hurd: Fix build with GCC 9
[glibc.git] / malloc / malloc.c
blob6d7a6a8cabb4edbf00881cb7503473a8ed4ec0b7
1 /* Malloc implementation for multiple threads without lock contention.
2 Copyright (C) 1996-2018 Free Software Foundation, Inc.
3 This file is part of the GNU C Library.
4 Contributed by Wolfram Gloger <wg@malloc.de>
5 and Doug Lea <dl@cs.oswego.edu>, 2001.
7 The GNU C Library is free software; you can redistribute it and/or
8 modify it under the terms of the GNU Lesser General Public License as
9 published by the Free Software Foundation; either version 2.1 of the
10 License, or (at your option) any later version.
12 The GNU C Library is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 Lesser General Public License for more details.
17 You should have received a copy of the GNU Lesser General Public
18 License along with the GNU C Library; see the file COPYING.LIB. If
19 not, see <http://www.gnu.org/licenses/>. */
22 This is a version (aka ptmalloc2) of malloc/free/realloc written by
23 Doug Lea and adapted to multiple threads/arenas by Wolfram Gloger.
25 There have been substantial changes made after the integration into
26 glibc in all parts of the code. Do not look for much commonality
27 with the ptmalloc2 version.
29 * Version ptmalloc2-20011215
30 based on:
31 VERSION 2.7.0 Sun Mar 11 14:14:06 2001 Doug Lea (dl at gee)
33 * Quickstart
35 In order to compile this implementation, a Makefile is provided with
36 the ptmalloc2 distribution, which has pre-defined targets for some
37 popular systems (e.g. "make posix" for Posix threads). All that is
38 typically required with regard to compiler flags is the selection of
39 the thread package via defining one out of USE_PTHREADS, USE_THR or
40 USE_SPROC. Check the thread-m.h file for what effects this has.
41 Many/most systems will additionally require USE_TSD_DATA_HACK to be
42 defined, so this is the default for "make posix".
44 * Why use this malloc?
46 This is not the fastest, most space-conserving, most portable, or
47 most tunable malloc ever written. However it is among the fastest
48 while also being among the most space-conserving, portable and tunable.
49 Consistent balance across these factors results in a good general-purpose
50 allocator for malloc-intensive programs.
52 The main properties of the algorithms are:
53 * For large (>= 512 bytes) requests, it is a pure best-fit allocator,
54 with ties normally decided via FIFO (i.e. least recently used).
55 * For small (<= 64 bytes by default) requests, it is a caching
56 allocator, that maintains pools of quickly recycled chunks.
57 * In between, and for combinations of large and small requests, it does
58 the best it can trying to meet both goals at once.
59 * For very large requests (>= 128KB by default), it relies on system
60 memory mapping facilities, if supported.
62 For a longer but slightly out of date high-level description, see
63 http://gee.cs.oswego.edu/dl/html/malloc.html
65 You may already by default be using a C library containing a malloc
66 that is based on some version of this malloc (for example in
67 linux). You might still want to use the one in this file in order to
68 customize settings or to avoid overheads associated with library
69 versions.
71 * Contents, described in more detail in "description of public routines" below.
73 Standard (ANSI/SVID/...) functions:
74 malloc(size_t n);
75 calloc(size_t n_elements, size_t element_size);
76 free(void* p);
77 realloc(void* p, size_t n);
78 memalign(size_t alignment, size_t n);
79 valloc(size_t n);
80 mallinfo()
81 mallopt(int parameter_number, int parameter_value)
83 Additional functions:
84 independent_calloc(size_t n_elements, size_t size, void* chunks[]);
85 independent_comalloc(size_t n_elements, size_t sizes[], void* chunks[]);
86 pvalloc(size_t n);
87 malloc_trim(size_t pad);
88 malloc_usable_size(void* p);
89 malloc_stats();
91 * Vital statistics:
93 Supported pointer representation: 4 or 8 bytes
94 Supported size_t representation: 4 or 8 bytes
95 Note that size_t is allowed to be 4 bytes even if pointers are 8.
96 You can adjust this by defining INTERNAL_SIZE_T
98 Alignment: 2 * sizeof(size_t) (default)
99 (i.e., 8 byte alignment with 4byte size_t). This suffices for
100 nearly all current machines and C compilers. However, you can
101 define MALLOC_ALIGNMENT to be wider than this if necessary.
103 Minimum overhead per allocated chunk: 4 or 8 bytes
104 Each malloced chunk has a hidden word of overhead holding size
105 and status information.
107 Minimum allocated size: 4-byte ptrs: 16 bytes (including 4 overhead)
108 8-byte ptrs: 24/32 bytes (including, 4/8 overhead)
110 When a chunk is freed, 12 (for 4byte ptrs) or 20 (for 8 byte
111 ptrs but 4 byte size) or 24 (for 8/8) additional bytes are
112 needed; 4 (8) for a trailing size field and 8 (16) bytes for
113 free list pointers. Thus, the minimum allocatable size is
114 16/24/32 bytes.
116 Even a request for zero bytes (i.e., malloc(0)) returns a
117 pointer to something of the minimum allocatable size.
119 The maximum overhead wastage (i.e., number of extra bytes
120 allocated than were requested in malloc) is less than or equal
121 to the minimum size, except for requests >= mmap_threshold that
122 are serviced via mmap(), where the worst case wastage is 2 *
123 sizeof(size_t) bytes plus the remainder from a system page (the
124 minimal mmap unit); typically 4096 or 8192 bytes.
126 Maximum allocated size: 4-byte size_t: 2^32 minus about two pages
127 8-byte size_t: 2^64 minus about two pages
129 It is assumed that (possibly signed) size_t values suffice to
130 represent chunk sizes. `Possibly signed' is due to the fact
131 that `size_t' may be defined on a system as either a signed or
132 an unsigned type. The ISO C standard says that it must be
133 unsigned, but a few systems are known not to adhere to this.
134 Additionally, even when size_t is unsigned, sbrk (which is by
135 default used to obtain memory from system) accepts signed
136 arguments, and may not be able to handle size_t-wide arguments
137 with negative sign bit. Generally, values that would
138 appear as negative after accounting for overhead and alignment
139 are supported only via mmap(), which does not have this
140 limitation.
142 Requests for sizes outside the allowed range will perform an optional
143 failure action and then return null. (Requests may also
144 also fail because a system is out of memory.)
146 Thread-safety: thread-safe
148 Compliance: I believe it is compliant with the 1997 Single Unix Specification
149 Also SVID/XPG, ANSI C, and probably others as well.
151 * Synopsis of compile-time options:
153 People have reported using previous versions of this malloc on all
154 versions of Unix, sometimes by tweaking some of the defines
155 below. It has been tested most extensively on Solaris and Linux.
156 People also report using it in stand-alone embedded systems.
158 The implementation is in straight, hand-tuned ANSI C. It is not
159 at all modular. (Sorry!) It uses a lot of macros. To be at all
160 usable, this code should be compiled using an optimizing compiler
161 (for example gcc -O3) that can simplify expressions and control
162 paths. (FAQ: some macros import variables as arguments rather than
163 declare locals because people reported that some debuggers
164 otherwise get confused.)
166 OPTION DEFAULT VALUE
168 Compilation Environment options:
170 HAVE_MREMAP 0
172 Changing default word sizes:
174 INTERNAL_SIZE_T size_t
176 Configuration and functionality options:
178 USE_PUBLIC_MALLOC_WRAPPERS NOT defined
179 USE_MALLOC_LOCK NOT defined
180 MALLOC_DEBUG NOT defined
181 REALLOC_ZERO_BYTES_FREES 1
182 TRIM_FASTBINS 0
184 Options for customizing MORECORE:
186 MORECORE sbrk
187 MORECORE_FAILURE -1
188 MORECORE_CONTIGUOUS 1
189 MORECORE_CANNOT_TRIM NOT defined
190 MORECORE_CLEARS 1
191 MMAP_AS_MORECORE_SIZE (1024 * 1024)
193 Tuning options that are also dynamically changeable via mallopt:
195 DEFAULT_MXFAST 64 (for 32bit), 128 (for 64bit)
196 DEFAULT_TRIM_THRESHOLD 128 * 1024
197 DEFAULT_TOP_PAD 0
198 DEFAULT_MMAP_THRESHOLD 128 * 1024
199 DEFAULT_MMAP_MAX 65536
201 There are several other #defined constants and macros that you
202 probably don't want to touch unless you are extending or adapting malloc. */
205 void* is the pointer type that malloc should say it returns
208 #ifndef void
209 #define void void
210 #endif /*void*/
212 #include <stddef.h> /* for size_t */
213 #include <stdlib.h> /* for getenv(), abort() */
214 #include <unistd.h> /* for __libc_enable_secure */
216 #include <atomic.h>
217 #include <_itoa.h>
218 #include <bits/wordsize.h>
219 #include <sys/sysinfo.h>
221 #include <ldsodefs.h>
223 #include <unistd.h>
224 #include <stdio.h> /* needed for malloc_stats */
225 #include <errno.h>
226 #include <assert.h>
228 #include <shlib-compat.h>
230 /* For uintptr_t. */
231 #include <stdint.h>
233 /* For va_arg, va_start, va_end. */
234 #include <stdarg.h>
236 /* For MIN, MAX, powerof2. */
237 #include <sys/param.h>
239 /* For ALIGN_UP et. al. */
240 #include <libc-pointer-arith.h>
242 /* For DIAG_PUSH/POP_NEEDS_COMMENT et al. */
243 #include <libc-diag.h>
245 #include <malloc/malloc-internal.h>
247 /* For SINGLE_THREAD_P. */
248 #include <sysdep-cancel.h>
251 Debugging:
253 Because freed chunks may be overwritten with bookkeeping fields, this
254 malloc will often die when freed memory is overwritten by user
255 programs. This can be very effective (albeit in an annoying way)
256 in helping track down dangling pointers.
258 If you compile with -DMALLOC_DEBUG, a number of assertion checks are
259 enabled that will catch more memory errors. You probably won't be
260 able to make much sense of the actual assertion errors, but they
261 should help you locate incorrectly overwritten memory. The checking
262 is fairly extensive, and will slow down execution
263 noticeably. Calling malloc_stats or mallinfo with MALLOC_DEBUG set
264 will attempt to check every non-mmapped allocated and free chunk in
265 the course of computing the summmaries. (By nature, mmapped regions
266 cannot be checked very much automatically.)
268 Setting MALLOC_DEBUG may also be helpful if you are trying to modify
269 this code. The assertions in the check routines spell out in more
270 detail the assumptions and invariants underlying the algorithms.
272 Setting MALLOC_DEBUG does NOT provide an automated mechanism for
273 checking that all accesses to malloced memory stay within their
274 bounds. However, there are several add-ons and adaptations of this
275 or other mallocs available that do this.
278 #ifndef MALLOC_DEBUG
279 #define MALLOC_DEBUG 0
280 #endif
282 #ifndef NDEBUG
283 # define __assert_fail(assertion, file, line, function) \
284 __malloc_assert(assertion, file, line, function)
286 extern const char *__progname;
288 static void
289 __malloc_assert (const char *assertion, const char *file, unsigned int line,
290 const char *function)
292 (void) __fxprintf (NULL, "%s%s%s:%u: %s%sAssertion `%s' failed.\n",
293 __progname, __progname[0] ? ": " : "",
294 file, line,
295 function ? function : "", function ? ": " : "",
296 assertion);
297 fflush (stderr);
298 abort ();
300 #endif
302 #if USE_TCACHE
303 /* We want 64 entries. This is an arbitrary limit, which tunables can reduce. */
304 # define TCACHE_MAX_BINS 64
305 # define MAX_TCACHE_SIZE tidx2usize (TCACHE_MAX_BINS-1)
307 /* Only used to pre-fill the tunables. */
308 # define tidx2usize(idx) (((size_t) idx) * MALLOC_ALIGNMENT + MINSIZE - SIZE_SZ)
310 /* When "x" is from chunksize(). */
311 # define csize2tidx(x) (((x) - MINSIZE + MALLOC_ALIGNMENT - 1) / MALLOC_ALIGNMENT)
312 /* When "x" is a user-provided size. */
313 # define usize2tidx(x) csize2tidx (request2size (x))
315 /* With rounding and alignment, the bins are...
316 idx 0 bytes 0..24 (64-bit) or 0..12 (32-bit)
317 idx 1 bytes 25..40 or 13..20
318 idx 2 bytes 41..56 or 21..28
319 etc. */
321 /* This is another arbitrary limit, which tunables can change. Each
322 tcache bin will hold at most this number of chunks. */
323 # define TCACHE_FILL_COUNT 7
324 #endif
328 REALLOC_ZERO_BYTES_FREES should be set if a call to
329 realloc with zero bytes should be the same as a call to free.
330 This is required by the C standard. Otherwise, since this malloc
331 returns a unique pointer for malloc(0), so does realloc(p, 0).
334 #ifndef REALLOC_ZERO_BYTES_FREES
335 #define REALLOC_ZERO_BYTES_FREES 1
336 #endif
339 TRIM_FASTBINS controls whether free() of a very small chunk can
340 immediately lead to trimming. Setting to true (1) can reduce memory
341 footprint, but will almost always slow down programs that use a lot
342 of small chunks.
344 Define this only if you are willing to give up some speed to more
345 aggressively reduce system-level memory footprint when releasing
346 memory in programs that use many small chunks. You can get
347 essentially the same effect by setting MXFAST to 0, but this can
348 lead to even greater slowdowns in programs using many small chunks.
349 TRIM_FASTBINS is an in-between compile-time option, that disables
350 only those chunks bordering topmost memory from being placed in
351 fastbins.
354 #ifndef TRIM_FASTBINS
355 #define TRIM_FASTBINS 0
356 #endif
359 /* Definition for getting more memory from the OS. */
360 #define MORECORE (*__morecore)
361 #define MORECORE_FAILURE 0
362 void * __default_morecore (ptrdiff_t);
363 void *(*__morecore)(ptrdiff_t) = __default_morecore;
366 #include <string.h>
369 MORECORE-related declarations. By default, rely on sbrk
374 MORECORE is the name of the routine to call to obtain more memory
375 from the system. See below for general guidance on writing
376 alternative MORECORE functions, as well as a version for WIN32 and a
377 sample version for pre-OSX macos.
380 #ifndef MORECORE
381 #define MORECORE sbrk
382 #endif
385 MORECORE_FAILURE is the value returned upon failure of MORECORE
386 as well as mmap. Since it cannot be an otherwise valid memory address,
387 and must reflect values of standard sys calls, you probably ought not
388 try to redefine it.
391 #ifndef MORECORE_FAILURE
392 #define MORECORE_FAILURE (-1)
393 #endif
396 If MORECORE_CONTIGUOUS is true, take advantage of fact that
397 consecutive calls to MORECORE with positive arguments always return
398 contiguous increasing addresses. This is true of unix sbrk. Even
399 if not defined, when regions happen to be contiguous, malloc will
400 permit allocations spanning regions obtained from different
401 calls. But defining this when applicable enables some stronger
402 consistency checks and space efficiencies.
405 #ifndef MORECORE_CONTIGUOUS
406 #define MORECORE_CONTIGUOUS 1
407 #endif
410 Define MORECORE_CANNOT_TRIM if your version of MORECORE
411 cannot release space back to the system when given negative
412 arguments. This is generally necessary only if you are using
413 a hand-crafted MORECORE function that cannot handle negative arguments.
416 /* #define MORECORE_CANNOT_TRIM */
418 /* MORECORE_CLEARS (default 1)
419 The degree to which the routine mapped to MORECORE zeroes out
420 memory: never (0), only for newly allocated space (1) or always
421 (2). The distinction between (1) and (2) is necessary because on
422 some systems, if the application first decrements and then
423 increments the break value, the contents of the reallocated space
424 are unspecified.
427 #ifndef MORECORE_CLEARS
428 # define MORECORE_CLEARS 1
429 #endif
433 MMAP_AS_MORECORE_SIZE is the minimum mmap size argument to use if
434 sbrk fails, and mmap is used as a backup. The value must be a
435 multiple of page size. This backup strategy generally applies only
436 when systems have "holes" in address space, so sbrk cannot perform
437 contiguous expansion, but there is still space available on system.
438 On systems for which this is known to be useful (i.e. most linux
439 kernels), this occurs only when programs allocate huge amounts of
440 memory. Between this, and the fact that mmap regions tend to be
441 limited, the size should be large, to avoid too many mmap calls and
442 thus avoid running out of kernel resources. */
444 #ifndef MMAP_AS_MORECORE_SIZE
445 #define MMAP_AS_MORECORE_SIZE (1024 * 1024)
446 #endif
449 Define HAVE_MREMAP to make realloc() use mremap() to re-allocate
450 large blocks.
453 #ifndef HAVE_MREMAP
454 #define HAVE_MREMAP 0
455 #endif
457 /* We may need to support __malloc_initialize_hook for backwards
458 compatibility. */
460 #if SHLIB_COMPAT (libc, GLIBC_2_0, GLIBC_2_24)
461 # define HAVE_MALLOC_INIT_HOOK 1
462 #else
463 # define HAVE_MALLOC_INIT_HOOK 0
464 #endif
468 This version of malloc supports the standard SVID/XPG mallinfo
469 routine that returns a struct containing usage properties and
470 statistics. It should work on any SVID/XPG compliant system that has
471 a /usr/include/malloc.h defining struct mallinfo. (If you'd like to
472 install such a thing yourself, cut out the preliminary declarations
473 as described above and below and save them in a malloc.h file. But
474 there's no compelling reason to bother to do this.)
476 The main declaration needed is the mallinfo struct that is returned
477 (by-copy) by mallinfo(). The SVID/XPG malloinfo struct contains a
478 bunch of fields that are not even meaningful in this version of
479 malloc. These fields are are instead filled by mallinfo() with
480 other numbers that might be of interest.
484 /* ---------- description of public routines ------------ */
487 malloc(size_t n)
488 Returns a pointer to a newly allocated chunk of at least n bytes, or null
489 if no space is available. Additionally, on failure, errno is
490 set to ENOMEM on ANSI C systems.
492 If n is zero, malloc returns a minumum-sized chunk. (The minimum
493 size is 16 bytes on most 32bit systems, and 24 or 32 bytes on 64bit
494 systems.) On most systems, size_t is an unsigned type, so calls
495 with negative arguments are interpreted as requests for huge amounts
496 of space, which will often fail. The maximum supported value of n
497 differs across systems, but is in all cases less than the maximum
498 representable value of a size_t.
500 void* __libc_malloc(size_t);
501 libc_hidden_proto (__libc_malloc)
504 free(void* p)
505 Releases the chunk of memory pointed to by p, that had been previously
506 allocated using malloc or a related routine such as realloc.
507 It has no effect if p is null. It can have arbitrary (i.e., bad!)
508 effects if p has already been freed.
510 Unless disabled (using mallopt), freeing very large spaces will
511 when possible, automatically trigger operations that give
512 back unused memory to the system, thus reducing program footprint.
514 void __libc_free(void*);
515 libc_hidden_proto (__libc_free)
518 calloc(size_t n_elements, size_t element_size);
519 Returns a pointer to n_elements * element_size bytes, with all locations
520 set to zero.
522 void* __libc_calloc(size_t, size_t);
525 realloc(void* p, size_t n)
526 Returns a pointer to a chunk of size n that contains the same data
527 as does chunk p up to the minimum of (n, p's size) bytes, or null
528 if no space is available.
530 The returned pointer may or may not be the same as p. The algorithm
531 prefers extending p when possible, otherwise it employs the
532 equivalent of a malloc-copy-free sequence.
534 If p is null, realloc is equivalent to malloc.
536 If space is not available, realloc returns null, errno is set (if on
537 ANSI) and p is NOT freed.
539 if n is for fewer bytes than already held by p, the newly unused
540 space is lopped off and freed if possible. Unless the #define
541 REALLOC_ZERO_BYTES_FREES is set, realloc with a size argument of
542 zero (re)allocates a minimum-sized chunk.
544 Large chunks that were internally obtained via mmap will always be
545 grown using malloc-copy-free sequences unless the system supports
546 MREMAP (currently only linux).
548 The old unix realloc convention of allowing the last-free'd chunk
549 to be used as an argument to realloc is not supported.
551 void* __libc_realloc(void*, size_t);
552 libc_hidden_proto (__libc_realloc)
555 memalign(size_t alignment, size_t n);
556 Returns a pointer to a newly allocated chunk of n bytes, aligned
557 in accord with the alignment argument.
559 The alignment argument should be a power of two. If the argument is
560 not a power of two, the nearest greater power is used.
561 8-byte alignment is guaranteed by normal malloc calls, so don't
562 bother calling memalign with an argument of 8 or less.
564 Overreliance on memalign is a sure way to fragment space.
566 void* __libc_memalign(size_t, size_t);
567 libc_hidden_proto (__libc_memalign)
570 valloc(size_t n);
571 Equivalent to memalign(pagesize, n), where pagesize is the page
572 size of the system. If the pagesize is unknown, 4096 is used.
574 void* __libc_valloc(size_t);
579 mallopt(int parameter_number, int parameter_value)
580 Sets tunable parameters The format is to provide a
581 (parameter-number, parameter-value) pair. mallopt then sets the
582 corresponding parameter to the argument value if it can (i.e., so
583 long as the value is meaningful), and returns 1 if successful else
584 0. SVID/XPG/ANSI defines four standard param numbers for mallopt,
585 normally defined in malloc.h. Only one of these (M_MXFAST) is used
586 in this malloc. The others (M_NLBLKS, M_GRAIN, M_KEEP) don't apply,
587 so setting them has no effect. But this malloc also supports four
588 other options in mallopt. See below for details. Briefly, supported
589 parameters are as follows (listed defaults are for "typical"
590 configurations).
592 Symbol param # default allowed param values
593 M_MXFAST 1 64 0-80 (0 disables fastbins)
594 M_TRIM_THRESHOLD -1 128*1024 any (-1U disables trimming)
595 M_TOP_PAD -2 0 any
596 M_MMAP_THRESHOLD -3 128*1024 any (or 0 if no MMAP support)
597 M_MMAP_MAX -4 65536 any (0 disables use of mmap)
599 int __libc_mallopt(int, int);
600 libc_hidden_proto (__libc_mallopt)
604 mallinfo()
605 Returns (by copy) a struct containing various summary statistics:
607 arena: current total non-mmapped bytes allocated from system
608 ordblks: the number of free chunks
609 smblks: the number of fastbin blocks (i.e., small chunks that
610 have been freed but not use resused or consolidated)
611 hblks: current number of mmapped regions
612 hblkhd: total bytes held in mmapped regions
613 usmblks: always 0
614 fsmblks: total bytes held in fastbin blocks
615 uordblks: current total allocated space (normal or mmapped)
616 fordblks: total free space
617 keepcost: the maximum number of bytes that could ideally be released
618 back to system via malloc_trim. ("ideally" means that
619 it ignores page restrictions etc.)
