1 | /* Malloc implementation for multiple threads without lock contention. |
2 | Copyright (C) 1996-2023 Free Software Foundation, Inc. |
3 | Copyright The GNU Toolchain Authors. |
4 | This file is part of the GNU C Library. |
5 | |
6 | The GNU C Library is free software; you can redistribute it and/or |
7 | modify it under the terms of the GNU Lesser General Public License as |
8 | published by the Free Software Foundation; either version 2.1 of the |
9 | License, or (at your option) any later version. |
10 | |
11 | The GNU C Library is distributed in the hope that it will be useful, |
12 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
13 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
14 | Lesser General Public License for more details. |
15 | |
16 | You should have received a copy of the GNU Lesser General Public |
17 | License along with the GNU C Library; see the file COPYING.LIB. If |
18 | not, see <https://www.gnu.org/licenses/>. */ |
19 | |
20 | /* |
21 | This is a version (aka ptmalloc2) of malloc/free/realloc written by |
22 | Doug Lea and adapted to multiple threads/arenas by Wolfram Gloger. |
23 | |
24 | There have been substantial changes made after the integration into |
25 | glibc in all parts of the code. Do not look for much commonality |
26 | with the ptmalloc2 version. |
27 | |
28 | * Version ptmalloc2-20011215 |
29 | based on: |
30 | VERSION 2.7.0 Sun Mar 11 14:14:06 2001 Doug Lea (dl at gee) |
31 | |
32 | * Quickstart |
33 | |
34 | In order to compile this implementation, a Makefile is provided with |
35 | the ptmalloc2 distribution, which has pre-defined targets for some |
36 | popular systems (e.g. "make posix" for Posix threads). All that is |
37 | typically required with regard to compiler flags is the selection of |
38 | the thread package via defining one out of USE_PTHREADS, USE_THR or |
39 | USE_SPROC. Check the thread-m.h file for what effects this has. |
40 | Many/most systems will additionally require USE_TSD_DATA_HACK to be |
41 | defined, so this is the default for "make posix". |
42 | |
43 | * Why use this malloc? |
44 | |
45 | This is not the fastest, most space-conserving, most portable, or |
46 | most tunable malloc ever written. However it is among the fastest |
47 | while also being among the most space-conserving, portable and tunable. |
48 | Consistent balance across these factors results in a good general-purpose |
49 | allocator for malloc-intensive programs. |
50 | |
51 | The main properties of the algorithms are: |
52 | * For large (>= 512 bytes) requests, it is a pure best-fit allocator, |
53 | with ties normally decided via FIFO (i.e. least recently used). |
54 | * For small (<= 64 bytes by default) requests, it is a caching |
55 | allocator, that maintains pools of quickly recycled chunks. |
56 | * In between, and for combinations of large and small requests, it does |
57 | the best it can trying to meet both goals at once. |
58 | * For very large requests (>= 128KB by default), it relies on system |
59 | memory mapping facilities, if supported. |
60 | |
61 | For a longer but slightly out of date high-level description, see |
62 | http://gee.cs.oswego.edu/dl/html/malloc.html |
63 | |
64 | You may already by default be using a C library containing a malloc |
65 | that is based on some version of this malloc (for example in |
66 | linux). You might still want to use the one in this file in order to |
67 | customize settings or to avoid overheads associated with library |
68 | versions. |
69 | |
70 | * Contents, described in more detail in "description of public routines" below. |
71 | |
72 | Standard (ANSI/SVID/...) functions: |
73 | malloc(size_t n); |
74 | calloc(size_t n_elements, size_t element_size); |
75 | free(void* p); |
76 | realloc(void* p, size_t n); |
77 | memalign(size_t alignment, size_t n); |
78 | valloc(size_t n); |
79 | mallinfo() |
80 | mallopt(int parameter_number, int parameter_value) |
81 | |
82 | Additional functions: |
83 | independent_calloc(size_t n_elements, size_t size, void* chunks[]); |
84 | independent_comalloc(size_t n_elements, size_t sizes[], void* chunks[]); |
85 | pvalloc(size_t n); |
86 | malloc_trim(size_t pad); |
87 | malloc_usable_size(void* p); |
88 | malloc_stats(); |
89 | |
90 | * Vital statistics: |
91 | |
92 | Supported pointer representation: 4 or 8 bytes |
93 | Supported size_t representation: 4 or 8 bytes |
94 | Note that size_t is allowed to be 4 bytes even if pointers are 8. |
95 | You can adjust this by defining INTERNAL_SIZE_T |
96 | |
97 | Alignment: 2 * sizeof(size_t) (default) |
98 | (i.e., 8 byte alignment with 4byte size_t). This suffices for |
99 | nearly all current machines and C compilers. However, you can |
100 | define MALLOC_ALIGNMENT to be wider than this if necessary. |
101 | |
102 | Minimum overhead per allocated chunk: 4 or 8 bytes |
103 | Each malloced chunk has a hidden word of overhead holding size |
104 | and status information. |
105 | |
106 | Minimum allocated size: 4-byte ptrs: 16 bytes (including 4 overhead) |
107 | 8-byte ptrs: 24/32 bytes (including, 4/8 overhead) |
108 | |
109 | When a chunk is freed, 12 (for 4byte ptrs) or 20 (for 8 byte |
110 | ptrs but 4 byte size) or 24 (for 8/8) additional bytes are |
111 | needed; 4 (8) for a trailing size field and 8 (16) bytes for |
112 | free list pointers. Thus, the minimum allocatable size is |
113 | 16/24/32 bytes. |
114 | |
115 | Even a request for zero bytes (i.e., malloc(0)) returns a |
116 | pointer to something of the minimum allocatable size. |
117 | |
118 | The maximum overhead wastage (i.e., number of extra bytes |
119 | allocated than were requested in malloc) is less than or equal |
120 | to the minimum size, except for requests >= mmap_threshold that |
121 | are serviced via mmap(), where the worst case wastage is 2 * |
122 | sizeof(size_t) bytes plus the remainder from a system page (the |
123 | minimal mmap unit); typically 4096 or 8192 bytes. |
124 | |
125 | Maximum allocated size: 4-byte size_t: 2^32 minus about two pages |
126 | 8-byte size_t: 2^64 minus about two pages |
127 | |
128 | It is assumed that (possibly signed) size_t values suffice to |
129 | represent chunk sizes. `Possibly signed' is due to the fact |
130 | that `size_t' may be defined on a system as either a signed or |
131 | an unsigned type. The ISO C standard says that it must be |
132 | unsigned, but a few systems are known not to adhere to this. |
133 | Additionally, even when size_t is unsigned, sbrk (which is by |
134 | default used to obtain memory from system) accepts signed |
135 | arguments, and may not be able to handle size_t-wide arguments |
136 | with negative sign bit. Generally, values that would |
137 | appear as negative after accounting for overhead and alignment |
138 | are supported only via mmap(), which does not have this |
139 | limitation. |
140 | |
141 | Requests for sizes outside the allowed range will perform an optional |
142 | failure action and then return null. (Requests may also |
143 | also fail because a system is out of memory.) |
144 | |
145 | Thread-safety: thread-safe |
146 | |
147 | Compliance: I believe it is compliant with the 1997 Single Unix Specification |
148 | Also SVID/XPG, ANSI C, and probably others as well. |
149 | |
150 | * Synopsis of compile-time options: |
151 | |
152 | People have reported using previous versions of this malloc on all |
153 | versions of Unix, sometimes by tweaking some of the defines |
154 | below. It has been tested most extensively on Solaris and Linux. |
155 | People also report using it in stand-alone embedded systems. |
156 | |
157 | The implementation is in straight, hand-tuned ANSI C. It is not |
158 | at all modular. (Sorry!) It uses a lot of macros. To be at all |
159 | usable, this code should be compiled using an optimizing compiler |
160 | (for example gcc -O3) that can simplify expressions and control |
161 | paths. (FAQ: some macros import variables as arguments rather than |
162 | declare locals because people reported that some debuggers |
163 | otherwise get confused.) |
164 | |
165 | OPTION DEFAULT VALUE |
166 | |
167 | Compilation Environment options: |
168 | |
169 | HAVE_MREMAP 0 |
170 | |
171 | Changing default word sizes: |
172 | |
173 | INTERNAL_SIZE_T size_t |
174 | |
175 | Configuration and functionality options: |
176 | |
177 | USE_PUBLIC_MALLOC_WRAPPERS NOT defined |
178 | USE_MALLOC_LOCK NOT defined |
179 | MALLOC_DEBUG NOT defined |
180 | REALLOC_ZERO_BYTES_FREES 1 |
181 | TRIM_FASTBINS 0 |
182 | |
183 | Options for customizing MORECORE: |
184 | |
185 | MORECORE sbrk |
186 | MORECORE_FAILURE -1 |
187 | MORECORE_CONTIGUOUS 1 |
188 | MORECORE_CANNOT_TRIM NOT defined |
189 | MORECORE_CLEARS 1 |
190 | MMAP_AS_MORECORE_SIZE (1024 * 1024) |
191 | |
192 | Tuning options that are also dynamically changeable via mallopt: |
193 | |
194 | DEFAULT_MXFAST 64 (for 32bit), 128 (for 64bit) |
195 | DEFAULT_TRIM_THRESHOLD 128 * 1024 |
196 | DEFAULT_TOP_PAD 0 |
197 | DEFAULT_MMAP_THRESHOLD 128 * 1024 |
198 | DEFAULT_MMAP_MAX 65536 |
199 | |
200 | There are several other #defined constants and macros that you |
201 | probably don't want to touch unless you are extending or adapting malloc. */ |
202 | |
203 | /* |
204 | void* is the pointer type that malloc should say it returns |
205 | */ |
206 | |
207 | #ifndef void |
208 | #define void void |
209 | #endif /*void*/ |
210 | |
211 | #include <stddef.h> /* for size_t */ |
212 | #include <stdlib.h> /* for getenv(), abort() */ |
213 | #include <unistd.h> /* for __libc_enable_secure */ |
214 | |
215 | #include <atomic.h> |
216 | #include <_itoa.h> |
217 | #include <bits/wordsize.h> |
218 | #include <sys/sysinfo.h> |
219 | |
220 | #include <ldsodefs.h> |
221 | |
222 | #include <unistd.h> |
223 | #include <stdio.h> /* needed for malloc_stats */ |
224 | #include <errno.h> |
225 | #include <assert.h> |
226 | |
227 | #include <shlib-compat.h> |
228 | |
229 | /* For uintptr_t. */ |
230 | #include <stdint.h> |
231 | |
232 | /* For va_arg, va_start, va_end. */ |
233 | #include <stdarg.h> |
234 | |
235 | /* For MIN, MAX, powerof2. */ |
236 | #include <sys/param.h> |
237 | |
238 | /* For ALIGN_UP et. al. */ |
239 | #include <libc-pointer-arith.h> |
240 | |
241 | /* For DIAG_PUSH/POP_NEEDS_COMMENT et al. */ |
242 | #include <libc-diag.h> |
243 | |
244 | /* For memory tagging. */ |
245 | #include <libc-mtag.h> |
246 | |
247 | #include <malloc/malloc-internal.h> |
248 | |
249 | /* For SINGLE_THREAD_P. */ |
250 | #include <sysdep-cancel.h> |
251 | |
252 | #include <libc-internal.h> |
253 | |
254 | /* For tcache double-free check. */ |
255 | #include <random-bits.h> |
256 | #include <sys/random.h> |
257 | #include <not-cancel.h> |
258 | |
259 | /* |
260 | Debugging: |
261 | |
262 | Because freed chunks may be overwritten with bookkeeping fields, this |
263 | malloc will often die when freed memory is overwritten by user |
264 | programs. This can be very effective (albeit in an annoying way) |
265 | in helping track down dangling pointers. |
266 | |
267 | If you compile with -DMALLOC_DEBUG, a number of assertion checks are |
268 | enabled that will catch more memory errors. You probably won't be |
269 | able to make much sense of the actual assertion errors, but they |
270 | should help you locate incorrectly overwritten memory. The checking |
271 | is fairly extensive, and will slow down execution |
272 | noticeably. Calling malloc_stats or mallinfo with MALLOC_DEBUG set |
273 | will attempt to check every non-mmapped allocated and free chunk in |
274 | the course of computing the summaries. (By nature, mmapped regions |
275 | cannot be checked very much automatically.) |
276 | |
277 | Setting MALLOC_DEBUG may also be helpful if you are trying to modify |
278 | this code. The assertions in the check routines spell out in more |
279 | detail the assumptions and invariants underlying the algorithms. |
280 | |
281 | Setting MALLOC_DEBUG does NOT provide an automated mechanism for |
282 | checking that all accesses to malloced memory stay within their |
283 | bounds. However, there are several add-ons and adaptations of this |
284 | or other mallocs available that do this. |
285 | */ |
286 | |
287 | #ifndef MALLOC_DEBUG |
288 | #define MALLOC_DEBUG 0 |
289 | #endif |
290 | |
291 | #if USE_TCACHE |
292 | /* We want 64 entries. This is an arbitrary limit, which tunables can reduce. */ |
293 | # define TCACHE_MAX_BINS 64 |
294 | # define MAX_TCACHE_SIZE tidx2usize (TCACHE_MAX_BINS-1) |
295 | |
296 | /* Only used to pre-fill the tunables. */ |
297 | # define tidx2usize(idx) (((size_t) idx) * MALLOC_ALIGNMENT + MINSIZE - SIZE_SZ) |
298 | |
299 | /* When "x" is from chunksize(). */ |
300 | # define csize2tidx(x) (((x) - MINSIZE + MALLOC_ALIGNMENT - 1) / MALLOC_ALIGNMENT) |
301 | /* When "x" is a user-provided size. */ |
302 | # define usize2tidx(x) csize2tidx (request2size (x)) |
303 | |
304 | /* With rounding and alignment, the bins are... |
305 | idx 0 bytes 0..24 (64-bit) or 0..12 (32-bit) |
306 | idx 1 bytes 25..40 or 13..20 |
307 | idx 2 bytes 41..56 or 21..28 |
308 | etc. */ |
309 | |
310 | /* This is another arbitrary limit, which tunables can change. Each |
311 | tcache bin will hold at most this number of chunks. */ |
312 | # define TCACHE_FILL_COUNT 7 |
313 | |
314 | /* Maximum chunks in tcache bins for tunables. This value must fit the range |
315 | of tcache->counts[] entries, else they may overflow. */ |
316 | # define MAX_TCACHE_COUNT UINT16_MAX |
317 | #endif |
318 | |
319 | /* Safe-Linking: |
320 | Use randomness from ASLR (mmap_base) to protect single-linked lists |
321 | of Fast-Bins and TCache. That is, mask the "next" pointers of the |
322 | lists' chunks, and also perform allocation alignment checks on them. |
323 | This mechanism reduces the risk of pointer hijacking, as was done with |
324 | Safe-Unlinking in the double-linked lists of Small-Bins. |
325 | It assumes a minimum page size of 4096 bytes (12 bits). Systems with |
326 | larger pages provide less entropy, although the pointer mangling |
327 | still works. */ |
328 | #define PROTECT_PTR(pos, ptr) \ |
329 | ((__typeof (ptr)) ((((size_t) pos) >> 12) ^ ((size_t) ptr))) |
330 | #define REVEAL_PTR(ptr) PROTECT_PTR (&ptr, ptr) |
331 | |
332 | /* |
333 | The REALLOC_ZERO_BYTES_FREES macro controls the behavior of realloc (p, 0) |
334 | when p is nonnull. If the macro is nonzero, the realloc call returns NULL; |
335 | otherwise, the call returns what malloc (0) would. In either case, |
336 | p is freed. Glibc uses a nonzero REALLOC_ZERO_BYTES_FREES, which |
337 | implements common historical practice. |
338 | |
339 | ISO C17 says the realloc call has implementation-defined behavior, |
340 | and it might not even free p. |
341 | */ |
342 | |
343 | #ifndef REALLOC_ZERO_BYTES_FREES |
344 | #define REALLOC_ZERO_BYTES_FREES 1 |
345 | #endif |
346 | |
347 | /* |
348 | TRIM_FASTBINS controls whether free() of a very small chunk can |
349 | immediately lead to trimming. Setting to true (1) can reduce memory |
350 | footprint, but will almost always slow down programs that use a lot |
351 | of small chunks. |
352 | |
353 | Define this only if you are willing to give up some speed to more |
354 | aggressively reduce system-level memory footprint when releasing |
355 | memory in programs that use many small chunks. You can get |
356 | essentially the same effect by setting MXFAST to 0, but this can |
357 | lead to even greater slowdowns in programs using many small chunks. |
358 | TRIM_FASTBINS is an in-between compile-time option, that disables |
359 | only those chunks bordering topmost memory from being placed in |
360 | fastbins. |
361 | */ |
362 | |
363 | #ifndef TRIM_FASTBINS |
364 | #define TRIM_FASTBINS 0 |
365 | #endif |
366 | |
367 | /* Definition for getting more memory from the OS. */ |
368 | #include "morecore.c" |
369 | |
370 | #define MORECORE (*__glibc_morecore) |
371 | #define MORECORE_FAILURE 0 |
372 | |
373 | /* Memory tagging. */ |
374 | |
375 | /* Some systems support the concept of tagging (sometimes known as |
376 | coloring) memory locations on a fine grained basis. Each memory |
377 | location is given a color (normally allocated randomly) and |
378 | pointers are also colored. When the pointer is dereferenced, the |
379 | pointer's color is checked against the memory's color and if they |
380 | differ the access is faulted (sometimes lazily). |
381 | |
382 | We use this in glibc by maintaining a single color for the malloc |
383 | data structures that are interleaved with the user data and then |
384 | assigning separate colors for each block allocation handed out. In |
385 | this way simple buffer overruns will be rapidly detected. When |
386 | memory is freed, the memory is recolored back to the glibc default |
387 | so that simple use-after-free errors can also be detected. |
388 | |
389 | If memory is reallocated the buffer is recolored even if the |
390 | address remains the same. This has a performance impact, but |
391 | guarantees that the old pointer cannot mistakenly be reused (code |
392 | that compares old against new will see a mismatch and will then |
393 | need to behave as though realloc moved the data to a new location). |
394 | |
395 | Internal API for memory tagging support. |
396 | |
397 | The aim is to keep the code for memory tagging support as close to |
398 | the normal APIs in glibc as possible, so that if tagging is not |
399 | enabled in the library, or is disabled at runtime then standard |
400 | operations can continue to be used. Support macros are used to do |
401 | this: |
402 | |
403 | void *tag_new_zero_region (void *ptr, size_t size) |
404 | |
405 | Allocates a new tag, colors the memory with that tag, zeros the |
406 | memory and returns a pointer that is correctly colored for that |
407 | location. The non-tagging version will simply call memset with 0. |
408 | |
409 | void *tag_region (void *ptr, size_t size) |
410 | |
411 | Color the region of memory pointed to by PTR and size SIZE with |
412 | the color of PTR. Returns the original pointer. |
413 | |
414 | void *tag_new_usable (void *ptr) |
415 | |
416 | Allocate a new random color and use it to color the user region of |
417 | a chunk; this may include data from the subsequent chunk's header |
418 | if tagging is sufficiently fine grained. Returns PTR suitably |
419 | recolored for accessing the memory there. |
420 | |
421 | void *tag_at (void *ptr) |
422 | |
423 | Read the current color of the memory at the address pointed to by |
424 | PTR (ignoring it's current color) and return PTR recolored to that |
425 | color. PTR must be valid address in all other respects. When |
426 | tagging is not enabled, it simply returns the original pointer. |
427 | */ |
428 | |
429 | #ifdef USE_MTAG |
430 | static bool mtag_enabled = false; |
431 | static int mtag_mmap_flags = 0; |
432 | #else |
433 | # define mtag_enabled false |
434 | # define mtag_mmap_flags 0 |
435 | #endif |
436 | |
437 | static __always_inline void * |
438 | tag_region (void *ptr, size_t size) |
439 | { |
440 | if (__glibc_unlikely (mtag_enabled)) |
441 | return __libc_mtag_tag_region (ptr, size); |
442 | return ptr; |
443 | } |
444 | |
445 | static __always_inline void * |
446 | tag_new_zero_region (void *ptr, size_t size) |
447 | { |
448 | if (__glibc_unlikely (mtag_enabled)) |
449 | return __libc_mtag_tag_zero_region (__libc_mtag_new_tag (ptr), size); |
450 | return memset (ptr, 0, size); |
451 | } |
452 | |
453 | /* Defined later. */ |
454 | static void * |
455 | tag_new_usable (void *ptr); |
456 | |
457 | static __always_inline void * |
458 | tag_at (void *ptr) |
459 | { |
460 | if (__glibc_unlikely (mtag_enabled)) |
461 | return __libc_mtag_address_get_tag (ptr); |
462 | return ptr; |
463 | } |
464 | |
465 | #include <string.h> |
466 | |
467 | /* |
468 | MORECORE-related declarations. By default, rely on sbrk |
469 | */ |
470 | |
471 | |
472 | /* |
473 | MORECORE is the name of the routine to call to obtain more memory |
474 | from the system. See below for general guidance on writing |
475 | alternative MORECORE functions, as well as a version for WIN32 and a |
476 | sample version for pre-OSX macos. |
477 | */ |
478 | |
479 | #ifndef MORECORE |
480 | #define MORECORE sbrk |
481 | #endif |
482 | |
483 | /* |
484 | MORECORE_FAILURE is the value returned upon failure of MORECORE |
485 | as well as mmap. Since it cannot be an otherwise valid memory address, |
486 | and must reflect values of standard sys calls, you probably ought not |
487 | try to redefine it. |
488 | */ |
489 | |
490 | #ifndef MORECORE_FAILURE |
491 | #define MORECORE_FAILURE (-1) |
492 | #endif |
493 | |
494 | /* |
495 | If MORECORE_CONTIGUOUS is true, take advantage of fact that |
496 | consecutive calls to MORECORE with positive arguments always return |
497 | contiguous increasing addresses. This is true of unix sbrk. Even |
498 | if not defined, when regions happen to be contiguous, malloc will |
499 | permit allocations spanning regions obtained from different |
500 | calls. But defining this when applicable enables some stronger |
501 | consistency checks and space efficiencies. |
502 | */ |
503 | |
504 | #ifndef MORECORE_CONTIGUOUS |
505 | #define MORECORE_CONTIGUOUS 1 |
506 | #endif |
507 | |
508 | /* |
509 | Define MORECORE_CANNOT_TRIM if your version of MORECORE |
510 | cannot release space back to the system when given negative |
511 | arguments. This is generally necessary only if you are using |
512 | a hand-crafted MORECORE function that cannot handle negative arguments. |
513 | */ |
514 | |
515 | /* #define MORECORE_CANNOT_TRIM */ |
516 | |
517 | /* MORECORE_CLEARS (default 1) |
518 | The degree to which the routine mapped to MORECORE zeroes out |
519 | memory: never (0), only for newly allocated space (1) or always |
520 | (2). The distinction between (1) and (2) is necessary because on |
521 | some systems, if the application first decrements and then |
522 | increments the break value, the contents of the reallocated space |
523 | are unspecified. |
524 | */ |
525 | |
526 | #ifndef MORECORE_CLEARS |
527 | # define MORECORE_CLEARS 1 |
528 | #endif |
529 | |
530 | |
531 | /* |
532 | MMAP_AS_MORECORE_SIZE is the minimum mmap size argument to use if |
533 | sbrk fails, and mmap is used as a backup. The value must be a |
534 | multiple of page size. This backup strategy generally applies only |
535 | when systems have "holes" in address space, so sbrk cannot perform |
536 | contiguous expansion, but there is still space available on system. |
537 | On systems for which this is known to be useful (i.e. most linux |
538 | kernels), this occurs only when programs allocate huge amounts of |
539 | memory. Between this, and the fact that mmap regions tend to be |
540 | limited, the size should be large, to avoid too many mmap calls and |
541 | thus avoid running out of kernel resources. */ |
542 | |
543 | #ifndef MMAP_AS_MORECORE_SIZE |
544 | #define MMAP_AS_MORECORE_SIZE (1024 * 1024) |
545 | #endif |
546 | |
547 | /* |
548 | Define HAVE_MREMAP to make realloc() use mremap() to re-allocate |
549 | large blocks. |
550 | */ |
551 | |
552 | #ifndef HAVE_MREMAP |
553 | #define HAVE_MREMAP 0 |
554 | #endif |
555 | |
556 | /* |
557 | This version of malloc supports the standard SVID/XPG mallinfo |
558 | routine that returns a struct containing usage properties and |
559 | statistics. It should work on any SVID/XPG compliant system that has |
560 | a /usr/include/malloc.h defining struct mallinfo. (If you'd like to |
561 | install such a thing yourself, cut out the preliminary declarations |
562 | as described above and below and save them in a malloc.h file. But |
563 | there's no compelling reason to bother to do this.) |
564 | |
565 | The main declaration needed is the mallinfo struct that is returned |
566 | (by-copy) by mallinfo(). The SVID/XPG malloinfo struct contains a |
567 | bunch of fields that are not even meaningful in this version of |
568 | malloc. These fields are are instead filled by mallinfo() with |
569 | other numbers that might be of interest. |
570 | */ |
571 | |
572 | |
573 | /* ---------- description of public routines ------------ */ |
574 | |
575 | #if IS_IN (libc) |
576 | /* |
577 | malloc(size_t n) |
578 | Returns a pointer to a newly allocated chunk of at least n bytes, or null |
579 | if no space is available. Additionally, on failure, errno is |
580 | set to ENOMEM on ANSI C systems. |
581 | |
582 | If n is zero, malloc returns a minimum-sized chunk. (The minimum |
583 | size is 16 bytes on most 32bit systems, and 24 or 32 bytes on 64bit |
584 | systems.) On most systems, size_t is an unsigned type, so calls |
585 | with negative arguments are interpreted as requests for huge amounts |
586 | of space, which will often fail. The maximum supported value of n |
587 | differs across systems, but is in all cases less than the maximum |
588 | representable value of a size_t. |
589 | */ |
590 | void* __libc_malloc(size_t); |
591 | libc_hidden_proto (__libc_malloc) |
592 | |
593 | /* |
594 | free(void* p) |
595 | Releases the chunk of memory pointed to by p, that had been previously |
596 | allocated using malloc or a related routine such as realloc. |
597 | It has no effect if p is null. It can have arbitrary (i.e., bad!) |
598 | effects if p has already been freed. |
599 | |
600 | Unless disabled (using mallopt), freeing very large spaces will |
601 | when possible, automatically trigger operations that give |
602 | back unused memory to the system, thus reducing program footprint. |
603 | */ |
604 | void __libc_free(void*); |
605 | libc_hidden_proto (__libc_free) |
606 | |
607 | /* |
608 | calloc(size_t n_elements, size_t element_size); |
609 | Returns a pointer to n_elements * element_size bytes, with all locations |
610 | set to zero. |
611 | */ |
612 | void* __libc_calloc(size_t, size_t); |
613 | |
614 | /* |
615 | realloc(void* p, size_t n) |
616 | Returns a pointer to a chunk of size n that contains the same data |
617 | as does chunk p up to the minimum of (n, p's size) bytes, or null |
618 | if no space is available. |
619 | |
620 | The returned pointer may or may not be the same as p. The algorithm |
621 | prefers extending p when possible, otherwise it employs the |
622 | equivalent of a malloc-copy-free sequence. |
623 | |
624 | If p is null, realloc is equivalent to malloc. |
625 | |
626 | If space is not available, realloc returns null, errno is set (if on |
627 | ANSI) and p is NOT freed. |
628 | |
629 | if n is for fewer bytes than already held by p, the newly unused |
630 | space is lopped off and freed if possible. Unless the #define |
631 | REALLOC_ZERO_BYTES_FREES is set, realloc with a size argument of |
632 | zero (re)allocates a minimum-sized chunk. |
633 | |
634 | Large chunks that were internally obtained via mmap will always be |
635 | grown using malloc-copy-free sequences unless the system supports |
636 | MREMAP (currently only linux). |
637 | |
638 | The old unix realloc convention of allowing the last-free'd chunk |
639 | to be used as an argument to realloc is not supported. |
640 | */ |
641 | void* __libc_realloc(void*, size_t); |
642 | libc_hidden_proto (__libc_realloc) |
643 | |
644 | /* |
645 | memalign(size_t alignment, size_t n); |
646 | Returns a pointer to a newly allocated chunk of n bytes, aligned |
647 | in accord with the alignment argument. |
648 | |
649 | The alignment argument should be a power of two. If the argument is |
650 | not a power of two, the nearest greater power is used. |
651 | 8-byte alignment is guaranteed by normal malloc calls, so don't |
652 | bother calling memalign with an argument of 8 or less. |
653 | |
654 | Overreliance on memalign is a sure way to fragment space. |
655 | */ |
656 | void* __libc_memalign(size_t, size_t); |
657 | libc_hidden_proto (__libc_memalign) |
658 | |
659 | /* |
660 | valloc(size_t n); |
661 | Equivalent to memalign(pagesize, n), where pagesize is the page |
662 | size of the system. If the pagesize is unknown, 4096 is used. |
663 | */ |
664 | void* __libc_valloc(size_t); |
665 | |
666 | |
667 | |
668 | /* |
669 | mallinfo() |
670 | Returns (by copy) a struct containing various summary statistics: |
671 | |
