FD.io VPP  v19.04.3-1-gdfec10d13
Vector Packet Processing
dlmalloc.h
Go to the documentation of this file.
1 /*
2  This is a version (aka dlmalloc) of malloc/free/realloc written by
3  Doug Lea and released to the public domain, as explained at
4  http://creativecommons.org/publicdomain/zero/1.0/ Send questions,
5  comments, complaints, performance data, etc to dl@cs.oswego.edu
6 
7 * Version 2.8.6 Wed Aug 29 06:57:58 2012 Doug Lea
8  Note: There may be an updated version of this malloc obtainable at
9  ftp://gee.cs.oswego.edu/pub/misc/malloc.c
10  Check before installing!
11 
12 * Quickstart
13 
14  This library is all in one file to simplify the most common usage:
15  ftp it, compile it (-O3), and link it into another program. All of
16  the compile-time options default to reasonable values for use on
17  most platforms. You might later want to step through various
18  compile-time and dynamic tuning options.
19 
20  For convenience, an include file for code using this malloc is at:
21  ftp://gee.cs.oswego.edu/pub/misc/malloc-2.8.6.h
22  You don't really need this .h file unless you call functions not
23  defined in your system include files. The .h file contains only the
24  excerpts from this file needed for using this malloc on ANSI C/C++
25  systems, so long as you haven't changed compile-time options about
26  naming and tuning parameters. If you do, then you can create your
27  own malloc.h that does include all settings by cutting at the point
28  indicated below. Note that you may already by default be using a C
29  library containing a malloc that is based on some version of this
30  malloc (for example in linux). You might still want to use the one
31  in this file to customize settings or to avoid overheads associated
32  with library versions.
33 
34 * Vital statistics:
35 
36  Supported pointer/size_t representation: 4 or 8 bytes
37  size_t MUST be an unsigned type of the same width as
38  pointers. (If you are using an ancient system that declares
39  size_t as a signed type, or need it to be a different width
40  than pointers, you can use a previous release of this malloc
41  (e.g. 2.7.2) supporting these.)
42 
43  Alignment: 8 bytes (minimum)
44  This suffices for nearly all current machines and C compilers.
45  However, you can define MALLOC_ALIGNMENT to be wider than this
46  if necessary (up to 128bytes), at the expense of using more space.
47 
48  Minimum overhead per allocated chunk: 4 or 8 bytes (if 4byte sizes)
49  8 or 16 bytes (if 8byte sizes)
50  Each malloced chunk has a hidden word of overhead holding size
51  and status information, and additional cross-check word
52  if FOOTERS is defined.
53 
54  Minimum allocated size: 4-byte ptrs: 16 bytes (including overhead)
55  8-byte ptrs: 32 bytes (including overhead)
56 
57  Even a request for zero bytes (i.e., malloc(0)) returns a
58  pointer to something of the minimum allocatable size.
59  The maximum overhead wastage (i.e., number of extra bytes
60  allocated than were requested in malloc) is less than or equal
61  to the minimum size, except for requests >= mmap_threshold that
62  are serviced via mmap(), where the worst case wastage is about
63  32 bytes plus the remainder from a system page (the minimal
64  mmap unit); typically 4096 or 8192 bytes.
65 
66  Security: static-safe; optionally more or less
67  The "security" of malloc refers to the ability of malicious
68  code to accentuate the effects of errors (for example, freeing
69  space that is not currently malloc'ed or overwriting past the
70  ends of chunks) in code that calls malloc. This malloc
71  guarantees not to modify any memory locations below the base of
72  heap, i.e., static variables, even in the presence of usage
73  errors. The routines additionally detect most improper frees
74  and reallocs. All this holds as long as the static bookkeeping
75  for malloc itself is not corrupted by some other means. This
76  is only one aspect of security -- these checks do not, and
77  cannot, detect all possible programming errors.
78 
79  If FOOTERS is defined nonzero, then each allocated chunk
80  carries an additional check word to verify that it was malloced
81  from its space. These check words are the same within each
82  execution of a program using malloc, but differ across
83  executions, so externally crafted fake chunks cannot be
84  freed. This improves security by rejecting frees/reallocs that
85  could corrupt heap memory, in addition to the checks preventing
86  writes to statics that are always on. This may further improve
87  security at the expense of time and space overhead. (Note that
88  FOOTERS may also be worth using with MSPACES.)
89 
90  By default detected errors cause the program to abort (calling
91  "abort()"). You can override this to instead proceed past
92  errors by defining PROCEED_ON_ERROR. In this case, a bad free
93  has no effect, and a malloc that encounters a bad address
94  caused by user overwrites will ignore the bad address by
95  dropping pointers and indices to all known memory. This may
96  be appropriate for programs that should continue if at all
97  possible in the face of programming errors, although they may
98  run out of memory because dropped memory is never reclaimed.
99 
100  If you don't like either of these options, you can define
101  CORRUPTION_ERROR_ACTION and USAGE_ERROR_ACTION to do anything
102  else. And if if you are sure that your program using malloc has
103  no errors or vulnerabilities, you can define INSECURE to 1,
104  which might (or might not) provide a small performance improvement.
105 
106  It is also possible to limit the maximum total allocatable
107  space, using malloc_set_footprint_limit. This is not
108  designed as a security feature in itself (calls to set limits
109  are not screened or privileged), but may be useful as one
110  aspect of a secure implementation.
111 
112  Thread-safety: NOT thread-safe unless USE_LOCKS defined non-zero
113  When USE_LOCKS is defined, each public call to malloc, free,
114  etc is surrounded with a lock. By default, this uses a plain
115  pthread mutex, win32 critical section, or a spin-lock if if
116  available for the platform and not disabled by setting
117  USE_SPIN_LOCKS=0. However, if USE_RECURSIVE_LOCKS is defined,
118  recursive versions are used instead (which are not required for
119  base functionality but may be needed in layered extensions).
120  Using a global lock is not especially fast, and can be a major
121  bottleneck. It is designed only to provide minimal protection
122  in concurrent environments, and to provide a basis for
123  extensions. If you are using malloc in a concurrent program,
124  consider instead using nedmalloc
125  (http://www.nedprod.com/programs/portable/nedmalloc/) or
126  ptmalloc (See http://www.malloc.de), which are derived from
127  versions of this malloc.
128 
129  System requirements: Any combination of MORECORE and/or MMAP/MUNMAP
130  This malloc can use unix sbrk or any emulation (invoked using
131  the CALL_MORECORE macro) and/or mmap/munmap or any emulation
132  (invoked using CALL_MMAP/CALL_MUNMAP) to get and release system
133  memory. On most unix systems, it tends to work best if both
134  MORECORE and MMAP are enabled. On Win32, it uses emulations
135  based on VirtualAlloc. It also uses common C library functions
136  like memset.