621 Because these fields are ints, but internal bookkeeping may
622 be kept as longs, the reported values may wrap around zero and
623 thus be inaccurate.
625 struct mallinfo __libc_mallinfo(void);
629 pvalloc(size_t n);
630 Equivalent to valloc(minimum-page-that-holds(n)), that is,
631 round up n to nearest pagesize.
633 void* __libc_pvalloc(size_t);
636 malloc_trim(size_t pad);
638 If possible, gives memory back to the system (via negative
639 arguments to sbrk) if there is unused memory at the `high' end of
640 the malloc pool. You can call this after freeing large blocks of
641 memory to potentially reduce the system-level memory requirements
642 of a program. However, it cannot guarantee to reduce memory. Under
643 some allocation patterns, some large free blocks of memory will be
644 locked between two used chunks, so they cannot be given back to
645 the system.
647 The `pad' argument to malloc_trim represents the amount of free
648 trailing space to leave untrimmed. If this argument is zero,
649 only the minimum amount of memory to maintain internal data
650 structures will be left (one page or less). Non-zero arguments
651 can be supplied to maintain enough trailing space to service
652 future expected allocations without having to re-obtain memory
653 from the system.
655 Malloc_trim returns 1 if it actually released any memory, else 0.
656 On systems that do not support "negative sbrks", it will always
657 return 0.
659 int __malloc_trim(size_t);
662 malloc_usable_size(void* p);
664 Returns the number of bytes you can actually use in
665 an allocated chunk, which may be more than you requested (although
666 often not) due to alignment and minimum size constraints.
667 You can use this many bytes without worrying about
668 overwriting other allocated objects. This is not a particularly great
669 programming practice. malloc_usable_size can be more useful in
670 debugging and assertions, for example:
672 p = malloc(n);
673 assert(malloc_usable_size(p) >= 256);
676 size_t __malloc_usable_size(void*);
679 malloc_stats();
680 Prints on stderr the amount of space obtained from the system (both
681 via sbrk and mmap), the maximum amount (which may be more than
682 current if malloc_trim and/or munmap got called), and the current
683 number of bytes allocated via malloc (or realloc, etc) but not yet
684 freed. Note that this is the number of bytes allocated, not the
685 number requested. It will be larger than the number requested
686 because of alignment and bookkeeping overhead. Because it includes
687 alignment wastage as being in use, this figure may be greater than
688 zero even when no user-level chunks are allocated.
690 The reported current and maximum system memory can be inaccurate if
691 a program makes other calls to system memory allocation functions
692 (normally sbrk) outside of malloc.
694 malloc_stats prints only the most commonly interesting statistics.
695 More information can be obtained by calling mallinfo.
698 void __malloc_stats(void);
701 posix_memalign(void **memptr, size_t alignment, size_t size);
703 POSIX wrapper like memalign(), checking for validity of size.
705 int __posix_memalign(void **, size_t, size_t);
707 /* mallopt tuning options */
710 M_MXFAST is the maximum request size used for "fastbins", special bins
711 that hold returned chunks without consolidating their spaces. This
712 enables future requests for chunks of the same size to be handled
713 very quickly, but can increase fragmentation, and thus increase the
714 overall memory footprint of a program.
716 This malloc manages fastbins very conservatively yet still
717 efficiently, so fragmentation is rarely a problem for values less
718 than or equal to the default. The maximum supported value of MXFAST
719 is 80. You wouldn't want it any higher than this anyway. Fastbins
720 are designed especially for use with many small structs, objects or
721 strings -- the default handles structs/objects/arrays with sizes up
722 to 8 4byte fields, or small strings representing words, tokens,
723 etc. Using fastbins for larger objects normally worsens
724 fragmentation without improving speed.
726 M_MXFAST is set in REQUEST size units. It is internally used in
727 chunksize units, which adds padding and alignment. You can reduce
728 M_MXFAST to 0 to disable all use of fastbins. This causes the malloc
729 algorithm to be a closer approximation of fifo-best-fit in all cases,
730 not just for larger requests, but will generally cause it to be
731 slower.
735 /* M_MXFAST is a standard SVID/XPG tuning option, usually listed in malloc.h */
736 #ifndef M_MXFAST
737 #define M_MXFAST 1
738 #endif
740 #ifndef DEFAULT_MXFAST
741 #define DEFAULT_MXFAST (64 * SIZE_SZ / 4)
742 #endif
746 M_TRIM_THRESHOLD is the maximum amount of unused top-most memory
747 to keep before releasing via malloc_trim in free().
749 Automatic trimming is mainly useful in long-lived programs.
750 Because trimming via sbrk can be slow on some systems, and can
751 sometimes be wasteful (in cases where programs immediately
752 afterward allocate more large chunks) the value should be high
753 enough so that your overall system performance would improve by
754 releasing this much memory.
756 The trim threshold and the mmap control parameters (see below)
757 can be traded off with one another. Trimming and mmapping are
758 two different ways of releasing unused memory back to the
759 system. Between these two, it is often possible to keep
760 system-level demands of a long-lived program down to a bare
761 minimum. For example, in one test suite of sessions measuring
762 the XF86 X server on Linux, using a trim threshold of 128K and a
763 mmap threshold of 192K led to near-minimal long term resource
764 consumption.
766 If you are using this malloc in a long-lived program, it should
767 pay to experiment with these values. As a rough guide, you
768 might set to a value close to the average size of a process
769 (program) running on your system. Releasing this much memory
770 would allow such a process to run in memory. Generally, it's
771 worth it to tune for trimming rather tham memory mapping when a
772 program undergoes phases where several large chunks are
773 allocated and released in ways that can reuse each other's
774 storage, perhaps mixed with phases where there are no such
775 chunks at all. And in well-behaved long-lived programs,
776 controlling release of large blocks via trimming versus mapping
777 is usually faster.
779 However, in most programs, these parameters serve mainly as
780 protection against the system-level effects of carrying around
781 massive amounts of unneeded memory. Since frequent calls to
782 sbrk, mmap, and munmap otherwise degrade performance, the default
783 parameters are set to relatively high values that serve only as
784 safeguards.
786 The trim value It must be greater than page size to have any useful
787 effect. To disable trimming completely, you can set to
788 (unsigned long)(-1)
790 Trim settings interact with fastbin (MXFAST) settings: Unless
791 TRIM_FASTBINS is defined, automatic trimming never takes place upon
792 freeing a chunk with size less than or equal to MXFAST. Trimming is
793 instead delayed until subsequent freeing of larger chunks. However,
794 you can still force an attempted trim by calling malloc_trim.
796 Also, trimming is not generally possible in cases where
797 the main arena is obtained via mmap.
799 Note that the trick some people use of mallocing a huge space and
800 then freeing it at program startup, in an attempt to reserve system
801 memory, doesn't have the intended effect under automatic trimming,
802 since that memory will immediately be returned to the system.
805 #define M_TRIM_THRESHOLD -1
807 #ifndef DEFAULT_TRIM_THRESHOLD
808 #define DEFAULT_TRIM_THRESHOLD (128 * 1024)
809 #endif
812 M_TOP_PAD is the amount of extra `padding' space to allocate or
813 retain whenever sbrk is called. It is used in two ways internally:
815 * When sbrk is called to extend the top of the arena to satisfy
816 a new malloc request, this much padding is added to the sbrk
817 request.
819 * When malloc_trim is called automatically from free(),
820 it is used as the `pad' argument.
822 In both cases, the actual amount of padding is rounded
823 so that the end of the arena is always a system page boundary.
825 The main reason for using padding is to avoid calling sbrk so
826 often. Having even a small pad greatly reduces the likelihood
827 that nearly every malloc request during program start-up (or
828 after trimming) will invoke sbrk, which needlessly wastes
829 time.
831 Automatic rounding-up to page-size units is normally sufficient
832 to avoid measurable overhead, so the default is 0. However, in
833 systems where sbrk is relatively slow, it can pay to increase
834 this value, at the expense of carrying around more memory than
835 the program needs.
838 #define M_TOP_PAD -2
840 #ifndef DEFAULT_TOP_PAD
841 #define DEFAULT_TOP_PAD (0)
842 #endif
845 MMAP_THRESHOLD_MAX and _MIN are the bounds on the dynamically
846 adjusted MMAP_THRESHOLD.
849 #ifndef DEFAULT_MMAP_THRESHOLD_MIN
850 #define DEFAULT_MMAP_THRESHOLD_MIN (128 * 1024)
851 #endif
853 #ifndef DEFAULT_MMAP_THRESHOLD_MAX
854 /* For 32-bit platforms we cannot increase the maximum mmap
855 threshold much because it is also the minimum value for the
856 maximum heap size and its alignment. Going above 512k (i.e., 1M
857 for new heaps) wastes too much address space. */
858 # if __WORDSIZE == 32
859 # define DEFAULT_MMAP_THRESHOLD_MAX (512 * 1024)
860 # else
861 # define DEFAULT_MMAP_THRESHOLD_MAX (4 * 1024 * 1024 * sizeof(long))
862 # endif
863 #endif
866 M_MMAP_THRESHOLD is the request size threshold for using mmap()
867 to service a request. Requests of at least this size that cannot
868 be allocated using already-existing space will be serviced via mmap.
869 (If enough normal freed space already exists it is used instead.)
871 Using mmap segregates relatively large chunks of memory so that
872 they can be individually obtained and released from the host
873 system. A request serviced through mmap is never reused by any
874 other request (at least not directly; the system may just so
875 happen to remap successive requests to the same locations).
877 Segregating space in this way has the benefits that:
879 1. Mmapped space can ALWAYS be individually released back
880 to the system, which helps keep the system level memory
881 demands of a long-lived program low.
882 2. Mapped memory can never become `locked' between
883 other chunks, as can happen with normally allocated chunks, which
884 means that even trimming via malloc_trim would not release them.
885 3. On some systems with "holes" in address spaces, mmap can obtain
886 memory that sbrk cannot.
888 However, it has the disadvantages that:
890 1. The space cannot be reclaimed, consolidated, and then
891 used to service later requests, as happens with normal chunks.
892 2. It can lead to more wastage because of mmap page alignment
893 requirements
894 3. It causes malloc performance to be more dependent on host
895 system memory management support routines which may vary in
896 implementation quality and may impose arbitrary
897 limitations. Generally, servicing a request via normal
898 malloc steps is faster than going through a system's mmap.
900 The advantages of mmap nearly always outweigh disadvantages for
901 "large" chunks, but the value of "large" varies across systems. The
902 default is an empirically derived value that works well in most
903 systems.
906 Update in 2006:
907 The above was written in 2001. Since then the world has changed a lot.
908 Memory got bigger. Applications got bigger. The virtual address space
909 layout in 32 bit linux changed.
911 In the new situation, brk() and mmap space is shared and there are no
912 artificial limits on brk size imposed by the kernel. What is more,
913 applications have started using transient allocations larger than the
914 128Kb as was imagined in 2001.
916 The price for mmap is also high now; each time glibc mmaps from the
917 kernel, the kernel is forced to zero out the memory it gives to the
918 application. Zeroing memory is expensive and eats a lot of cache and
919 memory bandwidth. This has nothing to do with the efficiency of the
920 virtual memory system, by doing mmap the kernel just has no choice but
921 to zero.
923 In 2001, the kernel had a maximum size for brk() which was about 800
924 megabytes on 32 bit x86, at that point brk() would hit the first
925 mmaped shared libaries and couldn't expand anymore. With current 2.6
926 kernels, the VA space layout is different and brk() and mmap
927 both can span the entire heap at will.
929 Rather than using a static threshold for the brk/mmap tradeoff,
930 we are now using a simple dynamic one. The goal is still to avoid
931 fragmentation. The old goals we kept are
932 1) try to get the long lived large allocations to use mmap()
933 2) really large allocations should always use mmap()
934 and we're adding now:
935 3) transient allocations should use brk() to avoid forcing the kernel
936 having to zero memory over and over again
938 The implementation works with a sliding threshold, which is by default
939 limited to go between 128Kb and 32Mb (64Mb for 64 bitmachines) and starts
940 out at 128Kb as per the 2001 default.
942 This allows us to satisfy requirement 1) under the assumption that long
943 lived allocations are made early in the process' lifespan, before it has
944 started doing dynamic allocations of the same size (which will
945 increase the threshold).
947 The upperbound on the threshold satisfies requirement 2)
949 The threshold goes up in value when the application frees memory that was
950 allocated with the mmap allocator. The idea is that once the application
951 starts freeing memory of a certain size, it's highly probable that this is
952 a size the application uses for transient allocations. This estimator
953 is there to satisfy the new third requirement.
957 #define M_MMAP_THRESHOLD -3
959 #ifndef DEFAULT_MMAP_THRESHOLD
960 #define DEFAULT_MMAP_THRESHOLD DEFAULT_MMAP_THRESHOLD_MIN
961 #endif
964 M_MMAP_MAX is the maximum number of requests to simultaneously
965 service using mmap. This parameter exists because
966 some systems have a limited number of internal tables for
967 use by mmap, and using more than a few of them may degrade
968 performance.
970 The default is set to a value that serves only as a safeguard.
971 Setting to 0 disables use of mmap for servicing large requests.
974 #define M_MMAP_MAX -4
976 #ifndef DEFAULT_MMAP_MAX
977 #define DEFAULT_MMAP_MAX (65536)
978 #endif
980 #include <malloc.h>
982 #ifndef RETURN_ADDRESS
983 #define RETURN_ADDRESS(X_) (NULL)
984 #endif
986 /* Forward declarations. */
987 struct malloc_chunk;
988 typedef struct malloc_chunk* mchunkptr;
990 /* Internal routines. */
992 static void* _int_malloc(mstate, size_t);
993 static void _int_free(mstate, mchunkptr, int);
994 static void* _int_realloc(mstate, mchunkptr, INTERNAL_SIZE_T,
995 INTERNAL_SIZE_T);
996 static void* _int_memalign(mstate, size_t, size_t);
997 static void* _mid_memalign(size_t, size_t, void *);
999 static void malloc_printerr(const char *str) __attribute__ ((noreturn));
1001 static void* mem2mem_check(void *p, size_t sz);
1002 static void top_check(void);
1003 static void munmap_chunk(mchunkptr p);
1004 #if HAVE_MREMAP
1005 static mchunkptr mremap_chunk(mchunkptr p, size_t new_size);
1006 #endif
1008 static void* malloc_check(size_t sz, const void *caller);
1009 static void free_check(void* mem, const void *caller);
1010 static void* realloc_check(void* oldmem, size_t bytes,
1011 const void *caller);
1012 static void* memalign_check(size_t alignment, size_t bytes,
1013 const void *caller);
1015 /* ------------------ MMAP support ------------------ */
1018 #include <fcntl.h>
1019 #include <sys/mman.h>
1021 #if !defined(MAP_ANONYMOUS) && defined(MAP_ANON)
1022 # define MAP_ANONYMOUS MAP_ANON
1023 #endif
1025 #ifndef MAP_NORESERVE
1026 # define MAP_NORESERVE 0
1027 #endif
1029 #define MMAP(addr, size, prot, flags) \
1030 __mmap((addr), (size), (prot), (flags)|MAP_ANONYMOUS|MAP_PRIVATE, -1, 0)
1034 ----------------------- Chunk representations -----------------------
1039 This struct declaration is misleading (but accurate and necessary).
1040 It declares a "view" into memory allowing access to necessary
1041 fields at known offsets from a given base. See explanation below.
1044 struct malloc_chunk {
1046 INTERNAL_SIZE_T mchunk_prev_size; /* Size of previous chunk (if free). */
1047 INTERNAL_SIZE_T mchunk_size; /* Size in bytes, including overhead. */
1049 struct malloc_chunk* fd; /* double links -- used only if free. */
1050 struct malloc_chunk* bk;
1052 /* Only used for large blocks: pointer to next larger size. */
1053 struct malloc_chunk* fd_nextsize; /* double links -- used only if free. */
1054 struct malloc_chunk* bk_nextsize;
1059 malloc_chunk details:
1061 (The following includes lightly edited explanations by Colin Plumb.)
1063 Chunks of memory are maintained using a `boundary tag' method as
1064 described in e.g., Knuth or Standish. (See the paper by Paul
1065 Wilson ftp://ftp.cs.utexas.edu/pub/garbage/allocsrv.ps for a
1066 survey of such techniques.) Sizes of free chunks are stored both
1067 in the front of each chunk and at the end. This makes
1068 consolidating fragmented chunks into bigger chunks very fast. The
1069 size fields also hold bits representing whether chunks are free or
1070 in use.
1072 An allocated chunk looks like this:
1075 chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1076 | Size of previous chunk, if unallocated (P clear) |
1077 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1078 | Size of chunk, in bytes |A|M|P|
1079 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1080 | User data starts here... .
1082 . (malloc_usable_size() bytes) .
1084 nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1085 | (size of chunk, but used for application data) |
1086 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1087 | Size of next chunk, in bytes |A|0|1|
1088 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1090 Where "chunk" is the front of the chunk for the purpose of most of
1091 the malloc code, but "mem" is the pointer that is returned to the
1092 user. "Nextchunk" is the beginning of the next contiguous chunk.
1094 Chunks always begin on even word boundaries, so the mem portion
1095 (which is returned to the user) is also on an even word boundary, and
1096 thus at least double-word aligned.
1098 Free chunks are stored in circular doubly-linked lists, and look like this:
1100 chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1101 | Size of previous chunk, if unallocated (P clear) |
1102 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1103 `head:' | Size of chunk, in bytes |A|0|P|
1104 mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1105 | Forward pointer to next chunk in list |
1106 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1107 | Back pointer to previous chunk in list |
1108 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1109 | Unused space (may be 0 bytes long) .
1112 nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1113 `foot:' | Size of chunk, in bytes |
1114 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1115 | Size of next chunk, in bytes |A|0|0|
1116 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1118 The P (PREV_INUSE) bit, stored in the unused low-order bit of the
1119 chunk size (which is always a multiple of two words), is an in-use
1120 bit for the *previous* chunk. If that bit is *clear*, then the
1121 word before the current chunk size contains the previous chunk
1122 size, and can be used to find the front of the previous chunk.
1123 The very first chunk allocated always has this bit set,
1124 preventing access to non-existent (or non-owned) memory. If
1125 prev_inuse is set for any given chunk, then you CANNOT determine
1126 the size of the previous chunk, and might even get a memory
1127 addressing fault when trying to do so.
1129 The A (NON_MAIN_ARENA) bit is cleared for chunks on the initial,
1130 main arena, described by the main_arena variable. When additional
1131 threads are spawned, each thread receives its own arena (up to a
1132 configurable limit, after which arenas are reused for multiple
1133 threads), and the chunks in these arenas have the A bit set. To
1134 find the arena for a chunk on such a non-main arena, heap_for_ptr
1135 performs a bit mask operation and indirection through the ar_ptr
1136 member of the per-heap header heap_info (see arena.c).
1138 Note that the `foot' of the current chunk is actually represented
1139 as the prev_size of the NEXT chunk. This makes it easier to
1140 deal with alignments etc but can be very confusing when trying
1141 to extend or adapt this code.
1143 The three exceptions to all this are:
1145 1. The special chunk `top' doesn't bother using the
1146 trailing size field since there is no next contiguous chunk
1147 that would have to index off it. After initialization, `top'
1148 is forced to always exist. If it would become less than
1149 MINSIZE bytes long, it is replenished.
1151 2. Chunks allocated via mmap, which have the second-lowest-order
1152 bit M (IS_MMAPPED) set in their size fields. Because they are
1153 allocated one-by-one, each must contain its own trailing size
1154 field. If the M bit is set, the other bits are ignored
1155 (because mmapped chunks are neither in an arena, nor adjacent
1156 to a freed chunk). The M bit is also used for chunks which
1157 originally came from a dumped heap via malloc_set_state in
1158 hooks.c.
1160 3. Chunks in fastbins are treated as allocated chunks from the
1161 point of view of the chunk allocator. They are consolidated
1162 with their neighbors only in bulk, in malloc_consolidate.
1166 ---------- Size and alignment checks and conversions ----------
1169 /* conversion from malloc headers to user pointers, and back */
1171 #define chunk2mem(p) ((void*)((char*)(p) + 2*SIZE_SZ))
1172 #define mem2chunk(mem) ((mchunkptr)((char*)(mem) - 2*SIZE_SZ))
1174 /* The smallest possible chunk */
1175 #define MIN_CHUNK_SIZE (offsetof(struct malloc_chunk, fd_nextsize))
1177 /* The smallest size we can malloc is an aligned minimal chunk */
1179 #define MINSIZE \
1180 (unsigned long)(((MIN_CHUNK_SIZE+MALLOC_ALIGN_MASK) & ~MALLOC_ALIGN_MASK))
1182 /* Check if m has acceptable alignment */
1184 #define aligned_OK(m) (((unsigned long)(m) & MALLOC_ALIGN_MASK) == 0)
1186 #define misaligned_chunk(p) \
1187 ((uintptr_t)(MALLOC_ALIGNMENT == 2 * SIZE_SZ ? (p) : chunk2mem (p)) \
1188 & MALLOC_ALIGN_MASK)
1192 Check if a request is so large that it would wrap around zero when
1193 padded and aligned. To simplify some other code, the bound is made
1194 low enough so that adding MINSIZE will also not wrap around zero.
