]>
Commit | Line | Data |
---|---|---|
56137dbc | 1 | /* Malloc implementation for multiple threads without lock contention. |
bfff8b1b | 2 | Copyright (C) 1996-2017 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 | ||
bb2ce416 | 25 | There have been substantial changes made after the integration into |
da2d2fb6 UD |
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); |
fa8d436c | 87 | malloc_trim(size_t pad); |
22a89187 | 88 | malloc_usable_size(void* p); |
fa8d436c | 89 | malloc_stats(); |
f65fd747 UD |
90 | |
91 | * Vital statistics: | |
92 | ||
fa8d436c | 93 | Supported pointer representation: 4 or 8 bytes |
a9177ff5 | 94 | Supported size_t representation: 4 or 8 bytes |
f65fd747 | 95 | Note that size_t is allowed to be 4 bytes even if pointers are 8. |
fa8d436c UD |
96 | You can adjust this by defining INTERNAL_SIZE_T |
97 | ||
98 | Alignment: 2 * sizeof(size_t) (default) | |
99 | (i.e., 8 byte alignment with 4byte size_t). This suffices for | |
100 | nearly all current machines and C compilers. However, you can | |
101 | define MALLOC_ALIGNMENT to be wider than this if necessary. | |
f65fd747 | 102 | |
fa8d436c UD |
103 | Minimum overhead per allocated chunk: 4 or 8 bytes |
104 | Each malloced chunk has a hidden word of overhead holding size | |
f65fd747 UD |
105 | and status information. |
106 | ||
107 | Minimum allocated size: 4-byte ptrs: 16 bytes (including 4 overhead) | |
72f90263 | 108 | 8-byte ptrs: 24/32 bytes (including, 4/8 overhead) |
f65fd747 UD |
109 | |
110 | When a chunk is freed, 12 (for 4byte ptrs) or 20 (for 8 byte | |
111 | ptrs but 4 byte size) or 24 (for 8/8) additional bytes are | |
fa8d436c UD |
112 | needed; 4 (8) for a trailing size field and 8 (16) bytes for |
113 | free list pointers. Thus, the minimum allocatable size is | |
114 | 16/24/32 bytes. | |
f65fd747 UD |
115 | |
116 | Even a request for zero bytes (i.e., malloc(0)) returns a | |
117 | pointer to something of the minimum allocatable size. | |
118 | ||
fa8d436c UD |
119 | The maximum overhead wastage (i.e., number of extra bytes |
120 | allocated than were requested in malloc) is less than or equal | |
121 | to the minimum size, except for requests >= mmap_threshold that | |
122 | are serviced via mmap(), where the worst case wastage is 2 * | |
123 | sizeof(size_t) bytes plus the remainder from a system page (the | |
124 | minimal mmap unit); typically 4096 or 8192 bytes. | |
f65fd747 | 125 | |
a9177ff5 | 126 | Maximum allocated size: 4-byte size_t: 2^32 minus about two pages |
72f90263 | 127 | 8-byte size_t: 2^64 minus about two pages |
fa8d436c UD |
128 | |
129 | It is assumed that (possibly signed) size_t values suffice to | |
f65fd747 UD |
130 | represent chunk sizes. `Possibly signed' is due to the fact |
131 | that `size_t' may be defined on a system as either a signed or | |
fa8d436c UD |
132 | an unsigned type. The ISO C standard says that it must be |
133 | unsigned, but a few systems are known not to adhere to this. | |
134 | Additionally, even when size_t is unsigned, sbrk (which is by | |
135 | default used to obtain memory from system) accepts signed | |
136 | arguments, and may not be able to handle size_t-wide arguments | |
137 | with negative sign bit. Generally, values that would | |
138 | appear as negative after accounting for overhead and alignment | |
139 | are supported only via mmap(), which does not have this | |
140 | limitation. | |
141 | ||
142 | Requests for sizes outside the allowed range will perform an optional | |
143 | failure action and then return null. (Requests may also | |
144 | also fail because a system is out of memory.) | |
145 | ||
22a89187 | 146 | Thread-safety: thread-safe |
fa8d436c UD |
147 | |
148 | Compliance: I believe it is compliant with the 1997 Single Unix Specification | |
2b0fba75 | 149 | Also SVID/XPG, ANSI C, and probably others as well. |
f65fd747 UD |
150 | |
151 | * Synopsis of compile-time options: | |
152 | ||
153 | People have reported using previous versions of this malloc on all | |
154 | versions of Unix, sometimes by tweaking some of the defines | |
22a89187 | 155 | below. It has been tested most extensively on Solaris and Linux. |
fa8d436c UD |
156 | People also report using it in stand-alone embedded systems. |
157 | ||
158 | The implementation is in straight, hand-tuned ANSI C. It is not | |
159 | at all modular. (Sorry!) It uses a lot of macros. To be at all | |
160 | usable, this code should be compiled using an optimizing compiler | |
161 | (for example gcc -O3) that can simplify expressions and control | |
162 | paths. (FAQ: some macros import variables as arguments rather than | |
163 | declare locals because people reported that some debuggers | |
164 | otherwise get confused.) | |
165 | ||
166 | OPTION DEFAULT VALUE | |
167 | ||
168 | Compilation Environment options: | |
169 | ||
2a26ef3a | 170 | HAVE_MREMAP 0 |
fa8d436c UD |
171 | |
172 | Changing default word sizes: | |
173 | ||
174 | INTERNAL_SIZE_T size_t | |
fa8d436c UD |
175 | |
176 | Configuration and functionality options: | |
177 | ||
fa8d436c UD |
178 | USE_PUBLIC_MALLOC_WRAPPERS NOT defined |
179 | USE_MALLOC_LOCK NOT defined | |
180 | MALLOC_DEBUG NOT defined | |
181 | REALLOC_ZERO_BYTES_FREES 1 | |
fa8d436c UD |
182 | TRIM_FASTBINS 0 |
183 | ||
184 | Options for customizing MORECORE: | |
185 | ||
186 | MORECORE sbrk | |
187 | MORECORE_FAILURE -1 | |
a9177ff5 | 188 | MORECORE_CONTIGUOUS 1 |
fa8d436c UD |
189 | MORECORE_CANNOT_TRIM NOT defined |
190 | MORECORE_CLEARS 1 | |
a9177ff5 | 191 | MMAP_AS_MORECORE_SIZE (1024 * 1024) |
fa8d436c UD |
192 | |
193 | Tuning options that are also dynamically changeable via mallopt: | |
194 | ||
425ce2ed | 195 | DEFAULT_MXFAST 64 (for 32bit), 128 (for 64bit) |
fa8d436c UD |
196 | DEFAULT_TRIM_THRESHOLD 128 * 1024 |
197 | DEFAULT_TOP_PAD 0 | |
198 | DEFAULT_MMAP_THRESHOLD 128 * 1024 | |
199 | DEFAULT_MMAP_MAX 65536 | |
200 | ||
201 | There are several other #defined constants and macros that you | |
202 | probably don't want to touch unless you are extending or adapting malloc. */ | |
f65fd747 UD |
203 | |
204 | /* | |
22a89187 | 205 | void* is the pointer type that malloc should say it returns |
f65fd747 UD |
206 | */ |
207 | ||
22a89187 UD |
208 | #ifndef void |
209 | #define void void | |
210 | #endif /*void*/ | |
f65fd747 | 211 | |
fa8d436c UD |
212 | #include <stddef.h> /* for size_t */ |
213 | #include <stdlib.h> /* for getenv(), abort() */ | |
2a26ef3a | 214 | #include <unistd.h> /* for __libc_enable_secure */ |
f65fd747 | 215 | |
425ce2ed | 216 | #include <atomic.h> |
eb96ffb0 | 217 | #include <_itoa.h> |
e404fb16 | 218 | #include <bits/wordsize.h> |
425ce2ed | 219 | #include <sys/sysinfo.h> |
c56da3a3 | 220 | |
02d46fc4 UD |
221 | #include <ldsodefs.h> |
222 | ||
fa8d436c | 223 | #include <unistd.h> |
fa8d436c | 224 | #include <stdio.h> /* needed for malloc_stats */ |
8e58439c | 225 | #include <errno.h> |
f65fd747 | 226 | |
66274218 AJ |
227 | #include <shlib-compat.h> |
228 | ||
5d78bb43 UD |
229 | /* For uintptr_t. */ |
230 | #include <stdint.h> | |
f65fd747 | 231 | |
3e030bd5 UD |
232 | /* For va_arg, va_start, va_end. */ |
233 | #include <stdarg.h> | |
234 | ||
070906ff RM |
235 | /* For MIN, MAX, powerof2. */ |
236 | #include <sys/param.h> | |
237 | ||
ca6be165 | 238 | /* For ALIGN_UP et. al. */ |
9090848d | 239 | #include <libc-pointer-arith.h> |
8a35c3fe | 240 | |
d5c3fafc DD |
241 | /* For DIAG_PUSH/POP_NEEDS_COMMENT et al. */ |
242 | #include <libc-diag.h> | |
243 | ||
29d79486 | 244 | #include <malloc/malloc-internal.h> |
c0f62c56 | 245 | |
fa8d436c UD |
246 | /* |
247 | Debugging: | |
248 | ||
249 | Because freed chunks may be overwritten with bookkeeping fields, this | |
250 | malloc will often die when freed memory is overwritten by user | |
251 | programs. This can be very effective (albeit in an annoying way) | |
252 | in helping track down dangling pointers. | |
253 | ||
254 | If you compile with -DMALLOC_DEBUG, a number of assertion checks are | |
255 | enabled that will catch more memory errors. You probably won't be | |
256 | able to make much sense of the actual assertion errors, but they | |
257 | should help you locate incorrectly overwritten memory. The checking | |
258 | is fairly extensive, and will slow down execution | |
259 | noticeably. Calling malloc_stats or mallinfo with MALLOC_DEBUG set | |
260 | will attempt to check every non-mmapped allocated and free chunk in | |
261 | the course of computing the summmaries. (By nature, mmapped regions | |
262 | cannot be checked very much automatically.) | |
263 | ||
264 | Setting MALLOC_DEBUG may also be helpful if you are trying to modify | |
265 | this code. The assertions in the check routines spell out in more | |
266 | detail the assumptions and invariants underlying the algorithms. | |
267 | ||
268 | Setting MALLOC_DEBUG does NOT provide an automated mechanism for | |
269 | checking that all accesses to malloced memory stay within their | |
270 | bounds. However, there are several add-ons and adaptations of this | |
271 | or other mallocs available that do this. | |
f65fd747 UD |
272 | */ |
273 | ||
439bda32 WN |
274 | #ifndef MALLOC_DEBUG |
275 | #define MALLOC_DEBUG 0 | |
276 | #endif | |
277 | ||
72f90263 UD |
278 | #ifdef NDEBUG |
279 | # define assert(expr) ((void) 0) | |
280 | #else | |
281 | # define assert(expr) \ | |
282 | ((expr) \ | |
283 | ? ((void) 0) \ | |
8ba14398 | 284 | : __malloc_assert (#expr, __FILE__, __LINE__, __func__)) |
72f90263 UD |
285 | |
286 | extern const char *__progname; | |
287 | ||
288 | static void | |
289 | __malloc_assert (const char *assertion, const char *file, unsigned int line, | |
290 | const char *function) | |
291 | { | |
292 | (void) __fxprintf (NULL, "%s%s%s:%u: %s%sAssertion `%s' failed.\n", | |
293 | __progname, __progname[0] ? ": " : "", | |
294 | file, line, | |
295 | function ? function : "", function ? ": " : "", | |
296 | assertion); | |
297 | fflush (stderr); | |
298 | abort (); | |
299 | } | |
300 | #endif | |
f65fd747 | 301 | |
d5c3fafc DD |
302 | #if USE_TCACHE |
303 | /* We want 64 entries. This is an arbitrary limit, which tunables can reduce. */ | |
304 | # define TCACHE_MAX_BINS 64 | |
305 | # define MAX_TCACHE_SIZE tidx2usize (TCACHE_MAX_BINS-1) | |
306 | ||
307 | /* Only used to pre-fill the tunables. */ | |
308 | # define tidx2usize(idx) (((size_t) idx) * MALLOC_ALIGNMENT + MINSIZE - SIZE_SZ) | |
309 | ||
310 | /* When "x" is from chunksize(). */ | |
311 | # define csize2tidx(x) (((x) - MINSIZE + MALLOC_ALIGNMENT - 1) / MALLOC_ALIGNMENT) | |
312 | /* When "x" is a user-provided size. */ | |
313 | # define usize2tidx(x) csize2tidx (request2size (x)) | |
314 | ||
315 | /* With rounding and alignment, the bins are... | |
316 | idx 0 bytes 0..24 (64-bit) or 0..12 (32-bit) | |
317 | idx 1 bytes 25..40 or 13..20 | |
318 | idx 2 bytes 41..56 or 21..28 | |
319 | etc. */ | |
320 | ||
321 | /* This is another arbitrary limit, which tunables can change. Each | |
322 | tcache bin will hold at most this number of chunks. */ | |
323 | # define TCACHE_FILL_COUNT 7 | |
324 | #endif | |
325 | ||
f65fd747 | 326 | |
fa8d436c UD |
327 | /* |
328 | REALLOC_ZERO_BYTES_FREES should be set if a call to | |
329 | realloc with zero bytes should be the same as a call to free. | |
330 | This is required by the C standard. Otherwise, since this malloc | |
331 | returns a unique pointer for malloc(0), so does realloc(p, 0). | |
332 | */ | |
333 | ||
334 | #ifndef REALLOC_ZERO_BYTES_FREES | |
335 | #define REALLOC_ZERO_BYTES_FREES 1 | |
336 | #endif | |
337 | ||
338 | /* | |
339 | TRIM_FASTBINS controls whether free() of a very small chunk can | |
340 | immediately lead to trimming. Setting to true (1) can reduce memory | |
341 | footprint, but will almost always slow down programs that use a lot | |
342 | of small chunks. | |
343 | ||
344 | Define this only if you are willing to give up some speed to more | |
345 | aggressively reduce system-level memory footprint when releasing | |
346 | memory in programs that use many small chunks. You can get | |
347 | essentially the same effect by setting MXFAST to 0, but this can | |
348 | lead to even greater slowdowns in programs using many small chunks. | |
349 | TRIM_FASTBINS is an in-between compile-time option, that disables | |
350 | only those chunks bordering topmost memory from being placed in | |
351 | fastbins. | |
352 | */ | |
353 | ||
354 | #ifndef TRIM_FASTBINS | |
355 | #define TRIM_FASTBINS 0 | |
356 | #endif | |
357 | ||
358 | ||
3b49edc0 | 359 | /* Definition for getting more memory from the OS. */ |
fa8d436c UD |
360 | #define MORECORE (*__morecore) |
361 | #define MORECORE_FAILURE 0 | |
22a89187 UD |
362 | void * __default_morecore (ptrdiff_t); |
363 | void *(*__morecore)(ptrdiff_t) = __default_morecore; | |
f65fd747 | 364 | |
f65fd747 | 365 | |
22a89187 | 366 | #include <string.h> |
f65fd747 | 367 | |
fa8d436c UD |
368 | /* |
369 | MORECORE-related declarations. By default, rely on sbrk | |
370 | */ | |
09f5e163 | 371 | |
f65fd747 | 372 | |
fa8d436c UD |
373 | /* |
374 | MORECORE is the name of the routine to call to obtain more memory | |
375 | from the system. See below for general guidance on writing | |
376 | alternative MORECORE functions, as well as a version for WIN32 and a | |
377 | sample version for pre-OSX macos. | |
378 | */ | |
f65fd747 | 379 | |
fa8d436c UD |
380 | #ifndef MORECORE |
381 | #define MORECORE sbrk | |
382 | #endif | |
f65fd747 | 383 | |
fa8d436c UD |
384 | /* |
385 | MORECORE_FAILURE is the value returned upon failure of MORECORE | |
386 | as well as mmap. Since it cannot be an otherwise valid memory address, | |
387 | and must reflect values of standard sys calls, you probably ought not | |
388 | try to redefine it. | |
389 | */ | |
09f5e163 | 390 | |
fa8d436c UD |
391 | #ifndef MORECORE_FAILURE |
392 | #define MORECORE_FAILURE (-1) | |
393 | #endif | |
394 | ||
395 | /* | |
396 | If MORECORE_CONTIGUOUS is true, take advantage of fact that | |
397 | consecutive calls to MORECORE with positive arguments always return | |
398 | contiguous increasing addresses. This is true of unix sbrk. Even | |
399 | if not defined, when regions happen to be contiguous, malloc will | |
400 | permit allocations spanning regions obtained from different | |
401 | calls. But defining this when applicable enables some stronger | |
402 | consistency checks and space efficiencies. | |
403 | */ | |
f65fd747 | 404 | |
fa8d436c UD |
405 | #ifndef MORECORE_CONTIGUOUS |
406 | #define MORECORE_CONTIGUOUS 1 | |
f65fd747 UD |
407 | #endif |
408 | ||
fa8d436c UD |
409 | /* |
410 | Define MORECORE_CANNOT_TRIM if your version of MORECORE | |
411 | cannot release space back to the system when given negative | |
412 | arguments. This is generally necessary only if you are using | |
413 | a hand-crafted MORECORE function that cannot handle negative arguments. | |
414 | */ | |
415 | ||
416 | /* #define MORECORE_CANNOT_TRIM */ | |
f65fd747 | 417 | |
fa8d436c UD |
418 | /* MORECORE_CLEARS (default 1) |
419 | The degree to which the routine mapped to MORECORE zeroes out | |
420 | memory: never (0), only for newly allocated space (1) or always | |
421 | (2). The distinction between (1) and (2) is necessary because on | |
422 | some systems, if the application first decrements and then | |
423 | increments the break value, the contents of the reallocated space | |
424 | are unspecified. | |
6c8dbf00 | 425 | */ |
fa8d436c UD |
426 | |
427 | #ifndef MORECORE_CLEARS | |
6c8dbf00 | 428 | # define MORECORE_CLEARS 1 |
7cabd57c UD |
429 | #endif |
430 | ||
fa8d436c | 431 | |
a9177ff5 | 432 | /* |
fa8d436c | 433 | MMAP_AS_MORECORE_SIZE is the minimum mmap size argument to use if |
22a89187 UD |
434 | sbrk fails, and mmap is used as a backup. The value must be a |
435 | multiple of page size. This backup strategy generally applies only | |
436 | when systems have "holes" in address space, so sbrk cannot perform | |
437 | contiguous expansion, but there is still space available on system. | |
438 | On systems for which this is known to be useful (i.e. most linux | |
439 | kernels), this occurs only when programs allocate huge amounts of | |
440 | memory. Between this, and the fact that mmap regions tend to be | |
441 | limited, the size should be large, to avoid too many mmap calls and | |
442 | thus avoid running out of kernel resources. */ | |
fa8d436c UD |
443 | |
444 | #ifndef MMAP_AS_MORECORE_SIZE | |
445 | #define MMAP_AS_MORECORE_SIZE (1024 * 1024) | |
f65fd747 UD |
446 | #endif |
447 | ||
448 | /* | |
449 | Define HAVE_MREMAP to make realloc() use mremap() to re-allocate | |
2a26ef3a | 450 | large blocks. |
f65fd747 UD |
451 | */ |
452 | ||
453 | #ifndef HAVE_MREMAP | |
fa8d436c | 454 | #define HAVE_MREMAP 0 |
f65fd747 UD |
455 | #endif |
456 | ||
2ba3cfa1 FW |
457 | /* We may need to support __malloc_initialize_hook for backwards |
458 | compatibility. */ | |
459 | ||
460 | #if SHLIB_COMPAT (libc, GLIBC_2_0, GLIBC_2_24) | |
461 | # define HAVE_MALLOC_INIT_HOOK 1 | |
462 | #else | |
463 | # define HAVE_MALLOC_INIT_HOOK 0 | |
464 | #endif | |
465 | ||
f65fd747 | 466 | |
f65fd747 | 467 | /* |
f65fd747 | 468 | This version of malloc supports the standard SVID/XPG mallinfo |
fa8d436c UD |
469 | routine that returns a struct containing usage properties and |
470 | statistics. It should work on any SVID/XPG compliant system that has | |
471 | a /usr/include/malloc.h defining struct mallinfo. (If you'd like to | |
472 | install such a thing yourself, cut out the preliminary declarations | |
473 | as described above and below and save them in a malloc.h file. But | |
474 | there's no compelling reason to bother to do this.) | |
f65fd747 UD |
475 | |
476 | The main declaration needed is the mallinfo struct that is returned | |
477 | (by-copy) by mallinfo(). The SVID/XPG malloinfo struct contains a | |
fa8d436c UD |
478 | bunch of fields that are not even meaningful in this version of |
479 | malloc. These fields are are instead filled by mallinfo() with | |
480 | other numbers that might be of interest. | |
f65fd747 UD |
481 | */ |
482 | ||
f65fd747 | 483 | |
fa8d436c | 484 | /* ---------- description of public routines ------------ */ |
f65fd747 UD |
485 | |
486 | /* | |
fa8d436c UD |
487 | malloc(size_t n) |
488 | Returns a pointer to a newly allocated chunk of at least n bytes, or null | |
489 | if no space is available. Additionally, on failure, errno is | |
490 | set to ENOMEM on ANSI C systems. | |
491 | ||
492 | If n is zero, malloc returns a minumum-sized chunk. (The minimum | |
493 | size is 16 bytes on most 32bit systems, and 24 or 32 bytes on 64bit | |
494 | systems.) On most systems, size_t is an unsigned type, so calls | |
495 | with negative arguments are interpreted as requests for huge amounts | |
496 | of space, which will often fail. The maximum supported value of n | |
497 | differs across systems, but is in all cases less than the maximum | |
498 | representable value of a size_t. | |
f65fd747 | 499 | */ |
3b49edc0 UD |
500 | void* __libc_malloc(size_t); |
501 | libc_hidden_proto (__libc_malloc) | |
f65fd747 | 502 | |
fa8d436c | 503 | /* |
22a89187 | 504 | free(void* p) |
fa8d436c UD |
505 | Releases the chunk of memory pointed to by p, that had been previously |
506 | allocated using malloc or a related routine such as realloc. | |
507 | It has no effect if p is null. It can have arbitrary (i.e., bad!) | |
508 | effects if p has already been freed. | |
509 | ||
510 | Unless disabled (using mallopt), freeing very large spaces will | |
511 | when possible, automatically trigger operations that give | |
512 | back unused memory to the system, thus reducing program footprint. | |
513 | */ | |
3b49edc0 UD |
514 | void __libc_free(void*); |
515 | libc_hidden_proto (__libc_free) | |
f65fd747 | 516 | |
fa8d436c UD |
517 | /* |
518 | calloc(size_t n_elements, size_t element_size); | |
519 | Returns a pointer to n_elements * element_size bytes, with all locations | |
520 | set to zero. | |
521 | */ | |
3b49edc0 | 522 | void* __libc_calloc(size_t, size_t); |
f65fd747 UD |
523 | |
524 | /* | |
22a89187 | 525 | realloc(void* p, size_t n) |
fa8d436c UD |
526 | Returns a pointer to a chunk of size n that contains the same data |
527 | as does chunk p up to the minimum of (n, p's size) bytes, or null | |
a9177ff5 | 528 | if no space is available. |
f65fd747 | 529 | |
fa8d436c UD |
530 | The returned pointer may or may not be the same as p. The algorithm |
531 | prefers extending p when possible, otherwise it employs the | |
532 | equivalent of a malloc-copy-free sequence. | |
f65fd747 | 533 | |
a9177ff5 | 534 | If p is null, realloc is equivalent to malloc. |
f65fd747 | 535 | |
fa8d436c UD |
536 | If space is not available, realloc returns null, errno is set (if on |
537 | ANSI) and p is NOT freed. | |
f65fd747 | 538 | |
fa8d436c UD |
539 | if n is for fewer bytes than already held by p, the newly unused |
540 | space is lopped off and freed if possible. Unless the #define | |
541 | REALLOC_ZERO_BYTES_FREES is set, realloc with a size argument of | |
542 | zero (re)allocates a minimum-sized chunk. | |
f65fd747 | 543 | |
3b5f801d DD |
544 | Large chunks that were internally obtained via mmap will always be |
545 | grown using malloc-copy-free sequences unless the system supports | |
546 | MREMAP (currently only linux). | |
f65fd747 | 547 | |
fa8d436c UD |
548 | The old unix realloc convention of allowing the last-free'd chunk |
549 | to be used as an argument to realloc is not supported. | |
f65fd747 | 550 | */ |
3b49edc0 UD |
551 | void* __libc_realloc(void*, size_t); |
552 | libc_hidden_proto (__libc_realloc) | |
f65fd747 | 553 | |
fa8d436c UD |
554 | /* |
555 | memalign(size_t alignment, size_t n); | |
556 | Returns a pointer to a newly allocated chunk of n bytes, aligned | |
557 | in accord with the alignment argument. | |
558 | ||
559 | The alignment argument should be a power of two. If the argument is | |
560 | not a power of two, the nearest greater power is used. | |
561 | 8-byte alignment is guaranteed by normal malloc calls, so don't | |
562 | bother calling memalign with an argument of 8 or less. | |
563 | ||
564 | Overreliance on memalign is a sure way to fragment space. | |
565 | */ | |
3b49edc0 UD |
566 | void* __libc_memalign(size_t, size_t); |
567 | libc_hidden_proto (__libc_memalign) | |
f65fd747 UD |
568 | |
569 | /* | |
fa8d436c UD |
570 | valloc(size_t n); |
571 | Equivalent to memalign(pagesize, n), where pagesize is the page | |
572 | size of the system. If the pagesize is unknown, 4096 is used. | |
573 | */ | |
3b49edc0 | 574 | void* __libc_valloc(size_t); |
fa8d436c | 575 | |
f65fd747 | 576 | |
f65fd747 | 577 | |
fa8d436c UD |
578 | /* |
579 | mallopt(int parameter_number, int parameter_value) | |
580 | Sets tunable parameters The format is to provide a | |
581 | (parameter-number, parameter-value) pair. mallopt then sets the | |
582 | corresponding parameter to the argument value if it can (i.e., so | |
583 | long as the value is meaningful), and returns 1 if successful else | |
584 | 0. SVID/XPG/ANSI defines four standard param numbers for mallopt, | |
585 | normally defined in malloc.h. Only one of these (M_MXFAST) is used | |
586 | in this malloc. The others (M_NLBLKS, M_GRAIN, M_KEEP) don't apply, | |
587 | so setting them has no effect. But this malloc also supports four | |
588 | other options in mallopt. See below for details. Briefly, supported | |
589 | parameters are as follows (listed defaults are for "typical" | |
590 | configurations). | |
591 | ||
592 | Symbol param # default allowed param values | |
593 | M_MXFAST 1 64 0-80 (0 disables fastbins) | |
594 | M_TRIM_THRESHOLD -1 128*1024 any (-1U disables trimming) | |
a9177ff5 | 595 | M_TOP_PAD -2 0 any |
fa8d436c UD |
596 | M_MMAP_THRESHOLD -3 128*1024 any (or 0 if no MMAP support) |
597 | M_MMAP_MAX -4 65536 any (0 disables use of mmap) | |
598 | */ | |
3b49edc0 UD |
599 | int __libc_mallopt(int, int); |
600 | libc_hidden_proto (__libc_mallopt) | |
fa8d436c UD |
601 | |
602 | ||
603 | /* | |
604 | mallinfo() | |
605 | Returns (by copy) a struct containing various summary statistics: | |
606 | ||
a9177ff5 RM |
607 | arena: current total non-mmapped bytes allocated from system |
608 | ordblks: the number of free chunks | |
fa8d436c | 609 | smblks: the number of fastbin blocks (i.e., small chunks that |
72f90263 | 610 | have been freed but not use resused or consolidated) |
a9177ff5 RM |
611 | hblks: current number of mmapped regions |
612 | hblkhd: total bytes held in mmapped regions | |
ca135f82 | 613 | usmblks: always 0 |
a9177ff5 | 614 | fsmblks: total bytes held in fastbin blocks |
fa8d436c | 615 | uordblks: current total allocated space (normal or mmapped) |
a9177ff5 | 616 | fordblks: total free space |
fa8d436c | 617 | keepcost: the maximum number of bytes that could ideally be released |
72f90263 UD |
618 | back to system via malloc_trim. ("ideally" means that |
619 | it ignores page restrictions etc.) | |
fa8d436c UD |
620 | |
621 | Because these fields are ints, but internal bookkeeping may | |
a9177ff5 | 622 | be kept as longs, the reported values may wrap around zero and |
fa8d436c UD |
623 | thus be inaccurate. |
624 | */ | |
3b49edc0 | 625 | struct mallinfo __libc_mallinfo(void); |
88764ae2 | 626 | |
f65fd747 | 627 | |
fa8d436c UD |
628 | /* |
629 | pvalloc(size_t n); | |
630 | Equivalent to valloc(minimum-page-that-holds(n)), that is, | |
631 | round up n to nearest pagesize. | |
632 | */ | |
3b49edc0 | 633 | void* __libc_pvalloc(size_t); |
fa8d436c UD |
634 | |
635 | /* | |
636 | malloc_trim(size_t pad); | |
637 | ||
638 | If possible, gives memory back to the system (via negative | |
639 | arguments to sbrk) if there is unused memory at the `high' end of | |
640 | the malloc pool. You can call this after freeing large blocks of | |
641 | memory to potentially reduce the system-level memory requirements | |
642 | of a program. However, it cannot guarantee to reduce memory. Under | |
643 | some allocation patterns, some large free blocks of memory will be | |
644 | locked between two used chunks, so they cannot be given back to | |
645 | the system. | |
a9177ff5 | 646 | |
fa8d436c UD |
647 | The `pad' argument to malloc_trim represents the amount of free |
648 | trailing space to leave untrimmed. If this argument is zero, | |
649 | only the minimum amount of memory to maintain internal data | |
650 | structures will be left (one page or less). Non-zero arguments | |
651 | can be supplied to maintain enough trailing space to service | |
652 | future expected allocations without having to re-obtain memory | |
653 | from the system. | |
a9177ff5 | 654 | |
fa8d436c UD |
655 | Malloc_trim returns 1 if it actually released any memory, else 0. |
656 | On systems that do not support "negative sbrks", it will always | |
c958a6a4 | 657 | return 0. |
fa8d436c | 658 | */ |
3b49edc0 | 659 | int __malloc_trim(size_t); |
fa8d436c UD |
660 | |
661 | /* | |
22a89187 | 662 | malloc_usable_size(void* p); |
fa8d436c UD |
663 | |
664 | Returns the number of bytes you can actually use in | |
665 | an allocated chunk, which may be more than you requested (although | |
666 | often not) due to alignment and minimum size constraints. | |
667 | You can use this many bytes without worrying about | |
668 | overwriting other allocated objects. This is not a particularly great | |
669 | programming practice. malloc_usable_size can be more useful in | |
670 | debugging and assertions, for example: | |
671 | ||
672 | p = malloc(n); | |
673 | assert(malloc_usable_size(p) >= 256); | |
674 | ||
675 | */ | |
3b49edc0 | 676 | size_t __malloc_usable_size(void*); |
fa8d436c UD |
677 | |
678 | /* | |
679 | malloc_stats(); | |
680 | Prints on stderr the amount of space obtained from the system (both | |
681 | via sbrk and mmap), the maximum amount (which may be more than | |
682 | current if malloc_trim and/or munmap got called), and the current | |
683 | number of bytes allocated via malloc (or realloc, etc) but not yet | |
684 | freed. Note that this is the number of bytes allocated, not the | |
685 | number requested. It will be larger than the number requested | |
686 | because of alignment and bookkeeping overhead. Because it includes | |