137 
138  Compliance: I believe it is compliant with the Single Unix Specification
139  (See http://www.unix.org). Also SVID/XPG, ANSI C, and probably
140  others as well.
141 
142 * Overview of algorithms
143 
144  This is not the fastest, most space-conserving, most portable, or
145  most tunable malloc ever written. However it is among the fastest
146  while also being among the most space-conserving, portable and
147  tunable. Consistent balance across these factors results in a good
148  general-purpose allocator for malloc-intensive programs.
149 
150  In most ways, this malloc is a best-fit allocator. Generally, it
151  chooses the best-fitting existing chunk for a request, with ties
152  broken in approximately least-recently-used order. (This strategy
153  normally maintains low fragmentation.) However, for requests less
154  than 256bytes, it deviates from best-fit when there is not an
155  exactly fitting available chunk by preferring to use space adjacent
156  to that used for the previous small request, as well as by breaking
157  ties in approximately most-recently-used order. (These enhance
158  locality of series of small allocations.) And for very large requests
159  (>= 256Kb by default), it relies on system memory mapping
160  facilities, if supported. (This helps avoid carrying around and
161  possibly fragmenting memory used only for large chunks.)
162 
163  All operations (except malloc_stats and mallinfo) have execution
164  times that are bounded by a constant factor of the number of bits in
165  a size_t, not counting any clearing in calloc or copying in realloc,
166  or actions surrounding MORECORE and MMAP that have times
167  proportional to the number of non-contiguous regions returned by
168  system allocation routines, which is often just 1. In real-time
169  applications, you can optionally suppress segment traversals using
170  NO_SEGMENT_TRAVERSAL, which assures bounded execution even when
171  system allocators return non-contiguous spaces, at the typical
172  expense of carrying around more memory and increased fragmentation.
173 
174  The implementation is not very modular and seriously overuses
175  macros. Perhaps someday all C compilers will do as good a job
176  inlining modular code as can now be done by brute-force expansion,
177  but now, enough of them seem not to.
178 
179  Some compilers issue a lot of warnings about code that is
180  dead/unreachable only on some platforms, and also about intentional
181  uses of negation on unsigned types. All known cases of each can be
182  ignored.
183 
184  For a longer but out of date high-level description, see
185  http://gee.cs.oswego.edu/dl/html/malloc.html
186 
187 * MSPACES
188  If MSPACES is defined, then in addition to malloc, free, etc.,
189  this file also defines mspace_malloc, mspace_free, etc. These
190  are versions of malloc routines that take an "mspace" argument
191  obtained using create_mspace, to control all internal bookkeeping.
192  If ONLY_MSPACES is defined, only these versions are compiled.
193  So if you would like to use this allocator for only some allocations,
194  and your system malloc for others, you can compile with
195  ONLY_MSPACES and then do something like...
196  static mspace mymspace = create_mspace(0,0); // for example
197  #define mymalloc(bytes) mspace_malloc(mymspace, bytes)
198 
199  (Note: If you only need one instance of an mspace, you can instead
200  use "USE_DL_PREFIX" to relabel the global malloc.)
201 
202  You can similarly create thread-local allocators by storing
203  mspaces as thread-locals. For example:
204  static __thread mspace tlms = 0;
205  void* tlmalloc(size_t bytes) {
206  if (tlms == 0) tlms = create_mspace(0, 0);
207  return mspace_malloc(tlms, bytes);
208  }
209  void tlfree(void* mem) { mspace_free(tlms, mem); }
210 
211  Unless FOOTERS is defined, each mspace is completely independent.
212  You cannot allocate from one and free to another (although
213  conformance is only weakly checked, so usage errors are not always
214  caught). If FOOTERS is defined, then each chunk carries around a tag
215  indicating its originating mspace, and frees are directed to their
216  originating spaces. Normally, this requires use of locks.
217 
218  ------------------------- Compile-time options ---------------------------
219 
220 Be careful in setting #define values for numerical constants of type
221 size_t. On some systems, literal values are not automatically extended
222 to size_t precision unless they are explicitly casted. You can also
223 use the symbolic values MAX_SIZE_T, SIZE_T_ONE, etc below.
224 
225 WIN32 default: defined if _WIN32 defined
226  Defining WIN32 sets up defaults for MS environment and compilers.
227  Otherwise defaults are for unix. Beware that there seem to be some
228  cases where this malloc might not be a pure drop-in replacement for
229  Win32 malloc: Random-looking failures from Win32 GDI API's (eg;
230  SetDIBits()) may be due to bugs in some video driver implementations
231  when pixel buffers are malloc()ed, and the region spans more than
232  one VirtualAlloc()ed region. Because dlmalloc uses a small (64Kb)
233  default granularity, pixel buffers may straddle virtual allocation
234  regions more often than when using the Microsoft allocator. You can
235  avoid this by using VirtualAlloc() and VirtualFree() for all pixel
236  buffers rather than using malloc(). If this is not possible,
237  recompile this malloc with a larger DEFAULT_GRANULARITY. Note:
238  in cases where MSC and gcc (cygwin) are known to differ on WIN32,
239  conditions use _MSC_VER to distinguish them.
240 
241 DLMALLOC_EXPORT default: extern
242  Defines how public APIs are declared. If you want to export via a
243  Windows DLL, you might define this as
244  #define DLMALLOC_EXPORT extern __declspec(dllexport)
245  If you want a POSIX ELF shared object, you might use
246  #define DLMALLOC_EXPORT extern __attribute__((visibility("default")))
247 
248 MALLOC_ALIGNMENT default: (size_t)(2 * sizeof(void *))
249  Controls the minimum alignment for malloc'ed chunks. It must be a
250  power of two and at least 8, even on machines for which smaller
251  alignments would suffice. It may be defined as larger than this
252  though. Note however that code and data structures are optimized for
253  the case of 8-byte alignment.
254 
255 MSPACES default: 0 (false)
256  If true, compile in support for independent allocation spaces.
257  This is only supported if HAVE_MMAP is true.
258 
259 ONLY_MSPACES default: 0 (false)
260  If true, only compile in mspace versions, not regular versions.
261 
262 USE_LOCKS default: 0 (false)
263  Causes each call to each public routine to be surrounded with
264  pthread or WIN32 mutex lock/unlock. (If set true, this can be
265  overridden on a per-mspace basis for mspace versions.) If set to a
266  non-zero value other than 1, locks are used, but their
267  implementation is left out, so lock functions must be supplied manually,
268  as described below.
269 
270 USE_SPIN_LOCKS default: 1 iff USE_LOCKS and spin locks available
271  If true, uses custom spin locks for locking. This is currently
272  supported only gcc >= 4.1, older gccs on x86 platforms, and recent
273  MS compilers. Otherwise, posix locks or win32 critical sections are
274  used.
275 
276 USE_RECURSIVE_LOCKS default: not defined
277  If defined nonzero, uses recursive (aka reentrant) locks, otherwise
278  uses plain mutexes. This is not required for malloc proper, but may
279  be needed for layered allocators such as nedmalloc.