672 | arena: current total non-mmapped bytes allocated from system |
673 | ordblks: the number of free chunks |
674 | smblks: the number of fastbin blocks (i.e., small chunks that |
675 | have been freed but not reused or consolidated) |
676 | hblks: current number of mmapped regions |
677 | hblkhd: total bytes held in mmapped regions |
678 | usmblks: always 0 |
679 | fsmblks: total bytes held in fastbin blocks |
680 | uordblks: current total allocated space (normal or mmapped) |
681 | fordblks: total free space |
682 | keepcost: the maximum number of bytes that could ideally be released |
683 | back to system via malloc_trim. ("ideally" means that |
684 | it ignores page restrictions etc.) |
685 | |
686 | Because these fields are ints, but internal bookkeeping may |
687 | be kept as longs, the reported values may wrap around zero and |
688 | thus be inaccurate. |
689 | */ |
690 | struct mallinfo2 __libc_mallinfo2(void); |
691 | libc_hidden_proto (__libc_mallinfo2) |
692 | |
693 | struct mallinfo __libc_mallinfo(void); |
694 | |
695 | |
696 | /* |
697 | pvalloc(size_t n); |
698 | Equivalent to valloc(minimum-page-that-holds(n)), that is, |
699 | round up n to nearest pagesize. |
700 | */ |
701 | void* __libc_pvalloc(size_t); |
702 | |
703 | /* |
704 | malloc_trim(size_t pad); |
705 | |
706 | If possible, gives memory back to the system (via negative |
707 | arguments to sbrk) if there is unused memory at the `high' end of |
708 | the malloc pool. You can call this after freeing large blocks of |
709 | memory to potentially reduce the system-level memory requirements |
710 | of a program. However, it cannot guarantee to reduce memory. Under |
711 | some allocation patterns, some large free blocks of memory will be |
712 | locked between two used chunks, so they cannot be given back to |
713 | the system. |
714 | |
715 | The `pad' argument to malloc_trim represents the amount of free |
716 | trailing space to leave untrimmed. If this argument is zero, |
717 | only the minimum amount of memory to maintain internal data |
718 | structures will be left (one page or less). Non-zero arguments |
719 | can be supplied to maintain enough trailing space to service |
720 | future expected allocations without having to re-obtain memory |
721 | from the system. |
722 | |
723 | Malloc_trim returns 1 if it actually released any memory, else 0. |
724 | On systems that do not support "negative sbrks", it will always |
725 | return 0. |
726 | */ |
727 | int __malloc_trim(size_t); |
728 | |
729 | /* |
730 | malloc_usable_size(void* p); |
731 | |
732 | Returns the number of bytes you can actually use in |
733 | an allocated chunk, which may be more than you requested (although |
734 | often not) due to alignment and minimum size constraints. |
735 | You can use this many bytes without worrying about |
736 | overwriting other allocated objects. This is not a particularly great |
737 | programming practice. malloc_usable_size can be more useful in |
738 | debugging and assertions, for example: |
739 | |
740 | p = malloc(n); |
741 | assert(malloc_usable_size(p) >= 256); |
742 | |
743 | */ |
744 | size_t __malloc_usable_size(void*); |
745 | |
746 | /* |
747 | malloc_stats(); |
748 | Prints on stderr the amount of space obtained from the system (both |
749 | via sbrk and mmap), the maximum amount (which may be more than |
750 | current if malloc_trim and/or munmap got called), and the current |
751 | number of bytes allocated via malloc (or realloc, etc) but not yet |
752 | freed. Note that this is the number of bytes allocated, not the |
753 | number requested. It will be larger than the number requested |
754 | because of alignment and bookkeeping overhead. Because it includes |
755 | alignment wastage as being in use, this figure may be greater than |
756 | zero even when no user-level chunks are allocated. |
757 | |
758 | The reported current and maximum system memory can be inaccurate if |
759 | a program makes other calls to system memory allocation functions |
760 | (normally sbrk) outside of malloc. |
761 | |
762 | malloc_stats prints only the most commonly interesting statistics. |
763 | More information can be obtained by calling mallinfo. |
764 | |
765 | */ |
766 | void __malloc_stats(void); |
767 | |
768 | /* |
769 | posix_memalign(void **memptr, size_t alignment, size_t size); |
770 | |
771 | POSIX wrapper like memalign(), checking for validity of size. |
772 | */ |
773 | int __posix_memalign(void **, size_t, size_t); |
774 | #endif /* IS_IN (libc) */ |
775 | |
776 | /* |
777 | mallopt(int parameter_number, int parameter_value) |
778 | Sets tunable parameters The format is to provide a |
779 | (parameter-number, parameter-value) pair. mallopt then sets the |
780 | corresponding parameter to the argument value if it can (i.e., so |
781 | long as the value is meaningful), and returns 1 if successful else |
782 | 0. SVID/XPG/ANSI defines four standard param numbers for mallopt, |
783 | normally defined in malloc.h. Only one of these (M_MXFAST) is used |
784 | in this malloc. The others (M_NLBLKS, M_GRAIN, M_KEEP) don't apply, |
785 | so setting them has no effect. But this malloc also supports four |
786 | other options in mallopt. See below for details. Briefly, supported |
787 | parameters are as follows (listed defaults are for "typical" |
788 | configurations). |
789 | |
790 | Symbol param # default allowed param values |
791 | M_MXFAST 1 64 0-80 (0 disables fastbins) |
792 | M_TRIM_THRESHOLD -1 128*1024 any (-1U disables trimming) |
793 | M_TOP_PAD -2 0 any |
794 | M_MMAP_THRESHOLD -3 128*1024 any (or 0 if no MMAP support) |
795 | M_MMAP_MAX -4 65536 any (0 disables use of mmap) |
796 | */ |
797 | int __libc_mallopt(int, int); |
798 | #if IS_IN (libc) |
799 | libc_hidden_proto (__libc_mallopt) |
800 | #endif |
801 | |
802 | /* mallopt tuning options */ |
803 | |
804 | /* |
805 | M_MXFAST is the maximum request size used for "fastbins", special bins |
806 | that hold returned chunks without consolidating their spaces. This |
807 | enables future requests for chunks of the same size to be handled |
808 | very quickly, but can increase fragmentation, and thus increase the |
809 | overall memory footprint of a program. |
810 | |
811 | This malloc manages fastbins very conservatively yet still |
812 | efficiently, so fragmentation is rarely a problem for values less |
813 | than or equal to the default. The maximum supported value of MXFAST |
814 | is 80. You wouldn't want it any higher than this anyway. Fastbins |
815 | are designed especially for use with many small structs, objects or |
816 | strings -- the default handles structs/objects/arrays with sizes up |
817 | to 8 4byte fields, or small strings representing words, tokens, |
818 | etc. Using fastbins for larger objects normally worsens |
819 | fragmentation without improving speed. |
820 | |
821 | M_MXFAST is set in REQUEST size units. It is internally used in |
822 | chunksize units, which adds padding and alignment. You can reduce |
823 | M_MXFAST to 0 to disable all use of fastbins. This causes the malloc |
824 | algorithm to be a closer approximation of fifo-best-fit in all cases, |
825 | not just for larger requests, but will generally cause it to be |
826 | slower. |
827 | */ |
828 | |
829 | |
830 | /* M_MXFAST is a standard SVID/XPG tuning option, usually listed in malloc.h */ |
831 | #ifndef M_MXFAST |
832 | #define M_MXFAST 1 |
833 | #endif |
834 | |
835 | #ifndef DEFAULT_MXFAST |
836 | #define DEFAULT_MXFAST (64 * SIZE_SZ / 4) |
837 | #endif |
838 | |
839 | |
840 | /* |
841 | M_TRIM_THRESHOLD is the maximum amount of unused top-most memory |
842 | to keep before releasing via malloc_trim in free(). |
843 | |
844 | Automatic trimming is mainly useful in long-lived programs. |
845 | Because trimming via sbrk can be slow on some systems, and can |
846 | sometimes be wasteful (in cases where programs immediately |
847 | afterward allocate more large chunks) the value should be high |
848 | enough so that your overall system performance would improve by |
849 | releasing this much memory. |
850 | |
851 | The trim threshold and the mmap control parameters (see below) |
852 | can be traded off with one another. Trimming and mmapping are |
853 | two different ways of releasing unused memory back to the |
854 | system. Between these two, it is often possible to keep |
855 | system-level demands of a long-lived program down to a bare |
856 | minimum. For example, in one test suite of sessions measuring |
857 | the XF86 X server on Linux, using a trim threshold of 128K and a |
858 | mmap threshold of 192K led to near-minimal long term resource |
859 | consumption. |
860 | |
861 | If you are using this malloc in a long-lived program, it should |
862 | pay to experiment with these values. As a rough guide, you |
863 | might set to a value close to the average size of a process |
864 | (program) running on your system. Releasing this much memory |
865 | would allow such a process to run in memory. Generally, it's |
866 | worth it to tune for trimming rather tham memory mapping when a |
867 | program undergoes phases where several large chunks are |
868 | allocated and released in ways that can reuse each other's |
869 | storage, perhaps mixed with phases where there are no such |
870 | chunks at all. And in well-behaved long-lived programs, |
871 | controlling release of large blocks via trimming versus mapping |
872 | is usually faster. |
873 | |
874 | However, in most programs, these parameters serve mainly as |
875 | protection against the system-level effects of carrying around |
876 | massive amounts of unneeded memory. Since frequent calls to |
877 | sbrk, mmap, and munmap otherwise degrade performance, the default |
878 | parameters are set to relatively high values that serve only as |
879 | safeguards. |
880 | |
881 | The trim value It must be greater than page size to have any useful |
882 | effect. To disable trimming completely, you can set to |
883 | (unsigned long)(-1) |
884 | |
885 | Trim settings interact with fastbin (MXFAST) settings: Unless |
886 | TRIM_FASTBINS is defined, automatic trimming never takes place upon |
887 | freeing a chunk with size less than or equal to MXFAST. Trimming is |
888 | instead delayed until subsequent freeing of larger chunks. However, |
889 | you can still force an attempted trim by calling malloc_trim. |
890 | |
891 | Also, trimming is not generally possible in cases where |
892 | the main arena is obtained via mmap. |
893 | |
894 | Note that the trick some people use of mallocing a huge space and |
895 | then freeing it at program startup, in an attempt to reserve system |
896 | memory, doesn't have the intended effect under automatic trimming, |
897 | since that memory will immediately be returned to the system. |
898 | */ |
899 | |
900 | #define M_TRIM_THRESHOLD -1 |
901 | |
902 | #ifndef DEFAULT_TRIM_THRESHOLD |
903 | #define DEFAULT_TRIM_THRESHOLD (128 * 1024) |
904 | #endif |
905 | |
906 | /* |
907 | M_TOP_PAD is the amount of extra `padding' space to allocate or |
908 | retain whenever sbrk is called. It is used in two ways internally: |
909 | |
910 | * When sbrk is called to extend the top of the arena to satisfy |
911 | a new malloc request, this much padding is added to the sbrk |
912 | request. |
913 | |
914 | * When malloc_trim is called automatically from free(), |
915 | it is used as the `pad' argument. |
916 | |
917 | In both cases, the actual amount of padding is rounded |
918 | so that the end of the arena is always a system page boundary. |
919 | |
920 | The main reason for using padding is to avoid calling sbrk so |
921 | often. Having even a small pad greatly reduces the likelihood |
922 | that nearly every malloc request during program start-up (or |
923 | after trimming) will invoke sbrk, which needlessly wastes |
924 | time. |
925 | |
926 | Automatic rounding-up to page-size units is normally sufficient |
927 | to avoid measurable overhead, so the default is 0. However, in |
928 | systems where sbrk is relatively slow, it can pay to increase |
929 | this value, at the expense of carrying around more memory than |
930 | the program needs. |
931 | */ |
932 | |
933 | #define M_TOP_PAD -2 |
934 | |
935 | #ifndef DEFAULT_TOP_PAD |
936 | #define DEFAULT_TOP_PAD (0) |
937 | #endif |
938 | |
939 | /* |
940 | MMAP_THRESHOLD_MAX and _MIN are the bounds on the dynamically |
941 | adjusted MMAP_THRESHOLD. |
942 | */ |
943 | |
944 | #ifndef DEFAULT_MMAP_THRESHOLD_MIN |
945 | #define DEFAULT_MMAP_THRESHOLD_MIN (128 * 1024) |
946 | #endif |
947 | |
948 | #ifndef DEFAULT_MMAP_THRESHOLD_MAX |
949 | /* For 32-bit platforms we cannot increase the maximum mmap |
950 | threshold much because it is also the minimum value for the |
951 | maximum heap size and its alignment. Going above 512k (i.e., 1M |
952 | for new heaps) wastes too much address space. */ |
953 | # if __WORDSIZE == 32 |
954 | # define DEFAULT_MMAP_THRESHOLD_MAX (512 * 1024) |
955 | # else |
956 | # define DEFAULT_MMAP_THRESHOLD_MAX (4 * 1024 * 1024 * sizeof(long)) |
957 | # endif |
958 | #endif |
959 | |
960 | /* |
961 | M_MMAP_THRESHOLD is the request size threshold for using mmap() |
962 | to service a request. Requests of at least this size that cannot |
963 | be allocated using already-existing space will be serviced via mmap. |
964 | (If enough normal freed space already exists it is used instead.) |
965 | |
966 | Using mmap segregates relatively large chunks of memory so that |
967 | they can be individually obtained and released from the host |
968 | system. A request serviced through mmap is never reused by any |
969 | other request (at least not directly; the system may just so |
970 | happen to remap successive requests to the same locations). |
971 | |
972 | Segregating space in this way has the benefits that: |
973 | |
974 | 1. Mmapped space can ALWAYS be individually released back |
975 | to the system, which helps keep the system level memory |
976 | demands of a long-lived program low. |
977 | 2. Mapped memory can never become `locked' between |
978 | other chunks, as can happen with normally allocated chunks, which |
979 | means that even trimming via malloc_trim would not release them. |
980 | 3. On some systems with "holes" in address spaces, mmap can obtain |
981 | memory that sbrk cannot. |
982 | |
983 | However, it has the disadvantages that: |
984 | |
985 | 1. The space cannot be reclaimed, consolidated, and then |
986 | used to service later requests, as happens with normal chunks. |
987 | 2. It can lead to more wastage because of mmap page alignment |
988 | requirements |
989 | 3. It causes malloc performance to be more dependent on host |
990 | system memory management support routines which may vary in |
991 | implementation quality and may impose arbitrary |
992 | limitations. Generally, servicing a request via normal |
993 | malloc steps is faster than going through a system's mmap. |
994 | |
995 | The advantages of mmap nearly always outweigh disadvantages for |
996 | "large" chunks, but the value of "large" varies across systems. The |
997 | default is an empirically derived value that works well in most |
998 | systems. |
999 | |
1000 | |
1001 | Update in 2006: |
1002 | The above was written in 2001. Since then the world has changed a lot. |
1003 | Memory got bigger. Applications got bigger. The virtual address space |
1004 | layout in 32 bit linux changed. |
1005 | |
1006 | In the new situation, brk() and mmap space is shared and there are no |
1007 | artificial limits on brk size imposed by the kernel. What is more, |
1008 | applications have started using transient allocations larger than the |
1009 | 128Kb as was imagined in 2001. |
1010 | |
1011 | The price for mmap is also high now; each time glibc mmaps from the |
1012 | kernel, the kernel is forced to zero out the memory it gives to the |
1013 | application. Zeroing memory is expensive and eats a lot of cache and |
1014 | memory bandwidth. This has nothing to do with the efficiency of the |
1015 | virtual memory system, by doing mmap the kernel just has no choice but |
1016 | to zero. |
1017 | |
1018 | In 2001, the kernel had a maximum size for brk() which was about 800 |
1019 | megabytes on 32 bit x86, at that point brk() would hit the first |
1020 | mmaped shared libraries and couldn't expand anymore. With current 2.6 |
1021 | kernels, the VA space layout is different and brk() and mmap |
1022 | both can span the entire heap at will. |
1023 | |
1024 | Rather than using a static threshold for the brk/mmap tradeoff, |
1025 | we are now using a simple dynamic one. The goal is still to avoid |
1026 | fragmentation. The old goals we kept are |
1027 | 1) try to get the long lived large allocations to use mmap() |
1028 | 2) really large allocations should always use mmap() |
1029 | and we're adding now: |
1030 | 3) transient allocations should use brk() to avoid forcing the kernel |
1031 | having to zero memory over and over again |
1032 | |
1033 | The implementation works with a sliding threshold, which is by default |
1034 | limited to go between 128Kb and 32Mb (64Mb for 64 bitmachines) and starts |
1035 | out at 128Kb as per the 2001 default. |
1036 | |
1037 | This allows us to satisfy requirement 1) under the assumption that long |
1038 | lived allocations are made early in the process' lifespan, before it has |
1039 | started doing dynamic allocations of the same size (which will |
1040 | increase the threshold). |
1041 | |
1042 | The upperbound on the threshold satisfies requirement 2) |
1043 | |
1044 | The threshold goes up in value when the application frees memory that was |
1045 | allocated with the mmap allocator. The idea is that once the application |
1046 | starts freeing memory of a certain size, it's highly probable that this is |
1047 | a size the application uses for transient allocations. This estimator |
1048 | is there to satisfy the new third requirement. |
1049 | |
1050 | */ |
1051 | |
1052 | #define M_MMAP_THRESHOLD -3 |
1053 | |
1054 | #ifndef DEFAULT_MMAP_THRESHOLD |
1055 | #define DEFAULT_MMAP_THRESHOLD DEFAULT_MMAP_THRESHOLD_MIN |
1056 | #endif |
1057 | |
1058 | /* |
1059 | M_MMAP_MAX is the maximum number of requests to simultaneously |
1060 | service using mmap. This parameter exists because |
1061 | some systems have a limited number of internal tables for |
1062 | use by mmap, and using more than a few of them may degrade |
1063 | performance. |
1064 | |
1065 | The default is set to a value that serves only as a safeguard. |
1066 | Setting to 0 disables use of mmap for servicing large requests. |
1067 | */ |
1068 | |
1069 | #define M_MMAP_MAX -4 |
1070 | |
1071 | #ifndef DEFAULT_MMAP_MAX |
1072 | #define DEFAULT_MMAP_MAX (65536) |
1073 | #endif |
1074 | |
1075 | #include <malloc.h> |
1076 | |
1077 | #ifndef RETURN_ADDRESS |
1078 | #define RETURN_ADDRESS(X_) (NULL) |
1079 | #endif |
1080 | |
1081 | /* Forward declarations. */ |
1082 | struct malloc_chunk; |
1083 | typedef struct malloc_chunk* mchunkptr; |
1084 | |
1085 | /* Internal routines. */ |
1086 | |
1087 | static void* _int_malloc(mstate, size_t); |
1088 | static void _int_free(mstate, mchunkptr, int); |
1089 | static void* _int_realloc(mstate, mchunkptr, INTERNAL_SIZE_T, |
1090 | INTERNAL_SIZE_T); |
1091 | static void* _int_memalign(mstate, size_t, size_t); |
1092 | #if IS_IN (libc) |
1093 | static void* _mid_memalign(size_t, size_t, void *); |
1094 | #endif |
1095 | |
1096 | static void malloc_printerr(const char *str) __attribute__ ((noreturn)); |
1097 | |
1098 | static void munmap_chunk(mchunkptr p); |
1099 | #if HAVE_MREMAP |
1100 | static mchunkptr mremap_chunk(mchunkptr p, size_t new_size); |
1101 | #endif |
1102 | |
1103 | static size_t musable (void *mem); |
1104 | |
1105 | /* ------------------ MMAP support ------------------ */ |
1106 | |
1107 | |
1108 | #include <fcntl.h> |
1109 | #include <sys/mman.h> |
1110 | |
1111 | #if !defined(MAP_ANONYMOUS) && defined(MAP_ANON) |
1112 | # define MAP_ANONYMOUS MAP_ANON |
1113 | #endif |
1114 | |
1115 | #define MMAP(addr, size, prot, flags) \ |
1116 | __mmap((addr), (size), (prot), (flags)|MAP_ANONYMOUS|MAP_PRIVATE, -1, 0) |
1117 | |
1118 | |
1119 | /* |
1120 | ----------------------- Chunk representations ----------------------- |
1121 | */ |
1122 | |
1123 | |
1124 | /* |
1125 | This struct declaration is misleading (but accurate and necessary). |
1126 | It declares a "view" into memory allowing access to necessary |
1127 | fields at known offsets from a given base. See explanation below. |
1128 | */ |
1129 | |
1130 | struct malloc_chunk { |
1131 | |
1132 | INTERNAL_SIZE_T mchunk_prev_size; /* Size of previous chunk (if free). */ |
1133 | INTERNAL_SIZE_T mchunk_size; /* Size in bytes, including overhead. */ |
1134 | |
1135 | struct malloc_chunk* fd; /* double links -- used only if free. */ |
1136 | struct malloc_chunk* bk; |
1137 | |
1138 | /* Only used for large blocks: pointer to next larger size. */ |
1139 | struct malloc_chunk* fd_nextsize; /* double links -- used only if free. */ |
1140 | struct malloc_chunk* bk_nextsize; |
1141 | }; |
1142 | |
1143 | |
1144 | /* |
1145 | malloc_chunk details: |
1146 | |
1147 | (The following includes lightly edited explanations by Colin Plumb.) |
1148 | |
1149 | Chunks of memory are maintained using a `boundary tag' method as |
1150 | described in e.g., Knuth or Standish. (See the paper by Paul |
1151 | Wilson ftp://ftp.cs.utexas.edu/pub/garbage/allocsrv.ps for a |
1152 | survey of such techniques.) Sizes of free chunks are stored both |
1153 | in the front of each chunk and at the end. This makes |
1154 | consolidating fragmented chunks into bigger chunks very fast. The |
1155 | size fields also hold bits representing whether chunks are free or |
1156 | in use. |
1157 | |
1158 | An allocated chunk looks like this: |
1159 | |
1160 | |
1161 | chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
1162 | | Size of previous chunk, if unallocated (P clear) | |
1163 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
1164 | | Size of chunk, in bytes |A|M|P| |
1165 | mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
1166 | | User data starts here... . |
1167 | . . |
1168 | . (malloc_usable_size() bytes) . |
1169 | . | |
1170 | nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
1171 | | (size of chunk, but used for application data) | |
1172 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
1173 | | Size of next chunk, in bytes |A|0|1| |
1174 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
1175 | |
1176 | Where "chunk" is the front of the chunk for the purpose of most of |
1177 | the malloc code, but "mem" is the pointer that is returned to the |
1178 | user. "Nextchunk" is the beginning of the next contiguous chunk. |
1179 | |
1180 | Chunks always begin on even word boundaries, so the mem portion |
1181 | (which is returned to the user) is also on an even word boundary, and |
1182 | thus at least double-word aligned. |
1183 | |
1184 | Free chunks are stored in circular doubly-linked lists, and look like this: |
1185 | |
1186 | chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
1187 | | Size of previous chunk, if unallocated (P clear) | |
1188 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
1189 | `head:' | Size of chunk, in bytes |A|0|P| |
1190 | mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
1191 | | Forward pointer to next chunk in list | |
1192 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
1193 | | Back pointer to previous chunk in list | |
1194 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
1195 | | Unused space (may be 0 bytes long) . |
1196 | . . |
1197 | . | |
1198 | nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
1199 | `foot:' | Size of chunk, in bytes | |
1200 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
1201 | | Size of next chunk, in bytes |A|0|0| |
1202 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
1203 | |
1204 | The P (PREV_INUSE) bit, stored in the unused low-order bit of the |
1205 | chunk size (which is always a multiple of two words), is an in-use |
1206 | bit for the *previous* chunk. If that bit is *clear*, then the |
1207 | word before the current chunk size contains the previous chunk |
1208 | size, and can be used to find the front of the previous chunk. |
1209 | The very first chunk allocated always has this bit set, |
1210 | preventing access to non-existent (or non-owned) memory. If |
1211 | prev_inuse is set for any given chunk, then you CANNOT determine |
1212 | the size of the previous chunk, and might even get a memory |
1213 | addressing fault when trying to do so. |
1214 | |
1215 | The A (NON_MAIN_ARENA) bit is cleared for chunks on the initial, |
1216 | main arena, described by the main_arena variable. When additional |
1217 | threads are spawned, each thread receives its own arena (up to a |
1218 | configurable limit, after which arenas are reused for multiple |
1219 | threads), and the chunks in these arenas have the A bit set. To |
1220 | find the arena for a chunk on such a non-main arena, heap_for_ptr |
1221 | performs a bit mask operation and indirection through the ar_ptr |
1222 | member of the per-heap header heap_info (see arena.c). |
1223 | |
1224 | Note that the `foot' of the current chunk is actually represented |
1225 | as the prev_size of the NEXT chunk. This makes it easier to |
1226 | deal with alignments etc but can be very confusing when trying |
1227 | to extend or adapt this code. |
1228 | |
1229 | The three exceptions to all this are: |
1230 | |
1231 | 1. The special chunk `top' doesn't bother using the |
1232 | trailing size field since there is no next contiguous chunk |
1233 | that would have to index off it. After initialization, `top' |
1234 | is forced to always exist. If it would become less than |
1235 | MINSIZE bytes long, it is replenished. |
1236 | |
1237 | 2. Chunks allocated via mmap, which have the second-lowest-order |
1238 | bit M (IS_MMAPPED) set in their size fields. Because they are |
1239 | allocated one-by-one, each must contain its own trailing size |
1240 | field. If the M bit is set, the other bits are ignored |
1241 | (because mmapped chunks are neither in an arena, nor adjacent |
1242 | to a freed chunk). The M bit is also used for chunks which |
1243 | originally came from a dumped heap via malloc_set_state in |
1244 | hooks.c. |
1245 | |
1246 | 3. Chunks in fastbins are treated as allocated chunks from the |
1247 | point of view of the chunk allocator. They are consolidated |
1248 | with their neighbors only in bulk, in malloc_consolidate. |
1249 | */ |
1250 | |
1251 | /* |
1252 | ---------- Size and alignment checks and conversions ---------- |
1253 | */ |
1254 | |
1255 | /* Conversion from malloc headers to user pointers, and back. When |
1256 | using memory tagging the user data and the malloc data structure |
1257 | headers have distinct tags. Converting fully from one to the other |
1258 | involves extracting the tag at the other address and creating a |
1259 | suitable pointer using it. That can be quite expensive. There are |
1260 | cases when the pointers are not dereferenced (for example only used |
1261 | for alignment check) so the tags are not relevant, and there are |
1262 | cases when user data is not tagged distinctly from malloc headers |
1263 | (user data is untagged because tagging is done late in malloc and |
1264 | early in free). User memory tagging across internal interfaces: |
1265 | |
1266 | sysmalloc: Returns untagged memory. |
1267 | _int_malloc: Returns untagged memory. |
1268 | _int_free: Takes untagged memory. |
1269 | _int_memalign: Returns untagged memory. |
1270 | _int_memalign: Returns untagged memory. |
1271 | _mid_memalign: Returns tagged memory. |
1272 | _int_realloc: Takes and returns tagged memory. |
1273 | */ |
1274 | |
1275 | /* The chunk header is two SIZE_SZ elements, but this is used widely, so |
1276 | we define it here for clarity later. */ |
1277 | #define CHUNK_HDR_SZ (2 * SIZE_SZ) |
1278 | |
1279 | /* Convert a chunk address to a user mem pointer without correcting |
1280 | the tag. */ |
1281 | #define chunk2mem(p) ((void*)((char*)(p) + CHUNK_HDR_SZ)) |
1282 | |
1283 | /* Convert a chunk address to a user mem pointer and extract the right tag. */ |