687 | alignment wastage as being in use, this figure may be greater than | |
688 | zero even when no user-level chunks are allocated. | |
689 | ||
690 | The reported current and maximum system memory can be inaccurate if | |
691 | a program makes other calls to system memory allocation functions | |
692 | (normally sbrk) outside of malloc. | |
693 | ||
694 | malloc_stats prints only the most commonly interesting statistics. | |
695 | More information can be obtained by calling mallinfo. | |
696 | ||
697 | */ | |
3b49edc0 | 698 | void __malloc_stats(void); |
f65fd747 | 699 | |
f7ddf3d3 UD |
700 | /* |
701 | malloc_get_state(void); | |
702 | ||
703 | Returns the state of all malloc variables in an opaque data | |
704 | structure. | |
705 | */ | |
3b49edc0 | 706 | void* __malloc_get_state(void); |
f7ddf3d3 UD |
707 | |
708 | /* | |
22a89187 | 709 | malloc_set_state(void* state); |
f7ddf3d3 UD |
710 | |
711 | Restore the state of all malloc variables from data obtained with | |
712 | malloc_get_state(). | |
713 | */ | |
3b49edc0 | 714 | int __malloc_set_state(void*); |
f7ddf3d3 | 715 | |
f7ddf3d3 UD |
716 | /* |
717 | posix_memalign(void **memptr, size_t alignment, size_t size); | |
718 | ||
719 | POSIX wrapper like memalign(), checking for validity of size. | |
720 | */ | |
721 | int __posix_memalign(void **, size_t, size_t); | |
f7ddf3d3 | 722 | |
fa8d436c UD |
723 | /* mallopt tuning options */ |
724 | ||
f65fd747 | 725 | /* |
fa8d436c UD |
726 | M_MXFAST is the maximum request size used for "fastbins", special bins |
727 | that hold returned chunks without consolidating their spaces. This | |
728 | enables future requests for chunks of the same size to be handled | |
729 | very quickly, but can increase fragmentation, and thus increase the | |
730 | overall memory footprint of a program. | |
731 | ||
732 | This malloc manages fastbins very conservatively yet still | |
733 | efficiently, so fragmentation is rarely a problem for values less | |
734 | than or equal to the default. The maximum supported value of MXFAST | |
735 | is 80. You wouldn't want it any higher than this anyway. Fastbins | |
736 | are designed especially for use with many small structs, objects or | |
737 | strings -- the default handles structs/objects/arrays with sizes up | |
738 | to 8 4byte fields, or small strings representing words, tokens, | |
739 | etc. Using fastbins for larger objects normally worsens | |
740 | fragmentation without improving speed. | |
741 | ||
742 | M_MXFAST is set in REQUEST size units. It is internally used in | |
743 | chunksize units, which adds padding and alignment. You can reduce | |
744 | M_MXFAST to 0 to disable all use of fastbins. This causes the malloc | |
745 | algorithm to be a closer approximation of fifo-best-fit in all cases, | |
746 | not just for larger requests, but will generally cause it to be | |
747 | slower. | |
f65fd747 UD |
748 | */ |
749 | ||
750 | ||
fa8d436c UD |
751 | /* M_MXFAST is a standard SVID/XPG tuning option, usually listed in malloc.h */ |
752 | #ifndef M_MXFAST | |
a9177ff5 | 753 | #define M_MXFAST 1 |
fa8d436c | 754 | #endif |
f65fd747 | 755 | |
fa8d436c | 756 | #ifndef DEFAULT_MXFAST |
425ce2ed | 757 | #define DEFAULT_MXFAST (64 * SIZE_SZ / 4) |
10dc2a90 UD |
758 | #endif |
759 | ||
10dc2a90 | 760 | |
fa8d436c UD |
761 | /* |
762 | M_TRIM_THRESHOLD is the maximum amount of unused top-most memory | |
763 | to keep before releasing via malloc_trim in free(). | |
764 | ||
765 | Automatic trimming is mainly useful in long-lived programs. | |
766 | Because trimming via sbrk can be slow on some systems, and can | |
767 | sometimes be wasteful (in cases where programs immediately | |
768 | afterward allocate more large chunks) the value should be high | |
769 | enough so that your overall system performance would improve by | |
770 | releasing this much memory. | |
771 | ||
772 | The trim threshold and the mmap control parameters (see below) | |
773 | can be traded off with one another. Trimming and mmapping are | |
774 | two different ways of releasing unused memory back to the | |
775 | system. Between these two, it is often possible to keep | |
776 | system-level demands of a long-lived program down to a bare | |
777 | minimum. For example, in one test suite of sessions measuring | |
778 | the XF86 X server on Linux, using a trim threshold of 128K and a | |
779 | mmap threshold of 192K led to near-minimal long term resource | |
780 | consumption. | |
781 | ||
782 | If you are using this malloc in a long-lived program, it should | |
783 | pay to experiment with these values. As a rough guide, you | |
784 | might set to a value close to the average size of a process | |
785 | (program) running on your system. Releasing this much memory | |
786 | would allow such a process to run in memory. Generally, it's | |
787 | worth it to tune for trimming rather tham memory mapping when a | |
788 | program undergoes phases where several large chunks are | |
789 | allocated and released in ways that can reuse each other's | |
790 | storage, perhaps mixed with phases where there are no such | |
791 | chunks at all. And in well-behaved long-lived programs, | |
792 | controlling release of large blocks via trimming versus mapping | |
793 | is usually faster. | |
794 | ||
795 | However, in most programs, these parameters serve mainly as | |
796 | protection against the system-level effects of carrying around | |
797 | massive amounts of unneeded memory. Since frequent calls to | |
798 | sbrk, mmap, and munmap otherwise degrade performance, the default | |
799 | parameters are set to relatively high values that serve only as | |
800 | safeguards. | |
801 | ||
802 | The trim value It must be greater than page size to have any useful | |
a9177ff5 | 803 | effect. To disable trimming completely, you can set to |
fa8d436c UD |
804 | (unsigned long)(-1) |
805 | ||
806 | Trim settings interact with fastbin (MXFAST) settings: Unless | |
807 | TRIM_FASTBINS is defined, automatic trimming never takes place upon | |
808 | freeing a chunk with size less than or equal to MXFAST. Trimming is | |
809 | instead delayed until subsequent freeing of larger chunks. However, | |
810 | you can still force an attempted trim by calling malloc_trim. | |
811 | ||
812 | Also, trimming is not generally possible in cases where | |
813 | the main arena is obtained via mmap. | |
814 | ||
815 | Note that the trick some people use of mallocing a huge space and | |
816 | then freeing it at program startup, in an attempt to reserve system | |
817 | memory, doesn't have the intended effect under automatic trimming, | |
818 | since that memory will immediately be returned to the system. | |
819 | */ | |
820 | ||
821 | #define M_TRIM_THRESHOLD -1 | |
822 | ||
823 | #ifndef DEFAULT_TRIM_THRESHOLD | |
824 | #define DEFAULT_TRIM_THRESHOLD (128 * 1024) | |
825 | #endif | |
826 | ||
827 | /* | |
828 | M_TOP_PAD is the amount of extra `padding' space to allocate or | |
829 | retain whenever sbrk is called. It is used in two ways internally: | |
830 | ||
831 | * When sbrk is called to extend the top of the arena to satisfy | |
832 | a new malloc request, this much padding is added to the sbrk | |
833 | request. | |
834 | ||
835 | * When malloc_trim is called automatically from free(), | |
836 | it is used as the `pad' argument. | |
837 | ||
838 | In both cases, the actual amount of padding is rounded | |
839 | so that the end of the arena is always a system page boundary. | |
840 | ||
841 | The main reason for using padding is to avoid calling sbrk so | |
842 | often. Having even a small pad greatly reduces the likelihood | |
843 | that nearly every malloc request during program start-up (or | |
844 | after trimming) will invoke sbrk, which needlessly wastes | |
845 | time. | |
846 | ||
847 | Automatic rounding-up to page-size units is normally sufficient | |
848 | to avoid measurable overhead, so the default is 0. However, in | |
849 | systems where sbrk is relatively slow, it can pay to increase | |
850 | this value, at the expense of carrying around more memory than | |
851 | the program needs. | |
852 | */ | |
10dc2a90 | 853 | |
fa8d436c | 854 | #define M_TOP_PAD -2 |
10dc2a90 | 855 | |
fa8d436c UD |
856 | #ifndef DEFAULT_TOP_PAD |
857 | #define DEFAULT_TOP_PAD (0) | |
858 | #endif | |
f65fd747 | 859 | |
1d05c2fb UD |
860 | /* |
861 | MMAP_THRESHOLD_MAX and _MIN are the bounds on the dynamically | |
862 | adjusted MMAP_THRESHOLD. | |
863 | */ | |
864 | ||
865 | #ifndef DEFAULT_MMAP_THRESHOLD_MIN | |
866 | #define DEFAULT_MMAP_THRESHOLD_MIN (128 * 1024) | |
867 | #endif | |
868 | ||
869 | #ifndef DEFAULT_MMAP_THRESHOLD_MAX | |
e404fb16 UD |
870 | /* For 32-bit platforms we cannot increase the maximum mmap |
871 | threshold much because it is also the minimum value for the | |
bd2c2341 UD |
872 | maximum heap size and its alignment. Going above 512k (i.e., 1M |
873 | for new heaps) wastes too much address space. */ | |
e404fb16 | 874 | # if __WORDSIZE == 32 |
bd2c2341 | 875 | # define DEFAULT_MMAP_THRESHOLD_MAX (512 * 1024) |
e404fb16 | 876 | # else |
bd2c2341 | 877 | # define DEFAULT_MMAP_THRESHOLD_MAX (4 * 1024 * 1024 * sizeof(long)) |
e404fb16 | 878 | # endif |
1d05c2fb UD |
879 | #endif |
880 | ||
fa8d436c UD |
881 | /* |
882 | M_MMAP_THRESHOLD is the request size threshold for using mmap() | |
883 | to service a request. Requests of at least this size that cannot | |
884 | be allocated using already-existing space will be serviced via mmap. | |
885 | (If enough normal freed space already exists it is used instead.) | |
886 | ||
887 | Using mmap segregates relatively large chunks of memory so that | |
888 | they can be individually obtained and released from the host | |
889 | system. A request serviced through mmap is never reused by any | |
890 | other request (at least not directly; the system may just so | |
891 | happen to remap successive requests to the same locations). | |
892 | ||
893 | Segregating space in this way has the benefits that: | |
894 | ||
a9177ff5 RM |
895 | 1. Mmapped space can ALWAYS be individually released back |
896 | to the system, which helps keep the system level memory | |
897 | demands of a long-lived program low. | |
fa8d436c UD |
898 | 2. Mapped memory can never become `locked' between |
899 | other chunks, as can happen with normally allocated chunks, which | |
900 | means that even trimming via malloc_trim would not release them. | |
901 | 3. On some systems with "holes" in address spaces, mmap can obtain | |
902 | memory that sbrk cannot. | |
903 | ||
904 | However, it has the disadvantages that: | |
905 | ||
906 | 1. The space cannot be reclaimed, consolidated, and then | |
907 | used to service later requests, as happens with normal chunks. | |
908 | 2. It can lead to more wastage because of mmap page alignment | |
909 | requirements | |
910 | 3. It causes malloc performance to be more dependent on host | |
911 | system memory management support routines which may vary in | |
912 | implementation quality and may impose arbitrary | |
913 | limitations. Generally, servicing a request via normal | |
914 | malloc steps is faster than going through a system's mmap. | |
915 | ||
916 | The advantages of mmap nearly always outweigh disadvantages for | |
917 | "large" chunks, but the value of "large" varies across systems. The | |
918 | default is an empirically derived value that works well in most | |
919 | systems. | |
1d05c2fb UD |
920 | |
921 | ||
922 | Update in 2006: | |
923 | The above was written in 2001. Since then the world has changed a lot. | |
924 | Memory got bigger. Applications got bigger. The virtual address space | |
925 | layout in 32 bit linux changed. | |
926 | ||
927 | In the new situation, brk() and mmap space is shared and there are no | |
928 | artificial limits on brk size imposed by the kernel. What is more, | |
929 | applications have started using transient allocations larger than the | |
930 | 128Kb as was imagined in 2001. | |
931 | ||
932 | The price for mmap is also high now; each time glibc mmaps from the | |
933 | kernel, the kernel is forced to zero out the memory it gives to the | |
934 | application. Zeroing memory is expensive and eats a lot of cache and | |
935 | memory bandwidth. This has nothing to do with the efficiency of the | |
936 | virtual memory system, by doing mmap the kernel just has no choice but | |
937 | to zero. | |
938 | ||
939 | In 2001, the kernel had a maximum size for brk() which was about 800 | |
940 | megabytes on 32 bit x86, at that point brk() would hit the first | |
941 | mmaped shared libaries and couldn't expand anymore. With current 2.6 | |
942 | kernels, the VA space layout is different and brk() and mmap | |
943 | both can span the entire heap at will. | |
944 | ||
945 | Rather than using a static threshold for the brk/mmap tradeoff, | |
946 | we are now using a simple dynamic one. The goal is still to avoid | |
947 | fragmentation. The old goals we kept are | |
948 | 1) try to get the long lived large allocations to use mmap() | |
949 | 2) really large allocations should always use mmap() | |
950 | and we're adding now: | |
951 | 3) transient allocations should use brk() to avoid forcing the kernel | |
952 | having to zero memory over and over again | |
953 | ||
954 | The implementation works with a sliding threshold, which is by default | |
955 | limited to go between 128Kb and 32Mb (64Mb for 64 bitmachines) and starts | |
956 | out at 128Kb as per the 2001 default. | |
957 | ||
958 | This allows us to satisfy requirement 1) under the assumption that long | |
959 | lived allocations are made early in the process' lifespan, before it has | |
960 | started doing dynamic allocations of the same size (which will | |
961 | increase the threshold). | |
962 | ||
963 | The upperbound on the threshold satisfies requirement 2) | |
964 | ||
965 | The threshold goes up in value when the application frees memory that was | |
966 | allocated with the mmap allocator. The idea is that once the application | |
967 | starts freeing memory of a certain size, it's highly probable that this is | |
968 | a size the application uses for transient allocations. This estimator | |
969 | is there to satisfy the new third requirement. | |
970 | ||
f65fd747 UD |
971 | */ |
972 | ||
fa8d436c | 973 | #define M_MMAP_THRESHOLD -3 |
f65fd747 | 974 | |
fa8d436c | 975 | #ifndef DEFAULT_MMAP_THRESHOLD |
1d05c2fb | 976 | #define DEFAULT_MMAP_THRESHOLD DEFAULT_MMAP_THRESHOLD_MIN |
fa8d436c UD |
977 | #endif |
978 | ||
979 | /* | |
980 | M_MMAP_MAX is the maximum number of requests to simultaneously | |
981 | service using mmap. This parameter exists because | |
982 | some systems have a limited number of internal tables for | |
983 | use by mmap, and using more than a few of them may degrade | |
984 | performance. | |
985 | ||
986 | The default is set to a value that serves only as a safeguard. | |
22a89187 | 987 | Setting to 0 disables use of mmap for servicing large requests. |
fa8d436c | 988 | */ |
f65fd747 | 989 | |
fa8d436c UD |
990 | #define M_MMAP_MAX -4 |
991 | ||
992 | #ifndef DEFAULT_MMAP_MAX | |
fa8d436c | 993 | #define DEFAULT_MMAP_MAX (65536) |
f65fd747 UD |
994 | #endif |
995 | ||
100351c3 | 996 | #include <malloc.h> |
f65fd747 | 997 | |
fa8d436c UD |
998 | #ifndef RETURN_ADDRESS |
999 | #define RETURN_ADDRESS(X_) (NULL) | |
9ae6fc54 | 1000 | #endif |
431c33c0 | 1001 | |
fa8d436c UD |
1002 | /* Forward declarations. */ |
1003 | struct malloc_chunk; | |
1004 | typedef struct malloc_chunk* mchunkptr; | |
431c33c0 | 1005 | |
fa8d436c | 1006 | /* Internal routines. */ |
f65fd747 | 1007 | |
22a89187 | 1008 | static void* _int_malloc(mstate, size_t); |
425ce2ed | 1009 | static void _int_free(mstate, mchunkptr, int); |
22a89187 | 1010 | static void* _int_realloc(mstate, mchunkptr, INTERNAL_SIZE_T, |
6e4b2107 | 1011 | INTERNAL_SIZE_T); |
22a89187 | 1012 | static void* _int_memalign(mstate, size_t, size_t); |
10ad46bc OB |
1013 | static void* _mid_memalign(size_t, size_t, void *); |
1014 | ||
ac3ed168 | 1015 | static void malloc_printerr(const char *str) __attribute__ ((noreturn)); |
fa8d436c | 1016 | |
0c71122c FW |
1017 | static void* mem2mem_check(void *p, size_t sz); |
1018 | static void top_check(void); | |
1019 | static void munmap_chunk(mchunkptr p); | |
a9177ff5 | 1020 | #if HAVE_MREMAP |
0c71122c | 1021 | static mchunkptr mremap_chunk(mchunkptr p, size_t new_size); |
a9177ff5 | 1022 | #endif |
fa8d436c | 1023 | |
22a89187 UD |
1024 | static void* malloc_check(size_t sz, const void *caller); |
1025 | static void free_check(void* mem, const void *caller); | |
1026 | static void* realloc_check(void* oldmem, size_t bytes, | |
1027 | const void *caller); | |
1028 | static void* memalign_check(size_t alignment, size_t bytes, | |
1029 | const void *caller); | |
f65fd747 | 1030 | |
fa8d436c | 1031 | /* ------------------ MMAP support ------------------ */ |
f65fd747 | 1032 | |
f65fd747 | 1033 | |
fa8d436c | 1034 | #include <fcntl.h> |
fa8d436c | 1035 | #include <sys/mman.h> |
f65fd747 | 1036 | |
fa8d436c UD |
1037 | #if !defined(MAP_ANONYMOUS) && defined(MAP_ANON) |
1038 | # define MAP_ANONYMOUS MAP_ANON | |
1039 | #endif | |
f65fd747 | 1040 | |
fa8d436c | 1041 | #ifndef MAP_NORESERVE |
3b49edc0 | 1042 | # define MAP_NORESERVE 0 |
f65fd747 UD |
1043 | #endif |
1044 | ||
fa8d436c | 1045 | #define MMAP(addr, size, prot, flags) \ |
3b49edc0 | 1046 | __mmap((addr), (size), (prot), (flags)|MAP_ANONYMOUS|MAP_PRIVATE, -1, 0) |
f65fd747 | 1047 | |
f65fd747 UD |
1048 | |
1049 | /* | |
fa8d436c | 1050 | ----------------------- Chunk representations ----------------------- |
f65fd747 UD |
1051 | */ |
1052 | ||
1053 | ||
fa8d436c UD |
1054 | /* |
1055 | This struct declaration is misleading (but accurate and necessary). | |
1056 | It declares a "view" into memory allowing access to necessary | |
1057 | fields at known offsets from a given base. See explanation below. | |
1058 | */ | |
1059 | ||
1060 | struct malloc_chunk { | |
1061 | ||
e9c4fe93 FW |
1062 | INTERNAL_SIZE_T mchunk_prev_size; /* Size of previous chunk (if free). */ |
1063 | INTERNAL_SIZE_T mchunk_size; /* Size in bytes, including overhead. */ | |
fa8d436c UD |
1064 | |
1065 | struct malloc_chunk* fd; /* double links -- used only if free. */ | |
f65fd747 | 1066 | struct malloc_chunk* bk; |
7ecfbd38 UD |
1067 | |
1068 | /* Only used for large blocks: pointer to next larger size. */ | |
1069 | struct malloc_chunk* fd_nextsize; /* double links -- used only if free. */ | |
1070 | struct malloc_chunk* bk_nextsize; | |
f65fd747 UD |
1071 | }; |
1072 | ||
f65fd747 UD |
1073 | |
1074 | /* | |
f65fd747 UD |
1075 | malloc_chunk details: |
1076 | ||
1077 | (The following includes lightly edited explanations by Colin Plumb.) | |
1078 | ||
1079 | Chunks of memory are maintained using a `boundary tag' method as | |
1080 | described in e.g., Knuth or Standish. (See the paper by Paul | |
1081 | Wilson ftp://ftp.cs.utexas.edu/pub/garbage/allocsrv.ps for a | |
1082 | survey of such techniques.) Sizes of free chunks are stored both | |
1083 | in the front of each chunk and at the end. This makes | |
1084 | consolidating fragmented chunks into bigger chunks very fast. The | |
1085 | size fields also hold bits representing whether chunks are free or | |
1086 | in use. | |
1087 | ||
1088 | An allocated chunk looks like this: | |
1089 | ||
1090 | ||
1091 | chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
ae9166f2 | 1092 | | Size of previous chunk, if unallocated (P clear) | |
72f90263 | 1093 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
ae9166f2 | 1094 | | Size of chunk, in bytes |A|M|P| |
f65fd747 | 1095 | mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
72f90263 UD |
1096 | | User data starts here... . |
1097 | . . | |
1098 | . (malloc_usable_size() bytes) . | |
1099 | . | | |
f65fd747 | 1100 | nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
ae9166f2 FW |
1101 | | (size of chunk, but used for application data) | |
1102 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
1103 | | Size of next chunk, in bytes |A|0|1| | |
72f90263 | 1104 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
f65fd747 UD |
1105 | |
1106 | Where "chunk" is the front of the chunk for the purpose of most of | |
1107 | the malloc code, but "mem" is the pointer that is returned to the | |
1108 | user. "Nextchunk" is the beginning of the next contiguous chunk. | |
1109 | ||
6f65e668 | 1110 | Chunks always begin on even word boundaries, so the mem portion |
f65fd747 | 1111 | (which is returned to the user) is also on an even word boundary, and |
fa8d436c | 1112 | thus at least double-word aligned. |
f65fd747 UD |
1113 | |
1114 | Free chunks are stored in circular doubly-linked lists, and look like this: | |
1115 | ||
1116 | chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
ae9166f2 | 1117 | | Size of previous chunk, if unallocated (P clear) | |
72f90263 | 1118 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
ae9166f2 | 1119 | `head:' | Size of chunk, in bytes |A|0|P| |
f65fd747 | 1120 | mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
72f90263 UD |
1121 | | Forward pointer to next chunk in list | |
1122 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
1123 | | Back pointer to previous chunk in list | | |
1124 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
1125 | | Unused space (may be 0 bytes long) . | |
1126 | . . | |
1127 | . | | |
f65fd747 UD |
1128 | nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
1129 | `foot:' | Size of chunk, in bytes | | |
72f90263 | 1130 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
ae9166f2 FW |
1131 | | Size of next chunk, in bytes |A|0|0| |
1132 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | |
f65fd747 UD |
1133 | |
1134 | The P (PREV_INUSE) bit, stored in the unused low-order bit of the | |
1135 | chunk size (which is always a multiple of two words), is an in-use | |
1136 | bit for the *previous* chunk. If that bit is *clear*, then the | |
1137 | word before the current chunk size contains the previous chunk | |
1138 | size, and can be used to find the front of the previous chunk. | |
fa8d436c UD |
1139 | The very first chunk allocated always has this bit set, |
1140 | preventing access to non-existent (or non-owned) memory. If | |
1141 | prev_inuse is set for any given chunk, then you CANNOT determine | |
1142 | the size of the previous chunk, and might even get a memory | |
1143 | addressing fault when trying to do so. | |
f65fd747 | 1144 | |
ae9166f2 FW |
1145 | The A (NON_MAIN_ARENA) bit is cleared for chunks on the initial, |
1146 | main arena, described by the main_arena variable. When additional | |
1147 | threads are spawned, each thread receives its own arena (up to a | |
1148 | configurable limit, after which arenas are reused for multiple | |
1149 | threads), and the chunks in these arenas have the A bit set. To | |
1150 | find the arena for a chunk on such a non-main arena, heap_for_ptr | |
1151 | performs a bit mask operation and indirection through the ar_ptr | |
1152 | member of the per-heap header heap_info (see arena.c). | |
1153 | ||
f65fd747 | 1154 | Note that the `foot' of the current chunk is actually represented |
fa8d436c UD |
1155 | as the prev_size of the NEXT chunk. This makes it easier to |
1156 | deal with alignments etc but can be very confusing when trying | |
1157 | to extend or adapt this code. | |
f65fd747 | 1158 | |
ae9166f2 | 1159 | The three exceptions to all this are: |
f65fd747 | 1160 | |
fa8d436c | 1161 | 1. The special chunk `top' doesn't bother using the |
72f90263 UD |
1162 | trailing size field since there is no next contiguous chunk |
1163 | that would have to index off it. After initialization, `top' | |
1164 | is forced to always exist. If it would become less than | |
1165 | MINSIZE bytes long, it is replenished. | |
f65fd747 UD |
1166 | |
1167 | 2. Chunks allocated via mmap, which have the second-lowest-order | |
72f90263 | 1168 | bit M (IS_MMAPPED) set in their size fields. Because they are |
ae9166f2 FW |
1169 | allocated one-by-one, each must contain its own trailing size |
1170 | field. If the M bit is set, the other bits are ignored | |
1171 | (because mmapped chunks are neither in an arena, nor adjacent | |
1172 | to a freed chunk). The M bit is also used for chunks which | |
1173 | originally came from a dumped heap via malloc_set_state in | |
1174 | hooks.c. | |
1175 | ||
1176 | 3. Chunks in fastbins are treated as allocated chunks from the | |
1177 | point of view of the chunk allocator. They are consolidated | |
1178 | with their neighbors only in bulk, in malloc_consolidate. | |
f65fd747 UD |
1179 | */ |
1180 | ||
1181 | /* | |
fa8d436c UD |
1182 | ---------- Size and alignment checks and conversions ---------- |
1183 | */ | |
f65fd747 | 1184 | |
fa8d436c | 1185 | /* conversion from malloc headers to user pointers, and back */ |
f65fd747 | 1186 | |
22a89187 | 1187 | #define chunk2mem(p) ((void*)((char*)(p) + 2*SIZE_SZ)) |
fa8d436c | 1188 | #define mem2chunk(mem) ((mchunkptr)((char*)(mem) - 2*SIZE_SZ)) |
f65fd747 | 1189 | |
fa8d436c | 1190 | /* The smallest possible chunk */ |
7ecfbd38 | 1191 | #define MIN_CHUNK_SIZE (offsetof(struct malloc_chunk, fd_nextsize)) |
f65fd747 | 1192 | |
fa8d436c | 1193 | /* The smallest size we can malloc is an aligned minimal chunk */ |
f65fd747 | 1194 | |
fa8d436c UD |
1195 | #define MINSIZE \ |
1196 | (unsigned long)(((MIN_CHUNK_SIZE+MALLOC_ALIGN_MASK) & ~MALLOC_ALIGN_MASK)) | |
f65fd747 | 1197 | |
fa8d436c | 1198 | /* Check if m has acceptable alignment */ |
f65fd747 | 1199 | |
073f560e UD |
1200 | #define aligned_OK(m) (((unsigned long)(m) & MALLOC_ALIGN_MASK) == 0) |
1201 | ||
1202 | #define misaligned_chunk(p) \ | |
1203 | ((uintptr_t)(MALLOC_ALIGNMENT == 2 * SIZE_SZ ? (p) : chunk2mem (p)) \ | |
1204 | & MALLOC_ALIGN_MASK) | |
f65fd747 | 1205 | |
f65fd747 | 1206 | |
a9177ff5 | 1207 | /* |
fa8d436c UD |
1208 | Check if a request is so large that it would wrap around zero when |
1209 | padded and aligned. To simplify some other code, the bound is made | |
1210 | low enough so that adding MINSIZE will also not wrap around zero. | |
6c8dbf00 | 1211 | */ |
f65fd747 | 1212 | |
fa8d436c | 1213 | #define REQUEST_OUT_OF_RANGE(req) \ |
6c8dbf00 OB |
1214 | ((unsigned long) (req) >= \ |
1215 | (unsigned long) (INTERNAL_SIZE_T) (-2 * MINSIZE)) | |
f65fd747 | 1216 | |
fa8d436c | 1217 | /* pad request bytes into a usable size -- internal version */ |
f65fd747 | 1218 | |
fa8d436c UD |
1219 | #define request2size(req) \ |
1220 | (((req) + SIZE_SZ + MALLOC_ALIGN_MASK < MINSIZE) ? \ | |
1221 | MINSIZE : \ | |
1222 | ((req) + SIZE_SZ + MALLOC_ALIGN_MASK) & ~MALLOC_ALIGN_MASK) | |
f65fd747 | 1223 | |
fa8d436c | 1224 | /* Same, except also perform argument check */ |
f65fd747 | 1225 | |
fa8d436c | 1226 | #define checked_request2size(req, sz) \ |
6c8dbf00 OB |
1227 | if (REQUEST_OUT_OF_RANGE (req)) { \ |
1228 | __set_errno (ENOMEM); \ | |
1229 | return 0; \ | |
1230 | } \ | |
1231 | (sz) = request2size (req); | |
f65fd747 UD |
1232 | |
1233 | /* | |
6c8dbf00 OB |
1234 | --------------- Physical chunk operations --------------- |
1235 | */ | |
f65fd747 | 1236 | |
10dc2a90 | 1237 | |
fa8d436c UD |
1238 | /* size field is or'ed with PREV_INUSE when previous adjacent chunk in use */ |
1239 | #define PREV_INUSE 0x1 | |
f65fd747 | 1240 | |
fa8d436c | 1241 | /* extract inuse bit of previous chunk */ |
e9c4fe93 | 1242 | #define prev_inuse(p) ((p)->mchunk_size & PREV_INUSE) |
f65fd747 | 1243 | |
f65fd747 | 1244 | |
fa8d436c UD |
1245 | /* size field is or'ed with IS_MMAPPED if the chunk was obtained with mmap() */ |
1246 | #define IS_MMAPPED 0x2 | |
f65fd747 | 1247 | |
fa8d436c | 1248 | /* check for mmap()'ed chunk */ |
e9c4fe93 | 1249 | #define chunk_is_mmapped(p) ((p)->mchunk_size & IS_MMAPPED) |
f65fd747 | 1250 | |
f65fd747 | 1251 | |
fa8d436c UD |
1252 | /* size field is or'ed with NON_MAIN_ARENA if the chunk was obtained |
1253 | from a non-main arena. This is only set immediately before handing | |
1254 | the chunk to the user, if necessary. */ | |
1255 | #define NON_MAIN_ARENA 0x4 | |
f65fd747 | 1256 | |
ae9166f2 | 1257 | /* Check for chunk from main arena. */ |
e9c4fe93 FW |