280 
281 LOCK_AT_FORK default: not defined
282  If defined nonzero, performs pthread_atfork upon initialization
283  to initialize child lock while holding parent lock. The implementation
284  assumes that pthread locks (not custom locks) are being used. In other
285  cases, you may need to customize the implementation.
286 
287 FOOTERS default: 0
288  If true, provide extra checking and dispatching by placing
289  information in the footers of allocated chunks. This adds
290  space and time overhead.
291 
292 INSECURE default: 0
293  If true, omit checks for usage errors and heap space overwrites.
294 
295 USE_DL_PREFIX default: NOT defined
296  Causes compiler to prefix all public routines with the string 'dl'.
297  This can be useful when you only want to use this malloc in one part
298  of a program, using your regular system malloc elsewhere.
299 
300 MALLOC_INSPECT_ALL default: NOT defined
301  If defined, compiles malloc_inspect_all and mspace_inspect_all, that
302  perform traversal of all heap space. Unless access to these
303  functions is otherwise restricted, you probably do not want to
304  include them in secure implementations.
305 
306 DLM_ABORT default: defined as abort()
307  Defines how to abort on failed checks. On most systems, a failed
308  check cannot die with an "assert" or even print an informative
309  message, because the underlying print routines in turn call malloc,
310  which will fail again. Generally, the best policy is to simply call
311  abort(). It's not very useful to do more than this because many
312  errors due to overwriting will show up as address faults (null, odd
313  addresses etc) rather than malloc-triggered checks, so will also
314  abort. Also, most compilers know that abort() does not return, so
315  can better optimize code conditionally calling it.
316 
317 PROCEED_ON_ERROR default: defined as 0 (false)
318  Controls whether detected bad addresses cause them to bypassed
319  rather than aborting. If set, detected bad arguments to free and
320  realloc are ignored. And all bookkeeping information is zeroed out
321  upon a detected overwrite of freed heap space, thus losing the
322  ability to ever return it from malloc again, but enabling the
323  application to proceed. If PROCEED_ON_ERROR is defined, the
324  static variable malloc_corruption_error_count is compiled in
325  and can be examined to see if errors have occurred. This option
326  generates slower code than the default abort policy.
327 
328 DEBUG default: NOT defined
329  The DEBUG setting is mainly intended for people trying to modify
330  this code or diagnose problems when porting to new platforms.
331  However, it may also be able to better isolate user errors than just
332  using runtime checks. The assertions in the check routines spell
333  out in more detail the assumptions and invariants underlying the
334  algorithms. The checking is fairly extensive, and will slow down
335  execution noticeably. Calling malloc_stats or mallinfo with DEBUG
336  set will attempt to check every non-mmapped allocated and free chunk
337  in the course of computing the summaries.
338 
339 DLM_ABORT_ON_ASSERT_FAILURE default: defined as 1 (true)
340  Debugging assertion failures can be nearly impossible if your
341  version of the assert macro causes malloc to be called, which will
342  lead to a cascade of further failures, blowing the runtime stack.
343  DLM_ABORT_ON_ASSERT_FAILURE cause assertions failures to call abort(),
344  which will usually make debugging easier.
345 
346 MALLOC_FAILURE_ACTION default: sets errno to ENOMEM, or no-op on win32
347  The action to take before "return 0" when malloc fails to be able to
348  return memory because there is none available.
349 
350 HAVE_MORECORE default: 1 (true) unless win32 or ONLY_MSPACES
351  True if this system supports sbrk or an emulation of it.
352 
353 MORECORE default: sbrk
354  The name of the sbrk-style system routine to call to obtain more
355  memory. See below for guidance on writing custom MORECORE
356  functions. The type of the argument to sbrk/MORECORE varies across
357  systems. It cannot be size_t, because it supports negative
358  arguments, so it is normally the signed type of the same width as
359  size_t (sometimes declared as "intptr_t"). It doesn't much matter
360  though. Internally, we only call it with arguments less than half
361  the max value of a size_t, which should work across all reasonable
362  possibilities, although sometimes generating compiler warnings.
363 
364 MORECORE_CONTIGUOUS default: 1 (true) if HAVE_MORECORE
365  If true, take advantage of fact that consecutive calls to MORECORE
366  with positive arguments always return contiguous increasing
367  addresses. This is true of unix sbrk. It does not hurt too much to
368  set it true anyway, since malloc copes with non-contiguities.
369  Setting it false when definitely non-contiguous saves time
370  and possibly wasted space it would take to discover this though.
371 
372 MORECORE_CANNOT_TRIM default: NOT defined
373  True if MORECORE cannot release space back to the system when given
374  negative arguments. This is generally necessary only if you are
375  using a hand-crafted MORECORE function that cannot handle negative
376  arguments.
377 
378 NO_SEGMENT_TRAVERSAL default: 0
379  If non-zero, suppresses traversals of memory segments
380  returned by either MORECORE or CALL_MMAP. This disables
381  merging of segments that are contiguous, and selectively
382  releasing them to the OS if unused, but bounds execution times.
383 
384 HAVE_MMAP default: 1 (true)
385  True if this system supports mmap or an emulation of it. If so, and
386  HAVE_MORECORE is not true, MMAP is used for all system
387  allocation. If set and HAVE_MORECORE is true as well, MMAP is
388  primarily used to directly allocate very large blocks. It is also
389  used as a backup strategy in cases where MORECORE fails to provide
390  space from system. Note: A single call to MUNMAP is assumed to be
391  able to unmap memory that may have be allocated using multiple calls
392  to MMAP, so long as they are adjacent.
393 
394 HAVE_MREMAP default: 1 on linux, else 0
395  If true realloc() uses mremap() to re-allocate large blocks and
396  extend or shrink allocation spaces.
397 
398 MMAP_CLEARS default: 1 except on WINCE.
399  True if mmap clears memory so calloc doesn't need to. This is true
400  for standard unix mmap using /dev/zero and on WIN32 except for WINCE.
401 
402 USE_BUILTIN_FFS default: 0 (i.e., not used)
403  Causes malloc to use the builtin ffs() function to compute indices.
404  Some compilers may recognize and intrinsify ffs to be faster than the
405  supplied C version. Also, the case of x86 using gcc is special-cased
406  to an asm instruction, so is already as fast as it can be, and so
407  this setting has no effect. Similarly for Win32 under recent MS compilers.
408  (On most x86s, the asm version is only slightly faster than the C version.)
409 
410 malloc_getpagesize default: derive from system includes, or 4096.
411  The system page size. To the extent possible, this malloc manages
412  memory from the system in page-size units. This may be (and
413  usually is) a function rather than a constant. This is ignored
414  if WIN32, where page size is determined using getSystemInfo during
415  initialization.