1197 #define REQUEST_OUT_OF_RANGE(req) \
1198 ((unsigned long) (req) >= \
1199 (unsigned long) (INTERNAL_SIZE_T) (-2 * MINSIZE))
1201 /* pad request bytes into a usable size -- internal version */
1203 #define request2size(req) \
1204 (((req) + SIZE_SZ + MALLOC_ALIGN_MASK < MINSIZE) ? \
1205 MINSIZE : \
1206 ((req) + SIZE_SZ + MALLOC_ALIGN_MASK) & ~MALLOC_ALIGN_MASK)
1208 /* Same, except also perform an argument and result check. First, we check
1209 that the padding done by request2size didn't result in an integer
1210 overflow. Then we check (using REQUEST_OUT_OF_RANGE) that the resulting
1211 size isn't so large that a later alignment would lead to another integer
1212 overflow. */
1213 #define checked_request2size(req, sz) \
1214 ({ \
1215 (sz) = request2size (req); \
1216 if (((sz) < (req)) \
1217 || REQUEST_OUT_OF_RANGE (sz)) \
1219 __set_errno (ENOMEM); \
1220 return 0; \
1225 --------------- Physical chunk operations ---------------
1229 /* size field is or'ed with PREV_INUSE when previous adjacent chunk in use */
1230 #define PREV_INUSE 0x1
1232 /* extract inuse bit of previous chunk */
1233 #define prev_inuse(p) ((p)->mchunk_size & PREV_INUSE)
1236 /* size field is or'ed with IS_MMAPPED if the chunk was obtained with mmap() */
1237 #define IS_MMAPPED 0x2
1239 /* check for mmap()'ed chunk */
1240 #define chunk_is_mmapped(p) ((p)->mchunk_size & IS_MMAPPED)
1243 /* size field is or'ed with NON_MAIN_ARENA if the chunk was obtained
1244 from a non-main arena. This is only set immediately before handing
1245 the chunk to the user, if necessary. */
1246 #define NON_MAIN_ARENA 0x4
1248 /* Check for chunk from main arena. */
1249 #define chunk_main_arena(p) (((p)->mchunk_size & NON_MAIN_ARENA) == 0)
1251 /* Mark a chunk as not being on the main arena. */
1252 #define set_non_main_arena(p) ((p)->mchunk_size |= NON_MAIN_ARENA)
1256 Bits to mask off when extracting size
1258 Note: IS_MMAPPED is intentionally not masked off from size field in
1259 macros for which mmapped chunks should never be seen. This should
1260 cause helpful core dumps to occur if it is tried by accident by
1261 people extending or adapting this malloc.
1263 #define SIZE_BITS (PREV_INUSE | IS_MMAPPED | NON_MAIN_ARENA)
1265 /* Get size, ignoring use bits */
1266 #define chunksize(p) (chunksize_nomask (p) & ~(SIZE_BITS))
1268 /* Like chunksize, but do not mask SIZE_BITS. */
1269 #define chunksize_nomask(p) ((p)->mchunk_size)
1271 /* Ptr to next physical malloc_chunk. */
1272 #define next_chunk(p) ((mchunkptr) (((char *) (p)) + chunksize (p)))
1274 /* Size of the chunk below P. Only valid if !prev_inuse (P). */
1275 #define prev_size(p) ((p)->mchunk_prev_size)
1277 /* Set the size of the chunk below P. Only valid if !prev_inuse (P). */
1278 #define set_prev_size(p, sz) ((p)->mchunk_prev_size = (sz))
1280 /* Ptr to previous physical malloc_chunk. Only valid if !prev_inuse (P). */
1281 #define prev_chunk(p) ((mchunkptr) (((char *) (p)) - prev_size (p)))
1283 /* Treat space at ptr + offset as a chunk */
1284 #define chunk_at_offset(p, s) ((mchunkptr) (((char *) (p)) + (s)))
1286 /* extract p's inuse bit */
1287 #define inuse(p) \
1288 ((((mchunkptr) (((char *) (p)) + chunksize (p)))->mchunk_size) & PREV_INUSE)
1290 /* set/clear chunk as being inuse without otherwise disturbing */
1291 #define set_inuse(p) \
1292 ((mchunkptr) (((char *) (p)) + chunksize (p)))->mchunk_size |= PREV_INUSE
1294 #define clear_inuse(p) \
1295 ((mchunkptr) (((char *) (p)) + chunksize (p)))->mchunk_size &= ~(PREV_INUSE)
1298 /* check/set/clear inuse bits in known places */
1299 #define inuse_bit_at_offset(p, s) \
1300 (((mchunkptr) (((char *) (p)) + (s)))->mchunk_size & PREV_INUSE)
1302 #define set_inuse_bit_at_offset(p, s) \
1303 (((mchunkptr) (((char *) (p)) + (s)))->mchunk_size |= PREV_INUSE)
1305 #define clear_inuse_bit_at_offset(p, s) \
1306 (((mchunkptr) (((char *) (p)) + (s)))->mchunk_size &= ~(PREV_INUSE))
1309 /* Set size at head, without disturbing its use bit */
1310 #define set_head_size(p, s) ((p)->mchunk_size = (((p)->mchunk_size & SIZE_BITS) | (s)))
1312 /* Set size/use field */
1313 #define set_head(p, s) ((p)->mchunk_size = (s))
1315 /* Set size at footer (only when chunk is not in use) */
1316 #define set_foot(p, s) (((mchunkptr) ((char *) (p) + (s)))->mchunk_prev_size = (s))
1319 /* Add an item to the atomic fastbin list at *ROOT. Returns the old
1320 value at *ROOT. Note that properties of the old chunk are only
1321 stable if the caller has acquired the arena lock. With out the
1322 lock, it can be deallocated at any time. */
1323 static inline struct malloc_chunk *
1324 fastbin_push_entry (struct malloc_chunk **root, struct malloc_chunk *e)
1326 struct malloc_chunk *head;
1327 if (SINGLE_THREAD_P)
1329 /* Check that the top of the bin is not the record we are going
1330 to add (i.e., double free). */
1331 head = *root;
1332 if (head == e)
1333 malloc_printerr ("double free or corruption (fasttop)");
1334 e->fd = head;
1335 *root = e;
1337 else
1340 /* Synchronize with the release release MO CAS below. We do
1341 not need synchronization locally, but fastbin_pop_entry and
1342 (especially) malloc_consolidate read the entire list after
1343 synchronizing on the head, so we need to make sure that the
1344 writes to the next (fd) pointers have happened. */
1345 head = atomic_load_acquire (root);
1346 /* Check that the top of the bin is not the record we are
1347 going to add (i.e., double free). */
1348 if (head == e)
1349 malloc_printerr ("double free or corruption (fasttop)");
1350 e->fd = head;
1352 /* Synchronizes with the acquire MO CAS in */
1353 while (!atomic_compare_exchange_weak_release (root, &head, e));
1354 return head;
1357 /* Remove an item from the atomic fastbin list at *ROOT. The caller
1358 must have acquired the arena lock. */
1359 static inline struct malloc_chunk *
1360 fastbin_pop_entry (struct malloc_chunk **root)
1362 struct malloc_chunk *head;
1363 if (SINGLE_THREAD_P)
1365 head = *root;
1366 if (head != NULL)
1367 *root = head->fd;
1369 else
1371 /* Synchromizes with the release MO store in fastbin_push_entry.
1372 Synchronization is needed because we read the next list
1373 pointer. */
1374 head = atomic_load_acquire (root);
1375 struct malloc_chunk *tail;
1378 if (head == NULL)
1379 return NULL;
1380 tail = head->fd;
1382 /* Synchronizes with the release MO store in fastbin_push_entry.
1383 We do not have an ABA issue here because the caller has
1384 acquired the arena lock, which ensures that there is only one
1385 thread which removes elements from this list. */
1386 while (!atomic_compare_exchange_weak_acquire (root, &head, tail));
1388 return head;
1391 #pragma GCC poison mchunk_size
1392 #pragma GCC poison mchunk_prev_size
1395 -------------------- Internal data structures --------------------
1397 All internal state is held in an instance of malloc_state defined
1398 below. There are no other static variables, except in two optional
1399 cases:
1400 * If USE_MALLOC_LOCK is defined, the mALLOC_MUTEx declared above.
1401 * If mmap doesn't support MAP_ANONYMOUS, a dummy file descriptor
1402 for mmap.
1404 Beware of lots of tricks that minimize the total bookkeeping space
1405 requirements. The result is a little over 1K bytes (for 4byte
1406 pointers and size_t.)
1410 Bins
1412 An array of bin headers for free chunks. Each bin is doubly
1413 linked. The bins are approximately proportionally (log) spaced.
1414 There are a lot of these bins (128). This may look excessive, but
1415 works very well in practice. Most bins hold sizes that are
1416 unusual as malloc request sizes, but are more usual for fragments
1417 and consolidated sets of chunks, which is what these bins hold, so
1418 they can be found quickly. All procedures maintain the invariant
1419 that no consolidated chunk physically borders another one, so each
1420 chunk in a list is known to be preceeded and followed by either
1421 inuse chunks or the ends of memory.
1423 Chunks in bins are kept in size order, with ties going to the
1424 approximately least recently used chunk. Ordering isn't needed
1425 for the small bins, which all contain the same-sized chunks, but
1426 facilitates best-fit allocation for larger chunks. These lists
1427 are just sequential. Keeping them in order almost never requires
1428 enough traversal to warrant using fancier ordered data
1429 structures.
1431 Chunks of the same size are linked with the most
1432 recently freed at the front, and allocations are taken from the
1433 back. This results in LRU (FIFO) allocation order, which tends
1434 to give each chunk an equal opportunity to be consolidated with
1435 adjacent freed chunks, resulting in larger free chunks and less
1436 fragmentation.
1438 To simplify use in double-linked lists, each bin header acts
1439 as a malloc_chunk. This avoids special-casing for headers.
1440 But to conserve space and improve locality, we allocate
1441 only the fd/bk pointers of bins, and then use repositioning tricks
1442 to treat these as the fields of a malloc_chunk*.
1445 typedef struct malloc_chunk *mbinptr;
1447 /* addressing -- note that bin_at(0) does not exist */
1448 #define bin_at(m, i) \
1449 (mbinptr) (((char *) &((m)->bins[((i) - 1) * 2])) \
1450 - offsetof (struct malloc_chunk, fd))
1452 /* analog of ++bin */
1453 #define next_bin(b) ((mbinptr) ((char *) (b) + (sizeof (mchunkptr) << 1)))
1455 /* Reminders about list directionality within bins */
1456 #define first(b) ((b)->fd)
1457 #define last(b) ((b)->bk)
1460 Indexing
1462 Bins for sizes < 512 bytes contain chunks of all the same size, spaced
1463 8 bytes apart. Larger bins are approximately logarithmically spaced:
1465 64 bins of size 8
1466 32 bins of size 64
1467 16 bins of size 512
1468 8 bins of size 4096
1469 4 bins of size 32768
1470 2 bins of size 262144
1471 1 bin of size what's left
1473 There is actually a little bit of slop in the numbers in bin_index
1474 for the sake of speed. This makes no difference elsewhere.
1476 The bins top out around 1MB because we expect to service large
1477 requests via mmap.
1479 Bin 0 does not exist. Bin 1 is the unordered list; if that would be
1480 a valid chunk size the small bins are bumped up one.
1483 #define NBINS 128
1484 #define NSMALLBINS 64
1485 #define SMALLBIN_WIDTH MALLOC_ALIGNMENT
1486 #define SMALLBIN_CORRECTION (MALLOC_ALIGNMENT > 2 * SIZE_SZ)
1487 #define MIN_LARGE_SIZE ((NSMALLBINS - SMALLBIN_CORRECTION) * SMALLBIN_WIDTH)
1489 #define in_smallbin_range(sz) \
1490 ((unsigned long) (sz) < (unsigned long) MIN_LARGE_SIZE)
1492 #define smallbin_index(sz) \
1493 ((SMALLBIN_WIDTH == 16 ? (((unsigned) (sz)) >> 4) : (((unsigned) (sz)) >> 3))\
1494 + SMALLBIN_CORRECTION)
1496 #define largebin_index_32(sz) \
1497 (((((unsigned long) (sz)) >> 6) <= 38) ? 56 + (((unsigned long) (sz)) >> 6) :\
1498 ((((unsigned long) (sz)) >> 9) <= 20) ? 91 + (((unsigned long) (sz)) >> 9) :\
1499 ((((unsigned long) (sz)) >> 12) <= 10) ? 110 + (((unsigned long) (sz)) >> 12) :\
1500 ((((unsigned long) (sz)) >> 15) <= 4) ? 119 + (((unsigned long) (sz)) >> 15) :\
1501 ((((unsigned long) (sz)) >> 18) <= 2) ? 124 + (((unsigned long) (sz)) >> 18) :\
1502 126)
1504 #define largebin_index_32_big(sz) \
1505 (((((unsigned long) (sz)) >> 6) <= 45) ? 49 + (((unsigned long) (sz)) >> 6) :\
1506 ((((unsigned long) (sz)) >> 9) <= 20) ? 91 + (((unsigned long) (sz)) >> 9) :\
1507 ((((unsigned long) (sz)) >> 12) <= 10) ? 110 + (((unsigned long) (sz)) >> 12) :\
1508 ((((unsigned long) (sz)) >> 15) <= 4) ? 119 + (((unsigned long) (sz)) >> 15) :\
1509 ((((unsigned long) (sz)) >> 18) <= 2) ? 124 + (((unsigned long) (sz)) >> 18) :\
1510 126)
1512 // XXX It remains to be seen whether it is good to keep the widths of
1513 // XXX the buckets the same or whether it should be scaled by a factor
1514 // XXX of two as well.
1515 #define largebin_index_64(sz) \
1516 (((((unsigned long) (sz)) >> 6) <= 48) ? 48 + (((unsigned long) (sz)) >> 6) :\
1517 ((((unsigned long) (sz)) >> 9) <= 20) ? 91 + (((unsigned long) (sz)) >> 9) :\
1518 ((((unsigned long) (sz)) >> 12) <= 10) ? 110 + (((unsigned long) (sz)) >> 12) :\
1519 ((((unsigned long) (sz)) >> 15) <= 4) ? 119 + (((unsigned long) (sz)) >> 15) :\
1520 ((((unsigned long) (sz)) >> 18) <= 2) ? 124 + (((unsigned long) (sz)) >> 18) :\
1521 126)
1523 #define largebin_index(sz) \
1524 (SIZE_SZ == 8 ? largebin_index_64 (sz) \
1525 : MALLOC_ALIGNMENT == 16 ? largebin_index_32_big (sz) \
1526 : largebin_index_32 (sz))
1528 #define bin_index(sz) \
1529 ((in_smallbin_range (sz)) ? smallbin_index (sz) : largebin_index (sz))
1531 /* Take a chunk off a bin list. */
1532 static void
1533 unlink_chunk (mstate av, mchunkptr p)
1535 if (chunksize (p) != prev_size (next_chunk (p)))
1536 malloc_printerr ("corrupted size vs. prev_size");
1538 mchunkptr fd = p->fd;
1539 mchunkptr bk = p->bk;
1541 if (__builtin_expect (fd->bk != p || bk->fd != p, 0))
1542 malloc_printerr ("corrupted double-linked list");
1544 fd->bk = bk;
1545 bk->fd = fd;
1546 if (!in_smallbin_range (chunksize_nomask (p)) && p->fd_nextsize != NULL)
1548 if (p->fd_nextsize->bk_nextsize != p
1549 || p->bk_nextsize->fd_nextsize != p)
1550 malloc_printerr ("corrupted double-linked list (not small)");
1552 if (fd->fd_nextsize == NULL)
1554 if (p->fd_nextsize == p)
1555 fd->fd_nextsize = fd->bk_nextsize = fd;
1556 else
1558 fd->fd_nextsize = p->fd_nextsize;
1559 fd->bk_nextsize = p->bk_nextsize;
1560 p->fd_nextsize->bk_nextsize = fd;
1561 p->bk_nextsize->fd_nextsize = fd;
1564 else
1566 p->fd_nextsize->bk_nextsize = p->bk_nextsize;
1567 p->bk_nextsize->fd_nextsize = p->fd_nextsize;
1573 Unsorted chunks
1575 All remainders from chunk splits, as well as all returned chunks,
1576 are first placed in the "unsorted" bin. They are then placed
1577 in regular bins after malloc gives them ONE chance to be used before
1578 binning. So, basically, the unsorted_chunks list acts as a queue,
1579 with chunks being placed on it in free (and malloc_consolidate),
1580 and taken off (to be either used or placed in bins) in malloc.
1582 The NON_MAIN_ARENA flag is never set for unsorted chunks, so it
1583 does not have to be taken into account in size comparisons.
1586 /* The otherwise unindexable 1-bin is used to hold unsorted chunks. */
1587 #define unsorted_chunks(M) (bin_at (M, 1))
1592 The top-most available chunk (i.e., the one bordering the end of
1593 available memory) is treated specially. It is never included in
1594 any bin, is used only if no other chunk is available, and is
1595 released back to the system if it is very large (see
1596 M_TRIM_THRESHOLD). Because top initially
1597 points to its own bin with initial zero size, thus forcing
1598 extension on the first malloc request, we avoid having any special
1599 code in malloc to check whether it even exists yet. But we still
1600 need to do so when getting memory from system, so we make
1601 initial_top treat the bin as a legal but unusable chunk during the
1602 interval between initialization and the first call to
1603 sysmalloc. (This is somewhat delicate, since it relies on
1604 the 2 preceding words to be zero during this interval as well.)
1607 /* Conveniently, the unsorted bin can be used as dummy top on first call */
1608 #define initial_top(M) (unsorted_chunks (M))
1611 Binmap
1613 To help compensate for the large number of bins, a one-level index
1614 structure is used for bin-by-bin searching. `binmap' is a
1615 bitvector recording whether bins are definitely empty so they can
1616 be skipped over during during traversals. The bits are NOT always
1617 cleared as soon as bins are empty, but instead only
1618 when they are noticed to be empty during traversal in malloc.
1621 /* Conservatively use 32 bits per map word, even if on 64bit system */
1622 #define BINMAPSHIFT 5
1623 #define BITSPERMAP (1U << BINMAPSHIFT)
1624 #define BINMAPSIZE (NBINS / BITSPERMAP)
1626 #define idx2block(i) ((i) >> BINMAPSHIFT)
1627 #define idx2bit(i) ((1U << ((i) & ((1U << BINMAPSHIFT) - 1))))
1629 #define mark_bin(m, i) ((m)->binmap[idx2block (i)] |= idx2bit (i))
1630 #define unmark_bin(m, i) ((m)->binmap[idx2block (i)] &= ~(idx2bit (i)))
1631 #define get_binmap(m, i) ((m)->binmap[idx2block (i)] & idx2bit (i))
1634 Fastbins
1636 An array of lists holding recently freed small chunks. Fastbins
1637 are not doubly linked. It is faster to single-link them, and
1638 since chunks are never removed from the middles of these lists,
1639 double linking is not necessary. Also, unlike regular bins, they
1640 are not even processed in FIFO order (they use faster LIFO) since
1641 ordering doesn't much matter in the transient contexts in which
1642 fastbins are normally used.
1644 Chunks in fastbins keep their inuse bit set, so they cannot
1645 be consolidated with other free chunks. malloc_consolidate
1646 releases all chunks in fastbins and consolidates them with
1647 other free chunks.
1650 typedef struct malloc_chunk *mfastbinptr;
1651 #define fastbin(ar_ptr, idx) ((ar_ptr)->fastbinsY[idx])
1653 /* offset 2 to use otherwise unindexable first 2 bins */
1654 #define fastbin_index(sz) \
1655 ((((unsigned int) (sz)) >> (SIZE_SZ == 8 ? 4 : 3)) - 2)
1658 /* The maximum fastbin request size we support */
1659 #define MAX_FAST_SIZE (80 * SIZE_SZ / 4)
1661 #define NFASTBINS (fastbin_index (request2size (MAX_FAST_SIZE)) + 1)
1664 FASTBIN_CONSOLIDATION_THRESHOLD is the size of a chunk in free()
1665 that triggers automatic consolidation of possibly-surrounding
1666 fastbin chunks. This is a heuristic, so the exact value should not
1667 matter too much. It is defined at half the default trim threshold as a
1668 compromise heuristic to only attempt consolidation if it is likely
1669 to lead to trimming. However, it is not dynamically tunable, since
1670 consolidation reduces fragmentation surrounding large chunks even
1671 if trimming is not used.
1674 #define FASTBIN_CONSOLIDATION_THRESHOLD (65536UL)
1677 NONCONTIGUOUS_BIT indicates that MORECORE does not return contiguous
1678 regions. Otherwise, contiguity is exploited in merging together,
1679 when possible, results from consecutive MORECORE calls.
1681 The initial value comes from MORECORE_CONTIGUOUS, but is
1682 changed dynamically if mmap is ever used as an sbrk substitute.
1685 #define NONCONTIGUOUS_BIT (2U)
1687 #define contiguous(M) (((M)->flags & NONCONTIGUOUS_BIT) == 0)
1688 #define noncontiguous(M) (((M)->flags & NONCONTIGUOUS_BIT) != 0)
1689 #define set_noncontiguous(M) ((M)->flags |= NONCONTIGUOUS_BIT)
1690 #define set_contiguous(M) ((M)->flags &= ~NONCONTIGUOUS_BIT)
1692 /* Maximum size of memory handled in fastbins. */
1693 static INTERNAL_SIZE_T global_max_fast;
1696 Set value of max_fast.
1697 Use impossibly small value if 0.
1698 Precondition: there are no existing fastbin chunks in the main arena.
1699 Since do_check_malloc_state () checks this, we call malloc_consolidate ()
1700 before changing max_fast. Note other arenas will leak their fast bin
1701 entries if max_fast is reduced.
1704 #define set_max_fast(s) \
1705 global_max_fast = (((s) == 0) \
1706 ? SMALLBIN_WIDTH : ((s + SIZE_SZ) & ~MALLOC_ALIGN_MASK))
1708 static inline INTERNAL_SIZE_T
1709 get_max_fast (void)
1711 /* Tell the GCC optimizers that global_max_fast is never larger
1712 than MAX_FAST_SIZE. This avoids out-of-bounds array accesses in
1713 _int_malloc after constant propagation of the size parameter.
1714 (The code never executes because malloc preserves the
1715 global_max_fast invariant, but the optimizers may not recognize
1716 this.) */
1717 if (global_max_fast > MAX_FAST_SIZE)
1718 __builtin_unreachable ();
1719 return global_max_fast;
1723 ----------- Internal state representation and initialization -----------
1727 have_fastchunks indicates that there are probably some fastbin chunks.
1728 It is set true on entering a chunk into any fastbin, and cleared early in
1729 malloc_consolidate. The value is approximate since it may be set when there
1730 are no fastbin chunks, or it may be clear even if there are fastbin chunks
1731 available. Given it's sole purpose is to reduce number of redundant calls to
1732 malloc_consolidate, it does not affect correctness. As a result we can safely
1733 use relaxed atomic accesses.