1258 | #define chunk_main_arena(p) (((p)->mchunk_size & NON_MAIN_ARENA) == 0) |
1259 | ||
1260 | /* Mark a chunk as not being on the main arena. */ | |
1261 | #define set_non_main_arena(p) ((p)->mchunk_size |= NON_MAIN_ARENA) | |
f65fd747 UD |
1262 | |
1263 | ||
a9177ff5 | 1264 | /* |
6c8dbf00 | 1265 | Bits to mask off when extracting size |
f65fd747 | 1266 | |
6c8dbf00 OB |
1267 | Note: IS_MMAPPED is intentionally not masked off from size field in |
1268 | macros for which mmapped chunks should never be seen. This should | |
1269 | cause helpful core dumps to occur if it is tried by accident by | |
1270 | people extending or adapting this malloc. | |
1271 | */ | |
1272 | #define SIZE_BITS (PREV_INUSE | IS_MMAPPED | NON_MAIN_ARENA) | |
f65fd747 | 1273 | |
fa8d436c | 1274 | /* Get size, ignoring use bits */ |
e9c4fe93 | 1275 | #define chunksize(p) (chunksize_nomask (p) & ~(SIZE_BITS)) |
f65fd747 | 1276 | |
e9c4fe93 FW |
1277 | /* Like chunksize, but do not mask SIZE_BITS. */ |
1278 | #define chunksize_nomask(p) ((p)->mchunk_size) | |
f65fd747 | 1279 | |
fa8d436c | 1280 | /* Ptr to next physical malloc_chunk. */ |
e9c4fe93 FW |
1281 | #define next_chunk(p) ((mchunkptr) (((char *) (p)) + chunksize (p))) |
1282 | ||
1283 | /* Size of the chunk below P. Only valid if prev_inuse (P). */ | |
1284 | #define prev_size(p) ((p)->mchunk_prev_size) | |
1285 | ||
1286 | /* Set the size of the chunk below P. Only valid if prev_inuse (P). */ | |
1287 | #define set_prev_size(p, sz) ((p)->mchunk_prev_size = (sz)) | |
f65fd747 | 1288 | |
e9c4fe93 FW |
1289 | /* Ptr to previous physical malloc_chunk. Only valid if prev_inuse (P). */ |
1290 | #define prev_chunk(p) ((mchunkptr) (((char *) (p)) - prev_size (p))) | |
f65fd747 | 1291 | |
fa8d436c | 1292 | /* Treat space at ptr + offset as a chunk */ |
6c8dbf00 | 1293 | #define chunk_at_offset(p, s) ((mchunkptr) (((char *) (p)) + (s))) |
fa8d436c UD |
1294 | |
1295 | /* extract p's inuse bit */ | |
6c8dbf00 | 1296 | #define inuse(p) \ |
e9c4fe93 | 1297 | ((((mchunkptr) (((char *) (p)) + chunksize (p)))->mchunk_size) & PREV_INUSE) |
f65fd747 | 1298 | |
fa8d436c | 1299 | /* set/clear chunk as being inuse without otherwise disturbing */ |
6c8dbf00 | 1300 | #define set_inuse(p) \ |
e9c4fe93 | 1301 | ((mchunkptr) (((char *) (p)) + chunksize (p)))->mchunk_size |= PREV_INUSE |
f65fd747 | 1302 | |
6c8dbf00 | 1303 | #define clear_inuse(p) \ |
e9c4fe93 | 1304 | ((mchunkptr) (((char *) (p)) + chunksize (p)))->mchunk_size &= ~(PREV_INUSE) |
f65fd747 UD |
1305 | |
1306 | ||
fa8d436c | 1307 | /* check/set/clear inuse bits in known places */ |
6c8dbf00 | 1308 | #define inuse_bit_at_offset(p, s) \ |
e9c4fe93 | 1309 | (((mchunkptr) (((char *) (p)) + (s)))->mchunk_size & PREV_INUSE) |
f65fd747 | 1310 | |
6c8dbf00 | 1311 | #define set_inuse_bit_at_offset(p, s) \ |
e9c4fe93 | 1312 | (((mchunkptr) (((char *) (p)) + (s)))->mchunk_size |= PREV_INUSE) |
f65fd747 | 1313 | |
6c8dbf00 | 1314 | #define clear_inuse_bit_at_offset(p, s) \ |
e9c4fe93 | 1315 | (((mchunkptr) (((char *) (p)) + (s)))->mchunk_size &= ~(PREV_INUSE)) |
f65fd747 | 1316 | |
f65fd747 | 1317 | |
fa8d436c | 1318 | /* Set size at head, without disturbing its use bit */ |
e9c4fe93 | 1319 | #define set_head_size(p, s) ((p)->mchunk_size = (((p)->mchunk_size & SIZE_BITS) | (s))) |
f65fd747 | 1320 | |
fa8d436c | 1321 | /* Set size/use field */ |
e9c4fe93 | 1322 | #define set_head(p, s) ((p)->mchunk_size = (s)) |
f65fd747 | 1323 | |
fa8d436c | 1324 | /* Set size at footer (only when chunk is not in use) */ |
e9c4fe93 | 1325 | #define set_foot(p, s) (((mchunkptr) ((char *) (p) + (s)))->mchunk_prev_size = (s)) |
f65fd747 UD |
1326 | |
1327 | ||
e9c4fe93 FW |
1328 | #pragma GCC poison mchunk_size |
1329 | #pragma GCC poison mchunk_prev_size | |
1330 | ||
fa8d436c | 1331 | /* |
6c8dbf00 | 1332 | -------------------- Internal data structures -------------------- |
fa8d436c UD |
1333 | |
1334 | All internal state is held in an instance of malloc_state defined | |
1335 | below. There are no other static variables, except in two optional | |
a9177ff5 | 1336 | cases: |
6c8dbf00 OB |
1337 | * If USE_MALLOC_LOCK is defined, the mALLOC_MUTEx declared above. |
1338 | * If mmap doesn't support MAP_ANONYMOUS, a dummy file descriptor | |
22a89187 | 1339 | for mmap. |
fa8d436c UD |
1340 | |
1341 | Beware of lots of tricks that minimize the total bookkeeping space | |
1342 | requirements. The result is a little over 1K bytes (for 4byte | |
1343 | pointers and size_t.) | |
6c8dbf00 | 1344 | */ |
f65fd747 UD |
1345 | |
1346 | /* | |
6c8dbf00 | 1347 | Bins |
fa8d436c UD |
1348 | |
1349 | An array of bin headers for free chunks. Each bin is doubly | |
1350 | linked. The bins are approximately proportionally (log) spaced. | |
1351 | There are a lot of these bins (128). This may look excessive, but | |
1352 | works very well in practice. Most bins hold sizes that are | |
1353 | unusual as malloc request sizes, but are more usual for fragments | |
1354 | and consolidated sets of chunks, which is what these bins hold, so | |
1355 | they can be found quickly. All procedures maintain the invariant | |
1356 | that no consolidated chunk physically borders another one, so each | |
1357 | chunk in a list is known to be preceeded and followed by either | |
1358 | inuse chunks or the ends of memory. | |
1359 | ||
1360 | Chunks in bins are kept in size order, with ties going to the | |
1361 | approximately least recently used chunk. Ordering isn't needed | |
1362 | for the small bins, which all contain the same-sized chunks, but | |
1363 | facilitates best-fit allocation for larger chunks. These lists | |
1364 | are just sequential. Keeping them in order almost never requires | |
1365 | enough traversal to warrant using fancier ordered data | |
a9177ff5 | 1366 | structures. |
fa8d436c UD |
1367 | |
1368 | Chunks of the same size are linked with the most | |
1369 | recently freed at the front, and allocations are taken from the | |
1370 | back. This results in LRU (FIFO) allocation order, which tends | |
1371 | to give each chunk an equal opportunity to be consolidated with | |
1372 | adjacent freed chunks, resulting in larger free chunks and less | |
1373 | fragmentation. | |
1374 | ||
1375 | To simplify use in double-linked lists, each bin header acts | |
1376 | as a malloc_chunk. This avoids special-casing for headers. | |
1377 | But to conserve space and improve locality, we allocate | |
1378 | only the fd/bk pointers of bins, and then use repositioning tricks | |
a9177ff5 | 1379 | to treat these as the fields of a malloc_chunk*. |
6c8dbf00 | 1380 | */ |
f65fd747 | 1381 | |
6c8dbf00 | 1382 | typedef struct malloc_chunk *mbinptr; |
f65fd747 | 1383 | |
fa8d436c | 1384 | /* addressing -- note that bin_at(0) does not exist */ |
41999a1a UD |
1385 | #define bin_at(m, i) \ |
1386 | (mbinptr) (((char *) &((m)->bins[((i) - 1) * 2])) \ | |
6c8dbf00 | 1387 | - offsetof (struct malloc_chunk, fd)) |
f65fd747 | 1388 | |
fa8d436c | 1389 | /* analog of ++bin */ |
6c8dbf00 | 1390 | #define next_bin(b) ((mbinptr) ((char *) (b) + (sizeof (mchunkptr) << 1))) |
f65fd747 | 1391 | |
fa8d436c UD |
1392 | /* Reminders about list directionality within bins */ |
1393 | #define first(b) ((b)->fd) | |
1394 | #define last(b) ((b)->bk) | |
f65fd747 | 1395 | |
fa8d436c | 1396 | /* Take a chunk off a bin list */ |
fff94fa2 | 1397 | #define unlink(AV, P, BK, FD) { \ |
17f487b7 | 1398 | if (__builtin_expect (chunksize(P) != prev_size (next_chunk(P)), 0)) \ |
ac3ed168 | 1399 | malloc_printerr ("corrupted size vs. prev_size"); \ |
6c8dbf00 OB |
1400 | FD = P->fd; \ |
1401 | BK = P->bk; \ | |
1402 | if (__builtin_expect (FD->bk != P || BK->fd != P, 0)) \ | |
ac3ed168 | 1403 | malloc_printerr ("corrupted double-linked list"); \ |
6c8dbf00 OB |
1404 | else { \ |
1405 | FD->bk = BK; \ | |
1406 | BK->fd = FD; \ | |
e9c4fe93 | 1407 | if (!in_smallbin_range (chunksize_nomask (P)) \ |
6c8dbf00 | 1408 | && __builtin_expect (P->fd_nextsize != NULL, 0)) { \ |
52ffbdf2 FW |
1409 | if (__builtin_expect (P->fd_nextsize->bk_nextsize != P, 0) \ |
1410 | || __builtin_expect (P->bk_nextsize->fd_nextsize != P, 0)) \ | |
ac3ed168 | 1411 | malloc_printerr ("corrupted double-linked list (not small)"); \ |
6c8dbf00 OB |
1412 | if (FD->fd_nextsize == NULL) { \ |
1413 | if (P->fd_nextsize == P) \ | |
1414 | FD->fd_nextsize = FD->bk_nextsize = FD; \ | |
1415 | else { \ | |
1416 | FD->fd_nextsize = P->fd_nextsize; \ | |
1417 | FD->bk_nextsize = P->bk_nextsize; \ | |
1418 | P->fd_nextsize->bk_nextsize = FD; \ | |
1419 | P->bk_nextsize->fd_nextsize = FD; \ | |
1420 | } \ | |
1421 | } else { \ | |
1422 | P->fd_nextsize->bk_nextsize = P->bk_nextsize; \ | |
1423 | P->bk_nextsize->fd_nextsize = P->fd_nextsize; \ | |
1424 | } \ | |
1425 | } \ | |
1426 | } \ | |
fa8d436c | 1427 | } |
f65fd747 | 1428 | |
fa8d436c | 1429 | /* |
6c8dbf00 | 1430 | Indexing |
fa8d436c UD |
1431 | |
1432 | Bins for sizes < 512 bytes contain chunks of all the same size, spaced | |
1433 | 8 bytes apart. Larger bins are approximately logarithmically spaced: | |
f65fd747 | 1434 | |
fa8d436c UD |
1435 | 64 bins of size 8 |
1436 | 32 bins of size 64 | |
1437 | 16 bins of size 512 | |
1438 | 8 bins of size 4096 | |
1439 | 4 bins of size 32768 | |
1440 | 2 bins of size 262144 | |
1441 | 1 bin of size what's left | |
f65fd747 | 1442 | |
fa8d436c UD |
1443 | There is actually a little bit of slop in the numbers in bin_index |
1444 | for the sake of speed. This makes no difference elsewhere. | |
f65fd747 | 1445 | |
fa8d436c UD |
1446 | The bins top out around 1MB because we expect to service large |
1447 | requests via mmap. | |
b5a2bbe6 L |
1448 | |
1449 | Bin 0 does not exist. Bin 1 is the unordered list; if that would be | |
1450 | a valid chunk size the small bins are bumped up one. | |
6c8dbf00 | 1451 | */ |
f65fd747 | 1452 | |
fa8d436c UD |
1453 | #define NBINS 128 |
1454 | #define NSMALLBINS 64 | |
1d47e92f | 1455 | #define SMALLBIN_WIDTH MALLOC_ALIGNMENT |
b5a2bbe6 L |
1456 | #define SMALLBIN_CORRECTION (MALLOC_ALIGNMENT > 2 * SIZE_SZ) |
1457 | #define MIN_LARGE_SIZE ((NSMALLBINS - SMALLBIN_CORRECTION) * SMALLBIN_WIDTH) | |
f65fd747 | 1458 | |
fa8d436c | 1459 | #define in_smallbin_range(sz) \ |
6c8dbf00 | 1460 | ((unsigned long) (sz) < (unsigned long) MIN_LARGE_SIZE) |
f65fd747 | 1461 | |
1d47e92f | 1462 | #define smallbin_index(sz) \ |
6c8dbf00 | 1463 | ((SMALLBIN_WIDTH == 16 ? (((unsigned) (sz)) >> 4) : (((unsigned) (sz)) >> 3))\ |
b5a2bbe6 | 1464 | + SMALLBIN_CORRECTION) |
f65fd747 | 1465 | |
1d47e92f | 1466 | #define largebin_index_32(sz) \ |
6c8dbf00 OB |
1467 | (((((unsigned long) (sz)) >> 6) <= 38) ? 56 + (((unsigned long) (sz)) >> 6) :\ |
1468 | ((((unsigned long) (sz)) >> 9) <= 20) ? 91 + (((unsigned long) (sz)) >> 9) :\ | |
1469 | ((((unsigned long) (sz)) >> 12) <= 10) ? 110 + (((unsigned long) (sz)) >> 12) :\ | |
1470 | ((((unsigned long) (sz)) >> 15) <= 4) ? 119 + (((unsigned long) (sz)) >> 15) :\ | |
1471 | ((((unsigned long) (sz)) >> 18) <= 2) ? 124 + (((unsigned long) (sz)) >> 18) :\ | |
1472 | 126) | |
f65fd747 | 1473 | |
b5a2bbe6 | 1474 | #define largebin_index_32_big(sz) \ |
6c8dbf00 OB |
1475 | (((((unsigned long) (sz)) >> 6) <= 45) ? 49 + (((unsigned long) (sz)) >> 6) :\ |
1476 | ((((unsigned long) (sz)) >> 9) <= 20) ? 91 + (((unsigned long) (sz)) >> 9) :\ | |
1477 | ((((unsigned long) (sz)) >> 12) <= 10) ? 110 + (((unsigned long) (sz)) >> 12) :\ | |
1478 | ((((unsigned long) (sz)) >> 15) <= 4) ? 119 + (((unsigned long) (sz)) >> 15) :\ | |
1479 | ((((unsigned long) (sz)) >> 18) <= 2) ? 124 + (((unsigned long) (sz)) >> 18) :\ | |
1480 | 126) | |
b5a2bbe6 | 1481 | |
1d47e92f UD |
1482 | // XXX It remains to be seen whether it is good to keep the widths of |
1483 | // XXX the buckets the same or whether it should be scaled by a factor | |
1484 | // XXX of two as well. | |
1485 | #define largebin_index_64(sz) \ | |
6c8dbf00 OB |
1486 | (((((unsigned long) (sz)) >> 6) <= 48) ? 48 + (((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) | |
1d47e92f UD |
1492 | |
1493 | #define largebin_index(sz) \ | |
b5a2bbe6 L |
1494 | (SIZE_SZ == 8 ? largebin_index_64 (sz) \ |
1495 | : MALLOC_ALIGNMENT == 16 ? largebin_index_32_big (sz) \ | |
1496 | : largebin_index_32 (sz)) | |
1d47e92f | 1497 | |
fa8d436c | 1498 | #define bin_index(sz) \ |
6c8dbf00 | 1499 | ((in_smallbin_range (sz)) ? smallbin_index (sz) : largebin_index (sz)) |
f65fd747 | 1500 | |
f65fd747 UD |
1501 | |
1502 | /* | |
6c8dbf00 | 1503 | Unsorted chunks |
fa8d436c UD |
1504 | |
1505 | All remainders from chunk splits, as well as all returned chunks, | |
1506 | are first placed in the "unsorted" bin. They are then placed | |
1507 | in regular bins after malloc gives them ONE chance to be used before | |
1508 | binning. So, basically, the unsorted_chunks list acts as a queue, | |
1509 | with chunks being placed on it in free (and malloc_consolidate), | |
1510 | and taken off (to be either used or placed in bins) in malloc. | |
1511 | ||
1512 | The NON_MAIN_ARENA flag is never set for unsorted chunks, so it | |
1513 | does not have to be taken into account in size comparisons. | |
6c8dbf00 | 1514 | */ |
f65fd747 | 1515 | |
fa8d436c | 1516 | /* The otherwise unindexable 1-bin is used to hold unsorted chunks. */ |
6c8dbf00 | 1517 | #define unsorted_chunks(M) (bin_at (M, 1)) |
f65fd747 | 1518 | |
fa8d436c | 1519 | /* |
6c8dbf00 | 1520 | Top |
fa8d436c UD |
1521 | |
1522 | The top-most available chunk (i.e., the one bordering the end of | |
1523 | available memory) is treated specially. It is never included in | |
1524 | any bin, is used only if no other chunk is available, and is | |
1525 | released back to the system if it is very large (see | |
1526 | M_TRIM_THRESHOLD). Because top initially | |
1527 | points to its own bin with initial zero size, thus forcing | |
1528 | extension on the first malloc request, we avoid having any special | |
1529 | code in malloc to check whether it even exists yet. But we still | |
1530 | need to do so when getting memory from system, so we make | |
1531 | initial_top treat the bin as a legal but unusable chunk during the | |
1532 | interval between initialization and the first call to | |
3b49edc0 | 1533 | sysmalloc. (This is somewhat delicate, since it relies on |
fa8d436c | 1534 | the 2 preceding words to be zero during this interval as well.) |
6c8dbf00 | 1535 | */ |
f65fd747 | 1536 | |
fa8d436c | 1537 | /* Conveniently, the unsorted bin can be used as dummy top on first call */ |
6c8dbf00 | 1538 | #define initial_top(M) (unsorted_chunks (M)) |
f65fd747 | 1539 | |
fa8d436c | 1540 | /* |
6c8dbf00 | 1541 | Binmap |
f65fd747 | 1542 | |
fa8d436c UD |
1543 | To help compensate for the large number of bins, a one-level index |
1544 | structure is used for bin-by-bin searching. `binmap' is a | |
1545 | bitvector recording whether bins are definitely empty so they can | |
1546 | be skipped over during during traversals. The bits are NOT always | |
1547 | cleared as soon as bins are empty, but instead only | |
1548 | when they are noticed to be empty during traversal in malloc. | |
6c8dbf00 | 1549 | */ |
f65fd747 | 1550 | |
fa8d436c UD |
1551 | /* Conservatively use 32 bits per map word, even if on 64bit system */ |
1552 | #define BINMAPSHIFT 5 | |
1553 | #define BITSPERMAP (1U << BINMAPSHIFT) | |
1554 | #define BINMAPSIZE (NBINS / BITSPERMAP) | |
f65fd747 | 1555 | |
fa8d436c | 1556 | #define idx2block(i) ((i) >> BINMAPSHIFT) |
6c8dbf00 | 1557 | #define idx2bit(i) ((1U << ((i) & ((1U << BINMAPSHIFT) - 1)))) |
f65fd747 | 1558 | |
6c8dbf00 OB |
1559 | #define mark_bin(m, i) ((m)->binmap[idx2block (i)] |= idx2bit (i)) |
1560 | #define unmark_bin(m, i) ((m)->binmap[idx2block (i)] &= ~(idx2bit (i))) | |
1561 | #define get_binmap(m, i) ((m)->binmap[idx2block (i)] & idx2bit (i)) | |
f65fd747 | 1562 | |
fa8d436c | 1563 | /* |
6c8dbf00 | 1564 | Fastbins |
fa8d436c UD |
1565 | |
1566 | An array of lists holding recently freed small chunks. Fastbins | |
1567 | are not doubly linked. It is faster to single-link them, and | |
1568 | since chunks are never removed from the middles of these lists, | |
1569 | double linking is not necessary. Also, unlike regular bins, they | |
1570 | are not even processed in FIFO order (they use faster LIFO) since | |
1571 | ordering doesn't much matter in the transient contexts in which | |
1572 | fastbins are normally used. | |
1573 | ||
1574 | Chunks in fastbins keep their inuse bit set, so they cannot | |
1575 | be consolidated with other free chunks. malloc_consolidate | |
1576 | releases all chunks in fastbins and consolidates them with | |
a9177ff5 | 1577 | other free chunks. |
6c8dbf00 | 1578 | */ |
f65fd747 | 1579 | |
6c8dbf00 | 1580 | typedef struct malloc_chunk *mfastbinptr; |
425ce2ed | 1581 | #define fastbin(ar_ptr, idx) ((ar_ptr)->fastbinsY[idx]) |
f65fd747 | 1582 | |
fa8d436c | 1583 | /* offset 2 to use otherwise unindexable first 2 bins */ |
425ce2ed | 1584 | #define fastbin_index(sz) \ |
6c8dbf00 | 1585 | ((((unsigned int) (sz)) >> (SIZE_SZ == 8 ? 4 : 3)) - 2) |
425ce2ed | 1586 | |
f65fd747 | 1587 | |
fa8d436c | 1588 | /* The maximum fastbin request size we support */ |
425ce2ed | 1589 | #define MAX_FAST_SIZE (80 * SIZE_SZ / 4) |
f65fd747 | 1590 | |
6c8dbf00 | 1591 | #define NFASTBINS (fastbin_index (request2size (MAX_FAST_SIZE)) + 1) |
f65fd747 UD |
1592 | |
1593 | /* | |
6c8dbf00 OB |
1594 | FASTBIN_CONSOLIDATION_THRESHOLD is the size of a chunk in free() |
1595 | that triggers automatic consolidation of possibly-surrounding | |
1596 | fastbin chunks. This is a heuristic, so the exact value should not | |
1597 | matter too much. It is defined at half the default trim threshold as a | |
1598 | compromise heuristic to only attempt consolidation if it is likely | |
1599 | to lead to trimming. However, it is not dynamically tunable, since | |
1600 | consolidation reduces fragmentation surrounding large chunks even | |
1601 | if trimming is not used. | |
1602 | */ | |
f65fd747 | 1603 | |
fa8d436c | 1604 | #define FASTBIN_CONSOLIDATION_THRESHOLD (65536UL) |
f65fd747 | 1605 | |
f65fd747 | 1606 | /* |
6c8dbf00 OB |
1607 | NONCONTIGUOUS_BIT indicates that MORECORE does not return contiguous |
1608 | regions. Otherwise, contiguity is exploited in merging together, | |
1609 | when possible, results from consecutive MORECORE calls. | |
f65fd747 | 1610 | |
6c8dbf00 OB |
1611 | The initial value comes from MORECORE_CONTIGUOUS, but is |
1612 | changed dynamically if mmap is ever used as an sbrk substitute. | |
1613 | */ | |
f65fd747 | 1614 | |
fa8d436c | 1615 | #define NONCONTIGUOUS_BIT (2U) |
f65fd747 | 1616 | |
6c8dbf00 OB |
1617 | #define contiguous(M) (((M)->flags & NONCONTIGUOUS_BIT) == 0) |
1618 | #define noncontiguous(M) (((M)->flags & NONCONTIGUOUS_BIT) != 0) | |
1619 | #define set_noncontiguous(M) ((M)->flags |= NONCONTIGUOUS_BIT) | |
9bf248c6 | 1620 | #define set_contiguous(M) ((M)->flags &= ~NONCONTIGUOUS_BIT) |
f65fd747 | 1621 | |
eac43cbb FW |
1622 | /* Maximum size of memory handled in fastbins. */ |
1623 | static INTERNAL_SIZE_T global_max_fast; | |
1624 | ||
a9177ff5 RM |
1625 | /* |
1626 | Set value of max_fast. | |
fa8d436c | 1627 | Use impossibly small value if 0. |
3381be5c WD |
1628 | Precondition: there are no existing fastbin chunks in the main arena. |
1629 | Since do_check_malloc_state () checks this, we call malloc_consolidate () | |
1630 | before changing max_fast. Note other arenas will leak their fast bin | |
1631 | entries if max_fast is reduced. | |
6c8dbf00 | 1632 | */ |
f65fd747 | 1633 | |
9bf248c6 | 1634 | #define set_max_fast(s) \ |
991eda1e | 1635 | global_max_fast = (((s) == 0) \ |
6c8dbf00 | 1636 | ? SMALLBIN_WIDTH : ((s + SIZE_SZ) & ~MALLOC_ALIGN_MASK)) |
f65fd747 | 1637 | |
eac43cbb FW |
1638 | static inline INTERNAL_SIZE_T |
1639 | get_max_fast (void) | |
1640 | { | |
1641 | /* Tell the GCC optimizers that global_max_fast is never larger | |
1642 | than MAX_FAST_SIZE. This avoids out-of-bounds array accesses in | |
1643 | _int_malloc after constant propagation of the size parameter. | |
1644 | (The code never executes because malloc preserves the | |
1645 | global_max_fast invariant, but the optimizers may not recognize | |
1646 | this.) */ | |
1647 | if (global_max_fast > MAX_FAST_SIZE) | |
1648 | __builtin_unreachable (); | |
1649 | return global_max_fast; | |
1650 | } | |
f65fd747 UD |
1651 | |
1652 | /* | |
fa8d436c | 1653 | ----------- Internal state representation and initialization ----------- |
6c8dbf00 | 1654 | */ |
f65fd747 | 1655 | |
e956075a WD |
1656 | /* |
1657 | have_fastchunks indicates that there are probably some fastbin chunks. | |
1658 | It is set true on entering a chunk into any fastbin, and cleared early in | |
1659 | malloc_consolidate. The value is approximate since it may be set when there | |
1660 | are no fastbin chunks, or it may be clear even if there are fastbin chunks | |
1661 | available. Given it's sole purpose is to reduce number of redundant calls to | |
1662 | malloc_consolidate, it does not affect correctness. As a result we can safely | |
1663 | use relaxed atomic accesses. | |
1664 | */ | |
1665 | ||
1666 | ||
6c8dbf00 OB |
1667 | struct malloc_state |
1668 | { | |
fa8d436c | 1669 | /* Serialize access. */ |
cbb47fa1 | 1670 | __libc_lock_define (, mutex); |
9bf248c6 UD |
1671 | |
1672 | /* Flags (formerly in max_fast). */ | |
1673 | int flags; | |
f65fd747 | 1674 | |
e956075a | 1675 | /* Set if the fastbin chunks contain recently inserted free blocks. */ |
2c2245b9 WD |
1676 | /* Note this is a bool but not all targets support atomics on booleans. */ |
1677 | int have_fastchunks; | |
e956075a | 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 | |
a62719ba | 1697 | /* Linked list for free arenas. Access to this field is serialized |
90c400bd | 1698 | by free_list_lock in arena.c. */ |
425ce2ed | 1699 | struct malloc_state *next_free; |
425ce2ed | 1700 | |
a62719ba | 1701 | /* Number of threads attached to this arena. 0 if the arena is on |
90c400bd FW |
1702 | the free list. Access to this field is serialized by |
1703 | free_list_lock in arena.c. */ | |
a62719ba FW |
1704 | INTERNAL_SIZE_T attached_threads; |
1705 | ||
fa8d436c UD |
1706 | /* Memory allocated from the system in this arena. */ |
1707 | INTERNAL_SIZE_T system_mem; | |
1708 | INTERNAL_SIZE_T max_system_mem; | |
1709 | }; | |
f65fd747 | 1710 | |
6c8dbf00 OB |
1711 | struct malloc_par |
1712 | { | |
fa8d436c | 1713 | /* Tunable parameters */ |
6c8dbf00 OB |
1714 | unsigned long trim_threshold; |
1715 | INTERNAL_SIZE_T top_pad; | |
1716 | INTERNAL_SIZE_T mmap_threshold; | |
1717 | INTERNAL_SIZE_T arena_test; | |
1718 | INTERNAL_SIZE_T arena_max; | |
fa8d436c UD |
1719 | |
1720 | /* Memory map support */ | |
6c8dbf00 OB |
1721 | int n_mmaps; |
1722 | int n_mmaps_max; | |
1723 | int max_n_mmaps; | |
1d05c2fb UD |
1724 | /* the mmap_threshold is dynamic, until the user sets |
1725 | it manually, at which point we need to disable any | |
1726 | dynamic behavior. */ | |
6c8dbf00 | 1727 | int no_dyn_threshold; |
fa8d436c | 1728 | |
fa8d436c | 1729 | /* Statistics */ |
6c8dbf00 | 1730 | INTERNAL_SIZE_T mmapped_mem; |
6c8dbf00 | 1731 | INTERNAL_SIZE_T max_mmapped_mem; |
fa8d436c UD |
1732 | |
1733 | /* First address handed out by MORECORE/sbrk. */ | |
6c8dbf00 | 1734 | char *sbrk_base; |
d5c3fafc DD |
1735 | |
1736 | #if USE_TCACHE | |
1737 | /* Maximum number of buckets to use. */ | |
1738 | size_t tcache_bins; | |
1739 | size_t tcache_max_bytes; | |
1740 | /* Maximum number of chunks in each bucket. */ | |
1741 | size_t tcache_count; | |
1742 | /* Maximum number of chunks to remove from the unsorted list, which | |
1743 | aren't used to prefill the cache. */ | |
1744 | size_t tcache_unsorted_limit; | |
1745 | #endif | |
fa8d436c | 1746 | }; |
f65fd747 | 1747 | |
fa8d436c UD |
1748 | /* There are several instances of this struct ("arenas") in this |
1749 | malloc. If you are adapting this malloc in a way that does NOT use | |
1750 | a static or mmapped malloc_state, you MUST explicitly zero-fill it | |
1751 | before using. This malloc relies on the property that malloc_state | |
1752 | is initialized to all zeroes (as is true of C statics). */ | |
f65fd747 | 1753 | |
02d46fc4 | 1754 | static struct malloc_state main_arena = |
6c8dbf00 | 1755 | { |
400e1226 | 1756 | .mutex = _LIBC_LOCK_INITIALIZER, |
a62719ba FW |
1757 | .next = &main_arena, |
1758 | .attached_threads = 1 | |
6c8dbf00 | 1759 | }; |
f65fd747 | 1760 | |
4cf6c72f FW |
1761 | /* These variables are used for undumping support. Chunked are marked |
1762 | as using mmap, but we leave them alone if they fall into this | |
1e8a8875 FW |
1763 | range. NB: The chunk size for these chunks only includes the |
1764 | initial size field (of SIZE_SZ bytes), there is no trailing size | |
1765 | field (unlike with regular mmapped chunks). */ | |
4cf6c72f FW |
1766 | static mchunkptr dumped_main_arena_start; /* Inclusive. */ |
1767 | static mchunkptr dumped_main_arena_end; /* Exclusive. */ | |
1768 | ||
1769 | /* True if the pointer falls into the dumped arena. Use this after | |
1770 | chunk_is_mmapped indicates a chunk is mmapped. */ | |
1771 | #define DUMPED_MAIN_ARENA_CHUNK(p) \ | |
1772 | ((p) >= dumped_main_arena_start && (p) < dumped_main_arena_end) | |
1773 | ||
fa8d436c | 1774 | /* There is only one instance of the malloc parameters. */ |
f65fd747 | 1775 | |
02d46fc4 | 1776 | static struct malloc_par mp_ = |
6c8dbf00 OB |
1777 | { |
1778 | .top_pad = DEFAULT_TOP_PAD, | |
1779 | .n_mmaps_max = DEFAULT_MMAP_MAX, | |
1780 | .mmap_threshold = DEFAULT_MMAP_THRESHOLD, | |
1781 | .trim_threshold = DEFAULT_TRIM_THRESHOLD, | |
1782 | #define NARENAS_FROM_NCORES(n) ((n) * (sizeof (long) == 4 ? 2 : 8)) | |
1783 | .arena_test = NARENAS_FROM_NCORES (1) | |
d5c3fafc DD |
1784 | #if USE_TCACHE |
1785 | , | |
1786 | .tcache_count = TCACHE_FILL_COUNT, | |
1787 | .tcache_bins = TCACHE_MAX_BINS, | |
1788 | .tcache_max_bytes = tidx2usize (TCACHE_MAX_BINS-1), | |
1789 | .tcache_unsorted_limit = 0 /* No limit. */ | |
1790 | #endif | |
6c8dbf00 | 1791 | }; |
f65fd747 | 1792 | |
fa8d436c | 1793 | /* |
6c8dbf00 | 1794 | Initialize a malloc_state struct. |
f65fd747 | 1795 | |
3381be5c WD |
1796 | This is called from ptmalloc_init () or from _int_new_arena () |
1797 | when creating a new arena. | |
6c8dbf00 | 1798 | */ |
f65fd747 | 1799 | |
6c8dbf00 OB |
1800 | static void |
1801 | malloc_init_state (mstate av) | |
fa8d436c | 1802 | { |
6c8dbf00 | 1803 | int i; |
fa8d436c | 1804 | mbinptr bin; |
a9177ff5 | 1805 | |
fa8d436c | 1806 | /* Establish circular links for normal bins */ |
6c8dbf00 OB |
1807 | for (i = 1; i < NBINS; ++i) |
1808 | { | |
1809 | bin = bin_at (av, i); | |
1810 | bin->fd = bin->bk = bin; | |
1811 | } | |
f65fd747 | 1812 | |
fa8d436c UD |
1813 | #if MORECORE_CONTIGUOUS |
1814 | if (av != &main_arena) | |
1815 | #endif | |
6c8dbf00 | 1816 | set_noncontiguous (av); |
9bf248c6 | 1817 | if (av == &main_arena) |
6c8dbf00 | 1818 | set_max_fast (DEFAULT_MXFAST); |
e956075a | 1819 | atomic_store_relaxed (&av->have_fastchunks, false); |
f65fd747 | 1820 | |
6c8dbf00 | 1821 | av->top = initial_top (av); |
fa8d436c | 1822 | } |
e9b3e3c5 | 1823 | |
a9177ff5 | 1824 | /* |
fa8d436c | 1825 | Other internal utilities operating on mstates |
6c8dbf00 | 1826 | */ |
f65fd747 | 1827 | |
6c8dbf00 OB |
1828 | static void *sysmalloc (INTERNAL_SIZE_T, mstate); |
1829 | static int systrim (size_t, mstate); | |
1830 | static void malloc_consolidate (mstate); | |
7e3be507 | 1831 | |
404d4cef RM |
1832 | |
1833 | /* -------------- Early definitions for debugging hooks ---------------- */ | |
1834 | ||
1835 | /* Define and initialize the hook variables. These weak definitions must | |
1836 | appear before any use of the variables in a function (arena.c uses one). */ | |
1837 | #ifndef weak_variable | |
404d4cef RM |
1838 | /* In GNU libc we want the hook variables to be weak definitions to |
1839 | avoid a problem with Emacs. */ | |
22a89187 | 1840 | # define weak_variable weak_function |
404d4cef RM |
1841 | #endif |
1842 | ||
1843 | /* Forward declarations. */ | |
6c8dbf00 OB |
1844 | static void *malloc_hook_ini (size_t sz, |
1845 | const void *caller) __THROW; | |
1846 | static void *realloc_hook_ini (void *ptr, size_t sz, | |