416 
417 USE_DEV_RANDOM default: 0 (i.e., not used)
418  Causes malloc to use /dev/random to initialize secure magic seed for
419  stamping footers. Otherwise, the current time is used.
420 
421 NO_MALLINFO default: 0
422  If defined, don't compile "mallinfo". This can be a simple way
423  of dealing with mismatches between system declarations and
424  those in this file.
425 
426 MALLINFO_FIELD_TYPE default: size_t
427  The type of the fields in the mallinfo struct. This was originally
428  defined as "int" in SVID etc, but is more usefully defined as
429  size_t. The value is used only if HAVE_USR_INCLUDE_MALLOC_H is not set
430 
431 NO_MALLOC_STATS default: 0
432  If defined, don't compile "malloc_stats". This avoids calls to
433  fprintf and bringing in stdio dependencies you might not want.
434 
435 REALLOC_ZERO_BYTES_FREES default: not defined
436  This should be set if a call to realloc with zero bytes should
437  be the same as a call to free. Some people think it should. Otherwise,
438  since this malloc returns a unique pointer for malloc(0), so does
439  realloc(p, 0).
440 
441 LACKS_UNISTD_H, LACKS_FCNTL_H, LACKS_SYS_PARAM_H, LACKS_SYS_MMAN_H
442 LACKS_STRINGS_H, LACKS_STRING_H, LACKS_SYS_TYPES_H, LACKS_ERRNO_H
443 LACKS_STDLIB_H LACKS_SCHED_H LACKS_TIME_H default: NOT defined unless on WIN32
444  Define these if your system does not have these header files.
445  You might need to manually insert some of the declarations they provide.
446 
447 DEFAULT_GRANULARITY default: page size if MORECORE_CONTIGUOUS,
448  system_info.dwAllocationGranularity in WIN32,
449  otherwise 64K.
450  Also settable using mallopt(M_GRANULARITY, x)
451  The unit for allocating and deallocating memory from the system. On
452  most systems with contiguous MORECORE, there is no reason to
453  make this more than a page. However, systems with MMAP tend to
454  either require or encourage larger granularities. You can increase
455  this value to prevent system allocation functions to be called so
456  often, especially if they are slow. The value must be at least one
457  page and must be a power of two. Setting to 0 causes initialization
458  to either page size or win32 region size. (Note: In previous
459  versions of malloc, the equivalent of this option was called
460  "TOP_PAD")
461 
462 DEFAULT_TRIM_THRESHOLD default: 2MB
463  Also settable using mallopt(M_TRIM_THRESHOLD, x)
464  The maximum amount of unused top-most memory to keep before
465  releasing via malloc_trim in free(). Automatic trimming is mainly
466  useful in long-lived programs using contiguous MORECORE. Because
467  trimming via sbrk can be slow on some systems, and can sometimes be
468  wasteful (in cases where programs immediately afterward allocate
469  more large chunks) the value should be high enough so that your
470  overall system performance would improve by releasing this much
471  memory. As a rough guide, you might set to a value close to the
472  average size of a process (program) running on your system.
473  Releasing this much memory would allow such a process to run in
474  memory. Generally, it is worth tuning trim thresholds when a
475  program undergoes phases where several large chunks are allocated
476  and released in ways that can reuse each other's storage, perhaps
477  mixed with phases where there are no such chunks at all. The trim
478  value must be greater than page size to have any useful effect. To
479  disable trimming completely, you can set to MAX_SIZE_T. Note that the trick
480  some people use of mallocing a huge space and then freeing it at
481  program startup, in an attempt to reserve system memory, doesn't
482  have the intended effect under automatic trimming, since that memory
483  will immediately be returned to the system.
484 
485 DEFAULT_MMAP_THRESHOLD default: 256K
486  Also settable using mallopt(M_MMAP_THRESHOLD, x)
487  The request size threshold for using MMAP to directly service a
488  request. Requests of at least this size that cannot be allocated
489  using already-existing space will be serviced via mmap. (If enough
490  normal freed space already exists it is used instead.) Using mmap
491  segregates relatively large chunks of memory so that they can be
492  individually obtained and released from the host system. A request
493  serviced through mmap is never reused by any other request (at least
494  not directly; the system may just so happen to remap successive
495  requests to the same locations). Segregating space in this way has
496  the benefits that: Mmapped space can always be individually released
497  back to the system, which helps keep the system level memory demands
498  of a long-lived program low. Also, mapped memory doesn't become
499  `locked' between other chunks, as can happen with normally allocated
500  chunks, which means that even trimming via malloc_trim would not
501  release them. However, it has the disadvantage that the space
502  cannot be reclaimed, consolidated, and then used to service later
503  requests, as happens with normal chunks. The advantages of mmap
504  nearly always outweigh disadvantages for "large" chunks, but the
505  value of "large" may vary across systems. The default is an
506  empirically derived value that works well in most systems. You can
507  disable mmap by setting to MAX_SIZE_T.
508 
509 MAX_RELEASE_CHECK_RATE default: 4095 unless not HAVE_MMAP
510  The number of consolidated frees between checks to release
511  unused segments when freeing. When using non-contiguous segments,
512  especially with multiple mspaces, checking only for topmost space
513  doesn't always suffice to trigger trimming. To compensate for this,
514  free() will, with a period of MAX_RELEASE_CHECK_RATE (or the
515  current number of segments, if greater) try to release unused
516  segments to the OS when freeing chunks that result in
517  consolidation. The best value for this parameter is a compromise
518  between slowing down frees with relatively costly checks that
519  rarely trigger versus holding on to unused memory. To effectively
520  disable, set to MAX_SIZE_T. This may lead to a very slight speed
521  improvement at the expense of carrying around more memory.
