 | 3019 | |
6c8dbf00 OB |
3020 | (void) mutex_unlock (&ar_ptr->mutex); |
3021 | assert (!newp || chunk_is_mmapped (mem2chunk (newp)) || | |
3022 | ar_ptr == arena_for_chunk (mem2chunk (newp))); | |
07014fca UD |
3023 | |
3024 | if (newp == NULL) | |
3025 | { | |
3026 | /* Try harder to allocate memory in other arenas. */ | |
35fed6f1 | 3027 | LIBC_PROBE (memory_realloc_retry, 2, bytes, oldmem); |
6c8dbf00 | 3028 | newp = __libc_malloc (bytes); |
07014fca | 3029 | if (newp != NULL) |
6c8dbf00 OB |
3030 | { |
3031 | memcpy (newp, oldmem, oldsize - SIZE_SZ); | |
3032 | _int_free (ar_ptr, oldp, 0); | |
3033 | } | |
07014fca UD |
3034 | } |
3035 | ||
fa8d436c UD |
3036 | return newp; |
3037 | } | |
3b49edc0 | 3038 | libc_hidden_def (__libc_realloc) |
f65fd747 | 3039 | |
6c8dbf00 OB |
3040 | void * |
3041 | __libc_memalign (size_t alignment, size_t bytes) | |
10ad46bc OB |
3042 | { |
3043 | void *address = RETURN_ADDRESS (0); | |
3044 | return _mid_memalign (alignment, bytes, address); | |
3045 | } | |
3046 | ||
3047 | static void * | |
3048 | _mid_memalign (size_t alignment, size_t bytes, void *address) | |
fa8d436c UD |
3049 | { |
3050 | mstate ar_ptr; | |
22a89187 | 3051 | void *p; |
f65fd747 | 3052 | |
a222d91a | 3053 | void *(*hook) (size_t, size_t, const void *) = |
f3eeb3fc | 3054 | atomic_forced_read (__memalign_hook); |
bfacf1af | 3055 | if (__builtin_expect (hook != NULL, 0)) |
10ad46bc | 3056 | return (*hook)(alignment, bytes, address); |
f65fd747 | 3057 | |
10ad46bc | 3058 | /* If we need less alignment than we give anyway, just relay to malloc. */ |
6c8dbf00 OB |
3059 | if (alignment <= MALLOC_ALIGNMENT) |
3060 | return __libc_malloc (bytes); | |
1228ed5c | 3061 | |
fa8d436c | 3062 | /* Otherwise, ensure that it is at least a minimum chunk size */ |
6c8dbf00 OB |
3063 | if (alignment < MINSIZE) |
3064 | alignment = MINSIZE; | |
f65fd747 | 3065 | |
a56ee40b WN |
3066 | /* If the alignment is greater than SIZE_MAX / 2 + 1 it cannot be a |
3067 | power of 2 and will cause overflow in the check below. */ | |
3068 | if (alignment > SIZE_MAX / 2 + 1) | |
3069 | { | |
3070 | __set_errno (EINVAL); | |
3071 | return 0; | |
3072 | } | |
3073 | ||
b73ed247 WN |
3074 | /* Check for overflow. */ |
3075 | if (bytes > SIZE_MAX - alignment - MINSIZE) | |
3076 | { | |
3077 | __set_errno (ENOMEM); | |
3078 | return 0; | |
3079 | } | |
3080 | ||
10ad46bc OB |
3081 | |
3082 | /* Make sure alignment is power of 2. */ | |
6c8dbf00 OB |
3083 | if (!powerof2 (alignment)) |
3084 | { | |
3085 | size_t a = MALLOC_ALIGNMENT * 2; | |
3086 | while (a < alignment) | |
3087 | a <<= 1; | |
3088 | alignment = a; | |
3089 | } | |
10ad46bc | 3090 | |
6c8dbf00 OB |
3091 | arena_get (ar_ptr, bytes + alignment + MINSIZE); |
3092 | if (!ar_ptr) | |
fa8d436c | 3093 | return 0; |
6c8dbf00 OB |
3094 | |
3095 | p = _int_memalign (ar_ptr, alignment, bytes); | |
3096 | if (!p) | |
3097 | { | |
3098 | LIBC_PROBE (memory_memalign_retry, 2, bytes, alignment); | |
3099 | ar_ptr = arena_get_retry (ar_ptr, bytes); | |
3100 | if (__builtin_expect (ar_ptr != NULL, 1)) | |
3101 | { | |
3102 | p = _int_memalign (ar_ptr, alignment, bytes); | |
3103 | (void) mutex_unlock (&ar_ptr->mutex); | |
3104 | } | |
f65fd747 | 3105 | } |
6c8dbf00 OB |
3106 | else |
3107 | (void) mutex_unlock (&ar_ptr->mutex); | |
3108 | assert (!p || chunk_is_mmapped (mem2chunk (p)) || | |
3109 | ar_ptr == arena_for_chunk (mem2chunk (p))); | |
fa8d436c | 3110 | return p; |
f65fd747 | 3111 | } |
380d7e87 | 3112 | /* For ISO C11. */ |
3b49edc0 UD |
3113 | weak_alias (__libc_memalign, aligned_alloc) |
3114 | libc_hidden_def (__libc_memalign) | |
f65fd747 | 3115 | |
6c8dbf00 OB |
3116 | void * |
3117 | __libc_valloc (size_t bytes) | |
fa8d436c | 3118 | { |
6c8dbf00 | 3119 | if (__malloc_initialized < 0) |
fa8d436c | 3120 | ptmalloc_init (); |
8088488d | 3121 | |
10ad46bc | 3122 | void *address = RETURN_ADDRESS (0); |
6c8dbf00 | 3123 | size_t pagesz = GLRO (dl_pagesize); |
10ad46bc | 3124 | return _mid_memalign (pagesz, bytes, address); |
fa8d436c | 3125 | } |
f65fd747 | 3126 | |
6c8dbf00 OB |
3127 | void * |
3128 | __libc_pvalloc (size_t bytes) | |
fa8d436c | 3129 | { |
6c8dbf00 | 3130 | if (__malloc_initialized < 0) |
fa8d436c | 3131 | ptmalloc_init (); |
8088488d | 3132 | |
10ad46bc | 3133 | void *address = RETURN_ADDRESS (0); |
6c8dbf00 OB |
3134 | size_t pagesz = GLRO (dl_pagesize); |
3135 | size_t page_mask = GLRO (dl_pagesize) - 1; | |
dba38551 UD |
3136 | size_t rounded_bytes = (bytes + page_mask) & ~(page_mask); |
3137 | ||
1159a193 | 3138 | /* Check for overflow. */ |
6c8dbf00 | 3139 | if (bytes > SIZE_MAX - 2 * pagesz - MINSIZE) |
1159a193 WN |
3140 | { |
3141 | __set_errno (ENOMEM); | |
3142 | return 0; | |
3143 | } | |
3144 | ||
10ad46bc | 3145 | return _mid_memalign (pagesz, rounded_bytes, address); |
fa8d436c | 3146 | } |
f65fd747 | 3147 | |
6c8dbf00 OB |
3148 | void * |
3149 | __libc_calloc (size_t n, size_t elem_size) | |
f65fd747 | 3150 | { |
d6285c9f CD |
3151 | mstate av; |
3152 | mchunkptr oldtop, p; | |
3153 | INTERNAL_SIZE_T bytes, sz, csz, oldtopsize; | |
6c8dbf00 | 3154 | void *mem; |
d6285c9f CD |
3155 | unsigned long clearsize; |
3156 | unsigned long nclears; | |
3157 | INTERNAL_SIZE_T *d; | |
0950889b UD |
3158 | |
3159 | /* size_t is unsigned so the behavior on overflow is defined. */ | |
3160 | bytes = n * elem_size; | |
d9af917d UD |
3161 | #define HALF_INTERNAL_SIZE_T \ |
3162 | (((INTERNAL_SIZE_T) 1) << (8 * sizeof (INTERNAL_SIZE_T) / 2)) | |
6c8dbf00 OB |
3163 | if (__builtin_expect ((n | elem_size) >= HALF_INTERNAL_SIZE_T, 0)) |
3164 | { | |
3165 | if (elem_size != 0 && bytes / elem_size != n) | |
3166 | { | |
3167 | __set_errno (ENOMEM); | |
3168 | return 0; | |
3169 | } | |
d9af917d | 3170 | } |
0950889b | 3171 | |
a222d91a | 3172 | void *(*hook) (size_t, const void *) = |
f3eeb3fc | 3173 | atomic_forced_read (__malloc_hook); |
6c8dbf00 OB |
3174 | if (__builtin_expect (hook != NULL, 0)) |
3175 | { | |
d6285c9f CD |
3176 | sz = bytes; |
3177 | mem = (*hook)(sz, RETURN_ADDRESS (0)); | |
3178 | if (mem == 0) | |
3179 | return 0; | |
3180 | ||
3181 | return memset (mem, 0, sz); | |
7799b7b3 | 3182 | } |
f65fd747 | 3183 | |
d6285c9f CD |
3184 | sz = bytes; |
3185 | ||
3186 | arena_get (av, sz); | |
3187 | if (!av) | |
4248f0da | 3188 | return 0; |
f65fd747 | 3189 | |
d6285c9f CD |
3190 | /* Check if we hand out the top chunk, in which case there may be no |
3191 | need to clear. */ | |
3192 | #if MORECORE_CLEARS | |
3193 | oldtop = top (av); | |
3194 | oldtopsize = chunksize (top (av)); | |
3195 | # if MORECORE_CLEARS < 2 | |
3196 | /* Only newly allocated memory is guaranteed to be cleared. */ | |
3197 | if (av == &main_arena && | |
3198 | oldtopsize < mp_.sbrk_base + av->max_system_mem - (char *) oldtop) | |
3199 | oldtopsize = (mp_.sbrk_base + av->max_system_mem - (char *) oldtop); | |
3200 | # endif | |
3201 | if (av != &main_arena) | |
3202 | { | |
3203 | heap_info *heap = heap_for_ptr (oldtop); | |
3204 | if (oldtopsize < (char *) heap + heap->mprotect_size - (char *) oldtop) | |
3205 | oldtopsize = (char *) heap + heap->mprotect_size - (char *) oldtop; | |
3206 | } | |
3207 | #endif | |
3208 | mem = _int_malloc (av, sz); | |
3209 | ||
3210 | ||
3211 | assert (!mem || chunk_is_mmapped (mem2chunk (mem)) || | |
3212 | av == arena_for_chunk (mem2chunk (mem))); | |
3213 | ||
3214 | if (mem == 0) | |
3215 | { | |
3216 | LIBC_PROBE (memory_calloc_retry, 1, sz); | |
3217 | av = arena_get_retry (av, sz); | |
3218 | if (__builtin_expect (av != NULL, 1)) | |
3219 | { | |
3220 | mem = _int_malloc (av, sz); | |
3221 | (void) mutex_unlock (&av->mutex); | |
3222 | } | |
3223 | if (mem == 0) | |
3224 | return 0; | |
3225 | } | |
3226 | else | |
3227 | (void) mutex_unlock (&av->mutex); | |
3228 | p = mem2chunk (mem); | |
3229 | ||
3230 | /* Two optional cases in which clearing not necessary */ | |
3231 | if (chunk_is_mmapped (p)) | |
3232 | { | |
3233 | if (__builtin_expect (perturb_byte, 0)) | |
3234 | return memset (mem, 0, sz); | |
3235 | ||
3236 | return mem; | |
3237 | } | |
3238 | ||
3239 | csz = chunksize (p); | |
3240 | ||
3241 | #if MORECORE_CLEARS | |
3242 | if (perturb_byte == 0 && (p == oldtop && csz > oldtopsize)) | |
3243 | { | |
3244 | /* clear only the bytes from non-freshly-sbrked memory */ | |
3245 | csz = oldtopsize; | |
3246 | } | |
3247 | #endif | |
3248 | ||
3249 | /* Unroll clear of <= 36 bytes (72 if 8byte sizes). We know that | |
3250 | contents have an odd number of INTERNAL_SIZE_T-sized words; | |
3251 | minimally 3. */ | |
3252 | d = (INTERNAL_SIZE_T *) mem; | |
3253 | clearsize = csz - SIZE_SZ; | |
3254 | nclears = clearsize / sizeof (INTERNAL_SIZE_T); | |
3255 | assert (nclears >= 3); | |
3256 | ||
3257 | if (nclears > 9) | |
3258 | return memset (d, 0, clearsize); | |
3259 | ||
3260 | else | |
3261 | { | |
3262 | *(d + 0) = 0; | |
3263 | *(d + 1) = 0; | |
3264 | *(d + 2) = 0; | |
3265 | if (nclears > 4) | |
3266 | { | |
3267 | *(d + 3) = 0; | |
3268 | *(d + 4) = 0; | |
3269 | if (nclears > 6) | |
3270 | { | |
3271 | *(d + 5) = 0; | |
3272 | *(d + 6) = 0; | |
3273 | if (nclears > 8) | |
3274 | { | |
3275 | *(d + 7) = 0; | |
3276 | *(d + 8) = 0; | |
3277 | } | |
3278 | } | |
3279 | } | |
3280 | } | |
3281 | ||
3282 | return mem; | |
fa8d436c | 3283 | } |
f65fd747 | 3284 | |
f65fd747 | 3285 | /* |
6c8dbf00 OB |
3286 | ------------------------------ malloc ------------------------------ |
3287 | */ | |
f65fd747 | 3288 | |
6c8dbf00 OB |
3289 | static void * |
3290 | _int_malloc (mstate av, size_t bytes) | |
f65fd747 | 3291 | { |
fa8d436c | 3292 | INTERNAL_SIZE_T nb; /* normalized request size */ |
6c8dbf00 OB |
3293 | unsigned int idx; /* associated bin index */ |
3294 | mbinptr bin; /* associated bin */ | |
f65fd747 | 3295 | |
6c8dbf00 | 3296 | mchunkptr victim; /* inspected/selected chunk */ |
fa8d436c | 3297 | INTERNAL_SIZE_T size; /* its size */ |
6c8dbf00 | 3298 | int victim_index; /* its bin index */ |
f65fd747 | 3299 | |
6c8dbf00 OB |
3300 | mchunkptr remainder; /* remainder from a split */ |
3301 | unsigned long remainder_size; /* its size */ | |
8a4b65b4 | 3302 | |
6c8dbf00 OB |
3303 | unsigned int block; /* bit map traverser */ |