1847 | const void *caller) __THROW; | |
1848 | static void *memalign_hook_ini (size_t alignment, size_t sz, | |
1849 | const void *caller) __THROW; | |
404d4cef | 1850 | |
2ba3cfa1 | 1851 | #if HAVE_MALLOC_INIT_HOOK |
92e1ab0e FW |
1852 | void weak_variable (*__malloc_initialize_hook) (void) = NULL; |
1853 | compat_symbol (libc, __malloc_initialize_hook, | |
1854 | __malloc_initialize_hook, GLIBC_2_0); | |
2ba3cfa1 FW |
1855 | #endif |
1856 | ||
a222d91a | 1857 | void weak_variable (*__free_hook) (void *__ptr, |
6c8dbf00 | 1858 | const void *) = NULL; |
a222d91a | 1859 | void *weak_variable (*__malloc_hook) |
6c8dbf00 | 1860 | (size_t __size, const void *) = malloc_hook_ini; |
a222d91a | 1861 | void *weak_variable (*__realloc_hook) |
6c8dbf00 OB |
1862 | (void *__ptr, size_t __size, const void *) |
1863 | = realloc_hook_ini; | |
a222d91a | 1864 | void *weak_variable (*__memalign_hook) |
6c8dbf00 OB |
1865 | (size_t __alignment, size_t __size, const void *) |
1866 | = memalign_hook_ini; | |
06d6611a | 1867 | void weak_variable (*__after_morecore_hook) (void) = NULL; |
404d4cef RM |
1868 | |
1869 | ||
854278df UD |
1870 | /* ------------------ Testing support ----------------------------------*/ |
1871 | ||
1872 | static int perturb_byte; | |
1873 | ||
af102d95 | 1874 | static void |
e8349efd OB |
1875 | alloc_perturb (char *p, size_t n) |
1876 | { | |
1877 | if (__glibc_unlikely (perturb_byte)) | |
1878 | memset (p, perturb_byte ^ 0xff, n); | |
1879 | } | |
1880 | ||
af102d95 | 1881 | static void |
e8349efd OB |
1882 | free_perturb (char *p, size_t n) |
1883 | { | |
1884 | if (__glibc_unlikely (perturb_byte)) | |
1885 | memset (p, perturb_byte, n); | |
1886 | } | |
1887 | ||
854278df UD |
1888 | |
1889 | ||
3ea5be54 AO |
1890 | #include <stap-probe.h> |
1891 | ||
fa8d436c UD |
1892 | /* ------------------- Support for multiple arenas -------------------- */ |
1893 | #include "arena.c" | |
f65fd747 | 1894 | |
fa8d436c | 1895 | /* |
6c8dbf00 | 1896 | Debugging support |
f65fd747 | 1897 | |
6c8dbf00 OB |
1898 | These routines make a number of assertions about the states |
1899 | of data structures that should be true at all times. If any | |
1900 | are not true, it's very likely that a user program has somehow | |
1901 | trashed memory. (It's also possible that there is a coding error | |
1902 | in malloc. In which case, please report it!) | |
1903 | */ | |
ee74a442 | 1904 | |
6c8dbf00 | 1905 | #if !MALLOC_DEBUG |
d8f00d46 | 1906 | |
6c8dbf00 OB |
1907 | # define check_chunk(A, P) |
1908 | # define check_free_chunk(A, P) | |
1909 | # define check_inuse_chunk(A, P) | |
1910 | # define check_remalloced_chunk(A, P, N) | |
1911 | # define check_malloced_chunk(A, P, N) | |
1912 | # define check_malloc_state(A) | |
d8f00d46 | 1913 | |
fa8d436c | 1914 | #else |
ca34d7a7 | 1915 | |
6c8dbf00 OB |
1916 | # define check_chunk(A, P) do_check_chunk (A, P) |
1917 | # define check_free_chunk(A, P) do_check_free_chunk (A, P) | |
1918 | # define check_inuse_chunk(A, P) do_check_inuse_chunk (A, P) | |
1919 | # define check_remalloced_chunk(A, P, N) do_check_remalloced_chunk (A, P, N) | |
1920 | # define check_malloced_chunk(A, P, N) do_check_malloced_chunk (A, P, N) | |
1921 | # define check_malloc_state(A) do_check_malloc_state (A) | |
ca34d7a7 | 1922 | |
fa8d436c | 1923 | /* |
6c8dbf00 OB |
1924 | Properties of all chunks |
1925 | */ | |
ca34d7a7 | 1926 | |
6c8dbf00 OB |
1927 | static void |
1928 | do_check_chunk (mstate av, mchunkptr p) | |
ca34d7a7 | 1929 | { |
6c8dbf00 | 1930 | unsigned long sz = chunksize (p); |
fa8d436c | 1931 | /* min and max possible addresses assuming contiguous allocation */ |
6c8dbf00 OB |
1932 | char *max_address = (char *) (av->top) + chunksize (av->top); |
1933 | char *min_address = max_address - av->system_mem; | |
fa8d436c | 1934 | |
6c8dbf00 OB |
1935 | if (!chunk_is_mmapped (p)) |
1936 | { | |
1937 | /* Has legal address ... */ | |
1938 | if (p != av->top) | |
1939 | { | |
1940 | if (contiguous (av)) | |
1941 | { | |
1942 | assert (((char *) p) >= min_address); | |
1943 | assert (((char *) p + sz) <= ((char *) (av->top))); | |
1944 | } | |
1945 | } | |
1946 | else | |
1947 | { | |
1948 | /* top size is always at least MINSIZE */ | |
1949 | assert ((unsigned long) (sz) >= MINSIZE); | |
1950 | /* top predecessor always marked inuse */ | |
1951 | assert (prev_inuse (p)); | |
1952 | } | |
fa8d436c | 1953 | } |
4cf6c72f | 1954 | else if (!DUMPED_MAIN_ARENA_CHUNK (p)) |
6c8dbf00 OB |
1955 | { |
1956 | /* address is outside main heap */ | |
1957 | if (contiguous (av) && av->top != initial_top (av)) | |
1958 | { | |
1959 | assert (((char *) p) < min_address || ((char *) p) >= max_address); | |
1960 | } | |
1961 | /* chunk is page-aligned */ | |
e9c4fe93 | 1962 | assert (((prev_size (p) + sz) & (GLRO (dl_pagesize) - 1)) == 0); |
6c8dbf00 OB |
1963 | /* mem is aligned */ |
1964 | assert (aligned_OK (chunk2mem (p))); | |
fa8d436c | 1965 | } |
eb406346 UD |
1966 | } |
1967 | ||
fa8d436c | 1968 | /* |
6c8dbf00 OB |
1969 | Properties of free chunks |
1970 | */ | |
ee74a442 | 1971 | |
6c8dbf00 OB |
1972 | static void |
1973 | do_check_free_chunk (mstate av, mchunkptr p) | |
67c94753 | 1974 | { |
3381be5c | 1975 | INTERNAL_SIZE_T sz = chunksize_nomask (p) & ~(PREV_INUSE | NON_MAIN_ARENA); |
6c8dbf00 | 1976 | mchunkptr next = chunk_at_offset (p, sz); |
67c94753 | 1977 | |
6c8dbf00 | 1978 | do_check_chunk (av, p); |
67c94753 | 1979 | |
fa8d436c | 1980 | /* Chunk must claim to be free ... */ |
6c8dbf00 OB |
1981 | assert (!inuse (p)); |
1982 | assert (!chunk_is_mmapped (p)); | |
67c94753 | 1983 | |
fa8d436c | 1984 | /* Unless a special marker, must have OK fields */ |
6c8dbf00 OB |
1985 | if ((unsigned long) (sz) >= MINSIZE) |
1986 | { | |
1987 | assert ((sz & MALLOC_ALIGN_MASK) == 0); | |
1988 | assert (aligned_OK (chunk2mem (p))); | |
1989 | /* ... matching footer field */ | |
3381be5c | 1990 | assert (prev_size (next_chunk (p)) == sz); |
6c8dbf00 OB |
1991 | /* ... and is fully consolidated */ |
1992 | assert (prev_inuse (p)); | |
1993 | assert (next == av->top || inuse (next)); | |
1994 | ||
1995 | /* ... and has minimally sane links */ | |
1996 | assert (p->fd->bk == p); | |
1997 | assert (p->bk->fd == p); | |
1998 | } | |
fa8d436c | 1999 | else /* markers are always of size SIZE_SZ */ |
6c8dbf00 | 2000 | assert (sz == SIZE_SZ); |
67c94753 | 2001 | } |
67c94753 | 2002 | |
fa8d436c | 2003 | /* |
6c8dbf00 OB |
2004 | Properties of inuse chunks |
2005 | */ | |
fa8d436c | 2006 | |
6c8dbf00 OB |
2007 | static void |
2008 | do_check_inuse_chunk (mstate av, mchunkptr p) | |
f65fd747 | 2009 | { |
fa8d436c | 2010 | mchunkptr next; |
f65fd747 | 2011 | |
6c8dbf00 | 2012 | do_check_chunk (av, p); |
f65fd747 | 2013 | |
6c8dbf00 | 2014 | if (chunk_is_mmapped (p)) |
fa8d436c | 2015 | return; /* mmapped chunks have no next/prev */ |
ca34d7a7 | 2016 | |
fa8d436c | 2017 | /* Check whether it claims to be in use ... */ |
6c8dbf00 | 2018 | assert (inuse (p)); |
10dc2a90 | 2019 | |
6c8dbf00 | 2020 | next = next_chunk (p); |
10dc2a90 | 2021 | |
fa8d436c | 2022 | /* ... and is surrounded by OK chunks. |
6c8dbf00 OB |
2023 | Since more things can be checked with free chunks than inuse ones, |
2024 | if an inuse chunk borders them and debug is on, it's worth doing them. | |
2025 | */ | |
2026 | if (!prev_inuse (p)) | |
2027 | { | |
2028 | /* Note that we cannot even look at prev unless it is not inuse */ | |
2029 | mchunkptr prv = prev_chunk (p); | |
2030 | assert (next_chunk (prv) == p); | |
2031 | do_check_free_chunk (av, prv); | |
2032 | } | |
fa8d436c | 2033 | |
6c8dbf00 OB |
2034 | if (next == av->top) |
2035 | { | |
2036 | assert (prev_inuse (next)); | |
2037 | assert (chunksize (next) >= MINSIZE); | |
2038 | } | |
2039 | else if (!inuse (next)) | |
2040 | do_check_free_chunk (av, next); | |
10dc2a90 UD |
2041 | } |
2042 | ||
fa8d436c | 2043 | /* |
6c8dbf00 OB |
2044 | Properties of chunks recycled from fastbins |
2045 | */ | |
fa8d436c | 2046 | |
6c8dbf00 OB |
2047 | static void |
2048 | do_check_remalloced_chunk (mstate av, mchunkptr p, INTERNAL_SIZE_T s) | |
10dc2a90 | 2049 | { |
3381be5c | 2050 | INTERNAL_SIZE_T sz = chunksize_nomask (p) & ~(PREV_INUSE | NON_MAIN_ARENA); |
fa8d436c | 2051 | |
6c8dbf00 OB |
2052 | if (!chunk_is_mmapped (p)) |
2053 | { | |
2054 | assert (av == arena_for_chunk (p)); | |
e9c4fe93 | 2055 | if (chunk_main_arena (p)) |
6c8dbf00 | 2056 | assert (av == &main_arena); |
e9c4fe93 FW |
2057 | else |
2058 | assert (av != &main_arena); | |
6c8dbf00 | 2059 | } |
fa8d436c | 2060 | |
6c8dbf00 | 2061 | do_check_inuse_chunk (av, p); |
fa8d436c UD |
2062 | |
2063 | /* Legal size ... */ | |
6c8dbf00 OB |
2064 | assert ((sz & MALLOC_ALIGN_MASK) == 0); |
2065 | assert ((unsigned long) (sz) >= MINSIZE); | |
fa8d436c | 2066 | /* ... and alignment */ |
6c8dbf00 | 2067 | assert (aligned_OK (chunk2mem (p))); |
fa8d436c | 2068 | /* chunk is less than MINSIZE more than request */ |
6c8dbf00 OB |
2069 | assert ((long) (sz) - (long) (s) >= 0); |
2070 | assert ((long) (sz) - (long) (s + MINSIZE) < 0); | |
10dc2a90 UD |
2071 | } |
2072 | ||
fa8d436c | 2073 | /* |
6c8dbf00 OB |
2074 | Properties of nonrecycled chunks at the point they are malloced |
2075 | */ | |
fa8d436c | 2076 | |
6c8dbf00 OB |
2077 | static void |
2078 | do_check_malloced_chunk (mstate av, mchunkptr p, INTERNAL_SIZE_T s) | |
10dc2a90 | 2079 | { |
fa8d436c | 2080 | /* same as recycled case ... */ |
6c8dbf00 | 2081 | do_check_remalloced_chunk (av, p, s); |
10dc2a90 | 2082 | |
fa8d436c | 2083 | /* |
6c8dbf00 OB |
2084 | ... plus, must obey implementation invariant that prev_inuse is |
2085 | always true of any allocated chunk; i.e., that each allocated | |
2086 | chunk borders either a previously allocated and still in-use | |
2087 | chunk, or the base of its memory arena. This is ensured | |
2088 | by making all allocations from the `lowest' part of any found | |
2089 | chunk. This does not necessarily hold however for chunks | |
2090 | recycled via fastbins. | |
2091 | */ | |
2092 | ||
2093 | assert (prev_inuse (p)); | |
fa8d436c | 2094 | } |
10dc2a90 | 2095 | |
f65fd747 | 2096 | |
fa8d436c | 2097 | /* |
6c8dbf00 | 2098 | Properties of malloc_state. |
f65fd747 | 2099 | |
6c8dbf00 OB |
2100 | This may be useful for debugging malloc, as well as detecting user |
2101 | programmer errors that somehow write into malloc_state. | |
f65fd747 | 2102 | |
6c8dbf00 OB |
2103 | If you are extending or experimenting with this malloc, you can |
2104 | probably figure out how to hack this routine to print out or | |
2105 | display chunk addresses, sizes, bins, and other instrumentation. | |
2106 | */ | |
f65fd747 | 2107 | |
6c8dbf00 OB |
2108 | static void |
2109 | do_check_malloc_state (mstate av) | |
fa8d436c UD |
2110 | { |
2111 | int i; | |
2112 | mchunkptr p; | |
2113 | mchunkptr q; | |
2114 | mbinptr b; | |
fa8d436c UD |
2115 | unsigned int idx; |
2116 | INTERNAL_SIZE_T size; | |
2117 | unsigned long total = 0; | |
2118 | int max_fast_bin; | |
f65fd747 | 2119 | |
fa8d436c | 2120 | /* internal size_t must be no wider than pointer type */ |
6c8dbf00 | 2121 | assert (sizeof (INTERNAL_SIZE_T) <= sizeof (char *)); |
f65fd747 | 2122 | |
fa8d436c | 2123 | /* alignment is a power of 2 */ |
6c8dbf00 | 2124 | assert ((MALLOC_ALIGNMENT & (MALLOC_ALIGNMENT - 1)) == 0); |
f65fd747 | 2125 | |
3381be5c WD |
2126 | /* Check the arena is initialized. */ |
2127 | assert (av->top != 0); | |
2128 | ||
2129 | /* No memory has been allocated yet, so doing more tests is not possible. */ | |
2130 | if (av->top == initial_top (av)) | |
fa8d436c | 2131 | return; |
f65fd747 | 2132 | |
fa8d436c | 2133 | /* pagesize is a power of 2 */ |
8a35c3fe | 2134 | assert (powerof2(GLRO (dl_pagesize))); |
f65fd747 | 2135 | |
fa8d436c | 2136 | /* A contiguous main_arena is consistent with sbrk_base. */ |
6c8dbf00 OB |
2137 | if (av == &main_arena && contiguous (av)) |
2138 | assert ((char *) mp_.sbrk_base + av->system_mem == | |
2139 | (char *) av->top + chunksize (av->top)); | |
fa8d436c UD |
2140 | |
2141 | /* properties of fastbins */ | |
2142 | ||
2143 | /* max_fast is in allowed range */ | |
6c8dbf00 OB |
2144 | assert ((get_max_fast () & ~1) <= request2size (MAX_FAST_SIZE)); |
2145 | ||
2146 | max_fast_bin = fastbin_index (get_max_fast ()); | |
2147 | ||
2148 | for (i = 0; i < NFASTBINS; ++i) | |
2149 | { | |
2150 | p = fastbin (av, i); | |
2151 | ||
2152 | /* The following test can only be performed for the main arena. | |
2153 | While mallopt calls malloc_consolidate to get rid of all fast | |
2154 | bins (especially those larger than the new maximum) this does | |
2155 | only happen for the main arena. Trying to do this for any | |
2156 | other arena would mean those arenas have to be locked and | |
2157 | malloc_consolidate be called for them. This is excessive. And | |
2158 | even if this is acceptable to somebody it still cannot solve | |
2159 | the problem completely since if the arena is locked a | |
2160 | concurrent malloc call might create a new arena which then | |
2161 | could use the newly invalid fast bins. */ | |
2162 | ||
2163 | /* all bins past max_fast are empty */ | |
2164 | if (av == &main_arena && i > max_fast_bin) | |
2165 | assert (p == 0); | |
2166 | ||
2167 | while (p != 0) | |
2168 | { | |
2169 | /* each chunk claims to be inuse */ | |
2170 | do_check_inuse_chunk (av, p); | |
2171 | total += chunksize (p); | |
2172 | /* chunk belongs in this bin */ | |
2173 | assert (fastbin_index (chunksize (p)) == i); | |
2174 | p = p->fd; | |
2175 | } | |
fa8d436c | 2176 | } |
fa8d436c | 2177 | |
fa8d436c | 2178 | /* check normal bins */ |
6c8dbf00 OB |
2179 | for (i = 1; i < NBINS; ++i) |
2180 | { | |
2181 | b = bin_at (av, i); | |
2182 | ||
2183 | /* binmap is accurate (except for bin 1 == unsorted_chunks) */ | |
2184 | if (i >= 2) | |
2185 | { | |
2186 | unsigned int binbit = get_binmap (av, i); | |
2187 | int empty = last (b) == b; | |
2188 | if (!binbit) | |
2189 | assert (empty); | |
2190 | else if (!empty) | |
2191 | assert (binbit); | |
2192 | } | |
2193 | ||
2194 | for (p = last (b); p != b; p = p->bk) | |
2195 | { | |
2196 | /* each chunk claims to be free */ | |
2197 | do_check_free_chunk (av, p); | |
2198 | size = chunksize (p); | |
2199 | total += size; | |
2200 | if (i >= 2) | |
2201 | { | |
2202 | /* chunk belongs in bin */ | |
2203 | idx = bin_index (size); | |
2204 | assert (idx == i); | |
2205 | /* lists are sorted */ | |
2206 | assert (p->bk == b || | |
2207 | (unsigned long) chunksize (p->bk) >= (unsigned long) chunksize (p)); | |
2208 | ||
2209 | if (!in_smallbin_range (size)) | |
2210 | { | |
2211 | if (p->fd_nextsize != NULL) | |
2212 | { | |
2213 | if (p->fd_nextsize == p) | |
2214 | assert (p->bk_nextsize == p); | |
2215 | else | |
2216 | { | |
2217 | if (p->fd_nextsize == first (b)) | |
2218 | assert (chunksize (p) < chunksize (p->fd_nextsize)); | |
2219 | else | |
2220 | assert (chunksize (p) > chunksize (p->fd_nextsize)); | |
2221 | ||
2222 | if (p == first (b)) | |
2223 | assert (chunksize (p) > chunksize (p->bk_nextsize)); | |
2224 | else | |
2225 | assert (chunksize (p) < chunksize (p->bk_nextsize)); | |
2226 | } | |
2227 | } | |
2228 | else | |
2229 | assert (p->bk_nextsize == NULL); | |
2230 | } | |
2231 | } | |
2232 | else if (!in_smallbin_range (size)) | |
2233 | assert (p->fd_nextsize == NULL && p->bk_nextsize == NULL); | |
2234 | /* chunk is followed by a legal chain of inuse chunks */ | |
2235 | for (q = next_chunk (p); | |
2236 | (q != av->top && inuse (q) && | |
2237 | (unsigned long) (chunksize (q)) >= MINSIZE); | |
2238 | q = next_chunk (q)) | |
2239 | do_check_inuse_chunk (av, q); | |
2240 | } | |
fa8d436c | 2241 | } |
f65fd747 | 2242 | |
fa8d436c | 2243 | /* top chunk is OK */ |
6c8dbf00 | 2244 | check_chunk (av, av->top); |
fa8d436c UD |
2245 | } |
2246 | #endif | |
2247 | ||
2248 | ||
2249 | /* ----------------- Support for debugging hooks -------------------- */ | |
2250 | #include "hooks.c" | |
2251 | ||
2252 | ||
2253 | /* ----------- Routines dealing with system allocation -------------- */ | |
2254 | ||
2255 | /* | |
6c8dbf00 OB |
2256 | sysmalloc handles malloc cases requiring more memory from the system. |
2257 | On entry, it is assumed that av->top does not have enough | |
2258 | space to service request for nb bytes, thus requiring that av->top | |
2259 | be extended or replaced. | |
2260 | */ | |
fa8d436c | 2261 | |
6c8dbf00 OB |
2262 | static void * |
2263 | sysmalloc (INTERNAL_SIZE_T nb, mstate av) | |
f65fd747 | 2264 | { |
6c8dbf00 | 2265 | mchunkptr old_top; /* incoming value of av->top */ |
fa8d436c | 2266 | INTERNAL_SIZE_T old_size; /* its size */ |
6c8dbf00 | 2267 | char *old_end; /* its end address */ |
f65fd747 | 2268 | |
6c8dbf00 OB |
2269 | long size; /* arg to first MORECORE or mmap call */ |
2270 | char *brk; /* return value from MORECORE */ | |
f65fd747 | 2271 | |
6c8dbf00 OB |
2272 | long correction; /* arg to 2nd MORECORE call */ |
2273 | char *snd_brk; /* 2nd return val */ | |
f65fd747 | 2274 | |
fa8d436c UD |
2275 | INTERNAL_SIZE_T front_misalign; /* unusable bytes at front of new space */ |
2276 | INTERNAL_SIZE_T end_misalign; /* partial page left at end of new space */ | |
6c8dbf00 | 2277 | char *aligned_brk; /* aligned offset into brk */ |
f65fd747 | 2278 | |
6c8dbf00 OB |
2279 | mchunkptr p; /* the allocated/returned chunk */ |
2280 | mchunkptr remainder; /* remainder from allocation */ | |
2281 | unsigned long remainder_size; /* its size */ | |
fa8d436c | 2282 | |
fa8d436c | 2283 | |
8a35c3fe | 2284 | size_t pagesize = GLRO (dl_pagesize); |
6c8dbf00 | 2285 | bool tried_mmap = false; |
fa8d436c UD |
2286 | |
2287 | ||
fa8d436c | 2288 | /* |
6c8dbf00 OB |
2289 | If have mmap, and the request size meets the mmap threshold, and |
2290 | the system supports mmap, and there are few enough currently | |
2291 | allocated mmapped regions, try to directly map this request | |
2292 | rather than expanding top. | |
2293 | */ | |
2294 | ||
fff94fa2 SP |
2295 | if (av == NULL |
2296 | || ((unsigned long) (nb) >= (unsigned long) (mp_.mmap_threshold) | |
2297 | && (mp_.n_mmaps < mp_.n_mmaps_max))) | |
6c8dbf00 OB |
2298 | { |
2299 | char *mm; /* return value from mmap call*/ | |
a9177ff5 | 2300 | |
6c8dbf00 OB |
2301 | try_mmap: |
2302 | /* | |
2303 | Round up size to nearest page. For mmapped chunks, the overhead | |
2304 | is one SIZE_SZ unit larger than for normal chunks, because there | |
2305 | is no following chunk whose prev_size field could be used. | |
2306 | ||
2307 | See the front_misalign handling below, for glibc there is no | |
2308 | need for further alignments unless we have have high alignment. | |
2309 | */ | |
2310 | if (MALLOC_ALIGNMENT == 2 * SIZE_SZ) | |
8a35c3fe | 2311 | size = ALIGN_UP (nb + SIZE_SZ, pagesize); |
6c8dbf00 | 2312 | else |
8a35c3fe | 2313 | size = ALIGN_UP (nb + SIZE_SZ + MALLOC_ALIGN_MASK, pagesize); |
6c8dbf00 OB |
2314 | tried_mmap = true; |
2315 | ||
2316 | /* Don't try if size wraps around 0 */ | |
2317 | if ((unsigned long) (size) > (unsigned long) (nb)) | |
2318 | { | |
2319 | mm = (char *) (MMAP (0, size, PROT_READ | PROT_WRITE, 0)); | |
2320 | ||
2321 | if (mm != MAP_FAILED) | |
2322 | { | |
2323 | /* | |
2324 | The offset to the start of the mmapped region is stored | |
2325 | in the prev_size field of the chunk. This allows us to adjust | |
2326 | returned start address to meet alignment requirements here | |
2327 | and in memalign(), and still be able to compute proper | |
2328 | address argument for later munmap in free() and realloc(). | |
2329 | */ | |
2330 | ||
2331 | if (MALLOC_ALIGNMENT == 2 * SIZE_SZ) | |
2332 | { | |
2333 | /* For glibc, chunk2mem increases the address by 2*SIZE_SZ and | |
2334 | MALLOC_ALIGN_MASK is 2*SIZE_SZ-1. Each mmap'ed area is page | |
2335 | aligned and therefore definitely MALLOC_ALIGN_MASK-aligned. */ | |
2336 | assert (((INTERNAL_SIZE_T) chunk2mem (mm) & MALLOC_ALIGN_MASK) == 0); | |
2337 | front_misalign = 0; | |
2338 | } | |
2339 | else | |
2340 | front_misalign = (INTERNAL_SIZE_T) chunk2mem (mm) & MALLOC_ALIGN_MASK; | |
2341 | if (front_misalign > 0) | |
2342 | { | |
2343 | correction = MALLOC_ALIGNMENT - front_misalign; | |
2344 | p = (mchunkptr) (mm + correction); | |
e9c4fe93 | 2345 | set_prev_size (p, correction); |
6c8dbf00 OB |
2346 | set_head (p, (size - correction) | IS_MMAPPED); |
2347 | } | |
2348 | else | |
2349 | { | |
2350 | p = (mchunkptr) mm; | |
681421f3 | 2351 | set_prev_size (p, 0); |
6c8dbf00 OB |
2352 | set_head (p, size | IS_MMAPPED); |
2353 | } | |
2354 | ||
2355 | /* update statistics */ | |
2356 | ||
2357 | int new = atomic_exchange_and_add (&mp_.n_mmaps, 1) + 1; | |
2358 | atomic_max (&mp_.max_n_mmaps, new); | |
2359 | ||
2360 | unsigned long sum; | |
2361 | sum = atomic_exchange_and_add (&mp_.mmapped_mem, size) + size; | |
2362 | atomic_max (&mp_.max_mmapped_mem, sum); | |
2363 | ||
2364 | check_chunk (av, p); | |
2365 | ||
2366 | return chunk2mem (p); | |
2367 | } | |
2368 | } | |
fa8d436c | 2369 | } |
fa8d436c | 2370 | |
fff94fa2 SP |
2371 | /* There are no usable arenas and mmap also failed. */ |
2372 | if (av == NULL) | |
2373 | return 0; | |
2374 | ||
fa8d436c UD |
2375 | /* Record incoming configuration of top */ |
2376 | ||
6c8dbf00 OB |
2377 | old_top = av->top; |
2378 | old_size = chunksize (old_top); | |
2379 | old_end = (char *) (chunk_at_offset (old_top, old_size)); | |
fa8d436c | 2380 | |
6c8dbf00 | 2381 | brk = snd_brk = (char *) (MORECORE_FAILURE); |
fa8d436c | 2382 | |
a9177ff5 | 2383 | /* |
fa8d436c UD |
2384 | If not the first time through, we require old_size to be |
2385 | at least MINSIZE and to have prev_inuse set. | |
6c8dbf00 | 2386 | */ |
fa8d436c | 2387 | |
6c8dbf00 OB |
2388 | assert ((old_top == initial_top (av) && old_size == 0) || |
2389 | ((unsigned long) (old_size) >= MINSIZE && | |
2390 | prev_inuse (old_top) && | |
8a35c3fe | 2391 | ((unsigned long) old_end & (pagesize - 1)) == 0)); |
fa8d436c UD |
2392 | |
2393 | /* Precondition: not enough current space to satisfy nb request */ | |
6c8dbf00 | 2394 | assert ((unsigned long) (old_size) < (unsigned long) (nb + MINSIZE)); |
a9177ff5 | 2395 | |
72f90263 | 2396 | |
6c8dbf00 OB |
2397 | if (av != &main_arena) |
2398 | { | |
2399 | heap_info *old_heap, *heap; | |
2400 | size_t old_heap_size; | |
2401 | ||
2402 | /* First try to extend the current heap. */ | |
2403 | old_heap = heap_for_ptr (old_top); | |
2404 | old_heap_size = old_heap->size; | |
2405 | if ((long) (MINSIZE + nb - old_size) > 0 | |
2406 | && grow_heap (old_heap, MINSIZE + nb - old_size) == 0) | |
2407 | { | |
2408 | av->system_mem += old_heap->size - old_heap_size; | |
6c8dbf00 OB |
2409 | set_head (old_top, (((char *) old_heap + old_heap->size) - (char *) old_top) |
2410 | | PREV_INUSE); | |
2411 | } | |
2412 | else if ((heap = new_heap (nb + (MINSIZE + sizeof (*heap)), mp_.top_pad))) | |
2413 | { | |
2414 | /* Use a newly allocated heap. */ | |
2415 | heap->ar_ptr = av; | |
2416 | heap->prev = old_heap; | |
2417 | av->system_mem += heap->size; | |
6c8dbf00 OB |
2418 | /* Set up the new top. */ |
2419 | top (av) = chunk_at_offset (heap, sizeof (*heap)); | |
2420 | set_head (top (av), (heap->size - sizeof (*heap)) | PREV_INUSE); | |
2421 | ||
2422 | /* Setup fencepost and free the old top chunk with a multiple of | |
2423 | MALLOC_ALIGNMENT in size. */ | |
2424 | /* The fencepost takes at least MINSIZE bytes, because it might | |
2425 | become the top chunk again later. Note that a footer is set | |
2426 | up, too, although the chunk is marked in use. */ | |
2427 | old_size = (old_size - MINSIZE) & ~MALLOC_ALIGN_MASK; | |
2428 | set_head (chunk_at_offset (old_top, old_size + 2 * SIZE_SZ), 0 | PREV_INUSE); | |
2429 | if (old_size >= MINSIZE) | |
2430 | { | |
2431 | set_head (chunk_at_offset (old_top, old_size), (2 * SIZE_SZ) | PREV_INUSE); | |
2432 | set_foot (chunk_at_offset (old_top, old_size), (2 * SIZE_SZ)); | |
2433 | set_head (old_top, old_size | PREV_INUSE | NON_MAIN_ARENA); | |
2434 | _int_free (av, old_top, 1); | |
2435 | } | |
2436 | else | |
2437 | { | |
2438 | set_head (old_top, (old_size + 2 * SIZE_SZ) | PREV_INUSE); | |
2439 | set_foot (old_top, (old_size + 2 * SIZE_SZ)); | |
2440 | } | |
2441 | } | |
2442 | else if (!tried_mmap) | |
2443 | /* We can at least try to use to mmap memory. */ | |
2444 | goto try_mmap; | |
fa8d436c | 2445 | } |
6c8dbf00 | 2446 | else /* av == main_arena */ |
fa8d436c | 2447 | |
fa8d436c | 2448 | |
6c8dbf00 OB |
2449 | { /* Request enough space for nb + pad + overhead */ |
2450 | size = nb + mp_.top_pad + MINSIZE; | |
a9177ff5 | 2451 | |
6c8dbf00 OB |
2452 | /* |
2453 | If contiguous, we can subtract out existing space that we hope to | |
2454 | combine with new space. We add it back later only if | |
2455 | we don't actually get contiguous space. | |
2456 | */ | |
a9177ff5 | 2457 | |
6c8dbf00 OB |
2458 | if (contiguous (av)) |
2459 | size -= old_size; | |
fa8d436c | 2460 | |
6c8dbf00 OB |
2461 | /* |
2462 | Round to a multiple of page size. | |
2463 | If MORECORE is not contiguous, this ensures that we only call it | |
2464 | with whole-page arguments. And if MORECORE is contiguous and | |
2465 | this is not first time through, this preserves page-alignment of | |
2466 | previous calls. Otherwise, we correct to page-align below. | |
2467 | */ | |
fa8d436c | 2468 | |
8a35c3fe | 2469 | size = ALIGN_UP (size, pagesize); |
fa8d436c | 2470 | |
6c8dbf00 OB |
2471 | /* |
2472 | Don't try to call MORECORE if argument is so big as to appear | |
2473 | negative. Note that since mmap takes size_t arg, it may succeed | |
2474 | below even if we cannot call MORECORE. | |
2475 | */ | |
2476 | ||
2477 | if (size > 0) | |
2478 | { | |
2479 | brk = (char *) (MORECORE (size)); | |
2480 | LIBC_PROBE (memory_sbrk_more, 2, brk, size); | |
2481 | } | |
2482 | ||
2483 | if (brk != (char *) (MORECORE_FAILURE)) | |
2484 | { | |
2485 | /* Call the `morecore' hook if necessary. */ | |
2486 | void (*hook) (void) = atomic_forced_read (__after_morecore_hook); | |
2487 | if (__builtin_expect (hook != NULL, 0)) | |
2488 | (*hook)(); | |
2489 | } | |
2490 | else | |
2491 | { | |
2492 | /* | |
2493 | If have mmap, try using it as a backup when MORECORE fails or | |
2494 | cannot be used. This is worth doing on systems that have "holes" in | |
2495 | address space, so sbrk cannot extend to give contiguous space, but | |
2496 | space is available elsewhere. Note that we ignore mmap max count | |
2497 | and threshold limits, since the space will not be used as a | |
2498 | segregated mmap region. | |
2499 | */ | |
2500 | ||
2501 | /* Cannot merge with old top, so add its size back in */ | |
2502 | if (contiguous (av)) | |
8a35c3fe | 2503 | size = ALIGN_UP (size + old_size, pagesize); |
6c8dbf00 OB |
2504 | |
2505 | /* If we are relying on mmap as backup, then use larger units */ | |
2506 | if ((unsigned long) (size) < (unsigned long) (MMAP_AS_MORECORE_SIZE)) | |
2507 | size = MMAP_AS_MORECORE_SIZE; | |
2508 | ||
2509 | /* Don't try if size wraps around 0 */ | |
2510 | if ((unsigned long) (size) > (unsigned long) (nb)) | |
2511 | { | |
2512 | char *mbrk = (char *) (MMAP (0, size, PROT_READ | PROT_WRITE, 0)); | |
2513 | ||
2514 | if (mbrk != MAP_FAILED) | |
2515 | { | |
2516 | /* We do not need, and cannot use, another sbrk call to find end */ | |
2517 | brk = mbrk; | |
2518 | snd_brk = brk + size; | |
2519 | ||
2520 | /* | |
2521 | Record that we no longer have a contiguous sbrk region. | |
2522 | After the first time mmap is used as backup, we do not | |
2523 | ever rely on contiguous space since this could incorrectly | |
2524 | bridge regions. | |
2525 | */ | |
2526 | set_noncontiguous (av); | |
2527 | } | |
2528 | } | |
2529 | } | |
2530 | ||
2531 | if (brk != (char *) (MORECORE_FAILURE)) | |
2532 | { | |
2533 | if (mp_.sbrk_base == 0) | |
2534 | mp_.sbrk_base = brk; | |
2535 | av->system_mem += size; | |
2536 | ||
2537 | /* | |
2538 | If MORECORE extends previous space, we can likewise extend top size. | |
2539 | */ | |
2540 | ||
2541 | if (brk == old_end && snd_brk == (char *) (MORECORE_FAILURE)) | |
2542 | set_head (old_top, (size + old_size) | PREV_INUSE); | |
2543 | ||
2544 | else if (contiguous (av) && old_size && brk < old_end) | |
ac3ed168 FW |
2545 | /* Oops! Someone else killed our space.. Can't touch anything. */ |