1284 | #define chunk2mem_tag(p) ((void*)tag_at ((char*)(p) + CHUNK_HDR_SZ)) |
1285 | |
1286 | /* Convert a user mem pointer to a chunk address and extract the right tag. */ |
1287 | #define mem2chunk(mem) ((mchunkptr)tag_at (((char*)(mem) - CHUNK_HDR_SZ))) |
1288 | |
1289 | /* The smallest possible chunk */ |
1290 | #define MIN_CHUNK_SIZE (offsetof(struct malloc_chunk, fd_nextsize)) |
1291 | |
1292 | /* The smallest size we can malloc is an aligned minimal chunk */ |
1293 | |
1294 | #define MINSIZE \ |
1295 | (unsigned long)(((MIN_CHUNK_SIZE+MALLOC_ALIGN_MASK) & ~MALLOC_ALIGN_MASK)) |
1296 | |
1297 | /* Check if m has acceptable alignment */ |
1298 | |
1299 | #define aligned_OK(m) (((unsigned long)(m) & MALLOC_ALIGN_MASK) == 0) |
1300 | |
1301 | #define misaligned_chunk(p) \ |
1302 | ((uintptr_t)(MALLOC_ALIGNMENT == CHUNK_HDR_SZ ? (p) : chunk2mem (p)) \ |
1303 | & MALLOC_ALIGN_MASK) |
1304 | |
1305 | /* pad request bytes into a usable size -- internal version */ |
1306 | /* Note: This must be a macro that evaluates to a compile time constant |
1307 | if passed a literal constant. */ |
1308 | #define request2size(req) \ |
1309 | (((req) + SIZE_SZ + MALLOC_ALIGN_MASK < MINSIZE) ? \ |
1310 | MINSIZE : \ |
1311 | ((req) + SIZE_SZ + MALLOC_ALIGN_MASK) & ~MALLOC_ALIGN_MASK) |
1312 | |
1313 | /* Check if REQ overflows when padded and aligned and if the resulting |
1314 | value is less than PTRDIFF_T. Returns the requested size or |
1315 | MINSIZE in case the value is less than MINSIZE, or 0 if any of the |
1316 | previous checks fail. */ |
1317 | static inline size_t |
1318 | checked_request2size (size_t req) __nonnull (1) |
1319 | { |
1320 | if (__glibc_unlikely (req > PTRDIFF_MAX)) |
1321 | return 0; |
1322 | |
1323 | /* When using tagged memory, we cannot share the end of the user |
1324 | block with the header for the next chunk, so ensure that we |
1325 | allocate blocks that are rounded up to the granule size. Take |
1326 | care not to overflow from close to MAX_SIZE_T to a small |
1327 | number. Ideally, this would be part of request2size(), but that |
1328 | must be a macro that produces a compile time constant if passed |
1329 | a constant literal. */ |
1330 | if (__glibc_unlikely (mtag_enabled)) |
1331 | { |
1332 | /* Ensure this is not evaluated if !mtag_enabled, see gcc PR 99551. */ |
1333 | asm ("" ); |
1334 | |
1335 | req = (req + (__MTAG_GRANULE_SIZE - 1)) & |
1336 | ~(size_t)(__MTAG_GRANULE_SIZE - 1); |
1337 | } |
1338 | |
1339 | return request2size (req); |
1340 | } |
1341 | |
1342 | /* |
1343 | --------------- Physical chunk operations --------------- |
1344 | */ |
1345 | |
1346 | |
1347 | /* size field is or'ed with PREV_INUSE when previous adjacent chunk in use */ |
1348 | #define PREV_INUSE 0x1 |
1349 | |
1350 | /* extract inuse bit of previous chunk */ |
1351 | #define prev_inuse(p) ((p)->mchunk_size & PREV_INUSE) |
1352 | |
1353 | |
1354 | /* size field is or'ed with IS_MMAPPED if the chunk was obtained with mmap() */ |
1355 | #define IS_MMAPPED 0x2 |
1356 | |
1357 | /* check for mmap()'ed chunk */ |
1358 | #define chunk_is_mmapped(p) ((p)->mchunk_size & IS_MMAPPED) |
1359 | |
1360 | |
1361 | /* size field is or'ed with NON_MAIN_ARENA if the chunk was obtained |
1362 | from a non-main arena. This is only set immediately before handing |
1363 | the chunk to the user, if necessary. */ |
1364 | #define NON_MAIN_ARENA 0x4 |
1365 | |
1366 | /* Check for chunk from main arena. */ |
1367 | #define chunk_main_arena(p) (((p)->mchunk_size & NON_MAIN_ARENA) == 0) |
1368 | |
1369 | /* Mark a chunk as not being on the main arena. */ |
1370 | #define set_non_main_arena(p) ((p)->mchunk_size |= NON_MAIN_ARENA) |
1371 | |
1372 | |
1373 | /* |
1374 | Bits to mask off when extracting size |
1375 | |
1376 | Note: IS_MMAPPED is intentionally not masked off from size field in |
1377 | macros for which mmapped chunks should never be seen. This should |
1378 | cause helpful core dumps to occur if it is tried by accident by |
1379 | people extending or adapting this malloc. |
1380 | */ |
1381 | #define SIZE_BITS (PREV_INUSE | IS_MMAPPED | NON_MAIN_ARENA) |
1382 | |
1383 | /* Get size, ignoring use bits */ |
1384 | #define chunksize(p) (chunksize_nomask (p) & ~(SIZE_BITS)) |
1385 | |
1386 | /* Like chunksize, but do not mask SIZE_BITS. */ |
1387 | #define chunksize_nomask(p) ((p)->mchunk_size) |
1388 | |
1389 | /* Ptr to next physical malloc_chunk. */ |
1390 | #define next_chunk(p) ((mchunkptr) (((char *) (p)) + chunksize (p))) |
1391 | |
1392 | /* Size of the chunk below P. Only valid if !prev_inuse (P). */ |
1393 | #define prev_size(p) ((p)->mchunk_prev_size) |
1394 | |
1395 | /* Set the size of the chunk below P. Only valid if !prev_inuse (P). */ |
1396 | #define set_prev_size(p, sz) ((p)->mchunk_prev_size = (sz)) |
1397 | |
1398 | /* Ptr to previous physical malloc_chunk. Only valid if !prev_inuse (P). */ |
1399 | #define prev_chunk(p) ((mchunkptr) (((char *) (p)) - prev_size (p))) |
1400 | |
1401 | /* Treat space at ptr + offset as a chunk */ |
1402 | #define chunk_at_offset(p, s) ((mchunkptr) (((char *) (p)) + (s))) |
1403 | |
1404 | /* extract p's inuse bit */ |
1405 | #define inuse(p) \ |
1406 | ((((mchunkptr) (((char *) (p)) + chunksize (p)))->mchunk_size) & PREV_INUSE) |
1407 | |
1408 | /* set/clear chunk as being inuse without otherwise disturbing */ |
1409 | #define set_inuse(p) \ |
1410 | ((mchunkptr) (((char *) (p)) + chunksize (p)))->mchunk_size |= PREV_INUSE |
1411 | |
1412 | #define clear_inuse(p) \ |
1413 | ((mchunkptr) (((char *) (p)) + chunksize (p)))->mchunk_size &= ~(PREV_INUSE) |
1414 | |
1415 | |
1416 | /* check/set/clear inuse bits in known places */ |
1417 | #define inuse_bit_at_offset(p, s) \ |
1418 | (((mchunkptr) (((char *) (p)) + (s)))->mchunk_size & PREV_INUSE) |
1419 | |
1420 | #define set_inuse_bit_at_offset(p, s) \ |
1421 | (((mchunkptr) (((char *) (p)) + (s)))->mchunk_size |= PREV_INUSE) |
1422 | |
1423 | #define clear_inuse_bit_at_offset(p, s) \ |
1424 | (((mchunkptr) (((char *) (p)) + (s)))->mchunk_size &= ~(PREV_INUSE)) |
1425 | |
1426 | |
1427 | /* Set size at head, without disturbing its use bit */ |
1428 | #define set_head_size(p, s) ((p)->mchunk_size = (((p)->mchunk_size & SIZE_BITS) | (s))) |
1429 | |
1430 | /* Set size/use field */ |
1431 | #define set_head(p, s) ((p)->mchunk_size = (s)) |
1432 | |
1433 | /* Set size at footer (only when chunk is not in use) */ |
1434 | #define (p, s) (((mchunkptr) ((char *) (p) + (s)))->mchunk_prev_size = (s)) |
1435 | |
1436 | #pragma GCC poison mchunk_size |
1437 | #pragma GCC poison mchunk_prev_size |
1438 | |
1439 | /* This is the size of the real usable data in the chunk. Not valid for |
1440 | dumped heap chunks. */ |
1441 | #define memsize(p) \ |
1442 | (__MTAG_GRANULE_SIZE > SIZE_SZ && __glibc_unlikely (mtag_enabled) ? \ |
1443 | chunksize (p) - CHUNK_HDR_SZ : \ |
1444 | chunksize (p) - CHUNK_HDR_SZ + (chunk_is_mmapped (p) ? 0 : SIZE_SZ)) |
1445 | |
1446 | /* If memory tagging is enabled the layout changes to accommodate the granule |
1447 | size, this is wasteful for small allocations so not done by default. |
1448 | Both the chunk header and user data has to be granule aligned. */ |
1449 | _Static_assert (__MTAG_GRANULE_SIZE <= CHUNK_HDR_SZ, |
1450 | "memory tagging is not supported with large granule." ); |
1451 | |
1452 | static __always_inline void * |
1453 | tag_new_usable (void *ptr) |
1454 | { |
1455 | if (__glibc_unlikely (mtag_enabled) && ptr) |
1456 | { |
1457 | mchunkptr cp = mem2chunk(ptr); |
1458 | ptr = __libc_mtag_tag_region (__libc_mtag_new_tag (ptr), memsize (cp)); |
1459 | } |
1460 | return ptr; |
1461 | } |
1462 | |
1463 | /* |
1464 | -------------------- Internal data structures -------------------- |
1465 | |
1466 | All internal state is held in an instance of malloc_state defined |
1467 | below. There are no other static variables, except in two optional |
1468 | cases: |
1469 | * If USE_MALLOC_LOCK is defined, the mALLOC_MUTEx declared above. |
1470 | * If mmap doesn't support MAP_ANONYMOUS, a dummy file descriptor |
1471 | for mmap. |
1472 | |
1473 | Beware of lots of tricks that minimize the total bookkeeping space |
1474 | requirements. The result is a little over 1K bytes (for 4byte |
1475 | pointers and size_t.) |
1476 | */ |
1477 | |
1478 | /* |
1479 | Bins |
1480 | |
1481 | An array of bin headers for free chunks. Each bin is doubly |
1482 | linked. The bins are approximately proportionally (log) spaced. |
1483 | There are a lot of these bins (128). This may look excessive, but |
1484 | works very well in practice. Most bins hold sizes that are |
1485 | unusual as malloc request sizes, but are more usual for fragments |
1486 | and consolidated sets of chunks, which is what these bins hold, so |
1487 | they can be found quickly. All procedures maintain the invariant |
1488 | that no consolidated chunk physically borders another one, so each |
1489 | chunk in a list is known to be preceded and followed by either |
1490 | inuse chunks or the ends of memory. |
1491 | |
1492 | Chunks in bins are kept in size order, with ties going to the |
1493 | approximately least recently used chunk. Ordering isn't needed |
1494 | for the small bins, which all contain the same-sized chunks, but |
1495 | facilitates best-fit allocation for larger chunks. These lists |
1496 | are just sequential. Keeping them in order almost never requires |
1497 | enough traversal to warrant using fancier ordered data |
1498 | structures. |
1499 | |
1500 | Chunks of the same size are linked with the most |
1501 | recently freed at the front, and allocations are taken from the |
1502 | back. This results in LRU (FIFO) allocation order, which tends |
1503 | to give each chunk an equal opportunity to be consolidated with |
1504 | adjacent freed chunks, resulting in larger free chunks and less |
1505 | fragmentation. |
1506 | |
1507 | To simplify use in double-linked lists, each bin header acts |
1508 | as a malloc_chunk. This avoids special-casing for headers. |
1509 | But to conserve space and improve locality, we allocate |
1510 | only the fd/bk pointers of bins, and then use repositioning tricks |
1511 | to treat these as the fields of a malloc_chunk*. |
1512 | */ |
1513 | |
1514 | typedef struct malloc_chunk *mbinptr; |
1515 | |
1516 | /* addressing -- note that bin_at(0) does not exist */ |
1517 | #define bin_at(m, i) \ |
1518 | (mbinptr) (((char *) &((m)->bins[((i) - 1) * 2])) \ |
1519 | - offsetof (struct malloc_chunk, fd)) |
1520 | |
1521 | /* analog of ++bin */ |
1522 | #define next_bin(b) ((mbinptr) ((char *) (b) + (sizeof (mchunkptr) << 1))) |
1523 | |
1524 | /* Reminders about list directionality within bins */ |
1525 | #define first(b) ((b)->fd) |
1526 | #define last(b) ((b)->bk) |
1527 | |
1528 | /* |
1529 | Indexing |
1530 | |
1531 | Bins for sizes < 512 bytes contain chunks of all the same size, spaced |
1532 | 8 bytes apart. Larger bins are approximately logarithmically spaced: |
1533 | |
1534 | 64 bins of size 8 |
1535 | 32 bins of size 64 |
1536 | 16 bins of size 512 |
1537 | 8 bins of size 4096 |
1538 | 4 bins of size 32768 |
1539 | 2 bins of size 262144 |
1540 | 1 bin of size what's left |
1541 | |
1542 | There is actually a little bit of slop in the numbers in bin_index |
1543 | for the sake of speed. This makes no difference elsewhere. |
1544 | |
1545 | The bins top out around 1MB because we expect to service large |
1546 | requests via mmap. |
1547 | |
1548 | Bin 0 does not exist. Bin 1 is the unordered list; if that would be |
1549 | a valid chunk size the small bins are bumped up one. |
1550 | */ |
1551 | |
1552 | #define NBINS 128 |
1553 | #define NSMALLBINS 64 |
1554 | #define SMALLBIN_WIDTH MALLOC_ALIGNMENT |
1555 | #define SMALLBIN_CORRECTION (MALLOC_ALIGNMENT > CHUNK_HDR_SZ) |
1556 | #define MIN_LARGE_SIZE ((NSMALLBINS - SMALLBIN_CORRECTION) * SMALLBIN_WIDTH) |
1557 | |
1558 | #define in_smallbin_range(sz) \ |
1559 | ((unsigned long) (sz) < (unsigned long) MIN_LARGE_SIZE) |
1560 | |
1561 | #define smallbin_index(sz) \ |
1562 | ((SMALLBIN_WIDTH == 16 ? (((unsigned) (sz)) >> 4) : (((unsigned) (sz)) >> 3))\ |
1563 | + SMALLBIN_CORRECTION) |
1564 | |
1565 | #define largebin_index_32(sz) \ |
1566 | (((((unsigned long) (sz)) >> 6) <= 38) ? 56 + (((unsigned long) (sz)) >> 6) :\ |
1567 | ((((unsigned long) (sz)) >> 9) <= 20) ? 91 + (((unsigned long) (sz)) >> 9) :\ |
1568 | ((((unsigned long) (sz)) >> 12) <= 10) ? 110 + (((unsigned long) (sz)) >> 12) :\ |
1569 | ((((unsigned long) (sz)) >> 15) <= 4) ? 119 + (((unsigned long) (sz)) >> 15) :\ |
1570 | ((((unsigned long) (sz)) >> 18) <= 2) ? 124 + (((unsigned long) (sz)) >> 18) :\ |
1571 | 126) |
1572 | |
1573 | #define largebin_index_32_big(sz) \ |
1574 | (((((unsigned long) (sz)) >> 6) <= 45) ? 49 + (((unsigned long) (sz)) >> 6) :\ |
1575 | ((((unsigned long) (sz)) >> 9) <= 20) ? 91 + (((unsigned long) (sz)) >> 9) :\ |
1576 | ((((unsigned long) (sz)) >> 12) <= 10) ? 110 + (((unsigned long) (sz)) >> 12) :\ |
1577 | ((((unsigned long) (sz)) >> 15) <= 4) ? 119 + (((unsigned long) (sz)) >> 15) :\ |
1578 | ((((unsigned long) (sz)) >> 18) <= 2) ? 124 + (((unsigned long) (sz)) >> 18) :\ |
1579 | 126) |
1580 | |
1581 | // XXX It remains to be seen whether it is good to keep the widths of |
1582 | // XXX the buckets the same or whether it should be scaled by a factor |
1583 | // XXX of two as well. |
1584 | #define largebin_index_64(sz) \ |
1585 | (((((unsigned long) (sz)) >> 6) <= 48) ? 48 + (((unsigned long) (sz)) >> 6) :\ |
1586 | ((((unsigned long) (sz)) >> 9) <= 20) ? 91 + (((unsigned long) (sz)) >> 9) :\ |
1587 | ((((unsigned long) (sz)) >> 12) <= 10) ? 110 + (((unsigned long) (sz)) >> 12) :\ |
1588 | ((((unsigned long) (sz)) >> 15) <= 4) ? 119 + (((unsigned long) (sz)) >> 15) :\ |
1589 | ((((unsigned long) (sz)) >> 18) <= 2) ? 124 + (((unsigned long) (sz)) >> 18) :\ |
1590 | 126) |
1591 | |
1592 | #define largebin_index(sz) \ |
1593 | (SIZE_SZ == 8 ? largebin_index_64 (sz) \ |
1594 | : MALLOC_ALIGNMENT == 16 ? largebin_index_32_big (sz) \ |
1595 | : largebin_index_32 (sz)) |
1596 | |
1597 | #define bin_index(sz) \ |
1598 | ((in_smallbin_range (sz)) ? smallbin_index (sz) : largebin_index (sz)) |
1599 | |
1600 | /* Take a chunk off a bin list. */ |
1601 | static void |
1602 | unlink_chunk (mstate av, mchunkptr p) |
1603 | { |
1604 | if (chunksize (p) != prev_size (next_chunk (p))) |
1605 | malloc_printerr ("corrupted size vs. prev_size" ); |
1606 | |
1607 | mchunkptr fd = p->fd; |
1608 | mchunkptr bk = p->bk; |
1609 | |
1610 | if (__builtin_expect (fd->bk != p || bk->fd != p, 0)) |
1611 | malloc_printerr ("corrupted double-linked list" ); |
1612 | |
1613 | fd->bk = bk; |
1614 | bk->fd = fd; |
1615 | if (!in_smallbin_range (chunksize_nomask (p)) && p->fd_nextsize != NULL) |
1616 | { |
1617 | if (p->fd_nextsize->bk_nextsize != p |
1618 | || p->bk_nextsize->fd_nextsize != p) |
1619 | malloc_printerr ("corrupted double-linked list (not small)" ); |
1620 | |
1621 | if (fd->fd_nextsize == NULL) |
1622 | { |
1623 | if (p->fd_nextsize == p) |
1624 | fd->fd_nextsize = fd->bk_nextsize = fd; |
1625 | else |
1626 | { |
1627 | fd->fd_nextsize = p->fd_nextsize; |
1628 | fd->bk_nextsize = p->bk_nextsize; |
1629 | p->fd_nextsize->bk_nextsize = fd; |
1630 | p->bk_nextsize->fd_nextsize = fd; |
1631 | } |
1632 | } |
1633 | else |
1634 | { |
1635 | p->fd_nextsize->bk_nextsize = p->bk_nextsize; |
1636 | p->bk_nextsize->fd_nextsize = p->fd_nextsize; |
1637 | } |
1638 | } |
1639 | } |
1640 | |
1641 | /* |
1642 | Unsorted chunks |
1643 | |
1644 | All remainders from chunk splits, as well as all returned chunks, |
1645 | are first placed in the "unsorted" bin. They are then placed |
1646 | in regular bins after malloc gives them ONE chance to be used before |
1647 | binning. So, basically, the unsorted_chunks list acts as a queue, |
1648 | with chunks being placed on it in free (and malloc_consolidate), |
1649 | and taken off (to be either used or placed in bins) in malloc. |
1650 | |
1651 | The NON_MAIN_ARENA flag is never set for unsorted chunks, so it |
1652 | does not have to be taken into account in size comparisons. |
1653 | */ |
1654 | |
1655 | /* The otherwise unindexable 1-bin is used to hold unsorted chunks. */ |
1656 | #define unsorted_chunks(M) (bin_at (M, 1)) |
1657 | |
1658 | /* |
1659 | Top |
1660 | |
1661 | The top-most available chunk (i.e., the one bordering the end of |
1662 | available memory) is treated specially. It is never included in |
1663 | any bin, is used only if no other chunk is available, and is |
1664 | released back to the system if it is very large (see |
1665 | M_TRIM_THRESHOLD). Because top initially |
1666 | points to its own bin with initial zero size, thus forcing |
1667 | extension on the first malloc request, we avoid having any special |
1668 | code in malloc to check whether it even exists yet. But we still |
1669 | need to do so when getting memory from system, so we make |
1670 | initial_top treat the bin as a legal but unusable chunk during the |
1671 | interval between initialization and the first call to |
1672 | sysmalloc. (This is somewhat delicate, since it relies on |
1673 | the 2 preceding words to be zero during this interval as well.) |
1674 | */ |
1675 | |
1676 | /* Conveniently, the unsorted bin can be used as dummy top on first call */ |
1677 | #define initial_top(M) (unsorted_chunks (M)) |
1678 | |
1679 | /* |
1680 | Binmap |
1681 | |
1682 | To help compensate for the large number of bins, a one-level index |
1683 | structure is used for bin-by-bin searching. `binmap' is a |
1684 | bitvector recording whether bins are definitely empty so they can |
1685 | be skipped over during during traversals. The bits are NOT always |
1686 | cleared as soon as bins are empty, but instead only |
1687 | when they are noticed to be empty during traversal in malloc. |
1688 | */ |
1689 | |
1690 | /* Conservatively use 32 bits per map word, even if on 64bit system */ |
1691 | #define BINMAPSHIFT 5 |
1692 | #define BITSPERMAP (1U << BINMAPSHIFT) |
1693 | #define BINMAPSIZE (NBINS / BITSPERMAP) |
1694 | |
1695 | #define idx2block(i) ((i) >> BINMAPSHIFT) |
1696 | #define idx2bit(i) ((1U << ((i) & ((1U << BINMAPSHIFT) - 1)))) |
1697 | |
1698 | #define mark_bin(m, i) ((m)->binmap[idx2block (i)] |= idx2bit (i)) |
1699 | #define unmark_bin(m, i) ((m)->binmap[idx2block (i)] &= ~(idx2bit (i))) |
1700 | #define get_binmap(m, i) ((m)->binmap[idx2block (i)] & idx2bit (i)) |
1701 | |
1702 | /* |
1703 | Fastbins |
1704 | |
1705 | An array of lists holding recently freed small chunks. Fastbins |
1706 | are not doubly linked. It is faster to single-link them, and |
1707 | since chunks are never removed from the middles of these lists, |
1708 | double linking is not necessary. Also, unlike regular bins, they |
1709 | are not even processed in FIFO order (they use faster LIFO) since |
1710 | ordering doesn't much matter in the transient contexts in which |
1711 | fastbins are normally used. |
1712 | |
1713 | Chunks in fastbins keep their inuse bit set, so they cannot |
1714 | be consolidated with other free chunks. malloc_consolidate |
1715 | releases all chunks in fastbins and consolidates them with |
1716 | other free chunks. |
1717 | */ |
1718 | |
1719 | typedef struct malloc_chunk *mfastbinptr; |
1720 | #define fastbin(ar_ptr, idx) ((ar_ptr)->fastbinsY[idx]) |
1721 | |
1722 | /* offset 2 to use otherwise unindexable first 2 bins */ |
1723 | #define fastbin_index(sz) \ |
1724 | ((((unsigned int) (sz)) >> (SIZE_SZ == 8 ? 4 : 3)) - 2) |
1725 | |
1726 | |
1727 | /* The maximum fastbin request size we support */ |
1728 | #define MAX_FAST_SIZE (80 * SIZE_SZ / 4) |
1729 | |
1730 | #define NFASTBINS (fastbin_index (request2size (MAX_FAST_SIZE)) + 1) |
1731 | |
1732 | /* |
1733 | FASTBIN_CONSOLIDATION_THRESHOLD is the size of a chunk in free() |
1734 | that triggers automatic consolidation of possibly-surrounding |
1735 | fastbin chunks. This is a heuristic, so the exact value should not |
1736 | matter too much. It is defined at half the default trim threshold as a |
1737 | compromise heuristic to only attempt consolidation if it is likely |
1738 | to lead to trimming. However, it is not dynamically tunable, since |
1739 | consolidation reduces fragmentation surrounding large chunks even |
1740 | if trimming is not used. |
1741 | */ |
1742 | |
1743 | #define FASTBIN_CONSOLIDATION_THRESHOLD (65536UL) |
1744 | |
1745 | /* |
1746 | NONCONTIGUOUS_BIT indicates that MORECORE does not return contiguous |
1747 | regions. Otherwise, contiguity is exploited in merging together, |
1748 | when possible, results from consecutive MORECORE calls. |
1749 | |
1750 | The initial value comes from MORECORE_CONTIGUOUS, but is |
1751 | changed dynamically if mmap is ever used as an sbrk substitute. |
1752 | */ |
1753 | |
1754 | #define NONCONTIGUOUS_BIT (2U) |
1755 | |
1756 | #define contiguous(M) (((M)->flags & NONCONTIGUOUS_BIT) == 0) |
1757 | #define noncontiguous(M) (((M)->flags & NONCONTIGUOUS_BIT) != 0) |
1758 | #define set_noncontiguous(M) ((M)->flags |= NONCONTIGUOUS_BIT) |
1759 | #define set_contiguous(M) ((M)->flags &= ~NONCONTIGUOUS_BIT) |
1760 | |
1761 | /* Maximum size of memory handled in fastbins. */ |
1762 | static uint8_t global_max_fast; |
1763 | |
1764 | /* |
1765 | Set value of max_fast. |
1766 | Use impossibly small value if 0. |
1767 | Precondition: there are no existing fastbin chunks in the main arena. |
1768 | Since do_check_malloc_state () checks this, we call malloc_consolidate () |
1769 | before changing max_fast. Note other arenas will leak their fast bin |
1770 | entries if max_fast is reduced. |
1771 | */ |
1772 | |
1773 | #define set_max_fast(s) \ |
1774 | global_max_fast = (((size_t) (s) <= MALLOC_ALIGN_MASK - SIZE_SZ) \ |
1775 | ? MIN_CHUNK_SIZE / 2 : ((s + SIZE_SZ) & ~MALLOC_ALIGN_MASK)) |
1776 | |
1777 | static inline INTERNAL_SIZE_T |
1778 | get_max_fast (void) |
1779 | { |
1780 | /* Tell the GCC optimizers that global_max_fast is never larger |
1781 | than MAX_FAST_SIZE. This avoids out-of-bounds array accesses in |
1782 | _int_malloc after constant propagation of the size parameter. |
1783 | (The code never executes because malloc preserves the |
1784 | global_max_fast invariant, but the optimizers may not recognize |
1785 | this.) */ |
1786 | if (global_max_fast > MAX_FAST_SIZE) |
1787 | __builtin_unreachable (); |
1788 | return global_max_fast; |
1789 | } |
1790 | |
1791 | /* |
1792 | ----------- Internal state representation and initialization ----------- |
1793 | */ |
1794 | |
1795 | /* |
1796 | have_fastchunks indicates that there are probably some fastbin chunks. |
1797 | It is set true on entering a chunk into any fastbin, and cleared early in |
1798 | malloc_consolidate. The value is approximate since it may be set when there |
1799 | are no fastbin chunks, or it may be clear even if there are fastbin chunks |
1800 | available. Given it's sole purpose is to reduce number of redundant calls to |
1801 | malloc_consolidate, it does not affect correctness. As a result we can safely |
1802 | use relaxed atomic accesses. |
1803 | */ |
1804 | |
1805 | |
1806 | struct malloc_state |
1807 | { |
1808 | /* Serialize access. */ |
1809 | __libc_lock_define (, mutex); |
1810 | |
1811 | /* Flags (formerly in max_fast). */ |
1812 | int flags; |
1813 | |
1814 | /* Set if the fastbin chunks contain recently inserted free blocks. */ |
1815 | /* Note this is a bool but not all targets support atomics on booleans. */ |
1816 | int have_fastchunks; |
1817 | |
1818 | /* Fastbins */ |
1819 | mfastbinptr fastbinsY[NFASTBINS]; |
1820 | |
1821 | /* Base of the topmost chunk -- not otherwise kept in a bin */ |
1822 | mchunkptr top; |
1823 | |
1824 | /* The remainder from the most recent split of a small request */ |
1825 | mchunkptr last_remainder; |
1826 | |
1827 | /* Normal bins packed as described above */ |
1828 | mchunkptr bins[NBINS * 2 - 2]; |
1829 | |
1830 | /* Bitmap of bins */ |
1831 | unsigned int binmap[BINMAPSIZE]; |
1832 | |
1833 | /* Linked list */ |
1834 | struct malloc_state *next; |
1835 | |
1836 | /* Linked list for free arenas. Access to this field is serialized |
1837 | by free_list_lock in arena.c. */ |
1838 | struct malloc_state *next_free; |
1839 | |
1840 | /* Number of threads attached to this arena. 0 if the arena is on |
1841 | the free list. Access to this field is serialized by |
1842 | free_list_lock in arena.c. */ |
1843 | INTERNAL_SIZE_T attached_threads; |
1844 | |
1845 | /* Memory allocated from the system in this arena. */ |
1846 | INTERNAL_SIZE_T system_mem; |
1847 | INTERNAL_SIZE_T max_system_mem; |
1848 | }; |
1849 | |
1850 | struct malloc_par |
1851 | { |
1852 | /* Tunable parameters */ |
1853 | unsigned long trim_threshold; |
1854 | INTERNAL_SIZE_T top_pad; |
1855 | INTERNAL_SIZE_T mmap_threshold; |
1856 | INTERNAL_SIZE_T arena_test; |
1857 | INTERNAL_SIZE_T arena_max; |
1858 | |
1859 | /* Transparent Large Page support. */ |
1860 | INTERNAL_SIZE_T thp_pagesize; |
1861 | /* A value different than 0 means to align mmap allocation to hp_pagesize |
1862 | add hp_flags on flags. */ |
1863 | INTERNAL_SIZE_T hp_pagesize; |
1864 | int hp_flags; |
1865 | |
1866 | /* Memory map support */ |
1867 | int n_mmaps; |
1868 | int n_mmaps_max; |
1869 | int max_n_mmaps; |
1870 | /* the mmap_threshold is dynamic, until the user sets |
1871 | it manually, at which point we need to disable any |
1872 | dynamic behavior. */ |
1873 | int no_dyn_threshold; |
1874 | |
1875 | /* Statistics */ |
1876 | INTERNAL_SIZE_T mmapped_mem; |
1877 | INTERNAL_SIZE_T max_mmapped_mem; |
1878 | |
1879 | /* First address handed out by MORECORE/sbrk. */ |
1880 | char *sbrk_base; |
1881 | |
1882 | #if USE_TCACHE |
1883 | /* Maximum number of buckets to use. */ |
1884 | size_t tcache_bins; |
1885 | size_t tcache_max_bytes; |
1886 | /* Maximum number of chunks in each bucket. */ |
1887 | size_t tcache_count; |
1888 | /* Maximum number of chunks to remove from the unsorted list, which |
1889 | aren't used to prefill the cache. */ |
1890 | size_t tcache_unsorted_limit; |
1891 | #endif |
1892 | }; |
1893 | |
1894 | /* There are several instances of this struct ("arenas") in this |
1895 | malloc. If you are adapting this malloc in a way that does NOT use |
1896 | a static or mmapped malloc_state, you MUST explicitly zero-fill it |
1897 | before using. This malloc relies on the property that malloc_state |
1898 | is initialized to all zeroes (as is true of C statics). */ |
1899 | |
1900 | static struct malloc_state main_arena = |
1901 | { |
1902 | .mutex = _LIBC_LOCK_INITIALIZER, |
1903 | .next = &main_arena, |
1904 | .attached_threads = 1 |
1905 | }; |
1906 | |
1907 | /* There is only one instance of the malloc parameters. */ |
1908 | |
1909 | static struct malloc_par mp_ = |
1910 | { |
1911 | .top_pad = DEFAULT_TOP_PAD, |
1912 | .n_mmaps_max = DEFAULT_MMAP_MAX, |
1913 | .mmap_threshold = DEFAULT_MMAP_THRESHOLD, |
1914 | .trim_threshold = DEFAULT_TRIM_THRESHOLD, |
1915 | #define NARENAS_FROM_NCORES(n) ((n) * (sizeof (long) == 4 ? 2 : 8)) |
1916 | .arena_test = NARENAS_FROM_NCORES (1) |
1917 | #if USE_TCACHE |
1918 | , |
1919 | .tcache_count = TCACHE_FILL_COUNT, |
1920 | .tcache_bins = TCACHE_MAX_BINS, |
1921 | .tcache_max_bytes = tidx2usize (TCACHE_MAX_BINS-1), |
1922 | .tcache_unsorted_limit = 0 /* No limit. */ |
1923 | #endif |
1924 | }; |
1925 | |
1926 | /* |
1927 | Initialize a malloc_state struct. |
1928 | |
1929 | This is called from ptmalloc_init () or from _int_new_arena () |
1930 | when creating a new arena. |
1931 | */ |
1932 | |
1933 | static void |
1934 | malloc_init_state (mstate av) |
1935 | { |
1936 | int i; |
1937 | mbinptr bin; |
1938 | |
1939 | /* Establish circular links for normal bins */ |
1940 | for (i = 1; i < NBINS; ++i) |
1941 | { |
1942 | bin = bin_at (av, i); |
1943 | bin->fd = bin->bk = bin; |
1944 | } |
1945 | |
1946 | #if MORECORE_CONTIGUOUS |
1947 | if (av != &main_arena) |
1948 | #endif |
1949 | set_noncontiguous (av); |
1950 | if (av == &main_arena) |
1951 | set_max_fast (DEFAULT_MXFAST); |
1952 | atomic_store_relaxed (&av->have_fastchunks, false); |
1953 | |
1954 | av->top = initial_top (av); |
1955 | } |
1956 | |
1957 | /* |