522 */
523 
524 #include <vppinfra/clib.h>
525 #include <vppinfra/cache.h>
526 
527 /* --- begin vpp customizations --- */
528 
529 #if CLIB_DEBUG > 0
530 #define FOOTERS 1 /* extra debugging */
531 #define DLM_MAGIC_CONSTANT 0xdeaddabe
532 #endif
533 #define USE_LOCKS 1
534 #define DLM_ABORT {extern void os_panic(void); os_panic(); abort();}
535 #define ONLY_MSPACES 1
536 
537 /* --- end vpp customizations --- */
538 
539 /* Version identifier to allow people to support multiple versions */
540 #ifndef DLMALLOC_VERSION
541 #define DLMALLOC_VERSION 20806
542 #endif /* DLMALLOC_VERSION */
543 
544 #ifndef DLMALLOC_EXPORT
545 #define DLMALLOC_EXPORT extern
546 #endif
547 
548 #ifndef WIN32
549 #ifdef _WIN32
550 #define WIN32 1
551 #endif /* _WIN32 */
552 #ifdef _WIN32_WCE
553 #define LACKS_FCNTL_H
554 #define WIN32 1
555 #endif /* _WIN32_WCE */
556 #endif /* WIN32 */
557 #ifdef WIN32
558 #define WIN32_LEAN_AND_MEAN
559 #include <windows.h>
560 #include <tchar.h>
561 #define HAVE_MMAP 1
562 #define HAVE_MORECORE 0
563 #define LACKS_UNISTD_H
564 #define LACKS_SYS_PARAM_H
565 #define LACKS_SYS_MMAN_H
566 #define LACKS_STRING_H
567 #define LACKS_STRINGS_H
568 #define LACKS_SYS_TYPES_H
569 #define LACKS_ERRNO_H
570 #define LACKS_SCHED_H
571 #ifndef MALLOC_FAILURE_ACTION
572 #define MALLOC_FAILURE_ACTION
573 #endif /* MALLOC_FAILURE_ACTION */
574 #ifndef MMAP_CLEARS
575 #ifdef _WIN32_WCE /* WINCE reportedly does not clear */
576 #define MMAP_CLEARS 0
577 #else
578 #define MMAP_CLEARS 1
579 #endif /* _WIN32_WCE */
580 #endif /*MMAP_CLEARS */
581 #endif /* WIN32 */
582 
583 #if defined(DARWIN) || defined(_DARWIN)
584 /* Mac OSX docs advise not to use sbrk; it seems better to use mmap */
585 #ifndef HAVE_MORECORE
586 #define HAVE_MORECORE 0
587 #define HAVE_MMAP 1
588 /* OSX allocators provide 16 byte alignment */
589 #ifndef MALLOC_ALIGNMENT
590 #define MALLOC_ALIGNMENT ((size_t)16U)
591 #endif
592 #endif /* HAVE_MORECORE */
593 #endif /* DARWIN */
594 
595 #ifndef LACKS_SYS_TYPES_H
596 #include <sys/types.h> /* For size_t */
597 #endif /* LACKS_SYS_TYPES_H */
598 
599 /* The maximum possible size_t value has all bits set */
600 #define MAX_SIZE_T (~(size_t)0)
601 
602 #ifndef USE_LOCKS /* ensure true if spin or recursive locks set */
603 #define USE_LOCKS ((defined(USE_SPIN_LOCKS) && USE_SPIN_LOCKS != 0) || \
604  (defined(USE_RECURSIVE_LOCKS) && USE_RECURSIVE_LOCKS != 0))
605 #endif /* USE_LOCKS */
606 
607 #if USE_LOCKS /* Spin locks for gcc >= 4.1, older gcc on x86, MSC >= 1310 */
608 #if ((defined(__GNUC__) && \
609  ((__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 1)) || \
610  defined(__i386__) || defined(__x86_64__))) || \
611  (defined(_MSC_VER) && _MSC_VER>=1310))
612 #ifndef USE_SPIN_LOCKS
613 #define USE_SPIN_LOCKS 1
614 #endif /* USE_SPIN_LOCKS */
615 #elif USE_SPIN_LOCKS
616 #error "USE_SPIN_LOCKS defined without implementation"
617 #endif /* ... locks available... */
618 #elif !defined(USE_SPIN_LOCKS)
619 #define USE_SPIN_LOCKS 0
620 #endif /* USE_LOCKS */
621 
622 #ifndef ONLY_MSPACES
623 #define ONLY_MSPACES 1
624 #endif /* ONLY_MSPACES */
625 #ifndef MSPACES
626 #if ONLY_MSPACES
627 #define MSPACES 1
628 #else /* ONLY_MSPACES */
629 #define MSPACES 0
630 #endif /* ONLY_MSPACES */
631 #endif /* MSPACES */
632 #ifndef MALLOC_ALIGNMENT
633 #define MALLOC_ALIGNMENT ((size_t)(2 * sizeof(void *)))
634 #endif /* MALLOC_ALIGNMENT */
635 #ifndef FOOTERS
636 #define FOOTERS 0
637 #endif /* FOOTERS */
638 #ifndef DLM_ABORT
639 #define DLM_ABORT abort()
640 #endif /* DLM_ABORT */
641 #ifndef DLM_ABORT_ON_ASSERT_FAILURE
642 #define DLM_ABORT_ON_ASSERT_FAILURE 1
643 #endif /* DLM_ABORT_ON_ASSERT_FAILURE */
644 #ifndef PROCEED_ON_ERROR
645 #define PROCEED_ON_ERROR 0
646 #endif /* PROCEED_ON_ERROR */
647 
648 #ifndef INSECURE
649 #define INSECURE 0
650 #endif /* INSECURE */
651 #ifndef MALLOC_INSPECT_ALL
652 #define MALLOC_INSPECT_ALL 0
653 #endif /* MALLOC_INSPECT_ALL */
654 #ifndef HAVE_MMAP
655 #define HAVE_MMAP 1
656 #endif /* HAVE_MMAP */
657 #ifndef MMAP_CLEARS
658 #define MMAP_CLEARS 1
659 #endif /* MMAP_CLEARS */
660 #ifndef HAVE_MREMAP
661 #ifdef linux
662 #define HAVE_MREMAP 1
663 #define _GNU_SOURCE /* Turns on mremap() definition */
664 #else /* linux */
665 #define HAVE_MREMAP 0
666 #endif /* linux */
667 #endif /* HAVE_MREMAP */
668 #ifndef MALLOC_FAILURE_ACTION
669 #define MALLOC_FAILURE_ACTION errno = ENOMEM;
670 #endif /* MALLOC_FAILURE_ACTION */
671 #ifndef HAVE_MORECORE
672 #if ONLY_MSPACES
673 #define HAVE_MORECORE 0
674 #else /* ONLY_MSPACES */
675 #define HAVE_MORECORE 1
676 #endif /* ONLY_MSPACES */
677 #endif /* HAVE_MORECORE */
678 #if !HAVE_MORECORE
679 #define MORECORE_CONTIGUOUS 0
680 #else /* !HAVE_MORECORE */
681 #define MORECORE_DEFAULT sbrk
682 #ifndef MORECORE_CONTIGUOUS
683 #define MORECORE_CONTIGUOUS 1
684 #endif /* MORECORE_CONTIGUOUS */
685 #endif /* HAVE_MORECORE */
686 #ifndef DEFAULT_GRANULARITY
687 #if (MORECORE_CONTIGUOUS || defined(WIN32))
688 #define DEFAULT_GRANULARITY (0) /* 0 means to compute in init_mparams */
689 #else /* MORECORE_CONTIGUOUS */
690 #define DEFAULT_GRANULARITY ((size_t)64U * (size_t)1024U)
691 #endif /* MORECORE_CONTIGUOUS */
692 #endif /* DEFAULT_GRANULARITY */
693 #ifndef DEFAULT_TRIM_THRESHOLD
694 #ifndef MORECORE_CANNOT_TRIM
695 #define DEFAULT_TRIM_THRESHOLD ((size_t)2U * (size_t)1024U * (size_t)1024U)
696 #else /* MORECORE_CANNOT_TRIM */
697 #define DEFAULT_TRIM_THRESHOLD MAX_SIZE_T
698 #endif /* MORECORE_CANNOT_TRIM */
699 #endif /* DEFAULT_TRIM_THRESHOLD */
700 #ifndef DEFAULT_MMAP_THRESHOLD
701 #if HAVE_MMAP
702 /*
703  * The default value in the dlmalloc was set as follows:
704  *
705  * #define DEFAULT_MMAP_THRESHOLD ((size_t)256U * (size_t)1024U)
706  *
707  * Above this threshold the sys_alloc() calls mmap_alloc() to directly mmap the memory.