3304 | unsigned int bit; /* bit map traverser */ | |
3305 | unsigned int map; /* current word of binmap */ | |
8a4b65b4 | 3306 | |
6c8dbf00 OB |
3307 | mchunkptr fwd; /* misc temp for linking */ |
3308 | mchunkptr bck; /* misc temp for linking */ | |
8a4b65b4 | 3309 | |
f6887a0d UD |
3310 | const char *errstr = NULL; |
3311 | ||
fa8d436c | 3312 | /* |
6c8dbf00 OB |
3313 | Convert request size to internal form by adding SIZE_SZ bytes |
3314 | overhead plus possibly more to obtain necessary alignment and/or | |
3315 | to obtain a size of at least MINSIZE, the smallest allocatable | |
3316 | size. Also, checked_request2size traps (returning 0) request sizes | |
3317 | that are so large that they wrap around zero when padded and | |
3318 | aligned. | |
3319 | */ | |
f65fd747 | 3320 | |
6c8dbf00 | 3321 | checked_request2size (bytes, nb); |
f65fd747 | 3322 | |
fa8d436c | 3323 | /* |
6c8dbf00 OB |
3324 | If the size qualifies as a fastbin, first check corresponding bin. |
3325 | This code is safe to execute even if av is not yet initialized, so we | |
3326 | can try it without checking, which saves some time on this fast path. | |
3327 | */ | |
f65fd747 | 3328 | |
6c8dbf00 OB |
3329 | if ((unsigned long) (nb) <= (unsigned long) (get_max_fast ())) |
3330 | { | |
3331 | idx = fastbin_index (nb); | |
3332 | mfastbinptr *fb = &fastbin (av, idx); | |
3333 | mchunkptr pp = *fb; | |
3334 | do | |
3335 | { | |
3336 | victim = pp; | |
3337 | if (victim == NULL) | |
3338 | break; | |
3339 | } | |
3340 | while ((pp = catomic_compare_and_exchange_val_acq (fb, victim->fd, victim)) | |
3341 | != victim); | |
3342 | if (victim != 0) | |
3343 | { | |
3344 | if (__builtin_expect (fastbin_index (chunksize (victim)) != idx, 0)) | |
3345 | { | |
3346 | errstr = "malloc(): memory corruption (fast)"; | |
3347 | errout: | |
3348 | malloc_printerr (check_action, errstr, chunk2mem (victim)); | |
3349 | return NULL; | |
3350 | } | |
3351 | check_remalloced_chunk (av, victim, nb); | |
3352 | void *p = chunk2mem (victim); | |
3353 | alloc_perturb (p, bytes); | |
3354 | return p; | |
3355 | } | |
fa8d436c | 3356 | } |
f65fd747 | 3357 | |
fa8d436c | 3358 | /* |
6c8dbf00 OB |
3359 | If a small request, check regular bin. Since these "smallbins" |
3360 | hold one size each, no searching within bins is necessary. | |
3361 | (For a large request, we need to wait until unsorted chunks are | |
3362 | processed to find best fit. But for small ones, fits are exact | |
3363 | anyway, so we can check now, which is faster.) | |
3364 | */ | |
3365 | ||
3366 | if (in_smallbin_range (nb)) | |
3367 | { | |
3368 | idx = smallbin_index (nb); | |
3369 | bin = bin_at (av, idx); | |
3370 | ||
3371 | if ((victim = last (bin)) != bin) | |
3372 | { | |
3373 | if (victim == 0) /* initialization check */ | |
3374 | malloc_consolidate (av); | |
3375 | else | |
3376 | { | |
3377 | bck = victim->bk; | |
a1ffb40e | 3378 | if (__glibc_unlikely (bck->fd != victim)) |
6c8dbf00 OB |
3379 | { |
3380 | errstr = "malloc(): smallbin double linked list corrupted"; | |
3381 | goto errout; | |
3382 | } | |
3383 | set_inuse_bit_at_offset (victim, nb); | |
3384 | bin->bk = bck; | |
3385 | bck->fd = bin; | |
3386 | ||
3387 | if (av != &main_arena) | |
3388 | victim->size |= NON_MAIN_ARENA; | |
3389 | check_malloced_chunk (av, victim, nb); | |
3390 | void *p = chunk2mem (victim); | |
3391 | alloc_perturb (p, bytes); | |
3392 | return p; | |
3393 | } | |
3394 | } | |
fa8d436c | 3395 | } |
f65fd747 | 3396 | |
a9177ff5 | 3397 | /* |
fa8d436c UD |
3398 | If this is a large request, consolidate fastbins before continuing. |
3399 | While it might look excessive to kill all fastbins before | |
3400 | even seeing if there is space available, this avoids | |
3401 | fragmentation problems normally associated with fastbins. | |
3402 | Also, in practice, programs tend to have runs of either small or | |
a9177ff5 | 3403 | large requests, but less often mixtures, so consolidation is not |
fa8d436c UD |
3404 | invoked all that often in most programs. And the programs that |
3405 | it is called frequently in otherwise tend to fragment. | |
6c8dbf00 | 3406 | */ |
7799b7b3 | 3407 | |
6c8dbf00 OB |
3408 | else |
3409 | { | |
3410 | idx = largebin_index (nb); | |
3411 | if (have_fastchunks (av)) | |
3412 | malloc_consolidate (av); | |
3413 | } | |
f65fd747 | 3414 | |
fa8d436c | 3415 | /* |
6c8dbf00 OB |
3416 | Process recently freed or remaindered chunks, taking one only if |
3417 | it is exact fit, or, if this a small request, the chunk is remainder from | |
3418 | the most recent non-exact fit. Place other traversed chunks in | |
3419 | bins. Note that this step is the only place in any routine where | |
3420 | chunks are placed in bins. | |
3421 | ||
3422 | The outer loop here is needed because we might not realize until | |
3423 | near the end of malloc that we should have consolidated, so must | |
3424 | do so and retry. This happens at most once, and only when we would | |
3425 | otherwise need to expand memory to service a "small" request. | |
3426 | */ | |
3427 | ||
3428 | for (;; ) | |
3429 | { | |
3430 | int iters = 0; | |
3431 | while ((victim = unsorted_chunks (av)->bk) != unsorted_chunks (av)) | |
3432 | { | |
3433 | bck = victim->bk; | |
3434 | if (__builtin_expect (victim->size <= 2 * SIZE_SZ, 0) | |
3435 | || __builtin_expect (victim->size > av->system_mem, 0)) | |
3436 | malloc_printerr (check_action, "malloc(): memory corruption", | |
3437 | chunk2mem (victim)); | |
3438 | size = chunksize (victim); | |
3439 | ||
3440 | /* | |
3441 | If a small request, try to use last remainder if it is the | |
3442 | only chunk in unsorted bin. This helps promote locality for | |
3443 | runs of consecutive small requests. This is the only | |
3444 | exception to best-fit, and applies only when there is | |
3445 | no exact fit for a small chunk. | |
3446 | */ | |
3447 | ||
3448 | if (in_smallbin_range (nb) && | |
3449 | bck == unsorted_chunks (av) && | |
3450 | victim == av->last_remainder && | |
3451 | (unsigned long) (size) > (unsigned long) (nb + MINSIZE)) | |
3452 | { | |
3453 | /* split and reattach remainder */ | |
3454 | remainder_size = size - nb; | |
3455 | remainder = chunk_at_offset (victim, nb); | |
3456 | unsorted_chunks (av)->bk = unsorted_chunks (av)->fd = remainder; | |
3457 | av->last_remainder = remainder; | |
3458 | remainder->bk = remainder->fd = unsorted_chunks (av); | |
3459 | if (!in_smallbin_range (remainder_size)) | |
3460 | { | |
3461 | remainder->fd_nextsize = NULL; | |
3462 | remainder->bk_nextsize = NULL; | |
3463 | } | |
3464 | ||
3465 | set_head (victim, nb | PREV_INUSE | | |
3466 | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
3467 | set_head (remainder, remainder_size | PREV_INUSE); | |
3468 | set_foot (remainder, remainder_size); | |
3469 | ||
3470 | check_malloced_chunk (av, victim, nb); | |
3471 | void *p = chunk2mem (victim); | |
3472 | alloc_perturb (p, bytes); | |
3473 | return p; | |
3474 | } | |
3475 | ||
3476 | /* remove from unsorted list */ | |
3477 | unsorted_chunks (av)->bk = bck; | |
3478 | bck->fd = unsorted_chunks (av); | |
3479 | ||
3480 | /* Take now instead of binning if exact fit */ | |
3481 | ||
3482 | if (size == nb) | |
3483 | { | |
3484 | set_inuse_bit_at_offset (victim, size); | |
3485 | if (av != &main_arena) | |
3486 | victim->size |= NON_MAIN_ARENA; | |
3487 | check_malloced_chunk (av, victim, nb); | |
3488 | void *p = chunk2mem (victim); | |
3489 | alloc_perturb (p, bytes); | |
3490 | return p; | |
3491 | } | |
3492 | ||
3493 | /* place chunk in bin */ | |
3494 | ||
3495 | if (in_smallbin_range (size)) | |
3496 | { | |
3497 | victim_index = smallbin_index (size); | |
3498 | bck = bin_at (av, victim_index); | |
3499 | fwd = bck->fd; | |
3500 | } | |
3501 | else | |
3502 | { | |
3503 | victim_index = largebin_index (size); | |
3504 | bck = bin_at (av, victim_index); | |
3505 | fwd = bck->fd; | |
3506 | ||
3507 | /* maintain large bins in sorted order */ | |
3508 | if (fwd != bck) | |
3509 | { | |
3510 | /* Or with inuse bit to speed comparisons */ | |
3511 | size |= PREV_INUSE; | |
3512 | /* if smaller than smallest, bypass loop below */ | |
3513 | assert ((bck->bk->size & NON_MAIN_ARENA) == 0); | |
3514 | if ((unsigned long) (size) < (unsigned long) (bck->bk->size)) | |
3515 | { | |
3516 | fwd = bck; | |
3517 | bck = bck->bk; | |
3518 | ||
3519 | victim->fd_nextsize = fwd->fd; | |
3520 | victim->bk_nextsize = fwd->fd->bk_nextsize; | |
3521 | fwd->fd->bk_nextsize = victim->bk_nextsize->fd_nextsize = victim; | |
3522 | } | |
3523 | else | |
3524 | { | |
3525 | assert ((fwd->size & NON_MAIN_ARENA) == 0); | |
3526 | while ((unsigned long) size < fwd->size) | |
3527 | { | |
3528 | fwd = fwd->fd_nextsize; | |
3529 | assert ((fwd->size & NON_MAIN_ARENA) == 0); | |
3530 | } | |
3531 | ||
3532 | if ((unsigned long) size == (unsigned long) fwd->size) | |
3533 | /* Always insert in the second position. */ | |
3534 | fwd = fwd->fd; | |
3535 | else | |
3536 | { | |
3537 | victim->fd_nextsize = fwd; | |
3538 | victim->bk_nextsize = fwd->bk_nextsize; | |
3539 | fwd->bk_nextsize = victim; | |
3540 | victim->bk_nextsize->fd_nextsize = victim; | |
3541 | } | |
3542 | bck = fwd->bk; | |
3543 | } | |
3544 | } | |
3545 | else | |
3546 | victim->fd_nextsize = victim->bk_nextsize = victim; | |
3547 | } | |
3548 | ||
3549 | mark_bin (av, victim_index); | |
3550 | victim->bk = bck; | |
3551 | victim->fd = fwd; | |
3552 | fwd->bk = victim; | |
3553 | bck->fd = victim; | |
3554 | ||
3555 | #define MAX_ITERS 10000 | |
3556 | if (++iters >= MAX_ITERS) | |
3557 | break; | |
3558 | } | |
fa8d436c | 3559 | |
a9177ff5 | 3560 | /* |
6c8dbf00 OB |
3561 | If a large request, scan through the chunks of current bin in |
3562 | sorted order to find smallest that fits. Use the skip list for this. | |
3563 | */ | |
3564 | ||
3565 | if (!in_smallbin_range (nb)) | |
3566 | { | |
3567 | bin = bin_at (av, idx); | |
3568 | ||
3569 | /* skip scan if empty or largest chunk is too small */ | |
3570 | if ((victim = first (bin)) != bin && | |
3571 | (unsigned long) (victim->size) >= (unsigned long) (nb)) | |
3572 | { | |
3573 | victim = victim->bk_nextsize; | |
3574 | while (((unsigned long) (size = chunksize (victim)) < | |