1958 | Other internal utilities operating on mstates |
1959 | */ |
1960 | |
1961 | static void *sysmalloc (INTERNAL_SIZE_T, mstate); |
1962 | static int systrim (size_t, mstate); |
1963 | static void malloc_consolidate (mstate); |
1964 | |
1965 | |
1966 | /* -------------- Early definitions for debugging hooks ---------------- */ |
1967 | |
1968 | /* This function is called from the arena shutdown hook, to free the |
1969 | thread cache (if it exists). */ |
1970 | static void tcache_thread_shutdown (void); |
1971 | |
1972 | /* ------------------ Testing support ----------------------------------*/ |
1973 | |
1974 | static int perturb_byte; |
1975 | |
1976 | static void |
1977 | alloc_perturb (char *p, size_t n) |
1978 | { |
1979 | if (__glibc_unlikely (perturb_byte)) |
1980 | memset (p, perturb_byte ^ 0xff, n); |
1981 | } |
1982 | |
1983 | static void |
1984 | free_perturb (char *p, size_t n) |
1985 | { |
1986 | if (__glibc_unlikely (perturb_byte)) |
1987 | memset (p, perturb_byte, n); |
1988 | } |
1989 | |
1990 | |
1991 | |
1992 | #include <stap-probe.h> |
1993 | |
1994 | /* ----------- Routines dealing with transparent huge pages ----------- */ |
1995 | |
1996 | static inline void |
1997 | madvise_thp (void *p, INTERNAL_SIZE_T size) |
1998 | { |
1999 | #ifdef MADV_HUGEPAGE |
2000 | /* Do not consider areas smaller than a huge page or if the tunable is |
2001 | not active. */ |
2002 | if (mp_.thp_pagesize == 0 || size < mp_.thp_pagesize) |
2003 | return; |
2004 | |
2005 | /* Linux requires the input address to be page-aligned, and unaligned |
2006 | inputs happens only for initial data segment. */ |
2007 | if (__glibc_unlikely (!PTR_IS_ALIGNED (p, GLRO (dl_pagesize)))) |
2008 | { |
2009 | void *q = PTR_ALIGN_DOWN (p, GLRO (dl_pagesize)); |
2010 | size += PTR_DIFF (p, q); |
2011 | p = q; |
2012 | } |
2013 | |
2014 | __madvise (p, size, MADV_HUGEPAGE); |
2015 | #endif |
2016 | } |
2017 | |
2018 | /* ------------------- Support for multiple arenas -------------------- */ |
2019 | #include "arena.c" |
2020 | |
2021 | /* |
2022 | Debugging support |
2023 | |
2024 | These routines make a number of assertions about the states |
2025 | of data structures that should be true at all times. If any |
2026 | are not true, it's very likely that a user program has somehow |
2027 | trashed memory. (It's also possible that there is a coding error |
2028 | in malloc. In which case, please report it!) |
2029 | */ |
2030 | |
2031 | #if !MALLOC_DEBUG |
2032 | |
2033 | # define check_chunk(A, P) |
2034 | # define check_free_chunk(A, P) |
2035 | # define check_inuse_chunk(A, P) |
2036 | # define check_remalloced_chunk(A, P, N) |
2037 | # define check_malloced_chunk(A, P, N) |
2038 | # define check_malloc_state(A) |
2039 | |
2040 | #else |
2041 | |
2042 | # define check_chunk(A, P) do_check_chunk (A, P) |
2043 | # define check_free_chunk(A, P) do_check_free_chunk (A, P) |
2044 | # define check_inuse_chunk(A, P) do_check_inuse_chunk (A, P) |
2045 | # define check_remalloced_chunk(A, P, N) do_check_remalloced_chunk (A, P, N) |
2046 | # define check_malloced_chunk(A, P, N) do_check_malloced_chunk (A, P, N) |
2047 | # define check_malloc_state(A) do_check_malloc_state (A) |
2048 | |
2049 | /* |
2050 | Properties of all chunks |
2051 | */ |
2052 | |
2053 | static void |
2054 | do_check_chunk (mstate av, mchunkptr p) |
2055 | { |
2056 | unsigned long sz = chunksize (p); |
2057 | /* min and max possible addresses assuming contiguous allocation */ |
2058 | char *max_address = (char *) (av->top) + chunksize (av->top); |
2059 | char *min_address = max_address - av->system_mem; |
2060 | |
2061 | if (!chunk_is_mmapped (p)) |
2062 | { |
2063 | /* Has legal address ... */ |
2064 | if (p != av->top) |
2065 | { |
2066 | if (contiguous (av)) |
2067 | { |
2068 | assert (((char *) p) >= min_address); |
2069 | assert (((char *) p + sz) <= ((char *) (av->top))); |
2070 | } |
2071 | } |
2072 | else |
2073 | { |
2074 | /* top size is always at least MINSIZE */ |
2075 | assert ((unsigned long) (sz) >= MINSIZE); |
2076 | /* top predecessor always marked inuse */ |
2077 | assert (prev_inuse (p)); |
2078 | } |
2079 | } |
2080 | else |
2081 | { |
2082 | /* address is outside main heap */ |
2083 | if (contiguous (av) && av->top != initial_top (av)) |
2084 | { |
2085 | assert (((char *) p) < min_address || ((char *) p) >= max_address); |
2086 | } |
2087 | /* chunk is page-aligned */ |
2088 | assert (((prev_size (p) + sz) & (GLRO (dl_pagesize) - 1)) == 0); |
2089 | /* mem is aligned */ |
2090 | assert (aligned_OK (chunk2mem (p))); |
2091 | } |
2092 | } |
2093 | |
2094 | /* |
2095 | Properties of free chunks |
2096 | */ |
2097 | |
2098 | static void |
2099 | do_check_free_chunk (mstate av, mchunkptr p) |
2100 | { |
2101 | INTERNAL_SIZE_T sz = chunksize_nomask (p) & ~(PREV_INUSE | NON_MAIN_ARENA); |
2102 | mchunkptr next = chunk_at_offset (p, sz); |
2103 | |
2104 | do_check_chunk (av, p); |
2105 | |
2106 | /* Chunk must claim to be free ... */ |
2107 | assert (!inuse (p)); |
2108 | assert (!chunk_is_mmapped (p)); |
2109 | |
2110 | /* Unless a special marker, must have OK fields */ |
2111 | if ((unsigned long) (sz) >= MINSIZE) |
2112 | { |
2113 | assert ((sz & MALLOC_ALIGN_MASK) == 0); |
2114 | assert (aligned_OK (chunk2mem (p))); |
2115 | /* ... matching footer field */ |
2116 | assert (prev_size (next_chunk (p)) == sz); |
2117 | /* ... and is fully consolidated */ |
2118 | assert (prev_inuse (p)); |
2119 | assert (next == av->top || inuse (next)); |
2120 | |
2121 | /* ... and has minimally sane links */ |
2122 | assert (p->fd->bk == p); |
2123 | assert (p->bk->fd == p); |
2124 | } |
2125 | else /* markers are always of size SIZE_SZ */ |
2126 | assert (sz == SIZE_SZ); |
2127 | } |
2128 | |
2129 | /* |
2130 | Properties of inuse chunks |
2131 | */ |
2132 | |
2133 | static void |
2134 | do_check_inuse_chunk (mstate av, mchunkptr p) |
2135 | { |
2136 | mchunkptr next; |
2137 | |
2138 | do_check_chunk (av, p); |
2139 | |
2140 | if (chunk_is_mmapped (p)) |
2141 | return; /* mmapped chunks have no next/prev */ |
2142 | |
2143 | /* Check whether it claims to be in use ... */ |
2144 | assert (inuse (p)); |
2145 | |
2146 | next = next_chunk (p); |
2147 | |
2148 | /* ... and is surrounded by OK chunks. |
2149 | Since more things can be checked with free chunks than inuse ones, |
2150 | if an inuse chunk borders them and debug is on, it's worth doing them. |
2151 | */ |
2152 | if (!prev_inuse (p)) |
2153 | { |
2154 | /* Note that we cannot even look at prev unless it is not inuse */ |
2155 | mchunkptr prv = prev_chunk (p); |
2156 | assert (next_chunk (prv) == p); |
2157 | do_check_free_chunk (av, prv); |
2158 | } |
2159 | |
2160 | if (next == av->top) |
2161 | { |
2162 | assert (prev_inuse (next)); |
2163 | assert (chunksize (next) >= MINSIZE); |
2164 | } |
2165 | else if (!inuse (next)) |
2166 | do_check_free_chunk (av, next); |
2167 | } |
2168 | |
2169 | /* |
2170 | Properties of chunks recycled from fastbins |
2171 | */ |
2172 | |
2173 | static void |
2174 | do_check_remalloced_chunk (mstate av, mchunkptr p, INTERNAL_SIZE_T s) |
2175 | { |
2176 | INTERNAL_SIZE_T sz = chunksize_nomask (p) & ~(PREV_INUSE | NON_MAIN_ARENA); |
2177 | |
2178 | if (!chunk_is_mmapped (p)) |
2179 | { |
2180 | assert (av == arena_for_chunk (p)); |
2181 | if (chunk_main_arena (p)) |
2182 | assert (av == &main_arena); |
2183 | else |
2184 | assert (av != &main_arena); |
2185 | } |
2186 | |
2187 | do_check_inuse_chunk (av, p); |
2188 | |
2189 | /* Legal size ... */ |
2190 | assert ((sz & MALLOC_ALIGN_MASK) == 0); |
2191 | assert ((unsigned long) (sz) >= MINSIZE); |
2192 | /* ... and alignment */ |
2193 | assert (aligned_OK (chunk2mem (p))); |
2194 | /* chunk is less than MINSIZE more than request */ |
2195 | assert ((long) (sz) - (long) (s) >= 0); |
2196 | assert ((long) (sz) - (long) (s + MINSIZE) < 0); |
2197 | } |
2198 | |
2199 | /* |
2200 | Properties of nonrecycled chunks at the point they are malloced |
2201 | */ |
2202 | |
2203 | static void |
2204 | do_check_malloced_chunk (mstate av, mchunkptr p, INTERNAL_SIZE_T s) |
2205 | { |
2206 | /* same as recycled case ... */ |
2207 | do_check_remalloced_chunk (av, p, s); |
2208 | |
2209 | /* |
2210 | ... plus, must obey implementation invariant that prev_inuse is |
2211 | always true of any allocated chunk; i.e., that each allocated |
2212 | chunk borders either a previously allocated and still in-use |
2213 | chunk, or the base of its memory arena. This is ensured |
2214 | by making all allocations from the `lowest' part of any found |
2215 | chunk. This does not necessarily hold however for chunks |
2216 | recycled via fastbins. |
2217 | */ |
2218 | |
2219 | assert (prev_inuse (p)); |
2220 | } |
2221 | |
2222 | |
2223 | /* |
2224 | Properties of malloc_state. |
2225 | |
2226 | This may be useful for debugging malloc, as well as detecting user |
2227 | programmer errors that somehow write into malloc_state. |
2228 | |
2229 | If you are extending or experimenting with this malloc, you can |
2230 | probably figure out how to hack this routine to print out or |
2231 | display chunk addresses, sizes, bins, and other instrumentation. |
2232 | */ |
2233 | |
2234 | static void |
2235 | do_check_malloc_state (mstate av) |
2236 | { |
2237 | int i; |
2238 | mchunkptr p; |
2239 | mchunkptr q; |
2240 | mbinptr b; |
2241 | unsigned int idx; |
2242 | INTERNAL_SIZE_T size; |
2243 | unsigned long total = 0; |
2244 | int max_fast_bin; |
2245 | |
2246 | /* internal size_t must be no wider than pointer type */ |
2247 | assert (sizeof (INTERNAL_SIZE_T) <= sizeof (char *)); |
2248 | |
2249 | /* alignment is a power of 2 */ |
2250 | assert ((MALLOC_ALIGNMENT & (MALLOC_ALIGNMENT - 1)) == 0); |
2251 | |
2252 | /* Check the arena is initialized. */ |
2253 | assert (av->top != 0); |
2254 | |
2255 | /* No memory has been allocated yet, so doing more tests is not possible. */ |
2256 | if (av->top == initial_top (av)) |
2257 | return; |
2258 | |
2259 | /* pagesize is a power of 2 */ |
2260 | assert (powerof2(GLRO (dl_pagesize))); |
2261 | |
2262 | /* A contiguous main_arena is consistent with sbrk_base. */ |
2263 | if (av == &main_arena && contiguous (av)) |
2264 | assert ((char *) mp_.sbrk_base + av->system_mem == |
2265 | (char *) av->top + chunksize (av->top)); |
2266 | |
2267 | /* properties of fastbins */ |
2268 | |
2269 | /* max_fast is in allowed range */ |
2270 | assert ((get_max_fast () & ~1) <= request2size (MAX_FAST_SIZE)); |
2271 | |
2272 | max_fast_bin = fastbin_index (get_max_fast ()); |
2273 | |
2274 | for (i = 0; i < NFASTBINS; ++i) |
2275 | { |
2276 | p = fastbin (av, i); |
2277 | |
2278 | /* The following test can only be performed for the main arena. |
2279 | While mallopt calls malloc_consolidate to get rid of all fast |
2280 | bins (especially those larger than the new maximum) this does |
2281 | only happen for the main arena. Trying to do this for any |
2282 | other arena would mean those arenas have to be locked and |
2283 | malloc_consolidate be called for them. This is excessive. And |
2284 | even if this is acceptable to somebody it still cannot solve |
2285 | the problem completely since if the arena is locked a |
2286 | concurrent malloc call might create a new arena which then |
2287 | could use the newly invalid fast bins. */ |
2288 | |
2289 | /* all bins past max_fast are empty */ |
2290 | if (av == &main_arena && i > max_fast_bin) |
2291 | assert (p == 0); |
2292 | |
2293 | while (p != 0) |
2294 | { |
2295 | if (__glibc_unlikely (misaligned_chunk (p))) |
2296 | malloc_printerr ("do_check_malloc_state(): " |
2297 | "unaligned fastbin chunk detected" ); |
2298 | /* each chunk claims to be inuse */ |
2299 | do_check_inuse_chunk (av, p); |
2300 | total += chunksize (p); |
2301 | /* chunk belongs in this bin */ |
2302 | assert (fastbin_index (chunksize (p)) == i); |
2303 | p = REVEAL_PTR (p->fd); |
2304 | } |
2305 | } |
2306 | |
2307 | /* check normal bins */ |
2308 | for (i = 1; i < NBINS; ++i) |
2309 | { |
2310 | b = bin_at (av, i); |
2311 | |
2312 | /* binmap is accurate (except for bin 1 == unsorted_chunks) */ |
2313 | if (i >= 2) |
2314 | { |
2315 | unsigned int binbit = get_binmap (av, i); |
2316 | int empty = last (b) == b; |
2317 | if (!binbit) |
2318 | assert (empty); |
2319 | else if (!empty) |
2320 | assert (binbit); |
2321 | } |
2322 | |
2323 | for (p = last (b); p != b; p = p->bk) |
2324 | { |
2325 | /* each chunk claims to be free */ |
2326 | do_check_free_chunk (av, p); |
2327 | size = chunksize (p); |
2328 | total += size; |
2329 | if (i >= 2) |
2330 | { |
2331 | /* chunk belongs in bin */ |
2332 | idx = bin_index (size); |
2333 | assert (idx == i); |
2334 | /* lists are sorted */ |
2335 | assert (p->bk == b || |
2336 | (unsigned long) chunksize (p->bk) >= (unsigned long) chunksize (p)); |
2337 | |
2338 | if (!in_smallbin_range (size)) |
2339 | { |
2340 | if (p->fd_nextsize != NULL) |
2341 | { |
2342 | if (p->fd_nextsize == p) |
2343 | assert (p->bk_nextsize == p); |
2344 | else |
2345 | { |
2346 | if (p->fd_nextsize == first (b)) |
2347 | assert (chunksize (p) < chunksize (p->fd_nextsize)); |
2348 | else |
2349 | assert (chunksize (p) > chunksize (p->fd_nextsize)); |
2350 | |
2351 | if (p == first (b)) |
2352 | assert (chunksize (p) > chunksize (p->bk_nextsize)); |
2353 | else |
2354 | assert (chunksize (p) < chunksize (p->bk_nextsize)); |
2355 | } |
2356 | } |
2357 | else |
2358 | assert (p->bk_nextsize == NULL); |
2359 | } |
2360 | } |
2361 | else if (!in_smallbin_range (size)) |
2362 | assert (p->fd_nextsize == NULL && p->bk_nextsize == NULL); |
2363 | /* chunk is followed by a legal chain of inuse chunks */ |
2364 | for (q = next_chunk (p); |
2365 | (q != av->top && inuse (q) && |
2366 | (unsigned long) (chunksize (q)) >= MINSIZE); |
2367 | q = next_chunk (q)) |
2368 | do_check_inuse_chunk (av, q); |
2369 | } |
2370 | } |
2371 | |
2372 | /* top chunk is OK */ |
2373 | check_chunk (av, av->top); |
2374 | } |
2375 | #endif |
2376 | |
2377 | |
2378 | /* ----------------- Support for debugging hooks -------------------- */ |
2379 | #if IS_IN (libc) |
2380 | #include "hooks.c" |
2381 | #endif |
2382 | |
2383 | |
2384 | /* ----------- Routines dealing with system allocation -------------- */ |
2385 | |
2386 | /* |
2387 | sysmalloc handles malloc cases requiring more memory from the system. |
2388 | On entry, it is assumed that av->top does not have enough |
2389 | space to service request for nb bytes, thus requiring that av->top |
2390 | be extended or replaced. |
2391 | */ |
2392 | |
2393 | static void * |
2394 | sysmalloc_mmap (INTERNAL_SIZE_T nb, size_t pagesize, int , mstate av) |
2395 | { |
2396 | long int size; |
2397 | |
2398 | /* |
2399 | Round up size to nearest page. For mmapped chunks, the overhead is one |
2400 | SIZE_SZ unit larger than for normal chunks, because there is no |
2401 | following chunk whose prev_size field could be used. |
2402 | |
2403 | See the front_misalign handling below, for glibc there is no need for |
2404 | further alignments unless we have have high alignment. |
2405 | */ |
2406 | if (MALLOC_ALIGNMENT == CHUNK_HDR_SZ) |
2407 | size = ALIGN_UP (nb + SIZE_SZ, pagesize); |
2408 | else |
2409 | size = ALIGN_UP (nb + SIZE_SZ + MALLOC_ALIGN_MASK, pagesize); |
2410 | |
2411 | /* Don't try if size wraps around 0. */ |
2412 | if ((unsigned long) (size) <= (unsigned long) (nb)) |
2413 | return MAP_FAILED; |
2414 | |
2415 | char *mm = (char *) MMAP (0, size, |
2416 | mtag_mmap_flags | PROT_READ | PROT_WRITE, |
2417 | extra_flags); |
2418 | if (mm == MAP_FAILED) |
2419 | return mm; |
2420 | |
2421 | #ifdef MAP_HUGETLB |
2422 | if (!(extra_flags & MAP_HUGETLB)) |
2423 | madvise_thp (mm, size); |
2424 | #endif |
2425 | |
2426 | /* |
2427 | The offset to the start of the mmapped region is stored in the prev_size |
2428 | field of the chunk. This allows us to adjust returned start address to |
2429 | meet alignment requirements here and in memalign(), and still be able to |
2430 | compute proper address argument for later munmap in free() and realloc(). |
2431 | */ |
2432 | |
2433 | INTERNAL_SIZE_T front_misalign; /* unusable bytes at front of new space */ |
2434 | |
2435 | if (MALLOC_ALIGNMENT == CHUNK_HDR_SZ) |
2436 | { |
2437 | /* For glibc, chunk2mem increases the address by CHUNK_HDR_SZ and |
2438 | MALLOC_ALIGN_MASK is CHUNK_HDR_SZ-1. Each mmap'ed area is page |
2439 | aligned and therefore definitely MALLOC_ALIGN_MASK-aligned. */ |
2440 | assert (((INTERNAL_SIZE_T) chunk2mem (mm) & MALLOC_ALIGN_MASK) == 0); |
2441 | front_misalign = 0; |
2442 | } |
2443 | else |
2444 | front_misalign = (INTERNAL_SIZE_T) chunk2mem (mm) & MALLOC_ALIGN_MASK; |
2445 | |
2446 | mchunkptr p; /* the allocated/returned chunk */ |
2447 | |
2448 | if (front_misalign > 0) |
2449 | { |
2450 | ptrdiff_t correction = MALLOC_ALIGNMENT - front_misalign; |
2451 | p = (mchunkptr) (mm + correction); |
2452 | set_prev_size (p, correction); |
2453 | set_head (p, (size - correction) | IS_MMAPPED); |
2454 | } |
2455 | else |
2456 | { |
2457 | p = (mchunkptr) mm; |
2458 | set_prev_size (p, 0); |
2459 | set_head (p, size | IS_MMAPPED); |
2460 | } |
2461 | |
2462 | /* update statistics */ |
2463 | int new = atomic_fetch_add_relaxed (&mp_.n_mmaps, 1) + 1; |
2464 | atomic_max (&mp_.max_n_mmaps, new); |
2465 | |
2466 | unsigned long sum; |
2467 | sum = atomic_fetch_add_relaxed (&mp_.mmapped_mem, size) + size; |
2468 | atomic_max (&mp_.max_mmapped_mem, sum); |
2469 | |
2470 | check_chunk (av, p); |
2471 | |
2472 | return chunk2mem (p); |
2473 | } |
2474 | |
2475 | /* |
2476 | Allocate memory using mmap() based on S and NB requested size, aligning to |
2477 | PAGESIZE if required. The EXTRA_FLAGS is used on mmap() call. If the call |
2478 | succeeds S is updated with the allocated size. This is used as a fallback |
2479 | if MORECORE fails. |
2480 | */ |
2481 | static void * |
2482 | sysmalloc_mmap_fallback (long int *s, INTERNAL_SIZE_T nb, |
2483 | INTERNAL_SIZE_T old_size, size_t minsize, |
2484 | size_t pagesize, int , mstate av) |
2485 | { |
2486 | long int size = *s; |
2487 | |
2488 | /* Cannot merge with old top, so add its size back in */ |
2489 | if (contiguous (av)) |
2490 | size = ALIGN_UP (size + old_size, pagesize); |
2491 | |
2492 | /* If we are relying on mmap as backup, then use larger units */ |
2493 | if ((unsigned long) (size) < minsize) |
2494 | size = minsize; |
2495 | |
2496 | /* Don't try if size wraps around 0 */ |
2497 | if ((unsigned long) (size) <= (unsigned long) (nb)) |
2498 | return MORECORE_FAILURE; |
2499 | |
2500 | char *mbrk = (char *) (MMAP (0, size, |
2501 | mtag_mmap_flags | PROT_READ | PROT_WRITE, |
2502 | extra_flags)); |
2503 | if (mbrk == MAP_FAILED) |
2504 | return MAP_FAILED; |
2505 | |
2506 | #ifdef MAP_HUGETLB |
2507 | if (!(extra_flags & MAP_HUGETLB)) |
2508 | madvise_thp (mbrk, size); |
2509 | #endif |
2510 | |
2511 | /* Record that we no longer have a contiguous sbrk region. After the first |
2512 | time mmap is used as backup, we do not ever rely on contiguous space |
2513 | since this could incorrectly bridge regions. */ |
2514 | set_noncontiguous (av); |
2515 | |
2516 | *s = size; |
2517 | return mbrk; |
2518 | } |
2519 | |
2520 | static void * |
2521 | sysmalloc (INTERNAL_SIZE_T nb, mstate av) |
2522 | { |
2523 | mchunkptr old_top; /* incoming value of av->top */ |
2524 | INTERNAL_SIZE_T old_size; /* its size */ |
2525 | char *old_end; /* its end address */ |
2526 | |
2527 | long size; /* arg to first MORECORE or mmap call */ |
2528 | char *brk; /* return value from MORECORE */ |
2529 | |
2530 | long correction; /* arg to 2nd MORECORE call */ |
2531 | char *snd_brk; /* 2nd return val */ |
2532 | |
2533 | INTERNAL_SIZE_T front_misalign; /* unusable bytes at front of new space */ |
2534 | INTERNAL_SIZE_T end_misalign; /* partial page left at end of new space */ |
2535 | char *aligned_brk; /* aligned offset into brk */ |
2536 | |
2537 | mchunkptr p; /* the allocated/returned chunk */ |
2538 | mchunkptr remainder; /* remainder from allocation */ |
2539 | unsigned long remainder_size; /* its size */ |
2540 | |
2541 | |
2542 | size_t pagesize = GLRO (dl_pagesize); |
2543 | bool tried_mmap = false; |
2544 | |
2545 | |
2546 | /* |
2547 | If have mmap, and the request size meets the mmap threshold, and |
2548 | the system supports mmap, and there are few enough currently |
2549 | allocated mmapped regions, try to directly map this request |
2550 | rather than expanding top. |
2551 | */ |
2552 | |
2553 | if (av == NULL |
2554 | || ((unsigned long) (nb) >= (unsigned long) (mp_.mmap_threshold) |
2555 | && (mp_.n_mmaps < mp_.n_mmaps_max))) |
2556 | { |
2557 | char *mm; |
2558 | if (mp_.hp_pagesize > 0 && nb >= mp_.hp_pagesize) |
2559 | { |
2560 | /* There is no need to issue the THP madvise call if Huge Pages are |
2561 | used directly. */ |
2562 | mm = sysmalloc_mmap (nb, mp_.hp_pagesize, mp_.hp_flags, av); |
2563 | if (mm != MAP_FAILED) |
2564 | return mm; |
2565 | } |
2566 | mm = sysmalloc_mmap (nb, pagesize, 0, av); |
2567 | if (mm != MAP_FAILED) |
2568 | return mm; |
2569 | tried_mmap = true; |
2570 | } |
2571 | |
2572 | /* There are no usable arenas and mmap also failed. */ |
2573 | if (av == NULL) |
2574 | return 0; |
2575 | |
2576 | /* Record incoming configuration of top */ |
2577 | |
2578 | old_top = av->top; |
2579 | old_size = chunksize (old_top); |
2580 | old_end = (char *) (chunk_at_offset (old_top, old_size)); |
2581 | |
2582 | brk = snd_brk = (char *) (MORECORE_FAILURE); |
2583 | |
2584 | /* |
2585 | If not the first time through, we require old_size to be |
2586 | at least MINSIZE and to have prev_inuse set. |
2587 | */ |
2588 | |
2589 | assert ((old_top == initial_top (av) && old_size == 0) || |
2590 | ((unsigned long) (old_size) >= MINSIZE && |
2591 | prev_inuse (old_top) && |
2592 | ((unsigned long) old_end & (pagesize - 1)) == 0)); |
2593 | |
2594 | /* Precondition: not enough current space to satisfy nb request */ |
2595 | assert ((unsigned long) (old_size) < (unsigned long) (nb + MINSIZE)); |
2596 | |
2597 | |
2598 | if (av != &main_arena) |
2599 | { |
2600 | heap_info *old_heap, *heap; |
2601 | size_t old_heap_size; |
2602 | |
2603 | /* First try to extend the current heap. */ |
2604 | old_heap = heap_for_ptr (old_top); |
2605 | old_heap_size = old_heap->size; |
2606 | if ((long) (MINSIZE + nb - old_size) > 0 |
2607 | && grow_heap (old_heap, MINSIZE + nb - old_size) == 0) |
2608 | { |
2609 | av->system_mem += old_heap->size - old_heap_size; |
2610 | set_head (old_top, (((char *) old_heap + old_heap->size) - (char *) old_top) |
2611 | | PREV_INUSE); |
2612 | } |
2613 | else if ((heap = new_heap (nb + (MINSIZE + sizeof (*heap)), mp_.top_pad))) |
2614 | { |
2615 | /* Use a newly allocated heap. */ |
2616 | heap->ar_ptr = av; |
2617 | heap->prev = old_heap; |
2618 | av->system_mem += heap->size; |
2619 | /* Set up the new top. */ |
2620 | top (av) = chunk_at_offset (heap, sizeof (*heap)); |
2621 | set_head (top (av), (heap->size - sizeof (*heap)) | PREV_INUSE); |
2622 | |
2623 | /* Setup fencepost and free the old top chunk with a multiple of |
2624 | MALLOC_ALIGNMENT in size. */ |
2625 | /* The fencepost takes at least MINSIZE bytes, because it might |
2626 | become the top chunk again later. Note that a footer is set |
2627 | up, too, although the chunk is marked in use. */ |
2628 | old_size = (old_size - MINSIZE) & ~MALLOC_ALIGN_MASK; |
2629 | set_head (chunk_at_offset (old_top, old_size + CHUNK_HDR_SZ), |
2630 | 0 | PREV_INUSE); |
2631 | if (old_size >= MINSIZE) |
2632 | { |
2633 | set_head (chunk_at_offset (old_top, old_size), |
2634 | CHUNK_HDR_SZ | PREV_INUSE); |
2635 | set_foot (chunk_at_offset (old_top, old_size), CHUNK_HDR_SZ); |
2636 | set_head (old_top, old_size | PREV_INUSE | NON_MAIN_ARENA); |
2637 | _int_free (av, old_top, 1); |
2638 | } |
2639 | else |
2640 | { |
2641 | set_head (old_top, (old_size + CHUNK_HDR_SZ) | PREV_INUSE); |
2642 | set_foot (old_top, (old_size + CHUNK_HDR_SZ)); |
2643 | } |
2644 | } |
2645 | else if (!tried_mmap) |
2646 | { |
2647 | /* We can at least try to use to mmap memory. If new_heap fails |
2648 | it is unlikely that trying to allocate huge pages will |
2649 | succeed. */ |
2650 | char *mm = sysmalloc_mmap (nb, pagesize, 0, av); |
2651 | if (mm != MAP_FAILED) |
2652 | return mm; |
2653 | } |
2654 | } |
2655 | else /* av == main_arena */ |
2656 | |
2657 | |
2658 | { /* Request enough space for nb + pad + overhead */ |
2659 | size = nb + mp_.top_pad + MINSIZE; |
2660 | |
2661 | /* |
2662 | If contiguous, we can subtract out existing space that we hope to |
2663 | combine with new space. We add it back later only if |
2664 | we don't actually get contiguous space. |
2665 | */ |
2666 | |
2667 | if (contiguous (av)) |
2668 | size -= old_size; |
2669 | |
2670 | /* |
2671 | Round to a multiple of page size or huge page size. |
2672 | If MORECORE is not contiguous, this ensures that we only call it |
2673 | with whole-page arguments. And if MORECORE is contiguous and |
2674 | this is not first time through, this preserves page-alignment of |
2675 | previous calls. Otherwise, we correct to page-align below. |
2676 | */ |
2677 | |
2678 | #ifdef MADV_HUGEPAGE |
2679 | /* Defined in brk.c. */ |
2680 | extern void *__curbrk; |
2681 | if (__glibc_unlikely (mp_.thp_pagesize != 0)) |
2682 | { |
2683 | uintptr_t top = ALIGN_UP ((uintptr_t) __curbrk + size, |
2684 | mp_.thp_pagesize); |
2685 | size = top - (uintptr_t) __curbrk; |
2686 | } |
2687 | else |
2688 | #endif |
2689 | size = ALIGN_UP (size, GLRO(dl_pagesize)); |
2690 | |
2691 | /* |
2692 | Don't try to call MORECORE if argument is so big as to appear |
2693 | negative. Note that since mmap takes size_t arg, it may succeed |
2694 | below even if we cannot call MORECORE. |
2695 | */ |
2696 | |
2697 | if (size > 0) |
2698 | { |
2699 | brk = (char *) (MORECORE (size)); |
2700 | if (brk != (char *) (MORECORE_FAILURE)) |
2701 | madvise_thp (brk, size); |
2702 | LIBC_PROBE (memory_sbrk_more, 2, brk, size); |
2703 | } |
2704 | |
2705 | if (brk == (char *) (MORECORE_FAILURE)) |
2706 | { |
2707 | /* |
2708 | If have mmap, try using it as a backup when MORECORE fails or |
2709 | cannot be used. This is worth doing on systems that have "holes" in |
2710 | address space, so sbrk cannot extend to give contiguous space, but |
2711 | space is available elsewhere. Note that we ignore mmap max count |
2712 | and threshold limits, since the space will not be used as a |
2713 | segregated mmap region. |
2714 | */ |
2715 | |
2716 | char *mbrk = MAP_FAILED; |
2717 | if (mp_.hp_pagesize > 0) |
2718 | mbrk = sysmalloc_mmap_fallback (&size, nb, old_size, |
2719 | mp_.hp_pagesize, mp_.hp_pagesize, |
2720 | mp_.hp_flags, av); |
2721 | if (mbrk == MAP_FAILED) |
2722 | mbrk = sysmalloc_mmap_fallback (&size, nb, old_size, MMAP_AS_MORECORE_SIZE, |
2723 | pagesize, 0, av); |
2724 | if (mbrk != MAP_FAILED) |
2725 | { |
2726 | /* We do not need, and cannot use, another sbrk call to find end */ |
2727 | brk = mbrk; |
2728 | snd_brk = brk + size; |
2729 | } |
2730 | } |
2731 | |
2732 | if (brk != (char *) (MORECORE_FAILURE)) |
2733 | { |
2734 | if (mp_.sbrk_base == 0) |
2735 | mp_.sbrk_base = brk; |
2736 | av->system_mem += size; |
2737 | |
2738 | /* |
2739 | If MORECORE extends previous space, we can likewise extend top size. |
2740 | */ |
2741 | |
2742 | if (brk == old_end && snd_brk == (char *) (MORECORE_FAILURE)) |
2743 | set_head (old_top, (size + old_size) | PREV_INUSE); |
2744 | |
2745 | else if (contiguous (av) && old_size && brk < old_end) |
2746 | /* Oops! Someone else killed our space.. Can't touch anything. */ |
2747 | malloc_printerr ("break adjusted to free malloc space" ); |
2748 | |
2749 | /* |
2750 | Otherwise, make adjustments: |
2751 | |
2752 | * If the first time through or noncontiguous, we need to call sbrk |
2753 | just to find out where the end of memory lies. |
2754 | |
2755 | * We need to ensure that all returned chunks from malloc will meet |
2756 | MALLOC_ALIGNMENT |
2757 | |
2758 | * If there was an intervening foreign sbrk, we need to adjust sbrk |
2759 | request size to account for fact that we will not be able to |
2760 | combine new space with existing space in old_top. |
2761 | |
2762 | * Almost all systems internally allocate whole pages at a time, in |