1737 struct malloc_state
1739 /* Serialize access. */
1740 __libc_lock_define (, mutex);
1742 /* Flags (formerly in max_fast). */
1743 int flags;
1745 /* Set if the fastbin chunks contain recently inserted free blocks. */
1746 /* Note this is a bool but not all targets support atomics on booleans. */
1747 int have_fastchunks;
1749 /* Fastbins */
1750 mfastbinptr fastbinsY[NFASTBINS];
1752 /* Base of the topmost chunk -- not otherwise kept in a bin */
1753 mchunkptr top;
1755 /* The remainder from the most recent split of a small request */
1756 mchunkptr last_remainder;
1758 /* Normal bins packed as described above */
1759 mchunkptr bins[NBINS * 2 - 2];
1761 /* Bitmap of bins */
1762 unsigned int binmap[BINMAPSIZE];
1764 /* Linked list */
1765 struct malloc_state *next;
1767 /* Linked list for free arenas. Access to this field is serialized
1768 by free_list_lock in arena.c. */
1769 struct malloc_state *next_free;
1771 /* Number of threads attached to this arena. 0 if the arena is on
1772 the free list. Access to this field is serialized by
1773 free_list_lock in arena.c. */
1774 INTERNAL_SIZE_T attached_threads;
1776 /* Memory allocated from the system in this arena. */
1777 INTERNAL_SIZE_T system_mem;
1778 INTERNAL_SIZE_T max_system_mem;
1781 struct malloc_par
1783 /* Tunable parameters */
1784 unsigned long trim_threshold;
1785 INTERNAL_SIZE_T top_pad;
1786 INTERNAL_SIZE_T mmap_threshold;
1787 INTERNAL_SIZE_T arena_test;
1788 INTERNAL_SIZE_T arena_max;
1790 /* Memory map support */
1791 int n_mmaps;
1792 int n_mmaps_max;
1793 int max_n_mmaps;
1794 /* the mmap_threshold is dynamic, until the user sets
1795 it manually, at which point we need to disable any
1796 dynamic behavior. */
1797 int no_dyn_threshold;
1799 /* Statistics */
1800 INTERNAL_SIZE_T mmapped_mem;
1801 INTERNAL_SIZE_T max_mmapped_mem;
1803 /* First address handed out by MORECORE/sbrk. */
1804 char *sbrk_base;
1806 #if USE_TCACHE
1807 /* Maximum number of buckets to use. */
1808 size_t tcache_bins;
1809 size_t tcache_max_bytes;
1810 /* Maximum number of chunks in each bucket. */
1811 size_t tcache_count;
1812 /* Maximum number of chunks to remove from the unsorted list, which
1813 aren't used to prefill the cache. */
1814 size_t tcache_unsorted_limit;
1815 #endif
1818 /* There are several instances of this struct ("arenas") in this
1819 malloc. If you are adapting this malloc in a way that does NOT use
1820 a static or mmapped malloc_state, you MUST explicitly zero-fill it
1821 before using. This malloc relies on the property that malloc_state
1822 is initialized to all zeroes (as is true of C statics). */
1824 static struct malloc_state main_arena =
1826 .mutex = _LIBC_LOCK_INITIALIZER,
1827 .next = &main_arena,
1828 .attached_threads = 1
1831 /* These variables are used for undumping support. Chunked are marked
1832 as using mmap, but we leave them alone if they fall into this
1833 range. NB: The chunk size for these chunks only includes the
1834 initial size field (of SIZE_SZ bytes), there is no trailing size
1835 field (unlike with regular mmapped chunks). */
1836 static mchunkptr dumped_main_arena_start; /* Inclusive. */
1837 static mchunkptr dumped_main_arena_end; /* Exclusive. */
1839 /* True if the pointer falls into the dumped arena. Use this after
1840 chunk_is_mmapped indicates a chunk is mmapped. */
1841 #define DUMPED_MAIN_ARENA_CHUNK(p) \
1842 ((p) >= dumped_main_arena_start && (p) < dumped_main_arena_end)
1844 /* There is only one instance of the malloc parameters. */
1846 static struct malloc_par mp_ =
1848 .top_pad = DEFAULT_TOP_PAD,
1849 .n_mmaps_max = DEFAULT_MMAP_MAX,
1850 .mmap_threshold = DEFAULT_MMAP_THRESHOLD,
1851 .trim_threshold = DEFAULT_TRIM_THRESHOLD,
1852 #define NARENAS_FROM_NCORES(n) ((n) * (sizeof (long) == 4 ? 2 : 8))
1853 .arena_test = NARENAS_FROM_NCORES (1)
1854 #if USE_TCACHE
1856 .tcache_count = TCACHE_FILL_COUNT,
1857 .tcache_bins = TCACHE_MAX_BINS,
1858 .tcache_max_bytes = tidx2usize (TCACHE_MAX_BINS-1),
1859 .tcache_unsorted_limit = 0 /* No limit. */
1860 #endif
1864 Initialize a malloc_state struct.
1866 This is called from ptmalloc_init () or from _int_new_arena ()
1867 when creating a new arena.
1870 static void
1871 malloc_init_state (mstate av)
1873 int i;
1874 mbinptr bin;
1876 /* Establish circular links for normal bins */
1877 for (i = 1; i < NBINS; ++i)
1879 bin = bin_at (av, i);
1880 bin->fd = bin->bk = bin;
1883 #if MORECORE_CONTIGUOUS
1884 if (av != &main_arena)
1885 #endif
1886 set_noncontiguous (av);
1887 if (av == &main_arena)
1888 set_max_fast (DEFAULT_MXFAST);
1889 atomic_store_relaxed (&av->have_fastchunks, false);
1891 av->top = initial_top (av);
1895 Other internal utilities operating on mstates
1898 static void *sysmalloc (INTERNAL_SIZE_T, mstate);
1899 static int systrim (size_t, mstate);
1900 static void malloc_consolidate (mstate);
1903 /* -------------- Early definitions for debugging hooks ---------------- */
1905 /* Define and initialize the hook variables. These weak definitions must
1906 appear before any use of the variables in a function (arena.c uses one). */
1907 #ifndef weak_variable
1908 /* In GNU libc we want the hook variables to be weak definitions to
1909 avoid a problem with Emacs. */
1910 # define weak_variable weak_function
1911 #endif
1913 /* Forward declarations. */
1914 static void *malloc_hook_ini (size_t sz,
1915 const void *caller) __THROW;
1916 static void *realloc_hook_ini (void *ptr, size_t sz,
1917 const void *caller) __THROW;
1918 static void *memalign_hook_ini (size_t alignment, size_t sz,
1919 const void *caller) __THROW;
1921 #if HAVE_MALLOC_INIT_HOOK
1922 void weak_variable (*__malloc_initialize_hook) (void) = NULL;
1923 compat_symbol (libc, __malloc_initialize_hook,
1924 __malloc_initialize_hook, GLIBC_2_0);
1925 #endif
1927 void weak_variable (*__free_hook) (void *__ptr,
1928 const void *) = NULL;
1929 void *weak_variable (*__malloc_hook)
1930 (size_t __size, const void *) = malloc_hook_ini;
1931 void *weak_variable (*__realloc_hook)
1932 (void *__ptr, size_t __size, const void *)
1933 = realloc_hook_ini;
1934 void *weak_variable (*__memalign_hook)
1935 (size_t __alignment, size_t __size, const void *)
1936 = memalign_hook_ini;
1937 void weak_variable (*__after_morecore_hook) (void) = NULL;
1939 /* This function is called from the arena shutdown hook, to free the
1940 thread cache (if it exists). */
1941 static void tcache_thread_shutdown (void);
1943 /* ------------------ Testing support ----------------------------------*/
1945 static int perturb_byte;
1947 static void
1948 alloc_perturb (char *p, size_t n)
1950 if (__glibc_unlikely (perturb_byte))
1951 memset (p, perturb_byte ^ 0xff, n);
1954 static void
1955 free_perturb (char *p, size_t n)
1957 if (__glibc_unlikely (perturb_byte))
1958 memset (p, perturb_byte, n);
1963 #include <stap-probe.h>
1965 /* ------------------- Support for multiple arenas -------------------- */
1966 #include "arena.c"
1969 Debugging support
1971 These routines make a number of assertions about the states
1972 of data structures that should be true at all times. If any
1973 are not true, it's very likely that a user program has somehow
1974 trashed memory. (It's also possible that there is a coding error
1975 in malloc. In which case, please report it!)
1978 #if !MALLOC_DEBUG
1980 # define check_chunk(A, P)
1981 # define check_free_chunk(A, P)
1982 # define check_inuse_chunk(A, P)
1983 # define check_remalloced_chunk(A, P, N)
1984 # define check_malloced_chunk(A, P, N)
1985 # define check_malloc_state(A)
1987 #else
1989 # define check_chunk(A, P) do_check_chunk (A, P)
1990 # define check_free_chunk(A, P) do_check_free_chunk (A, P)
1991 # define check_inuse_chunk(A, P) do_check_inuse_chunk (A, P)
1992 # define check_remalloced_chunk(A, P, N) do_check_remalloced_chunk (A, P, N)
1993 # define check_malloced_chunk(A, P, N) do_check_malloced_chunk (A, P, N)
1994 # define check_malloc_state(A) do_check_malloc_state (A)
1997 Properties of all chunks
2000 static void
2001 do_check_chunk (mstate av, mchunkptr p)
2003 unsigned long sz = chunksize (p);
2004 /* min and max possible addresses assuming contiguous allocation */
2005 char *max_address = (char *) (av->top) + chunksize (av->top);
2006 char *min_address = max_address - av->system_mem;
2008 if (!chunk_is_mmapped (p))
2010 /* Has legal address ... */
2011 if (p != av->top)
2013 if (contiguous (av))
2015 assert (((char *) p) >= min_address);
2016 assert (((char *) p + sz) <= ((char *) (av->top)));
2019 else
2021 /* top size is always at least MINSIZE */
2022 assert ((unsigned long) (sz) >= MINSIZE);
2023 /* top predecessor always marked inuse */
2024 assert (prev_inuse (p));
2027 else if (!DUMPED_MAIN_ARENA_CHUNK (p))
2029 /* address is outside main heap */
2030 if (contiguous (av) && av->top != initial_top (av))
2032 assert (((char *) p) < min_address || ((char *) p) >= max_address);
2034 /* chunk is page-aligned */
2035 assert (((prev_size (p) + sz) & (GLRO (dl_pagesize) - 1)) == 0);
2036 /* mem is aligned */
2037 assert (aligned_OK (chunk2mem (p)));
2042 Properties of free chunks
2045 static void
2046 do_check_free_chunk (mstate av, mchunkptr p)
2048 INTERNAL_SIZE_T sz = chunksize_nomask (p) & ~(PREV_INUSE | NON_MAIN_ARENA);
2049 mchunkptr next = chunk_at_offset (p, sz);
2051 do_check_chunk (av, p);
2053 /* Chunk must claim to be free ... */
2054 assert (!inuse (p));
2055 assert (!chunk_is_mmapped (p));
2057 /* Unless a special marker, must have OK fields */
2058 if ((unsigned long) (sz) >= MINSIZE)
2060 assert ((sz & MALLOC_ALIGN_MASK) == 0);
2061 assert (aligned_OK (chunk2mem (p)));
2062 /* ... matching footer field */
2063 assert (prev_size (next_chunk (p)) == sz);
2064 /* ... and is fully consolidated */
2065 assert (prev_inuse (p));
2066 assert (next == av->top || inuse (next));
2068 /* ... and has minimally sane links */
2069 assert (p->fd->bk == p);
2070 assert (p->bk->fd == p);
2072 else /* markers are always of size SIZE_SZ */
2073 assert (sz == SIZE_SZ);
2077 Properties of inuse chunks
2080 static void
2081 do_check_inuse_chunk (mstate av, mchunkptr p)
2083 mchunkptr next;
2085 do_check_chunk (av, p);
2087 if (chunk_is_mmapped (p))
2088 return; /* mmapped chunks have no next/prev */
2090 /* Check whether it claims to be in use ... */
2091 assert (inuse (p));
2093 next = next_chunk (p);
2095 /* ... and is surrounded by OK chunks.
2096 Since more things can be checked with free chunks than inuse ones,
2097 if an inuse chunk borders them and debug is on, it's worth doing them.
2099 if (!prev_inuse (p))
2101 /* Note that we cannot even look at prev unless it is not inuse */
2102 mchunkptr prv = prev_chunk (p);
2103 assert (next_chunk (prv) == p);
2104 do_check_free_chunk (av, prv);
2107 if (next == av->top)
2109 assert (prev_inuse (next));
2110 assert (chunksize (next) >= MINSIZE);
2112 else if (!inuse (next))
2113 do_check_free_chunk (av, next);
2117 Properties of chunks recycled from fastbins
2120 static void
2121 do_check_remalloced_chunk (mstate av, mchunkptr p, INTERNAL_SIZE_T s)
2123 INTERNAL_SIZE_T sz = chunksize_nomask (p) & ~(PREV_INUSE | NON_MAIN_ARENA);
2125 if (!chunk_is_mmapped (p))
2127 assert (av == arena_for_chunk (p));
2128 if (chunk_main_arena (p))
2129 assert (av == &main_arena);
2130 else
2131 assert (av != &main_arena);
2134 do_check_inuse_chunk (av, p);
2136 /* Legal size ... */
2137 assert ((sz & MALLOC_ALIGN_MASK) == 0);
2138 assert ((unsigned long) (sz) >= MINSIZE);
2139 /* ... and alignment */
2140 assert (aligned_OK (chunk2mem (p)));
2141 /* chunk is less than MINSIZE more than request */
2142 assert ((long) (sz) - (long) (s) >= 0);
2143 assert ((long) (sz) - (long) (s + MINSIZE) < 0);
2147 Properties of nonrecycled chunks at the point they are malloced
2150 static void
2151 do_check_malloced_chunk (mstate av, mchunkptr p, INTERNAL_SIZE_T s)
2153 /* same as recycled case ... */
2154 do_check_remalloced_chunk (av, p, s);
2157 ... plus, must obey implementation invariant that prev_inuse is
2158 always true of any allocated chunk; i.e., that each allocated
2159 chunk borders either a previously allocated and still in-use
2160 chunk, or the base of its memory arena. This is ensured
2161 by making all allocations from the `lowest' part of any found
2162 chunk. This does not necessarily hold however for chunks
2163 recycled via fastbins.
2166 assert (prev_inuse (p));
2171 Properties of malloc_state.
2173 This may be useful for debugging malloc, as well as detecting user
2174 programmer errors that somehow write into malloc_state.
2176 If you are extending or experimenting with this malloc, you can
2177 probably figure out how to hack this routine to print out or
2178 display chunk addresses, sizes, bins, and other instrumentation.
2181 static void
2182 do_check_malloc_state (mstate av)
2184 int i;
2185 mchunkptr p;
2186 mchunkptr q;
2187 mbinptr b;
2188 unsigned int idx;
2189 INTERNAL_SIZE_T size;
2190 unsigned long total = 0;
2191 int max_fast_bin;
2193 /* internal size_t must be no wider than pointer type */
2194 assert (sizeof (INTERNAL_SIZE_T) <= sizeof (char *));
2196 /* alignment is a power of 2 */
2197 assert ((MALLOC_ALIGNMENT & (MALLOC_ALIGNMENT - 1)) == 0);
2199 /* Check the arena is initialized. */
2200 assert (av->top != 0);
2202 /* No memory has been allocated yet, so doing more tests is not possible. */
2203 if (av->top == initial_top (av))
2204 return;
2206 /* pagesize is a power of 2 */
2207 assert (powerof2(GLRO (dl_pagesize)));
2209 /* A contiguous main_arena is consistent with sbrk_base. */
2210 if (av == &main_arena && contiguous (av))
2211 assert ((char *) mp_.sbrk_base + av->system_mem ==
2212 (char *) av->top + chunksize (av->top));
2214 /* properties of fastbins */
2216 /* max_fast is in allowed range */
2217 assert ((get_max_fast () & ~1) <= request2size (MAX_FAST_SIZE));
2219 max_fast_bin = fastbin_index (get_max_fast ());
2221 for (i = 0; i < NFASTBINS; ++i)
2223 p = fastbin (av, i);
2225 /* The following test can only be performed for the main arena.
2226 While mallopt calls malloc_consolidate to get rid of all fast
2227 bins (especially those larger than the new maximum) this does
2228 only happen for the main arena. Trying to do this for any
2229 other arena would mean those arenas have to be locked and
2230 malloc_consolidate be called for them. This is excessive. And
2231 even if this is acceptable to somebody it still cannot solve
2232 the problem completely since if the arena is locked a
2233 concurrent malloc call might create a new arena which then
2234 could use the newly invalid fast bins. */
2236 /* all bins past max_fast are empty */
2237 if (av == &main_arena && i > max_fast_bin)
2238 assert (p == 0);
2240 while (p != 0)
2242 /* each chunk claims to be inuse */
2243 do_check_inuse_chunk (av, p);
2244 total += chunksize (p);
2245 /* chunk belongs in this bin */
2246 assert (fastbin_index (chunksize (p)) == i);
2247 p = p->fd;
2251 /* check normal bins */
2252 for (i = 1; i < NBINS; ++i)
2254 b = bin_at (av, i);
2256 /* binmap is accurate (except for bin 1 == unsorted_chunks) */
2257 if (i >= 2)
2259 unsigned int binbit = get_binmap (av, i);
2260 int empty = last (b) == b;
2261 if (!binbit)
2262 assert (empty);
2263 else if (!empty)
2264 assert (binbit);
2267 for (p = last (b); p != b; p = p->bk)
2269 /* each chunk claims to be free */
2270 do_check_free_chunk (av, p);
2271 size = chunksize (p);
2272 total += size;
2273 if (i >= 2)
2275 /* chunk belongs in bin */
2276 idx = bin_index (size);
2277 assert (idx == i);
2278 /* lists are sorted */
2279 assert (p->bk == b ||
2280 (unsigned long) chunksize (p->bk) >= (unsigned long) chunksize (p));
2282 if (!in_smallbin_range (size))
2284 if (p->fd_nextsize != NULL)
2286 if (p->fd_nextsize == p)
2287 assert (p->bk_nextsize == p);
2288 else
2290 if (p->fd_nextsize == first (b))
2291 assert (chunksize (p) < chunksize (p->fd_nextsize));
2292 else
2293 assert (chunksize (p) > chunksize (p->fd_nextsize));
2295 if (p == first (b))
2296 assert (chunksize (p) > chunksize (p->bk_nextsize));
2297 else
2298 assert (chunksize (p) < chunksize (p->bk_nextsize));
2301 else
2302 assert (p->bk_nextsize == NULL);
2305 else if (!in_smallbin_range (size))
2306 assert (p->fd_nextsize == NULL && p->bk_nextsize == NULL);
2307 /* chunk is followed by a legal chain of inuse chunks */
2308 for (q = next_chunk (p);
2309 (q != av->top && inuse (q) &&
2310 (unsigned long) (chunksize (q)) >= MINSIZE);
2311 q = next_chunk (q))
2312 do_check_inuse_chunk (av, q);
2316 /* top chunk is OK */
2317 check_chunk (av, av->top);
2319 #endif
2322 /* ----------------- Support for debugging hooks -------------------- */
2323 #include "hooks.c"
2326 /* ----------- Routines dealing with system allocation -------------- */
2329 sysmalloc handles malloc cases requiring more memory from the system.
2330 On entry, it is assumed that av->top does not have enough
2331 space to service request for nb bytes, thus requiring that av->top
2332 be extended or replaced.
2335 static void *
2336 sysmalloc (INTERNAL_SIZE_T nb, mstate av)
2338 mchunkptr old_top; /* incoming value of av->top */
2339 INTERNAL_SIZE_T old_size; /* its size */
2340 char *old_end; /* its end address */
2342 long size; /* arg to first MORECORE or mmap call */
2343 char *brk; /* return value from MORECORE */
2345 long correction; /* arg to 2nd MORECORE call */
2346 char *snd_brk; /* 2nd return val */
2348 INTERNAL_SIZE_T front_misalign; /* unusable bytes at front of new space */
2349 INTERNAL_SIZE_T end_misalign; /* partial page left at end of new space */
2350 char *aligned_brk; /* aligned offset into brk */
2352 mchunkptr p; /* the allocated/returned chunk */
2353 mchunkptr remainder; /* remainder from allocation */
2354 unsigned long remainder_size; /* its size */
2357 size_t pagesize = GLRO (dl_pagesize);
2358 bool tried_mmap = false;
2362 If have mmap, and the request size meets the mmap threshold, and
2363 the system supports mmap, and there are few enough currently
2364 allocated mmapped regions, try to directly map this request
2365 rather than expanding top.
2368 if (av == NULL
2369 || ((unsigned long) (nb) >= (unsigned long) (mp_.mmap_threshold)
2370 && (mp_.n_mmaps < mp_.n_mmaps_max)))
2372 char *mm; /* return value from mmap call*/
2374 try_mmap:
2376 Round up size to nearest page. For mmapped chunks, the overhead
2377 is one SIZE_SZ unit larger than for normal chunks, because there
2378 is no following chunk whose prev_size field could be used.
2380 See the front_misalign handling below, for glibc there is no
2381 need for further alignments unless we have have high alignment.
2383 if (MALLOC_ALIGNMENT == 2 * SIZE_SZ)
2384 size = ALIGN_UP (nb + SIZE_SZ, pagesize);
2385 else
2386 size = ALIGN_UP (nb + SIZE_SZ + MALLOC_ALIGN_MASK, pagesize);
2387 tried_mmap = true;
2389 /* Don't try if size wraps around 0 */
2390 if ((unsigned long) (size) > (unsigned long) (nb))
2392 mm = (char *) (MMAP (0, size, PROT_READ | PROT_WRITE, 0));
2394 if (mm != MAP_FAILED)
2397 The offset to the start of the mmapped region is stored
2398 in the prev_size field of the chunk. This allows us to adjust
2399 returned start address to meet alignment requirements here
2400 and in memalign(), and still be able to compute proper
2401 address argument for later munmap in free() and realloc().