3575 | (unsigned long) (nb))) | |
3576 | victim = victim->bk_nextsize; | |
3577 | ||
3578 | /* Avoid removing the first entry for a size so that the skip | |
3579 | list does not have to be rerouted. */ | |
3580 | if (victim != last (bin) && victim->size == victim->fd->size) | |
3581 | victim = victim->fd; | |
3582 | ||
3583 | remainder_size = size - nb; | |
3584 | unlink (victim, bck, fwd); | |
3585 | ||
3586 | /* Exhaust */ | |
3587 | if (remainder_size < MINSIZE) | |
3588 | { | |
3589 | set_inuse_bit_at_offset (victim, size); | |
3590 | if (av != &main_arena) | |
3591 | victim->size |= NON_MAIN_ARENA; | |
3592 | } | |
3593 | /* Split */ | |
3594 | else | |
3595 | { | |
3596 | remainder = chunk_at_offset (victim, nb); | |
3597 | /* We cannot assume the unsorted list is empty and therefore | |
3598 | have to perform a complete insert here. */ | |
3599 | bck = unsorted_chunks (av); | |
3600 | fwd = bck->fd; | |
a1ffb40e | 3601 | if (__glibc_unlikely (fwd->bk != bck)) |
6c8dbf00 OB |
3602 | { |
3603 | errstr = "malloc(): corrupted unsorted chunks"; | |
3604 | goto errout; | |
3605 | } | |
3606 | remainder->bk = bck; | |
3607 | remainder->fd = fwd; | |
3608 | bck->fd = remainder; | |
3609 | fwd->bk = remainder; | |
3610 | if (!in_smallbin_range (remainder_size)) | |
3611 | { | |
3612 | remainder->fd_nextsize = NULL; | |
3613 | remainder->bk_nextsize = NULL; | |
3614 | } | |
3615 | set_head (victim, nb | PREV_INUSE | | |
3616 | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
3617 | set_head (remainder, remainder_size | PREV_INUSE); | |
3618 | set_foot (remainder, remainder_size); | |
3619 | } | |
3620 | check_malloced_chunk (av, victim, nb); | |
3621 | void *p = chunk2mem (victim); | |
3622 | alloc_perturb (p, bytes); | |
3623 | return p; | |
3624 | } | |
3625 | } | |
f65fd747 | 3626 | |
6c8dbf00 OB |
3627 | /* |
3628 | Search for a chunk by scanning bins, starting with next largest | |
3629 | bin. This search is strictly by best-fit; i.e., the smallest | |
3630 | (with ties going to approximately the least recently used) chunk | |
3631 | that fits is selected. | |
3632 | ||
3633 | The bitmap avoids needing to check that most blocks are nonempty. | |
3634 | The particular case of skipping all bins during warm-up phases | |
3635 | when no chunks have been returned yet is faster than it might look. | |
3636 | */ | |
3637 | ||
3638 | ++idx; | |
3639 | bin = bin_at (av, idx); | |
3640 | block = idx2block (idx); | |
3641 | map = av->binmap[block]; | |
3642 | bit = idx2bit (idx); | |
3643 | ||
3644 | for (;; ) | |
3645 | { | |
3646 | /* Skip rest of block if there are no more set bits in this block. */ | |
3647 | if (bit > map || bit == 0) | |
3648 | { | |
3649 | do | |
3650 | { | |
3651 | if (++block >= BINMAPSIZE) /* out of bins */ | |
3652 | goto use_top; | |
3653 | } | |
3654 | while ((map = av->binmap[block]) == 0); | |
3655 | ||
3656 | bin = bin_at (av, (block << BINMAPSHIFT)); | |
3657 | bit = 1; | |
3658 | } | |
3659 | ||
3660 | /* Advance to bin with set bit. There must be one. */ | |
3661 | while ((bit & map) == 0) | |
3662 | { | |
3663 | bin = next_bin (bin); | |
3664 | bit <<= 1; | |
3665 | assert (bit != 0); | |
3666 | } | |
3667 | ||
3668 | /* Inspect the bin. It is likely to be non-empty */ | |
3669 | victim = last (bin); | |
3670 | ||
3671 | /* If a false alarm (empty bin), clear the bit. */ | |
3672 | if (victim == bin) | |
3673 | { | |
3674 | av->binmap[block] = map &= ~bit; /* Write through */ | |
3675 | bin = next_bin (bin); | |
3676 | bit <<= 1; | |
3677 | } | |
3678 | ||
3679 | else | |
3680 | { | |
3681 | size = chunksize (victim); | |
3682 | ||
3683 | /* We know the first chunk in this bin is big enough to use. */ | |
3684 | assert ((unsigned long) (size) >= (unsigned long) (nb)); | |
3685 | ||
3686 | remainder_size = size - nb; | |
3687 | ||
3688 | /* unlink */ | |
3689 | unlink (victim, bck, fwd); | |
3690 | ||
3691 | /* Exhaust */ | |
3692 | if (remainder_size < MINSIZE) | |
3693 | { | |
3694 | set_inuse_bit_at_offset (victim, size); | |
3695 | if (av != &main_arena) | |
3696 | victim->size |= NON_MAIN_ARENA; | |
3697 | } | |
3698 | ||
3699 | /* Split */ | |
3700 | else | |
3701 | { | |
3702 | remainder = chunk_at_offset (victim, nb); | |
3703 | ||
3704 | /* We cannot assume the unsorted list is empty and therefore | |
3705 | have to perform a complete insert here. */ | |
3706 | bck = unsorted_chunks (av); | |
3707 | fwd = bck->fd; | |
a1ffb40e | 3708 | if (__glibc_unlikely (fwd->bk != bck)) |
6c8dbf00 OB |
3709 | { |
3710 | errstr = "malloc(): corrupted unsorted chunks 2"; | |
3711 | goto errout; | |
3712 | } | |
3713 | remainder->bk = bck; | |
3714 | remainder->fd = fwd; | |
3715 | bck->fd = remainder; | |
3716 | fwd->bk = remainder; | |
3717 | ||
3718 | /* advertise as last remainder */ | |
3719 | if (in_smallbin_range (nb)) | |
3720 | av->last_remainder = remainder; | |
3721 | if (!in_smallbin_range (remainder_size)) | |
3722 | { | |
3723 | remainder->fd_nextsize = NULL; | |
3724 | remainder->bk_nextsize = NULL; | |
3725 | } | |
3726 | set_head (victim, nb | PREV_INUSE | | |
3727 | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
3728 | set_head (remainder, remainder_size | PREV_INUSE); | |
3729 | set_foot (remainder, remainder_size); | |
3730 | } | |
3731 | check_malloced_chunk (av, victim, nb); | |
3732 | void *p = chunk2mem (victim); | |
3733 | alloc_perturb (p, bytes); | |
3734 | return p; | |
3735 | } | |
3736 | } | |
3737 | ||
3738 | use_top: | |
3739 | /* | |
3740 | If large enough, split off the chunk bordering the end of memory | |
3741 | (held in av->top). Note that this is in accord with the best-fit | |
3742 | search rule. In effect, av->top is treated as larger (and thus | |
3743 | less well fitting) than any other available chunk since it can | |
3744 | be extended to be as large as necessary (up to system | |
3745 | limitations). | |
3746 | ||
3747 | We require that av->top always exists (i.e., has size >= | |
3748 | MINSIZE) after initialization, so if it would otherwise be | |
3749 | exhausted by current request, it is replenished. (The main | |
3750 | reason for ensuring it exists is that we may need MINSIZE space | |
3751 | to put in fenceposts in sysmalloc.) | |
3752 | */ | |
3753 | ||
3754 | victim = av->top; | |
3755 | size = chunksize (victim); | |
3756 | ||
3757 | if ((unsigned long) (size) >= (unsigned long) (nb + MINSIZE)) | |
3758 | { | |
3759 | remainder_size = size - nb; | |
3760 | remainder = chunk_at_offset (victim, nb); | |
3761 | av->top = remainder; | |
3762 | set_head (victim, nb | PREV_INUSE | | |
3763 | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
3764 | set_head (remainder, remainder_size | PREV_INUSE); | |
3765 | ||
3766 | check_malloced_chunk (av, victim, nb); | |
3767 | void *p = chunk2mem (victim); | |
3768 | alloc_perturb (p, bytes); | |
3769 | return p; | |
3770 | } | |
3771 | ||
3772 | /* When we are using atomic ops to free fast chunks we can get | |
3773 | here for all block sizes. */ | |
3774 | else if (have_fastchunks (av)) | |
3775 | { | |
3776 | malloc_consolidate (av); | |
3777 | /* restore original bin index */ | |
3778 | if (in_smallbin_range (nb)) | |
3779 | idx = smallbin_index (nb); | |
3780 | else | |
3781 | idx = largebin_index (nb); | |
3782 | } | |
f65fd747 | 3783 | |
6c8dbf00 OB |
3784 | /* |
3785 | Otherwise, relay to handle system-dependent cases | |
3786 | */ | |
425ce2ed | 3787 | else |
6c8dbf00 OB |
3788 | { |
3789 | void *p = sysmalloc (nb, av); | |
3790 | if (p != NULL) | |
3791 | alloc_perturb (p, bytes); | |
3792 | return p; | |
3793 | } | |
425ce2ed | 3794 | } |
fa8d436c | 3795 | } |
f65fd747 | 3796 | |
fa8d436c | 3797 | /* |
6c8dbf00 OB |
3798 | ------------------------------ free ------------------------------ |
3799 | */ | |
f65fd747 | 3800 | |
78ac92ad | 3801 | static void |
6c8dbf00 | 3802 | _int_free (mstate av, mchunkptr p, int have_lock) |
f65fd747 | 3803 | { |
fa8d436c | 3804 | INTERNAL_SIZE_T size; /* its size */ |
6c8dbf00 OB |
3805 | mfastbinptr *fb; /* associated fastbin */ |
3806 | mchunkptr nextchunk; /* next contiguous chunk */ | |
fa8d436c | 3807 | INTERNAL_SIZE_T nextsize; /* its size */ |
6c8dbf00 | 3808 | int nextinuse; /* true if nextchunk is used */ |
fa8d436c | 3809 | INTERNAL_SIZE_T prevsize; /* size of previous contiguous chunk */ |
6c8dbf00 OB |
3810 | mchunkptr bck; /* misc temp for linking */ |
3811 | mchunkptr fwd; /* misc temp for linking */ | |
fa8d436c | 3812 | |
37fa1953 | 3813 | const char *errstr = NULL; |
425ce2ed | 3814 | int locked = 0; |
f65fd747 | 3815 | |
6c8dbf00 | 3816 | size = chunksize (p); |
f65fd747 | 3817 | |
37fa1953 UD |
3818 | /* Little security check which won't hurt performance: the |
3819 | allocator never wrapps around at the end of the address space. | |
3820 | Therefore we can exclude some size values which might appear | |
3821 | here by accident or by "design" from some intruder. */ | |
dc165f7b | 3822 | if (__builtin_expect ((uintptr_t) p > (uintptr_t) -size, 0) |
073f560e | 3823 | || __builtin_expect (misaligned_chunk (p), 0)) |
37fa1953 UD |
3824 | { |
3825 | errstr = "free(): invalid pointer"; | |
3826 | errout: | |
6c8dbf00 OB |
3827 | if (!have_lock && locked) |
3828 | (void) mutex_unlock (&av->mutex); | |
3829 | malloc_printerr (check_action, errstr, chunk2mem (p)); | |
37fa1953 | 3830 | return; |
fa8d436c | 3831 | } |
347c92e9 L |
3832 | /* We know that each chunk is at least MINSIZE bytes in size or a |
3833 | multiple of MALLOC_ALIGNMENT. */ | |
a1ffb40e | 3834 | if (__glibc_unlikely (size < MINSIZE || !aligned_OK (size))) |
bf589066 UD |
3835 | { |
3836 | errstr = "free(): invalid size"; | |
3837 | goto errout; | |
3838 | } | |
f65fd747 | 3839 | |
37fa1953 | 3840 | check_inuse_chunk(av, p); |
f65fd747 | 3841 | |
37fa1953 UD |
3842 | /* |
3843 | If eligible, place chunk on a fastbin so it can be found | |
3844 | and used quickly in malloc. | |
3845 | */ | |
6bf4302e | 3846 | |
9bf248c6 | 3847 | if ((unsigned long)(size) <= (unsigned long)(get_max_fast ()) |
6bf4302e | 3848 | |
37fa1953 UD |
3849 | #if TRIM_FASTBINS |
3850 | /* | |
3851 | If TRIM_FASTBINS set, don't place chunks | |
3852 | bordering top into fastbins | |