708  * However, the interaction of this path with the rest of the vpp infra results
709  * in vpp infra considering directly mmap-allocated pieces to not
710  * be part of the heap, with predictable consequence.
711  *
712  * A simple unit-test to show that behavior is to make a small private
713  * heap and repeatedly perform vec_add1() within that heap.
714  *
715  * The issue is because there is no tracking which mmap-allocated chunk
716  * belongs to which heap.
717  *
718  * The temporary approach is to dial up the threshold so that the problematic
719  * code path never gets called. The full fix needs to introduce
720  * introduce the vector of mmap-allocated chunks to mspace, and in general
721  * do some more thorough testing.
722  */
723 #define DEFAULT_MMAP_THRESHOLD ((size_t)~0ULL)
724 #else /* HAVE_MMAP */
725 #define DEFAULT_MMAP_THRESHOLD MAX_SIZE_T
726 #endif /* HAVE_MMAP */
727 #endif /* DEFAULT_MMAP_THRESHOLD */
728 #ifndef MAX_RELEASE_CHECK_RATE
729 #if HAVE_MMAP
730 #define MAX_RELEASE_CHECK_RATE 4095
731 #else
732 #define MAX_RELEASE_CHECK_RATE MAX_SIZE_T
733 #endif /* HAVE_MMAP */
734 #endif /* MAX_RELEASE_CHECK_RATE */
735 #ifndef USE_BUILTIN_FFS
736 #define USE_BUILTIN_FFS 0
737 #endif /* USE_BUILTIN_FFS */
738 #ifndef USE_DEV_RANDOM
739 #define USE_DEV_RANDOM 0
740 #endif /* USE_DEV_RANDOM */
741 #ifndef NO_MALLINFO
742 #define NO_MALLINFO 0
743 #endif /* NO_MALLINFO */
744 #ifndef MALLINFO_FIELD_TYPE
745 #define MALLINFO_FIELD_TYPE size_t
746 #endif /* MALLINFO_FIELD_TYPE */
747 #ifndef NO_MALLOC_STATS
748 #define NO_MALLOC_STATS 0
749 #endif /* NO_MALLOC_STATS */
750 #ifndef NO_SEGMENT_TRAVERSAL
751 #define NO_SEGMENT_TRAVERSAL 0
752 #endif /* NO_SEGMENT_TRAVERSAL */
753 
754 /*
755  mallopt tuning options. SVID/XPG defines four standard parameter
756  numbers for mallopt, normally defined in malloc.h. None of these
757  are used in this malloc, so setting them has no effect. But this
758  malloc does support the following options.
759 */
760 
761 #define M_TRIM_THRESHOLD (-1)
762 #define M_GRANULARITY (-2)
763 #define M_MMAP_THRESHOLD (-3)
764 
765 /* ------------------------ Mallinfo declarations ------------------------ */
766 
767 #if !NO_MALLINFO
768 /*
769  This version of malloc supports the standard SVID/XPG mallinfo
770  routine that returns a struct containing usage properties and
771  statistics. It should work on any system that has a
772  /usr/include/malloc.h defining struct mallinfo. The main
773  declaration needed is the mallinfo struct that is returned (by-copy)
774  by mallinfo(). The malloinfo struct contains a bunch of fields that
775  are not even meaningful in this version of malloc. These fields are
776  are instead filled by mallinfo() with other numbers that might be of
777  interest.
778 
779  HAVE_USR_INCLUDE_MALLOC_H should be set if you have a
780  /usr/include/malloc.h file that includes a declaration of struct
781  mallinfo. If so, it is included; else a compliant version is
782  declared below. These must be precisely the same for mallinfo() to
783  work. The original SVID version of this struct, defined on most
784  systems with mallinfo, declares all fields as ints. But some others
785  define as unsigned long. If your system defines the fields using a
786  type of different width than listed here, you MUST #include your
787  system version and #define HAVE_USR_INCLUDE_MALLOC_H.
788 */
789 
790 /* #define HAVE_USR_INCLUDE_MALLOC_H */
791 
792 #if 0 // def HAVE_USR_INCLUDE_MALLOC_H
793 #include "/usr/include/malloc.h"
794 #else /* HAVE_USR_INCLUDE_MALLOC_H */
795 #ifndef STRUCT_MALLINFO_DECLARED
796 /* HP-UX (and others?) redefines mallinfo unless _STRUCT_MALLINFO is defined */
797 #define _STRUCT_MALLINFO
798 #define STRUCT_MALLINFO_DECLARED 1
799 struct dlmallinfo {
800  MALLINFO_FIELD_TYPE arena; /* non-mmapped space allocated from system */
801  MALLINFO_FIELD_TYPE ordblks; /* number of free chunks */
802  MALLINFO_FIELD_TYPE smblks; /* always 0 */
803  MALLINFO_FIELD_TYPE hblks; /* always 0 */
804  MALLINFO_FIELD_TYPE hblkhd; /* space in mmapped regions */
805  MALLINFO_FIELD_TYPE usmblks; /* maximum total allocated space */
806  MALLINFO_FIELD_TYPE fsmblks; /* always 0 */
807  MALLINFO_FIELD_TYPE uordblks; /* total allocated space */
808  MALLINFO_FIELD_TYPE fordblks; /* total free space */
809  MALLINFO_FIELD_TYPE keepcost; /* releasable (via malloc_trim) space */
810 };
811 #endif /* STRUCT_MALLINFO_DECLARED */
812 #endif /* HAVE_USR_INCLUDE_MALLOC_H */
813 #endif /* NO_MALLINFO */
814 
815 /*
816  Try to persuade compilers to inline. The most critical functions for
817  inlining are defined as macros, so these aren't used for them.