2404 if (MALLOC_ALIGNMENT == 2 * SIZE_SZ)
2406 /* For glibc, chunk2mem increases the address by 2*SIZE_SZ and
2407 MALLOC_ALIGN_MASK is 2*SIZE_SZ-1. Each mmap'ed area is page
2408 aligned and therefore definitely MALLOC_ALIGN_MASK-aligned. */
2409 assert (((INTERNAL_SIZE_T) chunk2mem (mm) & MALLOC_ALIGN_MASK) == 0);
2410 front_misalign = 0;
2412 else
2413 front_misalign = (INTERNAL_SIZE_T) chunk2mem (mm) & MALLOC_ALIGN_MASK;
2414 if (front_misalign > 0)
2416 correction = MALLOC_ALIGNMENT - front_misalign;
2417 p = (mchunkptr) (mm + correction);
2418 set_prev_size (p, correction);
2419 set_head (p, (size - correction) | IS_MMAPPED);
2421 else
2423 p = (mchunkptr) mm;
2424 set_prev_size (p, 0);
2425 set_head (p, size | IS_MMAPPED);
2428 /* update statistics */
2430 int new = atomic_exchange_and_add (&mp_.n_mmaps, 1) + 1;
2431 atomic_max (&mp_.max_n_mmaps, new);
2433 unsigned long sum;
2434 sum = atomic_exchange_and_add (&mp_.mmapped_mem, size) + size;
2435 atomic_max (&mp_.max_mmapped_mem, sum);
2437 check_chunk (av, p);
2439 return chunk2mem (p);
2444 /* There are no usable arenas and mmap also failed. */
2445 if (av == NULL)
2446 return 0;
2448 /* Record incoming configuration of top */
2450 old_top = av->top;
2451 old_size = chunksize (old_top);
2452 old_end = (char *) (chunk_at_offset (old_top, old_size));
2454 brk = snd_brk = (char *) (MORECORE_FAILURE);
2457 If not the first time through, we require old_size to be
2458 at least MINSIZE and to have prev_inuse set.
2461 assert ((old_top == initial_top (av) && old_size == 0) ||
2462 ((unsigned long) (old_size) >= MINSIZE &&
2463 prev_inuse (old_top) &&
2464 ((unsigned long) old_end & (pagesize - 1)) == 0));
2466 /* Precondition: not enough current space to satisfy nb request */
2467 assert ((unsigned long) (old_size) < (unsigned long) (nb + MINSIZE));
2470 if (av != &main_arena)
2472 heap_info *old_heap, *heap;
2473 size_t old_heap_size;
2475 /* First try to extend the current heap. */
2476 old_heap = heap_for_ptr (old_top);
2477 old_heap_size = old_heap->size;
2478 if ((long) (MINSIZE + nb - old_size) > 0
2479 && grow_heap (old_heap, MINSIZE + nb - old_size) == 0)
2481 av->system_mem += old_heap->size - old_heap_size;
2482 set_head (old_top, (((char *) old_heap + old_heap->size) - (char *) old_top)
2483 | PREV_INUSE);
2485 else if ((heap = new_heap (nb + (MINSIZE + sizeof (*heap)), mp_.top_pad)))
2487 /* Use a newly allocated heap. */
2488 heap->ar_ptr = av;
2489 heap->prev = old_heap;
2490 av->system_mem += heap->size;
2491 /* Set up the new top. */
2492 top (av) = chunk_at_offset (heap, sizeof (*heap));
2493 set_head (top (av), (heap->size - sizeof (*heap)) | PREV_INUSE);
2495 /* Setup fencepost and free the old top chunk with a multiple of
2496 MALLOC_ALIGNMENT in size. */
2497 /* The fencepost takes at least MINSIZE bytes, because it might
2498 become the top chunk again later. Note that a footer is set
2499 up, too, although the chunk is marked in use. */
2500 old_size = (old_size - MINSIZE) & ~MALLOC_ALIGN_MASK;
2501 set_head (chunk_at_offset (old_top, old_size + 2 * SIZE_SZ), 0 | PREV_INUSE);
2502 if (old_size >= MINSIZE)
2504 set_head (chunk_at_offset (old_top, old_size), (2 * SIZE_SZ) | PREV_INUSE);
2505 set_foot (chunk_at_offset (old_top, old_size), (2 * SIZE_SZ));
2506 set_head (old_top, old_size | PREV_INUSE | NON_MAIN_ARENA);
2507 _int_free (av, old_top, 1);
2509 else
2511 set_head (old_top, (old_size + 2 * SIZE_SZ) | PREV_INUSE);
2512 set_foot (old_top, (old_size + 2 * SIZE_SZ));
2515 else if (!tried_mmap)
2516 /* We can at least try to use to mmap memory. */
2517 goto try_mmap;
2519 else /* av == main_arena */
2522 { /* Request enough space for nb + pad + overhead */
2523 size = nb + mp_.top_pad + MINSIZE;
2526 If contiguous, we can subtract out existing space that we hope to
2527 combine with new space. We add it back later only if
2528 we don't actually get contiguous space.
2531 if (contiguous (av))
2532 size -= old_size;
2535 Round to a multiple of page size.
2536 If MORECORE is not contiguous, this ensures that we only call it
2537 with whole-page arguments. And if MORECORE is contiguous and
2538 this is not first time through, this preserves page-alignment of
2539 previous calls. Otherwise, we correct to page-align below.
2542 size = ALIGN_UP (size, pagesize);
2545 Don't try to call MORECORE if argument is so big as to appear
2546 negative. Note that since mmap takes size_t arg, it may succeed
2547 below even if we cannot call MORECORE.
2550 if (size > 0)
2552 brk = (char *) (MORECORE (size));
2553 LIBC_PROBE (memory_sbrk_more, 2, brk, size);
2556 if (brk != (char *) (MORECORE_FAILURE))
2558 /* Call the `morecore' hook if necessary. */
2559 void (*hook) (void) = atomic_forced_read (__after_morecore_hook);
2560 if (__builtin_expect (hook != NULL, 0))
2561 (*hook)();
2563 else
2566 If have mmap, try using it as a backup when MORECORE fails or
2567 cannot be used. This is worth doing on systems that have "holes" in
2568 address space, so sbrk cannot extend to give contiguous space, but
2569 space is available elsewhere. Note that we ignore mmap max count
2570 and threshold limits, since the space will not be used as a
2571 segregated mmap region.
2574 /* Cannot merge with old top, so add its size back in */
2575 if (contiguous (av))
2576 size = ALIGN_UP (size + old_size, pagesize);
2578 /* If we are relying on mmap as backup, then use larger units */
2579 if ((unsigned long) (size) < (unsigned long) (MMAP_AS_MORECORE_SIZE))
2580 size = MMAP_AS_MORECORE_SIZE;
2582 /* Don't try if size wraps around 0 */
2583 if ((unsigned long) (size) > (unsigned long) (nb))
2585 char *mbrk = (char *) (MMAP (0, size, PROT_READ | PROT_WRITE, 0));
2587 if (mbrk != MAP_FAILED)
2589 /* We do not need, and cannot use, another sbrk call to find end */
2590 brk = mbrk;
2591 snd_brk = brk + size;
2594 Record that we no longer have a contiguous sbrk region.
2595 After the first time mmap is used as backup, we do not
2596 ever rely on contiguous space since this could incorrectly
2597 bridge regions.
2599 set_noncontiguous (av);
2604 if (brk != (char *) (MORECORE_FAILURE))
2606 if (mp_.sbrk_base == 0)
2607 mp_.sbrk_base = brk;
2608 av->system_mem += size;
2611 If MORECORE extends previous space, we can likewise extend top size.
2614 if (brk == old_end && snd_brk == (char *) (MORECORE_FAILURE))
2615 set_head (old_top, (size + old_size) | PREV_INUSE);
2617 else if (contiguous (av) && old_size && brk < old_end)
2618 /* Oops! Someone else killed our space.. Can't touch anything. */
2619 malloc_printerr ("break adjusted to free malloc space");
2622 Otherwise, make adjustments:
2624 * If the first time through or noncontiguous, we need to call sbrk
2625 just to find out where the end of memory lies.
2627 * We need to ensure that all returned chunks from malloc will meet
2628 MALLOC_ALIGNMENT
2630 * If there was an intervening foreign sbrk, we need to adjust sbrk
2631 request size to account for fact that we will not be able to
2632 combine new space with existing space in old_top.
2634 * Almost all systems internally allocate whole pages at a time, in
2635 which case we might as well use the whole last page of request.
2636 So we allocate enough more memory to hit a page boundary now,
2637 which in turn causes future contiguous calls to page-align.
2640 else
2642 front_misalign = 0;
2643 end_misalign = 0;
2644 correction = 0;
2645 aligned_brk = brk;
2647 /* handle contiguous cases */
2648 if (contiguous (av))
2650 /* Count foreign sbrk as system_mem. */
2651 if (old_size)
2652 av->system_mem += brk - old_end;
2654 /* Guarantee alignment of first new chunk made from this space */
2656 front_misalign = (INTERNAL_SIZE_T) chunk2mem (brk) & MALLOC_ALIGN_MASK;
2657 if (front_misalign > 0)
2660 Skip over some bytes to arrive at an aligned position.
2661 We don't need to specially mark these wasted front bytes.
2662 They will never be accessed anyway because
2663 prev_inuse of av->top (and any chunk created from its start)
2664 is always true after initialization.
2667 correction = MALLOC_ALIGNMENT - front_misalign;
2668 aligned_brk += correction;
2672 If this isn't adjacent to existing space, then we will not
2673 be able to merge with old_top space, so must add to 2nd request.
2676 correction += old_size;
2678 /* Extend the end address to hit a page boundary */
2679 end_misalign = (INTERNAL_SIZE_T) (brk + size + correction);
2680 correction += (ALIGN_UP (end_misalign, pagesize)) - end_misalign;
2682 assert (correction >= 0);
2683 snd_brk = (char *) (MORECORE (correction));
2686 If can't allocate correction, try to at least find out current
2687 brk. It might be enough to proceed without failing.
2689 Note that if second sbrk did NOT fail, we assume that space
2690 is contiguous with first sbrk. This is a safe assumption unless
2691 program is multithreaded but doesn't use locks and a foreign sbrk
2692 occurred between our first and second calls.
2695 if (snd_brk == (char *) (MORECORE_FAILURE))
2697 correction = 0;
2698 snd_brk = (char *) (MORECORE (0));
2700 else
2702 /* Call the `morecore' hook if necessary. */
2703 void (*hook) (void) = atomic_forced_read (__after_morecore_hook);
2704 if (__builtin_expect (hook != NULL, 0))
2705 (*hook)();
2709 /* handle non-contiguous cases */
2710 else
2712 if (MALLOC_ALIGNMENT == 2 * SIZE_SZ)
2713 /* MORECORE/mmap must correctly align */
2714 assert (((unsigned long) chunk2mem (brk) & MALLOC_ALIGN_MASK) == 0);
2715 else
2717 front_misalign = (INTERNAL_SIZE_T) chunk2mem (brk) & MALLOC_ALIGN_MASK;
2718 if (front_misalign > 0)
2721 Skip over some bytes to arrive at an aligned position.
2722 We don't need to specially mark these wasted front bytes.
2723 They will never be accessed anyway because
2724 prev_inuse of av->top (and any chunk created from its start)
2725 is always true after initialization.
2728 aligned_brk += MALLOC_ALIGNMENT - front_misalign;
2732 /* Find out current end of memory */
2733 if (snd_brk == (char *) (MORECORE_FAILURE))
2735 snd_brk = (char *) (MORECORE (0));
2739 /* Adjust top based on results of second sbrk */
2740 if (snd_brk != (char *) (MORECORE_FAILURE))
2742 av->top = (mchunkptr) aligned_brk;
2743 set_head (av->top, (snd_brk - aligned_brk + correction) | PREV_INUSE);
2744 av->system_mem += correction;
2747 If not the first time through, we either have a
2748 gap due to foreign sbrk or a non-contiguous region. Insert a
2749 double fencepost at old_top to prevent consolidation with space
2750 we don't own. These fenceposts are artificial chunks that are
2751 marked as inuse and are in any case too small to use. We need
2752 two to make sizes and alignments work out.
2755 if (old_size != 0)
2758 Shrink old_top to insert fenceposts, keeping size a
2759 multiple of MALLOC_ALIGNMENT. We know there is at least
2760 enough space in old_top to do this.
2762 old_size = (old_size - 4 * SIZE_SZ) & ~MALLOC_ALIGN_MASK;
2763 set_head (old_top, old_size | PREV_INUSE);
2766 Note that the following assignments completely overwrite
2767 old_top when old_size was previously MINSIZE. This is
2768 intentional. We need the fencepost, even if old_top otherwise gets
2769 lost.
2771 set_head (chunk_at_offset (old_top, old_size),
2772 (2 * SIZE_SZ) | PREV_INUSE);
2773 set_head (chunk_at_offset (old_top, old_size + 2 * SIZE_SZ),
2774 (2 * SIZE_SZ) | PREV_INUSE);
2776 /* If possible, release the rest. */
2777 if (old_size >= MINSIZE)
2779 _int_free (av, old_top, 1);
2785 } /* if (av != &main_arena) */
2787 if ((unsigned long) av->system_mem > (unsigned long) (av->max_system_mem))
2788 av->max_system_mem = av->system_mem;
2789 check_malloc_state (av);
2791 /* finally, do the allocation */
2792 p = av->top;
2793 size = chunksize (p);
2795 /* check that one of the above allocation paths succeeded */
2796 if ((unsigned long) (size) >= (unsigned long) (nb + MINSIZE))
2798 remainder_size = size - nb;
2799 remainder = chunk_at_offset (p, nb);
2800 av->top = remainder;
2801 set_head (p, nb | PREV_INUSE | (av != &main_arena ? NON_MAIN_ARENA : 0));
2802 set_head (remainder, remainder_size | PREV_INUSE);
2803 check_malloced_chunk (av, p, nb);
2804 return chunk2mem (p);
2807 /* catch all failure paths */
2808 __set_errno (ENOMEM);
2809 return 0;
2814 systrim is an inverse of sorts to sysmalloc. It gives memory back
2815 to the system (via negative arguments to sbrk) if there is unused
2816 memory at the `high' end of the malloc pool. It is called
2817 automatically by free() when top space exceeds the trim
2818 threshold. It is also called by the public malloc_trim routine. It
2819 returns 1 if it actually released any memory, else 0.
2822 static int
2823 systrim (size_t pad, mstate av)
2825 long top_size; /* Amount of top-most memory */
2826 long extra; /* Amount to release */
2827 long released; /* Amount actually released */
2828 char *current_brk; /* address returned by pre-check sbrk call */
2829 char *new_brk; /* address returned by post-check sbrk call */
2830 size_t pagesize;
2831 long top_area;
2833 pagesize = GLRO (dl_pagesize);
2834 top_size = chunksize (av->top);
2836 top_area = top_size - MINSIZE - 1;
2837 if (top_area <= pad)
2838 return 0;
2840 /* Release in pagesize units and round down to the nearest page. */
2841 extra = ALIGN_DOWN(top_area - pad, pagesize);
2843 if (extra == 0)
2844 return 0;
2847 Only proceed if end of memory is where we last set it.
2848 This avoids problems if there were foreign sbrk calls.
2850 current_brk = (char *) (MORECORE (0));
2851 if (current_brk == (char *) (av->top) + top_size)
2854 Attempt to release memory. We ignore MORECORE return value,
2855 and instead call again to find out where new end of memory is.
2856 This avoids problems if first call releases less than we asked,
2857 of if failure somehow altered brk value. (We could still
2858 encounter problems if it altered brk in some very bad way,
2859 but the only thing we can do is adjust anyway, which will cause
2860 some downstream failure.)
2863 MORECORE (-extra);
2864 /* Call the `morecore' hook if necessary. */
2865 void (*hook) (void) = atomic_forced_read (__after_morecore_hook);
2866 if (__builtin_expect (hook != NULL, 0))
2867 (*hook)();
2868 new_brk = (char *) (MORECORE (0));
2870 LIBC_PROBE (memory_sbrk_less, 2, new_brk, extra);
2872 if (new_brk != (char *) MORECORE_FAILURE)
2874 released = (long) (current_brk - new_brk);
2876 if (released != 0)
2878 /* Success. Adjust top. */
2879 av->system_mem -= released;
2880 set_head (av->top, (top_size - released) | PREV_INUSE);
2881 check_malloc_state (av);
2882 return 1;
2886 return 0;
2889 static void
2890 munmap_chunk (mchunkptr p)
2892 INTERNAL_SIZE_T size = chunksize (p);
2894 assert (chunk_is_mmapped (p));
2896 /* Do nothing if the chunk is a faked mmapped chunk in the dumped
2897 main arena. We never free this memory. */
2898 if (DUMPED_MAIN_ARENA_CHUNK (p))
2899 return;
2901 uintptr_t block = (uintptr_t) p - prev_size (p);
2902 size_t total_size = prev_size (p) + size;
2903 /* Unfortunately we have to do the compilers job by hand here. Normally
2904 we would test BLOCK and TOTAL-SIZE separately for compliance with the
2905 page size. But gcc does not recognize the optimization possibility
2906 (in the moment at least) so we combine the two values into one before
2907 the bit test. */
2908 if (__builtin_expect (((block | total_size) & (GLRO (dl_pagesize) - 1)) != 0, 0))
2909 malloc_printerr ("munmap_chunk(): invalid pointer");
2911 atomic_decrement (&mp_.n_mmaps);
2912 atomic_add (&mp_.mmapped_mem, -total_size);
2914 /* If munmap failed the process virtual memory address space is in a
2915 bad shape. Just leave the block hanging around, the process will
2916 terminate shortly anyway since not much can be done. */
2917 __munmap ((char *) block, total_size);
2920 #if HAVE_MREMAP
2922 static mchunkptr
2923 mremap_chunk (mchunkptr p, size_t new_size)
2925 size_t pagesize = GLRO (dl_pagesize);
2926 INTERNAL_SIZE_T offset = prev_size (p);
2927 INTERNAL_SIZE_T size = chunksize (p);
2928 char *cp;
2930 assert (chunk_is_mmapped (p));
2931 assert (((size + offset) & (GLRO (dl_pagesize) - 1)) == 0);
2933 /* Note the extra SIZE_SZ overhead as in mmap_chunk(). */
2934 new_size = ALIGN_UP (new_size + offset + SIZE_SZ, pagesize);
2936 /* No need to remap if the number of pages does not change. */
2937 if (size + offset == new_size)
2938 return p;
2940 cp = (char *) __mremap ((char *) p - offset, size + offset, new_size,
2941 MREMAP_MAYMOVE);
2943 if (cp == MAP_FAILED)
2944 return 0;
2946 p = (mchunkptr) (cp + offset);
2948 assert (aligned_OK (chunk2mem (p)));
2950 assert (prev_size (p) == offset);
2951 set_head (p, (new_size - offset) | IS_MMAPPED);
2953 INTERNAL_SIZE_T new;
2954 new = atomic_exchange_and_add (&mp_.mmapped_mem, new_size - size - offset)
2955 + new_size - size - offset;
2956 atomic_max (&mp_.max_mmapped_mem, new);
2957 return p;
2959 #endif /* HAVE_MREMAP */
2961 /*------------------------ Public wrappers. --------------------------------*/
2963 #if USE_TCACHE
2965 /* We overlay this structure on the user-data portion of a chunk when
2966 the chunk is stored in the per-thread cache. */
2967 typedef struct tcache_entry
2969 struct tcache_entry *next;
2970 } tcache_entry;
2972 /* There is one of these for each thread, which contains the
2973 per-thread cache (hence "tcache_perthread_struct"). Keeping
2974 overall size low is mildly important. Note that COUNTS and ENTRIES
2975 are redundant (we could have just counted the linked list each
2976 time), this is for performance reasons. */
2977 typedef struct tcache_perthread_struct
2979 char counts[TCACHE_MAX_BINS];
2980 tcache_entry *entries[TCACHE_MAX_BINS];
2981 } tcache_perthread_struct;
2983 static __thread bool tcache_shutting_down = false;
2984 static __thread tcache_perthread_struct *tcache = NULL;
2986 /* Caller must ensure that we know tc_idx is valid and there's room
2987 for more chunks. */
2988 static __always_inline void
2989 tcache_put (mchunkptr chunk, size_t tc_idx)
2991 tcache_entry *e = (tcache_entry *) chunk2mem (chunk);
2992 assert (tc_idx < TCACHE_MAX_BINS);
2993 e->next = tcache->entries[tc_idx];
2994 tcache->entries[tc_idx] = e;
2995 ++(tcache->counts[tc_idx]);
2998 /* Caller must ensure that we know tc_idx is valid and there's
2999 available chunks to remove. */
3000 static __always_inline void *
3001 tcache_get (size_t tc_idx)
3003 tcache_entry *e = tcache->entries[tc_idx];
3004 assert (tc_idx < TCACHE_MAX_BINS);
3005 assert (tcache->entries[tc_idx] > 0);
3006 tcache->entries[tc_idx] = e->next;
3007 --(tcache->counts[tc_idx]);
3008 return (void *) e;
3011 static void
3012 tcache_thread_shutdown (void)
3014 int i;
3015 tcache_perthread_struct *tcache_tmp = tcache;
3017 if (!tcache)
3018 return;
3020 /* Disable the tcache and prevent it from being reinitialized. */
3021 tcache = NULL;
3022 tcache_shutting_down = true;
3024 /* Free all of the entries and the tcache itself back to the arena
3025 heap for coalescing. */
3026 for (i = 0; i < TCACHE_MAX_BINS; ++i)
3028 while (tcache_tmp->entries[i])
3030 tcache_entry *e = tcache_tmp->entries[i];
3031 tcache_tmp->entries[i] = e->next;
3032 __libc_free (e);
3036 __libc_free (tcache_tmp);
3039 static void
3040 tcache_init(void)
3042 mstate ar_ptr;
3043 void *victim = 0;
3044 const size_t bytes = sizeof (tcache_perthread_struct);
3046 if (tcache_shutting_down)
3047 return;
3049 arena_get (ar_ptr, bytes);
3050 victim = _int_malloc (ar_ptr, bytes);
3051 if (!victim && ar_ptr != NULL)
3053 ar_ptr = arena_get_retry (ar_ptr, bytes);
3054 victim = _int_malloc (ar_ptr, bytes);
3058 if (ar_ptr != NULL)
3059 __libc_lock_unlock (ar_ptr->mutex);
3061 /* In a low memory situation, we may not be able to allocate memory
3062 - in which case, we just keep trying later. However, we
3063 typically do this very early, so either there is sufficient
3064 memory, or there isn't enough memory to do non-trivial
3065 allocations anyway. */
3066 if (victim)
3068 tcache = (tcache_perthread_struct *) victim;
3069 memset (tcache, 0, sizeof (tcache_perthread_struct));
3074 # define MAYBE_INIT_TCACHE() \
3075 if (__glibc_unlikely (tcache == NULL)) \
3076 tcache_init();
3078 #else /* !USE_TCACHE */
3079 # define MAYBE_INIT_TCACHE()
3081 static void
3082 tcache_thread_shutdown (void)
3084 /* Nothing to do if there is no thread cache. */
3087 #endif /* !USE_TCACHE */
3089 void *
3090 __libc_malloc (size_t bytes)
3092 mstate ar_ptr;
3093 void *victim;
3095 void *(*hook) (size_t, const void *)
3096 = atomic_forced_read (__malloc_hook);
3097 if (__builtin_expect (hook != NULL, 0))
3098 return (*hook)(bytes, RETURN_ADDRESS (0));
3099 #if USE_TCACHE
3100 /* int_free also calls request2size, be careful to not pad twice. */
3101 size_t tbytes;
3102 checked_request2size (bytes, tbytes);
3103 size_t tc_idx = csize2tidx (tbytes);
3105 MAYBE_INIT_TCACHE ();
3107 DIAG_PUSH_NEEDS_COMMENT;
3108 if (tc_idx < mp_.tcache_bins
3109 /*&& tc_idx < TCACHE_MAX_BINS*/ /* to appease gcc */
3110 && tcache
3111 && tcache->entries[tc_idx] != NULL)
3113 return tcache_get (tc_idx);
3115 DIAG_POP_NEEDS_COMMENT;
3116 #endif
3118 if (SINGLE_THREAD_P)
3120 victim = _int_malloc (&main_arena, bytes);
3121 assert (!victim || chunk_is_mmapped (mem2chunk (victim)) ||
3122 &main_arena == arena_for_chunk (mem2chunk (victim)));
3123 return victim;
3126 arena_get (ar_ptr, bytes);
3128 victim = _int_malloc (ar_ptr, bytes);
3129 /* Retry with another arena only if we were able to find a usable arena
3130 before. */
3131 if (!victim && ar_ptr != NULL)
3133 LIBC_PROBE (memory_malloc_retry, 1, bytes);
3134 ar_ptr = arena_get_retry (ar_ptr, bytes);
3135 victim = _int_malloc (ar_ptr, bytes);
3138 if (ar_ptr != NULL)
3139 __libc_lock_unlock (ar_ptr->mutex);
3141 assert (!victim || chunk_is_mmapped (mem2chunk (victim)) ||
3142 ar_ptr == arena_for_chunk (mem2chunk (victim)));
3143 return victim;
3145 libc_hidden_def (__libc_malloc)
3147 void
3148 __libc_free (void *mem)
3150 mstate ar_ptr;
3151 mchunkptr p; /* chunk corresponding to mem */
3153 void (*hook) (void *, const void *)
3154 = atomic_forced_read (__free_hook);
3155 if (__builtin_expect (hook != NULL, 0))
3157 (*hook)(mem, RETURN_ADDRESS (0));
3158 return;
3161 if (mem == 0) /* free(0) has no effect */
3162 return;
3164 p = mem2chunk (mem);
3166 if (chunk_is_mmapped (p)) /* release mmapped memory. */
3168 /* See if the dynamic brk/mmap threshold needs adjusting.