2546 | malloc_printerr ("break adjusted to free malloc space"); | |
6c8dbf00 OB |
2547 | |
2548 | /* | |
2549 | Otherwise, make adjustments: | |
2550 | ||
2551 | * If the first time through or noncontiguous, we need to call sbrk | |
2552 | just to find out where the end of memory lies. | |
2553 | ||
2554 | * We need to ensure that all returned chunks from malloc will meet | |
2555 | MALLOC_ALIGNMENT | |
2556 | ||
2557 | * If there was an intervening foreign sbrk, we need to adjust sbrk | |
2558 | request size to account for fact that we will not be able to | |
2559 | combine new space with existing space in old_top. | |
2560 | ||
2561 | * Almost all systems internally allocate whole pages at a time, in | |
2562 | which case we might as well use the whole last page of request. | |
2563 | So we allocate enough more memory to hit a page boundary now, | |
2564 | which in turn causes future contiguous calls to page-align. | |
2565 | */ | |
2566 | ||
2567 | else | |
2568 | { | |
2569 | front_misalign = 0; | |
2570 | end_misalign = 0; | |
2571 | correction = 0; | |
2572 | aligned_brk = brk; | |
2573 | ||
2574 | /* handle contiguous cases */ | |
2575 | if (contiguous (av)) | |
2576 | { | |
2577 | /* Count foreign sbrk as system_mem. */ | |
2578 | if (old_size) | |
2579 | av->system_mem += brk - old_end; | |
2580 | ||
2581 | /* Guarantee alignment of first new chunk made from this space */ | |
2582 | ||
2583 | front_misalign = (INTERNAL_SIZE_T) chunk2mem (brk) & MALLOC_ALIGN_MASK; | |
2584 | if (front_misalign > 0) | |
2585 | { | |
2586 | /* | |
2587 | Skip over some bytes to arrive at an aligned position. | |
2588 | We don't need to specially mark these wasted front bytes. | |
2589 | They will never be accessed anyway because | |
2590 | prev_inuse of av->top (and any chunk created from its start) | |
2591 | is always true after initialization. | |
2592 | */ | |
2593 | ||
2594 | correction = MALLOC_ALIGNMENT - front_misalign; | |
2595 | aligned_brk += correction; | |
2596 | } | |
2597 | ||
2598 | /* | |
2599 | If this isn't adjacent to existing space, then we will not | |
2600 | be able to merge with old_top space, so must add to 2nd request. | |
2601 | */ | |
2602 | ||
2603 | correction += old_size; | |
2604 | ||
2605 | /* Extend the end address to hit a page boundary */ | |
2606 | end_misalign = (INTERNAL_SIZE_T) (brk + size + correction); | |
8a35c3fe | 2607 | correction += (ALIGN_UP (end_misalign, pagesize)) - end_misalign; |
6c8dbf00 OB |
2608 | |
2609 | assert (correction >= 0); | |
2610 | snd_brk = (char *) (MORECORE (correction)); | |
2611 | ||
2612 | /* | |
2613 | If can't allocate correction, try to at least find out current | |
2614 | brk. It might be enough to proceed without failing. | |
2615 | ||
2616 | Note that if second sbrk did NOT fail, we assume that space | |
2617 | is contiguous with first sbrk. This is a safe assumption unless | |
2618 | program is multithreaded but doesn't use locks and a foreign sbrk | |
2619 | occurred between our first and second calls. | |
2620 | */ | |
2621 | ||
2622 | if (snd_brk == (char *) (MORECORE_FAILURE)) | |
2623 | { | |
2624 | correction = 0; | |
2625 | snd_brk = (char *) (MORECORE (0)); | |
2626 | } | |
2627 | else | |
2628 | { | |
2629 | /* Call the `morecore' hook if necessary. */ | |
2630 | void (*hook) (void) = atomic_forced_read (__after_morecore_hook); | |
2631 | if (__builtin_expect (hook != NULL, 0)) | |
2632 | (*hook)(); | |
2633 | } | |
2634 | } | |
2635 | ||
2636 | /* handle non-contiguous cases */ | |
2637 | else | |
2638 | { | |
2639 | if (MALLOC_ALIGNMENT == 2 * SIZE_SZ) | |
2640 | /* MORECORE/mmap must correctly align */ | |
2641 | assert (((unsigned long) chunk2mem (brk) & MALLOC_ALIGN_MASK) == 0); | |
2642 | else | |
2643 | { | |
2644 | front_misalign = (INTERNAL_SIZE_T) chunk2mem (brk) & MALLOC_ALIGN_MASK; | |
2645 | if (front_misalign > 0) | |
2646 | { | |
2647 | /* | |
2648 | Skip over some bytes to arrive at an aligned position. | |
2649 | We don't need to specially mark these wasted front bytes. | |
2650 | They will never be accessed anyway because | |
2651 | prev_inuse of av->top (and any chunk created from its start) | |
2652 | is always true after initialization. | |
2653 | */ | |
2654 | ||
2655 | aligned_brk += MALLOC_ALIGNMENT - front_misalign; | |
2656 | } | |
2657 | } | |
2658 | ||
2659 | /* Find out current end of memory */ | |
2660 | if (snd_brk == (char *) (MORECORE_FAILURE)) | |
2661 | { | |
2662 | snd_brk = (char *) (MORECORE (0)); | |
2663 | } | |
2664 | } | |
2665 | ||
2666 | /* Adjust top based on results of second sbrk */ | |
2667 | if (snd_brk != (char *) (MORECORE_FAILURE)) | |
2668 | { | |
2669 | av->top = (mchunkptr) aligned_brk; | |
2670 | set_head (av->top, (snd_brk - aligned_brk + correction) | PREV_INUSE); | |
2671 | av->system_mem += correction; | |
2672 | ||
2673 | /* | |
2674 | If not the first time through, we either have a | |
2675 | gap due to foreign sbrk or a non-contiguous region. Insert a | |
2676 | double fencepost at old_top to prevent consolidation with space | |
2677 | we don't own. These fenceposts are artificial chunks that are | |
2678 | marked as inuse and are in any case too small to use. We need | |
2679 | two to make sizes and alignments work out. | |
2680 | */ | |
2681 | ||
2682 | if (old_size != 0) | |
2683 | { | |
2684 | /* | |
2685 | Shrink old_top to insert fenceposts, keeping size a | |
2686 | multiple of MALLOC_ALIGNMENT. We know there is at least | |
2687 | enough space in old_top to do this. | |
2688 | */ | |
2689 | old_size = (old_size - 4 * SIZE_SZ) & ~MALLOC_ALIGN_MASK; | |
2690 | set_head (old_top, old_size | PREV_INUSE); | |
2691 | ||
2692 | /* | |
2693 | Note that the following assignments completely overwrite | |
2694 | old_top when old_size was previously MINSIZE. This is | |
2695 | intentional. We need the fencepost, even if old_top otherwise gets | |
2696 | lost. | |
2697 | */ | |
e9c4fe93 FW |
2698 | set_head (chunk_at_offset (old_top, old_size), |
2699 | (2 * SIZE_SZ) | PREV_INUSE); | |
2700 | set_head (chunk_at_offset (old_top, old_size + 2 * SIZE_SZ), | |
2701 | (2 * SIZE_SZ) | PREV_INUSE); | |
6c8dbf00 OB |
2702 | |
2703 | /* If possible, release the rest. */ | |
2704 | if (old_size >= MINSIZE) | |
2705 | { | |
2706 | _int_free (av, old_top, 1); | |
2707 | } | |
2708 | } | |
2709 | } | |
2710 | } | |
2711 | } | |
2712 | } /* if (av != &main_arena) */ | |
2713 | ||
2714 | if ((unsigned long) av->system_mem > (unsigned long) (av->max_system_mem)) | |
fa8d436c | 2715 | av->max_system_mem = av->system_mem; |
6c8dbf00 | 2716 | check_malloc_state (av); |
a9177ff5 | 2717 | |
fa8d436c UD |
2718 | /* finally, do the allocation */ |
2719 | p = av->top; | |
6c8dbf00 | 2720 | size = chunksize (p); |
fa8d436c UD |
2721 | |
2722 | /* check that one of the above allocation paths succeeded */ | |
6c8dbf00 OB |
2723 | if ((unsigned long) (size) >= (unsigned long) (nb + MINSIZE)) |
2724 | { | |
2725 | remainder_size = size - nb; | |
2726 | remainder = chunk_at_offset (p, nb); | |
2727 | av->top = remainder; | |
2728 | set_head (p, nb | PREV_INUSE | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
2729 | set_head (remainder, remainder_size | PREV_INUSE); | |
2730 | check_malloced_chunk (av, p, nb); | |
2731 | return chunk2mem (p); | |
2732 | } | |
fa8d436c UD |
2733 | |
2734 | /* catch all failure paths */ | |
8e58439c | 2735 | __set_errno (ENOMEM); |
fa8d436c UD |
2736 | return 0; |
2737 | } | |
2738 | ||
2739 | ||
2740 | /* | |
6c8dbf00 OB |
2741 | systrim is an inverse of sorts to sysmalloc. It gives memory back |
2742 | to the system (via negative arguments to sbrk) if there is unused | |
2743 | memory at the `high' end of the malloc pool. It is called | |
2744 | automatically by free() when top space exceeds the trim | |
2745 | threshold. It is also called by the public malloc_trim routine. It | |
2746 | returns 1 if it actually released any memory, else 0. | |
2747 | */ | |
fa8d436c | 2748 | |
6c8dbf00 OB |
2749 | static int |
2750 | systrim (size_t pad, mstate av) | |
fa8d436c | 2751 | { |
6c8dbf00 OB |
2752 | long top_size; /* Amount of top-most memory */ |
2753 | long extra; /* Amount to release */ | |
2754 | long released; /* Amount actually released */ | |
2755 | char *current_brk; /* address returned by pre-check sbrk call */ | |
2756 | char *new_brk; /* address returned by post-check sbrk call */ | |
8a35c3fe | 2757 | size_t pagesize; |
6c8dbf00 | 2758 | long top_area; |
fa8d436c | 2759 | |
8a35c3fe | 2760 | pagesize = GLRO (dl_pagesize); |
6c8dbf00 | 2761 | top_size = chunksize (av->top); |
a9177ff5 | 2762 | |
4b5b548c FS |
2763 | top_area = top_size - MINSIZE - 1; |
2764 | if (top_area <= pad) | |
2765 | return 0; | |
2766 | ||
ca6be165 CD |
2767 | /* Release in pagesize units and round down to the nearest page. */ |
2768 | extra = ALIGN_DOWN(top_area - pad, pagesize); | |
a9177ff5 | 2769 | |
51a7380b WN |
2770 | if (extra == 0) |
2771 | return 0; | |
2772 | ||
4b5b548c | 2773 | /* |
6c8dbf00 OB |
2774 | Only proceed if end of memory is where we last set it. |
2775 | This avoids problems if there were foreign sbrk calls. | |
2776 | */ | |
2777 | current_brk = (char *) (MORECORE (0)); | |
2778 | if (current_brk == (char *) (av->top) + top_size) | |
2779 | { | |
2780 | /* | |
2781 | Attempt to release memory. We ignore MORECORE return value, | |
2782 | and instead call again to find out where new end of memory is. | |
2783 | This avoids problems if first call releases less than we asked, | |
2784 | of if failure somehow altered brk value. (We could still | |
2785 | encounter problems if it altered brk in some very bad way, | |
2786 | but the only thing we can do is adjust anyway, which will cause | |
2787 | some downstream failure.) | |
2788 | */ | |
2789 | ||
2790 | MORECORE (-extra); | |
2791 | /* Call the `morecore' hook if necessary. */ | |
2792 | void (*hook) (void) = atomic_forced_read (__after_morecore_hook); | |
2793 | if (__builtin_expect (hook != NULL, 0)) | |
2794 | (*hook)(); | |
2795 | new_brk = (char *) (MORECORE (0)); | |
2796 | ||
2797 | LIBC_PROBE (memory_sbrk_less, 2, new_brk, extra); | |
2798 | ||
2799 | if (new_brk != (char *) MORECORE_FAILURE) | |
2800 | { | |
2801 | released = (long) (current_brk - new_brk); | |
2802 | ||
2803 | if (released != 0) | |
2804 | { | |
2805 | /* Success. Adjust top. */ | |
2806 | av->system_mem -= released; | |
2807 | set_head (av->top, (top_size - released) | PREV_INUSE); | |
2808 | check_malloc_state (av); | |
2809 | return 1; | |
2810 | } | |
2811 | } | |
fa8d436c | 2812 | } |
fa8d436c | 2813 | return 0; |
f65fd747 UD |
2814 | } |
2815 | ||
431c33c0 | 2816 | static void |
6c8dbf00 | 2817 | munmap_chunk (mchunkptr p) |
f65fd747 | 2818 | { |
6c8dbf00 | 2819 | INTERNAL_SIZE_T size = chunksize (p); |
f65fd747 | 2820 | |
6c8dbf00 | 2821 | assert (chunk_is_mmapped (p)); |
8e635611 | 2822 | |
4cf6c72f FW |
2823 | /* Do nothing if the chunk is a faked mmapped chunk in the dumped |
2824 | main arena. We never free this memory. */ | |
2825 | if (DUMPED_MAIN_ARENA_CHUNK (p)) | |
2826 | return; | |
2827 | ||
e9c4fe93 FW |
2828 | uintptr_t block = (uintptr_t) p - prev_size (p); |
2829 | size_t total_size = prev_size (p) + size; | |
8e635611 UD |
2830 | /* Unfortunately we have to do the compilers job by hand here. Normally |
2831 | we would test BLOCK and TOTAL-SIZE separately for compliance with the | |
2832 | page size. But gcc does not recognize the optimization possibility | |
2833 | (in the moment at least) so we combine the two values into one before | |
2834 | the bit test. */ | |
6c8dbf00 | 2835 | if (__builtin_expect (((block | total_size) & (GLRO (dl_pagesize) - 1)) != 0, 0)) |
ac3ed168 | 2836 | malloc_printerr ("munmap_chunk(): invalid pointer"); |
f65fd747 | 2837 | |
c6e4925d OB |
2838 | atomic_decrement (&mp_.n_mmaps); |
2839 | atomic_add (&mp_.mmapped_mem, -total_size); | |
f65fd747 | 2840 | |
6ef76f3b UD |
2841 | /* If munmap failed the process virtual memory address space is in a |
2842 | bad shape. Just leave the block hanging around, the process will | |
2843 | terminate shortly anyway since not much can be done. */ | |
6c8dbf00 | 2844 | __munmap ((char *) block, total_size); |
f65fd747 UD |
2845 | } |
2846 | ||
2847 | #if HAVE_MREMAP | |
2848 | ||
431c33c0 | 2849 | static mchunkptr |
6c8dbf00 | 2850 | mremap_chunk (mchunkptr p, size_t new_size) |
f65fd747 | 2851 | { |
8a35c3fe | 2852 | size_t pagesize = GLRO (dl_pagesize); |
e9c4fe93 | 2853 | INTERNAL_SIZE_T offset = prev_size (p); |
6c8dbf00 | 2854 | INTERNAL_SIZE_T size = chunksize (p); |
f65fd747 UD |
2855 | char *cp; |
2856 | ||
6c8dbf00 OB |
2857 | assert (chunk_is_mmapped (p)); |
2858 | assert (((size + offset) & (GLRO (dl_pagesize) - 1)) == 0); | |
f65fd747 UD |
2859 | |
2860 | /* Note the extra SIZE_SZ overhead as in mmap_chunk(). */ | |
8a35c3fe | 2861 | new_size = ALIGN_UP (new_size + offset + SIZE_SZ, pagesize); |
f65fd747 | 2862 | |
68f3802d UD |
2863 | /* No need to remap if the number of pages does not change. */ |
2864 | if (size + offset == new_size) | |
2865 | return p; | |
2866 | ||
6c8dbf00 OB |
2867 | cp = (char *) __mremap ((char *) p - offset, size + offset, new_size, |
2868 | MREMAP_MAYMOVE); | |
f65fd747 | 2869 | |
6c8dbf00 OB |
2870 | if (cp == MAP_FAILED) |
2871 | return 0; | |
f65fd747 | 2872 | |
6c8dbf00 | 2873 | p = (mchunkptr) (cp + offset); |
f65fd747 | 2874 | |
6c8dbf00 | 2875 | assert (aligned_OK (chunk2mem (p))); |
f65fd747 | 2876 | |
e9c4fe93 | 2877 | assert (prev_size (p) == offset); |
6c8dbf00 | 2878 | set_head (p, (new_size - offset) | IS_MMAPPED); |
f65fd747 | 2879 | |
c6e4925d OB |
2880 | INTERNAL_SIZE_T new; |
2881 | new = atomic_exchange_and_add (&mp_.mmapped_mem, new_size - size - offset) | |
6c8dbf00 | 2882 | + new_size - size - offset; |
c6e4925d | 2883 | atomic_max (&mp_.max_mmapped_mem, new); |
f65fd747 UD |
2884 | return p; |
2885 | } | |
f65fd747 UD |
2886 | #endif /* HAVE_MREMAP */ |
2887 | ||
fa8d436c | 2888 | /*------------------------ Public wrappers. --------------------------------*/ |
f65fd747 | 2889 | |
d5c3fafc DD |
2890 | #if USE_TCACHE |
2891 | ||
2892 | /* We overlay this structure on the user-data portion of a chunk when | |
2893 | the chunk is stored in the per-thread cache. */ | |
2894 | typedef struct tcache_entry | |
2895 | { | |
2896 | struct tcache_entry *next; | |
2897 | } tcache_entry; | |
2898 | ||
2899 | /* There is one of these for each thread, which contains the | |
2900 | per-thread cache (hence "tcache_perthread_struct"). Keeping | |
2901 | overall size low is mildly important. Note that COUNTS and ENTRIES | |
2902 | are redundant (we could have just counted the linked list each | |
2903 | time), this is for performance reasons. */ | |
2904 | typedef struct tcache_perthread_struct | |
2905 | { | |
2906 | char counts[TCACHE_MAX_BINS]; | |
2907 | tcache_entry *entries[TCACHE_MAX_BINS]; | |
2908 | } tcache_perthread_struct; | |
2909 | ||
1e26d351 | 2910 | static __thread bool tcache_shutting_down = false; |
d5c3fafc DD |
2911 | static __thread tcache_perthread_struct *tcache = NULL; |
2912 | ||
2913 | /* Caller must ensure that we know tc_idx is valid and there's room | |
2914 | for more chunks. */ | |
e4dd4ace | 2915 | static __always_inline void |
d5c3fafc DD |
2916 | tcache_put (mchunkptr chunk, size_t tc_idx) |
2917 | { | |
2918 | tcache_entry *e = (tcache_entry *) chunk2mem (chunk); | |
2919 | assert (tc_idx < TCACHE_MAX_BINS); | |
2920 | e->next = tcache->entries[tc_idx]; | |
2921 | tcache->entries[tc_idx] = e; | |
2922 | ++(tcache->counts[tc_idx]); | |
2923 | } | |
2924 | ||
2925 | /* Caller must ensure that we know tc_idx is valid and there's | |
2926 | available chunks to remove. */ | |
e4dd4ace | 2927 | static __always_inline void * |
d5c3fafc DD |
2928 | tcache_get (size_t tc_idx) |
2929 | { | |
2930 | tcache_entry *e = tcache->entries[tc_idx]; | |
2931 | assert (tc_idx < TCACHE_MAX_BINS); | |
2932 | assert (tcache->entries[tc_idx] > 0); | |
2933 | tcache->entries[tc_idx] = e->next; | |
2934 | --(tcache->counts[tc_idx]); | |
2935 | return (void *) e; | |
2936 | } | |
2937 | ||
2938 | static void __attribute__ ((section ("__libc_thread_freeres_fn"))) | |
2939 | tcache_thread_freeres (void) | |
2940 | { | |
2941 | int i; | |
2942 | tcache_perthread_struct *tcache_tmp = tcache; | |
2943 | ||
2944 | if (!tcache) | |
2945 | return; | |
2946 | ||
1e26d351 | 2947 | /* Disable the tcache and prevent it from being reinitialized. */ |
d5c3fafc | 2948 | tcache = NULL; |
1e26d351 | 2949 | tcache_shutting_down = true; |
d5c3fafc | 2950 | |
1e26d351 CD |
2951 | /* Free all of the entries and the tcache itself back to the arena |
2952 | heap for coalescing. */ | |
d5c3fafc DD |
2953 | for (i = 0; i < TCACHE_MAX_BINS; ++i) |
2954 | { | |
2955 | while (tcache_tmp->entries[i]) | |
2956 | { | |
2957 | tcache_entry *e = tcache_tmp->entries[i]; | |
2958 | tcache_tmp->entries[i] = e->next; | |
2959 | __libc_free (e); | |
2960 | } | |
2961 | } | |
2962 | ||
2963 | __libc_free (tcache_tmp); | |
d5c3fafc DD |
2964 | } |
2965 | text_set_element (__libc_thread_subfreeres, tcache_thread_freeres); | |
2966 | ||
2967 | static void | |
2968 | tcache_init(void) | |
2969 | { | |
2970 | mstate ar_ptr; | |
2971 | void *victim = 0; | |
2972 | const size_t bytes = sizeof (tcache_perthread_struct); | |
2973 | ||
2974 | if (tcache_shutting_down) | |
2975 | return; | |
2976 | ||
2977 | arena_get (ar_ptr, bytes); | |
2978 | victim = _int_malloc (ar_ptr, bytes); | |
2979 | if (!victim && ar_ptr != NULL) | |
2980 | { | |
2981 | ar_ptr = arena_get_retry (ar_ptr, bytes); | |
2982 | victim = _int_malloc (ar_ptr, bytes); | |
2983 | } | |
2984 | ||
2985 | ||
2986 | if (ar_ptr != NULL) | |
2987 | __libc_lock_unlock (ar_ptr->mutex); | |
2988 | ||
2989 | /* In a low memory situation, we may not be able to allocate memory | |
2990 | - in which case, we just keep trying later. However, we | |
2991 | typically do this very early, so either there is sufficient | |
2992 | memory, or there isn't enough memory to do non-trivial | |
2993 | allocations anyway. */ | |
2994 | if (victim) | |
2995 | { | |
2996 | tcache = (tcache_perthread_struct *) victim; | |
2997 | memset (tcache, 0, sizeof (tcache_perthread_struct)); | |
2998 | } | |
2999 | ||
3000 | } | |
3001 | ||
3002 | #define MAYBE_INIT_TCACHE() \ | |
3003 | if (__glibc_unlikely (tcache == NULL)) \ | |
3004 | tcache_init(); | |
3005 | ||
3006 | #else | |
3007 | #define MAYBE_INIT_TCACHE() | |
3008 | #endif | |
3009 | ||
6c8dbf00 OB |
3010 | void * |
3011 | __libc_malloc (size_t bytes) | |
fa8d436c UD |
3012 | { |
3013 | mstate ar_ptr; | |
22a89187 | 3014 | void *victim; |
f65fd747 | 3015 | |
a222d91a | 3016 | void *(*hook) (size_t, const void *) |
f3eeb3fc | 3017 | = atomic_forced_read (__malloc_hook); |
bfacf1af | 3018 | if (__builtin_expect (hook != NULL, 0)) |
fa8d436c | 3019 | return (*hook)(bytes, RETURN_ADDRESS (0)); |
d5c3fafc DD |
3020 | #if USE_TCACHE |
3021 | /* int_free also calls request2size, be careful to not pad twice. */ | |
3022 | size_t tbytes = request2size (bytes); | |
3023 | size_t tc_idx = csize2tidx (tbytes); | |
3024 | ||
3025 | MAYBE_INIT_TCACHE (); | |
3026 | ||
3027 | DIAG_PUSH_NEEDS_COMMENT; | |
3028 | if (tc_idx < mp_.tcache_bins | |
3029 | /*&& tc_idx < TCACHE_MAX_BINS*/ /* to appease gcc */ | |
3030 | && tcache | |
3031 | && tcache->entries[tc_idx] != NULL) | |
3032 | { | |
3033 | return tcache_get (tc_idx); | |
3034 | } | |
3035 | DIAG_POP_NEEDS_COMMENT; | |
3036 | #endif | |
f65fd747 | 3037 | |
94c5a52a | 3038 | arena_get (ar_ptr, bytes); |
425ce2ed | 3039 | |
6c8dbf00 | 3040 | victim = _int_malloc (ar_ptr, bytes); |
fff94fa2 SP |
3041 | /* Retry with another arena only if we were able to find a usable arena |
3042 | before. */ | |
3043 | if (!victim && ar_ptr != NULL) | |
6c8dbf00 OB |
3044 | { |
3045 | LIBC_PROBE (memory_malloc_retry, 1, bytes); | |
3046 | ar_ptr = arena_get_retry (ar_ptr, bytes); | |
fff94fa2 | 3047 | victim = _int_malloc (ar_ptr, bytes); |
60f0e64b | 3048 | } |
fff94fa2 SP |
3049 | |
3050 | if (ar_ptr != NULL) | |
4bf5f222 | 3051 | __libc_lock_unlock (ar_ptr->mutex); |
fff94fa2 | 3052 | |
6c8dbf00 OB |
3053 | assert (!victim || chunk_is_mmapped (mem2chunk (victim)) || |
3054 | ar_ptr == arena_for_chunk (mem2chunk (victim))); | |
fa8d436c | 3055 | return victim; |
f65fd747 | 3056 | } |
6c8dbf00 | 3057 | libc_hidden_def (__libc_malloc) |
f65fd747 | 3058 | |
fa8d436c | 3059 | void |
6c8dbf00 | 3060 | __libc_free (void *mem) |
f65fd747 | 3061 | { |
fa8d436c UD |
3062 | mstate ar_ptr; |
3063 | mchunkptr p; /* chunk corresponding to mem */ | |
3064 | ||
a222d91a | 3065 | void (*hook) (void *, const void *) |
f3eeb3fc | 3066 | = atomic_forced_read (__free_hook); |
6c8dbf00 OB |
3067 | if (__builtin_expect (hook != NULL, 0)) |
3068 | { | |
3069 | (*hook)(mem, RETURN_ADDRESS (0)); | |
3070 | return; | |
3071 | } | |
f65fd747 | 3072 | |
fa8d436c UD |
3073 | if (mem == 0) /* free(0) has no effect */ |
3074 | return; | |
f65fd747 | 3075 | |
6c8dbf00 | 3076 | p = mem2chunk (mem); |
f65fd747 | 3077 | |
6c8dbf00 OB |
3078 | if (chunk_is_mmapped (p)) /* release mmapped memory. */ |
3079 | { | |
4cf6c72f FW |
3080 | /* See if the dynamic brk/mmap threshold needs adjusting. |
3081 | Dumped fake mmapped chunks do not affect the threshold. */ | |
6c8dbf00 | 3082 | if (!mp_.no_dyn_threshold |
e9c4fe93 FW |
3083 | && chunksize_nomask (p) > mp_.mmap_threshold |
3084 | && chunksize_nomask (p) <= DEFAULT_MMAP_THRESHOLD_MAX | |
4cf6c72f | 3085 | && !DUMPED_MAIN_ARENA_CHUNK (p)) |
6c8dbf00 OB |
3086 | { |
3087 | mp_.mmap_threshold = chunksize (p); | |
3088 | mp_.trim_threshold = 2 * mp_.mmap_threshold; | |
3089 | LIBC_PROBE (memory_mallopt_free_dyn_thresholds, 2, | |
3090 | mp_.mmap_threshold, mp_.trim_threshold); | |
3091 | } | |
3092 | munmap_chunk (p); | |
3093 | return; | |
3094 | } | |
f65fd747 | 3095 | |
d5c3fafc DD |
3096 | MAYBE_INIT_TCACHE (); |
3097 | ||
6c8dbf00 OB |
3098 | ar_ptr = arena_for_chunk (p); |
3099 | _int_free (ar_ptr, p, 0); | |
f65fd747 | 3100 | } |
3b49edc0 | 3101 | libc_hidden_def (__libc_free) |
f65fd747 | 3102 | |
6c8dbf00 OB |
3103 | void * |
3104 | __libc_realloc (void *oldmem, size_t bytes) | |
f65fd747 | 3105 | { |
fa8d436c | 3106 | mstate ar_ptr; |
6c8dbf00 | 3107 | INTERNAL_SIZE_T nb; /* padded request size */ |
f65fd747 | 3108 | |
6c8dbf00 | 3109 | void *newp; /* chunk to return */ |
f65fd747 | 3110 | |
a222d91a | 3111 | void *(*hook) (void *, size_t, const void *) = |
f3eeb3fc | 3112 | atomic_forced_read (__realloc_hook); |
bfacf1af | 3113 | if (__builtin_expect (hook != NULL, 0)) |
fa8d436c | 3114 | return (*hook)(oldmem, bytes, RETURN_ADDRESS (0)); |
f65fd747 | 3115 | |
fa8d436c | 3116 | #if REALLOC_ZERO_BYTES_FREES |
6c8dbf00 OB |
3117 | if (bytes == 0 && oldmem != NULL) |
3118 | { | |
3119 | __libc_free (oldmem); return 0; | |
3120 | } | |
f65fd747 | 3121 | #endif |
f65fd747 | 3122 | |
fa8d436c | 3123 | /* realloc of null is supposed to be same as malloc */ |
6c8dbf00 OB |
3124 | if (oldmem == 0) |
3125 | return __libc_malloc (bytes); | |
f65fd747 | 3126 | |
78ac92ad | 3127 | /* chunk corresponding to oldmem */ |
6c8dbf00 | 3128 | const mchunkptr oldp = mem2chunk (oldmem); |
78ac92ad | 3129 | /* its size */ |
6c8dbf00 | 3130 | const INTERNAL_SIZE_T oldsize = chunksize (oldp); |
f65fd747 | 3131 | |
fff94fa2 SP |
3132 | if (chunk_is_mmapped (oldp)) |
3133 | ar_ptr = NULL; | |
3134 | else | |
d5c3fafc DD |
3135 | { |
3136 | MAYBE_INIT_TCACHE (); | |
3137 | ar_ptr = arena_for_chunk (oldp); | |
3138 | } | |
fff94fa2 | 3139 | |
4cf6c72f FW |
3140 | /* Little security check which won't hurt performance: the allocator |
3141 | never wrapps around at the end of the address space. Therefore | |
3142 | we can exclude some size values which might appear here by | |
3143 | accident or by "design" from some intruder. We need to bypass | |
3144 | this check for dumped fake mmap chunks from the old main arena | |
3145 | because the new malloc may provide additional alignment. */ | |
3146 | if ((__builtin_expect ((uintptr_t) oldp > (uintptr_t) -oldsize, 0) | |
3147 | || __builtin_expect (misaligned_chunk (oldp), 0)) | |
3148 | && !DUMPED_MAIN_ARENA_CHUNK (oldp)) | |
ac3ed168 | 3149 | malloc_printerr ("realloc(): invalid pointer"); |
dc165f7b | 3150 | |
6c8dbf00 | 3151 | checked_request2size (bytes, nb); |
f65fd747 | 3152 | |
6c8dbf00 OB |
3153 | if (chunk_is_mmapped (oldp)) |
3154 | { | |
4cf6c72f FW |
3155 | /* If this is a faked mmapped chunk from the dumped main arena, |
3156 | always make a copy (and do not free the old chunk). */ | |
3157 | if (DUMPED_MAIN_ARENA_CHUNK (oldp)) | |
3158 | { | |
3159 | /* Must alloc, copy, free. */ | |
3160 | void *newmem = __libc_malloc (bytes); | |
3161 | if (newmem == 0) | |
3162 | return NULL; | |
3163 | /* Copy as many bytes as are available from the old chunk | |
1e8a8875 FW |
3164 | and fit into the new size. NB: The overhead for faked |
3165 | mmapped chunks is only SIZE_SZ, not 2 * SIZE_SZ as for | |
3166 | regular mmapped chunks. */ | |
3167 | if (bytes > oldsize - SIZE_SZ) | |
3168 | bytes = oldsize - SIZE_SZ; | |
4cf6c72f FW |
3169 | memcpy (newmem, oldmem, bytes); |
3170 | return newmem; | |
3171 | } | |
3172 | ||
6c8dbf00 | 3173 | void *newmem; |
f65fd747 | 3174 | |
fa8d436c | 3175 | #if HAVE_MREMAP |
6c8dbf00 OB |
3176 | newp = mremap_chunk (oldp, nb); |
3177 | if (newp) | |
3178 | return chunk2mem (newp); | |
f65fd747 | 3179 | #endif |
6c8dbf00 OB |
3180 | /* Note the extra SIZE_SZ overhead. */ |
3181 | if (oldsize - SIZE_SZ >= nb) | |
3182 | return oldmem; /* do nothing */ | |
3183 | ||
3184 | /* Must alloc, copy, free. */ | |
3185 | newmem = __libc_malloc (bytes); | |
3186 | if (newmem == 0) | |
3187 | return 0; /* propagate failure */ | |
fa8d436c | 3188 | |
6c8dbf00 OB |
3189 | memcpy (newmem, oldmem, oldsize - 2 * SIZE_SZ); |
3190 | munmap_chunk (oldp); | |
3191 | return newmem; | |
3192 | } | |
3193 | ||
4bf5f222 | 3194 | __libc_lock_lock (ar_ptr->mutex); |
f65fd747 | 3195 | |
6c8dbf00 | 3196 | newp = _int_realloc (ar_ptr, oldp, oldsize, nb); |
f65fd747 | 3197 | |
4bf5f222 | 3198 | __libc_lock_unlock (ar_ptr->mutex); |
6c8dbf00 OB |
3199 | assert (!newp || chunk_is_mmapped (mem2chunk (newp)) || |
3200 | ar_ptr == arena_for_chunk (mem2chunk (newp))); | |
07014fca UD |
3201 | |
3202 | if (newp == NULL) | |
3203 | { | |
3204 | /* Try harder to allocate memory in other arenas. */ | |
35fed6f1 | 3205 | LIBC_PROBE (memory_realloc_retry, 2, bytes, oldmem); |
6c8dbf00 | 3206 | newp = __libc_malloc (bytes); |
07014fca | 3207 | if (newp != NULL) |
6c8dbf00 OB |
3208 | { |
3209 | memcpy (newp, oldmem, oldsize - SIZE_SZ); | |
3210 | _int_free (ar_ptr, oldp, 0); | |
3211 | } | |
07014fca UD |
3212 | } |
3213 | ||
fa8d436c UD |
3214 | return newp; |
3215 | } | |
3b49edc0 | 3216 | libc_hidden_def (__libc_realloc) |
f65fd747 | 3217 | |
6c8dbf00 OB |
3218 | void * |
3219 | __libc_memalign (size_t alignment, size_t bytes) | |
10ad46bc OB |
3220 | { |
3221 | void *address = RETURN_ADDRESS (0); | |
3222 | return _mid_memalign (alignment, bytes, address); | |
3223 | } | |
3224 | ||
3225 | static void * | |
3226 | _mid_memalign (size_t alignment, size_t bytes, void *address) | |
fa8d436c UD |
3227 | { |
3228 | mstate ar_ptr; | |
22a89187 | 3229 | void *p; |
f65fd747 | 3230 | |
a222d91a | 3231 | void *(*hook) (size_t, size_t, const void *) = |
f3eeb3fc | 3232 | atomic_forced_read (__memalign_hook); |
bfacf1af | 3233 | if (__builtin_expect (hook != NULL, 0)) |
10ad46bc | 3234 | return (*hook)(alignment, bytes, address); |
f65fd747 | 3235 | |
10ad46bc | 3236 | /* If we need less alignment than we give anyway, just relay to malloc. */ |
6c8dbf00 OB |
3237 | if (alignment <= MALLOC_ALIGNMENT) |
3238 | return __libc_malloc (bytes); | |
1228ed5c | 3239 | |
fa8d436c | 3240 | /* Otherwise, ensure that it is at least a minimum chunk size */ |
6c8dbf00 OB |
3241 | if (alignment < MINSIZE) |
3242 | alignment = MINSIZE; | |
f65fd747 | 3243 | |
a56ee40b WN |
3244 | /* If the alignment is greater than SIZE_MAX / 2 + 1 it cannot be a |
3245 | power of 2 and will cause overflow in the check below. */ | |
3246 | if (alignment > SIZE_MAX / 2 + 1) | |
3247 | { | |
3248 | __set_errno (EINVAL); | |
3249 | return 0; | |