3853 | */ | |
3854 | && (chunk_at_offset(p, size) != av->top) | |
3855 | #endif | |
3856 | ) { | |
fa8d436c | 3857 | |
893e6098 UD |
3858 | if (__builtin_expect (chunk_at_offset (p, size)->size <= 2 * SIZE_SZ, 0) |
3859 | || __builtin_expect (chunksize (chunk_at_offset (p, size)) | |
3860 | >= av->system_mem, 0)) | |
3861 | { | |
bec466d9 UD |
3862 | /* We might not have a lock at this point and concurrent modifications |
3863 | of system_mem might have let to a false positive. Redo the test | |
3864 | after getting the lock. */ | |
3865 | if (have_lock | |
3866 | || ({ assert (locked == 0); | |
3867 | mutex_lock(&av->mutex); | |
3868 | locked = 1; | |
3869 | chunk_at_offset (p, size)->size <= 2 * SIZE_SZ | |
3870 | || chunksize (chunk_at_offset (p, size)) >= av->system_mem; | |
3871 | })) | |
bec466d9 UD |
3872 | { |
3873 | errstr = "free(): invalid next size (fast)"; | |
3874 | goto errout; | |
3875 | } | |
bec466d9 UD |
3876 | if (! have_lock) |
3877 | { | |
3878 | (void)mutex_unlock(&av->mutex); | |
3879 | locked = 0; | |
3880 | } | |
893e6098 UD |
3881 | } |
3882 | ||
e8349efd | 3883 | free_perturb (chunk2mem(p), size - 2 * SIZE_SZ); |
425ce2ed | 3884 | |
37fa1953 | 3885 | set_fastchunks(av); |
90a3055e UD |
3886 | unsigned int idx = fastbin_index(size); |
3887 | fb = &fastbin (av, idx); | |
425ce2ed | 3888 | |
362b47fe MK |
3889 | /* Atomically link P to its fastbin: P->FD = *FB; *FB = P; */ |
3890 | mchunkptr old = *fb, old2; | |
5f24d53a | 3891 | unsigned int old_idx = ~0u; |
425ce2ed UD |
3892 | do |
3893 | { | |
362b47fe MK |
3894 | /* Check that the top of the bin is not the record we are going to add |
3895 | (i.e., double free). */ | |
425ce2ed UD |
3896 | if (__builtin_expect (old == p, 0)) |
3897 | { | |
3898 | errstr = "double free or corruption (fasttop)"; | |
3899 | goto errout; | |
3900 | } | |
362b47fe MK |
3901 | /* Check that size of fastbin chunk at the top is the same as |
3902 | size of the chunk that we are adding. We can dereference OLD | |
3903 | only if we have the lock, otherwise it might have already been | |
3904 | deallocated. See use of OLD_IDX below for the actual check. */ | |
3905 | if (have_lock && old != NULL) | |
5f24d53a | 3906 | old_idx = fastbin_index(chunksize(old)); |
362b47fe | 3907 | p->fd = old2 = old; |
425ce2ed | 3908 | } |
362b47fe | 3909 | while ((old = catomic_compare_and_exchange_val_rel (fb, p, old2)) != old2); |
5f24d53a | 3910 | |
362b47fe | 3911 | if (have_lock && old != NULL && __builtin_expect (old_idx != idx, 0)) |
5f24d53a UD |
3912 | { |
3913 | errstr = "invalid fastbin entry (free)"; | |
3914 | goto errout; | |
3915 | } | |
37fa1953 | 3916 | } |
f65fd747 | 3917 | |
37fa1953 UD |
3918 | /* |
3919 | Consolidate other non-mmapped chunks as they arrive. | |
3920 | */ | |
fa8d436c | 3921 | |
37fa1953 | 3922 | else if (!chunk_is_mmapped(p)) { |
425ce2ed | 3923 | if (! have_lock) { |
425ce2ed | 3924 | (void)mutex_lock(&av->mutex); |
425ce2ed UD |
3925 | locked = 1; |
3926 | } | |
425ce2ed | 3927 | |
37fa1953 | 3928 | nextchunk = chunk_at_offset(p, size); |
fa8d436c | 3929 | |
37fa1953 UD |
3930 | /* Lightweight tests: check whether the block is already the |
3931 | top block. */ | |
a1ffb40e | 3932 | if (__glibc_unlikely (p == av->top)) |
37fa1953 UD |
3933 | { |
3934 | errstr = "double free or corruption (top)"; | |
3935 | goto errout; | |
3936 | } | |
3937 | /* Or whether the next chunk is beyond the boundaries of the arena. */ | |
3938 | if (__builtin_expect (contiguous (av) | |
3939 | && (char *) nextchunk | |
3940 | >= ((char *) av->top + chunksize(av->top)), 0)) | |
3941 | { | |
3942 | errstr = "double free or corruption (out)"; | |
3943 | goto errout; | |
3944 | } | |
3945 | /* Or whether the block is actually not marked used. */ | |
a1ffb40e | 3946 | if (__glibc_unlikely (!prev_inuse(nextchunk))) |
37fa1953 UD |
3947 | { |
3948 | errstr = "double free or corruption (!prev)"; | |
3949 | goto errout; | |
3950 | } | |
fa8d436c | 3951 | |
37fa1953 | 3952 | nextsize = chunksize(nextchunk); |
893e6098 UD |
3953 | if (__builtin_expect (nextchunk->size <= 2 * SIZE_SZ, 0) |
3954 | || __builtin_expect (nextsize >= av->system_mem, 0)) | |
3955 | { | |
76761b63 | 3956 | errstr = "free(): invalid next size (normal)"; |
893e6098 UD |
3957 | goto errout; |
3958 | } | |
fa8d436c | 3959 | |
e8349efd | 3960 | free_perturb (chunk2mem(p), size - 2 * SIZE_SZ); |
854278df | 3961 | |
37fa1953 UD |
3962 | /* consolidate backward */ |
3963 | if (!prev_inuse(p)) { | |
3964 | prevsize = p->prev_size; | |
3965 | size += prevsize; | |
3966 | p = chunk_at_offset(p, -((long) prevsize)); | |
3967 | unlink(p, bck, fwd); | |
3968 | } | |
a9177ff5 | 3969 | |
37fa1953 UD |
3970 | if (nextchunk != av->top) { |
3971 | /* get and clear inuse bit */ | |
3972 | nextinuse = inuse_bit_at_offset(nextchunk, nextsize); | |
3973 | ||
3974 | /* consolidate forward */ | |
3975 | if (!nextinuse) { | |
3976 | unlink(nextchunk, bck, fwd); | |
3977 | size += nextsize; | |
3978 | } else | |
3979 | clear_inuse_bit_at_offset(nextchunk, 0); | |
10dc2a90 | 3980 | |
fa8d436c | 3981 | /* |
37fa1953 UD |
3982 | Place the chunk in unsorted chunk list. Chunks are |
3983 | not placed into regular bins until after they have | |
3984 | been given one chance to be used in malloc. | |
fa8d436c | 3985 | */ |
f65fd747 | 3986 | |
37fa1953 UD |
3987 | bck = unsorted_chunks(av); |
3988 | fwd = bck->fd; | |
a1ffb40e | 3989 | if (__glibc_unlikely (fwd->bk != bck)) |
f6887a0d UD |
3990 | { |
3991 | errstr = "free(): corrupted unsorted chunks"; | |
3992 | goto errout; | |
3993 | } | |
37fa1953 | 3994 | p->fd = fwd; |
7ecfbd38 UD |
3995 | p->bk = bck; |
3996 | if (!in_smallbin_range(size)) | |
3997 | { | |
3998 | p->fd_nextsize = NULL; | |
3999 | p->bk_nextsize = NULL; | |
4000 | } | |
37fa1953 UD |
4001 | bck->fd = p; |
4002 | fwd->bk = p; | |
8a4b65b4 | 4003 | |
37fa1953 UD |
4004 | set_head(p, size | PREV_INUSE); |
4005 | set_foot(p, size); | |
4006 | ||
4007 | check_free_chunk(av, p); | |
4008 | } | |
4009 | ||
4010 | /* | |
4011 | If the chunk borders the current high end of memory, | |
4012 | consolidate into top | |
4013 | */ | |
4014 | ||
4015 | else { | |
4016 | size += nextsize; | |
4017 | set_head(p, size | PREV_INUSE); | |
4018 | av->top = p; | |
4019 | check_chunk(av, p); | |
4020 | } | |
4021 | ||
4022 | /* | |
4023 | If freeing a large space, consolidate possibly-surrounding | |
4024 | chunks. Then, if the total unused topmost memory exceeds trim | |
4025 | threshold, ask malloc_trim to reduce top. | |
4026 | ||
4027 | Unless max_fast is 0, we don't know if there are fastbins | |
4028 | bordering top, so we cannot tell for sure whether threshold | |
4029 | has been reached unless fastbins are consolidated. But we | |
4030 | don't want to consolidate on each free. As a compromise, | |
4031 | consolidation is performed if FASTBIN_CONSOLIDATION_THRESHOLD | |
4032 | is reached. | |
4033 | */ | |
fa8d436c | 4034 | |
37fa1953 UD |
4035 | if ((unsigned long)(size) >= FASTBIN_CONSOLIDATION_THRESHOLD) { |
4036 | if (have_fastchunks(av)) | |
4037 | malloc_consolidate(av); | |
fa8d436c | 4038 | |
37fa1953 | 4039 | if (av == &main_arena) { |
a9177ff5 | 4040 | #ifndef MORECORE_CANNOT_TRIM |
37fa1953 UD |
4041 | if ((unsigned long)(chunksize(av->top)) >= |
4042 | (unsigned long)(mp_.trim_threshold)) | |
3b49edc0 | 4043 | systrim(mp_.top_pad, av); |
fa8d436c | 4044 | #endif |
37fa1953 UD |
4045 | } else { |
4046 | /* Always try heap_trim(), even if the top chunk is not | |
4047 | large, because the corresponding heap might go away. */ | |
4048 | heap_info *heap = heap_for_ptr(top(av)); | |
fa8d436c | 4049 | |
37fa1953 UD |
4050 | assert(heap->ar_ptr == av); |
4051 | heap_trim(heap, mp_.top_pad); | |
fa8d436c | 4052 | } |
fa8d436c | 4053 | } |
10dc2a90 | 4054 | |
425ce2ed UD |
4055 | if (! have_lock) { |
4056 | assert (locked); | |
4057 | (void)mutex_unlock(&av->mutex); | |
4058 | } | |
37fa1953 UD |
4059 | } |
4060 | /* | |
22a89187 | 4061 | If the chunk was allocated via mmap, release via munmap(). |
37fa1953 UD |
4062 | */ |
4063 | ||
4064 | else { | |
c120d94d | 4065 | munmap_chunk (p); |
fa8d436c | 4066 | } |
10dc2a90 UD |
4067 | } |
4068 | ||
fa8d436c UD |
4069 | /* |
4070 | ------------------------- malloc_consolidate ------------------------- | |
4071 | ||
4072 | malloc_consolidate is a specialized version of free() that tears | |
4073 | down chunks held in fastbins. Free itself cannot be used for this | |
4074 | purpose since, among other things, it might place chunks back onto | |
4075 | fastbins. So, instead, we need to use a minor variant of the same | |
4076 | code. | |
a9177ff5 | 4077 | |
fa8d436c UD |
4078 | Also, because this routine needs to be called the first time through |
4079 | malloc anyway, it turns out to be the perfect place to trigger | |
4080 | initialization code. | |
4081 | */ | |
4082 | ||
fa8d436c | 4083 | static void malloc_consolidate(mstate av) |
10dc2a90 | 4084 | { |
fa8d436c UD |
4085 | mfastbinptr* fb; /* current fastbin being consolidated */ |
4086 | mfastbinptr* maxfb; /* last fastbin (for loop control) */ | |
4087 | mchunkptr p; /* current chunk being consolidated */ | |
4088 | mchunkptr nextp; /* next chunk to consolidate */ | |
4089 | mchunkptr unsorted_bin; /* bin header */ | |
4090 | mchunkptr first_unsorted; /* chunk to link to */ | |
4091 | ||
4092 | /* These have same use as in free() */ | |
4093 | mchunkptr nextchunk; | |
4094 | INTERNAL_SIZE_T size; | |
4095 | INTERNAL_SIZE_T nextsize; | |
4096 | INTERNAL_SIZE_T prevsize; | |
4097 | int nextinuse; | |
4098 | mchunkptr bck; | |
4099 | mchunkptr fwd; | |
10dc2a90 | 4100 | |
fa8d436c UD |
4101 | /* |
4102 | If max_fast is 0, we know that av hasn't | |
4103 | yet been initialized, in which case do so below | |
4104 | */ | |
10dc2a90 | 4105 | |
9bf248c6 | 4106 | if (get_max_fast () != 0) { |
fa8d436c | 4107 | clear_fastchunks(av); |
10dc2a90 | 4108 | |
fa8d436c | 4109 | unsorted_bin = unsorted_chunks(av); |
10dc2a90 | 4110 | |
fa8d436c UD |
4111 | /* |
4112 | Remove each chunk from fast bin and consolidate it, placing it | |
4113 | then in unsorted bin. Among other reasons for doing this, | |