2763 | which case we might as well use the whole last page of request. |
2764 | So we allocate enough more memory to hit a page boundary now, |
2765 | which in turn causes future contiguous calls to page-align. |
2766 | */ |
2767 | |
2768 | else |
2769 | { |
2770 | front_misalign = 0; |
2771 | end_misalign = 0; |
2772 | correction = 0; |
2773 | aligned_brk = brk; |
2774 | |
2775 | /* handle contiguous cases */ |
2776 | if (contiguous (av)) |
2777 | { |
2778 | /* Count foreign sbrk as system_mem. */ |
2779 | if (old_size) |
2780 | av->system_mem += brk - old_end; |
2781 | |
2782 | /* Guarantee alignment of first new chunk made from this space */ |
2783 | |
2784 | front_misalign = (INTERNAL_SIZE_T) chunk2mem (brk) & MALLOC_ALIGN_MASK; |
2785 | if (front_misalign > 0) |
2786 | { |
2787 | /* |
2788 | Skip over some bytes to arrive at an aligned position. |
2789 | We don't need to specially mark these wasted front bytes. |
2790 | They will never be accessed anyway because |
2791 | prev_inuse of av->top (and any chunk created from its start) |
2792 | is always true after initialization. |
2793 | */ |
2794 | |
2795 | correction = MALLOC_ALIGNMENT - front_misalign; |
2796 | aligned_brk += correction; |
2797 | } |
2798 | |
2799 | /* |
2800 | If this isn't adjacent to existing space, then we will not |
2801 | be able to merge with old_top space, so must add to 2nd request. |
2802 | */ |
2803 | |
2804 | correction += old_size; |
2805 | |
2806 | /* Extend the end address to hit a page boundary */ |
2807 | end_misalign = (INTERNAL_SIZE_T) (brk + size + correction); |
2808 | correction += (ALIGN_UP (end_misalign, pagesize)) - end_misalign; |
2809 | |
2810 | assert (correction >= 0); |
2811 | snd_brk = (char *) (MORECORE (correction)); |
2812 | |
2813 | /* |
2814 | If can't allocate correction, try to at least find out current |
2815 | brk. It might be enough to proceed without failing. |
2816 | |
2817 | Note that if second sbrk did NOT fail, we assume that space |
2818 | is contiguous with first sbrk. This is a safe assumption unless |
2819 | program is multithreaded but doesn't use locks and a foreign sbrk |
2820 | occurred between our first and second calls. |
2821 | */ |
2822 | |
2823 | if (snd_brk == (char *) (MORECORE_FAILURE)) |
2824 | { |
2825 | correction = 0; |
2826 | snd_brk = (char *) (MORECORE (0)); |
2827 | } |
2828 | else |
2829 | madvise_thp (snd_brk, correction); |
2830 | } |
2831 | |
2832 | /* handle non-contiguous cases */ |
2833 | else |
2834 | { |
2835 | if (MALLOC_ALIGNMENT == CHUNK_HDR_SZ) |
2836 | /* MORECORE/mmap must correctly align */ |
2837 | assert (((unsigned long) chunk2mem (brk) & MALLOC_ALIGN_MASK) == 0); |
2838 | else |
2839 | { |
2840 | front_misalign = (INTERNAL_SIZE_T) chunk2mem (brk) & MALLOC_ALIGN_MASK; |
2841 | if (front_misalign > 0) |
2842 | { |
2843 | /* |
2844 | Skip over some bytes to arrive at an aligned position. |
2845 | We don't need to specially mark these wasted front bytes. |
2846 | They will never be accessed anyway because |
2847 | prev_inuse of av->top (and any chunk created from its start) |
2848 | is always true after initialization. |
2849 | */ |
2850 | |
2851 | aligned_brk += MALLOC_ALIGNMENT - front_misalign; |
2852 | } |
2853 | } |
2854 | |
2855 | /* Find out current end of memory */ |
2856 | if (snd_brk == (char *) (MORECORE_FAILURE)) |
2857 | { |
2858 | snd_brk = (char *) (MORECORE (0)); |
2859 | } |
2860 | } |
2861 | |
2862 | /* Adjust top based on results of second sbrk */ |
2863 | if (snd_brk != (char *) (MORECORE_FAILURE)) |
2864 | { |
2865 | av->top = (mchunkptr) aligned_brk; |
2866 | set_head (av->top, (snd_brk - aligned_brk + correction) | PREV_INUSE); |
2867 | av->system_mem += correction; |
2868 | |
2869 | /* |
2870 | If not the first time through, we either have a |
2871 | gap due to foreign sbrk or a non-contiguous region. Insert a |
2872 | double fencepost at old_top to prevent consolidation with space |
2873 | we don't own. These fenceposts are artificial chunks that are |
2874 | marked as inuse and are in any case too small to use. We need |
2875 | two to make sizes and alignments work out. |
2876 | */ |
2877 | |
2878 | if (old_size != 0) |
2879 | { |
2880 | /* |
2881 | Shrink old_top to insert fenceposts, keeping size a |
2882 | multiple of MALLOC_ALIGNMENT. We know there is at least |
2883 | enough space in old_top to do this. |
2884 | */ |
2885 | old_size = (old_size - 2 * CHUNK_HDR_SZ) & ~MALLOC_ALIGN_MASK; |
2886 | set_head (old_top, old_size | PREV_INUSE); |
2887 | |
2888 | /* |
2889 | Note that the following assignments completely overwrite |
2890 | old_top when old_size was previously MINSIZE. This is |
2891 | intentional. We need the fencepost, even if old_top otherwise gets |
2892 | lost. |
2893 | */ |
2894 | set_head (chunk_at_offset (old_top, old_size), |
2895 | CHUNK_HDR_SZ | PREV_INUSE); |
2896 | set_head (chunk_at_offset (old_top, |
2897 | old_size + CHUNK_HDR_SZ), |
2898 | CHUNK_HDR_SZ | PREV_INUSE); |
2899 | |
2900 | /* If possible, release the rest. */ |
2901 | if (old_size >= MINSIZE) |
2902 | { |
2903 | _int_free (av, old_top, 1); |
2904 | } |
2905 | } |
2906 | } |
2907 | } |
2908 | } |
2909 | } /* if (av != &main_arena) */ |
2910 | |
2911 | if ((unsigned long) av->system_mem > (unsigned long) (av->max_system_mem)) |
2912 | av->max_system_mem = av->system_mem; |
2913 | check_malloc_state (av); |
2914 | |
2915 | /* finally, do the allocation */ |
2916 | p = av->top; |
2917 | size = chunksize (p); |
2918 | |
2919 | /* check that one of the above allocation paths succeeded */ |
2920 | if ((unsigned long) (size) >= (unsigned long) (nb + MINSIZE)) |
2921 | { |
2922 | remainder_size = size - nb; |
2923 | remainder = chunk_at_offset (p, nb); |
2924 | av->top = remainder; |
2925 | set_head (p, nb | PREV_INUSE | (av != &main_arena ? NON_MAIN_ARENA : 0)); |
2926 | set_head (remainder, remainder_size | PREV_INUSE); |
2927 | check_malloced_chunk (av, p, nb); |
2928 | return chunk2mem (p); |
2929 | } |
2930 | |
2931 | /* catch all failure paths */ |
2932 | __set_errno (ENOMEM); |
2933 | return 0; |
2934 | } |
2935 | |
2936 | |
2937 | /* |
2938 | systrim is an inverse of sorts to sysmalloc. It gives memory back |
2939 | to the system (via negative arguments to sbrk) if there is unused |
2940 | memory at the `high' end of the malloc pool. It is called |
2941 | automatically by free() when top space exceeds the trim |
2942 | threshold. It is also called by the public malloc_trim routine. It |
2943 | returns 1 if it actually released any memory, else 0. |
2944 | */ |
2945 | |
2946 | static int |
2947 | systrim (size_t pad, mstate av) |
2948 | { |
2949 | long top_size; /* Amount of top-most memory */ |
2950 | long ; /* Amount to release */ |
2951 | long released; /* Amount actually released */ |
2952 | char *current_brk; /* address returned by pre-check sbrk call */ |
2953 | char *new_brk; /* address returned by post-check sbrk call */ |
2954 | long top_area; |
2955 | |
2956 | top_size = chunksize (av->top); |
2957 | |
2958 | top_area = top_size - MINSIZE - 1; |
2959 | if (top_area <= pad) |
2960 | return 0; |
2961 | |
2962 | /* Release in pagesize units and round down to the nearest page. */ |
2963 | #ifdef MADV_HUGEPAGE |
2964 | if (__glibc_unlikely (mp_.thp_pagesize != 0)) |
2965 | extra = ALIGN_DOWN (top_area - pad, mp_.thp_pagesize); |
2966 | else |
2967 | #endif |
2968 | extra = ALIGN_DOWN (top_area - pad, GLRO(dl_pagesize)); |
2969 | |
2970 | if (extra == 0) |
2971 | return 0; |
2972 | |
2973 | /* |
2974 | Only proceed if end of memory is where we last set it. |
2975 | This avoids problems if there were foreign sbrk calls. |
2976 | */ |
2977 | current_brk = (char *) (MORECORE (0)); |
2978 | if (current_brk == (char *) (av->top) + top_size) |
2979 | { |
2980 | /* |
2981 | Attempt to release memory. We ignore MORECORE return value, |
2982 | and instead call again to find out where new end of memory is. |
2983 | This avoids problems if first call releases less than we asked, |
2984 | of if failure somehow altered brk value. (We could still |
2985 | encounter problems if it altered brk in some very bad way, |
2986 | but the only thing we can do is adjust anyway, which will cause |
2987 | some downstream failure.) |
2988 | */ |
2989 | |
2990 | MORECORE (-extra); |
2991 | new_brk = (char *) (MORECORE (0)); |
2992 | |
2993 | LIBC_PROBE (memory_sbrk_less, 2, new_brk, extra); |
2994 | |
2995 | if (new_brk != (char *) MORECORE_FAILURE) |
2996 | { |
2997 | released = (long) (current_brk - new_brk); |
2998 | |
2999 | if (released != 0) |
3000 | { |
3001 | /* Success. Adjust top. */ |
3002 | av->system_mem -= released; |
3003 | set_head (av->top, (top_size - released) | PREV_INUSE); |
3004 | check_malloc_state (av); |
3005 | return 1; |
3006 | } |
3007 | } |
3008 | } |
3009 | return 0; |
3010 | } |
3011 | |
3012 | static void |
3013 | munmap_chunk (mchunkptr p) |
3014 | { |
3015 | size_t pagesize = GLRO (dl_pagesize); |
3016 | INTERNAL_SIZE_T size = chunksize (p); |
3017 | |
3018 | assert (chunk_is_mmapped (p)); |
3019 | |
3020 | uintptr_t mem = (uintptr_t) chunk2mem (p); |
3021 | uintptr_t block = (uintptr_t) p - prev_size (p); |
3022 | size_t total_size = prev_size (p) + size; |
3023 | /* Unfortunately we have to do the compilers job by hand here. Normally |
3024 | we would test BLOCK and TOTAL-SIZE separately for compliance with the |
3025 | page size. But gcc does not recognize the optimization possibility |
3026 | (in the moment at least) so we combine the two values into one before |
3027 | the bit test. */ |
3028 | if (__glibc_unlikely ((block | total_size) & (pagesize - 1)) != 0 |
3029 | || __glibc_unlikely (!powerof2 (mem & (pagesize - 1)))) |
3030 | malloc_printerr ("munmap_chunk(): invalid pointer" ); |
3031 | |
3032 | atomic_fetch_add_relaxed (&mp_.n_mmaps, -1); |
3033 | atomic_fetch_add_relaxed (&mp_.mmapped_mem, -total_size); |
3034 | |
3035 | /* If munmap failed the process virtual memory address space is in a |
3036 | bad shape. Just leave the block hanging around, the process will |
3037 | terminate shortly anyway since not much can be done. */ |
3038 | __munmap ((char *) block, total_size); |
3039 | } |
3040 | |
3041 | #if HAVE_MREMAP |
3042 | |
3043 | static mchunkptr |
3044 | mremap_chunk (mchunkptr p, size_t new_size) |
3045 | { |
3046 | size_t pagesize = GLRO (dl_pagesize); |
3047 | INTERNAL_SIZE_T offset = prev_size (p); |
3048 | INTERNAL_SIZE_T size = chunksize (p); |
3049 | char *cp; |
3050 | |
3051 | assert (chunk_is_mmapped (p)); |
3052 | |
3053 | uintptr_t block = (uintptr_t) p - offset; |
3054 | uintptr_t mem = (uintptr_t) chunk2mem(p); |
3055 | size_t total_size = offset + size; |
3056 | if (__glibc_unlikely ((block | total_size) & (pagesize - 1)) != 0 |
3057 | || __glibc_unlikely (!powerof2 (mem & (pagesize - 1)))) |
3058 | malloc_printerr("mremap_chunk(): invalid pointer" ); |
3059 | |
3060 | /* Note the extra SIZE_SZ overhead as in mmap_chunk(). */ |
3061 | new_size = ALIGN_UP (new_size + offset + SIZE_SZ, pagesize); |
3062 | |
3063 | /* No need to remap if the number of pages does not change. */ |
3064 | if (total_size == new_size) |
3065 | return p; |
3066 | |
3067 | cp = (char *) __mremap ((char *) block, total_size, new_size, |
3068 | MREMAP_MAYMOVE); |
3069 | |
3070 | if (cp == MAP_FAILED) |
3071 | return 0; |
3072 | |
3073 | madvise_thp (cp, new_size); |
3074 | |
3075 | p = (mchunkptr) (cp + offset); |
3076 | |
3077 | assert (aligned_OK (chunk2mem (p))); |
3078 | |
3079 | assert (prev_size (p) == offset); |
3080 | set_head (p, (new_size - offset) | IS_MMAPPED); |
3081 | |
3082 | INTERNAL_SIZE_T new; |
3083 | new = atomic_fetch_add_relaxed (&mp_.mmapped_mem, new_size - size - offset) |
3084 | + new_size - size - offset; |
3085 | atomic_max (&mp_.max_mmapped_mem, new); |
3086 | return p; |
3087 | } |
3088 | #endif /* HAVE_MREMAP */ |
3089 | |
3090 | /*------------------------ Public wrappers. --------------------------------*/ |
3091 | |
3092 | #if USE_TCACHE |
3093 | |
3094 | /* We overlay this structure on the user-data portion of a chunk when |
3095 | the chunk is stored in the per-thread cache. */ |
3096 | typedef struct tcache_entry |
3097 | { |
3098 | struct tcache_entry *next; |
3099 | /* This field exists to detect double frees. */ |
3100 | uintptr_t key; |
3101 | } tcache_entry; |
3102 | |
3103 | /* There is one of these for each thread, which contains the |
3104 | per-thread cache (hence "tcache_perthread_struct"). Keeping |
3105 | overall size low is mildly important. Note that COUNTS and ENTRIES |
3106 | are redundant (we could have just counted the linked list each |
3107 | time), this is for performance reasons. */ |
3108 | typedef struct tcache_perthread_struct |
3109 | { |
3110 | uint16_t counts[TCACHE_MAX_BINS]; |
3111 | tcache_entry *entries[TCACHE_MAX_BINS]; |
3112 | } tcache_perthread_struct; |
3113 | |
3114 | static __thread bool tcache_shutting_down = false; |
3115 | static __thread tcache_perthread_struct *tcache = NULL; |
3116 | |
3117 | /* Process-wide key to try and catch a double-free in the same thread. */ |
3118 | static uintptr_t tcache_key; |
3119 | |
3120 | /* The value of tcache_key does not really have to be a cryptographically |
3121 | secure random number. It only needs to be arbitrary enough so that it does |
3122 | not collide with values present in applications. If a collision does happen |
3123 | consistently enough, it could cause a degradation in performance since the |
3124 | entire list is checked to check if the block indeed has been freed the |
3125 | second time. The odds of this happening are exceedingly low though, about 1 |
3126 | in 2^wordsize. There is probably a higher chance of the performance |
3127 | degradation being due to a double free where the first free happened in a |
3128 | different thread; that's a case this check does not cover. */ |
3129 | static void |
3130 | tcache_key_initialize (void) |
3131 | { |
3132 | if (__getrandom_nocancel (&tcache_key, sizeof(tcache_key), GRND_NONBLOCK) |
3133 | != sizeof (tcache_key)) |
3134 | { |
3135 | tcache_key = random_bits (); |
3136 | #if __WORDSIZE == 64 |
3137 | tcache_key = (tcache_key << 32) | random_bits (); |
3138 | #endif |
3139 | } |
3140 | } |
3141 | |
3142 | /* Caller must ensure that we know tc_idx is valid and there's room |
3143 | for more chunks. */ |
3144 | static __always_inline void |
3145 | tcache_put (mchunkptr chunk, size_t tc_idx) |
3146 | { |
3147 | tcache_entry *e = (tcache_entry *) chunk2mem (chunk); |
3148 | |
3149 | /* Mark this chunk as "in the tcache" so the test in _int_free will |
3150 | detect a double free. */ |
3151 | e->key = tcache_key; |
3152 | |
3153 | e->next = PROTECT_PTR (&e->next, tcache->entries[tc_idx]); |
3154 | tcache->entries[tc_idx] = e; |
3155 | ++(tcache->counts[tc_idx]); |
3156 | } |
3157 | |
3158 | /* Caller must ensure that we know tc_idx is valid and there's |
3159 | available chunks to remove. Removes chunk from the middle of the |
3160 | list. */ |
3161 | static __always_inline void * |
3162 | tcache_get_n (size_t tc_idx, tcache_entry **ep) |
3163 | { |
3164 | tcache_entry *e; |
3165 | if (ep == &(tcache->entries[tc_idx])) |
3166 | e = *ep; |
3167 | else |
3168 | e = REVEAL_PTR (*ep); |
3169 | |
3170 | if (__glibc_unlikely (!aligned_OK (e))) |
3171 | malloc_printerr ("malloc(): unaligned tcache chunk detected" ); |
3172 | |
3173 | if (ep == &(tcache->entries[tc_idx])) |
3174 | *ep = REVEAL_PTR (e->next); |
3175 | else |
3176 | *ep = PROTECT_PTR (ep, REVEAL_PTR (e->next)); |
3177 | |
3178 | --(tcache->counts[tc_idx]); |
3179 | e->key = 0; |
3180 | return (void *) e; |
3181 | } |
3182 | |
3183 | /* Like the above, but removes from the head of the list. */ |
3184 | static __always_inline void * |
3185 | tcache_get (size_t tc_idx) |
3186 | { |
3187 | return tcache_get_n (tc_idx, & tcache->entries[tc_idx]); |
3188 | } |
3189 | |
3190 | /* Iterates through the tcache linked list. */ |
3191 | static __always_inline tcache_entry * |
3192 | tcache_next (tcache_entry *e) |
3193 | { |
3194 | return (tcache_entry *) REVEAL_PTR (e->next); |
3195 | } |
3196 | |
3197 | static void |
3198 | tcache_thread_shutdown (void) |
3199 | { |
3200 | int i; |
3201 | tcache_perthread_struct *tcache_tmp = tcache; |
3202 | |
3203 | tcache_shutting_down = true; |
3204 | |
3205 | if (!tcache) |
3206 | return; |
3207 | |
3208 | /* Disable the tcache and prevent it from being reinitialized. */ |
3209 | tcache = NULL; |
3210 | |
3211 | /* Free all of the entries and the tcache itself back to the arena |
3212 | heap for coalescing. */ |
3213 | for (i = 0; i < TCACHE_MAX_BINS; ++i) |
3214 | { |
3215 | while (tcache_tmp->entries[i]) |
3216 | { |
3217 | tcache_entry *e = tcache_tmp->entries[i]; |
3218 | if (__glibc_unlikely (!aligned_OK (e))) |
3219 | malloc_printerr ("tcache_thread_shutdown(): " |
3220 | "unaligned tcache chunk detected" ); |
3221 | tcache_tmp->entries[i] = REVEAL_PTR (e->next); |
3222 | __libc_free (e); |
3223 | } |
3224 | } |
3225 | |
3226 | __libc_free (tcache_tmp); |
3227 | } |
3228 | |
3229 | static void |
3230 | tcache_init(void) |
3231 | { |
3232 | mstate ar_ptr; |
3233 | void *victim = 0; |
3234 | const size_t bytes = sizeof (tcache_perthread_struct); |
3235 | |
3236 | if (tcache_shutting_down) |
3237 | return; |
3238 | |
3239 | arena_get (ar_ptr, bytes); |
3240 | victim = _int_malloc (ar_ptr, bytes); |
3241 | if (!victim && ar_ptr != NULL) |
3242 | { |
3243 | ar_ptr = arena_get_retry (ar_ptr, bytes); |
3244 | victim = _int_malloc (ar_ptr, bytes); |
3245 | } |
3246 | |
3247 | |
3248 | if (ar_ptr != NULL) |
3249 | __libc_lock_unlock (ar_ptr->mutex); |
3250 | |
3251 | /* In a low memory situation, we may not be able to allocate memory |
3252 | - in which case, we just keep trying later. However, we |
3253 | typically do this very early, so either there is sufficient |
3254 | memory, or there isn't enough memory to do non-trivial |
3255 | allocations anyway. */ |
3256 | if (victim) |
3257 | { |
3258 | tcache = (tcache_perthread_struct *) victim; |
3259 | memset (tcache, 0, sizeof (tcache_perthread_struct)); |
3260 | } |
3261 | |
3262 | } |
3263 | |
3264 | # define MAYBE_INIT_TCACHE() \ |
3265 | if (__glibc_unlikely (tcache == NULL)) \ |
3266 | tcache_init(); |
3267 | |
3268 | #else /* !USE_TCACHE */ |
3269 | # define MAYBE_INIT_TCACHE() |
3270 | |
3271 | static void |
3272 | tcache_thread_shutdown (void) |
3273 | { |
3274 | /* Nothing to do if there is no thread cache. */ |
3275 | } |
3276 | |
3277 | #endif /* !USE_TCACHE */ |
3278 | |
3279 | #if IS_IN (libc) |
3280 | void * |
3281 | __libc_malloc (size_t bytes) |
3282 | { |
3283 | mstate ar_ptr; |
3284 | void *victim; |
3285 | |
3286 | _Static_assert (PTRDIFF_MAX <= SIZE_MAX / 2, |
3287 | "PTRDIFF_MAX is not more than half of SIZE_MAX" ); |
3288 | |
3289 | if (!__malloc_initialized) |
3290 | ptmalloc_init (); |
3291 | #if USE_TCACHE |
3292 | /* int_free also calls request2size, be careful to not pad twice. */ |
3293 | size_t tbytes = checked_request2size (bytes); |
3294 | if (tbytes == 0) |
3295 | { |
3296 | __set_errno (ENOMEM); |
3297 | return NULL; |
3298 | } |
3299 | size_t tc_idx = csize2tidx (tbytes); |
3300 | |
3301 | MAYBE_INIT_TCACHE (); |
3302 | |
3303 | DIAG_PUSH_NEEDS_COMMENT; |
3304 | if (tc_idx < mp_.tcache_bins |
3305 | && tcache != NULL |
3306 | && tcache->counts[tc_idx] > 0) |
3307 | { |
3308 | victim = tcache_get (tc_idx); |
3309 | return tag_new_usable (victim); |
3310 | } |
3311 | DIAG_POP_NEEDS_COMMENT; |
3312 | #endif |
3313 | |
3314 | if (SINGLE_THREAD_P) |
3315 | { |
3316 | victim = tag_new_usable (_int_malloc (&main_arena, bytes)); |
3317 | assert (!victim || chunk_is_mmapped (mem2chunk (victim)) || |
3318 | &main_arena == arena_for_chunk (mem2chunk (victim))); |
3319 | return victim; |
3320 | } |
3321 | |
3322 | arena_get (ar_ptr, bytes); |
3323 | |
3324 | victim = _int_malloc (ar_ptr, bytes); |
3325 | /* Retry with another arena only if we were able to find a usable arena |
3326 | before. */ |
3327 | if (!victim && ar_ptr != NULL) |
3328 | { |
3329 | LIBC_PROBE (memory_malloc_retry, 1, bytes); |
3330 | ar_ptr = arena_get_retry (ar_ptr, bytes); |
3331 | victim = _int_malloc (ar_ptr, bytes); |
3332 | } |
3333 | |
3334 | if (ar_ptr != NULL) |
3335 | __libc_lock_unlock (ar_ptr->mutex); |
3336 | |
3337 | victim = tag_new_usable (victim); |
3338 | |
3339 | assert (!victim || chunk_is_mmapped (mem2chunk (victim)) || |
3340 | ar_ptr == arena_for_chunk (mem2chunk (victim))); |
3341 | return victim; |
3342 | } |
3343 | libc_hidden_def (__libc_malloc) |
3344 | |
3345 | void |
3346 | __libc_free (void *mem) |
3347 | { |
3348 | mstate ar_ptr; |
3349 | mchunkptr p; /* chunk corresponding to mem */ |
3350 | |
3351 | if (mem == 0) /* free(0) has no effect */ |
3352 | return; |
3353 | |
3354 | /* Quickly check that the freed pointer matches the tag for the memory. |
3355 | This gives a useful double-free detection. */ |
3356 | if (__glibc_unlikely (mtag_enabled)) |
3357 | *(volatile char *)mem; |
3358 | |
3359 | int err = errno; |
3360 | |
3361 | p = mem2chunk (mem); |
3362 | |
3363 | if (chunk_is_mmapped (p)) /* release mmapped memory. */ |
3364 | { |
3365 | /* See if the dynamic brk/mmap threshold needs adjusting. |
3366 | Dumped fake mmapped chunks do not affect the threshold. */ |
3367 | if (!mp_.no_dyn_threshold |
3368 | && chunksize_nomask (p) > mp_.mmap_threshold |
3369 | && chunksize_nomask (p) <= DEFAULT_MMAP_THRESHOLD_MAX) |
3370 | { |
3371 | mp_.mmap_threshold = chunksize (p); |
3372 | mp_.trim_threshold = 2 * mp_.mmap_threshold; |
3373 | LIBC_PROBE (memory_mallopt_free_dyn_thresholds, 2, |
3374 | mp_.mmap_threshold, mp_.trim_threshold); |
3375 | } |
3376 | munmap_chunk (p); |
3377 | } |
3378 | else |
3379 | { |
3380 | MAYBE_INIT_TCACHE (); |
3381 | |
3382 | /* Mark the chunk as belonging to the library again. */ |
3383 | (void)tag_region (chunk2mem (p), memsize (p)); |
3384 | |
3385 | ar_ptr = arena_for_chunk (p); |
3386 | _int_free (ar_ptr, p, 0); |
3387 | } |
3388 | |
3389 | __set_errno (err); |
3390 | } |
3391 | libc_hidden_def (__libc_free) |
3392 | |
3393 | void * |
3394 | __libc_realloc (void *oldmem, size_t bytes) |
3395 | { |
3396 | mstate ar_ptr; |
3397 | INTERNAL_SIZE_T nb; /* padded request size */ |
3398 | |
3399 | void *newp; /* chunk to return */ |
3400 | |
3401 | if (!__malloc_initialized) |
3402 | ptmalloc_init (); |
3403 | |
3404 | #if REALLOC_ZERO_BYTES_FREES |
3405 | if (bytes == 0 && oldmem != NULL) |
3406 | { |
3407 | __libc_free (oldmem); return 0; |
3408 | } |
3409 | #endif |
3410 | |
3411 | /* realloc of null is supposed to be same as malloc */ |
3412 | if (oldmem == 0) |
3413 | return __libc_malloc (bytes); |
3414 | |
3415 | /* Perform a quick check to ensure that the pointer's tag matches the |
3416 | memory's tag. */ |
3417 | if (__glibc_unlikely (mtag_enabled)) |
3418 | *(volatile char*) oldmem; |
3419 | |
3420 | /* chunk corresponding to oldmem */ |
3421 | const mchunkptr oldp = mem2chunk (oldmem); |
3422 | |
3423 | /* Return the chunk as is if the request grows within usable bytes, typically |
3424 | into the alignment padding. We want to avoid reusing the block for |
3425 | shrinkages because it ends up unnecessarily fragmenting the address space. |
3426 | This is also why the heuristic misses alignment padding for THP for |
3427 | now. */ |
3428 | size_t usable = musable (oldmem); |
3429 | if (bytes <= usable) |
3430 | { |
3431 | size_t difference = usable - bytes; |
3432 | if ((unsigned long) difference < 2 * sizeof (INTERNAL_SIZE_T) |
3433 | || (chunk_is_mmapped (oldp) && difference <= GLRO (dl_pagesize))) |
3434 | return oldmem; |
3435 | } |
3436 | |
3437 | /* its size */ |
3438 | const INTERNAL_SIZE_T oldsize = chunksize (oldp); |
3439 | |
3440 | if (chunk_is_mmapped (oldp)) |
3441 | ar_ptr = NULL; |
3442 | else |
3443 | { |
3444 | MAYBE_INIT_TCACHE (); |
3445 | ar_ptr = arena_for_chunk (oldp); |
3446 | } |
3447 | |
3448 | /* Little security check which won't hurt performance: the allocator |
3449 | never wraps around at the end of the address space. Therefore |
3450 | we can exclude some size values which might appear here by |
3451 | accident or by "design" from some intruder. */ |
3452 | if ((__builtin_expect ((uintptr_t) oldp > (uintptr_t) -oldsize, 0) |
3453 | || __builtin_expect (misaligned_chunk (oldp), 0))) |
3454 | malloc_printerr ("realloc(): invalid pointer" ); |
3455 | |
3456 | nb = checked_request2size (bytes); |
3457 | if (nb == 0) |
3458 | { |
3459 | __set_errno (ENOMEM); |
3460 | return NULL; |
3461 | } |
3462 | |
3463 | if (chunk_is_mmapped (oldp)) |
3464 | { |
3465 | void *newmem; |
3466 | |
3467 | #if HAVE_MREMAP |
3468 | newp = mremap_chunk (oldp, nb); |
3469 | if (newp) |
3470 | { |
3471 | void *newmem = chunk2mem_tag (newp); |
3472 | /* Give the new block a different tag. This helps to ensure |
3473 | that stale handles to the previous mapping are not |
3474 | reused. There's a performance hit for both us and the |
3475 | caller for doing this, so we might want to |
3476 | reconsider. */ |
3477 | return tag_new_usable (newmem); |
3478 | } |
3479 | #endif |
3480 | /* Note the extra SIZE_SZ overhead. */ |
3481 | if (oldsize - SIZE_SZ >= nb) |
3482 | return oldmem; /* do nothing */ |
3483 | |
3484 | /* Must alloc, copy, free. */ |
3485 | newmem = __libc_malloc (bytes); |
3486 | if (newmem == 0) |
3487 | return 0; /* propagate failure */ |
3488 | |
3489 | memcpy (newmem, oldmem, oldsize - CHUNK_HDR_SZ); |
3490 | munmap_chunk (oldp); |
3491 | return newmem; |
3492 | } |
3493 | |
3494 | if (SINGLE_THREAD_P) |
3495 | { |
3496 | newp = _int_realloc (ar_ptr, oldp, oldsize, nb); |
3497 | assert (!newp || chunk_is_mmapped (mem2chunk (newp)) || |
3498 | ar_ptr == arena_for_chunk (mem2chunk (newp))); |
3499 | |
3500 | return newp; |
3501 | } |
3502 | |
3503 | __libc_lock_lock (ar_ptr->mutex); |
3504 | |
3505 | newp = _int_realloc (ar_ptr, oldp, oldsize, nb); |
3506 | |
3507 | __libc_lock_unlock (ar_ptr->mutex); |
3508 | assert (!newp || chunk_is_mmapped (mem2chunk (newp)) || |
3509 | ar_ptr == arena_for_chunk (mem2chunk (newp))); |
3510 | |
3511 | if (newp == NULL) |
3512 | { |
3513 | /* Try harder to allocate memory in other arenas. */ |
3514 | LIBC_PROBE (memory_realloc_retry, 2, bytes, oldmem); |
3515 | newp = __libc_malloc (bytes); |
3516 | if (newp != NULL) |
3517 | { |
3518 | size_t sz = memsize (oldp); |
3519 | memcpy (newp, oldmem, sz); |
3520 | (void) tag_region (chunk2mem (oldp), sz); |
3521 | _int_free (ar_ptr, oldp, 0); |
3522 | } |
3523 | } |
3524 | |
3525 | return newp; |
3526 | } |
3527 | libc_hidden_def (__libc_realloc) |
3528 | |
3529 | void * |
3530 | __libc_memalign (size_t alignment, size_t bytes) |
3531 | { |
3532 | if (!__malloc_initialized) |
3533 | ptmalloc_init (); |
3534 | |
3535 | void *address = RETURN_ADDRESS (0); |
3536 | return _mid_memalign (alignment, bytes, address); |
3537 | } |
3538 | libc_hidden_def (__libc_memalign) |
3539 | |
3540 | /* For ISO C17. */ |
3541 | void * |
3542 | weak_function |
3543 | aligned_alloc (size_t alignment, size_t bytes) |
3544 | { |
3545 | if (!__malloc_initialized) |
3546 | ptmalloc_init (); |
3547 | |
3548 | /* Similar to memalign, but starting with ISO C17 the standard |
3549 | requires an error for alignments that are not supported by the |
3550 | implementation. Valid alignments for the current implementation |
3551 | are non-negative powers of two. */ |
3552 | if (!powerof2 (alignment) || alignment == 0) |
3553 | { |
3554 | __set_errno (EINVAL); |
3555 | return 0; |
3556 | } |
3557 | |
3558 | void *address = RETURN_ADDRESS (0); |
3559 | return _mid_memalign (alignment, bytes, address); |
3560 | } |
3561 | |
3562 | static void * |
3563 | _mid_memalign (size_t alignment, size_t bytes, void *address) |
3564 | { |
3565 | mstate ar_ptr; |
3566 | void *p; |
3567 | |
3568 | /* If we need less alignment than we give anyway, just relay to malloc. */ |
3569 | if (alignment <= MALLOC_ALIGNMENT) |
3570 | return __libc_malloc (bytes); |
3571 | |
3572 | /* Otherwise, ensure that it is at least a minimum chunk size */ |
3573 | if (alignment < MINSIZE) |
3574 | alignment = MINSIZE; |
3575 | |
3576 | /* If the alignment is greater than SIZE_MAX / 2 + 1 it cannot be a |