818 */
819 
820 #ifndef FORCEINLINE
821  #if defined(__GNUC__)
822 #define FORCEINLINE __inline __attribute__ ((always_inline))
823  #elif defined(_MSC_VER)
824  #define FORCEINLINE __forceinline
825  #endif
826 #endif
827 #ifndef NOINLINE
828  #if defined(__GNUC__)
829  #define NOINLINE __attribute__ ((noinline))
830  #elif defined(_MSC_VER)
831  #define NOINLINE __declspec(noinline)
832  #else
833  #define NOINLINE
834  #endif
835 #endif
836 
837 #ifdef __cplusplus
838 extern "C" {
839 #ifndef FORCEINLINE
840  #define FORCEINLINE inline
841 #endif
842 #endif /* __cplusplus */
843 #ifndef FORCEINLINE
844  #define FORCEINLINE
845 #endif
846 
847 #if !ONLY_MSPACES
848 
849 /* ------------------- Declarations of public routines ------------------- */
850 
851 #ifndef USE_DL_PREFIX
852 #define dlcalloc calloc
853 #define dlfree free
854 #define dlmalloc malloc
855 #define dlmemalign memalign
856 #define dlposix_memalign posix_memalign
857 #define dlrealloc realloc
858 #define dlrealloc_in_place realloc_in_place
859 #define dlvalloc valloc
860 #define dlpvalloc pvalloc
861 // #define dlmallinfo mallinfo
862 #define dlmallopt mallopt
863 #define dlmalloc_trim malloc_trim
864 #define dlmalloc_stats malloc_stats
865 #define dlmalloc_usable_size malloc_usable_size
866 #define dlmalloc_footprint malloc_footprint
867 #define dlmalloc_max_footprint malloc_max_footprint
868 #define dlmalloc_footprint_limit malloc_footprint_limit
869 #define dlmalloc_set_footprint_limit malloc_set_footprint_limit
870 #define dlmalloc_inspect_all malloc_inspect_all
871 #define dlindependent_calloc independent_calloc
872 #define dlindependent_comalloc independent_comalloc
873 #define dlbulk_free bulk_free
874 #endif /* USE_DL_PREFIX */
875 
876 /*
877  malloc(size_t n)
878  Returns a pointer to a newly allocated chunk of at least n bytes, or
879  null if no space is available, in which case errno is set to ENOMEM
880  on ANSI C systems.
881 
882  If n is zero, malloc returns a minimum-sized chunk. (The minimum
883  size is 16 bytes on most 32bit systems, and 32 bytes on 64bit
884  systems.) Note that size_t is an unsigned type, so calls with
885  arguments that would be negative if signed are interpreted as
886  requests for huge amounts of space, which will often fail. The
887  maximum supported value of n differs across systems, but is in all
888  cases less than the maximum representable value of a size_t.
889 */
890 DLMALLOC_EXPORT void* dlmalloc(size_t);
891 
892 /*
893  free(void* p)
894  Releases the chunk of memory pointed to by p, that had been previously
895  allocated using malloc or a related routine such as realloc.
896  It has no effect if p is null. If p was not malloced or already
897  freed, free(p) will by default cause the current program to abort.
898 */
899 DLMALLOC_EXPORT void dlfree(void*);
900 
901 /*
902  calloc(size_t n_elements, size_t element_size);
903  Returns a pointer to n_elements * element_size bytes, with all locations
904  set to zero.
905 */
906 DLMALLOC_EXPORT void* dlcalloc(size_t, size_t);
907 
908 /*
909  realloc(void* p, size_t n)
910  Returns a pointer to a chunk of size n that contains the same data
911  as does chunk p up to the minimum of (n, p's size) bytes, or null
912  if no space is available.
913 
914  The returned pointer may or may not be the same as p. The algorithm
915  prefers extending p in most cases when possible, otherwise it
916  employs the equivalent of a malloc-copy-free sequence.
917 
918  If p is null, realloc is equivalent to malloc.
919 
920  If space is not available, realloc returns null, errno is set (if on
921  ANSI) and p is NOT freed.
922 
923  if n is for fewer bytes than already held by p, the newly unused
924  space is lopped off and freed if possible. realloc with a size
925  argument of zero (re)allocates a minimum-sized chunk.
926 
927  The old unix realloc convention of allowing the last-free'd chunk
928  to be used as an argument to realloc is not supported.
929 */
930 DLMALLOC_EXPORT void* dlrealloc(void*, size_t);
931 
932 /*
933  realloc_in_place(void* p, size_t n)
934  Resizes the space allocated for p to size n, only if this can be
935  done without moving p (i.e., only if there is adjacent space
936  available if n is greater than p's current allocated size, or n is
937  less than or equal to p's size). This may be used instead of plain
938  realloc if an alternative allocation strategy is needed upon failure
939  to expand space; for example, reallocation of a buffer that must be
940  memory-aligned or cleared. You can use realloc_in_place to trigger
941  these alternatives only when needed.
942 
943  Returns p if successful; otherwise null.
944 */
945 DLMALLOC_EXPORT void* dlrealloc_in_place(void*, size_t);
946 
947 /*
948  memalign(size_t alignment, size_t n);
949  Returns a pointer to a newly allocated chunk of n bytes, aligned
950  in accord with the alignment argument.
951 
952  The alignment argument should be a power of two. If the argument is
953  not a power of two, the nearest greater power is used.
954  8-byte alignment is guaranteed by normal malloc calls, so don't
955  bother calling memalign with an argument of 8 or less.
956 
957  Overreliance on memalign is a sure way to fragment space.
958 */
959 DLMALLOC_EXPORT void* dlmemalign(size_t, size_t);
960 
961 /*
962  int posix_memalign(void** pp, size_t alignment, size_t n);
963  Allocates a chunk of n bytes, aligned in accord with the alignment
964  argument. Differs from memalign only in that it (1) assigns the
965  allocated memory to *pp rather than returning it, (2) fails and
966  returns EINVAL if the alignment is not a power of two (3) fails and
967  returns ENOMEM if memory cannot be allocated.
968 */
969 DLMALLOC_EXPORT int dlposix_memalign(void**, size_t, size_t);
970 
971 /*
972  valloc(size_t n);
973  Equivalent to memalign(pagesize, n), where pagesize is the page
974  size of the system. If the pagesize is unknown, 4096 is used.
975 */
976 DLMALLOC_EXPORT void* dlvalloc(size_t);
977 
978 /*
979  mallopt(int parameter_number, int parameter_value)
980  Sets tunable parameters The format is to provide a
981  (parameter-number, parameter-value) pair. mallopt then sets the
982  corresponding parameter to the argument value if it can (i.e., so
983  long as the value is meaningful), and returns 1 if successful else
984  0. To workaround the fact that mallopt is specified to use int,
985  not size_t parameters, the value -1 is specially treated as the
986  maximum unsigned size_t value.
987 
988  SVID/XPG/ANSI defines four standard param numbers for mallopt,
989  normally defined in malloc.h. None of these are use in this malloc,
990  so setting them has no effect. But this malloc also supports other
991  options in mallopt. See below for details. Briefly, supported
992  parameters are as follows (listed defaults are for "typical"
993  configurations).