4114 | placing in unsorted bin avoids needing to calculate actual bins | |
4115 | until malloc is sure that chunks aren't immediately going to be | |
4116 | reused anyway. | |
4117 | */ | |
a9177ff5 | 4118 | |
425ce2ed | 4119 | maxfb = &fastbin (av, NFASTBINS - 1); |
425ce2ed | 4120 | fb = &fastbin (av, 0); |
fa8d436c | 4121 | do { |
425ce2ed | 4122 | p = atomic_exchange_acq (fb, 0); |
425ce2ed | 4123 | if (p != 0) { |
72f90263 UD |
4124 | do { |
4125 | check_inuse_chunk(av, p); | |
4126 | nextp = p->fd; | |
4127 | ||
4128 | /* Slightly streamlined version of consolidation code in free() */ | |
4129 | size = p->size & ~(PREV_INUSE|NON_MAIN_ARENA); | |
4130 | nextchunk = chunk_at_offset(p, size); | |
4131 | nextsize = chunksize(nextchunk); | |
4132 | ||
4133 | if (!prev_inuse(p)) { | |
4134 | prevsize = p->prev_size; | |
4135 | size += prevsize; | |
4136 | p = chunk_at_offset(p, -((long) prevsize)); | |
4137 | unlink(p, bck, fwd); | |
4138 | } | |
4139 | ||
4140 | if (nextchunk != av->top) { | |
4141 | nextinuse = inuse_bit_at_offset(nextchunk, nextsize); | |
4142 | ||
4143 | if (!nextinuse) { | |
4144 | size += nextsize; | |
4145 | unlink(nextchunk, bck, fwd); | |
4146 | } else | |
fa8d436c | 4147 | clear_inuse_bit_at_offset(nextchunk, 0); |
a9177ff5 | 4148 | |
72f90263 UD |
4149 | first_unsorted = unsorted_bin->fd; |
4150 | unsorted_bin->fd = p; | |
4151 | first_unsorted->bk = p; | |
a9177ff5 | 4152 | |
72f90263 | 4153 | if (!in_smallbin_range (size)) { |
7ecfbd38 UD |
4154 | p->fd_nextsize = NULL; |
4155 | p->bk_nextsize = NULL; | |
4156 | } | |
4157 | ||
72f90263 UD |
4158 | set_head(p, size | PREV_INUSE); |
4159 | p->bk = unsorted_bin; | |
4160 | p->fd = first_unsorted; | |
4161 | set_foot(p, size); | |
4162 | } | |
a9177ff5 | 4163 | |
72f90263 UD |
4164 | else { |
4165 | size += nextsize; | |
4166 | set_head(p, size | PREV_INUSE); | |
4167 | av->top = p; | |
4168 | } | |
a9177ff5 | 4169 | |
72f90263 | 4170 | } while ( (p = nextp) != 0); |
a9177ff5 | 4171 | |
fa8d436c UD |
4172 | } |
4173 | } while (fb++ != maxfb); | |
4174 | } | |
4175 | else { | |
4176 | malloc_init_state(av); | |
4177 | check_malloc_state(av); | |
4178 | } | |
4179 | } | |
10dc2a90 | 4180 | |
fa8d436c UD |
4181 | /* |
4182 | ------------------------------ realloc ------------------------------ | |
4183 | */ | |
f65fd747 | 4184 | |
22a89187 | 4185 | void* |
4c8b8cc3 UD |
4186 | _int_realloc(mstate av, mchunkptr oldp, INTERNAL_SIZE_T oldsize, |
4187 | INTERNAL_SIZE_T nb) | |
fa8d436c | 4188 | { |
fa8d436c UD |
4189 | mchunkptr newp; /* chunk to return */ |
4190 | INTERNAL_SIZE_T newsize; /* its size */ | |
22a89187 | 4191 | void* newmem; /* corresponding user mem */ |
f65fd747 | 4192 | |
fa8d436c | 4193 | mchunkptr next; /* next contiguous chunk after oldp */ |
f65fd747 | 4194 | |
fa8d436c UD |
4195 | mchunkptr remainder; /* extra space at end of newp */ |
4196 | unsigned long remainder_size; /* its size */ | |
f65fd747 | 4197 | |
fa8d436c UD |
4198 | mchunkptr bck; /* misc temp for linking */ |
4199 | mchunkptr fwd; /* misc temp for linking */ | |
2ed5fd9a | 4200 | |
fa8d436c UD |
4201 | unsigned long copysize; /* bytes to copy */ |
4202 | unsigned int ncopies; /* INTERNAL_SIZE_T words to copy */ | |
a9177ff5 | 4203 | INTERNAL_SIZE_T* s; /* copy source */ |
fa8d436c | 4204 | INTERNAL_SIZE_T* d; /* copy destination */ |
f65fd747 | 4205 | |
76761b63 | 4206 | const char *errstr = NULL; |
f65fd747 | 4207 | |
6dd6a580 | 4208 | /* oldmem size */ |
76761b63 UD |
4209 | if (__builtin_expect (oldp->size <= 2 * SIZE_SZ, 0) |
4210 | || __builtin_expect (oldsize >= av->system_mem, 0)) | |
4211 | { | |
4b04154d | 4212 | errstr = "realloc(): invalid old size"; |
4c8b8cc3 | 4213 | errout: |
6c8dbf00 | 4214 | malloc_printerr (check_action, errstr, chunk2mem (oldp)); |
4c8b8cc3 | 4215 | return NULL; |
76761b63 UD |
4216 | } |
4217 | ||
6c8dbf00 | 4218 | check_inuse_chunk (av, oldp); |
f65fd747 | 4219 | |
4c8b8cc3 | 4220 | /* All callers already filter out mmap'ed chunks. */ |
6c8dbf00 | 4221 | assert (!chunk_is_mmapped (oldp)); |
f65fd747 | 4222 | |
6c8dbf00 OB |
4223 | next = chunk_at_offset (oldp, oldsize); |
4224 | INTERNAL_SIZE_T nextsize = chunksize (next); | |
22a89187 UD |
4225 | if (__builtin_expect (next->size <= 2 * SIZE_SZ, 0) |
4226 | || __builtin_expect (nextsize >= av->system_mem, 0)) | |
4227 | { | |
4228 | errstr = "realloc(): invalid next size"; | |
4229 | goto errout; | |
4230 | } | |
4231 | ||
6c8dbf00 OB |
4232 | if ((unsigned long) (oldsize) >= (unsigned long) (nb)) |
4233 | { | |
4234 | /* already big enough; split below */ | |
fa8d436c | 4235 | newp = oldp; |
6c8dbf00 | 4236 | newsize = oldsize; |
7799b7b3 | 4237 | } |
f65fd747 | 4238 | |
6c8dbf00 OB |
4239 | else |
4240 | { | |
4241 | /* Try to expand forward into top */ | |
4242 | if (next == av->top && | |
4243 | (unsigned long) (newsize = oldsize + nextsize) >= | |
4244 | (unsigned long) (nb + MINSIZE)) | |
4245 | { | |
4246 | set_head_size (oldp, nb | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
4247 | av->top = chunk_at_offset (oldp, nb); | |
4248 | set_head (av->top, (newsize - nb) | PREV_INUSE); | |
4249 | check_inuse_chunk (av, oldp); | |
4250 | return chunk2mem (oldp); | |
4251 | } | |
4252 | ||
4253 | /* Try to expand forward into next chunk; split off remainder below */ | |
4254 | else if (next != av->top && | |
4255 | !inuse (next) && | |
4256 | (unsigned long) (newsize = oldsize + nextsize) >= | |
4257 | (unsigned long) (nb)) | |
4258 | { | |
4259 | newp = oldp; | |
4260 | unlink (next, bck, fwd); | |
4261 | } | |
4262 | ||
4263 | /* allocate, copy, free */ | |
4264 | else | |
4265 | { | |
4266 | newmem = _int_malloc (av, nb - MALLOC_ALIGN_MASK); | |
4267 | if (newmem == 0) | |
4268 | return 0; /* propagate failure */ | |
4269 | ||
4270 | newp = mem2chunk (newmem); | |
4271 | newsize = chunksize (newp); | |
4272 | ||
4273 | /* | |
4274 | Avoid copy if newp is next chunk after oldp. | |
4275 | */ | |
4276 | if (newp == next) | |
4277 | { | |
4278 | newsize += oldsize; | |
4279 | newp = oldp; | |
4280 | } | |
4281 | else | |
4282 | { | |
4283 | /* | |
4284 | Unroll copy of <= 36 bytes (72 if 8byte sizes) | |
4285 | We know that contents have an odd number of | |
4286 | INTERNAL_SIZE_T-sized words; minimally 3. | |
4287 | */ | |
4288 | ||
4289 | copysize = oldsize - SIZE_SZ; | |
4290 | s = (INTERNAL_SIZE_T *) (chunk2mem (oldp)); | |
4291 | d = (INTERNAL_SIZE_T *) (newmem); | |
4292 | ncopies = copysize / sizeof (INTERNAL_SIZE_T); | |
4293 | assert (ncopies >= 3); | |
4294 | ||
4295 | if (ncopies > 9) | |
4296 | memcpy (d, s, copysize); | |
4297 | ||
4298 | else | |
4299 | { | |
4300 | *(d + 0) = *(s + 0); | |
4301 | *(d + 1) = *(s + 1); | |
4302 | *(d + 2) = *(s + 2); | |
4303 | if (ncopies > 4) | |
4304 | { | |
4305 | *(d + 3) = *(s + 3); | |
4306 | *(d + 4) = *(s + 4); | |
4307 | if (ncopies > 6) | |
4308 | { | |
4309 | *(d + 5) = *(s + 5); | |
4310 | *(d + 6) = *(s + 6); | |
4311 | if (ncopies > 8) | |
4312 | { | |
4313 | *(d + 7) = *(s + 7); | |
4314 | *(d + 8) = *(s + 8); | |
4315 | } | |
4316 | } | |
4317 | } | |
4318 | } | |
4319 | ||
4320 | _int_free (av, oldp, 1); | |
4321 | check_inuse_chunk (av, newp); | |
4322 | return chunk2mem (newp); | |
4323 | } | |
4324 | } | |
fa8d436c | 4325 | } |
f65fd747 | 4326 | |
22a89187 | 4327 | /* If possible, free extra space in old or extended chunk */ |
f65fd747 | 4328 | |
6c8dbf00 | 4329 | assert ((unsigned long) (newsize) >= (unsigned long) (nb)); |
f65fd747 | 4330 | |
22a89187 | 4331 | remainder_size = newsize - nb; |
10dc2a90 | 4332 | |
6c8dbf00 OB |
4333 | if (remainder_size < MINSIZE) /* not enough extra to split off */ |
4334 | { | |
4335 | set_head_size (newp, newsize | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
4336 | set_inuse_bit_at_offset (newp, newsize); | |
4337 | } | |
4338 | else /* split remainder */ | |
4339 | { | |
4340 | remainder = chunk_at_offset (newp, nb); | |
4341 | set_head_size (newp, nb | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
4342 | set_head (remainder, remainder_size | PREV_INUSE | | |
4343 | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
4344 | /* Mark remainder as inuse so free() won't complain */ | |
4345 | set_inuse_bit_at_offset (remainder, remainder_size); | |
4346 | _int_free (av, remainder, 1); | |
4347 | } | |
22a89187 | 4348 | |
6c8dbf00 OB |
4349 | check_inuse_chunk (av, newp); |
4350 | return chunk2mem (newp); | |
fa8d436c UD |
4351 | } |
4352 | ||
4353 | /* | |
6c8dbf00 OB |
4354 | ------------------------------ memalign ------------------------------ |
4355 | */ | |
fa8d436c | 4356 | |
6c8dbf00 OB |
4357 | static void * |
4358 | _int_memalign (mstate av, size_t alignment, size_t bytes) | |
fa8d436c UD |
4359 | { |
4360 | INTERNAL_SIZE_T nb; /* padded request size */ | |
6c8dbf00 OB |
4361 | char *m; /* memory returned by malloc call */ |
4362 | mchunkptr p; /* corresponding chunk */ | |
4363 | char *brk; /* alignment point within p */ | |
4364 | mchunkptr newp; /* chunk to return */ | |
fa8d436c UD |
4365 | INTERNAL_SIZE_T newsize; /* its size */ |
4366 | INTERNAL_SIZE_T leadsize; /* leading space before alignment point */ | |
6c8dbf00 OB |
4367 | mchunkptr remainder; /* spare room at end to split off */ |
4368 | unsigned long remainder_size; /* its size */ | |
fa8d436c | 4369 | INTERNAL_SIZE_T size; |
f65fd747 | 4370 | |
f65fd747 | 4371 | |
f65fd747 | 4372 | |
6c8dbf00 | 4373 | checked_request2size (bytes, nb); |
fa8d436c UD |
4374 | |
4375 | /* | |
6c8dbf00 OB |
4376 | Strategy: find a spot within that chunk that meets the alignment |
4377 | request, and then possibly free the leading and trailing space. | |
4378 | */ | |
fa8d436c UD |
4379 | |
4380 | ||
4381 | /* Call malloc with worst case padding to hit alignment. */ | |
4382 | ||
6c8dbf00 OB |
4383 | m = (char *) (_int_malloc (av, nb + alignment + MINSIZE)); |
4384 | ||
4385 | if (m == 0) | |
4386 | return 0; /* propagate failure */ | |
4387 | ||
4388 | p = mem2chunk (m); | |
4389 | ||
4390 | if ((((unsigned long) (m)) % alignment) != 0) /* misaligned */ | |
4391 | ||
4392 | { /* | |
4393 | Find an aligned spot inside chunk. Since we need to give back | |
4394 | leading space in a chunk of at least MINSIZE, if the first | |