3577 | power of 2 and will cause overflow in the check below. */ |
3578 | if (alignment > SIZE_MAX / 2 + 1) |
3579 | { |
3580 | __set_errno (EINVAL); |
3581 | return 0; |
3582 | } |
3583 | |
3584 | |
3585 | /* Make sure alignment is power of 2. */ |
3586 | if (!powerof2 (alignment)) |
3587 | { |
3588 | size_t a = MALLOC_ALIGNMENT * 2; |
3589 | while (a < alignment) |
3590 | a <<= 1; |
3591 | alignment = a; |
3592 | } |
3593 | |
3594 | #if USE_TCACHE |
3595 | { |
3596 | size_t tbytes; |
3597 | tbytes = checked_request2size (bytes); |
3598 | if (tbytes == 0) |
3599 | { |
3600 | __set_errno (ENOMEM); |
3601 | return NULL; |
3602 | } |
3603 | size_t tc_idx = csize2tidx (tbytes); |
3604 | |
3605 | if (tc_idx < mp_.tcache_bins |
3606 | && tcache != NULL |
3607 | && tcache->counts[tc_idx] > 0) |
3608 | { |
3609 | /* The tcache itself isn't encoded, but the chain is. */ |
3610 | tcache_entry **tep = & tcache->entries[tc_idx]; |
3611 | tcache_entry *te = *tep; |
3612 | while (te != NULL && !PTR_IS_ALIGNED (te, alignment)) |
3613 | { |
3614 | tep = & (te->next); |
3615 | te = tcache_next (te); |
3616 | } |
3617 | if (te != NULL) |
3618 | { |
3619 | void *victim = tcache_get_n (tc_idx, tep); |
3620 | return tag_new_usable (victim); |
3621 | } |
3622 | } |
3623 | } |
3624 | #endif |
3625 | |
3626 | if (SINGLE_THREAD_P) |
3627 | { |
3628 | p = _int_memalign (&main_arena, alignment, bytes); |
3629 | assert (!p || chunk_is_mmapped (mem2chunk (p)) || |
3630 | &main_arena == arena_for_chunk (mem2chunk (p))); |
3631 | return tag_new_usable (p); |
3632 | } |
3633 | |
3634 | arena_get (ar_ptr, bytes + alignment + MINSIZE); |
3635 | |
3636 | p = _int_memalign (ar_ptr, alignment, bytes); |
3637 | if (!p && ar_ptr != NULL) |
3638 | { |
3639 | LIBC_PROBE (memory_memalign_retry, 2, bytes, alignment); |
3640 | ar_ptr = arena_get_retry (ar_ptr, bytes); |
3641 | p = _int_memalign (ar_ptr, alignment, bytes); |
3642 | } |
3643 | |
3644 | if (ar_ptr != NULL) |
3645 | __libc_lock_unlock (ar_ptr->mutex); |
3646 | |
3647 | assert (!p || chunk_is_mmapped (mem2chunk (p)) || |
3648 | ar_ptr == arena_for_chunk (mem2chunk (p))); |
3649 | return tag_new_usable (p); |
3650 | } |
3651 | |
3652 | void * |
3653 | __libc_valloc (size_t bytes) |
3654 | { |
3655 | if (!__malloc_initialized) |
3656 | ptmalloc_init (); |
3657 | |
3658 | void *address = RETURN_ADDRESS (0); |
3659 | size_t pagesize = GLRO (dl_pagesize); |
3660 | return _mid_memalign (pagesize, bytes, address); |
3661 | } |
3662 | |
3663 | void * |
3664 | __libc_pvalloc (size_t bytes) |
3665 | { |
3666 | if (!__malloc_initialized) |
3667 | ptmalloc_init (); |
3668 | |
3669 | void *address = RETURN_ADDRESS (0); |
3670 | size_t pagesize = GLRO (dl_pagesize); |
3671 | size_t rounded_bytes; |
3672 | /* ALIGN_UP with overflow check. */ |
3673 | if (__glibc_unlikely (__builtin_add_overflow (bytes, |
3674 | pagesize - 1, |
3675 | &rounded_bytes))) |
3676 | { |
3677 | __set_errno (ENOMEM); |
3678 | return 0; |
3679 | } |
3680 | rounded_bytes = rounded_bytes & -(pagesize - 1); |
3681 | |
3682 | return _mid_memalign (pagesize, rounded_bytes, address); |
3683 | } |
3684 | |
3685 | void * |
3686 | __libc_calloc (size_t n, size_t elem_size) |
3687 | { |
3688 | mstate av; |
3689 | mchunkptr oldtop; |
3690 | INTERNAL_SIZE_T sz, oldtopsize; |
3691 | void *mem; |
3692 | unsigned long clearsize; |
3693 | unsigned long nclears; |
3694 | INTERNAL_SIZE_T *d; |
3695 | ptrdiff_t bytes; |
3696 | |
3697 | if (__glibc_unlikely (__builtin_mul_overflow (n, elem_size, &bytes))) |
3698 | { |
3699 | __set_errno (ENOMEM); |
3700 | return NULL; |
3701 | } |
3702 | |
3703 | sz = bytes; |
3704 | |
3705 | if (!__malloc_initialized) |
3706 | ptmalloc_init (); |
3707 | |
3708 | MAYBE_INIT_TCACHE (); |
3709 | |
3710 | if (SINGLE_THREAD_P) |
3711 | av = &main_arena; |
3712 | else |
3713 | arena_get (av, sz); |
3714 | |
3715 | if (av) |
3716 | { |
3717 | /* Check if we hand out the top chunk, in which case there may be no |
3718 | need to clear. */ |
3719 | #if MORECORE_CLEARS |
3720 | oldtop = top (av); |
3721 | oldtopsize = chunksize (top (av)); |
3722 | # if MORECORE_CLEARS < 2 |
3723 | /* Only newly allocated memory is guaranteed to be cleared. */ |
3724 | if (av == &main_arena && |
3725 | oldtopsize < mp_.sbrk_base + av->max_system_mem - (char *) oldtop) |
3726 | oldtopsize = (mp_.sbrk_base + av->max_system_mem - (char *) oldtop); |
3727 | # endif |
3728 | if (av != &main_arena) |
3729 | { |
3730 | heap_info *heap = heap_for_ptr (oldtop); |
3731 | if (oldtopsize < (char *) heap + heap->mprotect_size - (char *) oldtop) |
3732 | oldtopsize = (char *) heap + heap->mprotect_size - (char *) oldtop; |
3733 | } |
3734 | #endif |
3735 | } |
3736 | else |
3737 | { |
3738 | /* No usable arenas. */ |
3739 | oldtop = 0; |
3740 | oldtopsize = 0; |
3741 | } |
3742 | mem = _int_malloc (av, sz); |
3743 | |
3744 | assert (!mem || chunk_is_mmapped (mem2chunk (mem)) || |
3745 | av == arena_for_chunk (mem2chunk (mem))); |
3746 | |
3747 | if (!SINGLE_THREAD_P) |
3748 | { |
3749 | if (mem == 0 && av != NULL) |
3750 | { |
3751 | LIBC_PROBE (memory_calloc_retry, 1, sz); |
3752 | av = arena_get_retry (av, sz); |
3753 | mem = _int_malloc (av, sz); |
3754 | } |
3755 | |
3756 | if (av != NULL) |
3757 | __libc_lock_unlock (av->mutex); |
3758 | } |
3759 | |
3760 | /* Allocation failed even after a retry. */ |
3761 | if (mem == 0) |
3762 | return 0; |
3763 | |
3764 | mchunkptr p = mem2chunk (mem); |
3765 | |
3766 | /* If we are using memory tagging, then we need to set the tags |
3767 | regardless of MORECORE_CLEARS, so we zero the whole block while |
3768 | doing so. */ |
3769 | if (__glibc_unlikely (mtag_enabled)) |
3770 | return tag_new_zero_region (mem, memsize (p)); |
3771 | |
3772 | INTERNAL_SIZE_T csz = chunksize (p); |
3773 | |
3774 | /* Two optional cases in which clearing not necessary */ |
3775 | if (chunk_is_mmapped (p)) |
3776 | { |
3777 | if (__builtin_expect (perturb_byte, 0)) |
3778 | return memset (mem, 0, sz); |
3779 | |
3780 | return mem; |
3781 | } |
3782 | |
3783 | #if MORECORE_CLEARS |
3784 | if (perturb_byte == 0 && (p == oldtop && csz > oldtopsize)) |
3785 | { |
3786 | /* clear only the bytes from non-freshly-sbrked memory */ |
3787 | csz = oldtopsize; |
3788 | } |
3789 | #endif |
3790 | |
3791 | /* Unroll clear of <= 36 bytes (72 if 8byte sizes). We know that |
3792 | contents have an odd number of INTERNAL_SIZE_T-sized words; |
3793 | minimally 3. */ |
3794 | d = (INTERNAL_SIZE_T *) mem; |
3795 | clearsize = csz - SIZE_SZ; |
3796 | nclears = clearsize / sizeof (INTERNAL_SIZE_T); |
3797 | assert (nclears >= 3); |
3798 | |
3799 | if (nclears > 9) |
3800 | return memset (d, 0, clearsize); |
3801 | |
3802 | else |
3803 | { |
3804 | *(d + 0) = 0; |
3805 | *(d + 1) = 0; |
3806 | *(d + 2) = 0; |
3807 | if (nclears > 4) |
3808 | { |
3809 | *(d + 3) = 0; |
3810 | *(d + 4) = 0; |
3811 | if (nclears > 6) |
3812 | { |
3813 | *(d + 5) = 0; |
3814 | *(d + 6) = 0; |
3815 | if (nclears > 8) |
3816 | { |
3817 | *(d + 7) = 0; |
3818 | *(d + 8) = 0; |
3819 | } |
3820 | } |
3821 | } |
3822 | } |
3823 | |
3824 | return mem; |
3825 | } |
3826 | #endif /* IS_IN (libc) */ |
3827 | |
3828 | /* |
3829 | ------------------------------ malloc ------------------------------ |
3830 | */ |
3831 | |
3832 | static void * |
3833 | _int_malloc (mstate av, size_t bytes) |
3834 | { |
3835 | INTERNAL_SIZE_T nb; /* normalized request size */ |
3836 | unsigned int idx; /* associated bin index */ |
3837 | mbinptr bin; /* associated bin */ |
3838 | |
3839 | mchunkptr victim; /* inspected/selected chunk */ |
3840 | INTERNAL_SIZE_T size; /* its size */ |
3841 | int victim_index; /* its bin index */ |
3842 | |
3843 | mchunkptr remainder; /* remainder from a split */ |
3844 | unsigned long remainder_size; /* its size */ |
3845 | |
3846 | unsigned int block; /* bit map traverser */ |
3847 | unsigned int bit; /* bit map traverser */ |
3848 | unsigned int map; /* current word of binmap */ |
3849 | |
3850 | mchunkptr fwd; /* misc temp for linking */ |
3851 | mchunkptr bck; /* misc temp for linking */ |
3852 | |
3853 | #if USE_TCACHE |
3854 | size_t tcache_unsorted_count; /* count of unsorted chunks processed */ |
3855 | #endif |
3856 | |
3857 | /* |
3858 | Convert request size to internal form by adding SIZE_SZ bytes |
3859 | overhead plus possibly more to obtain necessary alignment and/or |
3860 | to obtain a size of at least MINSIZE, the smallest allocatable |
3861 | size. Also, checked_request2size returns false for request sizes |
3862 | that are so large that they wrap around zero when padded and |
3863 | aligned. |
3864 | */ |
3865 | |
3866 | nb = checked_request2size (bytes); |
3867 | if (nb == 0) |
3868 | { |
3869 | __set_errno (ENOMEM); |
3870 | return NULL; |
3871 | } |
3872 | |
3873 | /* There are no usable arenas. Fall back to sysmalloc to get a chunk from |
3874 | mmap. */ |
3875 | if (__glibc_unlikely (av == NULL)) |
3876 | { |
3877 | void *p = sysmalloc (nb, av); |
3878 | if (p != NULL) |
3879 | alloc_perturb (p, bytes); |
3880 | return p; |
3881 | } |
3882 | |
3883 | /* |
3884 | If the size qualifies as a fastbin, first check corresponding bin. |
3885 | This code is safe to execute even if av is not yet initialized, so we |
3886 | can try it without checking, which saves some time on this fast path. |
3887 | */ |
3888 | |
3889 | #define REMOVE_FB(fb, victim, pp) \ |
3890 | do \ |
3891 | { \ |
3892 | victim = pp; \ |
3893 | if (victim == NULL) \ |
3894 | break; \ |
3895 | pp = REVEAL_PTR (victim->fd); \ |
3896 | if (__glibc_unlikely (pp != NULL && misaligned_chunk (pp))) \ |
3897 | malloc_printerr ("malloc(): unaligned fastbin chunk detected"); \ |
3898 | } \ |
3899 | while ((pp = catomic_compare_and_exchange_val_acq (fb, pp, victim)) \ |
3900 | != victim); \ |
3901 | |
3902 | if ((unsigned long) (nb) <= (unsigned long) (get_max_fast ())) |
3903 | { |
3904 | idx = fastbin_index (nb); |
3905 | mfastbinptr *fb = &fastbin (av, idx); |
3906 | mchunkptr pp; |
3907 | victim = *fb; |
3908 | |
3909 | if (victim != NULL) |
3910 | { |
3911 | if (__glibc_unlikely (misaligned_chunk (victim))) |
3912 | malloc_printerr ("malloc(): unaligned fastbin chunk detected 2" ); |
3913 | |
3914 | if (SINGLE_THREAD_P) |
3915 | *fb = REVEAL_PTR (victim->fd); |
3916 | else |
3917 | REMOVE_FB (fb, pp, victim); |
3918 | if (__glibc_likely (victim != NULL)) |
3919 | { |
3920 | size_t victim_idx = fastbin_index (chunksize (victim)); |
3921 | if (__builtin_expect (victim_idx != idx, 0)) |
3922 | malloc_printerr ("malloc(): memory corruption (fast)" ); |
3923 | check_remalloced_chunk (av, victim, nb); |
3924 | #if USE_TCACHE |
3925 | /* While we're here, if we see other chunks of the same size, |
3926 | stash them in the tcache. */ |
3927 | size_t tc_idx = csize2tidx (nb); |
3928 | if (tcache != NULL && tc_idx < mp_.tcache_bins) |
3929 | { |
3930 | mchunkptr tc_victim; |
3931 | |
3932 | /* While bin not empty and tcache not full, copy chunks. */ |
3933 | while (tcache->counts[tc_idx] < mp_.tcache_count |
3934 | && (tc_victim = *fb) != NULL) |
3935 | { |
3936 | if (__glibc_unlikely (misaligned_chunk (tc_victim))) |
3937 | malloc_printerr ("malloc(): unaligned fastbin chunk detected 3" ); |
3938 | if (SINGLE_THREAD_P) |
3939 | *fb = REVEAL_PTR (tc_victim->fd); |
3940 | else |
3941 | { |
3942 | REMOVE_FB (fb, pp, tc_victim); |
3943 | if (__glibc_unlikely (tc_victim == NULL)) |
3944 | break; |
3945 | } |
3946 | tcache_put (tc_victim, tc_idx); |
3947 | } |
3948 | } |
3949 | #endif |
3950 | void *p = chunk2mem (victim); |
3951 | alloc_perturb (p, bytes); |
3952 | return p; |
3953 | } |
3954 | } |
3955 | } |
3956 | |
3957 | /* |
3958 | If a small request, check regular bin. Since these "smallbins" |
3959 | hold one size each, no searching within bins is necessary. |
3960 | (For a large request, we need to wait until unsorted chunks are |
3961 | processed to find best fit. But for small ones, fits are exact |
3962 | anyway, so we can check now, which is faster.) |
3963 | */ |
3964 | |
3965 | if (in_smallbin_range (nb)) |
3966 | { |
3967 | idx = smallbin_index (nb); |
3968 | bin = bin_at (av, idx); |
3969 | |
3970 | if ((victim = last (bin)) != bin) |
3971 | { |
3972 | bck = victim->bk; |
3973 | if (__glibc_unlikely (bck->fd != victim)) |
3974 | malloc_printerr ("malloc(): smallbin double linked list corrupted" ); |
3975 | set_inuse_bit_at_offset (victim, nb); |
3976 | bin->bk = bck; |
3977 | bck->fd = bin; |
3978 | |
3979 | if (av != &main_arena) |
3980 | set_non_main_arena (victim); |
3981 | check_malloced_chunk (av, victim, nb); |
3982 | #if USE_TCACHE |
3983 | /* While we're here, if we see other chunks of the same size, |
3984 | stash them in the tcache. */ |
3985 | size_t tc_idx = csize2tidx (nb); |
3986 | if (tcache != NULL && tc_idx < mp_.tcache_bins) |
3987 | { |
3988 | mchunkptr tc_victim; |
3989 | |
3990 | /* While bin not empty and tcache not full, copy chunks over. */ |
3991 | while (tcache->counts[tc_idx] < mp_.tcache_count |
3992 | && (tc_victim = last (bin)) != bin) |
3993 | { |
3994 | if (tc_victim != 0) |
3995 | { |
3996 | bck = tc_victim->bk; |
3997 | set_inuse_bit_at_offset (tc_victim, nb); |
3998 | if (av != &main_arena) |
3999 | set_non_main_arena (tc_victim); |
4000 | bin->bk = bck; |
4001 | bck->fd = bin; |
4002 | |
4003 | tcache_put (tc_victim, tc_idx); |
4004 | } |
4005 | } |
4006 | } |
4007 | #endif |
4008 | void *p = chunk2mem (victim); |
4009 | alloc_perturb (p, bytes); |
4010 | return p; |
4011 | } |
4012 | } |
4013 | |
4014 | /* |
4015 | If this is a large request, consolidate fastbins before continuing. |
4016 | While it might look excessive to kill all fastbins before |
4017 | even seeing if there is space available, this avoids |
4018 | fragmentation problems normally associated with fastbins. |
4019 | Also, in practice, programs tend to have runs of either small or |
4020 | large requests, but less often mixtures, so consolidation is not |
4021 | invoked all that often in most programs. And the programs that |
4022 | it is called frequently in otherwise tend to fragment. |
4023 | */ |
4024 | |
4025 | else |
4026 | { |
4027 | idx = largebin_index (nb); |
4028 | if (atomic_load_relaxed (&av->have_fastchunks)) |
4029 | malloc_consolidate (av); |
4030 | } |
4031 | |
4032 | /* |
4033 | Process recently freed or remaindered chunks, taking one only if |
4034 | it is exact fit, or, if this a small request, the chunk is remainder from |
4035 | the most recent non-exact fit. Place other traversed chunks in |
4036 | bins. Note that this step is the only place in any routine where |
4037 | chunks are placed in bins. |
4038 | |
4039 | The outer loop here is needed because we might not realize until |
4040 | near the end of malloc that we should have consolidated, so must |
4041 | do so and retry. This happens at most once, and only when we would |
4042 | otherwise need to expand memory to service a "small" request. |
4043 | */ |
4044 | |
4045 | #if USE_TCACHE |
4046 | INTERNAL_SIZE_T tcache_nb = 0; |
4047 | size_t tc_idx = csize2tidx (nb); |
4048 | if (tcache != NULL && tc_idx < mp_.tcache_bins) |
4049 | tcache_nb = nb; |
4050 | int return_cached = 0; |
4051 | |
4052 | tcache_unsorted_count = 0; |
4053 | #endif |
4054 | |
4055 | for (;; ) |
4056 | { |
4057 | int iters = 0; |
4058 | while ((victim = unsorted_chunks (av)->bk) != unsorted_chunks (av)) |
4059 | { |
4060 | bck = victim->bk; |
4061 | size = chunksize (victim); |
4062 | mchunkptr next = chunk_at_offset (victim, size); |
4063 | |
4064 | if (__glibc_unlikely (size <= CHUNK_HDR_SZ) |
4065 | || __glibc_unlikely (size > av->system_mem)) |
4066 | malloc_printerr ("malloc(): invalid size (unsorted)" ); |
4067 | if (__glibc_unlikely (chunksize_nomask (next) < CHUNK_HDR_SZ) |
4068 | || __glibc_unlikely (chunksize_nomask (next) > av->system_mem)) |
4069 | malloc_printerr ("malloc(): invalid next size (unsorted)" ); |
4070 | if (__glibc_unlikely ((prev_size (next) & ~(SIZE_BITS)) != size)) |
4071 | malloc_printerr ("malloc(): mismatching next->prev_size (unsorted)" ); |
4072 | if (__glibc_unlikely (bck->fd != victim) |
4073 | || __glibc_unlikely (victim->fd != unsorted_chunks (av))) |
4074 | malloc_printerr ("malloc(): unsorted double linked list corrupted" ); |
4075 | if (__glibc_unlikely (prev_inuse (next))) |
4076 | malloc_printerr ("malloc(): invalid next->prev_inuse (unsorted)" ); |
4077 | |
4078 | /* |
4079 | If a small request, try to use last remainder if it is the |
4080 | only chunk in unsorted bin. This helps promote locality for |
4081 | runs of consecutive small requests. This is the only |
4082 | exception to best-fit, and applies only when there is |
4083 | no exact fit for a small chunk. |
4084 | */ |
4085 | |
4086 | if (in_smallbin_range (nb) && |
4087 | bck == unsorted_chunks (av) && |
4088 | victim == av->last_remainder && |
4089 | (unsigned long) (size) > (unsigned long) (nb + MINSIZE)) |
4090 | { |
4091 | /* split and reattach remainder */ |
4092 | remainder_size = size - nb; |
4093 | remainder = chunk_at_offset (victim, nb); |
4094 | unsorted_chunks (av)->bk = unsorted_chunks (av)->fd = remainder; |
4095 | av->last_remainder = remainder; |
4096 | remainder->bk = remainder->fd = unsorted_chunks (av); |
4097 | if (!in_smallbin_range (remainder_size)) |
4098 | { |
4099 | remainder->fd_nextsize = NULL; |
4100 | remainder->bk_nextsize = NULL; |
4101 | } |
4102 | |
4103 | set_head (victim, nb | PREV_INUSE | |
4104 | (av != &main_arena ? NON_MAIN_ARENA : 0)); |
4105 | set_head (remainder, remainder_size | PREV_INUSE); |
4106 | set_foot (remainder, remainder_size); |
4107 | |
4108 | check_malloced_chunk (av, victim, nb); |
4109 | void *p = chunk2mem (victim); |
4110 | alloc_perturb (p, bytes); |
4111 | return p; |
4112 | } |
4113 | |
4114 | /* remove from unsorted list */ |
4115 | unsorted_chunks (av)->bk = bck; |
4116 | bck->fd = unsorted_chunks (av); |
4117 | |
4118 | /* Take now instead of binning if exact fit */ |
4119 | |
4120 | if (size == nb) |
4121 | { |
4122 | set_inuse_bit_at_offset (victim, size); |
4123 | if (av != &main_arena) |
4124 | set_non_main_arena (victim); |
4125 | #if USE_TCACHE |
4126 | /* Fill cache first, return to user only if cache fills. |
4127 | We may return one of these chunks later. */ |
4128 | if (tcache_nb > 0 |
4129 | && tcache->counts[tc_idx] < mp_.tcache_count) |
4130 | { |
4131 | tcache_put (victim, tc_idx); |
4132 | return_cached = 1; |
4133 | continue; |
4134 | } |
4135 | else |
4136 | { |
4137 | #endif |
4138 | check_malloced_chunk (av, victim, nb); |
4139 | void *p = chunk2mem (victim); |
4140 | alloc_perturb (p, bytes); |
4141 | return p; |
4142 | #if USE_TCACHE |
4143 | } |
4144 | #endif |
4145 | } |
4146 | |
4147 | /* place chunk in bin */ |
4148 | |
4149 | if (in_smallbin_range (size)) |
4150 | { |
4151 | victim_index = smallbin_index (size); |
4152 | bck = bin_at (av, victim_index); |
4153 | fwd = bck->fd; |
4154 | } |
4155 | else |
4156 | { |
4157 | victim_index = largebin_index (size); |
4158 | bck = bin_at (av, victim_index); |
4159 | fwd = bck->fd; |
4160 | |
4161 | /* maintain large bins in sorted order */ |
4162 | if (fwd != bck) |
4163 | { |
4164 | /* Or with inuse bit to speed comparisons */ |
4165 | size |= PREV_INUSE; |
4166 | /* if smaller than smallest, bypass loop below */ |
4167 | assert (chunk_main_arena (bck->bk)); |
4168 | if ((unsigned long) (size) |
4169 | < (unsigned long) chunksize_nomask (bck->bk)) |
4170 | { |
4171 | fwd = bck; |
4172 | bck = bck->bk; |
4173 | |
4174 | victim->fd_nextsize = fwd->fd; |
4175 | victim->bk_nextsize = fwd->fd->bk_nextsize; |
4176 | fwd->fd->bk_nextsize = victim->bk_nextsize->fd_nextsize = victim; |
4177 | } |
4178 | else |
4179 | { |
4180 | assert (chunk_main_arena (fwd)); |
4181 | while ((unsigned long) size < chunksize_nomask (fwd)) |
4182 | { |
4183 | fwd = fwd->fd_nextsize; |
4184 | assert (chunk_main_arena (fwd)); |
4185 | } |
4186 | |
4187 | if ((unsigned long) size |
4188 | == (unsigned long) chunksize_nomask (fwd)) |
4189 | /* Always insert in the second position. */ |
4190 | fwd = fwd->fd; |
4191 | else |
4192 | { |
4193 | victim->fd_nextsize = fwd; |
4194 | victim->bk_nextsize = fwd->bk_nextsize; |
4195 | if (__glibc_unlikely (fwd->bk_nextsize->fd_nextsize != fwd)) |
4196 | malloc_printerr ("malloc(): largebin double linked list corrupted (nextsize)" ); |
4197 | fwd->bk_nextsize = victim; |
4198 | victim->bk_nextsize->fd_nextsize = victim; |
4199 | } |
4200 | bck = fwd->bk; |
4201 | if (bck->fd != fwd) |
4202 | malloc_printerr ("malloc(): largebin double linked list corrupted (bk)" ); |
4203 | } |
4204 | } |
4205 | else |
4206 | victim->fd_nextsize = victim->bk_nextsize = victim; |
4207 | } |
4208 | |
4209 | mark_bin (av, victim_index); |
4210 | victim->bk = bck; |
4211 | victim->fd = fwd; |
4212 | fwd->bk = victim; |
4213 | bck->fd = victim; |
4214 | |
4215 | #if USE_TCACHE |
4216 | /* If we've processed as many chunks as we're allowed while |
4217 | filling the cache, return one of the cached ones. */ |
4218 | ++tcache_unsorted_count; |
4219 | if (return_cached |
4220 | && mp_.tcache_unsorted_limit > 0 |
4221 | && tcache_unsorted_count > mp_.tcache_unsorted_limit) |
4222 | { |
4223 | return tcache_get (tc_idx); |
4224 | } |
4225 | #endif |
4226 | |
4227 | #define MAX_ITERS 10000 |
4228 | if (++iters >= MAX_ITERS) |
4229 | break; |
4230 | } |
4231 | |
4232 | #if USE_TCACHE |
4233 | /* If all the small chunks we found ended up cached, return one now. */ |
4234 | if (return_cached) |
4235 | { |
4236 | return tcache_get (tc_idx); |
4237 | } |
4238 | #endif |
4239 | |
4240 | /* |
4241 | If a large request, scan through the chunks of current bin in |
4242 | sorted order to find smallest that fits. Use the skip list for this. |
4243 | */ |
4244 | |
4245 | if (!in_smallbin_range (nb)) |
4246 | { |
4247 | bin = bin_at (av, idx); |
4248 | |
4249 | /* skip scan if empty or largest chunk is too small */ |
4250 | if ((victim = first (bin)) != bin |
4251 | && (unsigned long) chunksize_nomask (victim) |
4252 | >= (unsigned long) (nb)) |
4253 | { |
4254 | victim = victim->bk_nextsize; |
4255 | while (((unsigned long) (size = chunksize (victim)) < |
4256 | (unsigned long) (nb))) |
4257 | victim = victim->bk_nextsize; |
4258 | |
4259 | /* Avoid removing the first entry for a size so that the skip |
4260 | list does not have to be rerouted. */ |
4261 | if (victim != last (bin) |
4262 | && chunksize_nomask (victim) |
4263 | == chunksize_nomask (victim->fd)) |
4264 | victim = victim->fd; |
4265 | |
4266 | remainder_size = size - nb; |
4267 | unlink_chunk (av, victim); |
4268 | |
4269 | /* Exhaust */ |
4270 | if (remainder_size < MINSIZE) |
4271 | { |
4272 | set_inuse_bit_at_offset (victim, size); |
4273 | if (av != &main_arena) |
4274 | set_non_main_arena (victim); |
4275 | } |
4276 | /* Split */ |
4277 | else |
4278 | { |
4279 | remainder = chunk_at_offset (victim, nb); |
4280 | /* We cannot assume the unsorted list is empty and therefore |
4281 | have to perform a complete insert here. */ |
4282 | bck = unsorted_chunks (av); |
4283 | fwd = bck->fd; |
4284 | if (__glibc_unlikely (fwd->bk != bck)) |
4285 | malloc_printerr ("malloc(): corrupted unsorted chunks" ); |
4286 | remainder->bk = bck; |
4287 | remainder->fd = fwd; |
4288 | bck->fd = remainder; |
4289 | fwd->bk = remainder; |
4290 | if (!in_smallbin_range (remainder_size)) |
4291 | { |
4292 | remainder->fd_nextsize = NULL; |
4293 | remainder->bk_nextsize = NULL; |
4294 | } |
4295 | set_head (victim, nb | PREV_INUSE | |
4296 | (av != &main_arena ? NON_MAIN_ARENA : 0)); |
4297 | set_head (remainder, remainder_size | PREV_INUSE); |
4298 | set_foot (remainder, remainder_size); |
4299 | } |
4300 | check_malloced_chunk (av, victim, nb); |
4301 | void *p = chunk2mem (victim); |
4302 | alloc_perturb (p, bytes); |
4303 | return p; |
4304 | } |
4305 | } |
4306 | |
4307 | /* |
4308 | Search for a chunk by scanning bins, starting with next largest |
4309 | bin. This search is strictly by best-fit; i.e., the smallest |
4310 | (with ties going to approximately the least recently used) chunk |
4311 | that fits is selected. |
4312 | |
4313 | The bitmap avoids needing to check that most blocks are nonempty. |
4314 | The particular case of skipping all bins during warm-up phases |
4315 | when no chunks have been returned yet is faster than it might look. |
4316 | */ |
4317 | |
4318 | ++idx; |
4319 | bin = bin_at (av, idx); |
4320 | block = idx2block (idx); |
4321 | map = av->binmap[block]; |
4322 | bit = idx2bit (idx); |
4323 | |
4324 | for (;; ) |
4325 | { |
4326 | /* Skip rest of block if there are no more set bits in this block. */ |
4327 | if (bit > map || bit == 0) |
4328 | { |
4329 | do |
4330 | { |
4331 | if (++block >= BINMAPSIZE) /* out of bins */ |
4332 | goto use_top; |
4333 | } |
4334 | while ((map = av->binmap[block]) == 0); |
4335 | |
4336 | bin = bin_at (av, (block << BINMAPSHIFT)); |
4337 | bit = 1; |
4338 | } |
4339 | |
4340 | /* Advance to bin with set bit. There must be one. */ |
4341 | while ((bit & map) == 0) |
4342 | { |
4343 | bin = next_bin (bin); |
4344 | bit <<= 1; |
4345 | assert (bit != 0); |
4346 | } |
4347 | |
4348 | /* Inspect the bin. It is likely to be non-empty */ |
4349 | victim = last (bin); |
4350 | |
4351 | /* If a false alarm (empty bin), clear the bit. */ |
4352 | if (victim == bin) |
4353 | { |
4354 | av->binmap[block] = map &= ~bit; /* Write through */ |
4355 | bin = next_bin (bin); |
4356 | bit <<= 1; |
4357 | } |
4358 | |
4359 | else |
4360 | { |
4361 | size = chunksize (victim); |
4362 | |
4363 | /* We know the first chunk in this bin is big enough to use. */ |
4364 | assert ((unsigned long) (size) >= (unsigned long) (nb)); |
4365 | |
4366 | remainder_size = size - nb; |
4367 | |
4368 | /* unlink */ |
4369 | unlink_chunk (av, victim); |
4370 | |
4371 | /* Exhaust */ |
4372 | if (remainder_size < MINSIZE) |
4373 | { |
4374 | set_inuse_bit_at_offset (victim, size); |
4375 | if (av != &main_arena) |
4376 | set_non_main_arena (victim); |
4377 | } |
4378 | |
4379 | /* Split */ |
4380 | else |
4381 | { |
4382 | remainder = chunk_at_offset (victim, nb); |
4383 | |
4384 | /* We cannot assume the unsorted list is empty and therefore |
4385 | have to perform a complete insert here. */ |
4386 | bck = unsorted_chunks (av); |
4387 | fwd = bck->fd; |
4388 | if (__glibc_unlikely (fwd->bk != bck)) |
4389 | malloc_printerr ("malloc(): corrupted unsorted chunks 2" ); |
4390 | remainder->bk = bck; |
4391 | remainder->fd = fwd; |
4392 | bck->fd = remainder; |
4393 | fwd->bk = remainder; |
4394 | |
4395 | /* advertise as last remainder */ |
4396 | if (in_smallbin_range (nb)) |
4397 | av->last_remainder = remainder; |
4398 | if (!in_smallbin_range (remainder_size)) |
4399 | { |
4400 | remainder->fd_nextsize = NULL; |
4401 | remainder->bk_nextsize = NULL; |
4402 | } |
4403 | set_head (victim, nb | PREV_INUSE | |
4404 | (av != &main_arena ? NON_MAIN_ARENA : 0)); |
4405 | set_head (remainder, remainder_size | PREV_INUSE); |
4406 | set_foot (remainder, remainder_size); |
4407 | } |
4408 | check_malloced_chunk (av, victim, nb); |
4409 | void *p = chunk2mem (victim); |
4410 | alloc_perturb (p, bytes); |
4411 | return p; |
4412 | } |
4413 | } |
4414 | |
4415 | use_top: |
4416 | /* |
4417 | If large enough, split off the chunk bordering the end of memory |
4418 | (held in av->top). Note that this is in accord with the best-fit |
4419 | search rule. In effect, av->top is treated as larger (and thus |
4420 | less well fitting) than any other available chunk since it can |
4421 | be extended to be as large as necessary (up to system |
4422 | limitations). |
4423 | |
4424 | We require that av->top always exists (i.e., has size >= |
4425 | MINSIZE) after initialization, so if it would otherwise be |
4426 | exhausted by current request, it is replenished. (The main |
4427 | reason for ensuring it exists is that we may need MINSIZE space |
4428 | to put in fenceposts in sysmalloc.) |
4429 | */ |
4430 | |
4431 | victim = av->top; |
4432 | size = chunksize (victim); |
4433 | |
4434 | if (__glibc_unlikely (size > av->system_mem)) |
4435 | malloc_printerr ("malloc(): corrupted top size" ); |
4436 | |
4437 | if ((unsigned long) (size) >= (unsigned long) (nb + MINSIZE)) |
4438 | { |
4439 | remainder_size = size - nb; |