3169 Dumped fake mmapped chunks do not affect the threshold. */
3170 if (!mp_.no_dyn_threshold
3171 && chunksize_nomask (p) > mp_.mmap_threshold
3172 && chunksize_nomask (p) <= DEFAULT_MMAP_THRESHOLD_MAX
3173 && !DUMPED_MAIN_ARENA_CHUNK (p))
3175 mp_.mmap_threshold = chunksize (p);
3176 mp_.trim_threshold = 2 * mp_.mmap_threshold;
3177 LIBC_PROBE (memory_mallopt_free_dyn_thresholds, 2,
3178 mp_.mmap_threshold, mp_.trim_threshold);
3180 munmap_chunk (p);
3181 return;
3184 MAYBE_INIT_TCACHE ();
3186 ar_ptr = arena_for_chunk (p);
3187 _int_free (ar_ptr, p, 0);
3189 libc_hidden_def (__libc_free)
3191 void *
3192 __libc_realloc (void *oldmem, size_t bytes)
3194 mstate ar_ptr;
3195 INTERNAL_SIZE_T nb; /* padded request size */
3197 void *newp; /* chunk to return */
3199 void *(*hook) (void *, size_t, const void *) =
3200 atomic_forced_read (__realloc_hook);
3201 if (__builtin_expect (hook != NULL, 0))
3202 return (*hook)(oldmem, bytes, RETURN_ADDRESS (0));
3204 #if REALLOC_ZERO_BYTES_FREES
3205 if (bytes == 0 && oldmem != NULL)
3207 __libc_free (oldmem); return 0;
3209 #endif
3211 /* realloc of null is supposed to be same as malloc */
3212 if (oldmem == 0)
3213 return __libc_malloc (bytes);
3215 /* chunk corresponding to oldmem */
3216 const mchunkptr oldp = mem2chunk (oldmem);
3217 /* its size */
3218 const INTERNAL_SIZE_T oldsize = chunksize (oldp);
3220 if (chunk_is_mmapped (oldp))
3221 ar_ptr = NULL;
3222 else
3224 MAYBE_INIT_TCACHE ();
3225 ar_ptr = arena_for_chunk (oldp);
3228 /* Little security check which won't hurt performance: the allocator
3229 never wrapps around at the end of the address space. Therefore
3230 we can exclude some size values which might appear here by
3231 accident or by "design" from some intruder. We need to bypass
3232 this check for dumped fake mmap chunks from the old main arena
3233 because the new malloc may provide additional alignment. */
3234 if ((__builtin_expect ((uintptr_t) oldp > (uintptr_t) -oldsize, 0)
3235 || __builtin_expect (misaligned_chunk (oldp), 0))
3236 && !DUMPED_MAIN_ARENA_CHUNK (oldp))
3237 malloc_printerr ("realloc(): invalid pointer");
3239 checked_request2size (bytes, nb);
3241 if (chunk_is_mmapped (oldp))
3243 /* If this is a faked mmapped chunk from the dumped main arena,
3244 always make a copy (and do not free the old chunk). */
3245 if (DUMPED_MAIN_ARENA_CHUNK (oldp))
3247 /* Must alloc, copy, free. */
3248 void *newmem = __libc_malloc (bytes);
3249 if (newmem == 0)
3250 return NULL;
3251 /* Copy as many bytes as are available from the old chunk
3252 and fit into the new size. NB: The overhead for faked
3253 mmapped chunks is only SIZE_SZ, not 2 * SIZE_SZ as for
3254 regular mmapped chunks. */
3255 if (bytes > oldsize - SIZE_SZ)
3256 bytes = oldsize - SIZE_SZ;
3257 memcpy (newmem, oldmem, bytes);
3258 return newmem;
3261 void *newmem;
3263 #if HAVE_MREMAP
3264 newp = mremap_chunk (oldp, nb);
3265 if (newp)
3266 return chunk2mem (newp);
3267 #endif
3268 /* Note the extra SIZE_SZ overhead. */
3269 if (oldsize - SIZE_SZ >= nb)
3270 return oldmem; /* do nothing */
3272 /* Must alloc, copy, free. */
3273 newmem = __libc_malloc (bytes);
3274 if (newmem == 0)
3275 return 0; /* propagate failure */
3277 memcpy (newmem, oldmem, oldsize - 2 * SIZE_SZ);
3278 munmap_chunk (oldp);
3279 return newmem;
3282 if (SINGLE_THREAD_P)
3284 newp = _int_realloc (ar_ptr, oldp, oldsize, nb);
3285 assert (!newp || chunk_is_mmapped (mem2chunk (newp)) ||
3286 ar_ptr == arena_for_chunk (mem2chunk (newp)));
3288 return newp;
3291 __libc_lock_lock (ar_ptr->mutex);
3293 newp = _int_realloc (ar_ptr, oldp, oldsize, nb);
3295 __libc_lock_unlock (ar_ptr->mutex);
3296 assert (!newp || chunk_is_mmapped (mem2chunk (newp)) ||
3297 ar_ptr == arena_for_chunk (mem2chunk (newp)));
3299 if (newp == NULL)
3301 /* Try harder to allocate memory in other arenas. */
3302 LIBC_PROBE (memory_realloc_retry, 2, bytes, oldmem);
3303 newp = __libc_malloc (bytes);
3304 if (newp != NULL)
3306 memcpy (newp, oldmem, oldsize - SIZE_SZ);
3307 _int_free (ar_ptr, oldp, 0);
3311 return newp;
3313 libc_hidden_def (__libc_realloc)
3315 void *
3316 __libc_memalign (size_t alignment, size_t bytes)
3318 void *address = RETURN_ADDRESS (0);
3319 return _mid_memalign (alignment, bytes, address);
3322 static void *
3323 _mid_memalign (size_t alignment, size_t bytes, void *address)
3325 mstate ar_ptr;
3326 void *p;
3328 void *(*hook) (size_t, size_t, const void *) =
3329 atomic_forced_read (__memalign_hook);
3330 if (__builtin_expect (hook != NULL, 0))
3331 return (*hook)(alignment, bytes, address);
3333 /* If we need less alignment than we give anyway, just relay to malloc. */
3334 if (alignment <= MALLOC_ALIGNMENT)
3335 return __libc_malloc (bytes);
3337 /* Otherwise, ensure that it is at least a minimum chunk size */
3338 if (alignment < MINSIZE)
3339 alignment = MINSIZE;
3341 /* If the alignment is greater than SIZE_MAX / 2 + 1 it cannot be a
3342 power of 2 and will cause overflow in the check below. */
3343 if (alignment > SIZE_MAX / 2 + 1)
3345 __set_errno (EINVAL);
3346 return 0;
3349 /* Check for overflow. */
3350 if (bytes > SIZE_MAX - alignment - MINSIZE)
3352 __set_errno (ENOMEM);
3353 return 0;
3357 /* Make sure alignment is power of 2. */
3358 if (!powerof2 (alignment))
3360 size_t a = MALLOC_ALIGNMENT * 2;
3361 while (a < alignment)
3362 a <<= 1;
3363 alignment = a;
3366 if (SINGLE_THREAD_P)
3368 p = _int_memalign (&main_arena, alignment, bytes);
3369 assert (!p || chunk_is_mmapped (mem2chunk (p)) ||
3370 &main_arena == arena_for_chunk (mem2chunk (p)));
3372 return p;
3375 arena_get (ar_ptr, bytes + alignment + MINSIZE);
3377 p = _int_memalign (ar_ptr, alignment, bytes);
3378 if (!p && ar_ptr != NULL)
3380 LIBC_PROBE (memory_memalign_retry, 2, bytes, alignment);
3381 ar_ptr = arena_get_retry (ar_ptr, bytes);
3382 p = _int_memalign (ar_ptr, alignment, bytes);
3385 if (ar_ptr != NULL)
3386 __libc_lock_unlock (ar_ptr->mutex);
3388 assert (!p || chunk_is_mmapped (mem2chunk (p)) ||
3389 ar_ptr == arena_for_chunk (mem2chunk (p)));
3390 return p;
3392 /* For ISO C11. */
3393 weak_alias (__libc_memalign, aligned_alloc)
3394 libc_hidden_def (__libc_memalign)
3396 void *
3397 __libc_valloc (size_t bytes)
3399 if (__malloc_initialized < 0)
3400 ptmalloc_init ();
3402 void *address = RETURN_ADDRESS (0);
3403 size_t pagesize = GLRO (dl_pagesize);
3404 return _mid_memalign (pagesize, bytes, address);
3407 void *
3408 __libc_pvalloc (size_t bytes)
3410 if (__malloc_initialized < 0)
3411 ptmalloc_init ();
3413 void *address = RETURN_ADDRESS (0);
3414 size_t pagesize = GLRO (dl_pagesize);
3415 size_t rounded_bytes = ALIGN_UP (bytes, pagesize);
3417 /* Check for overflow. */
3418 if (bytes > SIZE_MAX - 2 * pagesize - MINSIZE)
3420 __set_errno (ENOMEM);
3421 return 0;
3424 return _mid_memalign (pagesize, rounded_bytes, address);
3427 void *
3428 __libc_calloc (size_t n, size_t elem_size)
3430 mstate av;
3431 mchunkptr oldtop, p;
3432 INTERNAL_SIZE_T bytes, sz, csz, oldtopsize;
3433 void *mem;
3434 unsigned long clearsize;
3435 unsigned long nclears;
3436 INTERNAL_SIZE_T *d;
3438 /* size_t is unsigned so the behavior on overflow is defined. */
3439 bytes = n * elem_size;
3440 #define HALF_INTERNAL_SIZE_T \
3441 (((INTERNAL_SIZE_T) 1) << (8 * sizeof (INTERNAL_SIZE_T) / 2))
3442 if (__builtin_expect ((n | elem_size) >= HALF_INTERNAL_SIZE_T, 0))
3444 if (elem_size != 0 && bytes / elem_size != n)
3446 __set_errno (ENOMEM);
3447 return 0;
3451 void *(*hook) (size_t, const void *) =
3452 atomic_forced_read (__malloc_hook);
3453 if (__builtin_expect (hook != NULL, 0))
3455 sz = bytes;
3456 mem = (*hook)(sz, RETURN_ADDRESS (0));
3457 if (mem == 0)
3458 return 0;
3460 return memset (mem, 0, sz);
3463 sz = bytes;
3465 MAYBE_INIT_TCACHE ();
3467 if (SINGLE_THREAD_P)
3468 av = &main_arena;
3469 else
3470 arena_get (av, sz);
3472 if (av)
3474 /* Check if we hand out the top chunk, in which case there may be no
3475 need to clear. */
3476 #if MORECORE_CLEARS
3477 oldtop = top (av);
3478 oldtopsize = chunksize (top (av));
3479 # if MORECORE_CLEARS < 2
3480 /* Only newly allocated memory is guaranteed to be cleared. */
3481 if (av == &main_arena &&
3482 oldtopsize < mp_.sbrk_base + av->max_system_mem - (char *) oldtop)
3483 oldtopsize = (mp_.sbrk_base + av->max_system_mem - (char *) oldtop);
3484 # endif
3485 if (av != &main_arena)
3487 heap_info *heap = heap_for_ptr (oldtop);
3488 if (oldtopsize < (char *) heap + heap->mprotect_size - (char *) oldtop)
3489 oldtopsize = (char *) heap + heap->mprotect_size - (char *) oldtop;
3491 #endif
3493 else
3495 /* No usable arenas. */
3496 oldtop = 0;
3497 oldtopsize = 0;
3499 mem = _int_malloc (av, sz);
3501 assert (!mem || chunk_is_mmapped (mem2chunk (mem)) ||
3502 av == arena_for_chunk (mem2chunk (mem)));
3504 if (!SINGLE_THREAD_P)
3506 if (mem == 0 && av != NULL)
3508 LIBC_PROBE (memory_calloc_retry, 1, sz);
3509 av = arena_get_retry (av, sz);
3510 mem = _int_malloc (av, sz);
3513 if (av != NULL)
3514 __libc_lock_unlock (av->mutex);
3517 /* Allocation failed even after a retry. */
3518 if (mem == 0)
3519 return 0;
3521 p = mem2chunk (mem);
3523 /* Two optional cases in which clearing not necessary */
3524 if (chunk_is_mmapped (p))
3526 if (__builtin_expect (perturb_byte, 0))
3527 return memset (mem, 0, sz);
3529 return mem;
3532 csz = chunksize (p);
3534 #if MORECORE_CLEARS
3535 if (perturb_byte == 0 && (p == oldtop && csz > oldtopsize))
3537 /* clear only the bytes from non-freshly-sbrked memory */
3538 csz = oldtopsize;
3540 #endif
3542 /* Unroll clear of <= 36 bytes (72 if 8byte sizes). We know that
3543 contents have an odd number of INTERNAL_SIZE_T-sized words;
3544 minimally 3. */
3545 d = (INTERNAL_SIZE_T *) mem;
3546 clearsize = csz - SIZE_SZ;
3547 nclears = clearsize / sizeof (INTERNAL_SIZE_T);
3548 assert (nclears >= 3);
3550 if (nclears > 9)
3551 return memset (d, 0, clearsize);
3553 else
3555 *(d + 0) = 0;
3556 *(d + 1) = 0;
3557 *(d + 2) = 0;
3558 if (nclears > 4)
3560 *(d + 3) = 0;
3561 *(d + 4) = 0;
3562 if (nclears > 6)
3564 *(d + 5) = 0;
3565 *(d + 6) = 0;
3566 if (nclears > 8)
3568 *(d + 7) = 0;
3569 *(d + 8) = 0;
3575 return mem;
3579 ------------------------------ malloc ------------------------------
3582 static void *
3583 _int_malloc (mstate av, size_t bytes)
3585 INTERNAL_SIZE_T nb; /* normalized request size */
3586 unsigned int idx; /* associated bin index */
3587 mbinptr bin; /* associated bin */
3589 mchunkptr victim; /* inspected/selected chunk */
3590 INTERNAL_SIZE_T size; /* its size */
3591 int victim_index; /* its bin index */
3593 mchunkptr remainder; /* remainder from a split */
3594 unsigned long remainder_size; /* its size */
3596 unsigned int block; /* bit map traverser */
3597 unsigned int bit; /* bit map traverser */
3598 unsigned int map; /* current word of binmap */
3600 mchunkptr fwd; /* misc temp for linking */
3601 mchunkptr bck; /* misc temp for linking */
3603 #if USE_TCACHE
3604 size_t tcache_unsorted_count; /* count of unsorted chunks processed */
3605 #endif
3608 Convert request size to internal form by adding SIZE_SZ bytes
3609 overhead plus possibly more to obtain necessary alignment and/or
3610 to obtain a size of at least MINSIZE, the smallest allocatable
3611 size. Also, checked_request2size traps (returning 0) request sizes
3612 that are so large that they wrap around zero when padded and
3613 aligned.
3616 checked_request2size (bytes, nb);
3618 /* There are no usable arenas. Fall back to sysmalloc to get a chunk from
3619 mmap. */
3620 if (__glibc_unlikely (av == NULL))
3622 void *p = sysmalloc (nb, av);
3623 if (p != NULL)
3624 alloc_perturb (p, bytes);
3625 return p;
3629 If the size qualifies as a fastbin, first check corresponding bin.
3630 This code is safe to execute even if av is not yet initialized, so we
3631 can try it without checking, which saves some time on this fast path.
3634 if ((unsigned long) (nb) <= (unsigned long) (get_max_fast ()))
3636 idx = fastbin_index (nb);
3637 mfastbinptr *fb = &fastbin (av, idx);
3638 victim = fastbin_pop_entry (fb);
3639 if (victim != NULL)
3641 size_t victim_idx = fastbin_index (chunksize (victim));
3642 if (victim_idx != idx)
3643 malloc_printerr ("malloc(): memory corruption (fast)");
3644 check_remalloced_chunk (av, victim, nb);
3645 #if USE_TCACHE
3646 /* While we're here, if we see other chunks of the same size,
3647 stash them in the tcache. */
3648 size_t tc_idx = csize2tidx (nb);
3649 if (tcache && tc_idx < mp_.tcache_bins)
3651 /* While bin not empty and tcache not full, copy chunks. */
3652 while (tcache->counts[tc_idx] < mp_.tcache_count)
3654 mchunkptr tc_victim = fastbin_pop_entry (fb);
3655 if (tc_victim == NULL)
3656 break;
3657 tcache_put (tc_victim, tc_idx);
3660 #endif
3661 void *p = chunk2mem (victim);
3662 alloc_perturb (p, bytes);
3663 return p;
3668 If a small request, check regular bin. Since these "smallbins"
3669 hold one size each, no searching within bins is necessary.
3670 (For a large request, we need to wait until unsorted chunks are
3671 processed to find best fit. But for small ones, fits are exact
3672 anyway, so we can check now, which is faster.)
3675 if (in_smallbin_range (nb))
3677 idx = smallbin_index (nb);
3678 bin = bin_at (av, idx);
3680 if ((victim = last (bin)) != bin)
3682 bck = victim->bk;
3683 if (__glibc_unlikely (bck->fd != victim))
3684 malloc_printerr ("malloc(): smallbin double linked list corrupted");
3685 set_inuse_bit_at_offset (victim, nb);
3686 bin->bk = bck;
3687 bck->fd = bin;
3689 if (av != &main_arena)
3690 set_non_main_arena (victim);
3691 check_malloced_chunk (av, victim, nb);
3692 #if USE_TCACHE
3693 /* While we're here, if we see other chunks of the same size,
3694 stash them in the tcache. */
3695 size_t tc_idx = csize2tidx (nb);
3696 if (tcache && tc_idx < mp_.tcache_bins)
3698 mchunkptr tc_victim;
3700 /* While bin not empty and tcache not full, copy chunks over. */
3701 while (tcache->counts[tc_idx] < mp_.tcache_count
3702 && (tc_victim = last (bin)) != bin)
3704 if (tc_victim != 0)
3706 bck = tc_victim->bk;
3707 set_inuse_bit_at_offset (tc_victim, nb);
3708 if (av != &main_arena)
3709 set_non_main_arena (tc_victim);
3710 bin->bk = bck;
3711 bck->fd = bin;
3713 tcache_put (tc_victim, tc_idx);
3717 #endif
3718 void *p = chunk2mem (victim);
3719 alloc_perturb (p, bytes);
3720 return p;
3725 If this is a large request, consolidate fastbins before continuing.
3726 While it might look excessive to kill all fastbins before
3727 even seeing if there is space available, this avoids
3728 fragmentation problems normally associated with fastbins.
3729 Also, in practice, programs tend to have runs of either small or
3730 large requests, but less often mixtures, so consolidation is not
3731 invoked all that often in most programs. And the programs that
3732 it is called frequently in otherwise tend to fragment.