4395 | calculation places us at a spot with less than MINSIZE leader, | |
4396 | we can move to the next aligned spot -- we've allocated enough | |
4397 | total room so that this is always possible. | |
4398 | */ | |
4399 | brk = (char *) mem2chunk (((unsigned long) (m + alignment - 1)) & | |
4400 | - ((signed long) alignment)); | |
4401 | if ((unsigned long) (brk - (char *) (p)) < MINSIZE) | |
4402 | brk += alignment; | |
4403 | ||
4404 | newp = (mchunkptr) brk; | |
4405 | leadsize = brk - (char *) (p); | |
4406 | newsize = chunksize (p) - leadsize; | |
4407 | ||
4408 | /* For mmapped chunks, just adjust offset */ | |
4409 | if (chunk_is_mmapped (p)) | |
4410 | { | |
4411 | newp->prev_size = p->prev_size + leadsize; | |
4412 | set_head (newp, newsize | IS_MMAPPED); | |
4413 | return chunk2mem (newp); | |
4414 | } | |
4415 | ||
4416 | /* Otherwise, give back leader, use the rest */ | |
4417 | set_head (newp, newsize | PREV_INUSE | | |
4418 | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
4419 | set_inuse_bit_at_offset (newp, newsize); | |
4420 | set_head_size (p, leadsize | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
4421 | _int_free (av, p, 1); | |
4422 | p = newp; | |
4423 | ||
4424 | assert (newsize >= nb && | |
4425 | (((unsigned long) (chunk2mem (p))) % alignment) == 0); | |
f65fd747 | 4426 | } |
f65fd747 | 4427 | |
f65fd747 | 4428 | /* Also give back spare room at the end */ |
6c8dbf00 OB |
4429 | if (!chunk_is_mmapped (p)) |
4430 | { | |
4431 | size = chunksize (p); | |
4432 | if ((unsigned long) (size) > (unsigned long) (nb + MINSIZE)) | |
4433 | { | |
4434 | remainder_size = size - nb; | |
4435 | remainder = chunk_at_offset (p, nb); | |
4436 | set_head (remainder, remainder_size | PREV_INUSE | | |
4437 | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
4438 | set_head_size (p, nb); | |
4439 | _int_free (av, remainder, 1); | |
4440 | } | |
fa8d436c | 4441 | } |
f65fd747 | 4442 | |
6c8dbf00 OB |
4443 | check_inuse_chunk (av, p); |
4444 | return chunk2mem (p); | |
f65fd747 UD |
4445 | } |
4446 | ||
f65fd747 | 4447 | |
fa8d436c | 4448 | /* |
6c8dbf00 OB |
4449 | ------------------------------ malloc_trim ------------------------------ |
4450 | */ | |
8a4b65b4 | 4451 | |
6c8dbf00 OB |
4452 | static int |
4453 | mtrim (mstate av, size_t pad) | |
f65fd747 | 4454 | { |
fa8d436c | 4455 | /* Ensure initialization/consolidation */ |
68631c8e UD |
4456 | malloc_consolidate (av); |
4457 | ||
6c8dbf00 | 4458 | const size_t ps = GLRO (dl_pagesize); |
68631c8e UD |
4459 | int psindex = bin_index (ps); |
4460 | const size_t psm1 = ps - 1; | |
4461 | ||
4462 | int result = 0; | |
4463 | for (int i = 1; i < NBINS; ++i) | |
4464 | if (i == 1 || i >= psindex) | |
4465 | { | |
6c8dbf00 | 4466 | mbinptr bin = bin_at (av, i); |
68631c8e | 4467 | |
6c8dbf00 OB |
4468 | for (mchunkptr p = last (bin); p != bin; p = p->bk) |
4469 | { | |
4470 | INTERNAL_SIZE_T size = chunksize (p); | |
68631c8e | 4471 | |
6c8dbf00 OB |
4472 | if (size > psm1 + sizeof (struct malloc_chunk)) |
4473 | { | |
4474 | /* See whether the chunk contains at least one unused page. */ | |
4475 | char *paligned_mem = (char *) (((uintptr_t) p | |
4476 | + sizeof (struct malloc_chunk) | |
4477 | + psm1) & ~psm1); | |
68631c8e | 4478 | |
6c8dbf00 OB |
4479 | assert ((char *) chunk2mem (p) + 4 * SIZE_SZ <= paligned_mem); |
4480 | assert ((char *) p + size > paligned_mem); | |
68631c8e | 4481 | |
6c8dbf00 OB |
4482 | /* This is the size we could potentially free. */ |
4483 | size -= paligned_mem - (char *) p; | |
68631c8e | 4484 | |
6c8dbf00 OB |
4485 | if (size > psm1) |
4486 | { | |
439bda32 | 4487 | #if MALLOC_DEBUG |
6c8dbf00 OB |
4488 | /* When debugging we simulate destroying the memory |
4489 | content. */ | |
4490 | memset (paligned_mem, 0x89, size & ~psm1); | |
68631c8e | 4491 | #endif |
6c8dbf00 | 4492 | __madvise (paligned_mem, size & ~psm1, MADV_DONTNEED); |
68631c8e | 4493 | |
6c8dbf00 OB |
4494 | result = 1; |
4495 | } | |
4496 | } | |
4497 | } | |
68631c8e | 4498 | } |
8a4b65b4 | 4499 | |
a9177ff5 | 4500 | #ifndef MORECORE_CANNOT_TRIM |
3b49edc0 | 4501 | return result | (av == &main_arena ? systrim (pad, av) : 0); |
6c8dbf00 | 4502 | |
8a4b65b4 | 4503 | #else |
68631c8e | 4504 | return result; |
f65fd747 | 4505 | #endif |
f65fd747 UD |
4506 | } |
4507 | ||
f65fd747 | 4508 | |
3b49edc0 | 4509 | int |
6c8dbf00 | 4510 | __malloc_trim (size_t s) |
3b49edc0 UD |
4511 | { |
4512 | int result = 0; | |
4513 | ||
6c8dbf00 | 4514 | if (__malloc_initialized < 0) |
3b49edc0 UD |
4515 | ptmalloc_init (); |
4516 | ||
4517 | mstate ar_ptr = &main_arena; | |
4518 | do | |
4519 | { | |
4520 | (void) mutex_lock (&ar_ptr->mutex); | |
4521 | result |= mtrim (ar_ptr, s); | |
4522 | (void) mutex_unlock (&ar_ptr->mutex); | |
4523 | ||
4524 | ar_ptr = ar_ptr->next; | |
4525 | } | |
4526 | while (ar_ptr != &main_arena); | |
4527 | ||
4528 | return result; | |
4529 | } | |
4530 | ||
4531 | ||
f65fd747 | 4532 | /* |
6c8dbf00 OB |
4533 | ------------------------- malloc_usable_size ------------------------- |
4534 | */ | |
f65fd747 | 4535 | |
3b49edc0 | 4536 | static size_t |
6c8dbf00 | 4537 | musable (void *mem) |
f65fd747 UD |
4538 | { |
4539 | mchunkptr p; | |
6c8dbf00 OB |
4540 | if (mem != 0) |
4541 | { | |
4542 | p = mem2chunk (mem); | |
4543 | ||
4544 | if (__builtin_expect (using_malloc_checking == 1, 0)) | |
4545 | return malloc_check_get_size (p); | |
4546 | ||
4547 | if (chunk_is_mmapped (p)) | |
4548 | return chunksize (p) - 2 * SIZE_SZ; | |
4549 | else if (inuse (p)) | |
4550 | return chunksize (p) - SIZE_SZ; | |
4551 | } | |
fa8d436c | 4552 | return 0; |
f65fd747 UD |
4553 | } |
4554 | ||
3b49edc0 UD |
4555 | |
4556 | size_t | |
6c8dbf00 | 4557 | __malloc_usable_size (void *m) |
3b49edc0 UD |
4558 | { |
4559 | size_t result; | |
4560 | ||
6c8dbf00 | 4561 | result = musable (m); |
3b49edc0 UD |
4562 | return result; |
4563 | } | |
4564 | ||
fa8d436c | 4565 | /* |
6c8dbf00 OB |
4566 | ------------------------------ mallinfo ------------------------------ |
4567 | Accumulate malloc statistics for arena AV into M. | |
4568 | */ | |
f65fd747 | 4569 | |
bedee953 | 4570 | static void |
6c8dbf00 | 4571 | int_mallinfo (mstate av, struct mallinfo *m) |
f65fd747 | 4572 | { |
6dd67bd5 | 4573 | size_t i; |
f65fd747 UD |
4574 | mbinptr b; |
4575 | mchunkptr p; | |
f65fd747 | 4576 | INTERNAL_SIZE_T avail; |
fa8d436c UD |
4577 | INTERNAL_SIZE_T fastavail; |
4578 | int nblocks; | |
4579 | int nfastblocks; | |
f65fd747 | 4580 | |
fa8d436c | 4581 | /* Ensure initialization */ |
6c8dbf00 OB |
4582 | if (av->top == 0) |
4583 | malloc_consolidate (av); | |
8a4b65b4 | 4584 | |
6c8dbf00 | 4585 | check_malloc_state (av); |
8a4b65b4 | 4586 | |
fa8d436c | 4587 | /* Account for top */ |
6c8dbf00 | 4588 | avail = chunksize (av->top); |
fa8d436c | 4589 | nblocks = 1; /* top always exists */ |
f65fd747 | 4590 | |
fa8d436c UD |
4591 | /* traverse fastbins */ |
4592 | nfastblocks = 0; | |
4593 | fastavail = 0; | |
4594 | ||
6c8dbf00 OB |
4595 | for (i = 0; i < NFASTBINS; ++i) |
4596 | { | |
4597 | for (p = fastbin (av, i); p != 0; p = p->fd) | |
4598 | { | |
4599 | ++nfastblocks; | |
4600 | fastavail += chunksize (p); | |
4601 | } | |
fa8d436c | 4602 | } |
fa8d436c UD |
4603 | |
4604 | avail += fastavail; | |
f65fd747 | 4605 | |
fa8d436c | 4606 | /* traverse regular bins */ |
6c8dbf00 OB |
4607 | for (i = 1; i < NBINS; ++i) |
4608 | { | |
4609 | b = bin_at (av, i); | |
4610 | for (p = last (b); p != b; p = p->bk) | |
4611 | { | |
4612 | ++nblocks; | |
4613 | avail += chunksize (p); | |
4614 | } | |
fa8d436c | 4615 | } |
f65fd747 | 4616 | |
bedee953 PP |
4617 | m->smblks += nfastblocks; |
4618 | m->ordblks += nblocks; | |
4619 | m->fordblks += avail; | |
4620 | m->uordblks += av->system_mem - avail; | |
4621 | m->arena += av->system_mem; | |
4622 | m->fsmblks += fastavail; | |
4623 | if (av == &main_arena) | |
4624 | { | |
4625 | m->hblks = mp_.n_mmaps; | |
4626 | m->hblkhd = mp_.mmapped_mem; | |
4627 | m->usmblks = mp_.max_total_mem; | |
6c8dbf00 | 4628 | m->keepcost = chunksize (av->top); |
bedee953 | 4629 | } |
fa8d436c | 4630 | } |
f65fd747 | 4631 | |
3b49edc0 | 4632 | |
6c8dbf00 OB |
4633 | struct mallinfo |
4634 | __libc_mallinfo () | |
3b49edc0 UD |
4635 | { |
4636 | struct mallinfo m; | |
bedee953 | 4637 | mstate ar_ptr; |
3b49edc0 | 4638 | |
6c8dbf00 | 4639 | if (__malloc_initialized < 0) |
3b49edc0 | 4640 | ptmalloc_init (); |
bedee953 | 4641 | |
6c8dbf00 | 4642 | memset (&m, 0, sizeof (m)); |
bedee953 | 4643 | ar_ptr = &main_arena; |
6c8dbf00 OB |
4644 | do |
4645 | { | |
4646 | (void) mutex_lock (&ar_ptr->mutex); | |
4647 | int_mallinfo (ar_ptr, &m); | |
4648 | (void) mutex_unlock (&ar_ptr->mutex); | |
bedee953 | 4649 | |
6c8dbf00 OB |
4650 | ar_ptr = ar_ptr->next; |
4651 | } | |
4652 | while (ar_ptr != &main_arena); | |
bedee953 | 4653 | |
3b49edc0 UD |
4654 | return m; |
4655 | } | |
4656 | ||
fa8d436c | 4657 | /* |
6c8dbf00 OB |
4658 | ------------------------------ malloc_stats ------------------------------ |
4659 | */ | |
f65fd747 | 4660 | |
3b49edc0 | 4661 | void |
60d2f8f3 | 4662 | __malloc_stats (void) |
f65fd747 | 4663 | { |
8a4b65b4 | 4664 | int i; |
fa8d436c | 4665 | mstate ar_ptr; |
fa8d436c | 4666 | unsigned int in_use_b = mp_.mmapped_mem, system_b = in_use_b; |
8a4b65b4 | 4667 | |
6c8dbf00 | 4668 | if (__malloc_initialized < 0) |
a234e27d | 4669 | ptmalloc_init (); |
8dab36a1 UD |
4670 | _IO_flockfile (stderr); |
4671 | int old_flags2 = ((_IO_FILE *) stderr)->_flags2; | |
4672 | ((_IO_FILE *) stderr)->_flags2 |= _IO_FLAGS2_NOTCANCEL; | |
6c8dbf00 OB |
4673 | for (i = 0, ar_ptr = &main_arena;; i++) |
4674 | { | |
4675 | struct mallinfo mi; | |
4676 | ||
4677 | memset (&mi, 0, sizeof (mi)); | |
4678 | (void) mutex_lock (&ar_ptr->mutex); | |
4679 | int_mallinfo (ar_ptr, &mi); | |
4680 | fprintf (stderr, "Arena %d:\n", i); | |
4681 | fprintf (stderr, "system bytes = %10u\n", (unsigned int) mi.arena); | |
4682 | fprintf (stderr, "in use bytes = %10u\n", (unsigned int) mi.uordblks); | |
fa8d436c | 4683 | #if MALLOC_DEBUG > 1 |
6c8dbf00 OB |
4684 | if (i > 0) |
4685 | dump_heap (heap_for_ptr (top (ar_ptr))); | |
fa8d436c | 4686 | #endif |
6c8dbf00 OB |