3250 | } | |
3251 | ||
b73ed247 WN |
3252 | /* Check for overflow. */ |
3253 | if (bytes > SIZE_MAX - alignment - MINSIZE) | |
3254 | { | |
3255 | __set_errno (ENOMEM); | |
3256 | return 0; | |
3257 | } | |
3258 | ||
10ad46bc OB |
3259 | |
3260 | /* Make sure alignment is power of 2. */ | |
6c8dbf00 OB |
3261 | if (!powerof2 (alignment)) |
3262 | { | |
3263 | size_t a = MALLOC_ALIGNMENT * 2; | |
3264 | while (a < alignment) | |
3265 | a <<= 1; | |
3266 | alignment = a; | |
3267 | } | |
10ad46bc | 3268 | |
6c8dbf00 | 3269 | arena_get (ar_ptr, bytes + alignment + MINSIZE); |
6c8dbf00 OB |
3270 | |
3271 | p = _int_memalign (ar_ptr, alignment, bytes); | |
fff94fa2 | 3272 | if (!p && ar_ptr != NULL) |
6c8dbf00 OB |
3273 | { |
3274 | LIBC_PROBE (memory_memalign_retry, 2, bytes, alignment); | |
3275 | ar_ptr = arena_get_retry (ar_ptr, bytes); | |
fff94fa2 | 3276 | p = _int_memalign (ar_ptr, alignment, bytes); |
f65fd747 | 3277 | } |
fff94fa2 SP |
3278 | |
3279 | if (ar_ptr != NULL) | |
4bf5f222 | 3280 | __libc_lock_unlock (ar_ptr->mutex); |
fff94fa2 | 3281 | |
6c8dbf00 OB |
3282 | assert (!p || chunk_is_mmapped (mem2chunk (p)) || |
3283 | ar_ptr == arena_for_chunk (mem2chunk (p))); | |
fa8d436c | 3284 | return p; |
f65fd747 | 3285 | } |
380d7e87 | 3286 | /* For ISO C11. */ |
3b49edc0 UD |
3287 | weak_alias (__libc_memalign, aligned_alloc) |
3288 | libc_hidden_def (__libc_memalign) | |
f65fd747 | 3289 | |
6c8dbf00 OB |
3290 | void * |
3291 | __libc_valloc (size_t bytes) | |
fa8d436c | 3292 | { |
6c8dbf00 | 3293 | if (__malloc_initialized < 0) |
fa8d436c | 3294 | ptmalloc_init (); |
8088488d | 3295 | |
10ad46bc | 3296 | void *address = RETURN_ADDRESS (0); |
8a35c3fe CD |
3297 | size_t pagesize = GLRO (dl_pagesize); |
3298 | return _mid_memalign (pagesize, bytes, address); | |
fa8d436c | 3299 | } |
f65fd747 | 3300 | |
6c8dbf00 OB |
3301 | void * |
3302 | __libc_pvalloc (size_t bytes) | |
fa8d436c | 3303 | { |
6c8dbf00 | 3304 | if (__malloc_initialized < 0) |
fa8d436c | 3305 | ptmalloc_init (); |
8088488d | 3306 | |
10ad46bc | 3307 | void *address = RETURN_ADDRESS (0); |
8a35c3fe CD |
3308 | size_t pagesize = GLRO (dl_pagesize); |
3309 | size_t rounded_bytes = ALIGN_UP (bytes, pagesize); | |
dba38551 | 3310 | |
1159a193 | 3311 | /* Check for overflow. */ |
8a35c3fe | 3312 | if (bytes > SIZE_MAX - 2 * pagesize - MINSIZE) |
1159a193 WN |
3313 | { |
3314 | __set_errno (ENOMEM); | |
3315 | return 0; | |
3316 | } | |
3317 | ||
8a35c3fe | 3318 | return _mid_memalign (pagesize, rounded_bytes, address); |
fa8d436c | 3319 | } |
f65fd747 | 3320 | |
6c8dbf00 OB |
3321 | void * |
3322 | __libc_calloc (size_t n, size_t elem_size) | |
f65fd747 | 3323 | { |
d6285c9f CD |
3324 | mstate av; |
3325 | mchunkptr oldtop, p; | |
3326 | INTERNAL_SIZE_T bytes, sz, csz, oldtopsize; | |
6c8dbf00 | 3327 | void *mem; |
d6285c9f CD |
3328 | unsigned long clearsize; |
3329 | unsigned long nclears; | |
3330 | INTERNAL_SIZE_T *d; | |
0950889b UD |
3331 | |
3332 | /* size_t is unsigned so the behavior on overflow is defined. */ | |
3333 | bytes = n * elem_size; | |
d9af917d UD |
3334 | #define HALF_INTERNAL_SIZE_T \ |
3335 | (((INTERNAL_SIZE_T) 1) << (8 * sizeof (INTERNAL_SIZE_T) / 2)) | |
6c8dbf00 OB |
3336 | if (__builtin_expect ((n | elem_size) >= HALF_INTERNAL_SIZE_T, 0)) |
3337 | { | |
3338 | if (elem_size != 0 && bytes / elem_size != n) | |
3339 | { | |
3340 | __set_errno (ENOMEM); | |
3341 | return 0; | |
3342 | } | |
d9af917d | 3343 | } |
0950889b | 3344 | |
a222d91a | 3345 | void *(*hook) (size_t, const void *) = |
f3eeb3fc | 3346 | atomic_forced_read (__malloc_hook); |
6c8dbf00 OB |
3347 | if (__builtin_expect (hook != NULL, 0)) |
3348 | { | |
d6285c9f CD |
3349 | sz = bytes; |
3350 | mem = (*hook)(sz, RETURN_ADDRESS (0)); | |
3351 | if (mem == 0) | |
3352 | return 0; | |
3353 | ||
3354 | return memset (mem, 0, sz); | |
7799b7b3 | 3355 | } |
f65fd747 | 3356 | |
d6285c9f CD |
3357 | sz = bytes; |
3358 | ||
d5c3fafc DD |
3359 | MAYBE_INIT_TCACHE (); |
3360 | ||
d6285c9f | 3361 | arena_get (av, sz); |
fff94fa2 SP |
3362 | if (av) |
3363 | { | |
3364 | /* Check if we hand out the top chunk, in which case there may be no | |
3365 | need to clear. */ | |
d6285c9f | 3366 | #if MORECORE_CLEARS |
fff94fa2 SP |
3367 | oldtop = top (av); |
3368 | oldtopsize = chunksize (top (av)); | |
d6285c9f | 3369 | # if MORECORE_CLEARS < 2 |
fff94fa2 SP |
3370 | /* Only newly allocated memory is guaranteed to be cleared. */ |
3371 | if (av == &main_arena && | |
3372 | oldtopsize < mp_.sbrk_base + av->max_system_mem - (char *) oldtop) | |
3373 | oldtopsize = (mp_.sbrk_base + av->max_system_mem - (char *) oldtop); | |
d6285c9f | 3374 | # endif |
fff94fa2 SP |
3375 | if (av != &main_arena) |
3376 | { | |
3377 | heap_info *heap = heap_for_ptr (oldtop); | |
3378 | if (oldtopsize < (char *) heap + heap->mprotect_size - (char *) oldtop) | |
3379 | oldtopsize = (char *) heap + heap->mprotect_size - (char *) oldtop; | |
3380 | } | |
3381 | #endif | |
3382 | } | |
3383 | else | |
d6285c9f | 3384 | { |
fff94fa2 SP |
3385 | /* No usable arenas. */ |
3386 | oldtop = 0; | |
3387 | oldtopsize = 0; | |
d6285c9f | 3388 | } |
d6285c9f CD |
3389 | mem = _int_malloc (av, sz); |
3390 | ||
3391 | ||
3392 | assert (!mem || chunk_is_mmapped (mem2chunk (mem)) || | |
3393 | av == arena_for_chunk (mem2chunk (mem))); | |
3394 | ||
fff94fa2 | 3395 | if (mem == 0 && av != NULL) |
d6285c9f CD |
3396 | { |
3397 | LIBC_PROBE (memory_calloc_retry, 1, sz); | |
3398 | av = arena_get_retry (av, sz); | |
fff94fa2 | 3399 | mem = _int_malloc (av, sz); |
d6285c9f | 3400 | } |
fff94fa2 SP |
3401 | |
3402 | if (av != NULL) | |
4bf5f222 | 3403 | __libc_lock_unlock (av->mutex); |
fff94fa2 SP |
3404 | |
3405 | /* Allocation failed even after a retry. */ | |
3406 | if (mem == 0) | |
3407 | return 0; | |
3408 | ||
d6285c9f CD |
3409 | p = mem2chunk (mem); |
3410 | ||
3411 | /* Two optional cases in which clearing not necessary */ | |
3412 | if (chunk_is_mmapped (p)) | |
3413 | { | |
3414 | if (__builtin_expect (perturb_byte, 0)) | |
3415 | return memset (mem, 0, sz); | |
3416 | ||
3417 | return mem; | |
3418 | } | |
3419 | ||
3420 | csz = chunksize (p); | |
3421 | ||
3422 | #if MORECORE_CLEARS | |
3423 | if (perturb_byte == 0 && (p == oldtop && csz > oldtopsize)) | |
3424 | { | |
3425 | /* clear only the bytes from non-freshly-sbrked memory */ | |
3426 | csz = oldtopsize; | |
3427 | } | |
3428 | #endif | |
3429 | ||
3430 | /* Unroll clear of <= 36 bytes (72 if 8byte sizes). We know that | |
3431 | contents have an odd number of INTERNAL_SIZE_T-sized words; | |
3432 | minimally 3. */ | |
3433 | d = (INTERNAL_SIZE_T *) mem; | |
3434 | clearsize = csz - SIZE_SZ; | |
3435 | nclears = clearsize / sizeof (INTERNAL_SIZE_T); | |
3436 | assert (nclears >= 3); | |
3437 | ||
3438 | if (nclears > 9) | |
3439 | return memset (d, 0, clearsize); | |
3440 | ||
3441 | else | |
3442 | { | |
3443 | *(d + 0) = 0; | |
3444 | *(d + 1) = 0; | |
3445 | *(d + 2) = 0; | |
3446 | if (nclears > 4) | |
3447 | { | |
3448 | *(d + 3) = 0; | |
3449 | *(d + 4) = 0; | |
3450 | if (nclears > 6) | |
3451 | { | |
3452 | *(d + 5) = 0; | |
3453 | *(d + 6) = 0; | |
3454 | if (nclears > 8) | |
3455 | { | |
3456 | *(d + 7) = 0; | |
3457 | *(d + 8) = 0; | |
3458 | } | |
3459 | } | |
3460 | } | |
3461 | } | |
3462 | ||
3463 | return mem; | |
fa8d436c | 3464 | } |
f65fd747 | 3465 | |
f65fd747 | 3466 | /* |
6c8dbf00 OB |
3467 | ------------------------------ malloc ------------------------------ |
3468 | */ | |
f65fd747 | 3469 | |
6c8dbf00 OB |
3470 | static void * |
3471 | _int_malloc (mstate av, size_t bytes) | |
f65fd747 | 3472 | { |
fa8d436c | 3473 | INTERNAL_SIZE_T nb; /* normalized request size */ |
6c8dbf00 OB |
3474 | unsigned int idx; /* associated bin index */ |
3475 | mbinptr bin; /* associated bin */ | |
f65fd747 | 3476 | |
6c8dbf00 | 3477 | mchunkptr victim; /* inspected/selected chunk */ |
fa8d436c | 3478 | INTERNAL_SIZE_T size; /* its size */ |
6c8dbf00 | 3479 | int victim_index; /* its bin index */ |
f65fd747 | 3480 | |
6c8dbf00 OB |
3481 | mchunkptr remainder; /* remainder from a split */ |
3482 | unsigned long remainder_size; /* its size */ | |
8a4b65b4 | 3483 | |
6c8dbf00 OB |
3484 | unsigned int block; /* bit map traverser */ |
3485 | unsigned int bit; /* bit map traverser */ | |
3486 | unsigned int map; /* current word of binmap */ | |
8a4b65b4 | 3487 | |
6c8dbf00 OB |
3488 | mchunkptr fwd; /* misc temp for linking */ |
3489 | mchunkptr bck; /* misc temp for linking */ | |
8a4b65b4 | 3490 | |
d5c3fafc DD |
3491 | #if USE_TCACHE |
3492 | size_t tcache_unsorted_count; /* count of unsorted chunks processed */ | |
3493 | #endif | |
3494 | ||
fa8d436c | 3495 | /* |
6c8dbf00 OB |
3496 | Convert request size to internal form by adding SIZE_SZ bytes |
3497 | overhead plus possibly more to obtain necessary alignment and/or | |
3498 | to obtain a size of at least MINSIZE, the smallest allocatable | |
3499 | size. Also, checked_request2size traps (returning 0) request sizes | |
3500 | that are so large that they wrap around zero when padded and | |
3501 | aligned. | |
3502 | */ | |
f65fd747 | 3503 | |
6c8dbf00 | 3504 | checked_request2size (bytes, nb); |
f65fd747 | 3505 | |
fff94fa2 SP |
3506 | /* There are no usable arenas. Fall back to sysmalloc to get a chunk from |
3507 | mmap. */ | |
3508 | if (__glibc_unlikely (av == NULL)) | |
3509 | { | |
3510 | void *p = sysmalloc (nb, av); | |
3511 | if (p != NULL) | |
3512 | alloc_perturb (p, bytes); | |
3513 | return p; | |
3514 | } | |
3515 | ||
fa8d436c | 3516 | /* |
6c8dbf00 OB |
3517 | If the size qualifies as a fastbin, first check corresponding bin. |
3518 | This code is safe to execute even if av is not yet initialized, so we | |
3519 | can try it without checking, which saves some time on this fast path. | |
3520 | */ | |
f65fd747 | 3521 | |
d5c3fafc DD |
3522 | #define REMOVE_FB(fb, victim, pp) \ |
3523 | do \ | |
3524 | { \ | |
3525 | victim = pp; \ | |
3526 | if (victim == NULL) \ | |
3527 | break; \ | |
3528 | } \ | |
3529 | while ((pp = catomic_compare_and_exchange_val_acq (fb, victim->fd, victim)) \ | |
3530 | != victim); \ | |
3531 | ||
6c8dbf00 OB |
3532 | if ((unsigned long) (nb) <= (unsigned long) (get_max_fast ())) |
3533 | { | |
3534 | idx = fastbin_index (nb); | |
3535 | mfastbinptr *fb = &fastbin (av, idx); | |
3536 | mchunkptr pp = *fb; | |
d5c3fafc | 3537 | REMOVE_FB (fb, victim, pp); |
6c8dbf00 OB |
3538 | if (victim != 0) |
3539 | { | |
3540 | if (__builtin_expect (fastbin_index (chunksize (victim)) != idx, 0)) | |
ac3ed168 | 3541 | malloc_printerr ("malloc(): memory corruption (fast)"); |
6c8dbf00 | 3542 | check_remalloced_chunk (av, victim, nb); |
d5c3fafc DD |
3543 | #if USE_TCACHE |
3544 | /* While we're here, if we see other chunks of the same size, | |
3545 | stash them in the tcache. */ | |
3546 | size_t tc_idx = csize2tidx (nb); | |
3547 | if (tcache && tc_idx < mp_.tcache_bins) | |
3548 | { | |
3549 | mchunkptr tc_victim; | |
3550 | ||
3551 | /* While bin not empty and tcache not full, copy chunks over. */ | |
3552 | while (tcache->counts[tc_idx] < mp_.tcache_count | |
3553 | && (pp = *fb) != NULL) | |
3554 | { | |
3555 | REMOVE_FB (fb, tc_victim, pp); | |
3556 | if (tc_victim != 0) | |
3557 | { | |
3558 | tcache_put (tc_victim, tc_idx); | |
3559 | } | |
3560 | } | |
3561 | } | |
3562 | #endif | |
6c8dbf00 OB |
3563 | void *p = chunk2mem (victim); |
3564 | alloc_perturb (p, bytes); | |
3565 | return p; | |
3566 | } | |
fa8d436c | 3567 | } |
f65fd747 | 3568 | |
fa8d436c | 3569 | /* |
6c8dbf00 OB |
3570 | If a small request, check regular bin. Since these "smallbins" |
3571 | hold one size each, no searching within bins is necessary. | |
3572 | (For a large request, we need to wait until unsorted chunks are | |
3573 | processed to find best fit. But for small ones, fits are exact | |
3574 | anyway, so we can check now, which is faster.) | |
3575 | */ | |
3576 | ||
3577 | if (in_smallbin_range (nb)) | |
3578 | { | |
3579 | idx = smallbin_index (nb); | |
3580 | bin = bin_at (av, idx); | |
3581 | ||
3582 | if ((victim = last (bin)) != bin) | |
3583 | { | |
3381be5c WD |
3584 | bck = victim->bk; |
3585 | if (__glibc_unlikely (bck->fd != victim)) | |
3586 | malloc_printerr ("malloc(): smallbin double linked list corrupted"); | |
3587 | set_inuse_bit_at_offset (victim, nb); | |
3588 | bin->bk = bck; | |
3589 | bck->fd = bin; | |
3590 | ||
3591 | if (av != &main_arena) | |
3592 | set_non_main_arena (victim); | |
3593 | check_malloced_chunk (av, victim, nb); | |
d5c3fafc DD |
3594 | #if USE_TCACHE |
3595 | /* While we're here, if we see other chunks of the same size, | |
3596 | stash them in the tcache. */ | |
3597 | size_t tc_idx = csize2tidx (nb); | |
3598 | if (tcache && tc_idx < mp_.tcache_bins) | |
3599 | { | |
3600 | mchunkptr tc_victim; | |
3601 | ||
3602 | /* While bin not empty and tcache not full, copy chunks over. */ | |
3603 | while (tcache->counts[tc_idx] < mp_.tcache_count | |
3604 | && (tc_victim = last (bin)) != bin) | |
3605 | { | |
3606 | if (tc_victim != 0) | |
3607 | { | |
3608 | bck = tc_victim->bk; | |
3609 | set_inuse_bit_at_offset (tc_victim, nb); | |
3610 | if (av != &main_arena) | |
3611 | set_non_main_arena (tc_victim); | |
3612 | bin->bk = bck; | |
3613 | bck->fd = bin; | |
3614 | ||
3615 | tcache_put (tc_victim, tc_idx); | |
3616 | } | |
3617 | } | |
3618 | } | |
3619 | #endif | |
3381be5c WD |
3620 | void *p = chunk2mem (victim); |
3621 | alloc_perturb (p, bytes); | |
3622 | return p; | |
6c8dbf00 | 3623 | } |
fa8d436c | 3624 | } |
f65fd747 | 3625 | |
a9177ff5 | 3626 | /* |
fa8d436c UD |
3627 | If this is a large request, consolidate fastbins before continuing. |
3628 | While it might look excessive to kill all fastbins before | |
3629 | even seeing if there is space available, this avoids | |
3630 | fragmentation problems normally associated with fastbins. | |
3631 | Also, in practice, programs tend to have runs of either small or | |
a9177ff5 | 3632 | large requests, but less often mixtures, so consolidation is not |
fa8d436c UD |
3633 | invoked all that often in most programs. And the programs that |
3634 | it is called frequently in otherwise tend to fragment. | |
6c8dbf00 | 3635 | */ |
7799b7b3 | 3636 | |
6c8dbf00 OB |
3637 | else |
3638 | { | |
3639 | idx = largebin_index (nb); | |
e956075a | 3640 | if (atomic_load_relaxed (&av->have_fastchunks)) |
6c8dbf00 OB |
3641 | malloc_consolidate (av); |
3642 | } | |
f65fd747 | 3643 | |
fa8d436c | 3644 | /* |
6c8dbf00 OB |
3645 | Process recently freed or remaindered chunks, taking one only if |
3646 | it is exact fit, or, if this a small request, the chunk is remainder from | |
3647 | the most recent non-exact fit. Place other traversed chunks in | |
3648 | bins. Note that this step is the only place in any routine where | |
3649 | chunks are placed in bins. | |
3650 | ||
3651 | The outer loop here is needed because we might not realize until | |
3652 | near the end of malloc that we should have consolidated, so must | |
3653 | do so and retry. This happens at most once, and only when we would | |
3654 | otherwise need to expand memory to service a "small" request. | |
3655 | */ | |
3656 | ||
d5c3fafc DD |
3657 | #if USE_TCACHE |
3658 | INTERNAL_SIZE_T tcache_nb = 0; | |
3659 | size_t tc_idx = csize2tidx (nb); | |
3660 | if (tcache && tc_idx < mp_.tcache_bins) | |
3661 | tcache_nb = nb; | |
3662 | int return_cached = 0; | |
3663 | ||
3664 | tcache_unsorted_count = 0; | |
3665 | #endif | |
3666 | ||
6c8dbf00 OB |
3667 | for (;; ) |
3668 | { | |
3669 | int iters = 0; | |
3670 | while ((victim = unsorted_chunks (av)->bk) != unsorted_chunks (av)) | |
3671 | { | |
3672 | bck = victim->bk; | |
e9c4fe93 FW |
3673 | if (__builtin_expect (chunksize_nomask (victim) <= 2 * SIZE_SZ, 0) |
3674 | || __builtin_expect (chunksize_nomask (victim) | |
3675 | > av->system_mem, 0)) | |
ac3ed168 | 3676 | malloc_printerr ("malloc(): memory corruption"); |
6c8dbf00 OB |
3677 | size = chunksize (victim); |
3678 | ||
3679 | /* | |
3680 | If a small request, try to use last remainder if it is the | |
3681 | only chunk in unsorted bin. This helps promote locality for | |
3682 | runs of consecutive small requests. This is the only | |
3683 | exception to best-fit, and applies only when there is | |
3684 | no exact fit for a small chunk. | |
3685 | */ | |
3686 | ||
3687 | if (in_smallbin_range (nb) && | |
3688 | bck == unsorted_chunks (av) && | |
3689 | victim == av->last_remainder && | |
3690 | (unsigned long) (size) > (unsigned long) (nb + MINSIZE)) | |
3691 | { | |
3692 | /* split and reattach remainder */ | |
3693 | remainder_size = size - nb; | |
3694 | remainder = chunk_at_offset (victim, nb); | |
3695 | unsorted_chunks (av)->bk = unsorted_chunks (av)->fd = remainder; | |
3696 | av->last_remainder = remainder; | |
3697 | remainder->bk = remainder->fd = unsorted_chunks (av); | |
3698 | if (!in_smallbin_range (remainder_size)) | |
3699 | { | |
3700 | remainder->fd_nextsize = NULL; | |
3701 | remainder->bk_nextsize = NULL; | |
3702 | } | |
3703 | ||
3704 | set_head (victim, nb | PREV_INUSE | | |
3705 | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
3706 | set_head (remainder, remainder_size | PREV_INUSE); | |
3707 | set_foot (remainder, remainder_size); | |
3708 | ||
3709 | check_malloced_chunk (av, victim, nb); | |
3710 | void *p = chunk2mem (victim); | |
3711 | alloc_perturb (p, bytes); | |
3712 | return p; | |
3713 | } | |
3714 | ||
3715 | /* remove from unsorted list */ | |
3716 | unsorted_chunks (av)->bk = bck; | |
3717 | bck->fd = unsorted_chunks (av); | |
3718 | ||
3719 | /* Take now instead of binning if exact fit */ | |
3720 | ||
3721 | if (size == nb) | |
3722 | { | |
3723 | set_inuse_bit_at_offset (victim, size); | |
3724 | if (av != &main_arena) | |
e9c4fe93 | 3725 | set_non_main_arena (victim); |
d5c3fafc DD |
3726 | #if USE_TCACHE |
3727 | /* Fill cache first, return to user only if cache fills. | |
3728 | We may return one of these chunks later. */ | |
3729 | if (tcache_nb | |
3730 | && tcache->counts[tc_idx] < mp_.tcache_count) | |
3731 | { | |
3732 | tcache_put (victim, tc_idx); | |
3733 | return_cached = 1; | |
3734 | continue; | |
3735 | } | |
3736 | else | |
3737 | { | |
3738 | #endif | |
6c8dbf00 OB |
3739 | check_malloced_chunk (av, victim, nb); |
3740 | void *p = chunk2mem (victim); | |
3741 | alloc_perturb (p, bytes); | |
3742 | return p; | |
d5c3fafc DD |
3743 | #if USE_TCACHE |
3744 | } | |
3745 | #endif | |
6c8dbf00 OB |
3746 | } |
3747 | ||
3748 | /* place chunk in bin */ | |
3749 | ||
3750 | if (in_smallbin_range (size)) | |
3751 | { | |
3752 | victim_index = smallbin_index (size); | |
3753 | bck = bin_at (av, victim_index); | |
3754 | fwd = bck->fd; | |
3755 | } | |
3756 | else | |
3757 | { | |
3758 | victim_index = largebin_index (size); | |
3759 | bck = bin_at (av, victim_index); | |
3760 | fwd = bck->fd; | |
3761 | ||
3762 | /* maintain large bins in sorted order */ | |
3763 | if (fwd != bck) | |
3764 | { | |
3765 | /* Or with inuse bit to speed comparisons */ | |
3766 | size |= PREV_INUSE; | |
3767 | /* if smaller than smallest, bypass loop below */ | |
e9c4fe93 FW |
3768 | assert (chunk_main_arena (bck->bk)); |
3769 | if ((unsigned long) (size) | |
3770 | < (unsigned long) chunksize_nomask (bck->bk)) | |
6c8dbf00 OB |
3771 | { |
3772 | fwd = bck; | |
3773 | bck = bck->bk; | |
3774 | ||
3775 | victim->fd_nextsize = fwd->fd; | |
3776 | victim->bk_nextsize = fwd->fd->bk_nextsize; | |
3777 | fwd->fd->bk_nextsize = victim->bk_nextsize->fd_nextsize = victim; | |
3778 | } | |
3779 | else | |
3780 | { | |
e9c4fe93 FW |
3781 | assert (chunk_main_arena (fwd)); |
3782 | while ((unsigned long) size < chunksize_nomask (fwd)) | |
6c8dbf00 OB |
3783 | { |
3784 | fwd = fwd->fd_nextsize; | |
e9c4fe93 | 3785 | assert (chunk_main_arena (fwd)); |
6c8dbf00 OB |
3786 | } |
3787 | ||
e9c4fe93 FW |
3788 | if ((unsigned long) size |
3789 | == (unsigned long) chunksize_nomask (fwd)) | |
6c8dbf00 OB |
3790 | /* Always insert in the second position. */ |
3791 | fwd = fwd->fd; | |
3792 | else | |
3793 | { | |
3794 | victim->fd_nextsize = fwd; | |
3795 | victim->bk_nextsize = fwd->bk_nextsize; | |
3796 | fwd->bk_nextsize = victim; | |
3797 | victim->bk_nextsize->fd_nextsize = victim; | |
3798 | } | |
3799 | bck = fwd->bk; | |
3800 | } | |
3801 | } | |
3802 | else | |
3803 | victim->fd_nextsize = victim->bk_nextsize = victim; | |
3804 | } | |
3805 | ||
3806 | mark_bin (av, victim_index); | |
3807 | victim->bk = bck; | |
3808 | victim->fd = fwd; | |
3809 | fwd->bk = victim; | |
3810 | bck->fd = victim; | |
3811 | ||
d5c3fafc DD |
3812 | #if USE_TCACHE |
3813 | /* If we've processed as many chunks as we're allowed while | |
3814 | filling the cache, return one of the cached ones. */ | |
3815 | ++tcache_unsorted_count; | |
3816 | if (return_cached | |
3817 | && mp_.tcache_unsorted_limit > 0 | |
3818 | && tcache_unsorted_count > mp_.tcache_unsorted_limit) | |
3819 | { | |
3820 | return tcache_get (tc_idx); | |
3821 | } | |
3822 | #endif | |
3823 | ||
6c8dbf00 OB |
3824 | #define MAX_ITERS 10000 |
3825 | if (++iters >= MAX_ITERS) | |
3826 | break; | |
3827 | } | |
fa8d436c | 3828 | |
d5c3fafc DD |
3829 | #if USE_TCACHE |
3830 | /* If all the small chunks we found ended up cached, return one now. */ | |
3831 | if (return_cached) | |
3832 | { | |
3833 | return tcache_get (tc_idx); | |
3834 | } | |
3835 | #endif | |
3836 | ||
a9177ff5 | 3837 | /* |
6c8dbf00 OB |
3838 | If a large request, scan through the chunks of current bin in |
3839 | sorted order to find smallest that fits. Use the skip list for this. | |
3840 | */ | |
3841 | ||
3842 | if (!in_smallbin_range (nb)) | |
3843 | { | |
3844 | bin = bin_at (av, idx); | |
3845 | ||
3846 | /* skip scan if empty or largest chunk is too small */ | |
e9c4fe93 FW |
3847 | if ((victim = first (bin)) != bin |
3848 | && (unsigned long) chunksize_nomask (victim) | |
3849 | >= (unsigned long) (nb)) | |
6c8dbf00 OB |
3850 | { |
3851 | victim = victim->bk_nextsize; | |
3852 | while (((unsigned long) (size = chunksize (victim)) < | |
3853 | (unsigned long) (nb))) | |
3854 | victim = victim->bk_nextsize; | |
3855 | ||
3856 | /* Avoid removing the first entry for a size so that the skip | |
3857 | list does not have to be rerouted. */ | |
e9c4fe93 FW |
3858 | if (victim != last (bin) |
3859 | && chunksize_nomask (victim) | |
3860 | == chunksize_nomask (victim->fd)) | |
6c8dbf00 OB |
3861 | victim = victim->fd; |
3862 | ||
3863 | remainder_size = size - nb; | |
fff94fa2 | 3864 | unlink (av, victim, bck, fwd); |
6c8dbf00 OB |
3865 | |
3866 | /* Exhaust */ | |
3867 | if (remainder_size < MINSIZE) | |
3868 | { | |
3869 | set_inuse_bit_at_offset (victim, size); | |
3870 | if (av != &main_arena) | |
e9c4fe93 | 3871 | set_non_main_arena (victim); |
6c8dbf00 OB |
3872 | } |
3873 | /* Split */ | |
3874 | else | |
3875 | { | |
3876 | remainder = chunk_at_offset (victim, nb); | |
3877 | /* We cannot assume the unsorted list is empty and therefore | |
3878 | have to perform a complete insert here. */ | |
3879 | bck = unsorted_chunks (av); | |
3880 | fwd = bck->fd; | |
ac3ed168 FW |
3881 | if (__glibc_unlikely (fwd->bk != bck)) |
3882 | malloc_printerr ("malloc(): corrupted unsorted chunks"); | |
6c8dbf00 OB |
3883 | remainder->bk = bck; |
3884 | remainder->fd = fwd; | |
3885 | bck->fd = remainder; | |
3886 | fwd->bk = remainder; | |
3887 | if (!in_smallbin_range (remainder_size)) | |
3888 | { | |
3889 | remainder->fd_nextsize = NULL; | |
3890 | remainder->bk_nextsize = NULL; | |
3891 | } | |
3892 | set_head (victim, nb | PREV_INUSE | | |
3893 | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
3894 | set_head (remainder, remainder_size | PREV_INUSE); | |
3895 | set_foot (remainder, remainder_size); | |
3896 | } | |
3897 | check_malloced_chunk (av, victim, nb); | |
3898 | void *p = chunk2mem (victim); | |
3899 | alloc_perturb (p, bytes); | |
3900 | return p; | |
3901 | } | |
3902 | } | |
f65fd747 | 3903 | |
6c8dbf00 OB |
3904 | /* |
3905 | Search for a chunk by scanning bins, starting with next largest | |
3906 | bin. This search is strictly by best-fit; i.e., the smallest | |
3907 | (with ties going to approximately the least recently used) chunk | |
3908 | that fits is selected. | |
3909 | ||
3910 | The bitmap avoids needing to check that most blocks are nonempty. | |
3911 | The particular case of skipping all bins during warm-up phases | |
3912 | when no chunks have been returned yet is faster than it might look. | |
3913 | */ | |
3914 | ||
3915 | ++idx; | |
3916 | bin = bin_at (av, idx); | |
3917 | block = idx2block (idx); | |
3918 | map = av->binmap[block]; | |
3919 | bit = idx2bit (idx); | |
3920 | ||
3921 | for (;; ) | |
3922 | { | |
3923 | /* Skip rest of block if there are no more set bits in this block. */ | |
3924 | if (bit > map || bit == 0) | |
3925 | { | |
3926 | do | |
3927 | { | |
3928 | if (++block >= BINMAPSIZE) /* out of bins */ | |
3929 | goto use_top; | |
3930 | } | |
3931 | while ((map = av->binmap[block]) == 0); | |
3932 | ||
3933 | bin = bin_at (av, (block << BINMAPSHIFT)); | |
3934 | bit = 1; | |
3935 | } | |
3936 | ||
3937 | /* Advance to bin with set bit. There must be one. */ | |
3938 | while ((bit & map) == 0) | |
3939 | { | |
3940 | bin = next_bin (bin); | |
3941 | bit <<= 1; | |
3942 | assert (bit != 0); | |
3943 | } | |
3944 | ||
3945 | /* Inspect the bin. It is likely to be non-empty */ | |
3946 | victim = last (bin); | |
3947 | ||
3948 | /* If a false alarm (empty bin), clear the bit. */ | |
3949 | if (victim == bin) | |
3950 | { | |
3951 | av->binmap[block] = map &= ~bit; /* Write through */ | |
3952 | bin = next_bin (bin); | |
3953 | bit <<= 1; | |
3954 | } | |
3955 | ||
3956 | else | |
3957 | { | |
3958 | size = chunksize (victim); | |
3959 | ||
3960 | /* We know the first chunk in this bin is big enough to use. */ | |
3961 | assert ((unsigned long) (size) >= (unsigned long) (nb)); | |
3962 | ||
3963 | remainder_size = size - nb; | |
3964 | ||
3965 | /* unlink */ | |
fff94fa2 | 3966 | unlink (av, victim, bck, fwd); |
6c8dbf00 OB |
3967 | |
3968 | /* Exhaust */ | |
3969 | if (remainder_size < MINSIZE) | |
3970 | { | |
3971 | set_inuse_bit_at_offset (victim, size); | |
3972 | if (av != &main_arena) | |
e9c4fe93 | 3973 | set_non_main_arena (victim); |
6c8dbf00 OB |
3974 | } |
3975 | ||
3976 | /* Split */ | |
3977 | else | |
3978 | { | |
3979 | remainder = chunk_at_offset (victim, nb); | |
3980 | ||
3981 | /* We cannot assume the unsorted list is empty and therefore | |
3982 | have to perform a complete insert here. */ | |
3983 | bck = unsorted_chunks (av); | |
3984 | fwd = bck->fd; | |
ac3ed168 FW |
3985 | if (__glibc_unlikely (fwd->bk != bck)) |
3986 | malloc_printerr ("malloc(): corrupted unsorted chunks 2"); | |
6c8dbf00 OB |
3987 | remainder->bk = bck; |
3988 | remainder->fd = fwd; | |
3989 | bck->fd = remainder; | |
3990 | fwd->bk = remainder; | |
3991 | ||
3992 | /* advertise as last remainder */ | |
3993 | if (in_smallbin_range (nb)) | |
3994 | av->last_remainder = remainder; | |
3995 | if (!in_smallbin_range (remainder_size)) | |
3996 | { | |
3997 | remainder->fd_nextsize = NULL; | |
3998 | remainder->bk_nextsize = NULL; | |
3999 | } | |
4000 | set_head (victim, nb | PREV_INUSE | | |
4001 | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
4002 | set_head (remainder, remainder_size | PREV_INUSE); | |
4003 | set_foot (remainder, remainder_size); | |
4004 | } | |
4005 | check_malloced_chunk (av, victim, nb); | |
4006 | void *p = chunk2mem (victim); | |
4007 | alloc_perturb (p, bytes); | |
4008 | return p; | |
4009 | } | |
4010 | } | |
4011 | ||
4012 | use_top: | |
4013 | /* | |
4014 | If large enough, split off the chunk bordering the end of memory | |
4015 | (held in av->top). Note that this is in accord with the best-fit | |
4016 | search rule. In effect, av->top is treated as larger (and thus | |
4017 | less well fitting) than any other available chunk since it can | |
4018 | be extended to be as large as necessary (up to system | |
4019 | limitations). | |
4020 | ||
4021 | We require that av->top always exists (i.e., has size >= | |
4022 | MINSIZE) after initialization, so if it would otherwise be | |
4023 | exhausted by current request, it is replenished. (The main | |