4440 | remainder = chunk_at_offset (victim, nb); |
4441 | av->top = remainder; |
4442 | set_head (victim, nb | PREV_INUSE | |
4443 | (av != &main_arena ? NON_MAIN_ARENA : 0)); |
4444 | set_head (remainder, remainder_size | PREV_INUSE); |
4445 | |
4446 | check_malloced_chunk (av, victim, nb); |
4447 | void *p = chunk2mem (victim); |
4448 | alloc_perturb (p, bytes); |
4449 | return p; |
4450 | } |
4451 | |
4452 | /* When we are using atomic ops to free fast chunks we can get |
4453 | here for all block sizes. */ |
4454 | else if (atomic_load_relaxed (&av->have_fastchunks)) |
4455 | { |
4456 | malloc_consolidate (av); |
4457 | /* restore original bin index */ |
4458 | if (in_smallbin_range (nb)) |
4459 | idx = smallbin_index (nb); |
4460 | else |
4461 | idx = largebin_index (nb); |
4462 | } |
4463 | |
4464 | /* |
4465 | Otherwise, relay to handle system-dependent cases |
4466 | */ |
4467 | else |
4468 | { |
4469 | void *p = sysmalloc (nb, av); |
4470 | if (p != NULL) |
4471 | alloc_perturb (p, bytes); |
4472 | return p; |
4473 | } |
4474 | } |
4475 | } |
4476 | |
4477 | /* |
4478 | ------------------------------ free ------------------------------ |
4479 | */ |
4480 | |
4481 | static void |
4482 | _int_free (mstate av, mchunkptr p, int have_lock) |
4483 | { |
4484 | INTERNAL_SIZE_T size; /* its size */ |
4485 | mfastbinptr *fb; /* associated fastbin */ |
4486 | mchunkptr nextchunk; /* next contiguous chunk */ |
4487 | INTERNAL_SIZE_T nextsize; /* its size */ |
4488 | int nextinuse; /* true if nextchunk is used */ |
4489 | INTERNAL_SIZE_T prevsize; /* size of previous contiguous chunk */ |
4490 | mchunkptr bck; /* misc temp for linking */ |
4491 | mchunkptr fwd; /* misc temp for linking */ |
4492 | |
4493 | size = chunksize (p); |
4494 | |
4495 | /* Little security check which won't hurt performance: the |
4496 | allocator never wraps around at the end of the address space. |
4497 | Therefore we can exclude some size values which might appear |
4498 | here by accident or by "design" from some intruder. */ |
4499 | if (__builtin_expect ((uintptr_t) p > (uintptr_t) -size, 0) |
4500 | || __builtin_expect (misaligned_chunk (p), 0)) |
4501 | malloc_printerr ("free(): invalid pointer" ); |
4502 | /* We know that each chunk is at least MINSIZE bytes in size or a |
4503 | multiple of MALLOC_ALIGNMENT. */ |
4504 | if (__glibc_unlikely (size < MINSIZE || !aligned_OK (size))) |
4505 | malloc_printerr ("free(): invalid size" ); |
4506 | |
4507 | check_inuse_chunk(av, p); |
4508 | |
4509 | #if USE_TCACHE |
4510 | { |
4511 | size_t tc_idx = csize2tidx (size); |
4512 | if (tcache != NULL && tc_idx < mp_.tcache_bins) |
4513 | { |
4514 | /* Check to see if it's already in the tcache. */ |
4515 | tcache_entry *e = (tcache_entry *) chunk2mem (p); |
4516 | |
4517 | /* This test succeeds on double free. However, we don't 100% |
4518 | trust it (it also matches random payload data at a 1 in |
4519 | 2^<size_t> chance), so verify it's not an unlikely |
4520 | coincidence before aborting. */ |
4521 | if (__glibc_unlikely (e->key == tcache_key)) |
4522 | { |
4523 | tcache_entry *tmp; |
4524 | size_t cnt = 0; |
4525 | LIBC_PROBE (memory_tcache_double_free, 2, e, tc_idx); |
4526 | for (tmp = tcache->entries[tc_idx]; |
4527 | tmp; |
4528 | tmp = REVEAL_PTR (tmp->next), ++cnt) |
4529 | { |
4530 | if (cnt >= mp_.tcache_count) |
4531 | malloc_printerr ("free(): too many chunks detected in tcache" ); |
4532 | if (__glibc_unlikely (!aligned_OK (tmp))) |
4533 | malloc_printerr ("free(): unaligned chunk detected in tcache 2" ); |
4534 | if (tmp == e) |
4535 | malloc_printerr ("free(): double free detected in tcache 2" ); |
4536 | /* If we get here, it was a coincidence. We've wasted a |
4537 | few cycles, but don't abort. */ |
4538 | } |
4539 | } |
4540 | |
4541 | if (tcache->counts[tc_idx] < mp_.tcache_count) |
4542 | { |
4543 | tcache_put (p, tc_idx); |
4544 | return; |
4545 | } |
4546 | } |
4547 | } |
4548 | #endif |
4549 | |
4550 | /* |
4551 | If eligible, place chunk on a fastbin so it can be found |
4552 | and used quickly in malloc. |
4553 | */ |
4554 | |
4555 | if ((unsigned long)(size) <= (unsigned long)(get_max_fast ()) |
4556 | |
4557 | #if TRIM_FASTBINS |
4558 | /* |
4559 | If TRIM_FASTBINS set, don't place chunks |
4560 | bordering top into fastbins |
4561 | */ |
4562 | && (chunk_at_offset(p, size) != av->top) |
4563 | #endif |
4564 | ) { |
4565 | |
4566 | if (__builtin_expect (chunksize_nomask (chunk_at_offset (p, size)) |
4567 | <= CHUNK_HDR_SZ, 0) |
4568 | || __builtin_expect (chunksize (chunk_at_offset (p, size)) |
4569 | >= av->system_mem, 0)) |
4570 | { |
4571 | bool fail = true; |
4572 | /* We might not have a lock at this point and concurrent modifications |
4573 | of system_mem might result in a false positive. Redo the test after |
4574 | getting the lock. */ |
4575 | if (!have_lock) |
4576 | { |
4577 | __libc_lock_lock (av->mutex); |
4578 | fail = (chunksize_nomask (chunk_at_offset (p, size)) <= CHUNK_HDR_SZ |
4579 | || chunksize (chunk_at_offset (p, size)) >= av->system_mem); |
4580 | __libc_lock_unlock (av->mutex); |
4581 | } |
4582 | |
4583 | if (fail) |
4584 | malloc_printerr ("free(): invalid next size (fast)" ); |
4585 | } |
4586 | |
4587 | free_perturb (chunk2mem(p), size - CHUNK_HDR_SZ); |
4588 | |
4589 | atomic_store_relaxed (&av->have_fastchunks, true); |
4590 | unsigned int idx = fastbin_index(size); |
4591 | fb = &fastbin (av, idx); |
4592 | |
4593 | /* Atomically link P to its fastbin: P->FD = *FB; *FB = P; */ |
4594 | mchunkptr old = *fb, old2; |
4595 | |
4596 | if (SINGLE_THREAD_P) |
4597 | { |
4598 | /* Check that the top of the bin is not the record we are going to |
4599 | add (i.e., double free). */ |
4600 | if (__builtin_expect (old == p, 0)) |
4601 | malloc_printerr ("double free or corruption (fasttop)" ); |
4602 | p->fd = PROTECT_PTR (&p->fd, old); |
4603 | *fb = p; |
4604 | } |
4605 | else |
4606 | do |
4607 | { |
4608 | /* Check that the top of the bin is not the record we are going to |
4609 | add (i.e., double free). */ |
4610 | if (__builtin_expect (old == p, 0)) |
4611 | malloc_printerr ("double free or corruption (fasttop)" ); |
4612 | old2 = old; |
4613 | p->fd = PROTECT_PTR (&p->fd, old); |
4614 | } |
4615 | while ((old = catomic_compare_and_exchange_val_rel (fb, p, old2)) |
4616 | != old2); |
4617 | |
4618 | /* Check that size of fastbin chunk at the top is the same as |
4619 | size of the chunk that we are adding. We can dereference OLD |
4620 | only if we have the lock, otherwise it might have already been |
4621 | allocated again. */ |
4622 | if (have_lock && old != NULL |
4623 | && __builtin_expect (fastbin_index (chunksize (old)) != idx, 0)) |
4624 | malloc_printerr ("invalid fastbin entry (free)" ); |
4625 | } |
4626 | |
4627 | /* |
4628 | Consolidate other non-mmapped chunks as they arrive. |
4629 | */ |
4630 | |
4631 | else if (!chunk_is_mmapped(p)) { |
4632 | |
4633 | /* If we're single-threaded, don't lock the arena. */ |
4634 | if (SINGLE_THREAD_P) |
4635 | have_lock = true; |
4636 | |
4637 | if (!have_lock) |
4638 | __libc_lock_lock (av->mutex); |
4639 | |
4640 | nextchunk = chunk_at_offset(p, size); |
4641 | |
4642 | /* Lightweight tests: check whether the block is already the |
4643 | top block. */ |
4644 | if (__glibc_unlikely (p == av->top)) |
4645 | malloc_printerr ("double free or corruption (top)" ); |
4646 | /* Or whether the next chunk is beyond the boundaries of the arena. */ |
4647 | if (__builtin_expect (contiguous (av) |
4648 | && (char *) nextchunk |
4649 | >= ((char *) av->top + chunksize(av->top)), 0)) |
4650 | malloc_printerr ("double free or corruption (out)" ); |
4651 | /* Or whether the block is actually not marked used. */ |
4652 | if (__glibc_unlikely (!prev_inuse(nextchunk))) |
4653 | malloc_printerr ("double free or corruption (!prev)" ); |
4654 | |
4655 | nextsize = chunksize(nextchunk); |
4656 | if (__builtin_expect (chunksize_nomask (nextchunk) <= CHUNK_HDR_SZ, 0) |
4657 | || __builtin_expect (nextsize >= av->system_mem, 0)) |
4658 | malloc_printerr ("free(): invalid next size (normal)" ); |
4659 | |
4660 | free_perturb (chunk2mem(p), size - CHUNK_HDR_SZ); |
4661 | |
4662 | /* consolidate backward */ |
4663 | if (!prev_inuse(p)) { |
4664 | prevsize = prev_size (p); |
4665 | size += prevsize; |
4666 | p = chunk_at_offset(p, -((long) prevsize)); |
4667 | if (__glibc_unlikely (chunksize(p) != prevsize)) |
4668 | malloc_printerr ("corrupted size vs. prev_size while consolidating" ); |
4669 | unlink_chunk (av, p); |
4670 | } |
4671 | |
4672 | if (nextchunk != av->top) { |
4673 | /* get and clear inuse bit */ |
4674 | nextinuse = inuse_bit_at_offset(nextchunk, nextsize); |
4675 | |
4676 | /* consolidate forward */ |
4677 | if (!nextinuse) { |
4678 | unlink_chunk (av, nextchunk); |
4679 | size += nextsize; |
4680 | } else |
4681 | clear_inuse_bit_at_offset(nextchunk, 0); |
4682 | |
4683 | /* |
4684 | Place the chunk in unsorted chunk list. Chunks are |
4685 | not placed into regular bins until after they have |
4686 | been given one chance to be used in malloc. |
4687 | */ |
4688 | |
4689 | bck = unsorted_chunks(av); |
4690 | fwd = bck->fd; |
4691 | if (__glibc_unlikely (fwd->bk != bck)) |
4692 | malloc_printerr ("free(): corrupted unsorted chunks" ); |
4693 | p->fd = fwd; |
4694 | p->bk = bck; |
4695 | if (!in_smallbin_range(size)) |
4696 | { |
4697 | p->fd_nextsize = NULL; |
4698 | p->bk_nextsize = NULL; |
4699 | } |
4700 | bck->fd = p; |
4701 | fwd->bk = p; |
4702 | |
4703 | set_head(p, size | PREV_INUSE); |
4704 | set_foot(p, size); |
4705 | |
4706 | check_free_chunk(av, p); |
4707 | } |
4708 | |
4709 | /* |
4710 | If the chunk borders the current high end of memory, |
4711 | consolidate into top |
4712 | */ |
4713 | |
4714 | else { |
4715 | size += nextsize; |
4716 | set_head(p, size | PREV_INUSE); |
4717 | av->top = p; |
4718 | check_chunk(av, p); |
4719 | } |
4720 | |
4721 | /* |
4722 | If freeing a large space, consolidate possibly-surrounding |
4723 | chunks. Then, if the total unused topmost memory exceeds trim |
4724 | threshold, ask malloc_trim to reduce top. |
4725 | |
4726 | Unless max_fast is 0, we don't know if there are fastbins |
4727 | bordering top, so we cannot tell for sure whether threshold |
4728 | has been reached unless fastbins are consolidated. But we |
4729 | don't want to consolidate on each free. As a compromise, |
4730 | consolidation is performed if FASTBIN_CONSOLIDATION_THRESHOLD |
4731 | is reached. |
4732 | */ |
4733 | |
4734 | if ((unsigned long)(size) >= FASTBIN_CONSOLIDATION_THRESHOLD) { |
4735 | if (atomic_load_relaxed (&av->have_fastchunks)) |
4736 | malloc_consolidate(av); |
4737 | |
4738 | if (av == &main_arena) { |
4739 | #ifndef MORECORE_CANNOT_TRIM |
4740 | if ((unsigned long)(chunksize(av->top)) >= |
4741 | (unsigned long)(mp_.trim_threshold)) |
4742 | systrim(mp_.top_pad, av); |
4743 | #endif |
4744 | } else { |
4745 | /* Always try heap_trim(), even if the top chunk is not |
4746 | large, because the corresponding heap might go away. */ |
4747 | heap_info *heap = heap_for_ptr(top(av)); |
4748 | |
4749 | assert(heap->ar_ptr == av); |
4750 | heap_trim(heap, mp_.top_pad); |
4751 | } |
4752 | } |
4753 | |
4754 | if (!have_lock) |
4755 | __libc_lock_unlock (av->mutex); |
4756 | } |
4757 | /* |
4758 | If the chunk was allocated via mmap, release via munmap(). |
4759 | */ |
4760 | |
4761 | else { |
4762 | munmap_chunk (p); |
4763 | } |
4764 | } |
4765 | |
4766 | /* |
4767 | ------------------------- malloc_consolidate ------------------------- |
4768 | |
4769 | malloc_consolidate is a specialized version of free() that tears |
4770 | down chunks held in fastbins. Free itself cannot be used for this |
4771 | purpose since, among other things, it might place chunks back onto |
4772 | fastbins. So, instead, we need to use a minor variant of the same |
4773 | code. |
4774 | */ |
4775 | |
4776 | static void malloc_consolidate(mstate av) |
4777 | { |
4778 | mfastbinptr* fb; /* current fastbin being consolidated */ |
4779 | mfastbinptr* maxfb; /* last fastbin (for loop control) */ |
4780 | mchunkptr p; /* current chunk being consolidated */ |
4781 | mchunkptr nextp; /* next chunk to consolidate */ |
4782 | mchunkptr unsorted_bin; /* bin header */ |
4783 | mchunkptr first_unsorted; /* chunk to link to */ |
4784 | |
4785 | /* These have same use as in free() */ |
4786 | mchunkptr nextchunk; |
4787 | INTERNAL_SIZE_T size; |
4788 | INTERNAL_SIZE_T nextsize; |
4789 | INTERNAL_SIZE_T prevsize; |
4790 | int nextinuse; |
4791 | |
4792 | atomic_store_relaxed (&av->have_fastchunks, false); |
4793 | |
4794 | unsorted_bin = unsorted_chunks(av); |
4795 | |
4796 | /* |
4797 | Remove each chunk from fast bin and consolidate it, placing it |
4798 | then in unsorted bin. Among other reasons for doing this, |
4799 | placing in unsorted bin avoids needing to calculate actual bins |
4800 | until malloc is sure that chunks aren't immediately going to be |
4801 | reused anyway. |
4802 | */ |
4803 | |
4804 | maxfb = &fastbin (av, NFASTBINS - 1); |
4805 | fb = &fastbin (av, 0); |
4806 | do { |
4807 | p = atomic_exchange_acquire (fb, NULL); |
4808 | if (p != 0) { |
4809 | do { |
4810 | { |
4811 | if (__glibc_unlikely (misaligned_chunk (p))) |
4812 | malloc_printerr ("malloc_consolidate(): " |
4813 | "unaligned fastbin chunk detected" ); |
4814 | |
4815 | unsigned int idx = fastbin_index (chunksize (p)); |
4816 | if ((&fastbin (av, idx)) != fb) |
4817 | malloc_printerr ("malloc_consolidate(): invalid chunk size" ); |
4818 | } |
4819 | |
4820 | check_inuse_chunk(av, p); |
4821 | nextp = REVEAL_PTR (p->fd); |
4822 | |
4823 | /* Slightly streamlined version of consolidation code in free() */ |
4824 | size = chunksize (p); |
4825 | nextchunk = chunk_at_offset(p, size); |
4826 | nextsize = chunksize(nextchunk); |
4827 | |
4828 | if (!prev_inuse(p)) { |
4829 | prevsize = prev_size (p); |
4830 | size += prevsize; |
4831 | p = chunk_at_offset(p, -((long) prevsize)); |
4832 | if (__glibc_unlikely (chunksize(p) != prevsize)) |
4833 | malloc_printerr ("corrupted size vs. prev_size in fastbins" ); |
4834 | unlink_chunk (av, p); |
4835 | } |
4836 | |
4837 | if (nextchunk != av->top) { |
4838 | nextinuse = inuse_bit_at_offset(nextchunk, nextsize); |
4839 | |
4840 | if (!nextinuse) { |
4841 | size += nextsize; |
4842 | unlink_chunk (av, nextchunk); |
4843 | } else |
4844 | clear_inuse_bit_at_offset(nextchunk, 0); |
4845 | |
4846 | first_unsorted = unsorted_bin->fd; |
4847 | unsorted_bin->fd = p; |
4848 | first_unsorted->bk = p; |
4849 | |
4850 | if (!in_smallbin_range (size)) { |
4851 | p->fd_nextsize = NULL; |
4852 | p->bk_nextsize = NULL; |
4853 | } |
4854 | |
4855 | set_head(p, size | PREV_INUSE); |
4856 | p->bk = unsorted_bin; |
4857 | p->fd = first_unsorted; |
4858 | set_foot(p, size); |
4859 | } |
4860 | |
4861 | else { |
4862 | size += nextsize; |
4863 | set_head(p, size | PREV_INUSE); |
4864 | av->top = p; |
4865 | } |
4866 | |
4867 | } while ( (p = nextp) != 0); |
4868 | |
4869 | } |
4870 | } while (fb++ != maxfb); |
4871 | } |
4872 | |
4873 | /* |
4874 | ------------------------------ realloc ------------------------------ |
4875 | */ |
4876 | |
4877 | static void * |
4878 | _int_realloc (mstate av, mchunkptr oldp, INTERNAL_SIZE_T oldsize, |
4879 | INTERNAL_SIZE_T nb) |
4880 | { |
4881 | mchunkptr newp; /* chunk to return */ |
4882 | INTERNAL_SIZE_T newsize; /* its size */ |
4883 | void* newmem; /* corresponding user mem */ |
4884 | |
4885 | mchunkptr next; /* next contiguous chunk after oldp */ |
4886 | |
4887 | mchunkptr remainder; /* extra space at end of newp */ |
4888 | unsigned long remainder_size; /* its size */ |
4889 | |
4890 | /* oldmem size */ |
4891 | if (__builtin_expect (chunksize_nomask (oldp) <= CHUNK_HDR_SZ, 0) |
4892 | || __builtin_expect (oldsize >= av->system_mem, 0) |
4893 | || __builtin_expect (oldsize != chunksize (oldp), 0)) |
4894 | malloc_printerr ("realloc(): invalid old size" ); |
4895 | |
4896 | check_inuse_chunk (av, oldp); |
4897 | |
4898 | /* All callers already filter out mmap'ed chunks. */ |
4899 | assert (!chunk_is_mmapped (oldp)); |
4900 | |
4901 | next = chunk_at_offset (oldp, oldsize); |
4902 | INTERNAL_SIZE_T nextsize = chunksize (next); |
4903 | if (__builtin_expect (chunksize_nomask (next) <= CHUNK_HDR_SZ, 0) |
4904 | || __builtin_expect (nextsize >= av->system_mem, 0)) |
4905 | malloc_printerr ("realloc(): invalid next size" ); |
4906 | |
4907 | if ((unsigned long) (oldsize) >= (unsigned long) (nb)) |
4908 | { |
4909 | /* already big enough; split below */ |
4910 | newp = oldp; |
4911 | newsize = oldsize; |
4912 | } |
4913 | |
4914 | else |
4915 | { |
4916 | /* Try to expand forward into top */ |
4917 | if (next == av->top && |
4918 | (unsigned long) (newsize = oldsize + nextsize) >= |
4919 | (unsigned long) (nb + MINSIZE)) |
4920 | { |
4921 | set_head_size (oldp, nb | (av != &main_arena ? NON_MAIN_ARENA : 0)); |
4922 | av->top = chunk_at_offset (oldp, nb); |
4923 | set_head (av->top, (newsize - nb) | PREV_INUSE); |
4924 | check_inuse_chunk (av, oldp); |
4925 | return tag_new_usable (chunk2mem (oldp)); |
4926 | } |
4927 | |
4928 | /* Try to expand forward into next chunk; split off remainder below */ |
4929 | else if (next != av->top && |
4930 | !inuse (next) && |
4931 | (unsigned long) (newsize = oldsize + nextsize) >= |
4932 | (unsigned long) (nb)) |
4933 | { |
4934 | newp = oldp; |
4935 | unlink_chunk (av, next); |
4936 | } |
4937 | |
4938 | /* allocate, copy, free */ |
4939 | else |
4940 | { |
4941 | newmem = _int_malloc (av, nb - MALLOC_ALIGN_MASK); |
4942 | if (newmem == 0) |
4943 | return 0; /* propagate failure */ |
4944 | |
4945 | newp = mem2chunk (newmem); |
4946 | newsize = chunksize (newp); |
4947 | |
4948 | /* |
4949 | Avoid copy if newp is next chunk after oldp. |
4950 | */ |
4951 | if (newp == next) |
4952 | { |
4953 | newsize += oldsize; |
4954 | newp = oldp; |
4955 | } |
4956 | else |
4957 | { |
4958 | void *oldmem = chunk2mem (oldp); |
4959 | size_t sz = memsize (oldp); |
4960 | (void) tag_region (oldmem, sz); |
4961 | newmem = tag_new_usable (newmem); |
4962 | memcpy (newmem, oldmem, sz); |
4963 | _int_free (av, oldp, 1); |
4964 | check_inuse_chunk (av, newp); |
4965 | return newmem; |
4966 | } |
4967 | } |
4968 | } |
4969 | |
4970 | /* If possible, free extra space in old or extended chunk */ |
4971 | |
4972 | assert ((unsigned long) (newsize) >= (unsigned long) (nb)); |
4973 | |
4974 | remainder_size = newsize - nb; |
4975 | |
4976 | if (remainder_size < MINSIZE) /* not enough extra to split off */ |
4977 | { |
4978 | set_head_size (newp, newsize | (av != &main_arena ? NON_MAIN_ARENA : 0)); |
4979 | set_inuse_bit_at_offset (newp, newsize); |
4980 | } |
4981 | else /* split remainder */ |
4982 | { |
4983 | remainder = chunk_at_offset (newp, nb); |
4984 | /* Clear any user-space tags before writing the header. */ |
4985 | remainder = tag_region (remainder, remainder_size); |
4986 | set_head_size (newp, nb | (av != &main_arena ? NON_MAIN_ARENA : 0)); |
4987 | set_head (remainder, remainder_size | PREV_INUSE | |
4988 | (av != &main_arena ? NON_MAIN_ARENA : 0)); |
4989 | /* Mark remainder as inuse so free() won't complain */ |
4990 | set_inuse_bit_at_offset (remainder, remainder_size); |
4991 | _int_free (av, remainder, 1); |
4992 | } |
4993 | |
4994 | check_inuse_chunk (av, newp); |
4995 | return tag_new_usable (chunk2mem (newp)); |
4996 | } |
4997 | |
4998 | /* |
4999 | ------------------------------ memalign ------------------------------ |
5000 | */ |
5001 | |
5002 | /* Returns 0 if the chunk is not and does not contain the requested |
5003 | aligned sub-chunk, else returns the amount of "waste" from |
5004 | trimming. NB is the *chunk* byte size, not the user byte |
5005 | size. */ |
5006 | static size_t |
5007 | chunk_ok_for_memalign (mchunkptr p, size_t alignment, size_t nb) |
5008 | { |
5009 | void *m = chunk2mem (p); |
5010 | INTERNAL_SIZE_T size = chunksize (p); |
5011 | void *aligned_m = m; |
5012 | |
5013 | if (__glibc_unlikely (misaligned_chunk (p))) |
5014 | malloc_printerr ("_int_memalign(): unaligned chunk detected" ); |
5015 | |
5016 | aligned_m = PTR_ALIGN_UP (m, alignment); |
5017 | |
5018 | INTERNAL_SIZE_T = (intptr_t) aligned_m - (intptr_t) m; |
5019 | |
5020 | /* We can't trim off the front as it's too small. */ |
5021 | if (front_extra > 0 && front_extra < MINSIZE) |
5022 | return 0; |
5023 | |
5024 | /* If it's a perfect fit, it's an exception to the return value rule |
5025 | (we would return zero waste, which looks like "not usable"), so |
5026 | handle it here by returning a small non-zero value instead. */ |
5027 | if (size == nb && front_extra == 0) |
5028 | return 1; |
5029 | |
5030 | /* If the block we need fits in the chunk, calculate total waste. */ |
5031 | if (size > nb + front_extra) |
5032 | return size - nb; |
5033 | |
5034 | /* Can't use this chunk. */ |
5035 | return 0; |
5036 | } |
5037 | |
5038 | /* BYTES is user requested bytes, not requested chunksize bytes. */ |
5039 | static void * |
5040 | _int_memalign (mstate av, size_t alignment, size_t bytes) |
5041 | { |
5042 | INTERNAL_SIZE_T nb; /* padded request size */ |
5043 | char *m; /* memory returned by malloc call */ |
5044 | mchunkptr p; /* corresponding chunk */ |
5045 | char *brk; /* alignment point within p */ |
5046 | mchunkptr newp; /* chunk to return */ |
5047 | INTERNAL_SIZE_T newsize; /* its size */ |
5048 | INTERNAL_SIZE_T leadsize; /* leading space before alignment point */ |
5049 | mchunkptr remainder; /* spare room at end to split off */ |
5050 | unsigned long remainder_size; /* its size */ |
5051 | INTERNAL_SIZE_T size; |
5052 | mchunkptr victim; |
5053 | |
5054 | nb = checked_request2size (bytes); |
5055 | if (nb == 0) |
5056 | { |
5057 | __set_errno (ENOMEM); |
5058 | return NULL; |
5059 | } |
5060 | |
5061 | /* We can't check tcache here because we hold the arena lock, which |
5062 | tcache doesn't expect. We expect it has been checked |
5063 | earlier. */ |
5064 | |
5065 | /* Strategy: search the bins looking for an existing block that |
5066 | meets our needs. We scan a range of bins from "exact size" to |
5067 | "just under 2x", spanning the small/large barrier if needed. If |
5068 | we don't find anything in those bins, the common malloc code will |
5069 | scan starting at 2x. */ |
5070 | |
5071 | /* This will be set if we found a candidate chunk. */ |
5072 | victim = NULL; |
5073 | |
5074 | /* Fast bins are singly-linked, hard to remove a chunk from the middle |
5075 | and unlikely to meet our alignment requirements. We have not done |
5076 | any experimentation with searching for aligned fastbins. */ |
5077 | |
5078 | if (av != NULL) |
5079 | { |
5080 | int first_bin_index; |
5081 | int first_largebin_index; |
5082 | int last_bin_index; |
5083 | |
5084 | if (in_smallbin_range (nb)) |
5085 | first_bin_index = smallbin_index (nb); |
5086 | else |
5087 | first_bin_index = largebin_index (nb); |
5088 | |
5089 | if (in_smallbin_range (nb * 2)) |
5090 | last_bin_index = smallbin_index (nb * 2); |
5091 | else |
5092 | last_bin_index = largebin_index (nb * 2); |
5093 | |
5094 | first_largebin_index = largebin_index (MIN_LARGE_SIZE); |
5095 | |
5096 | int victim_index; /* its bin index */ |
5097 | |
5098 | for (victim_index = first_bin_index; |
5099 | victim_index < last_bin_index; |
5100 | victim_index ++) |
5101 | { |
5102 | victim = NULL; |
5103 | |
5104 | if (victim_index < first_largebin_index) |
5105 | { |
5106 | /* Check small bins. Small bin chunks are doubly-linked despite |
5107 | being the same size. */ |
5108 | |
5109 | mchunkptr fwd; /* misc temp for linking */ |
5110 | mchunkptr bck; /* misc temp for linking */ |
5111 | |
5112 | bck = bin_at (av, victim_index); |
5113 | fwd = bck->fd; |
5114 | while (fwd != bck) |
5115 | { |
5116 | if (chunk_ok_for_memalign (fwd, alignment, nb) > 0) |
5117 | { |
5118 | victim = fwd; |
5119 | |
5120 | /* Unlink it */ |
5121 | victim->fd->bk = victim->bk; |
5122 | victim->bk->fd = victim->fd; |
5123 | break; |
5124 | } |
5125 | |
5126 | fwd = fwd->fd; |
5127 | } |
5128 | } |
5129 | else |
5130 | { |
5131 | /* Check large bins. */ |
5132 | mchunkptr fwd; /* misc temp for linking */ |
5133 | mchunkptr bck; /* misc temp for linking */ |
5134 | mchunkptr best = NULL; |
5135 | size_t best_size = 0; |
5136 | |
5137 | bck = bin_at (av, victim_index); |
5138 | fwd = bck->fd; |
5139 | |
5140 | while (fwd != bck) |
5141 | { |
5142 | int ; |
5143 | |
5144 | if (chunksize (fwd) < nb) |
5145 | break; |
5146 | extra = chunk_ok_for_memalign (fwd, alignment, nb); |
5147 | if (extra > 0 |
5148 | && (extra <= best_size || best == NULL)) |
5149 | { |
5150 | best = fwd; |
5151 | best_size = extra; |
5152 | } |
5153 | |
5154 | fwd = fwd->fd; |
5155 | } |
5156 | victim = best; |
5157 | |
5158 | if (victim != NULL) |
5159 | { |
5160 | unlink_chunk (av, victim); |
5161 | break; |
5162 | } |
5163 | } |
5164 | |
5165 | if (victim != NULL) |
5166 | break; |
5167 | } |
5168 | } |
5169 | |
5170 | /* Strategy: find a spot within that chunk that meets the alignment |
5171 | request, and then possibly free the leading and trailing space. |
5172 | This strategy is incredibly costly and can lead to external |
5173 | fragmentation if header and footer chunks are unused. */ |
5174 | |
5175 | if (victim != NULL) |
5176 | { |
5177 | p = victim; |
5178 | m = chunk2mem (p); |
5179 | set_inuse (p); |
5180 | if (av != &main_arena) |
5181 | set_non_main_arena (p); |
5182 | } |
5183 | else |
5184 | { |
5185 | /* Call malloc with worst case padding to hit alignment. */ |
5186 | |
5187 | m = (char *) (_int_malloc (av, nb + alignment + MINSIZE)); |
5188 | |
5189 | if (m == 0) |
5190 | return 0; /* propagate failure */ |
5191 | |
5192 | p = mem2chunk (m); |
5193 | } |
5194 | |
5195 | if ((((unsigned long) (m)) % alignment) != 0) /* misaligned */ |
5196 | { |
5197 | /* Find an aligned spot inside chunk. Since we need to give back |
5198 | leading space in a chunk of at least MINSIZE, if the first |
5199 | calculation places us at a spot with less than MINSIZE leader, |
5200 | we can move to the next aligned spot -- we've allocated enough |
5201 | total room so that this is always possible. */ |
5202 | brk = (char *) mem2chunk (((unsigned long) (m + alignment - 1)) & |
5203 | - ((signed long) alignment)); |
5204 | if ((unsigned long) (brk - (char *) (p)) < MINSIZE) |
5205 | brk += alignment; |
5206 | |
5207 | newp = (mchunkptr) brk; |
5208 | leadsize = brk - (char *) (p); |
5209 | newsize = chunksize (p) - leadsize; |
5210 | |
5211 | /* For mmapped chunks, just adjust offset */ |
5212 | if (chunk_is_mmapped (p)) |
5213 | { |
5214 | set_prev_size (newp, prev_size (p) + leadsize); |
5215 | set_head (newp, newsize | IS_MMAPPED); |
5216 | return chunk2mem (newp); |
5217 | } |
5218 | |
5219 | /* Otherwise, give back leader, use the rest */ |
5220 | set_head (newp, newsize | PREV_INUSE | |
5221 | (av != &main_arena ? NON_MAIN_ARENA : 0)); |
5222 | set_inuse_bit_at_offset (newp, newsize); |
5223 | set_head_size (p, leadsize | (av != &main_arena ? NON_MAIN_ARENA : 0)); |
5224 | _int_free (av, p, 1); |
5225 | p = newp; |
5226 | |
5227 | assert (newsize >= nb && |
5228 | (((unsigned long) (chunk2mem (p))) % alignment) == 0); |
5229 | } |
5230 | |
5231 | /* Also give back spare room at the end */ |
5232 | if (!chunk_is_mmapped (p)) |
5233 | { |
5234 | size = chunksize (p); |
5235 | if ((unsigned long) (size) > (unsigned long) (nb + MINSIZE)) |
5236 | { |
5237 | remainder_size = size - nb; |
5238 | remainder = chunk_at_offset (p, nb); |
5239 | set_head (remainder, remainder_size | PREV_INUSE | |
5240 | (av != &main_arena ? NON_MAIN_ARENA : 0)); |
5241 | set_head_size (p, nb); |
5242 | _int_free (av, remainder, 1); |
5243 | } |
5244 | } |
5245 | |
5246 | check_inuse_chunk (av, p); |
5247 | return chunk2mem (p); |
5248 | } |
5249 | |
5250 | |
5251 | /* |
5252 | ------------------------------ malloc_trim ------------------------------ |
5253 | */ |
5254 | |
5255 | static int |
5256 | mtrim (mstate av, size_t pad) |
5257 | { |
5258 | /* Ensure all blocks are consolidated. */ |
5259 | malloc_consolidate (av); |
5260 | |
5261 | const size_t ps = GLRO (dl_pagesize); |
5262 | int psindex = bin_index (ps); |
5263 | const size_t psm1 = ps - 1; |