4687 | system_b += mi.arena; |
4688 | in_use_b += mi.uordblks; | |
6c8dbf00 OB |
4689 | (void) mutex_unlock (&ar_ptr->mutex); |
4690 | ar_ptr = ar_ptr->next; | |
4691 | if (ar_ptr == &main_arena) | |
4692 | break; | |
4693 | } | |
4694 | fprintf (stderr, "Total (incl. mmap):\n"); | |
4695 | fprintf (stderr, "system bytes = %10u\n", system_b); | |
4696 | fprintf (stderr, "in use bytes = %10u\n", in_use_b); | |
4697 | fprintf (stderr, "max mmap regions = %10u\n", (unsigned int) mp_.max_n_mmaps); | |
4698 | fprintf (stderr, "max mmap bytes = %10lu\n", | |
4699 | (unsigned long) mp_.max_mmapped_mem); | |
8dab36a1 UD |
4700 | ((_IO_FILE *) stderr)->_flags2 |= old_flags2; |
4701 | _IO_funlockfile (stderr); | |
f65fd747 UD |
4702 | } |
4703 | ||
f65fd747 UD |
4704 | |
4705 | /* | |
6c8dbf00 OB |
4706 | ------------------------------ mallopt ------------------------------ |
4707 | */ | |
f65fd747 | 4708 | |
6c8dbf00 OB |
4709 | int |
4710 | __libc_mallopt (int param_number, int value) | |
f65fd747 | 4711 | { |
fa8d436c UD |
4712 | mstate av = &main_arena; |
4713 | int res = 1; | |
f65fd747 | 4714 | |
6c8dbf00 | 4715 | if (__malloc_initialized < 0) |
0cb71e02 | 4716 | ptmalloc_init (); |
6c8dbf00 | 4717 | (void) mutex_lock (&av->mutex); |
fa8d436c | 4718 | /* Ensure initialization/consolidation */ |
6c8dbf00 | 4719 | malloc_consolidate (av); |
2f6d1f1b | 4720 | |
3ea5be54 AO |
4721 | LIBC_PROBE (memory_mallopt, 2, param_number, value); |
4722 | ||
6c8dbf00 OB |
4723 | switch (param_number) |
4724 | { | |
4725 | case M_MXFAST: | |
4726 | if (value >= 0 && value <= MAX_FAST_SIZE) | |
4727 | { | |
4728 | LIBC_PROBE (memory_mallopt_mxfast, 2, value, get_max_fast ()); | |
4729 | set_max_fast (value); | |
4730 | } | |
4731 | else | |
4732 | res = 0; | |
4733 | break; | |
4734 | ||
4735 | case M_TRIM_THRESHOLD: | |
4736 | LIBC_PROBE (memory_mallopt_trim_threshold, 3, value, | |
4737 | mp_.trim_threshold, mp_.no_dyn_threshold); | |
4738 | mp_.trim_threshold = value; | |
4739 | mp_.no_dyn_threshold = 1; | |
4740 | break; | |
4741 | ||
4742 | case M_TOP_PAD: | |
4743 | LIBC_PROBE (memory_mallopt_top_pad, 3, value, | |
4744 | mp_.top_pad, mp_.no_dyn_threshold); | |
4745 | mp_.top_pad = value; | |
4746 | mp_.no_dyn_threshold = 1; | |
4747 | break; | |
4748 | ||
4749 | case M_MMAP_THRESHOLD: | |
4750 | /* Forbid setting the threshold too high. */ | |
4751 | if ((unsigned long) value > HEAP_MAX_SIZE / 2) | |
4752 | res = 0; | |
4753 | else | |
4754 | { | |
4755 | LIBC_PROBE (memory_mallopt_mmap_threshold, 3, value, | |
4756 | mp_.mmap_threshold, mp_.no_dyn_threshold); | |
4757 | mp_.mmap_threshold = value; | |
4758 | mp_.no_dyn_threshold = 1; | |
4759 | } | |
4760 | break; | |
4761 | ||
4762 | case M_MMAP_MAX: | |
4763 | LIBC_PROBE (memory_mallopt_mmap_max, 3, value, | |
4764 | mp_.n_mmaps_max, mp_.no_dyn_threshold); | |
4765 | mp_.n_mmaps_max = value; | |
4766 | mp_.no_dyn_threshold = 1; | |
4767 | break; | |
4768 | ||
4769 | case M_CHECK_ACTION: | |
4770 | LIBC_PROBE (memory_mallopt_check_action, 2, value, check_action); | |
4771 | check_action = value; | |
4772 | break; | |
4773 | ||
4774 | case M_PERTURB: | |
4775 | LIBC_PROBE (memory_mallopt_perturb, 2, value, perturb_byte); | |
4776 | perturb_byte = value; | |
4777 | break; | |
4778 | ||
4779 | case M_ARENA_TEST: | |
4780 | if (value > 0) | |
4781 | { | |
4782 | LIBC_PROBE (memory_mallopt_arena_test, 2, value, mp_.arena_test); | |
4783 | mp_.arena_test = value; | |
4784 | } | |
4785 | break; | |
4786 | ||
4787 | case M_ARENA_MAX: | |
4788 | if (value > 0) | |
4789 | { | |
4790 | LIBC_PROBE (memory_mallopt_arena_max, 2, value, mp_.arena_max); | |
4791 | mp_.arena_max = value; | |
4792 | } | |
4793 | break; | |
4794 | } | |
4795 | (void) mutex_unlock (&av->mutex); | |
fa8d436c | 4796 | return res; |
b22fc5f5 | 4797 | } |
3b49edc0 | 4798 | libc_hidden_def (__libc_mallopt) |
b22fc5f5 | 4799 | |
10dc2a90 | 4800 | |
a9177ff5 | 4801 | /* |
6c8dbf00 OB |
4802 | -------------------- Alternative MORECORE functions -------------------- |
4803 | */ | |
10dc2a90 | 4804 | |
b22fc5f5 | 4805 | |
fa8d436c | 4806 | /* |
6c8dbf00 | 4807 | General Requirements for MORECORE. |
b22fc5f5 | 4808 | |
6c8dbf00 | 4809 | The MORECORE function must have the following properties: |
b22fc5f5 | 4810 | |
6c8dbf00 | 4811 | If MORECORE_CONTIGUOUS is false: |
10dc2a90 | 4812 | |
6c8dbf00 | 4813 | * MORECORE must allocate in multiples of pagesize. It will |
fa8d436c | 4814 | only be called with arguments that are multiples of pagesize. |
10dc2a90 | 4815 | |
6c8dbf00 | 4816 | * MORECORE(0) must return an address that is at least |
fa8d436c | 4817 | MALLOC_ALIGNMENT aligned. (Page-aligning always suffices.) |
10dc2a90 | 4818 | |
6c8dbf00 | 4819 | else (i.e. If MORECORE_CONTIGUOUS is true): |
10dc2a90 | 4820 | |
6c8dbf00 | 4821 | * Consecutive calls to MORECORE with positive arguments |
fa8d436c UD |
4822 | return increasing addresses, indicating that space has been |
4823 | contiguously extended. | |
10dc2a90 | 4824 | |
6c8dbf00 | 4825 | * MORECORE need not allocate in multiples of pagesize. |
fa8d436c | 4826 | Calls to MORECORE need not have args of multiples of pagesize. |
10dc2a90 | 4827 | |
6c8dbf00 | 4828 | * MORECORE need not page-align. |
10dc2a90 | 4829 | |
6c8dbf00 | 4830 | In either case: |
10dc2a90 | 4831 | |
6c8dbf00 | 4832 | * MORECORE may allocate more memory than requested. (Or even less, |
fa8d436c | 4833 | but this will generally result in a malloc failure.) |
10dc2a90 | 4834 | |
6c8dbf00 | 4835 | * MORECORE must not allocate memory when given argument zero, but |
fa8d436c UD |
4836 | instead return one past the end address of memory from previous |
4837 | nonzero call. This malloc does NOT call MORECORE(0) | |
4838 | until at least one call with positive arguments is made, so | |
4839 | the initial value returned is not important. | |
10dc2a90 | 4840 | |
6c8dbf00 | 4841 | * Even though consecutive calls to MORECORE need not return contiguous |
fa8d436c UD |
4842 | addresses, it must be OK for malloc'ed chunks to span multiple |
4843 | regions in those cases where they do happen to be contiguous. | |
10dc2a90 | 4844 | |
6c8dbf00 | 4845 | * MORECORE need not handle negative arguments -- it may instead |
fa8d436c UD |
4846 | just return MORECORE_FAILURE when given negative arguments. |
4847 | Negative arguments are always multiples of pagesize. MORECORE | |
4848 | must not misinterpret negative args as large positive unsigned | |
4849 | args. You can suppress all such calls from even occurring by defining | |
4850 | MORECORE_CANNOT_TRIM, | |
10dc2a90 | 4851 | |
6c8dbf00 OB |
4852 | There is some variation across systems about the type of the |
4853 | argument to sbrk/MORECORE. If size_t is unsigned, then it cannot | |
4854 | actually be size_t, because sbrk supports negative args, so it is | |
4855 | normally the signed type of the same width as size_t (sometimes | |
4856 | declared as "intptr_t", and sometimes "ptrdiff_t"). It doesn't much | |
4857 | matter though. Internally, we use "long" as arguments, which should | |
4858 | work across all reasonable possibilities. | |
4859 | ||
4860 | Additionally, if MORECORE ever returns failure for a positive | |
4861 | request, then mmap is used as a noncontiguous system allocator. This | |
4862 | is a useful backup strategy for systems with holes in address spaces | |
4863 | -- in this case sbrk cannot contiguously expand the heap, but mmap | |
4864 | may be able to map noncontiguous space. | |
4865 | ||
4866 | If you'd like mmap to ALWAYS be used, you can define MORECORE to be | |
4867 | a function that always returns MORECORE_FAILURE. | |
4868 | ||
4869 | If you are using this malloc with something other than sbrk (or its | |
4870 | emulation) to supply memory regions, you probably want to set | |
4871 | MORECORE_CONTIGUOUS as false. As an example, here is a custom | |
4872 | allocator kindly contributed for pre-OSX macOS. It uses virtually | |
4873 | but not necessarily physically contiguous non-paged memory (locked | |
4874 | in, present and won't get swapped out). You can use it by | |
4875 | uncommenting this section, adding some #includes, and setting up the | |
4876 | appropriate defines above: | |
4877 | ||
4878 | *#define MORECORE osMoreCore | |
4879 | *#define MORECORE_CONTIGUOUS 0 | |
4880 | ||
4881 | There is also a shutdown routine that should somehow be called for | |
4882 | cleanup upon program exit. | |
4883 | ||
4884 | *#define MAX_POOL_ENTRIES 100 | |
4885 | *#define MINIMUM_MORECORE_SIZE (64 * 1024) | |
4886 | static int next_os_pool; | |
4887 | void *our_os_pools[MAX_POOL_ENTRIES]; | |
4888 | ||
4889 | void *osMoreCore(int size) | |
4890 | { | |
fa8d436c UD |
4891 | void *ptr = 0; |
4892 | static void *sbrk_top = 0; | |
ca34d7a7 | 4893 | |
fa8d436c UD |
4894 | if (size > 0) |
4895 | { | |
4896 | if (size < MINIMUM_MORECORE_SIZE) | |
6c8dbf00 | 4897 | size = MINIMUM_MORECORE_SIZE; |
fa8d436c | 4898 | if (CurrentExecutionLevel() == kTaskLevel) |
6c8dbf00 | 4899 | ptr = PoolAllocateResident(size + RM_PAGE_SIZE, 0); |
fa8d436c UD |
4900 | if (ptr == 0) |
4901 | { | |
6c8dbf00 | 4902 | return (void *) MORECORE_FAILURE; |
fa8d436c UD |
4903 | } |
4904 | // save ptrs so they can be freed during cleanup | |
4905 | our_os_pools[next_os_pool] = ptr; | |
4906 | next_os_pool++; | |
4907 | ptr = (void *) ((((unsigned long) ptr) + RM_PAGE_MASK) & ~RM_PAGE_MASK); | |
4908 | sbrk_top = (char *) ptr + size; | |
4909 | return ptr; | |
4910 | } | |
4911 | else if (size < 0) | |
4912 | { | |
4913 | // we don't currently support shrink behavior | |
4914 | return (void *) MORECORE_FAILURE; | |
4915 | } | |
4916 | else | |
4917 | { | |
4918 | return sbrk_top; | |
431c33c0 | 4919 | } |
6c8dbf00 | 4920 | } |
ca34d7a7 | 4921 | |
6c8dbf00 OB |
4922 | // cleanup any allocated memory pools |
4923 | // called as last thing before shutting down driver | |
ca34d7a7 | 4924 | |
6c8dbf00 OB |
4925 | void osCleanupMem(void) |
4926 | { | |
fa8d436c | 4927 | void **ptr; |
ca34d7a7 | 4928 | |
fa8d436c UD |
4929 | for (ptr = our_os_pools; ptr < &our_os_pools[MAX_POOL_ENTRIES]; ptr++) |
4930 | if (*ptr) | |
4931 | { | |
6c8dbf00 OB |
4932 | PoolDeallocate(*ptr); |
4933 | * ptr = 0; | |