4024 | reason for ensuring it exists is that we may need MINSIZE space | |
4025 | to put in fenceposts in sysmalloc.) | |
4026 | */ | |
4027 | ||
4028 | victim = av->top; | |
4029 | size = chunksize (victim); | |
4030 | ||
4031 | if ((unsigned long) (size) >= (unsigned long) (nb + MINSIZE)) | |
4032 | { | |
4033 | remainder_size = size - nb; | |
4034 | remainder = chunk_at_offset (victim, nb); | |
4035 | av->top = remainder; | |
4036 | set_head (victim, nb | PREV_INUSE | | |
4037 | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
4038 | set_head (remainder, remainder_size | PREV_INUSE); | |
4039 | ||
4040 | check_malloced_chunk (av, victim, nb); | |
4041 | void *p = chunk2mem (victim); | |
4042 | alloc_perturb (p, bytes); | |
4043 | return p; | |
4044 | } | |
4045 | ||
4046 | /* When we are using atomic ops to free fast chunks we can get | |
4047 | here for all block sizes. */ | |
e956075a | 4048 | else if (atomic_load_relaxed (&av->have_fastchunks)) |
6c8dbf00 OB |
4049 | { |
4050 | malloc_consolidate (av); | |
4051 | /* restore original bin index */ | |
4052 | if (in_smallbin_range (nb)) | |
4053 | idx = smallbin_index (nb); | |
4054 | else | |
4055 | idx = largebin_index (nb); | |
4056 | } | |
f65fd747 | 4057 | |
6c8dbf00 OB |
4058 | /* |
4059 | Otherwise, relay to handle system-dependent cases | |
4060 | */ | |
425ce2ed | 4061 | else |
6c8dbf00 OB |
4062 | { |
4063 | void *p = sysmalloc (nb, av); | |
4064 | if (p != NULL) | |
4065 | alloc_perturb (p, bytes); | |
4066 | return p; | |
4067 | } | |
425ce2ed | 4068 | } |
fa8d436c | 4069 | } |
f65fd747 | 4070 | |
fa8d436c | 4071 | /* |
6c8dbf00 OB |
4072 | ------------------------------ free ------------------------------ |
4073 | */ | |
f65fd747 | 4074 | |
78ac92ad | 4075 | static void |
6c8dbf00 | 4076 | _int_free (mstate av, mchunkptr p, int have_lock) |
f65fd747 | 4077 | { |
fa8d436c | 4078 | INTERNAL_SIZE_T size; /* its size */ |
6c8dbf00 OB |
4079 | mfastbinptr *fb; /* associated fastbin */ |
4080 | mchunkptr nextchunk; /* next contiguous chunk */ | |
fa8d436c | 4081 | INTERNAL_SIZE_T nextsize; /* its size */ |
6c8dbf00 | 4082 | int nextinuse; /* true if nextchunk is used */ |
fa8d436c | 4083 | INTERNAL_SIZE_T prevsize; /* size of previous contiguous chunk */ |
6c8dbf00 OB |
4084 | mchunkptr bck; /* misc temp for linking */ |
4085 | mchunkptr fwd; /* misc temp for linking */ | |
fa8d436c | 4086 | |
6c8dbf00 | 4087 | size = chunksize (p); |
f65fd747 | 4088 | |
37fa1953 UD |
4089 | /* Little security check which won't hurt performance: the |
4090 | allocator never wrapps around at the end of the address space. | |
4091 | Therefore we can exclude some size values which might appear | |
4092 | here by accident or by "design" from some intruder. */ | |
dc165f7b | 4093 | if (__builtin_expect ((uintptr_t) p > (uintptr_t) -size, 0) |
073f560e | 4094 | || __builtin_expect (misaligned_chunk (p), 0)) |
ac3ed168 | 4095 | malloc_printerr ("free(): invalid pointer"); |
347c92e9 L |
4096 | /* We know that each chunk is at least MINSIZE bytes in size or a |
4097 | multiple of MALLOC_ALIGNMENT. */ | |
a1ffb40e | 4098 | if (__glibc_unlikely (size < MINSIZE || !aligned_OK (size))) |
ac3ed168 | 4099 | malloc_printerr ("free(): invalid size"); |
f65fd747 | 4100 | |
37fa1953 | 4101 | check_inuse_chunk(av, p); |
f65fd747 | 4102 | |
d5c3fafc DD |
4103 | #if USE_TCACHE |
4104 | { | |
4105 | size_t tc_idx = csize2tidx (size); | |
4106 | ||
4107 | if (tcache | |
4108 | && tc_idx < mp_.tcache_bins | |
4109 | && tcache->counts[tc_idx] < mp_.tcache_count) | |
4110 | { | |
4111 | tcache_put (p, tc_idx); | |
4112 | return; | |
4113 | } | |
4114 | } | |
4115 | #endif | |
4116 | ||
37fa1953 UD |
4117 | /* |
4118 | If eligible, place chunk on a fastbin so it can be found | |
4119 | and used quickly in malloc. | |
4120 | */ | |
6bf4302e | 4121 | |
9bf248c6 | 4122 | if ((unsigned long)(size) <= (unsigned long)(get_max_fast ()) |
6bf4302e | 4123 | |
37fa1953 UD |
4124 | #if TRIM_FASTBINS |
4125 | /* | |
4126 | If TRIM_FASTBINS set, don't place chunks | |
4127 | bordering top into fastbins | |
4128 | */ | |
4129 | && (chunk_at_offset(p, size) != av->top) | |
4130 | #endif | |
4131 | ) { | |
fa8d436c | 4132 | |
e9c4fe93 FW |
4133 | if (__builtin_expect (chunksize_nomask (chunk_at_offset (p, size)) |
4134 | <= 2 * SIZE_SZ, 0) | |
893e6098 UD |
4135 | || __builtin_expect (chunksize (chunk_at_offset (p, size)) |
4136 | >= av->system_mem, 0)) | |
4137 | { | |
bec466d9 UD |
4138 | /* We might not have a lock at this point and concurrent modifications |
4139 | of system_mem might have let to a false positive. Redo the test | |
4140 | after getting the lock. */ | |
24cffce7 FW |
4141 | if (!have_lock |
4142 | || ({ __libc_lock_lock (av->mutex); | |
e9c4fe93 | 4143 | chunksize_nomask (chunk_at_offset (p, size)) <= 2 * SIZE_SZ |
24cffce7 FW |
4144 | || chunksize (chunk_at_offset (p, size)) >= av->system_mem; |
4145 | })) | |
ac3ed168 | 4146 | malloc_printerr ("free(): invalid next size (fast)"); |
bec466d9 | 4147 | if (! have_lock) |
24cffce7 | 4148 | __libc_lock_unlock (av->mutex); |
893e6098 UD |
4149 | } |
4150 | ||
e8349efd | 4151 | free_perturb (chunk2mem(p), size - 2 * SIZE_SZ); |
425ce2ed | 4152 | |
e956075a | 4153 | atomic_store_relaxed (&av->have_fastchunks, true); |
90a3055e UD |
4154 | unsigned int idx = fastbin_index(size); |
4155 | fb = &fastbin (av, idx); | |
425ce2ed | 4156 | |
362b47fe MK |
4157 | /* Atomically link P to its fastbin: P->FD = *FB; *FB = P; */ |
4158 | mchunkptr old = *fb, old2; | |
5f24d53a | 4159 | unsigned int old_idx = ~0u; |
425ce2ed UD |
4160 | do |
4161 | { | |
362b47fe MK |
4162 | /* Check that the top of the bin is not the record we are going to add |
4163 | (i.e., double free). */ | |
425ce2ed | 4164 | if (__builtin_expect (old == p, 0)) |
ac3ed168 | 4165 | malloc_printerr ("double free or corruption (fasttop)"); |
362b47fe MK |
4166 | /* Check that size of fastbin chunk at the top is the same as |
4167 | size of the chunk that we are adding. We can dereference OLD | |
4168 | only if we have the lock, otherwise it might have already been | |
4169 | deallocated. See use of OLD_IDX below for the actual check. */ | |
4170 | if (have_lock && old != NULL) | |
5f24d53a | 4171 | old_idx = fastbin_index(chunksize(old)); |
362b47fe | 4172 | p->fd = old2 = old; |
425ce2ed | 4173 | } |
362b47fe | 4174 | while ((old = catomic_compare_and_exchange_val_rel (fb, p, old2)) != old2); |
5f24d53a | 4175 | |
362b47fe | 4176 | if (have_lock && old != NULL && __builtin_expect (old_idx != idx, 0)) |
ac3ed168 | 4177 | malloc_printerr ("invalid fastbin entry (free)"); |
37fa1953 | 4178 | } |
f65fd747 | 4179 | |
37fa1953 UD |
4180 | /* |
4181 | Consolidate other non-mmapped chunks as they arrive. | |
4182 | */ | |
fa8d436c | 4183 | |
37fa1953 | 4184 | else if (!chunk_is_mmapped(p)) { |
24cffce7 | 4185 | if (!have_lock) |
4bf5f222 | 4186 | __libc_lock_lock (av->mutex); |
425ce2ed | 4187 | |
37fa1953 | 4188 | nextchunk = chunk_at_offset(p, size); |
fa8d436c | 4189 | |
37fa1953 UD |
4190 | /* Lightweight tests: check whether the block is already the |
4191 | top block. */ | |
a1ffb40e | 4192 | if (__glibc_unlikely (p == av->top)) |
ac3ed168 | 4193 | malloc_printerr ("double free or corruption (top)"); |
37fa1953 UD |
4194 | /* Or whether the next chunk is beyond the boundaries of the arena. */ |
4195 | if (__builtin_expect (contiguous (av) | |
4196 | && (char *) nextchunk | |
4197 | >= ((char *) av->top + chunksize(av->top)), 0)) | |
ac3ed168 | 4198 | malloc_printerr ("double free or corruption (out)"); |
37fa1953 | 4199 | /* Or whether the block is actually not marked used. */ |
a1ffb40e | 4200 | if (__glibc_unlikely (!prev_inuse(nextchunk))) |
ac3ed168 | 4201 | malloc_printerr ("double free or corruption (!prev)"); |
fa8d436c | 4202 | |
37fa1953 | 4203 | nextsize = chunksize(nextchunk); |
e9c4fe93 | 4204 | if (__builtin_expect (chunksize_nomask (nextchunk) <= 2 * SIZE_SZ, 0) |
893e6098 | 4205 | || __builtin_expect (nextsize >= av->system_mem, 0)) |
ac3ed168 | 4206 | malloc_printerr ("free(): invalid next size (normal)"); |
fa8d436c | 4207 | |
e8349efd | 4208 | free_perturb (chunk2mem(p), size - 2 * SIZE_SZ); |
854278df | 4209 | |
37fa1953 UD |
4210 | /* consolidate backward */ |
4211 | if (!prev_inuse(p)) { | |
e9c4fe93 | 4212 | prevsize = prev_size (p); |
37fa1953 UD |
4213 | size += prevsize; |
4214 | p = chunk_at_offset(p, -((long) prevsize)); | |
fff94fa2 | 4215 | unlink(av, p, bck, fwd); |
37fa1953 | 4216 | } |
a9177ff5 | 4217 | |
37fa1953 UD |
4218 | if (nextchunk != av->top) { |
4219 | /* get and clear inuse bit */ | |
4220 | nextinuse = inuse_bit_at_offset(nextchunk, nextsize); | |
4221 | ||
4222 | /* consolidate forward */ | |
4223 | if (!nextinuse) { | |
fff94fa2 | 4224 | unlink(av, nextchunk, bck, fwd); |
37fa1953 UD |
4225 | size += nextsize; |
4226 | } else | |
4227 | clear_inuse_bit_at_offset(nextchunk, 0); | |
10dc2a90 | 4228 | |
fa8d436c | 4229 | /* |
37fa1953 UD |
4230 | Place the chunk in unsorted chunk list. Chunks are |
4231 | not placed into regular bins until after they have | |
4232 | been given one chance to be used in malloc. | |
fa8d436c | 4233 | */ |
f65fd747 | 4234 | |
37fa1953 UD |
4235 | bck = unsorted_chunks(av); |
4236 | fwd = bck->fd; | |
a1ffb40e | 4237 | if (__glibc_unlikely (fwd->bk != bck)) |
ac3ed168 | 4238 | malloc_printerr ("free(): corrupted unsorted chunks"); |
37fa1953 | 4239 | p->fd = fwd; |
7ecfbd38 UD |
4240 | p->bk = bck; |
4241 | if (!in_smallbin_range(size)) | |
4242 | { | |
4243 | p->fd_nextsize = NULL; | |
4244 | p->bk_nextsize = NULL; | |
4245 | } | |
37fa1953 UD |
4246 | bck->fd = p; |
4247 | fwd->bk = p; | |
8a4b65b4 | 4248 | |
37fa1953 UD |
4249 | set_head(p, size | PREV_INUSE); |
4250 | set_foot(p, size); | |
4251 | ||
4252 | check_free_chunk(av, p); | |
4253 | } | |
4254 | ||
4255 | /* | |
4256 | If the chunk borders the current high end of memory, | |
4257 | consolidate into top | |
4258 | */ | |
4259 | ||
4260 | else { | |
4261 | size += nextsize; | |
4262 | set_head(p, size | PREV_INUSE); | |
4263 | av->top = p; | |
4264 | check_chunk(av, p); | |
4265 | } | |
4266 | ||
4267 | /* | |
4268 | If freeing a large space, consolidate possibly-surrounding | |
4269 | chunks. Then, if the total unused topmost memory exceeds trim | |
4270 | threshold, ask malloc_trim to reduce top. | |
4271 | ||
4272 | Unless max_fast is 0, we don't know if there are fastbins | |
4273 | bordering top, so we cannot tell for sure whether threshold | |
4274 | has been reached unless fastbins are consolidated. But we | |
4275 | don't want to consolidate on each free. As a compromise, | |
4276 | consolidation is performed if FASTBIN_CONSOLIDATION_THRESHOLD | |
4277 | is reached. | |
4278 | */ | |
fa8d436c | 4279 | |
37fa1953 | 4280 | if ((unsigned long)(size) >= FASTBIN_CONSOLIDATION_THRESHOLD) { |
e956075a | 4281 | if (atomic_load_relaxed (&av->have_fastchunks)) |
37fa1953 | 4282 | malloc_consolidate(av); |
fa8d436c | 4283 | |
37fa1953 | 4284 | if (av == &main_arena) { |
a9177ff5 | 4285 | #ifndef MORECORE_CANNOT_TRIM |
37fa1953 UD |
4286 | if ((unsigned long)(chunksize(av->top)) >= |
4287 | (unsigned long)(mp_.trim_threshold)) | |
3b49edc0 | 4288 | systrim(mp_.top_pad, av); |
fa8d436c | 4289 | #endif |
37fa1953 UD |
4290 | } else { |
4291 | /* Always try heap_trim(), even if the top chunk is not | |
4292 | large, because the corresponding heap might go away. */ | |
4293 | heap_info *heap = heap_for_ptr(top(av)); | |
fa8d436c | 4294 | |
37fa1953 UD |
4295 | assert(heap->ar_ptr == av); |
4296 | heap_trim(heap, mp_.top_pad); | |
fa8d436c | 4297 | } |
fa8d436c | 4298 | } |
10dc2a90 | 4299 | |
24cffce7 | 4300 | if (!have_lock) |
4bf5f222 | 4301 | __libc_lock_unlock (av->mutex); |
37fa1953 UD |
4302 | } |
4303 | /* | |
22a89187 | 4304 | If the chunk was allocated via mmap, release via munmap(). |
37fa1953 UD |
4305 | */ |
4306 | ||
4307 | else { | |
c120d94d | 4308 | munmap_chunk (p); |
fa8d436c | 4309 | } |
10dc2a90 UD |
4310 | } |
4311 | ||
fa8d436c UD |
4312 | /* |
4313 | ------------------------- malloc_consolidate ------------------------- | |
4314 | ||
4315 | malloc_consolidate is a specialized version of free() that tears | |
4316 | down chunks held in fastbins. Free itself cannot be used for this | |
4317 | purpose since, among other things, it might place chunks back onto | |
4318 | fastbins. So, instead, we need to use a minor variant of the same | |
4319 | code. | |
fa8d436c UD |
4320 | */ |
4321 | ||
fa8d436c | 4322 | static void malloc_consolidate(mstate av) |
10dc2a90 | 4323 | { |
fa8d436c UD |
4324 | mfastbinptr* fb; /* current fastbin being consolidated */ |
4325 | mfastbinptr* maxfb; /* last fastbin (for loop control) */ | |
4326 | mchunkptr p; /* current chunk being consolidated */ | |
4327 | mchunkptr nextp; /* next chunk to consolidate */ | |
4328 | mchunkptr unsorted_bin; /* bin header */ | |
4329 | mchunkptr first_unsorted; /* chunk to link to */ | |
4330 | ||
4331 | /* These have same use as in free() */ | |
4332 | mchunkptr nextchunk; | |
4333 | INTERNAL_SIZE_T size; | |
4334 | INTERNAL_SIZE_T nextsize; | |
4335 | INTERNAL_SIZE_T prevsize; | |
4336 | int nextinuse; | |
4337 | mchunkptr bck; | |
4338 | mchunkptr fwd; | |
10dc2a90 | 4339 | |
3381be5c | 4340 | atomic_store_relaxed (&av->have_fastchunks, false); |
10dc2a90 | 4341 | |
3381be5c | 4342 | unsorted_bin = unsorted_chunks(av); |
a9177ff5 | 4343 | |
3381be5c WD |
4344 | /* |
4345 | Remove each chunk from fast bin and consolidate it, placing it | |
4346 | then in unsorted bin. Among other reasons for doing this, | |
4347 | placing in unsorted bin avoids needing to calculate actual bins | |
4348 | until malloc is sure that chunks aren't immediately going to be | |
4349 | reused anyway. | |
4350 | */ | |
72f90263 | 4351 | |
3381be5c WD |
4352 | maxfb = &fastbin (av, NFASTBINS - 1); |
4353 | fb = &fastbin (av, 0); | |
4354 | do { | |
4355 | p = atomic_exchange_acq (fb, NULL); | |
4356 | if (p != 0) { | |
4357 | do { | |
4358 | check_inuse_chunk(av, p); | |
4359 | nextp = p->fd; | |
4360 | ||
4361 | /* Slightly streamlined version of consolidation code in free() */ | |
4362 | size = chunksize (p); | |
4363 | nextchunk = chunk_at_offset(p, size); | |
4364 | nextsize = chunksize(nextchunk); | |
4365 | ||
4366 | if (!prev_inuse(p)) { | |
4367 | prevsize = prev_size (p); | |
4368 | size += prevsize; | |
4369 | p = chunk_at_offset(p, -((long) prevsize)); | |
4370 | unlink(av, p, bck, fwd); | |
4371 | } | |
72f90263 | 4372 | |
3381be5c WD |
4373 | if (nextchunk != av->top) { |
4374 | nextinuse = inuse_bit_at_offset(nextchunk, nextsize); | |
a9177ff5 | 4375 | |
3381be5c WD |
4376 | if (!nextinuse) { |
4377 | size += nextsize; | |
4378 | unlink(av, nextchunk, bck, fwd); | |
4379 | } else | |
4380 | clear_inuse_bit_at_offset(nextchunk, 0); | |
a9177ff5 | 4381 | |
3381be5c WD |
4382 | first_unsorted = unsorted_bin->fd; |
4383 | unsorted_bin->fd = p; | |
4384 | first_unsorted->bk = p; | |
7ecfbd38 | 4385 | |
3381be5c WD |
4386 | if (!in_smallbin_range (size)) { |
4387 | p->fd_nextsize = NULL; | |
4388 | p->bk_nextsize = NULL; | |
72f90263 | 4389 | } |
a9177ff5 | 4390 | |
3381be5c WD |
4391 | set_head(p, size | PREV_INUSE); |
4392 | p->bk = unsorted_bin; | |
4393 | p->fd = first_unsorted; | |
4394 | set_foot(p, size); | |
4395 | } | |
a9177ff5 | 4396 | |
3381be5c WD |
4397 | else { |
4398 | size += nextsize; | |
4399 | set_head(p, size | PREV_INUSE); | |
4400 | av->top = p; | |
4401 | } | |
a9177ff5 | 4402 | |
3381be5c WD |
4403 | } while ( (p = nextp) != 0); |
4404 | ||
4405 | } | |
4406 | } while (fb++ != maxfb); | |
fa8d436c | 4407 | } |
10dc2a90 | 4408 | |
fa8d436c UD |
4409 | /* |
4410 | ------------------------------ realloc ------------------------------ | |
4411 | */ | |
f65fd747 | 4412 | |
22a89187 | 4413 | void* |
4c8b8cc3 UD |
4414 | _int_realloc(mstate av, mchunkptr oldp, INTERNAL_SIZE_T oldsize, |
4415 | INTERNAL_SIZE_T nb) | |
fa8d436c | 4416 | { |
fa8d436c UD |
4417 | mchunkptr newp; /* chunk to return */ |
4418 | INTERNAL_SIZE_T newsize; /* its size */ | |
22a89187 | 4419 | void* newmem; /* corresponding user mem */ |
f65fd747 | 4420 | |
fa8d436c | 4421 | mchunkptr next; /* next contiguous chunk after oldp */ |
f65fd747 | 4422 | |
fa8d436c UD |
4423 | mchunkptr remainder; /* extra space at end of newp */ |
4424 | unsigned long remainder_size; /* its size */ | |
f65fd747 | 4425 | |
fa8d436c UD |
4426 | mchunkptr bck; /* misc temp for linking */ |
4427 | mchunkptr fwd; /* misc temp for linking */ | |
2ed5fd9a | 4428 | |
fa8d436c UD |
4429 | unsigned long copysize; /* bytes to copy */ |
4430 | unsigned int ncopies; /* INTERNAL_SIZE_T words to copy */ | |
a9177ff5 | 4431 | INTERNAL_SIZE_T* s; /* copy source */ |
fa8d436c | 4432 | INTERNAL_SIZE_T* d; /* copy destination */ |
f65fd747 | 4433 | |
6dd6a580 | 4434 | /* oldmem size */ |
e9c4fe93 | 4435 | if (__builtin_expect (chunksize_nomask (oldp) <= 2 * SIZE_SZ, 0) |
76761b63 | 4436 | || __builtin_expect (oldsize >= av->system_mem, 0)) |
ac3ed168 | 4437 | malloc_printerr ("realloc(): invalid old size"); |
76761b63 | 4438 | |
6c8dbf00 | 4439 | check_inuse_chunk (av, oldp); |
f65fd747 | 4440 | |
4c8b8cc3 | 4441 | /* All callers already filter out mmap'ed chunks. */ |
6c8dbf00 | 4442 | assert (!chunk_is_mmapped (oldp)); |
f65fd747 | 4443 | |
6c8dbf00 OB |
4444 | next = chunk_at_offset (oldp, oldsize); |
4445 | INTERNAL_SIZE_T nextsize = chunksize (next); | |
e9c4fe93 | 4446 | if (__builtin_expect (chunksize_nomask (next) <= 2 * SIZE_SZ, 0) |
22a89187 | 4447 | || __builtin_expect (nextsize >= av->system_mem, 0)) |
ac3ed168 | 4448 | malloc_printerr ("realloc(): invalid next size"); |
22a89187 | 4449 | |
6c8dbf00 OB |
4450 | if ((unsigned long) (oldsize) >= (unsigned long) (nb)) |
4451 | { | |
4452 | /* already big enough; split below */ | |
fa8d436c | 4453 | newp = oldp; |
6c8dbf00 | 4454 | newsize = oldsize; |
7799b7b3 | 4455 | } |
f65fd747 | 4456 | |
6c8dbf00 OB |
4457 | else |
4458 | { | |
4459 | /* Try to expand forward into top */ | |
4460 | if (next == av->top && | |
4461 | (unsigned long) (newsize = oldsize + nextsize) >= | |
4462 | (unsigned long) (nb + MINSIZE)) | |
4463 | { | |
4464 | set_head_size (oldp, nb | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
4465 | av->top = chunk_at_offset (oldp, nb); | |
4466 | set_head (av->top, (newsize - nb) | PREV_INUSE); | |
4467 | check_inuse_chunk (av, oldp); | |
4468 | return chunk2mem (oldp); | |
4469 | } | |
4470 | ||
4471 | /* Try to expand forward into next chunk; split off remainder below */ | |
4472 | else if (next != av->top && | |
4473 | !inuse (next) && | |
4474 | (unsigned long) (newsize = oldsize + nextsize) >= | |
4475 | (unsigned long) (nb)) | |
4476 | { | |
4477 | newp = oldp; | |
fff94fa2 | 4478 | unlink (av, next, bck, fwd); |
6c8dbf00 OB |
4479 | } |
4480 | ||
4481 | /* allocate, copy, free */ | |
4482 | else | |
4483 | { | |
4484 | newmem = _int_malloc (av, nb - MALLOC_ALIGN_MASK); | |
4485 | if (newmem == 0) | |
4486 | return 0; /* propagate failure */ | |
4487 | ||
4488 | newp = mem2chunk (newmem); | |
4489 | newsize = chunksize (newp); | |
4490 | ||
4491 | /* | |
4492 | Avoid copy if newp is next chunk after oldp. | |
4493 | */ | |
4494 | if (newp == next) | |
4495 | { | |
4496 | newsize += oldsize; | |
4497 | newp = oldp; | |
4498 | } | |
4499 | else | |
4500 | { | |
4501 | /* | |
4502 | Unroll copy of <= 36 bytes (72 if 8byte sizes) | |
4503 | We know that contents have an odd number of | |
4504 | INTERNAL_SIZE_T-sized words; minimally 3. | |
4505 | */ | |
4506 | ||
4507 | copysize = oldsize - SIZE_SZ; | |
4508 | s = (INTERNAL_SIZE_T *) (chunk2mem (oldp)); | |
4509 | d = (INTERNAL_SIZE_T *) (newmem); | |
4510 | ncopies = copysize / sizeof (INTERNAL_SIZE_T); | |
4511 | assert (ncopies >= 3); | |
4512 | ||
4513 | if (ncopies > 9) | |
4514 | memcpy (d, s, copysize); | |
4515 | ||
4516 | else | |
4517 | { | |
4518 | *(d + 0) = *(s + 0); | |
4519 | *(d + 1) = *(s + 1); | |
4520 | *(d + 2) = *(s + 2); | |
4521 | if (ncopies > 4) | |
4522 | { | |
4523 | *(d + 3) = *(s + 3); | |
4524 | *(d + 4) = *(s + 4); | |
4525 | if (ncopies > 6) | |
4526 | { | |
4527 | *(d + 5) = *(s + 5); | |
4528 | *(d + 6) = *(s + 6); | |
4529 | if (ncopies > 8) | |
4530 | { | |
4531 | *(d + 7) = *(s + 7); | |
4532 | *(d + 8) = *(s + 8); | |
4533 | } | |
4534 | } | |
4535 | } | |
4536 | } | |
4537 | ||
4538 | _int_free (av, oldp, 1); | |
4539 | check_inuse_chunk (av, newp); | |
4540 | return chunk2mem (newp); | |
4541 | } | |
4542 | } | |
fa8d436c | 4543 | } |
f65fd747 | 4544 | |
22a89187 | 4545 | /* If possible, free extra space in old or extended chunk */ |
f65fd747 | 4546 | |
6c8dbf00 | 4547 | assert ((unsigned long) (newsize) >= (unsigned long) (nb)); |
f65fd747 | 4548 | |
22a89187 | 4549 | remainder_size = newsize - nb; |
10dc2a90 | 4550 | |
6c8dbf00 OB |
4551 | if (remainder_size < MINSIZE) /* not enough extra to split off */ |
4552 | { | |
4553 | set_head_size (newp, newsize | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
4554 | set_inuse_bit_at_offset (newp, newsize); | |
4555 | } | |
4556 | else /* split remainder */ | |
4557 | { | |
4558 | remainder = chunk_at_offset (newp, nb); | |
4559 | set_head_size (newp, nb | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
4560 | set_head (remainder, remainder_size | PREV_INUSE | | |
4561 | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
4562 | /* Mark remainder as inuse so free() won't complain */ | |
4563 | set_inuse_bit_at_offset (remainder, remainder_size); | |
4564 | _int_free (av, remainder, 1); | |
4565 | } | |
22a89187 | 4566 | |
6c8dbf00 OB |
4567 | check_inuse_chunk (av, newp); |
4568 | return chunk2mem (newp); | |
fa8d436c UD |
4569 | } |
4570 | ||
4571 | /* | |
6c8dbf00 OB |
4572 | ------------------------------ memalign ------------------------------ |
4573 | */ | |
fa8d436c | 4574 | |
6c8dbf00 OB |
4575 | static void * |
4576 | _int_memalign (mstate av, size_t alignment, size_t bytes) | |
fa8d436c UD |
4577 | { |
4578 | INTERNAL_SIZE_T nb; /* padded request size */ | |
6c8dbf00 OB |
4579 | char *m; /* memory returned by malloc call */ |
4580 | mchunkptr p; /* corresponding chunk */ | |
4581 | char *brk; /* alignment point within p */ | |
4582 | mchunkptr newp; /* chunk to return */ | |
fa8d436c UD |
4583 | INTERNAL_SIZE_T newsize; /* its size */ |
4584 | INTERNAL_SIZE_T leadsize; /* leading space before alignment point */ | |
6c8dbf00 OB |
4585 | mchunkptr remainder; /* spare room at end to split off */ |
4586 | unsigned long remainder_size; /* its size */ | |
fa8d436c | 4587 | INTERNAL_SIZE_T size; |
f65fd747 | 4588 | |
f65fd747 | 4589 | |
f65fd747 | 4590 | |
6c8dbf00 | 4591 | checked_request2size (bytes, nb); |
fa8d436c UD |
4592 | |
4593 | /* | |
6c8dbf00 OB |
4594 | Strategy: find a spot within that chunk that meets the alignment |
4595 | request, and then possibly free the leading and trailing space. | |
4596 | */ | |
fa8d436c UD |
4597 | |
4598 | ||
4599 | /* Call malloc with worst case padding to hit alignment. */ | |
4600 | ||
6c8dbf00 OB |
4601 | m = (char *) (_int_malloc (av, nb + alignment + MINSIZE)); |
4602 | ||
4603 | if (m == 0) | |
4604 | return 0; /* propagate failure */ | |
4605 | ||
4606 | p = mem2chunk (m); | |
4607 | ||
4608 | if ((((unsigned long) (m)) % alignment) != 0) /* misaligned */ | |
4609 | ||
4610 | { /* | |
4611 | Find an aligned spot inside chunk. Since we need to give back | |
4612 | leading space in a chunk of at least MINSIZE, if the first | |
4613 | calculation places us at a spot with less than MINSIZE leader, | |
4614 | we can move to the next aligned spot -- we've allocated enough | |
4615 | total room so that this is always possible. | |
4616 | */ | |
4617 | brk = (char *) mem2chunk (((unsigned long) (m + alignment - 1)) & | |
4618 | - ((signed long) alignment)); | |
4619 | if ((unsigned long) (brk - (char *) (p)) < MINSIZE) | |
4620 | brk += alignment; | |
4621 | ||
4622 | newp = (mchunkptr) brk; | |
4623 | leadsize = brk - (char *) (p); | |
4624 | newsize = chunksize (p) - leadsize; | |
4625 | ||
4626 | /* For mmapped chunks, just adjust offset */ | |
4627 | if (chunk_is_mmapped (p)) | |
4628 | { | |
e9c4fe93 | 4629 | set_prev_size (newp, prev_size (p) + leadsize); |
6c8dbf00 OB |
4630 | set_head (newp, newsize | IS_MMAPPED); |
4631 | return chunk2mem (newp); | |
4632 | } | |
4633 | ||
4634 | /* Otherwise, give back leader, use the rest */ | |
4635 | set_head (newp, newsize | PREV_INUSE | | |
4636 | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
4637 | set_inuse_bit_at_offset (newp, newsize); | |
4638 | set_head_size (p, leadsize | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
4639 | _int_free (av, p, 1); | |
4640 | p = newp; | |
4641 | ||
4642 | assert (newsize >= nb && | |
4643 | (((unsigned long) (chunk2mem (p))) % alignment) == 0); | |
f65fd747 | 4644 | } |
f65fd747 | 4645 | |
f65fd747 | 4646 | /* Also give back spare room at the end */ |
6c8dbf00 OB |
4647 | if (!chunk_is_mmapped (p)) |
4648 | { | |
4649 | size = chunksize (p); | |
4650 | if ((unsigned long) (size) > (unsigned long) (nb + MINSIZE)) | |
4651 | { | |
4652 | remainder_size = size - nb; | |
4653 | remainder = chunk_at_offset (p, nb); | |
4654 | set_head (remainder, remainder_size | PREV_INUSE | | |
4655 | (av != &main_arena ? NON_MAIN_ARENA : 0)); | |
4656 | set_head_size (p, nb); | |
4657 | _int_free (av, remainder, 1); | |
4658 | } | |
fa8d436c | 4659 | } |
f65fd747 | 4660 | |
6c8dbf00 OB |
4661 | check_inuse_chunk (av, p); |
4662 | return chunk2mem (p); | |
f65fd747 UD |
4663 | } |
4664 | ||
f65fd747 | 4665 | |
fa8d436c | 4666 | /* |
6c8dbf00 OB |
4667 | ------------------------------ malloc_trim ------------------------------ |
4668 | */ | |
8a4b65b4 | 4669 | |
6c8dbf00 OB |
4670 | static int |
4671 | mtrim (mstate av, size_t pad) | |
f65fd747 | 4672 | { |
3381be5c | 4673 | /* Ensure all blocks are consolidated. */ |
68631c8e UD |
4674 | malloc_consolidate (av); |
4675 | ||
6c8dbf00 | 4676 | const size_t ps = GLRO (dl_pagesize); |
68631c8e UD |
4677 | int psindex = bin_index (ps); |
4678 | const size_t psm1 = ps - 1; | |
4679 | ||
4680 | int result = 0; | |
4681 | for (int i = 1; i < NBINS; ++i) | |
4682 | if (i == 1 || i >= psindex) | |
4683 | { | |
6c8dbf00 | 4684 | mbinptr bin = bin_at (av, i); |
68631c8e | 4685 | |
6c8dbf00 OB |
4686 | for (mchunkptr p = last (bin); p != bin; p = p->bk) |
4687 | { | |
4688 | INTERNAL_SIZE_T size = chunksize (p); | |
68631c8e | 4689 | |
6c8dbf00 OB |
4690 | if (size > psm1 + sizeof (struct malloc_chunk)) |
4691 | { | |
4692 | /* See whether the chunk contains at least one unused page. */ | |
4693 | char *paligned_mem = (char *) (((uintptr_t) p | |
4694 | + sizeof (struct malloc_chunk) | |
4695 | + psm1) & ~psm1); | |
68631c8e | 4696 | |
6c8dbf00 OB |
4697 | assert ((char *) chunk2mem (p) + 4 * SIZE_SZ <= paligned_mem); |
4698 | assert ((char *) p + size > paligned_mem); | |
68631c8e | 4699 | |
6c8dbf00 OB |
4700 | /* This is the size we could potentially free. */ |
4701 | size -= paligned_mem - (char *) p; | |
68631c8e | 4702 | |
6c8dbf00 OB |
4703 | if (size > psm1) |
4704 | { | |
439bda32 | 4705 | #if MALLOC_DEBUG |
6c8dbf00 OB |
4706 | /* When debugging we simulate destroying the memory |
4707 | content. */ | |
4708 | memset (paligned_mem, 0x89, size & ~psm1); | |
68631c8e | 4709 | #endif |
6c8dbf00 | 4710 | __madvise (paligned_mem, size & ~psm1, MADV_DONTNEED); |
68631c8e | 4711 | |
6c8dbf00 OB |
4712 | result = 1; |
4713 | } | |
4714 | } | |
4715 | } | |
68631c8e | 4716 | } |
8a4b65b4 | 4717 | |
a9177ff5 | 4718 | #ifndef MORECORE_CANNOT_TRIM |
3b49edc0 | 4719 | return result | (av == &main_arena ? systrim (pad, av) : 0); |
6c8dbf00 | 4720 | |
8a4b65b4 | 4721 | #else |
68631c8e | 4722 | return result; |
f65fd747 | 4723 | #endif |
f65fd747 UD |
4724 | } |
4725 | ||
f65fd747 | 4726 | |
3b49edc0 | 4727 | int |
6c8dbf00 | 4728 | __malloc_trim (size_t s) |
3b49edc0 UD |
4729 | { |
4730 | int result = 0; | |
4731 | ||
6c8dbf00 | 4732 | if (__malloc_initialized < 0) |
3b49edc0 UD |
4733 | ptmalloc_init (); |
4734 | ||
4735 | mstate ar_ptr = &main_arena; | |
4736 | do | |
4737 | { | |
4bf5f222 | 4738 | __libc_lock_lock (ar_ptr->mutex); |
3b49edc0 | 4739 | result |= mtrim (ar_ptr, s); |