5264 | |
5265 | int result = 0; |
5266 | for (int i = 1; i < NBINS; ++i) |
5267 | if (i == 1 || i >= psindex) |
5268 | { |
5269 | mbinptr bin = bin_at (av, i); |
5270 | |
5271 | for (mchunkptr p = last (bin); p != bin; p = p->bk) |
5272 | { |
5273 | INTERNAL_SIZE_T size = chunksize (p); |
5274 | |
5275 | if (size > psm1 + sizeof (struct malloc_chunk)) |
5276 | { |
5277 | /* See whether the chunk contains at least one unused page. */ |
5278 | char *paligned_mem = (char *) (((uintptr_t) p |
5279 | + sizeof (struct malloc_chunk) |
5280 | + psm1) & ~psm1); |
5281 | |
5282 | assert ((char *) chunk2mem (p) + 2 * CHUNK_HDR_SZ |
5283 | <= paligned_mem); |
5284 | assert ((char *) p + size > paligned_mem); |
5285 | |
5286 | /* This is the size we could potentially free. */ |
5287 | size -= paligned_mem - (char *) p; |
5288 | |
5289 | if (size > psm1) |
5290 | { |
5291 | #if MALLOC_DEBUG |
5292 | /* When debugging we simulate destroying the memory |
5293 | content. */ |
5294 | memset (paligned_mem, 0x89, size & ~psm1); |
5295 | #endif |
5296 | __madvise (paligned_mem, size & ~psm1, MADV_DONTNEED); |
5297 | |
5298 | result = 1; |
5299 | } |
5300 | } |
5301 | } |
5302 | } |
5303 | |
5304 | #ifndef MORECORE_CANNOT_TRIM |
5305 | return result | (av == &main_arena ? systrim (pad, av) : 0); |
5306 | |
5307 | #else |
5308 | return result; |
5309 | #endif |
5310 | } |
5311 | |
5312 | |
5313 | int |
5314 | __malloc_trim (size_t s) |
5315 | { |
5316 | int result = 0; |
5317 | |
5318 | if (!__malloc_initialized) |
5319 | ptmalloc_init (); |
5320 | |
5321 | mstate ar_ptr = &main_arena; |
5322 | do |
5323 | { |
5324 | __libc_lock_lock (ar_ptr->mutex); |
5325 | result |= mtrim (ar_ptr, s); |
5326 | __libc_lock_unlock (ar_ptr->mutex); |
5327 | |
5328 | ar_ptr = ar_ptr->next; |
5329 | } |
5330 | while (ar_ptr != &main_arena); |
5331 | |
5332 | return result; |
5333 | } |
5334 | |
5335 | |
5336 | /* |
5337 | ------------------------- malloc_usable_size ------------------------- |
5338 | */ |
5339 | |
5340 | static size_t |
5341 | musable (void *mem) |
5342 | { |
5343 | mchunkptr p = mem2chunk (mem); |
5344 | |
5345 | if (chunk_is_mmapped (p)) |
5346 | return chunksize (p) - CHUNK_HDR_SZ; |
5347 | else if (inuse (p)) |
5348 | return memsize (p); |
5349 | |
5350 | return 0; |
5351 | } |
5352 | |
5353 | #if IS_IN (libc) |
5354 | size_t |
5355 | __malloc_usable_size (void *m) |
5356 | { |
5357 | if (m == NULL) |
5358 | return 0; |
5359 | return musable (m); |
5360 | } |
5361 | #endif |
5362 | |
5363 | /* |
5364 | ------------------------------ mallinfo ------------------------------ |
5365 | Accumulate malloc statistics for arena AV into M. |
5366 | */ |
5367 | static void |
5368 | int_mallinfo (mstate av, struct mallinfo2 *m) |
5369 | { |
5370 | size_t i; |
5371 | mbinptr b; |
5372 | mchunkptr p; |
5373 | INTERNAL_SIZE_T avail; |
5374 | INTERNAL_SIZE_T fastavail; |
5375 | int nblocks; |
5376 | int nfastblocks; |
5377 | |
5378 | check_malloc_state (av); |
5379 | |
5380 | /* Account for top */ |
5381 | avail = chunksize (av->top); |
5382 | nblocks = 1; /* top always exists */ |
5383 | |
5384 | /* traverse fastbins */ |
5385 | nfastblocks = 0; |
5386 | fastavail = 0; |
5387 | |
5388 | for (i = 0; i < NFASTBINS; ++i) |
5389 | { |
5390 | for (p = fastbin (av, i); |
5391 | p != 0; |
5392 | p = REVEAL_PTR (p->fd)) |
5393 | { |
5394 | if (__glibc_unlikely (misaligned_chunk (p))) |
5395 | malloc_printerr ("int_mallinfo(): " |
5396 | "unaligned fastbin chunk detected" ); |
5397 | ++nfastblocks; |
5398 | fastavail += chunksize (p); |
5399 | } |
5400 | } |
5401 | |
5402 | avail += fastavail; |
5403 | |
5404 | /* traverse regular bins */ |
5405 | for (i = 1; i < NBINS; ++i) |
5406 | { |
5407 | b = bin_at (av, i); |
5408 | for (p = last (b); p != b; p = p->bk) |
5409 | { |
5410 | ++nblocks; |
5411 | avail += chunksize (p); |
5412 | } |
5413 | } |
5414 | |
5415 | m->smblks += nfastblocks; |
5416 | m->ordblks += nblocks; |
5417 | m->fordblks += avail; |
5418 | m->uordblks += av->system_mem - avail; |
5419 | m->arena += av->system_mem; |
5420 | m->fsmblks += fastavail; |
5421 | if (av == &main_arena) |
5422 | { |
5423 | m->hblks = mp_.n_mmaps; |
5424 | m->hblkhd = mp_.mmapped_mem; |
5425 | m->usmblks = 0; |
5426 | m->keepcost = chunksize (av->top); |
5427 | } |
5428 | } |
5429 | |
5430 | |
5431 | struct mallinfo2 |
5432 | __libc_mallinfo2 (void) |
5433 | { |
5434 | struct mallinfo2 m; |
5435 | mstate ar_ptr; |
5436 | |
5437 | if (!__malloc_initialized) |
5438 | ptmalloc_init (); |
5439 | |
5440 | memset (&m, 0, sizeof (m)); |
5441 | ar_ptr = &main_arena; |
5442 | do |
5443 | { |
5444 | __libc_lock_lock (ar_ptr->mutex); |
5445 | int_mallinfo (ar_ptr, &m); |
5446 | __libc_lock_unlock (ar_ptr->mutex); |
5447 | |
5448 | ar_ptr = ar_ptr->next; |
5449 | } |
5450 | while (ar_ptr != &main_arena); |
5451 | |
5452 | return m; |
5453 | } |
5454 | libc_hidden_def (__libc_mallinfo2) |
5455 | |
5456 | struct mallinfo |
5457 | __libc_mallinfo (void) |
5458 | { |
5459 | struct mallinfo m; |
5460 | struct mallinfo2 m2 = __libc_mallinfo2 (); |
5461 | |
5462 | m.arena = m2.arena; |
5463 | m.ordblks = m2.ordblks; |
5464 | m.smblks = m2.smblks; |
5465 | m.hblks = m2.hblks; |
5466 | m.hblkhd = m2.hblkhd; |
5467 | m.usmblks = m2.usmblks; |
5468 | m.fsmblks = m2.fsmblks; |
5469 | m.uordblks = m2.uordblks; |
5470 | m.fordblks = m2.fordblks; |
5471 | m.keepcost = m2.keepcost; |
5472 | |
5473 | return m; |
5474 | } |
5475 | |
5476 | |
5477 | /* |
5478 | ------------------------------ malloc_stats ------------------------------ |
5479 | */ |
5480 | |
5481 | void |
5482 | __malloc_stats (void) |
5483 | { |
5484 | int i; |
5485 | mstate ar_ptr; |
5486 | unsigned int in_use_b = mp_.mmapped_mem, system_b = in_use_b; |
5487 | |
5488 | if (!__malloc_initialized) |
5489 | ptmalloc_init (); |
5490 | _IO_flockfile (stderr); |
5491 | int old_flags2 = stderr->_flags2; |
5492 | stderr->_flags2 |= _IO_FLAGS2_NOTCANCEL; |
5493 | for (i = 0, ar_ptr = &main_arena;; i++) |
5494 | { |
5495 | struct mallinfo2 mi; |
5496 | |
5497 | memset (&mi, 0, sizeof (mi)); |
5498 | __libc_lock_lock (ar_ptr->mutex); |
5499 | int_mallinfo (ar_ptr, &mi); |
5500 | fprintf (stderr, "Arena %d:\n" , i); |
5501 | fprintf (stderr, "system bytes = %10u\n" , (unsigned int) mi.arena); |
5502 | fprintf (stderr, "in use bytes = %10u\n" , (unsigned int) mi.uordblks); |
5503 | #if MALLOC_DEBUG > 1 |
5504 | if (i > 0) |
5505 | dump_heap (heap_for_ptr (top (ar_ptr))); |
5506 | #endif |
5507 | system_b += mi.arena; |
5508 | in_use_b += mi.uordblks; |
5509 | __libc_lock_unlock (ar_ptr->mutex); |
5510 | ar_ptr = ar_ptr->next; |
5511 | if (ar_ptr == &main_arena) |
5512 | break; |
5513 | } |
5514 | fprintf (stderr, "Total (incl. mmap):\n" ); |
5515 | fprintf (stderr, "system bytes = %10u\n" , system_b); |
5516 | fprintf (stderr, "in use bytes = %10u\n" , in_use_b); |
5517 | fprintf (stderr, "max mmap regions = %10u\n" , (unsigned int) mp_.max_n_mmaps); |
5518 | fprintf (stderr, "max mmap bytes = %10lu\n" , |
5519 | (unsigned long) mp_.max_mmapped_mem); |
5520 | stderr->_flags2 = old_flags2; |
5521 | _IO_funlockfile (stderr); |
5522 | } |
5523 | |
5524 | |
5525 | /* |
5526 | ------------------------------ mallopt ------------------------------ |
5527 | */ |
5528 | static __always_inline int |
5529 | do_set_trim_threshold (size_t value) |
5530 | { |
5531 | LIBC_PROBE (memory_mallopt_trim_threshold, 3, value, mp_.trim_threshold, |
5532 | mp_.no_dyn_threshold); |
5533 | mp_.trim_threshold = value; |
5534 | mp_.no_dyn_threshold = 1; |
5535 | return 1; |
5536 | } |
5537 | |
5538 | static __always_inline int |
5539 | do_set_top_pad (size_t value) |
5540 | { |
5541 | LIBC_PROBE (memory_mallopt_top_pad, 3, value, mp_.top_pad, |
5542 | mp_.no_dyn_threshold); |
5543 | mp_.top_pad = value; |
5544 | mp_.no_dyn_threshold = 1; |
5545 | return 1; |
5546 | } |
5547 | |
5548 | static __always_inline int |
5549 | do_set_mmap_threshold (size_t value) |
5550 | { |
5551 | LIBC_PROBE (memory_mallopt_mmap_threshold, 3, value, mp_.mmap_threshold, |
5552 | mp_.no_dyn_threshold); |
5553 | mp_.mmap_threshold = value; |
5554 | mp_.no_dyn_threshold = 1; |
5555 | return 1; |
5556 | } |
5557 | |
5558 | static __always_inline int |
5559 | do_set_mmaps_max (int32_t value) |
5560 | { |
5561 | LIBC_PROBE (memory_mallopt_mmap_max, 3, value, mp_.n_mmaps_max, |
5562 | mp_.no_dyn_threshold); |
5563 | mp_.n_mmaps_max = value; |
5564 | mp_.no_dyn_threshold = 1; |
5565 | return 1; |
5566 | } |
5567 | |
5568 | static __always_inline int |
5569 | do_set_mallopt_check (int32_t value) |
5570 | { |
5571 | return 1; |
5572 | } |
5573 | |
5574 | static __always_inline int |
5575 | do_set_perturb_byte (int32_t value) |
5576 | { |
5577 | LIBC_PROBE (memory_mallopt_perturb, 2, value, perturb_byte); |
5578 | perturb_byte = value; |
5579 | return 1; |
5580 | } |
5581 | |
5582 | static __always_inline int |
5583 | do_set_arena_test (size_t value) |
5584 | { |
5585 | LIBC_PROBE (memory_mallopt_arena_test, 2, value, mp_.arena_test); |
5586 | mp_.arena_test = value; |
5587 | return 1; |
5588 | } |
5589 | |
5590 | static __always_inline int |
5591 | do_set_arena_max (size_t value) |
5592 | { |
5593 | LIBC_PROBE (memory_mallopt_arena_max, 2, value, mp_.arena_max); |
5594 | mp_.arena_max = value; |
5595 | return 1; |
5596 | } |
5597 | |
5598 | #if USE_TCACHE |
5599 | static __always_inline int |
5600 | do_set_tcache_max (size_t value) |
5601 | { |
5602 | if (value <= MAX_TCACHE_SIZE) |
5603 | { |
5604 | LIBC_PROBE (memory_tunable_tcache_max_bytes, 2, value, mp_.tcache_max_bytes); |
5605 | mp_.tcache_max_bytes = value; |
5606 | mp_.tcache_bins = csize2tidx (request2size(value)) + 1; |
5607 | return 1; |
5608 | } |
5609 | return 0; |
5610 | } |
5611 | |
5612 | static __always_inline int |
5613 | do_set_tcache_count (size_t value) |
5614 | { |
5615 | if (value <= MAX_TCACHE_COUNT) |
5616 | { |
5617 | LIBC_PROBE (memory_tunable_tcache_count, 2, value, mp_.tcache_count); |
5618 | mp_.tcache_count = value; |
5619 | return 1; |
5620 | } |
5621 | return 0; |
5622 | } |
5623 | |
5624 | static __always_inline int |
5625 | do_set_tcache_unsorted_limit (size_t value) |
5626 | { |
5627 | LIBC_PROBE (memory_tunable_tcache_unsorted_limit, 2, value, mp_.tcache_unsorted_limit); |
5628 | mp_.tcache_unsorted_limit = value; |
5629 | return 1; |
5630 | } |
5631 | #endif |
5632 | |
5633 | static __always_inline int |
5634 | do_set_mxfast (size_t value) |
5635 | { |
5636 | if (value <= MAX_FAST_SIZE) |
5637 | { |
5638 | LIBC_PROBE (memory_mallopt_mxfast, 2, value, get_max_fast ()); |
5639 | set_max_fast (value); |
5640 | return 1; |
5641 | } |
5642 | return 0; |
5643 | } |
5644 | |
5645 | static __always_inline int |
5646 | do_set_hugetlb (size_t value) |
5647 | { |
5648 | if (value == 1) |
5649 | { |
5650 | enum malloc_thp_mode_t thp_mode = __malloc_thp_mode (); |
5651 | /* |
5652 | Only enable THP madvise usage if system does support it and |
5653 | has 'madvise' mode. Otherwise the madvise() call is wasteful. |
5654 | */ |
5655 | if (thp_mode == malloc_thp_mode_madvise) |
5656 | mp_.thp_pagesize = __malloc_default_thp_pagesize (); |
5657 | } |
5658 | else if (value >= 2) |
5659 | __malloc_hugepage_config (value == 2 ? 0 : value, &mp_.hp_pagesize, |
5660 | &mp_.hp_flags); |
5661 | return 0; |
5662 | } |
5663 | |
5664 | int |
5665 | __libc_mallopt (int param_number, int value) |
5666 | { |
5667 | mstate av = &main_arena; |
5668 | int res = 1; |
5669 | |
5670 | if (!__malloc_initialized) |
5671 | ptmalloc_init (); |
5672 | __libc_lock_lock (av->mutex); |
5673 | |
5674 | LIBC_PROBE (memory_mallopt, 2, param_number, value); |
5675 | |
5676 | /* We must consolidate main arena before changing max_fast |
5677 | (see definition of set_max_fast). */ |
5678 | malloc_consolidate (av); |
5679 | |
5680 | /* Many of these helper functions take a size_t. We do not worry |
5681 | about overflow here, because negative int values will wrap to |
5682 | very large size_t values and the helpers have sufficient range |
5683 | checking for such conversions. Many of these helpers are also |
5684 | used by the tunables macros in arena.c. */ |
5685 | |
5686 | switch (param_number) |
5687 | { |
5688 | case M_MXFAST: |
5689 | res = do_set_mxfast (value); |
5690 | break; |
5691 | |
5692 | case M_TRIM_THRESHOLD: |
5693 | res = do_set_trim_threshold (value); |
5694 | break; |
5695 | |
5696 | case M_TOP_PAD: |
5697 | res = do_set_top_pad (value); |
5698 | break; |
5699 | |
5700 | case M_MMAP_THRESHOLD: |
5701 | res = do_set_mmap_threshold (value); |
5702 | break; |
5703 | |
5704 | case M_MMAP_MAX: |
5705 | res = do_set_mmaps_max (value); |
5706 | break; |
5707 | |
5708 | case M_CHECK_ACTION: |
5709 | res = do_set_mallopt_check (value); |
5710 | break; |
5711 | |
5712 | case M_PERTURB: |
5713 | res = do_set_perturb_byte (value); |
5714 | break; |
5715 | |
5716 | case M_ARENA_TEST: |
5717 | if (value > 0) |
5718 | res = do_set_arena_test (value); |
5719 | break; |
5720 | |
5721 | case M_ARENA_MAX: |
5722 | if (value > 0) |
5723 | res = do_set_arena_max (value); |
5724 | break; |
5725 | } |
5726 | __libc_lock_unlock (av->mutex); |
5727 | return res; |
5728 | } |
5729 | libc_hidden_def (__libc_mallopt) |
5730 | |
5731 | |
5732 | /* |
5733 | -------------------- Alternative MORECORE functions -------------------- |
5734 | */ |
5735 | |
5736 | |
5737 | /* |
5738 | General Requirements for MORECORE. |
5739 | |
5740 | The MORECORE function must have the following properties: |
5741 | |
5742 | If MORECORE_CONTIGUOUS is false: |
5743 | |
5744 | * MORECORE must allocate in multiples of pagesize. It will |
5745 | only be called with arguments that are multiples of pagesize. |
5746 | |
5747 | * MORECORE(0) must return an address that is at least |
5748 | MALLOC_ALIGNMENT aligned. (Page-aligning always suffices.) |
5749 | |
5750 | else (i.e. If MORECORE_CONTIGUOUS is true): |
5751 | |
5752 | * Consecutive calls to MORECORE with positive arguments |
5753 | return increasing addresses, indicating that space has been |
5754 | contiguously extended. |
5755 | |
5756 | * MORECORE need not allocate in multiples of pagesize. |
5757 | Calls to MORECORE need not have args of multiples of pagesize. |
5758 | |
5759 | * MORECORE need not page-align. |
5760 | |
5761 | In either case: |
5762 | |
5763 | * MORECORE may allocate more memory than requested. (Or even less, |
5764 | but this will generally result in a malloc failure.) |
5765 | |
5766 | * MORECORE must not allocate memory when given argument zero, but |
5767 | instead return one past the end address of memory from previous |
5768 | nonzero call. This malloc does NOT call MORECORE(0) |
5769 | until at least one call with positive arguments is made, so |
5770 | the initial value returned is not important. |
5771 | |
5772 | * Even though consecutive calls to MORECORE need not return contiguous |
5773 | addresses, it must be OK for malloc'ed chunks to span multiple |
5774 | regions in those cases where they do happen to be contiguous. |
5775 | |
5776 | * MORECORE need not handle negative arguments -- it may instead |
5777 | just return MORECORE_FAILURE when given negative arguments. |
5778 | Negative arguments are always multiples of pagesize. MORECORE |
5779 | must not misinterpret negative args as large positive unsigned |
5780 | args. You can suppress all such calls from even occurring by defining |
5781 | MORECORE_CANNOT_TRIM, |
5782 | |
5783 | There is some variation across systems about the type of the |
5784 | argument to sbrk/MORECORE. If size_t is unsigned, then it cannot |
5785 | actually be size_t, because sbrk supports negative args, so it is |
5786 | normally the signed type of the same width as size_t (sometimes |
5787 | declared as "intptr_t", and sometimes "ptrdiff_t"). It doesn't much |
5788 | matter though. Internally, we use "long" as arguments, which should |
5789 | work across all reasonable possibilities. |
5790 | |
5791 | Additionally, if MORECORE ever returns failure for a positive |
5792 | request, then mmap is used as a noncontiguous system allocator. This |
5793 | is a useful backup strategy for systems with holes in address spaces |
5794 | -- in this case sbrk cannot contiguously expand the heap, but mmap |
5795 | may be able to map noncontiguous space. |
5796 | |
5797 | If you'd like mmap to ALWAYS be used, you can define MORECORE to be |
5798 | a function that always returns MORECORE_FAILURE. |
5799 | |
5800 | If you are using this malloc with something other than sbrk (or its |
5801 | emulation) to supply memory regions, you probably want to set |
5802 | MORECORE_CONTIGUOUS as false. As an example, here is a custom |
5803 | allocator kindly contributed for pre-OSX macOS. It uses virtually |
5804 | but not necessarily physically contiguous non-paged memory (locked |
5805 | in, present and won't get swapped out). You can use it by |
5806 | uncommenting this section, adding some #includes, and setting up the |
5807 | appropriate defines above: |
5808 | |
5809 | *#define MORECORE osMoreCore |
5810 | *#define MORECORE_CONTIGUOUS 0 |
5811 | |
5812 | There is also a shutdown routine that should somehow be called for |
5813 | cleanup upon program exit. |
5814 | |
5815 | *#define MAX_POOL_ENTRIES 100 |
5816 | *#define MINIMUM_MORECORE_SIZE (64 * 1024) |
5817 | static int next_os_pool; |
5818 | void *our_os_pools[MAX_POOL_ENTRIES]; |
5819 | |
5820 | void *osMoreCore(int size) |
5821 | { |
5822 | void *ptr = 0; |
5823 | static void *sbrk_top = 0; |
5824 | |
5825 | if (size > 0) |
5826 | { |
5827 | if (size < MINIMUM_MORECORE_SIZE) |
5828 | size = MINIMUM_MORECORE_SIZE; |
5829 | if (CurrentExecutionLevel() == kTaskLevel) |
5830 | ptr = PoolAllocateResident(size + RM_PAGE_SIZE, 0); |
5831 | if (ptr == 0) |
5832 | { |
5833 | return (void *) MORECORE_FAILURE; |
5834 | } |
5835 | // save ptrs so they can be freed during cleanup |
5836 | our_os_pools[next_os_pool] = ptr; |
5837 | next_os_pool++; |
5838 | ptr = (void *) ((((unsigned long) ptr) + RM_PAGE_MASK) & ~RM_PAGE_MASK); |
5839 | sbrk_top = (char *) ptr + size; |
5840 | return ptr; |
5841 | } |
5842 | else if (size < 0) |
5843 | { |
5844 | // we don't currently support shrink behavior |
5845 | return (void *) MORECORE_FAILURE; |
5846 | } |
5847 | else |
5848 | { |
5849 | return sbrk_top; |
5850 | } |
5851 | } |
5852 | |
5853 | // cleanup any allocated memory pools |
5854 | // called as last thing before shutting down driver |
5855 | |
5856 | void osCleanupMem(void) |
5857 | { |
5858 | void **ptr; |
5859 | |
5860 | for (ptr = our_os_pools; ptr < &our_os_pools[MAX_POOL_ENTRIES]; ptr++) |
5861 | if (*ptr) |
5862 | { |
5863 | PoolDeallocate(*ptr); |
5864 | * ptr = 0; |
5865 | } |
5866 | } |
5867 | |
5868 | */ |
5869 | |
5870 | |
5871 | /* Helper code. */ |
5872 | |
5873 | extern char **__libc_argv attribute_hidden; |
5874 | |
5875 | static void |
5876 | malloc_printerr (const char *str) |
5877 | { |
5878 | #if IS_IN (libc) |
5879 | __libc_message ("%s\n" , str); |
5880 | #else |
5881 | __libc_fatal (str); |
5882 | #endif |
5883 | __builtin_unreachable (); |
5884 | } |
5885 | |
5886 | #if IS_IN (libc) |
5887 | /* We need a wrapper function for one of the additions of POSIX. */ |
5888 | int |
5889 | __posix_memalign (void **memptr, size_t alignment, size_t size) |
5890 | { |
5891 | void *mem; |
5892 | |
5893 | if (!__malloc_initialized) |
5894 | ptmalloc_init (); |
5895 | |
5896 | /* Test whether the SIZE argument is valid. It must be a power of |
5897 | two multiple of sizeof (void *). */ |
5898 | if (alignment % sizeof (void *) != 0 |
5899 | || !powerof2 (alignment / sizeof (void *)) |
5900 | || alignment == 0) |
5901 | return EINVAL; |
5902 | |
5903 | |
5904 | void *address = RETURN_ADDRESS (0); |
5905 | mem = _mid_memalign (alignment, size, address); |
5906 | |
5907 | if (mem != NULL) |
5908 | { |
5909 | *memptr = mem; |
5910 | return 0; |
5911 | } |
5912 | |
5913 | return ENOMEM; |
5914 | } |
5915 | weak_alias (__posix_memalign, posix_memalign) |
5916 | #endif |
5917 | |
5918 | |
5919 | int |
5920 | __malloc_info (int options, FILE *fp) |
5921 | { |
5922 | /* For now, at least. */ |
5923 | if (options != 0) |
5924 | return EINVAL; |
5925 | |
5926 | int n = 0; |
5927 | size_t total_nblocks = 0; |
5928 | size_t total_nfastblocks = 0; |
5929 | size_t total_avail = 0; |
5930 | size_t total_fastavail = 0; |
5931 | size_t total_system = 0; |
5932 | size_t total_max_system = 0; |
5933 | size_t total_aspace = 0; |
5934 | size_t total_aspace_mprotect = 0; |
5935 | |
5936 | |
5937 | |
5938 | if (!__malloc_initialized) |
5939 | ptmalloc_init (); |
5940 | |
5941 | fputs ("<malloc version=\"1\">\n" , fp); |
5942 | |
5943 | /* Iterate over all arenas currently in use. */ |
5944 | mstate ar_ptr = &main_arena; |
5945 | do |
5946 | { |
5947 | fprintf (fp, "<heap nr=\"%d\">\n<sizes>\n" , n++); |
5948 | |
5949 | size_t nblocks = 0; |
5950 | size_t nfastblocks = 0; |
5951 | size_t avail = 0; |
5952 | size_t fastavail = 0; |
5953 | struct |
5954 | { |
5955 | size_t from; |
5956 | size_t to; |
5957 | size_t total; |
5958 | size_t count; |
5959 | } sizes[NFASTBINS + NBINS - 1]; |
5960 | #define nsizes (sizeof (sizes) / sizeof (sizes[0])) |
5961 | |
5962 | __libc_lock_lock (ar_ptr->mutex); |
5963 | |
5964 | /* Account for top chunk. The top-most available chunk is |
5965 | treated specially and is never in any bin. See "initial_top" |
5966 | comments. */ |
5967 | avail = chunksize (ar_ptr->top); |
5968 | nblocks = 1; /* Top always exists. */ |
5969 | |
5970 | for (size_t i = 0; i < NFASTBINS; ++i) |
5971 | { |
5972 | mchunkptr p = fastbin (ar_ptr, i); |
5973 | if (p != NULL) |
5974 | { |
5975 | size_t nthissize = 0; |
5976 | size_t thissize = chunksize (p); |
5977 | |
5978 | while (p != NULL) |
5979 | { |
5980 | if (__glibc_unlikely (misaligned_chunk (p))) |
5981 | malloc_printerr ("__malloc_info(): " |
5982 | "unaligned fastbin chunk detected" ); |
5983 | ++nthissize; |
5984 | p = REVEAL_PTR (p->fd); |
5985 | } |
5986 | |
5987 | fastavail += nthissize * thissize; |
5988 | nfastblocks += nthissize; |
5989 | sizes[i].from = thissize - (MALLOC_ALIGNMENT - 1); |
5990 | sizes[i].to = thissize; |
5991 | sizes[i].count = nthissize; |
5992 | } |
5993 | else |
5994 | sizes[i].from = sizes[i].to = sizes[i].count = 0; |
5995 | |
5996 | sizes[i].total = sizes[i].count * sizes[i].to; |
5997 | } |
5998 | |
5999 | |
6000 | mbinptr bin; |
6001 | struct malloc_chunk *r; |
6002 | |
6003 | for (size_t i = 1; i < NBINS; ++i) |
6004 | { |
6005 | bin = bin_at (ar_ptr, i); |
6006 | r = bin->fd; |
6007 | sizes[NFASTBINS - 1 + i].from = ~((size_t) 0); |
6008 | sizes[NFASTBINS - 1 + i].to = sizes[NFASTBINS - 1 + i].total |
6009 | = sizes[NFASTBINS - 1 + i].count = 0; |
6010 | |
6011 | if (r != NULL) |
6012 | while (r != bin) |
6013 | { |
6014 | size_t r_size = chunksize_nomask (r); |
6015 | ++sizes[NFASTBINS - 1 + i].count; |
6016 | sizes[NFASTBINS - 1 + i].total += r_size; |
6017 | sizes[NFASTBINS - 1 + i].from |
6018 | = MIN (sizes[NFASTBINS - 1 + i].from, r_size); |
6019 | sizes[NFASTBINS - 1 + i].to = MAX (sizes[NFASTBINS - 1 + i].to, |
6020 | r_size); |
6021 | |
6022 | r = r->fd; |
6023 | } |
6024 | |
6025 | if (sizes[NFASTBINS - 1 + i].count == 0) |
6026 | sizes[NFASTBINS - 1 + i].from = 0; |
6027 | nblocks += sizes[NFASTBINS - 1 + i].count; |
6028 | avail += sizes[NFASTBINS - 1 + i].total; |
6029 | } |
6030 | |
6031 | size_t heap_size = 0; |
6032 | size_t heap_mprotect_size = 0; |
6033 | size_t heap_count = 0; |
6034 | if (ar_ptr != &main_arena) |
6035 | { |
6036 | /* Iterate over the arena heaps from back to front. */ |
6037 | heap_info *heap = heap_for_ptr (top (ar_ptr)); |
6038 | do |
6039 | { |
6040 | heap_size += heap->size; |
6041 | heap_mprotect_size += heap->mprotect_size; |
6042 | heap = heap->prev; |
6043 | ++heap_count; |
6044 | } |
6045 | while (heap != NULL); |
6046 | } |
6047 | |
6048 | __libc_lock_unlock (ar_ptr->mutex); |
6049 | |
6050 | total_nfastblocks += nfastblocks; |
6051 | total_fastavail += fastavail; |
6052 | |
6053 | total_nblocks += nblocks; |
6054 | total_avail += avail; |
6055 | |
6056 | for (size_t i = 0; i < nsizes; ++i) |
6057 | if (sizes[i].count != 0 && i != NFASTBINS) |
6058 | fprintf (fp, "\ |
6059 | <size from=\"%zu\" to=\"%zu\" total=\"%zu\" count=\"%zu\"/>\n" , |
6060 | sizes[i].from, sizes[i].to, sizes[i].total, sizes[i].count); |
6061 | |
6062 | if (sizes[NFASTBINS].count != 0) |
6063 | fprintf (fp, "\ |
6064 | <unsorted from=\"%zu\" to=\"%zu\" total=\"%zu\" count=\"%zu\"/>\n" , |
6065 | sizes[NFASTBINS].from, sizes[NFASTBINS].to, |
6066 | sizes[NFASTBINS].total, sizes[NFASTBINS].count); |
6067 | |
6068 | total_system += ar_ptr->system_mem; |
6069 | total_max_system += ar_ptr->max_system_mem; |
6070 | |
6071 | fprintf (fp, |
6072 | "</sizes>\n<total type=\"fast\" count=\"%zu\" size=\"%zu\"/>\n" |
6073 | "<total type=\"rest\" count=\"%zu\" size=\"%zu\"/>\n" |
6074 | "<system type=\"current\" size=\"%zu\"/>\n" |
6075 | "<system type=\"max\" size=\"%zu\"/>\n" , |
6076 | nfastblocks, fastavail, nblocks, avail, |
6077 | ar_ptr->system_mem, ar_ptr->max_system_mem); |
6078 | |
6079 | if (ar_ptr != &main_arena) |
6080 | { |
6081 | fprintf (fp, |
6082 | "<aspace type=\"total\" size=\"%zu\"/>\n" |
6083 | "<aspace type=\"mprotect\" size=\"%zu\"/>\n" |
6084 | "<aspace type=\"subheaps\" size=\"%zu\"/>\n" , |
6085 | heap_size, heap_mprotect_size, heap_count); |
6086 | total_aspace += heap_size; |
6087 | total_aspace_mprotect += heap_mprotect_size; |
6088 | } |
6089 | else |
6090 | { |
6091 | fprintf (fp, |
6092 | "<aspace type=\"total\" size=\"%zu\"/>\n" |
6093 | "<aspace type=\"mprotect\" size=\"%zu\"/>\n" , |
6094 | ar_ptr->system_mem, ar_ptr->system_mem); |
6095 | total_aspace += ar_ptr->system_mem; |
6096 | total_aspace_mprotect += ar_ptr->system_mem; |
6097 | } |
6098 | |
6099 | fputs ("</heap>\n" , fp); |
6100 | ar_ptr = ar_ptr->next; |
6101 | } |
6102 | while (ar_ptr != &main_arena); |
6103 | |
6104 | fprintf (fp, |
6105 | "<total type=\"fast\" count=\"%zu\" size=\"%zu\"/>\n" |
6106 | "<total type=\"rest\" count=\"%zu\" size=\"%zu\"/>\n" |
6107 | "<total type=\"mmap\" count=\"%d\" size=\"%zu\"/>\n" |
6108 | "<system type=\"current\" size=\"%zu\"/>\n" |
6109 | "<system type=\"max\" size=\"%zu\"/>\n" |
6110 | "<aspace type=\"total\" size=\"%zu\"/>\n" |
6111 | "<aspace type=\"mprotect\" size=\"%zu\"/>\n" |
6112 | "</malloc>\n" , |
6113 | total_nfastblocks, total_fastavail, total_nblocks, total_avail, |
6114 | mp_.n_mmaps, mp_.mmapped_mem, |
6115 | total_system, total_max_system, |
6116 | total_aspace, total_aspace_mprotect); |
6117 | |
6118 | return 0; |
6119 | } |
6120 | #if IS_IN (libc) |
6121 | weak_alias (__malloc_info, malloc_info) |
6122 | |
6123 | strong_alias (__libc_calloc, __calloc) weak_alias (__libc_calloc, calloc) |
6124 | strong_alias (__libc_free, __free) strong_alias (__libc_free, free) |
6125 | strong_alias (__libc_malloc, __malloc) strong_alias (__libc_malloc, malloc) |
6126 | strong_alias (__libc_memalign, __memalign) |
6127 | weak_alias (__libc_memalign, memalign) |
6128 | strong_alias (__libc_realloc, __realloc) strong_alias (__libc_realloc, realloc) |
6129 | strong_alias (__libc_valloc, __valloc) weak_alias (__libc_valloc, valloc) |
6130 | strong_alias (__libc_pvalloc, __pvalloc) weak_alias (__libc_pvalloc, pvalloc) |
6131 | strong_alias (__libc_mallinfo, __mallinfo) |
6132 | weak_alias (__libc_mallinfo, mallinfo) |
6133 | strong_alias (__libc_mallinfo2, __mallinfo2) |
6134 | weak_alias (__libc_mallinfo2, mallinfo2) |
6135 | strong_alias (__libc_mallopt, __mallopt) weak_alias (__libc_mallopt, mallopt) |
6136 | |
6137 | weak_alias (__malloc_stats, malloc_stats) |
6138 | weak_alias (__malloc_usable_size, malloc_usable_size) |
6139 | weak_alias (__malloc_trim, malloc_trim) |
6140 | #endif |
6141 | |
6142 | #if SHLIB_COMPAT (libc, GLIBC_2_0, GLIBC_2_26) |
6143 | compat_symbol (libc, __libc_free, cfree, GLIBC_2_0); |
6144 | #endif |
6145 | |
6146 | /* ------------------------------------------------------------ |
6147 | History: |
6148 | |
6149 | [see ftp://g.oswego.edu/pub/misc/malloc.c for the history of dlmalloc] |
6150 | |
6151 | */ |
6152 | /* |
6153 | * Local variables: |
6154 | * c-basic-offset: 2 |
6155 | * End: |
6156 | */ |
6157 | |