fa8d436c | 4934 | } |
6c8dbf00 | 4935 | } |
ee74a442 | 4936 | |
6c8dbf00 | 4937 | */ |
f65fd747 | 4938 | |
7e3be507 | 4939 | |
3e030bd5 UD |
4940 | /* Helper code. */ |
4941 | ||
ae7f5313 UD |
4942 | extern char **__libc_argv attribute_hidden; |
4943 | ||
3e030bd5 | 4944 | static void |
6c8dbf00 | 4945 | malloc_printerr (int action, const char *str, void *ptr) |
3e030bd5 | 4946 | { |
553cc5f9 UD |
4947 | if ((action & 5) == 5) |
4948 | __libc_message (action & 2, "%s\n", str); | |
4949 | else if (action & 1) | |
3e030bd5 | 4950 | { |
a9055cab | 4951 | char buf[2 * sizeof (uintptr_t) + 1]; |
3e030bd5 | 4952 | |
a9055cab UD |
4953 | buf[sizeof (buf) - 1] = '\0'; |
4954 | char *cp = _itoa_word ((uintptr_t) ptr, &buf[sizeof (buf) - 1], 16, 0); | |
4955 | while (cp > buf) | |
6c8dbf00 | 4956 | *--cp = '0'; |
a9055cab | 4957 | |
bb9510dc | 4958 | __libc_message (action & 2, "*** Error in `%s': %s: 0x%s ***\n", |
6c8dbf00 | 4959 | __libc_argv[0] ? : "<unknown>", str, cp); |
3e030bd5 | 4960 | } |
a9055cab | 4961 | else if (action & 2) |
3e030bd5 UD |
4962 | abort (); |
4963 | } | |
4964 | ||
a204dbb2 UD |
4965 | /* We need a wrapper function for one of the additions of POSIX. */ |
4966 | int | |
4967 | __posix_memalign (void **memptr, size_t alignment, size_t size) | |
4968 | { | |
4969 | void *mem; | |
4970 | ||
4971 | /* Test whether the SIZE argument is valid. It must be a power of | |
4972 | two multiple of sizeof (void *). */ | |
de02bd05 UD |
4973 | if (alignment % sizeof (void *) != 0 |
4974 | || !powerof2 (alignment / sizeof (void *)) != 0 | |
4975 | || alignment == 0) | |
a204dbb2 UD |
4976 | return EINVAL; |
4977 | ||
10ad46bc OB |
4978 | |
4979 | void *address = RETURN_ADDRESS (0); | |
4980 | mem = _mid_memalign (alignment, size, address); | |
a204dbb2 | 4981 | |
6c8dbf00 OB |
4982 | if (mem != NULL) |
4983 | { | |
4984 | *memptr = mem; | |
4985 | return 0; | |
4986 | } | |
a204dbb2 UD |
4987 | |
4988 | return ENOMEM; | |
4989 | } | |
4990 | weak_alias (__posix_memalign, posix_memalign) | |
4991 | ||
20c13899 OB |
4992 | |
4993 | int | |
4994 | malloc_info (int options, FILE *fp) | |
bb066545 | 4995 | { |
20c13899 OB |
4996 | /* For now, at least. */ |
4997 | if (options != 0) | |
4998 | return EINVAL; | |
bb066545 | 4999 | |
20c13899 OB |
5000 | int n = 0; |
5001 | size_t total_nblocks = 0; | |
5002 | size_t total_nfastblocks = 0; | |
5003 | size_t total_avail = 0; | |
5004 | size_t total_fastavail = 0; | |
5005 | size_t total_system = 0; | |
5006 | size_t total_max_system = 0; | |
5007 | size_t total_aspace = 0; | |
5008 | size_t total_aspace_mprotect = 0; | |
bb066545 | 5009 | |
6c8dbf00 | 5010 | |
6c8dbf00 | 5011 | |
987c0269 OB |
5012 | if (__malloc_initialized < 0) |
5013 | ptmalloc_init (); | |
bb066545 | 5014 | |
987c0269 | 5015 | fputs ("<malloc version=\"1\">\n", fp); |
bb066545 | 5016 | |
987c0269 OB |
5017 | /* Iterate over all arenas currently in use. */ |
5018 | mstate ar_ptr = &main_arena; | |
5019 | do | |
5020 | { | |
5021 | fprintf (fp, "<heap nr=\"%d\">\n<sizes>\n", n++); | |
8b35e35d | 5022 | |
987c0269 OB |
5023 | size_t nblocks = 0; |
5024 | size_t nfastblocks = 0; | |
5025 | size_t avail = 0; | |
5026 | size_t fastavail = 0; | |
5027 | struct | |
5028 | { | |
5029 | size_t from; | |
5030 | size_t to; | |
5031 | size_t total; | |
5032 | size_t count; | |
5033 | } sizes[NFASTBINS + NBINS - 1]; | |
5034 | #define nsizes (sizeof (sizes) / sizeof (sizes[0])) | |
6c8dbf00 | 5035 | |
987c0269 | 5036 | mutex_lock (&ar_ptr->mutex); |
bb066545 | 5037 | |
987c0269 OB |
5038 | for (size_t i = 0; i < NFASTBINS; ++i) |
5039 | { | |
5040 | mchunkptr p = fastbin (ar_ptr, i); | |
5041 | if (p != NULL) | |
5042 | { | |
5043 | size_t nthissize = 0; | |
5044 | size_t thissize = chunksize (p); | |
5045 | ||
5046 | while (p != NULL) | |
5047 | { | |
5048 | ++nthissize; | |
5049 | p = p->fd; | |
5050 | } | |
5051 | ||
5052 | fastavail += nthissize * thissize; | |
5053 | nfastblocks += nthissize; | |
5054 | sizes[i].from = thissize - (MALLOC_ALIGNMENT - 1); | |
5055 | sizes[i].to = thissize; | |
5056 | sizes[i].count = nthissize; | |
5057 | } | |
5058 | else | |
5059 | sizes[i].from = sizes[i].to = sizes[i].count = 0; | |
bb066545 | 5060 | |
987c0269 OB |
5061 | sizes[i].total = sizes[i].count * sizes[i].to; |
5062 | } | |
bb066545 | 5063 | |
bb066545 | 5064 | |
987c0269 OB |
5065 | mbinptr bin; |
5066 | struct malloc_chunk *r; | |
bb066545 | 5067 | |
987c0269 OB |
5068 | for (size_t i = 1; i < NBINS; ++i) |
5069 | { | |
5070 | bin = bin_at (ar_ptr, i); | |
5071 | r = bin->fd; | |
5072 | sizes[NFASTBINS - 1 + i].from = ~((size_t) 0); | |
5073 | sizes[NFASTBINS - 1 + i].to = sizes[NFASTBINS - 1 + i].total | |
5074 | = sizes[NFASTBINS - 1 + i].count = 0; | |
5075 | ||
5076 | if (r != NULL) | |
5077 | while (r != bin) | |
5078 | { | |
5079 | ++sizes[NFASTBINS - 1 + i].count; | |
5080 | sizes[NFASTBINS - 1 + i].total += r->size; | |
5081 | sizes[NFASTBINS - 1 + i].from | |
5082 | = MIN (sizes[NFASTBINS - 1 + i].from, r->size); | |
5083 | sizes[NFASTBINS - 1 + i].to = MAX (sizes[NFASTBINS - 1 + i].to, | |
5084 | r->size); | |
5085 | ||
5086 | r = r->fd; | |
5087 | } | |
5088 | ||
5089 | if (sizes[NFASTBINS - 1 + i].count == 0) | |
5090 | sizes[NFASTBINS - 1 + i].from = 0; | |
5091 | nblocks += sizes[NFASTBINS - 1 + i].count; | |
5092 | avail += sizes[NFASTBINS - 1 + i].total; | |
5093 | } | |
bb066545 | 5094 | |
987c0269 | 5095 | mutex_unlock (&ar_ptr->mutex); |
da2d2fb6 | 5096 | |
987c0269 OB |
5097 | total_nfastblocks += nfastblocks; |
5098 | total_fastavail += fastavail; | |
0588a9cb | 5099 | |
987c0269 OB |
5100 | total_nblocks += nblocks; |
5101 | total_avail += avail; | |
0588a9cb | 5102 | |
987c0269 OB |
5103 | for (size_t i = 0; i < nsizes; ++i) |
5104 | if (sizes[i].count != 0 && i != NFASTBINS) | |
5105 | fprintf (fp, " \ | |
5106 | <size from=\"%zu\" to=\"%zu\" total=\"%zu\" count=\"%zu\"/>\n", | |
5107 | sizes[i].from, sizes[i].to, sizes[i].total, sizes[i].count); | |
fdfd175d | 5108 | |
987c0269 OB |
5109 | if (sizes[NFASTBINS].count != 0) |
5110 | fprintf (fp, "\ | |
5111 | <unsorted from=\"%zu\" to=\"%zu\" total=\"%zu\" count=\"%zu\"/>\n", | |
5112 | sizes[NFASTBINS].from, sizes[NFASTBINS].to, | |
5113 | sizes[NFASTBINS].total, sizes[NFASTBINS].count); | |
fdfd175d | 5114 | |
987c0269 OB |
5115 | total_system += ar_ptr->system_mem; |
5116 | total_max_system += ar_ptr->max_system_mem; | |
bb066545 | 5117 | |
987c0269 OB |
5118 | fprintf (fp, |
5119 | "</sizes>\n<total type=\"fast\" count=\"%zu\" size=\"%zu\"/>\n" | |
5120 | "<total type=\"rest\" count=\"%zu\" size=\"%zu\"/>\n" | |
5121 | "<system type=\"current\" size=\"%zu\"/>\n" | |
5122 | "<system type=\"max\" size=\"%zu\"/>\n", | |
5123 | nfastblocks, fastavail, nblocks, avail, | |
5124 | ar_ptr->system_mem, ar_ptr->max_system_mem); | |
346bc35c | 5125 | |
987c0269 OB |
5126 | if (ar_ptr != &main_arena) |
5127 | { | |
5128 | heap_info *heap = heap_for_ptr (top (ar_ptr)); | |
5129 | fprintf (fp, | |
5130 | "<aspace type=\"total\" size=\"%zu\"/>\n" | |
5131 | "<aspace type=\"mprotect\" size=\"%zu\"/>\n", | |
5132 | heap->size, heap->mprotect_size); | |
5133 | total_aspace += heap->size; | |
5134 | total_aspace_mprotect += heap->mprotect_size; | |
5135 | } | |
5136 | else | |
5137 | { | |
5138 | fprintf (fp, | |
5139 | "<aspace type=\"total\" size=\"%zu\"/>\n" | |
5140 | "<aspace type=\"mprotect\" size=\"%zu\"/>\n", | |
5141 | ar_ptr->system_mem, ar_ptr->system_mem); | |
5142 | total_aspace += ar_ptr->system_mem; | |
5143 | total_aspace_mprotect += ar_ptr->system_mem; | |
5144 | } | |
bb066545 | 5145 | |
987c0269 | 5146 | fputs ("</heap>\n", fp); |
bb066545 UD |
5147 | ar_ptr = ar_ptr->next; |
5148 | } | |
5149 | while (ar_ptr != &main_arena); | |
5150 | ||
5151 | fprintf (fp, | |
62a58816 SP |
5152 | "<total type=\"fast\" count=\"%zu\" size=\"%zu\"/>\n" |
5153 | "<total type=\"rest\" count=\"%zu\" size=\"%zu\"/>\n" | |
9fa76613 | 5154 | "<total type=\"mmap\" count=\"%d\" size=\"%zu\"/>\n" |
62a58816 SP |
5155 | "<system type=\"current\" size=\"%zu\"/>\n" |
5156 | "<system type=\"max\" size=\"%zu\"/>\n" | |
5157 | "<aspace type=\"total\" size=\"%zu\"/>\n" | |
5158 | "<aspace type=\"mprotect\" size=\"%zu\"/>\n" | |
5159 | "</malloc>\n", | |
5160 | total_nfastblocks, total_fastavail, total_nblocks, total_avail, | |
4d653a59 | 5161 | mp_.n_mmaps, mp_.mmapped_mem, |
62a58816 SP |
5162 | total_system, total_max_system, |
5163 | total_aspace, total_aspace_mprotect); | |
bb066545 UD |
5164 | |
5165 | return 0; | |
5166 | } | |
5167 | ||
5168 | ||
eba19d2b UD |
5169 | strong_alias (__libc_calloc, __calloc) weak_alias (__libc_calloc, calloc) |
5170 | strong_alias (__libc_free, __cfree) weak_alias (__libc_free, cfree) | |
5171 | strong_alias (__libc_free, __free) strong_alias (__libc_free, free) | |
5172 | strong_alias (__libc_malloc, __malloc) strong_alias (__libc_malloc, malloc) | |
5173 | strong_alias (__libc_memalign, __memalign) | |
5174 | weak_alias (__libc_memalign, memalign) | |
5175 | strong_alias (__libc_realloc, __realloc) strong_alias (__libc_realloc, realloc) | |
5176 | strong_alias (__libc_valloc, __valloc) weak_alias (__libc_valloc, valloc) | |
5177 | strong_alias (__libc_pvalloc, __pvalloc) weak_alias (__libc_pvalloc, pvalloc) | |
5178 | strong_alias (__libc_mallinfo, __mallinfo) | |
5179 | weak_alias (__libc_mallinfo, mallinfo) | |
5180 | strong_alias (__libc_mallopt, __mallopt) weak_alias (__libc_mallopt, mallopt) | |
7e3be507 UD |
5181 | |
5182 | weak_alias (__malloc_stats, malloc_stats) | |
5183 | weak_alias (__malloc_usable_size, malloc_usable_size) | |
5184 | weak_alias (__malloc_trim, malloc_trim) | |
2f6d1f1b UD |
5185 | weak_alias (__malloc_get_state, malloc_get_state) |
5186 | weak_alias (__malloc_set_state, malloc_set_state) | |
7e3be507 | 5187 | |
f65fd747 | 5188 | |
fa8d436c | 5189 | /* ------------------------------------------------------------ |
6c8dbf00 | 5190 | History: |
f65fd747 | 5191 | |
6c8dbf00 | 5192 | [see ftp://g.oswego.edu/pub/misc/malloc.c for the history of dlmalloc] |
f65fd747 | 5193 | |
6c8dbf00 | 5194 | */ |
fa8d436c UD |
5195 | /* |
5196 | * Local variables: | |
5197 | * c-basic-offset: 2 | |
5198 | * End: | |
5199 | */ |