4bf5f222 | 4740 | __libc_lock_unlock (ar_ptr->mutex); |
3b49edc0 UD |
4741 | |
4742 | ar_ptr = ar_ptr->next; | |
4743 | } | |
4744 | while (ar_ptr != &main_arena); | |
4745 | ||
4746 | return result; | |
4747 | } | |
4748 | ||
4749 | ||
f65fd747 | 4750 | /* |
6c8dbf00 OB |
4751 | ------------------------- malloc_usable_size ------------------------- |
4752 | */ | |
f65fd747 | 4753 | |
3b49edc0 | 4754 | static size_t |
6c8dbf00 | 4755 | musable (void *mem) |
f65fd747 UD |
4756 | { |
4757 | mchunkptr p; | |
6c8dbf00 OB |
4758 | if (mem != 0) |
4759 | { | |
4760 | p = mem2chunk (mem); | |
4761 | ||
4762 | if (__builtin_expect (using_malloc_checking == 1, 0)) | |
4763 | return malloc_check_get_size (p); | |
4764 | ||
4765 | if (chunk_is_mmapped (p)) | |
073f8214 FW |
4766 | { |
4767 | if (DUMPED_MAIN_ARENA_CHUNK (p)) | |
4768 | return chunksize (p) - SIZE_SZ; | |
4769 | else | |
4770 | return chunksize (p) - 2 * SIZE_SZ; | |
4771 | } | |
6c8dbf00 OB |
4772 | else if (inuse (p)) |
4773 | return chunksize (p) - SIZE_SZ; | |
4774 | } | |
fa8d436c | 4775 | return 0; |
f65fd747 UD |
4776 | } |
4777 | ||
3b49edc0 UD |
4778 | |
4779 | size_t | |
6c8dbf00 | 4780 | __malloc_usable_size (void *m) |
3b49edc0 UD |
4781 | { |
4782 | size_t result; | |
4783 | ||
6c8dbf00 | 4784 | result = musable (m); |
3b49edc0 UD |
4785 | return result; |
4786 | } | |
4787 | ||
fa8d436c | 4788 | /* |
6c8dbf00 OB |
4789 | ------------------------------ mallinfo ------------------------------ |
4790 | Accumulate malloc statistics for arena AV into M. | |
4791 | */ | |
f65fd747 | 4792 | |
bedee953 | 4793 | static void |
6c8dbf00 | 4794 | int_mallinfo (mstate av, struct mallinfo *m) |
f65fd747 | 4795 | { |
6dd67bd5 | 4796 | size_t i; |
f65fd747 UD |
4797 | mbinptr b; |
4798 | mchunkptr p; | |
f65fd747 | 4799 | INTERNAL_SIZE_T avail; |
fa8d436c UD |
4800 | INTERNAL_SIZE_T fastavail; |
4801 | int nblocks; | |
4802 | int nfastblocks; | |
f65fd747 | 4803 | |
6c8dbf00 | 4804 | check_malloc_state (av); |
8a4b65b4 | 4805 | |
fa8d436c | 4806 | /* Account for top */ |
6c8dbf00 | 4807 | avail = chunksize (av->top); |
fa8d436c | 4808 | nblocks = 1; /* top always exists */ |
f65fd747 | 4809 | |
fa8d436c UD |
4810 | /* traverse fastbins */ |
4811 | nfastblocks = 0; | |
4812 | fastavail = 0; | |
4813 | ||
6c8dbf00 OB |
4814 | for (i = 0; i < NFASTBINS; ++i) |
4815 | { | |
4816 | for (p = fastbin (av, i); p != 0; p = p->fd) | |
4817 | { | |
4818 | ++nfastblocks; | |
4819 | fastavail += chunksize (p); | |
4820 | } | |
fa8d436c | 4821 | } |
fa8d436c UD |
4822 | |
4823 | avail += fastavail; | |
f65fd747 | 4824 | |
fa8d436c | 4825 | /* traverse regular bins */ |
6c8dbf00 OB |
4826 | for (i = 1; i < NBINS; ++i) |
4827 | { | |
4828 | b = bin_at (av, i); | |
4829 | for (p = last (b); p != b; p = p->bk) | |
4830 | { | |
4831 | ++nblocks; | |
4832 | avail += chunksize (p); | |
4833 | } | |
fa8d436c | 4834 | } |
f65fd747 | 4835 | |
bedee953 PP |
4836 | m->smblks += nfastblocks; |
4837 | m->ordblks += nblocks; | |
4838 | m->fordblks += avail; | |
4839 | m->uordblks += av->system_mem - avail; | |
4840 | m->arena += av->system_mem; | |
4841 | m->fsmblks += fastavail; | |
4842 | if (av == &main_arena) | |
4843 | { | |
4844 | m->hblks = mp_.n_mmaps; | |
4845 | m->hblkhd = mp_.mmapped_mem; | |
ca135f82 | 4846 | m->usmblks = 0; |
6c8dbf00 | 4847 | m->keepcost = chunksize (av->top); |
bedee953 | 4848 | } |
fa8d436c | 4849 | } |
f65fd747 | 4850 | |
3b49edc0 | 4851 | |
6c8dbf00 | 4852 | struct mallinfo |
9dd346ff | 4853 | __libc_mallinfo (void) |
3b49edc0 UD |
4854 | { |
4855 | struct mallinfo m; | |
bedee953 | 4856 | mstate ar_ptr; |
3b49edc0 | 4857 | |
6c8dbf00 | 4858 | if (__malloc_initialized < 0) |
3b49edc0 | 4859 | ptmalloc_init (); |
bedee953 | 4860 | |
6c8dbf00 | 4861 | memset (&m, 0, sizeof (m)); |
bedee953 | 4862 | ar_ptr = &main_arena; |
6c8dbf00 OB |
4863 | do |
4864 | { | |
4bf5f222 | 4865 | __libc_lock_lock (ar_ptr->mutex); |
6c8dbf00 | 4866 | int_mallinfo (ar_ptr, &m); |
4bf5f222 | 4867 | __libc_lock_unlock (ar_ptr->mutex); |
bedee953 | 4868 | |
6c8dbf00 OB |
4869 | ar_ptr = ar_ptr->next; |
4870 | } | |
4871 | while (ar_ptr != &main_arena); | |
bedee953 | 4872 | |
3b49edc0 UD |
4873 | return m; |
4874 | } | |
4875 | ||
fa8d436c | 4876 | /* |
6c8dbf00 OB |
4877 | ------------------------------ malloc_stats ------------------------------ |
4878 | */ | |
f65fd747 | 4879 | |
3b49edc0 | 4880 | void |
60d2f8f3 | 4881 | __malloc_stats (void) |
f65fd747 | 4882 | { |
8a4b65b4 | 4883 | int i; |
fa8d436c | 4884 | mstate ar_ptr; |
fa8d436c | 4885 | unsigned int in_use_b = mp_.mmapped_mem, system_b = in_use_b; |
8a4b65b4 | 4886 | |
6c8dbf00 | 4887 | if (__malloc_initialized < 0) |
a234e27d | 4888 | ptmalloc_init (); |
8dab36a1 UD |
4889 | _IO_flockfile (stderr); |
4890 | int old_flags2 = ((_IO_FILE *) stderr)->_flags2; | |
4891 | ((_IO_FILE *) stderr)->_flags2 |= _IO_FLAGS2_NOTCANCEL; | |
6c8dbf00 OB |
4892 | for (i = 0, ar_ptr = &main_arena;; i++) |
4893 | { | |
4894 | struct mallinfo mi; | |
4895 | ||
4896 | memset (&mi, 0, sizeof (mi)); | |
4bf5f222 | 4897 | __libc_lock_lock (ar_ptr->mutex); |
6c8dbf00 OB |
4898 | int_mallinfo (ar_ptr, &mi); |
4899 | fprintf (stderr, "Arena %d:\n", i); | |
4900 | fprintf (stderr, "system bytes = %10u\n", (unsigned int) mi.arena); | |
4901 | fprintf (stderr, "in use bytes = %10u\n", (unsigned int) mi.uordblks); | |
fa8d436c | 4902 | #if MALLOC_DEBUG > 1 |
6c8dbf00 OB |
4903 | if (i > 0) |
4904 | dump_heap (heap_for_ptr (top (ar_ptr))); | |
fa8d436c | 4905 | #endif |
6c8dbf00 OB |
4906 | system_b += mi.arena; |
4907 | in_use_b += mi.uordblks; | |
4bf5f222 | 4908 | __libc_lock_unlock (ar_ptr->mutex); |
6c8dbf00 OB |
4909 | ar_ptr = ar_ptr->next; |
4910 | if (ar_ptr == &main_arena) | |
4911 | break; | |
4912 | } | |
4913 | fprintf (stderr, "Total (incl. mmap):\n"); | |
4914 | fprintf (stderr, "system bytes = %10u\n", system_b); | |
4915 | fprintf (stderr, "in use bytes = %10u\n", in_use_b); | |
4916 | fprintf (stderr, "max mmap regions = %10u\n", (unsigned int) mp_.max_n_mmaps); | |
4917 | fprintf (stderr, "max mmap bytes = %10lu\n", | |
4918 | (unsigned long) mp_.max_mmapped_mem); | |
8dab36a1 UD |
4919 | ((_IO_FILE *) stderr)->_flags2 |= old_flags2; |
4920 | _IO_funlockfile (stderr); | |
f65fd747 UD |
4921 | } |
4922 | ||
f65fd747 UD |
4923 | |
4924 | /* | |
6c8dbf00 OB |
4925 | ------------------------------ mallopt ------------------------------ |
4926 | */ | |
be7991c0 SP |
4927 | static inline int |
4928 | __always_inline | |
4929 | do_set_trim_threshold (size_t value) | |
4930 | { | |
4931 | LIBC_PROBE (memory_mallopt_trim_threshold, 3, value, mp_.trim_threshold, | |
4932 | mp_.no_dyn_threshold); | |
4933 | mp_.trim_threshold = value; | |
4934 | mp_.no_dyn_threshold = 1; | |
4935 | return 1; | |
4936 | } | |
4937 | ||
4938 | static inline int | |
4939 | __always_inline | |
4940 | do_set_top_pad (size_t value) | |
4941 | { | |
4942 | LIBC_PROBE (memory_mallopt_top_pad, 3, value, mp_.top_pad, | |
4943 | mp_.no_dyn_threshold); | |
4944 | mp_.top_pad = value; | |
4945 | mp_.no_dyn_threshold = 1; | |
4946 | return 1; | |
4947 | } | |
4948 | ||
4949 | static inline int | |
4950 | __always_inline | |
4951 | do_set_mmap_threshold (size_t value) | |
4952 | { | |
4953 | /* Forbid setting the threshold too high. */ | |
4954 | if (value <= HEAP_MAX_SIZE / 2) | |
4955 | { | |
4956 | LIBC_PROBE (memory_mallopt_mmap_threshold, 3, value, mp_.mmap_threshold, | |
4957 | mp_.no_dyn_threshold); | |
4958 | mp_.mmap_threshold = value; | |
4959 | mp_.no_dyn_threshold = 1; | |
4960 | return 1; | |
4961 | } | |
4962 | return 0; | |
4963 | } | |
4964 | ||
4965 | static inline int | |
4966 | __always_inline | |
4967 | do_set_mmaps_max (int32_t value) | |
4968 | { | |
4969 | LIBC_PROBE (memory_mallopt_mmap_max, 3, value, mp_.n_mmaps_max, | |
4970 | mp_.no_dyn_threshold); | |
4971 | mp_.n_mmaps_max = value; | |
4972 | mp_.no_dyn_threshold = 1; | |
4973 | return 1; | |
4974 | } | |
4975 | ||
4976 | static inline int | |
4977 | __always_inline | |
4978 | do_set_mallopt_check (int32_t value) | |
4979 | { | |
be7991c0 SP |
4980 | return 1; |
4981 | } | |
4982 | ||
4983 | static inline int | |
4984 | __always_inline | |
4985 | do_set_perturb_byte (int32_t value) | |
4986 | { | |
4987 | LIBC_PROBE (memory_mallopt_perturb, 2, value, perturb_byte); | |
4988 | perturb_byte = value; | |
4989 | return 1; | |
4990 | } | |
4991 | ||
4992 | static inline int | |
4993 | __always_inline | |
4994 | do_set_arena_test (size_t value) | |
4995 | { | |
4996 | LIBC_PROBE (memory_mallopt_arena_test, 2, value, mp_.arena_test); | |
4997 | mp_.arena_test = value; | |
4998 | return 1; | |
4999 | } | |
5000 | ||
5001 | static inline int | |
5002 | __always_inline | |
5003 | do_set_arena_max (size_t value) | |
5004 | { | |
5005 | LIBC_PROBE (memory_mallopt_arena_max, 2, value, mp_.arena_max); | |
5006 | mp_.arena_max = value; | |
5007 | return 1; | |
5008 | } | |
5009 | ||
d5c3fafc DD |
5010 | #if USE_TCACHE |
5011 | static inline int | |
5012 | __always_inline | |
5013 | do_set_tcache_max (size_t value) | |
5014 | { | |
5015 | if (value >= 0 && value <= MAX_TCACHE_SIZE) | |
5016 | { | |
5017 | LIBC_PROBE (memory_tunable_tcache_max_bytes, 2, value, mp_.tcache_max_bytes); | |
5018 | mp_.tcache_max_bytes = value; | |
5019 | mp_.tcache_bins = csize2tidx (request2size(value)) + 1; | |
5020 | } | |
5021 | return 1; | |
5022 | } | |
5023 | ||
5024 | static inline int | |
5025 | __always_inline | |
5026 | do_set_tcache_count (size_t value) | |
5027 | { | |
5028 | LIBC_PROBE (memory_tunable_tcache_count, 2, value, mp_.tcache_count); | |
5029 | mp_.tcache_count = value; | |
5030 | return 1; | |
5031 | } | |
5032 | ||
5033 | static inline int | |
5034 | __always_inline | |
5035 | do_set_tcache_unsorted_limit (size_t value) | |
5036 | { | |
5037 | LIBC_PROBE (memory_tunable_tcache_unsorted_limit, 2, value, mp_.tcache_unsorted_limit); | |
5038 | mp_.tcache_unsorted_limit = value; | |
5039 | return 1; | |
5040 | } | |
5041 | #endif | |
f65fd747 | 5042 | |
6c8dbf00 OB |
5043 | int |
5044 | __libc_mallopt (int param_number, int value) | |
f65fd747 | 5045 | { |
fa8d436c UD |
5046 | mstate av = &main_arena; |
5047 | int res = 1; | |
f65fd747 | 5048 | |
6c8dbf00 | 5049 | if (__malloc_initialized < 0) |
0cb71e02 | 5050 | ptmalloc_init (); |
4bf5f222 | 5051 | __libc_lock_lock (av->mutex); |
2f6d1f1b | 5052 | |
3ea5be54 AO |
5053 | LIBC_PROBE (memory_mallopt, 2, param_number, value); |
5054 | ||
3381be5c WD |
5055 | /* We must consolidate main arena before changing max_fast |
5056 | (see definition of set_max_fast). */ | |
5057 | malloc_consolidate (av); | |
5058 | ||
6c8dbf00 OB |
5059 | switch (param_number) |
5060 | { | |
5061 | case M_MXFAST: | |
5062 | if (value >= 0 && value <= MAX_FAST_SIZE) | |
5063 | { | |
5064 | LIBC_PROBE (memory_mallopt_mxfast, 2, value, get_max_fast ()); | |
5065 | set_max_fast (value); | |
5066 | } | |
5067 | else | |
5068 | res = 0; | |
5069 | break; | |
5070 | ||
5071 | case M_TRIM_THRESHOLD: | |
be7991c0 | 5072 | do_set_trim_threshold (value); |
6c8dbf00 OB |
5073 | break; |
5074 | ||
5075 | case M_TOP_PAD: | |
be7991c0 | 5076 | do_set_top_pad (value); |
6c8dbf00 OB |
5077 | break; |
5078 | ||
5079 | case M_MMAP_THRESHOLD: | |
be7991c0 | 5080 | res = do_set_mmap_threshold (value); |
6c8dbf00 OB |
5081 | break; |
5082 | ||
5083 | case M_MMAP_MAX: | |
be7991c0 | 5084 | do_set_mmaps_max (value); |
6c8dbf00 OB |
5085 | break; |
5086 | ||
5087 | case M_CHECK_ACTION: | |
be7991c0 | 5088 | do_set_mallopt_check (value); |
6c8dbf00 OB |
5089 | break; |
5090 | ||
5091 | case M_PERTURB: | |
be7991c0 | 5092 | do_set_perturb_byte (value); |
6c8dbf00 OB |
5093 | break; |
5094 | ||
5095 | case M_ARENA_TEST: | |
5096 | if (value > 0) | |
be7991c0 | 5097 | do_set_arena_test (value); |
6c8dbf00 OB |
5098 | break; |
5099 | ||
5100 | case M_ARENA_MAX: | |
5101 | if (value > 0) | |
62222284 | 5102 | do_set_arena_max (value); |
6c8dbf00 OB |
5103 | break; |
5104 | } | |
4bf5f222 | 5105 | __libc_lock_unlock (av->mutex); |
fa8d436c | 5106 | return res; |
b22fc5f5 | 5107 | } |
3b49edc0 | 5108 | libc_hidden_def (__libc_mallopt) |
b22fc5f5 | 5109 | |
10dc2a90 | 5110 | |
a9177ff5 | 5111 | /* |
6c8dbf00 OB |
5112 | -------------------- Alternative MORECORE functions -------------------- |
5113 | */ | |
10dc2a90 | 5114 | |
b22fc5f5 | 5115 | |
fa8d436c | 5116 | /* |
6c8dbf00 | 5117 | General Requirements for MORECORE. |
b22fc5f5 | 5118 | |
6c8dbf00 | 5119 | The MORECORE function must have the following properties: |
b22fc5f5 | 5120 | |
6c8dbf00 | 5121 | If MORECORE_CONTIGUOUS is false: |
10dc2a90 | 5122 | |
6c8dbf00 | 5123 | * MORECORE must allocate in multiples of pagesize. It will |
fa8d436c | 5124 | only be called with arguments that are multiples of pagesize. |
10dc2a90 | 5125 | |
6c8dbf00 | 5126 | * MORECORE(0) must return an address that is at least |
fa8d436c | 5127 | MALLOC_ALIGNMENT aligned. (Page-aligning always suffices.) |
10dc2a90 | 5128 | |
6c8dbf00 | 5129 | else (i.e. If MORECORE_CONTIGUOUS is true): |
10dc2a90 | 5130 | |
6c8dbf00 | 5131 | * Consecutive calls to MORECORE with positive arguments |
fa8d436c UD |
5132 | return increasing addresses, indicating that space has been |
5133 | contiguously extended. | |
10dc2a90 | 5134 | |
6c8dbf00 | 5135 | * MORECORE need not allocate in multiples of pagesize. |
fa8d436c | 5136 | Calls to MORECORE need not have args of multiples of pagesize. |
10dc2a90 | 5137 | |
6c8dbf00 | 5138 | * MORECORE need not page-align. |
10dc2a90 | 5139 | |
6c8dbf00 | 5140 | In either case: |
10dc2a90 | 5141 | |
6c8dbf00 | 5142 | * MORECORE may allocate more memory than requested. (Or even less, |
fa8d436c | 5143 | but this will generally result in a malloc failure.) |
10dc2a90 | 5144 | |
6c8dbf00 | 5145 | * MORECORE must not allocate memory when given argument zero, but |
fa8d436c UD |
5146 | instead return one past the end address of memory from previous |
5147 | nonzero call. This malloc does NOT call MORECORE(0) | |
5148 | until at least one call with positive arguments is made, so | |
5149 | the initial value returned is not important. | |
10dc2a90 | 5150 | |
6c8dbf00 | 5151 | * Even though consecutive calls to MORECORE need not return contiguous |
fa8d436c UD |
5152 | addresses, it must be OK for malloc'ed chunks to span multiple |
5153 | regions in those cases where they do happen to be contiguous. | |
10dc2a90 | 5154 | |
6c8dbf00 | 5155 | * MORECORE need not handle negative arguments -- it may instead |
fa8d436c UD |
5156 | just return MORECORE_FAILURE when given negative arguments. |
5157 | Negative arguments are always multiples of pagesize. MORECORE | |
5158 | must not misinterpret negative args as large positive unsigned | |
5159 | args. You can suppress all such calls from even occurring by defining | |
5160 | MORECORE_CANNOT_TRIM, | |
10dc2a90 | 5161 | |
6c8dbf00 OB |
5162 | There is some variation across systems about the type of the |
5163 | argument to sbrk/MORECORE. If size_t is unsigned, then it cannot | |
5164 | actually be size_t, because sbrk supports negative args, so it is | |
5165 | normally the signed type of the same width as size_t (sometimes | |
5166 | declared as "intptr_t", and sometimes "ptrdiff_t"). It doesn't much | |
5167 | matter though. Internally, we use "long" as arguments, which should | |
5168 | work across all reasonable possibilities. | |
5169 | ||
5170 | Additionally, if MORECORE ever returns failure for a positive | |
5171 | request, then mmap is used as a noncontiguous system allocator. This | |
5172 | is a useful backup strategy for systems with holes in address spaces | |
5173 | -- in this case sbrk cannot contiguously expand the heap, but mmap | |
5174 | may be able to map noncontiguous space. | |
5175 | ||
5176 | If you'd like mmap to ALWAYS be used, you can define MORECORE to be | |
5177 | a function that always returns MORECORE_FAILURE. | |
5178 | ||
5179 | If you are using this malloc with something other than sbrk (or its | |
5180 | emulation) to supply memory regions, you probably want to set | |
5181 | MORECORE_CONTIGUOUS as false. As an example, here is a custom | |
5182 | allocator kindly contributed for pre-OSX macOS. It uses virtually | |
5183 | but not necessarily physically contiguous non-paged memory (locked | |
5184 | in, present and won't get swapped out). You can use it by | |
5185 | uncommenting this section, adding some #includes, and setting up the | |
5186 | appropriate defines above: | |
5187 | ||
5188 | *#define MORECORE osMoreCore | |
5189 | *#define MORECORE_CONTIGUOUS 0 | |
5190 | ||
5191 | There is also a shutdown routine that should somehow be called for | |
5192 | cleanup upon program exit. | |
5193 | ||
5194 | *#define MAX_POOL_ENTRIES 100 | |
5195 | *#define MINIMUM_MORECORE_SIZE (64 * 1024) | |
5196 | static int next_os_pool; | |
5197 | void *our_os_pools[MAX_POOL_ENTRIES]; | |
5198 | ||
5199 | void *osMoreCore(int size) | |
5200 | { | |
fa8d436c UD |
5201 | void *ptr = 0; |
5202 | static void *sbrk_top = 0; | |
ca34d7a7 | 5203 | |
fa8d436c UD |
5204 | if (size > 0) |
5205 | { | |
5206 | if (size < MINIMUM_MORECORE_SIZE) | |
6c8dbf00 | 5207 | size = MINIMUM_MORECORE_SIZE; |
fa8d436c | 5208 | if (CurrentExecutionLevel() == kTaskLevel) |
6c8dbf00 | 5209 | ptr = PoolAllocateResident(size + RM_PAGE_SIZE, 0); |
fa8d436c UD |
5210 | if (ptr == 0) |
5211 | { | |
6c8dbf00 | 5212 | return (void *) MORECORE_FAILURE; |
fa8d436c UD |
5213 | } |
5214 | // save ptrs so they can be freed during cleanup | |
5215 | our_os_pools[next_os_pool] = ptr; | |
5216 | next_os_pool++; | |
5217 | ptr = (void *) ((((unsigned long) ptr) + RM_PAGE_MASK) & ~RM_PAGE_MASK); | |
5218 | sbrk_top = (char *) ptr + size; | |
5219 | return ptr; | |
5220 | } | |
5221 | else if (size < 0) | |
5222 | { | |
5223 | // we don't currently support shrink behavior | |
5224 | return (void *) MORECORE_FAILURE; | |
5225 | } | |
5226 | else | |
5227 | { | |
5228 | return sbrk_top; | |
431c33c0 | 5229 | } |
6c8dbf00 | 5230 | } |
ca34d7a7 | 5231 | |
6c8dbf00 OB |
5232 | // cleanup any allocated memory pools |
5233 | // called as last thing before shutting down driver | |
ca34d7a7 | 5234 | |
6c8dbf00 OB |
5235 | void osCleanupMem(void) |
5236 | { | |
fa8d436c | 5237 | void **ptr; |
ca34d7a7 | 5238 | |
fa8d436c UD |
5239 | for (ptr = our_os_pools; ptr < &our_os_pools[MAX_POOL_ENTRIES]; ptr++) |
5240 | if (*ptr) | |
5241 | { | |
6c8dbf00 OB |
5242 | PoolDeallocate(*ptr); |
5243 | * ptr = 0; | |
fa8d436c | 5244 | } |
6c8dbf00 | 5245 | } |
ee74a442 | 5246 | |
6c8dbf00 | 5247 | */ |
f65fd747 | 5248 | |
7e3be507 | 5249 | |
3e030bd5 UD |
5250 | /* Helper code. */ |
5251 | ||
ae7f5313 UD |
5252 | extern char **__libc_argv attribute_hidden; |
5253 | ||
3e030bd5 | 5254 | static void |
ac3ed168 | 5255 | malloc_printerr (const char *str) |
3e030bd5 | 5256 | { |
ec2c1fce FW |
5257 | __libc_message (do_abort, "%s\n", str); |
5258 | __builtin_unreachable (); | |
3e030bd5 UD |
5259 | } |
5260 | ||
a204dbb2 UD |
5261 | /* We need a wrapper function for one of the additions of POSIX. */ |
5262 | int | |
5263 | __posix_memalign (void **memptr, size_t alignment, size_t size) | |
5264 | { | |
5265 | void *mem; | |
5266 | ||
5267 | /* Test whether the SIZE argument is valid. It must be a power of | |
5268 | two multiple of sizeof (void *). */ | |
de02bd05 | 5269 | if (alignment % sizeof (void *) != 0 |
fc56e970 | 5270 | || !powerof2 (alignment / sizeof (void *)) |
de02bd05 | 5271 | || alignment == 0) |
a204dbb2 UD |
5272 | return EINVAL; |
5273 | ||
10ad46bc OB |
5274 | |
5275 | void *address = RETURN_ADDRESS (0); | |
5276 | mem = _mid_memalign (alignment, size, address); | |
a204dbb2 | 5277 | |
6c8dbf00 OB |
5278 | if (mem != NULL) |
5279 | { | |
5280 | *memptr = mem; | |
5281 | return 0; | |
5282 | } | |
a204dbb2 UD |
5283 | |
5284 | return ENOMEM; | |
5285 | } | |
5286 | weak_alias (__posix_memalign, posix_memalign) | |
5287 | ||
20c13899 OB |
5288 | |
5289 | int | |
c52ff39e | 5290 | __malloc_info (int options, FILE *fp) |
bb066545 | 5291 | { |
20c13899 OB |
5292 | /* For now, at least. */ |
5293 | if (options != 0) | |
5294 | return EINVAL; | |
bb066545 | 5295 | |
20c13899 OB |
5296 | int n = 0; |
5297 | size_t total_nblocks = 0; | |
5298 | size_t total_nfastblocks = 0; | |
5299 | size_t total_avail = 0; | |
5300 | size_t total_fastavail = 0; | |
5301 | size_t total_system = 0; | |
5302 | size_t total_max_system = 0; | |
5303 | size_t total_aspace = 0; | |
5304 | size_t total_aspace_mprotect = 0; | |
bb066545 | 5305 | |
6c8dbf00 | 5306 | |
6c8dbf00 | 5307 | |
987c0269 OB |
5308 | if (__malloc_initialized < 0) |
5309 | ptmalloc_init (); | |
bb066545 | 5310 | |
987c0269 | 5311 | fputs ("<malloc version=\"1\">\n", fp); |
bb066545 | 5312 | |
987c0269 OB |
5313 | /* Iterate over all arenas currently in use. */ |
5314 | mstate ar_ptr = &main_arena; | |
5315 | do | |
5316 | { | |
5317 | fprintf (fp, "<heap nr=\"%d\">\n<sizes>\n", n++); | |
8b35e35d | 5318 | |
987c0269 OB |
5319 | size_t nblocks = 0; |
5320 | size_t nfastblocks = 0; | |
5321 | size_t avail = 0; | |
5322 | size_t fastavail = 0; | |
5323 | struct | |
5324 | { | |
5325 | size_t from; | |
5326 | size_t to; | |
5327 | size_t total; | |
5328 | size_t count; | |
5329 | } sizes[NFASTBINS + NBINS - 1]; | |
5330 | #define nsizes (sizeof (sizes) / sizeof (sizes[0])) | |
6c8dbf00 | 5331 | |
4bf5f222 | 5332 | __libc_lock_lock (ar_ptr->mutex); |
bb066545 | 5333 | |
987c0269 OB |
5334 | for (size_t i = 0; i < NFASTBINS; ++i) |
5335 | { | |
5336 | mchunkptr p = fastbin (ar_ptr, i); | |
5337 | if (p != NULL) | |
5338 | { | |
5339 | size_t nthissize = 0; | |
5340 | size_t thissize = chunksize (p); | |
5341 | ||
5342 | while (p != NULL) | |
5343 | { | |
5344 | ++nthissize; | |
5345 | p = p->fd; | |
5346 | } | |
5347 | ||
5348 | fastavail += nthissize * thissize; | |
5349 | nfastblocks += nthissize; | |
5350 | sizes[i].from = thissize - (MALLOC_ALIGNMENT - 1); | |
5351 | sizes[i].to = thissize; | |
5352 | sizes[i].count = nthissize; | |
5353 | } | |
5354 | else | |
5355 | sizes[i].from = sizes[i].to = sizes[i].count = 0; | |
bb066545 | 5356 | |
987c0269 OB |
5357 | sizes[i].total = sizes[i].count * sizes[i].to; |
5358 | } | |
bb066545 | 5359 | |
bb066545 | 5360 | |
987c0269 OB |
5361 | mbinptr bin; |
5362 | struct malloc_chunk *r; | |
bb066545 | 5363 | |
987c0269 OB |
5364 | for (size_t i = 1; i < NBINS; ++i) |
5365 | { | |
5366 | bin = bin_at (ar_ptr, i); | |
5367 | r = bin->fd; | |
5368 | sizes[NFASTBINS - 1 + i].from = ~((size_t) 0); | |
5369 | sizes[NFASTBINS - 1 + i].to = sizes[NFASTBINS - 1 + i].total | |
5370 | = sizes[NFASTBINS - 1 + i].count = 0; | |
5371 | ||
5372 | if (r != NULL) | |
5373 | while (r != bin) | |
5374 | { | |
e9c4fe93 | 5375 | size_t r_size = chunksize_nomask (r); |
987c0269 | 5376 | ++sizes[NFASTBINS - 1 + i].count; |
e9c4fe93 | 5377 | sizes[NFASTBINS - 1 + i].total += r_size; |
987c0269 | 5378 | sizes[NFASTBINS - 1 + i].from |
e9c4fe93 | 5379 | = MIN (sizes[NFASTBINS - 1 + i].from, r_size); |
987c0269 | 5380 | sizes[NFASTBINS - 1 + i].to = MAX (sizes[NFASTBINS - 1 + i].to, |
e9c4fe93 | 5381 | r_size); |
987c0269 OB |
5382 | |
5383 | r = r->fd; | |
5384 | } | |
5385 | ||
5386 | if (sizes[NFASTBINS - 1 + i].count == 0) | |
5387 | sizes[NFASTBINS - 1 + i].from = 0; | |
5388 | nblocks += sizes[NFASTBINS - 1 + i].count; | |
5389 | avail += sizes[NFASTBINS - 1 + i].total; | |
5390 | } | |
bb066545 | 5391 | |
4bf5f222 | 5392 | __libc_lock_unlock (ar_ptr->mutex); |
da2d2fb6 | 5393 | |
987c0269 OB |
5394 | total_nfastblocks += nfastblocks; |
5395 | total_fastavail += fastavail; | |
0588a9cb | 5396 | |
987c0269 OB |
5397 | total_nblocks += nblocks; |
5398 | total_avail += avail; | |
0588a9cb | 5399 | |
987c0269 OB |
5400 | for (size_t i = 0; i < nsizes; ++i) |
5401 | if (sizes[i].count != 0 && i != NFASTBINS) | |
5402 | fprintf (fp, " \ | |
5403 | <size from=\"%zu\" to=\"%zu\" total=\"%zu\" count=\"%zu\"/>\n", | |
5404 | sizes[i].from, sizes[i].to, sizes[i].total, sizes[i].count); | |
fdfd175d | 5405 | |
987c0269 OB |
5406 | if (sizes[NFASTBINS].count != 0) |
5407 | fprintf (fp, "\ | |
5408 | <unsorted from=\"%zu\" to=\"%zu\" total=\"%zu\" count=\"%zu\"/>\n", | |
5409 | sizes[NFASTBINS].from, sizes[NFASTBINS].to, | |
5410 | sizes[NFASTBINS].total, sizes[NFASTBINS].count); | |
fdfd175d | 5411 | |
987c0269 OB |
5412 | total_system += ar_ptr->system_mem; |
5413 | total_max_system += ar_ptr->max_system_mem; | |
bb066545 | 5414 | |
987c0269 OB |
5415 | fprintf (fp, |
5416 | "</sizes>\n<total type=\"fast\" count=\"%zu\" size=\"%zu\"/>\n" | |
5417 | "<total type=\"rest\" count=\"%zu\" size=\"%zu\"/>\n" | |
5418 | "<system type=\"current\" size=\"%zu\"/>\n" | |
5419 | "<system type=\"max\" size=\"%zu\"/>\n", | |
5420 | nfastblocks, fastavail, nblocks, avail, | |
5421 | ar_ptr->system_mem, ar_ptr->max_system_mem); | |
346bc35c | 5422 | |
987c0269 OB |
5423 | if (ar_ptr != &main_arena) |
5424 | { | |
5425 | heap_info *heap = heap_for_ptr (top (ar_ptr)); | |
5426 | fprintf (fp, | |
5427 | "<aspace type=\"total\" size=\"%zu\"/>\n" | |
5428 | "<aspace type=\"mprotect\" size=\"%zu\"/>\n", | |
5429 | heap->size, heap->mprotect_size); | |
5430 | total_aspace += heap->size; | |
5431 | total_aspace_mprotect += heap->mprotect_size; | |
5432 | } | |
5433 | else | |
5434 | { | |
5435 | fprintf (fp, | |
5436 | "<aspace type=\"total\" size=\"%zu\"/>\n" | |
5437 | "<aspace type=\"mprotect\" size=\"%zu\"/>\n", | |
5438 | ar_ptr->system_mem, ar_ptr->system_mem); | |
5439 | total_aspace += ar_ptr->system_mem; | |
5440 | total_aspace_mprotect += ar_ptr->system_mem; | |
5441 | } | |
bb066545 | 5442 | |
987c0269 | 5443 | fputs ("</heap>\n", fp); |
bb066545 UD |
5444 | ar_ptr = ar_ptr->next; |
5445 | } | |
5446 | while (ar_ptr != &main_arena); | |
5447 | ||
5448 | fprintf (fp, | |
62a58816 SP |
5449 | "<total type=\"fast\" count=\"%zu\" size=\"%zu\"/>\n" |
5450 | "<total type=\"rest\" count=\"%zu\" size=\"%zu\"/>\n" | |
9fa76613 | 5451 | "<total type=\"mmap\" count=\"%d\" size=\"%zu\"/>\n" |
62a58816 SP |
5452 | "<system type=\"current\" size=\"%zu\"/>\n" |
5453 | "<system type=\"max\" size=\"%zu\"/>\n" | |
5454 | "<aspace type=\"total\" size=\"%zu\"/>\n" | |
5455 | "<aspace type=\"mprotect\" size=\"%zu\"/>\n" | |
5456 | "</malloc>\n", | |
5457 | total_nfastblocks, total_fastavail, total_nblocks, total_avail, | |
4d653a59 | 5458 | mp_.n_mmaps, mp_.mmapped_mem, |
62a58816 SP |
5459 | total_system, total_max_system, |
5460 | total_aspace, total_aspace_mprotect); | |
bb066545 UD |
5461 | |
5462 | return 0; | |
5463 | } | |
c52ff39e | 5464 | weak_alias (__malloc_info, malloc_info) |
bb066545 UD |
5465 | |
5466 | ||
eba19d2b | 5467 | strong_alias (__libc_calloc, __calloc) weak_alias (__libc_calloc, calloc) |
eba19d2b UD |
5468 | strong_alias (__libc_free, __free) strong_alias (__libc_free, free) |
5469 | strong_alias (__libc_malloc, __malloc) strong_alias (__libc_malloc, malloc) | |
5470 | strong_alias (__libc_memalign, __memalign) | |
5471 | weak_alias (__libc_memalign, memalign) | |
5472 | strong_alias (__libc_realloc, __realloc) strong_alias (__libc_realloc, realloc) | |
5473 | strong_alias (__libc_valloc, __valloc) weak_alias (__libc_valloc, valloc) | |
5474 | strong_alias (__libc_pvalloc, __pvalloc) weak_alias (__libc_pvalloc, pvalloc) | |
5475 | strong_alias (__libc_mallinfo, __mallinfo) | |
5476 | weak_alias (__libc_mallinfo, mallinfo) | |
5477 | strong_alias (__libc_mallopt, __mallopt) weak_alias (__libc_mallopt, mallopt) | |
7e3be507 UD |
5478 | |
5479 | weak_alias (__malloc_stats, malloc_stats) | |
5480 | weak_alias (__malloc_usable_size, malloc_usable_size) | |
5481 | weak_alias (__malloc_trim, malloc_trim) | |
7e3be507 | 5482 | |
025b33ae FW |
5483 | #if SHLIB_COMPAT (libc, GLIBC_2_0, GLIBC_2_26) |
5484 | compat_symbol (libc, __libc_free, cfree, GLIBC_2_0); | |
5485 | #endif | |
f65fd747 | 5486 | |
fa8d436c | 5487 | /* ------------------------------------------------------------ |
6c8dbf00 | 5488 | History: |
f65fd747 | 5489 | |
6c8dbf00 | 5490 | [see ftp://g.oswego.edu/pub/misc/malloc.c for the history of dlmalloc] |
f65fd747 | 5491 | |
6c8dbf00 | 5492 | */ |
fa8d436c UD |
5493 | /* |
5494 | * Local variables: | |
5495 | * c-basic-offset: 2 | |
5496 | * End: | |
5497 | */ |