3735 else
3737 idx = largebin_index (nb);
3738 if (atomic_load_relaxed (&av->have_fastchunks))
3739 malloc_consolidate (av);
3743 Process recently freed or remaindered chunks, taking one only if
3744 it is exact fit, or, if this a small request, the chunk is remainder from
3745 the most recent non-exact fit. Place other traversed chunks in
3746 bins. Note that this step is the only place in any routine where
3747 chunks are placed in bins.
3749 The outer loop here is needed because we might not realize until
3750 near the end of malloc that we should have consolidated, so must
3751 do so and retry. This happens at most once, and only when we would
3752 otherwise need to expand memory to service a "small" request.
3755 #if USE_TCACHE
3756 INTERNAL_SIZE_T tcache_nb = 0;
3757 size_t tc_idx = csize2tidx (nb);
3758 if (tcache && tc_idx < mp_.tcache_bins)
3759 tcache_nb = nb;
3760 int return_cached = 0;
3762 tcache_unsorted_count = 0;
3763 #endif
3765 for (;; )
3767 int iters = 0;
3768 while ((victim = unsorted_chunks (av)->bk) != unsorted_chunks (av))
3770 bck = victim->bk;
3771 size = chunksize (victim);
3772 mchunkptr next = chunk_at_offset (victim, size);
3774 if (__glibc_unlikely (size <= 2 * SIZE_SZ)
3775 || __glibc_unlikely (size > av->system_mem))
3776 malloc_printerr ("malloc(): invalid size (unsorted)");
3777 if (__glibc_unlikely (chunksize_nomask (next) < 2 * SIZE_SZ)
3778 || __glibc_unlikely (chunksize_nomask (next) > av->system_mem))
3779 malloc_printerr ("malloc(): invalid next size (unsorted)");
3780 if (__glibc_unlikely ((prev_size (next) & ~(SIZE_BITS)) != size))
3781 malloc_printerr ("malloc(): mismatching next->prev_size (unsorted)");
3782 if (__glibc_unlikely (bck->fd != victim)
3783 || __glibc_unlikely (victim->fd != unsorted_chunks (av)))
3784 malloc_printerr ("malloc(): unsorted double linked list corrupted");
3785 if (__glibc_unlikely (prev_inuse (next)))
3786 malloc_printerr ("malloc(): invalid next->prev_inuse (unsorted)");
3789 If a small request, try to use last remainder if it is the
3790 only chunk in unsorted bin. This helps promote locality for
3791 runs of consecutive small requests. This is the only
3792 exception to best-fit, and applies only when there is
3793 no exact fit for a small chunk.
3796 if (in_smallbin_range (nb) &&
3797 bck == unsorted_chunks (av) &&
3798 victim == av->last_remainder &&
3799 (unsigned long) (size) > (unsigned long) (nb + MINSIZE))
3801 /* split and reattach remainder */
3802 remainder_size = size - nb;
3803 remainder = chunk_at_offset (victim, nb);
3804 unsorted_chunks (av)->bk = unsorted_chunks (av)->fd = remainder;
3805 av->last_remainder = remainder;
3806 remainder->bk = remainder->fd = unsorted_chunks (av);
3807 if (!in_smallbin_range (remainder_size))
3809 remainder->fd_nextsize = NULL;
3810 remainder->bk_nextsize = NULL;
3813 set_head (victim, nb | PREV_INUSE |
3814 (av != &main_arena ? NON_MAIN_ARENA : 0));
3815 set_head (remainder, remainder_size | PREV_INUSE);
3816 set_foot (remainder, remainder_size);
3818 check_malloced_chunk (av, victim, nb);
3819 void *p = chunk2mem (victim);
3820 alloc_perturb (p, bytes);
3821 return p;
3824 /* remove from unsorted list */
3825 if (__glibc_unlikely (bck->fd != victim))
3826 malloc_printerr ("malloc(): corrupted unsorted chunks 3");
3827 unsorted_chunks (av)->bk = bck;
3828 bck->fd = unsorted_chunks (av);
3830 /* Take now instead of binning if exact fit */
3832 if (size == nb)
3834 set_inuse_bit_at_offset (victim, size);
3835 if (av != &main_arena)
3836 set_non_main_arena (victim);
3837 #if USE_TCACHE
3838 /* Fill cache first, return to user only if cache fills.
3839 We may return one of these chunks later. */
3840 if (tcache_nb
3841 && tcache->counts[tc_idx] < mp_.tcache_count)
3843 tcache_put (victim, tc_idx);
3844 return_cached = 1;
3845 continue;
3847 else
3849 #endif
3850 check_malloced_chunk (av, victim, nb);
3851 void *p = chunk2mem (victim);
3852 alloc_perturb (p, bytes);
3853 return p;
3854 #if USE_TCACHE
3856 #endif
3859 /* place chunk in bin */
3861 if (in_smallbin_range (size))
3863 victim_index = smallbin_index (size);
3864 bck = bin_at (av, victim_index);
3865 fwd = bck->fd;
3867 else
3869 victim_index = largebin_index (size);
3870 bck = bin_at (av, victim_index);
3871 fwd = bck->fd;
3873 /* maintain large bins in sorted order */
3874 if (fwd != bck)
3876 /* Or with inuse bit to speed comparisons */
3877 size |= PREV_INUSE;
3878 /* if smaller than smallest, bypass loop below */
3879 assert (chunk_main_arena (bck->bk));
3880 if ((unsigned long) (size)
3881 < (unsigned long) chunksize_nomask (bck->bk))
3883 fwd = bck;
3884 bck = bck->bk;
3886 victim->fd_nextsize = fwd->fd;
3887 victim->bk_nextsize = fwd->fd->bk_nextsize;
3888 fwd->fd->bk_nextsize = victim->bk_nextsize->fd_nextsize = victim;
3890 else
3892 assert (chunk_main_arena (fwd));
3893 while ((unsigned long) size < chunksize_nomask (fwd))
3895 fwd = fwd->fd_nextsize;
3896 assert (chunk_main_arena (fwd));
3899 if ((unsigned long) size
3900 == (unsigned long) chunksize_nomask (fwd))
3901 /* Always insert in the second position. */
3902 fwd = fwd->fd;
3903 else
3905 victim->fd_nextsize = fwd;
3906 victim->bk_nextsize = fwd->bk_nextsize;
3907 fwd->bk_nextsize = victim;
3908 victim->bk_nextsize->fd_nextsize = victim;
3910 bck = fwd->bk;
3913 else
3914 victim->fd_nextsize = victim->bk_nextsize = victim;
3917 mark_bin (av, victim_index);
3918 victim->bk = bck;
3919 victim->fd = fwd;
3920 fwd->bk = victim;
3921 bck->fd = victim;
3923 #if USE_TCACHE
3924 /* If we've processed as many chunks as we're allowed while
3925 filling the cache, return one of the cached ones. */
3926 ++tcache_unsorted_count;
3927 if (return_cached
3928 && mp_.tcache_unsorted_limit > 0
3929 && tcache_unsorted_count > mp_.tcache_unsorted_limit)
3931 return tcache_get (tc_idx);
3933 #endif
3935 #define MAX_ITERS 10000
3936 if (++iters >= MAX_ITERS)
3937 break;
3940 #if USE_TCACHE
3941 /* If all the small chunks we found ended up cached, return one now. */
3942 if (return_cached)
3944 return tcache_get (tc_idx);
3946 #endif
3949 If a large request, scan through the chunks of current bin in
3950 sorted order to find smallest that fits. Use the skip list for this.
3953 if (!in_smallbin_range (nb))
3955 bin = bin_at (av, idx);
3957 /* skip scan if empty or largest chunk is too small */
3958 if ((victim = first (bin)) != bin
3959 && (unsigned long) chunksize_nomask (victim)
3960 >= (unsigned long) (nb))
3962 victim = victim->bk_nextsize;
3963 while (((unsigned long) (size = chunksize (victim)) <
3964 (unsigned long) (nb)))
3965 victim = victim->bk_nextsize;
3967 /* Avoid removing the first entry for a size so that the skip
3968 list does not have to be rerouted. */
3969 if (victim != last (bin)
3970 && chunksize_nomask (victim)
3971 == chunksize_nomask (victim->fd))
3972 victim = victim->fd;
3974 remainder_size = size - nb;
3975 unlink_chunk (av, victim);
3977 /* Exhaust */
3978 if (remainder_size < MINSIZE)
3980 set_inuse_bit_at_offset (victim, size);
3981 if (av != &main_arena)
3982 set_non_main_arena (victim);
3984 /* Split */
3985 else
3987 remainder = chunk_at_offset (victim, nb);
3988 /* We cannot assume the unsorted list is empty and therefore
3989 have to perform a complete insert here. */
3990 bck = unsorted_chunks (av);
3991 fwd = bck->fd;
3992 if (__glibc_unlikely (fwd->bk != bck))
3993 malloc_printerr ("malloc(): corrupted unsorted chunks");
3994 remainder->bk = bck;
3995 remainder->fd = fwd;
3996 bck->fd = remainder;
3997 fwd->bk = remainder;
3998 if (!in_smallbin_range (remainder_size))
4000 remainder->fd_nextsize = NULL;
4001 remainder->bk_nextsize = NULL;
4003 set_head (victim, nb | PREV_INUSE |
4004 (av != &main_arena ? NON_MAIN_ARENA : 0));
4005 set_head (remainder, remainder_size | PREV_INUSE);
4006 set_foot (remainder, remainder_size);
4008 check_malloced_chunk (av, victim, nb);
4009 void *p = chunk2mem (victim);
4010 alloc_perturb (p, bytes);
4011 return p;
4016 Search for a chunk by scanning bins, starting with next largest
4017 bin. This search is strictly by best-fit; i.e., the smallest
4018 (with ties going to approximately the least recently used) chunk
4019 that fits is selected.
4021 The bitmap avoids needing to check that most blocks are nonempty.
4022 The particular case of skipping all bins during warm-up phases
4023 when no chunks have been returned yet is faster than it might look.
4026 ++idx;
4027 bin = bin_at (av, idx);
4028 block = idx2block (idx);
4029 map = av->binmap[block];
4030 bit = idx2bit (idx);
4032 for (;; )
4034 /* Skip rest of block if there are no more set bits in this block. */
4035 if (bit > map || bit == 0)
4039 if (++block >= BINMAPSIZE) /* out of bins */
4040 goto use_top;
4042 while ((map = av->binmap[block]) == 0);
4044 bin = bin_at (av, (block << BINMAPSHIFT));
4045 bit = 1;
4048 /* Advance to bin with set bit. There must be one. */
4049 while ((bit & map) == 0)
4051 bin = next_bin (bin);
4052 bit <<= 1;
4053 assert (bit != 0);
4056 /* Inspect the bin. It is likely to be non-empty */
4057 victim = last (bin);
4059 /* If a false alarm (empty bin), clear the bit. */
4060 if (victim == bin)
4062 av->binmap[block] = map &= ~bit; /* Write through */
4063 bin = next_bin (bin);
4064 bit <<= 1;
4067 else
4069 size = chunksize (victim);
4071 /* We know the first chunk in this bin is big enough to use. */
4072 assert ((unsigned long) (size) >= (unsigned long) (nb));
4074 remainder_size = size - nb;
4076 /* unlink */
4077 unlink_chunk (av, victim);
4079 /* Exhaust */
4080 if (remainder_size < MINSIZE)
4082 set_inuse_bit_at_offset (victim, size);
4083 if (av != &main_arena)
4084 set_non_main_arena (victim);
4087 /* Split */
4088 else
4090 remainder = chunk_at_offset (victim, nb);
4092 /* We cannot assume the unsorted list is empty and therefore
4093 have to perform a complete insert here. */
4094 bck = unsorted_chunks (av);
4095 fwd = bck->fd;
4096 if (__glibc_unlikely (fwd->bk != bck))
4097 malloc_printerr ("malloc(): corrupted unsorted chunks 2");
4098 remainder->bk = bck;
4099 remainder->fd = fwd;
4100 bck->fd = remainder;
4101 fwd->bk = remainder;
4103 /* advertise as last remainder */
4104 if (in_smallbin_range (nb))
4105 av->last_remainder = remainder;
4106 if (!in_smallbin_range (remainder_size))
4108 remainder->fd_nextsize = NULL;
4109 remainder->bk_nextsize = NULL;
4111 set_head (victim, nb | PREV_INUSE |
4112 (av != &main_arena ? NON_MAIN_ARENA : 0));
4113 set_head (remainder, remainder_size | PREV_INUSE);
4114 set_foot (remainder, remainder_size);
4116 check_malloced_chunk (av, victim, nb);
4117 void *p = chunk2mem (victim);
4118 alloc_perturb (p, bytes);
4119 return p;
4123 use_top:
4125 If large enough, split off the chunk bordering the end of memory
4126 (held in av->top). Note that this is in accord with the best-fit
4127 search rule. In effect, av->top is treated as larger (and thus
4128 less well fitting) than any other available chunk since it can
4129 be extended to be as large as necessary (up to system
4130 limitations).
4132 We require that av->top always exists (i.e., has size >=
4133 MINSIZE) after initialization, so if it would otherwise be
4134 exhausted by current request, it is replenished. (The main
4135 reason for ensuring it exists is that we may need MINSIZE space
4136 to put in fenceposts in sysmalloc.)
4139 victim = av->top;
4140 size = chunksize (victim);
4142 if (__glibc_unlikely (size > av->system_mem))
4143 malloc_printerr ("malloc(): corrupted top size");
4145 if ((unsigned long) (size) >= (unsigned long) (nb + MINSIZE))
4147 remainder_size = size - nb;
4148 remainder = chunk_at_offset (victim, nb);
4149 av->top = remainder;
4150 set_head (victim, nb | PREV_INUSE |
4151 (av != &main_arena ? NON_MAIN_ARENA : 0));
4152 set_head (remainder, remainder_size | PREV_INUSE);
4154 check_malloced_chunk (av, victim, nb);
4155 void *p = chunk2mem (victim);
4156 alloc_perturb (p, bytes);
4157 return p;
4160 /* When we are using atomic ops to free fast chunks we can get
4161 here for all block sizes. */
4162 else if (atomic_load_relaxed (&av->have_fastchunks))
4164 malloc_consolidate (av);
4165 /* restore original bin index */
4166 if (in_smallbin_range (nb))
4167 idx = smallbin_index (nb);
4168 else
4169 idx = largebin_index (nb);
4173 Otherwise, relay to handle system-dependent cases
4175 else
4177 void *p = sysmalloc (nb, av);
4178 if (p != NULL)
4179 alloc_perturb (p, bytes);
4180 return p;
4186 ------------------------------ free ------------------------------
4189 static void
4190 _int_free (mstate av, mchunkptr p, int have_lock)
4192 INTERNAL_SIZE_T size; /* its size */
4193 mfastbinptr *fb; /* associated fastbin */
4194 mchunkptr nextchunk; /* next contiguous chunk */
4195 INTERNAL_SIZE_T nextsize; /* its size */
4196 int nextinuse; /* true if nextchunk is used */
4197 INTERNAL_SIZE_T prevsize; /* size of previous contiguous chunk */
4198 mchunkptr bck; /* misc temp for linking */
4199 mchunkptr fwd; /* misc temp for linking */
4201 size = chunksize (p);
4203 /* Little security check which won't hurt performance: the
4204 allocator never wrapps around at the end of the address space.
4205 Therefore we can exclude some size values which might appear
4206 here by accident or by "design" from some intruder. */
4207 if (__builtin_expect ((uintptr_t) p > (uintptr_t) -size, 0)
4208 || __builtin_expect (misaligned_chunk (p), 0))
4209 malloc_printerr ("free(): invalid pointer");
4210 /* We know that each chunk is at least MINSIZE bytes in size or a
4211 multiple of MALLOC_ALIGNMENT. */
4212 if (__glibc_unlikely (size < MINSIZE || !aligned_OK (size)))
4213 malloc_printerr ("free(): invalid size");
4215 check_inuse_chunk(av, p);
4217 #if USE_TCACHE
4219 size_t tc_idx = csize2tidx (size);
4221 if (tcache
4222 && tc_idx < mp_.tcache_bins
4223 && tcache->counts[tc_idx] < mp_.tcache_count)
4225 tcache_put (p, tc_idx);
4226 return;
4229 #endif
4232 If eligible, place chunk on a fastbin so it can be found
4233 and used quickly in malloc.
4236 if ((unsigned long)(size) <= (unsigned long)(get_max_fast ())
4238 #if TRIM_FASTBINS
4240 If TRIM_FASTBINS set, don't place chunks
4241 bordering top into fastbins
4243 && (chunk_at_offset(p, size) != av->top)
4244 #endif
4247 if (__builtin_expect (chunksize_nomask (chunk_at_offset (p, size))
4248 <= 2 * SIZE_SZ, 0)
4249 || __builtin_expect (chunksize (chunk_at_offset (p, size))
4250 >= av->system_mem, 0))
4252 bool fail = true;
4253 /* We might not have a lock at this point and concurrent modifications
4254 of system_mem might result in a false positive. Redo the test after
4255 getting the lock. */
4256 if (!have_lock)
4258 __libc_lock_lock (av->mutex);
4259 fail = (chunksize_nomask (chunk_at_offset (p, size)) <= 2 * SIZE_SZ
4260 || chunksize (chunk_at_offset (p, size)) >= av->system_mem);
4261 __libc_lock_unlock (av->mutex);
4264 if (fail)
4265 malloc_printerr ("free(): invalid next size (fast)");
4268 free_perturb (chunk2mem(p), size - 2 * SIZE_SZ);
4270 atomic_store_relaxed (&av->have_fastchunks, true);
4271 unsigned int idx = fastbin_index(size);
4272 fb = &fastbin (av, idx);
4274 /* Atomically link P to its fastbin: P->FD = *FB; *FB = P; */
4275 mchunkptr old = fastbin_push_entry (fb, p);
4277 /* Check that size of fastbin chunk at the top is the same as
4278 size of the chunk that we are adding. We can dereference OLD
4279 only if we have the lock, otherwise it might have already been
4280 allocated again. */
4281 if (have_lock && old != NULL
4282 && __builtin_expect (fastbin_index (chunksize (old)) != idx, 0))
4283 malloc_printerr ("invalid fastbin entry (free)");
4287 Consolidate other non-mmapped chunks as they arrive.
4290 else if (!chunk_is_mmapped(p)) {
4292 /* If we're single-threaded, don't lock the arena. */
4293 if (SINGLE_THREAD_P)
4294 have_lock = true;
4296 if (!have_lock)
4297 __libc_lock_lock (av->mutex);
4299 nextchunk = chunk_at_offset(p, size);
4301 /* Lightweight tests: check whether the block is already the
4302 top block. */
4303 if (__glibc_unlikely (p == av->top))
4304 malloc_printerr ("double free or corruption (top)");
4305 /* Or whether the next chunk is beyond the boundaries of the arena. */
4306 if (__builtin_expect (contiguous (av)
4307 && (char *) nextchunk
4308 >= ((char *) av->top + chunksize(av->top)), 0))
4309 malloc_printerr ("double free or corruption (out)");
4310 /* Or whether the block is actually not marked used. */
4311 if (__glibc_unlikely (!prev_inuse(nextchunk)))
4312 malloc_printerr ("double free or corruption (!prev)");
4314 nextsize = chunksize(nextchunk);
4315 if (__builtin_expect (chunksize_nomask (nextchunk) <= 2 * SIZE_SZ, 0)
4316 || __builtin_expect (nextsize >= av->system_mem, 0))
4317 malloc_printerr ("free(): invalid next size (normal)");
4319 free_perturb (chunk2mem(p), size - 2 * SIZE_SZ);
4321 /* consolidate backward */
4322 if (!prev_inuse(p)) {
4323 prevsize = prev_size (p);
4324 size += prevsize;
4325 p = chunk_at_offset(p, -((long) prevsize));
4326 if (__glibc_unlikely (chunksize(p) != prevsize))
4327 malloc_printerr ("corrupted size vs. prev_size while consolidating");
4328 unlink_chunk (av, p);
4331 if (nextchunk != av->top) {
4332 /* get and clear inuse bit */
4333 nextinuse = inuse_bit_at_offset(nextchunk, nextsize);
4335 /* consolidate forward */
4336 if (!nextinuse) {
4337 unlink_chunk (av, nextchunk);
4338 size += nextsize;
4339 } else
4340 clear_inuse_bit_at_offset(nextchunk, 0);
4343 Place the chunk in unsorted chunk list. Chunks are
4344 not placed into regular bins until after they have
4345 been given one chance to be used in malloc.
4348 bck = unsorted_chunks(av);
4349 fwd = bck->fd;
4350 if (__glibc_unlikely (fwd->bk != bck))
4351 malloc_printerr ("free(): corrupted unsorted chunks");
4352 p->fd = fwd;
4353 p->bk = bck;
4354 if (!in_smallbin_range(size))
4356 p->fd_nextsize = NULL;
4357 p->bk_nextsize = NULL;
4359 bck->fd = p;
4360 fwd->bk = p;
4362 set_head(p, size | PREV_INUSE);
4363 set_foot(p, size);
4365 check_free_chunk(av, p);
4369 If the chunk borders the current high end of memory,
4370 consolidate into top
4373 else {
4374 size += nextsize;
4375 set_head(p, size | PREV_INUSE);
4376 av->top = p;
4377 check_chunk(av, p);
4381 If freeing a large space, consolidate possibly-surrounding
4382 chunks. Then, if the total unused topmost memory exceeds trim
4383 threshold, ask malloc_trim to reduce top.
4385 Unless max_fast is 0, we don't know if there are fastbins
4386 bordering top, so we cannot tell for sure whether threshold
4387 has been reached unless fastbins are consolidated. But we
4388 don't want to consolidate on each free. As a compromise,
4389 consolidation is performed if FASTBIN_CONSOLIDATION_THRESHOLD
4390 is reached.
4393 if ((unsigned long)(size) >= FASTBIN_CONSOLIDATION_THRESHOLD) {
4394 if (atomic_load_relaxed (&av->have_fastchunks))
4395 malloc_consolidate(av);
4397 if (av == &main_arena) {
4398 #ifndef MORECORE_CANNOT_TRIM
4399 if ((unsigned long)(chunksize(av->top)) >=
4400 (unsigned long)(mp_.trim_threshold))
4401 systrim(mp_.top_pad, av);
4402 #endif
4403 } else {
4404 /* Always try heap_trim(), even if the top chunk is not
4405 large, because the corresponding heap might go away. */
4406 heap_info *heap = heap_for_ptr(top(av));
4408 assert(heap->ar_ptr == av);
4409 heap_trim(heap, mp_.top_pad);
4413 if (!have_lock)
4414 __libc_lock_unlock (av->mutex);
4417 If the chunk was allocated via mmap, release via munmap().
4420 else {
4421 munmap_chunk (p);
4426 ------------------------- malloc_consolidate -------------------------
4428 malloc_consolidate is a specialized version of free() that tears
4429 down chunks held in fastbins. Free itself cannot be used for this
4430 purpose since, among other things, it might place chunks back onto
4431 fastbins. So, instead, we need to use a minor variant of the same
4432 code.
4435 static void malloc_consolidate(mstate av)
4437 mfastbinptr* fb; /* current fastbin being consolidated */
4438 mfastbinptr* maxfb; /* last fastbin (for loop control) */
4439 mchunkptr p; /* current chunk being consolidated */
4440 mchunkptr nextp; /* next chunk to consolidate */
4441 mchunkptr unsorted_bin; /* bin header */
4442 mchunkptr first_unsorted; /* chunk to link to */
4444 /* These have same use as in free() */
4445 mchunkptr nextchunk;
4446 INTERNAL_SIZE_T size;
4447 INTERNAL_SIZE_T nextsize;
4448 INTERNAL_SIZE_T prevsize;
4449 int nextinuse;
4451 atomic_store_relaxed (&av->have_fastchunks, false);
4453 unsorted_bin = unsorted_chunks(av);
4456 Remove each chunk from fast bin and consolidate it, placing it
4457 then in unsorted bin. Among other reasons for doing this,
4458 placing in unsorted bin avoids needing to calculate actual bins
4459 until malloc is sure that chunks aren't immediately going to be
4460 reused anyway.
4463 maxfb = &fastbin (av, NFASTBINS - 1);
4464 fb = &fastbin (av, 0);
4465 do {
4466 /* Synchronizes with the release MO store in
4467 fastbin_push_entry. */
4468 p = atomic_exchange_acquire (fb, NULL);
4469 if (p != 0) {
4470 do {
4472 unsigned int idx = fastbin_index (chunksize (p));
4473 if ((&fastbin (av, idx)) != fb)
4474 malloc_printerr ("malloc_consolidate(): invalid chunk size");
4477 check_inuse_chunk(av, p);
4478 nextp = p->fd;
4480 /* Slightly streamlined version of consolidation code in free() */
4481 size = chunksize (p);
4482 nextchunk = chunk_at_offset(p, size);
4483 nextsize = chunksize(nextchunk);
4485 if (!prev_inuse(p)) {
4486 prevsize = prev_size (p);
4487 size += prevsize;
4488 p = chunk_at_offset(p, -((long) prevsize));
4489 if (__glibc_unlikely (chunksize(p) != prevsize))
4490 malloc_printerr ("corrupted size vs. prev_size in fastbins");
4491 unlink_chunk (av, p);
4494 if (nextchunk != av->top) {
4495 nextinuse = inuse_bit_at_offset(nextchunk, nextsize);
4497 if (!nextinuse) {
4498 size += nextsize;
4499 unlink_chunk (av, nextchunk);
4500 } else
4501 clear_inuse_bit_at_offset(nextchunk, 0);
4503 first_unsorted = unsorted_bin->fd;
4504 unsorted_bin->fd = p;
4505 first_unsorted->bk = p;
4507 if (!in_smallbin_range (size)) {
4508 p->fd_nextsize = NULL;
4509 p->bk_nextsize = NULL;
4512 set_head(p, size | PREV_INUSE);
4513 p->bk = unsorted_bin;
4514 p->fd = first_unsorted;
4515 set_foot(p, size);
4518 else {
4519 size += nextsize;
4520 set_head(p, size | PREV_INUSE);
4521 av->top = p;
4524 } while ( (p = nextp) != 0);
4527 } while (fb++ != maxfb);
4531 ------------------------------ realloc ------------------------------
4534 void*
4535 _int_realloc(mstate av, mchunkptr oldp, INTERNAL_SIZE_T oldsize,
4536 INTERNAL_SIZE_T nb)
4538 mchunkptr newp; /* chunk to return */
4539 INTERNAL_SIZE_T newsize; /* its size */
4540 void* newmem; /* corresponding user mem */
4542 mchunkptr next; /* next contiguous chunk after oldp */
4544 mchunkptr remainder; /* extra space at end of newp */
4545 unsigned long remainder_size; /* its size */
4547 unsigned long copysize; /* bytes to copy */
4548 unsigned int ncopies; /* INTERNAL_SIZE_T words to copy */
4549 INTERNAL_SIZE_T* s; /* copy source */
4550 INTERNAL_SIZE_T* d; /* copy destination */
4552 /* oldmem size */
4553 if (__builtin_expect (chunksize_nomask (oldp) <= 2 * SIZE_SZ, 0)
4554 || __builtin_expect (oldsize >= av->system_mem, 0))
4555 malloc_printerr ("realloc(): invalid old size");
4557 check_inuse_chunk (av, oldp);
4559 /* All callers already filter out mmap'ed chunks. */
4560 assert (!chunk_is_mmapped (oldp));
4562 next = chunk_at_offset (oldp, oldsize);
4563 INTERNAL_SIZE_T nextsize = chunksize (next);
4564 if (__builtin_expect (chunksize_nomask (next) <= 2 * SIZE_SZ, 0)
4565 || __builtin_expect (nextsize >= av->system_mem, 0))
4566 malloc_printerr ("realloc(): invalid next size");
4568 if ((unsigned long) (oldsize) >= (unsigned long) (nb))
4570 /* already big enough; split below */
4571 newp = oldp;
4572 newsize = oldsize;
4575 else
4577 /* Try to expand forward into top */
4578 if (next == av->top &&
4579 (unsigned long) (newsize = oldsize + nextsize) >=
4580 (unsigned long) (nb + MINSIZE))
4582 set_head_size (oldp, nb | (av != &main_arena ? NON_MAIN_ARENA : 0));
4583 av->top = chunk_at_offset (oldp, nb);
4584 set_head (av->top, (newsize - nb) | PREV_INUSE);
4585 check_inuse_chunk (av, oldp);
4586 return chunk2mem (oldp);
4589 /* Try to expand forward into next chunk; split off remainder below */
4590 else if (next != av->top &&
4591 !inuse (next) &&
4592 (unsigned long) (newsize = oldsize + nextsize) >=
4593 (unsigned long) (nb))
4595 newp = oldp;
4596 unlink_chunk (av, next);
4599 /* allocate, copy, free */
4600 else
4602 newmem = _int_malloc (av, nb - MALLOC_ALIGN_MASK);
4603 if (newmem == 0)
4604 return 0; /* propagate failure */
4606 newp = mem2chunk (newmem);
4607 newsize = chunksize (newp);
4610 Avoid copy if newp is next chunk after oldp.
4612 if (newp == next)
4614 newsize += oldsize;
4615 newp = oldp;
4617 else
4620 Unroll copy of <= 36 bytes (72 if 8byte sizes)
4621 We know that contents have an odd number of
4622 INTERNAL_SIZE_T-sized words; minimally 3.
4625 copysize = oldsize - SIZE_SZ;
4626 s = (INTERNAL_SIZE_T *) (chunk2mem (oldp));
4627 d = (INTERNAL_SIZE_T *) (newmem);
4628 ncopies = copysize / sizeof (INTERNAL_SIZE_T);
4629 assert (ncopies >= 3);
4631 if (ncopies > 9)
4632 memcpy (d, s, copysize);
4634 else
4636 *(d + 0) = *(s + 0);
4637 *(d + 1) = *(s + 1);
4638 *(d + 2) = *(s + 2);
4639 if (ncopies > 4)
4641 *(d + 3) = *(s + 3);
4642 *(d + 4) = *(s + 4);
4643 if (ncopies > 6)
4645 *(d + 5) = *(s + 5);
4646 *(d + 6) = *(s + 6);
4647 if (ncopies > 8)
4649 *(d + 7) = *(s + 7);
4650 *(d + 8) = *(s + 8);
4656 _int_free (av, oldp, 1);
4657 check_inuse_chunk (av, newp);
4658 return chunk2mem (newp);
4663 /* If possible, free extra space in old or extended chunk */
4665 assert ((unsigned long) (newsize) >= (unsigned long) (nb));
4667 remainder_size = newsize - nb;
4669 if (remainder_size < MINSIZE) /* not enough extra to split off */
4671 set_head_size (newp, newsize | (av != &main_arena ? NON_MAIN_ARENA : 0));
4672 set_inuse_bit_at_offset (newp, newsize);
4674 else /* split remainder */
4676 remainder = chunk_at_offset (newp, nb);
4677 set_head_size (newp, nb | (av != &main_arena ? NON_MAIN_ARENA : 0));
4678 set_head (remainder, remainder_size | PREV_INUSE |
4679 (av != &main_arena ? NON_MAIN_ARENA : 0));
4680 /* Mark remainder as inuse so free() won't complain */
4681 set_inuse_bit_at_offset (remainder, remainder_size);
4682 _int_free (av, remainder, 1);
4685 check_inuse_chunk (av, newp);
4686 return chunk2mem (newp);
4690 ------------------------------ memalign ------------------------------
4693 static void *
4694 _int_memalign (mstate av, size_t alignment, size_t bytes)
4696 INTERNAL_SIZE_T nb; /* padded request size */
4697 char *m; /* memory returned by malloc call */
4698 mchunkptr p; /* corresponding chunk */
4699 char *brk; /* alignment point within p */
4700 mchunkptr newp; /* chunk to return */
4701 INTERNAL_SIZE_T newsize; /* its size */
4702 INTERNAL_SIZE_T leadsize; /* leading space before alignment point */
4703 mchunkptr remainder; /* spare room at end to split off */
4704 unsigned long remainder_size; /* its size */
4705 INTERNAL_SIZE_T size;
4709 checked_request2size (bytes, nb);
4712 Strategy: find a spot within that chunk that meets the alignment
4713 request, and then possibly free the leading and trailing space.
4717 /* Check for overflow. */
4718 if (nb > SIZE_MAX - alignment - MINSIZE)
4720 __set_errno (ENOMEM);
4721 return 0;
4724 /* Call malloc with worst case padding to hit alignment. */
4726 m = (char *) (_int_malloc (av, nb + alignment + MINSIZE));
4728 if (m == 0)
4729 return 0; /* propagate failure */
4731 p = mem2chunk (m);
4733 if ((((unsigned long) (m)) % alignment) != 0) /* misaligned */
4735 { /*
4736 Find an aligned spot inside chunk. Since we need to give back
4737 leading space in a chunk of at least MINSIZE, if the first
4738 calculation places us at a spot with less than MINSIZE leader,
4739 we can move to the next aligned spot -- we've allocated enough
4740 total room so that this is always possible.
4742 brk = (char *) mem2chunk (((unsigned long) (m + alignment - 1)) &
4743 - ((signed long) alignment));
4744 if ((unsigned long) (brk - (char *) (p)) < MINSIZE)
4745 brk += alignment;
4747 newp = (mchunkptr) brk;
4748 leadsize = brk - (char *) (p);
4749 newsize = chunksize (p) - leadsize;
4751 /* For mmapped chunks, just adjust offset */
4752 if (chunk_is_mmapped (p))
4754 set_prev_size (newp, prev_size (p) + leadsize);
4755 set_head (newp, newsize | IS_MMAPPED);
4756 return chunk2mem (newp);
4759 /* Otherwise, give back leader, use the rest */
4760 set_head (newp, newsize | PREV_INUSE |
4761 (av != &main_arena ? NON_MAIN_ARENA : 0));
4762 set_inuse_bit_at_offset (newp, newsize);
4763 set_head_size (p, leadsize | (av != &main_arena ? NON_MAIN_ARENA : 0));
4764 _int_free (av, p, 1);
4765 p = newp;
4767 assert (newsize >= nb &&
4768 (((unsigned long) (chunk2mem (p))) % alignment) == 0);
4771 /* Also give back spare room at the end */
4772 if (!chunk_is_mmapped (p))
4774 size = chunksize (p);
4775 if ((unsigned long) (size) > (unsigned long) (nb + MINSIZE))
4777 remainder_size = size - nb;
4778 remainder = chunk_at_offset (p, nb);
4779 set_head (remainder, remainder_size | PREV_INUSE |
4780 (av != &main_arena ? NON_MAIN_ARENA : 0));
4781 set_head_size (p, nb);
4782 _int_free (av, remainder, 1);
4786 check_inuse_chunk (av, p);
4787 return chunk2mem (p);
4792 ------------------------------ malloc_trim ------------------------------
4795 static int
4796 mtrim (mstate av, size_t pad)
4798 /* Ensure all blocks are consolidated. */
4799 malloc_consolidate (av);
4801 const size_t ps = GLRO (dl_pagesize);
4802 int psindex = bin_index (ps);
4803 const size_t psm1 = ps - 1;
4805 int result = 0;
4806 for (int i = 1; i < NBINS; ++i)
4807 if (i == 1 || i >= psindex)
4809 mbinptr bin = bin_at (av, i);
4811 for (mchunkptr p = last (bin); p != bin; p = p->bk)
4813 INTERNAL_SIZE_T size = chunksize (p);
4815 if (size > psm1 + sizeof (struct malloc_chunk))
4817 /* See whether the chunk contains at least one unused page. */
4818 char *paligned_mem = (char *) (((uintptr_t) p
4819 + sizeof (struct malloc_chunk)
4820 + psm1) & ~psm1);
4822 assert ((char *) chunk2mem (p) + 4 * SIZE_SZ <= paligned_mem);
4823 assert ((char *) p + size > paligned_mem);
4825 /* This is the size we could potentially free. */
4826 size -= paligned_mem - (char *) p;
4828 if (size > psm1)
4830 #if MALLOC_DEBUG
4831 /* When debugging we simulate destroying the memory
4832 content. */
4833 memset (paligned_mem, 0x89, size & ~psm1);
4834 #endif
4835 __madvise (paligned_mem, size & ~psm1, MADV_DONTNEED);
4837 result = 1;
4843 #ifndef MORECORE_CANNOT_TRIM
4844 return result | (av == &main_arena ? systrim (pad, av) : 0);
4846 #else
4847 return result;
4848 #endif
4853 __malloc_trim (size_t s)
4855 int result = 0;
4857 if (__malloc_initialized < 0)
4858 ptmalloc_init ();
4860 mstate ar_ptr = &main_arena;
4863 __libc_lock_lock (ar_ptr->mutex);
4864 result |= mtrim (ar_ptr, s);
4865 __libc_lock_unlock (ar_ptr->mutex);
4867 ar_ptr = ar_ptr->next;
4869 while (ar_ptr != &main_arena);
4871 return result;
4876 ------------------------- malloc_usable_size -------------------------
4879 static size_t
4880 musable (void *mem)
4882 mchunkptr p;
4883 if (mem != 0)
4885 p = mem2chunk (mem);
4887 if (__builtin_expect (using_malloc_checking == 1, 0))
4888 return malloc_check_get_size (p);
4890 if (chunk_is_mmapped (p))
4892 if (DUMPED_MAIN_ARENA_CHUNK (p))
4893 return chunksize (p) - SIZE_SZ;
4894 else
4895 return chunksize (p) - 2 * SIZE_SZ;
4897 else if (inuse (p))
4898 return chunksize (p) - SIZE_SZ;
4900 return 0;
4904 size_t
4905 __malloc_usable_size (void *m)
4907 size_t result;
4909 result = musable (m);
4910 return result;
4914 ------------------------------ mallinfo ------------------------------
4915 Accumulate malloc statistics for arena AV into M.
4918 static void
4919 int_mallinfo (mstate av, struct mallinfo *m)
4921 size_t i;
4922 mbinptr b;
4923 mchunkptr p;
4924 INTERNAL_SIZE_T avail;
4925 INTERNAL_SIZE_T fastavail;
4926 int nblocks;
4927 int nfastblocks;
4929 check_malloc_state (av);
4931 /* Account for top */
4932 avail = chunksize (av->top);
4933 nblocks = 1; /* top always exists */
4935 /* traverse fastbins */
4936 nfastblocks = 0;
4937 fastavail = 0;
4939 for (i = 0; i < NFASTBINS; ++i)
4941 for (p = fastbin (av, i); p != 0; p = p->fd)
4943 ++nfastblocks;
4944 fastavail += chunksize (p);
4948 avail += fastavail;
4950 /* traverse regular bins */
4951 for (i = 1; i < NBINS; ++i)
4953 b = bin_at (av, i);
4954 for (p = last (b); p != b; p = p->bk)
4956 ++nblocks;
4957 avail += chunksize (p);
4961 m->smblks += nfastblocks;
4962 m->ordblks += nblocks;
4963 m->fordblks += avail;
4964 m->uordblks += av->system_mem - avail;
4965 m->arena += av->system_mem;
4966 m->fsmblks += fastavail;
4967 if (av == &main_arena)
4969 m->hblks = mp_.n_mmaps;
4970 m->hblkhd = mp_.mmapped_mem;
4971 m->usmblks = 0;
4972 m->keepcost = chunksize (av->top);
4977 struct mallinfo
4978 __libc_mallinfo (void)
4980 struct mallinfo m;
4981 mstate ar_ptr;
4983 if (__malloc_initialized < 0)
4984 ptmalloc_init ();
4986 memset (&m, 0, sizeof (m));
4987 ar_ptr = &main_arena;
4990 __libc_lock_lock (ar_ptr->mutex);
4991 int_mallinfo (ar_ptr, &m);
4992 __libc_lock_unlock (ar_ptr->mutex);
4994 ar_ptr = ar_ptr->next;
4996 while (ar_ptr != &main_arena);
4998 return m;
5002 ------------------------------ malloc_stats ------------------------------
5005 void
5006 __malloc_stats (void)
5008 int i;
5009 mstate ar_ptr;
5010 unsigned int in_use_b = mp_.mmapped_mem, system_b = in_use_b;
5012 if (__malloc_initialized < 0)
5013 ptmalloc_init ();
5014 _IO_flockfile (stderr);
5015 int old_flags2 = stderr->_flags2;
5016 stderr->_flags2 |= _IO_FLAGS2_NOTCANCEL;
5017 for (i = 0, ar_ptr = &main_arena;; i++)
5019 struct mallinfo mi;
5021 memset (&mi, 0, sizeof (mi));
5022 __libc_lock_lock (ar_ptr->mutex);
5023 int_mallinfo (ar_ptr, &mi);
5024 fprintf (stderr, "Arena %d:\n", i);
5025 fprintf (stderr, "system bytes = %10u\n", (unsigned int) mi.arena);
5026 fprintf (stderr, "in use bytes = %10u\n", (unsigned int) mi.uordblks);
5027 #if MALLOC_DEBUG > 1
5028 if (i > 0)
5029 dump_heap (heap_for_ptr (top (ar_ptr)));
5030 #endif
5031 system_b += mi.arena;
5032 in_use_b += mi.uordblks;
5033 __libc_lock_unlock (ar_ptr->mutex);
5034 ar_ptr = ar_ptr->next;
5035 if (ar_ptr == &main_arena)
5036 break;
5038 fprintf (stderr, "Total (incl. mmap):\n");
5039 fprintf (stderr, "system bytes = %10u\n", system_b);
5040 fprintf (stderr, "in use bytes = %10u\n", in_use_b);
5041 fprintf (stderr, "max mmap regions = %10u\n", (unsigned int) mp_.max_n_mmaps);
5042 fprintf (stderr, "max mmap bytes = %10lu\n",
5043 (unsigned long) mp_.max_mmapped_mem);
5044 stderr->_flags2 = old_flags2;
5045 _IO_funlockfile (stderr);
5050 ------------------------------ mallopt ------------------------------
5052 static inline int
5053 __always_inline
5054 do_set_trim_threshold (size_t value)
5056 LIBC_PROBE (memory_mallopt_trim_threshold, 3, value, mp_.trim_threshold,
5057 mp_.no_dyn_threshold);
5058 mp_.trim_threshold = value;
5059 mp_.no_dyn_threshold = 1;
5060 return 1;
5063 static inline int
5064 __always_inline
5065 do_set_top_pad (size_t value)
5067 LIBC_PROBE (memory_mallopt_top_pad, 3, value, mp_.top_pad,
5068 mp_.no_dyn_threshold);
5069 mp_.top_pad = value;
5070 mp_.no_dyn_threshold = 1;
5071 return 1;
5074 static inline int
5075 __always_inline
5076 do_set_mmap_threshold (size_t value)
5078 /* Forbid setting the threshold too high. */
5079 if (value <= HEAP_MAX_SIZE / 2)
5081 LIBC_PROBE (memory_mallopt_mmap_threshold, 3, value, mp_.mmap_threshold,
5082 mp_.no_dyn_threshold);
5083 mp_.mmap_threshold = value;
5084 mp_.no_dyn_threshold = 1;
5085 return 1;
5087 return 0;
5090 static inline int
5091 __always_inline
5092 do_set_mmaps_max (int32_t value)
5094 LIBC_PROBE (memory_mallopt_mmap_max, 3, value, mp_.n_mmaps_max,
5095 mp_.no_dyn_threshold);
5096 mp_.n_mmaps_max = value;
5097 mp_.no_dyn_threshold = 1;
5098 return 1;
5101 static inline int
5102 __always_inline
5103 do_set_mallopt_check (int32_t value)
5105 return 1;
5108 static inline int
5109 __always_inline
5110 do_set_perturb_byte (int32_t value)
5112 LIBC_PROBE (memory_mallopt_perturb, 2, value, perturb_byte);
5113 perturb_byte = value;
5114 return 1;
5117 static inline int
5118 __always_inline
5119 do_set_arena_test (size_t value)
5121 LIBC_PROBE (memory_mallopt_arena_test, 2, value, mp_.arena_test);
5122 mp_.arena_test = value;
5123 return 1;
5126 static inline int
5127 __always_inline
5128 do_set_arena_max (size_t value)
5130 LIBC_PROBE (memory_mallopt_arena_max, 2, value, mp_.arena_max);
5131 mp_.arena_max = value;
5132 return 1;
5135 #if USE_TCACHE
5136 static inline int
5137 __always_inline
5138 do_set_tcache_max (size_t value)
5140 if (value >= 0 && value <= MAX_TCACHE_SIZE)
5142 LIBC_PROBE (memory_tunable_tcache_max_bytes, 2, value, mp_.tcache_max_bytes);