994 
995  Symbol param # default allowed param values
996  M_TRIM_THRESHOLD -1 2*1024*1024 any (-1 disables)
997  M_GRANULARITY -2 page size any power of 2 >= page size
998  M_MMAP_THRESHOLD -3 256*1024 any (or 0 if no MMAP support)
999 */
1000 DLMALLOC_EXPORT int dlmallopt(int, int);
1001 
1002 /*
1003  malloc_footprint();
1004  Returns the number of bytes obtained from the system. The total
1005  number of bytes allocated by malloc, realloc etc., is less than this
1006  value. Unlike mallinfo, this function returns only a precomputed
1007  result, so can be called frequently to monitor memory consumption.
1008  Even if locks are otherwise defined, this function does not use them,
1009  so results might not be up to date.
1010 */
1011 DLMALLOC_EXPORT size_t dlmalloc_footprint(void);
1012 
1013 /*
1014  malloc_max_footprint();
1015  Returns the maximum number of bytes obtained from the system. This
1016  value will be greater than current footprint if deallocated space
1017  has been reclaimed by the system. The peak number of bytes allocated
1018  by malloc, realloc etc., is less than this value. Unlike mallinfo,
1019  this function returns only a precomputed result, so can be called
1020  frequently to monitor memory consumption. Even if locks are
1021  otherwise defined, this function does not use them, so results might
1022  not be up to date.
1023 */
1025 
1026 /*
1027  malloc_footprint_limit();
1028  Returns the number of bytes that the heap is allowed to obtain from
1029  the system, returning the last value returned by
1030  malloc_set_footprint_limit, or the maximum size_t value if
1031  never set. The returned value reflects a permission. There is no
1032  guarantee that this number of bytes can actually be obtained from
1033  the system.
1034 */
1036 
1037 /*
1038  malloc_set_footprint_limit();
1039  Sets the maximum number of bytes to obtain from the system, causing
1040  failure returns from malloc and related functions upon attempts to
1041  exceed this value. The argument value may be subject to page
1042  rounding to an enforceable limit; this actual value is returned.
1043  Using an argument of the maximum possible size_t effectively
1044  disables checks. If the argument is less than or equal to the
1045  current malloc_footprint, then all future allocations that require
1046  additional system memory will fail. However, invocation cannot
1047  retroactively deallocate existing used memory.
1048 */
1049 DLMALLOC_EXPORT size_t dlmalloc_set_footprint_limit(size_t bytes);
1050 
1051 #if MALLOC_INSPECT_ALL
1052 /*
1053  malloc_inspect_all(void(*handler)(void *start,
1054  void *end,
1055  size_t used_bytes,
1056  void* callback_arg),
1057  void* arg);
1058  Traverses the heap and calls the given handler for each managed
1059  region, skipping all bytes that are (or may be) used for bookkeeping
1060  purposes. Traversal does not include include chunks that have been
1061  directly memory mapped. Each reported region begins at the start
1062  address, and continues up to but not including the end address. The
1063  first used_bytes of the region contain allocated data. If
1064  used_bytes is zero, the region is unallocated. The handler is
1065  invoked with the given callback argument. If locks are defined, they
1066  are held during the entire traversal. It is a bad idea to invoke
1067  other malloc functions from within the handler.
1068 
1069  For example, to count the number of in-use chunks with size greater
1070  than 1000, you could write:
1071  static int count = 0;
1072  void count_chunks(void* start, void* end, size_t used, void* arg) {
1073  if (used >= 1000) ++count;
1074  }
1075  then:
1076  malloc_inspect_all(count_chunks, NULL);
1077 
1078  malloc_inspect_all is compiled only if MALLOC_INSPECT_ALL is defined.
1079 */
1080 DLMALLOC_EXPORT void dlmalloc_inspect_all(void(*handler)(void*, void *, size_t, void*),
1081  void* arg);
1082 
1083 #endif /* MALLOC_INSPECT_ALL */
1084 
1085 #if !NO_MALLINFO
1086 /*
1087  mallinfo()
1088  Returns (by copy) a struct containing various summary statistics:
1089 
1090  arena: current total non-mmapped bytes allocated from system
1091  ordblks: the number of free chunks
1092  smblks: always zero.
1093  hblks: current number of mmapped regions
1094  hblkhd: total bytes held in mmapped regions
1095  usmblks: the maximum total allocated space. This will be greater
1096  than current total if trimming has occurred.
1097  fsmblks: always zero
1098  uordblks: current total allocated space (normal or mmapped)
1099  fordblks: total free space
1100  keepcost: the maximum number of bytes that could ideally be released
1101  back to system via malloc_trim. ("ideally" means that
1102  it ignores page restrictions etc.)
1103 
1104  Because these fields are ints, but internal bookkeeping may
1105  be kept as longs, the reported values may wrap around zero and
1106  thus be inaccurate.
1107 */
1109 #endif /* NO_MALLINFO */
1110 
1111 /*
1112  independent_calloc(size_t n_elements, size_t element_size, void* chunks[]);
1113 
1114  independent_calloc is similar to calloc, but instead of returning a
1115  single cleared space, it returns an array of pointers to n_elements
1116  independent elements that can hold contents of size elem_size, each
1117  of which starts out cleared, and can be independently freed,
1118  realloc'ed etc. The elements are guaranteed to be adjacently
1119  allocated (this is not guaranteed to occur with multiple callocs or
1120  mallocs), which may also improve cache locality in some
1121  applications.
1122 
1123  The "chunks" argument is optional (i.e., may be null, which is
1124  probably the most typical usage). If it is null, the returned array
1125  is itself dynamically allocated and should also be freed when it is
1126  no longer needed. Otherwise, the chunks array must be of at least
1127  n_elements in length. It is filled in with the pointers to the
1128  chunks.
1129 
1130  In either case, independent_calloc returns this pointer array, or
1131  null if the allocation failed. If n_elements is zero and "chunks"
1132  is null, it returns a chunk representing an array with zero elements
1133  (which should be freed if not wanted).
1134 
1135  Each element must be freed when it is no longer needed. This can be
1136  done all at once using bulk_free.
1137 
1138  independent_calloc simplifies and speeds up implementations of many
1139  kinds of pools. It may also be useful when constructing large data
1140  structures that initially have a fixed number of fixed-sized nodes,
1141  but the number is not known at compile time, and some of the nodes
1142  may later need to be freed. For example:
1143 
1144  struct Node { int item; struct Node* next; };
1145 
1146  struct Node* build_list() {
1147  struct Node** pool;
1148  int n = read_number_of_nodes_needed();
1149  if (n <= 0) return 0;
1150  pool = (struct Node**)(independent_calloc(n, sizeof(struct Node), 0);
1151  if (pool == 0) die();
1152  // organize into a linked list...
1153  struct Node* first = pool[0];
1154  for (i = 0; i < n-1; ++i)
1155  pool[i]->next = pool[i+1];
1156  free(pool); // Can now free the array (or not, if it is needed later)
1157  return first;
1158  }
1159 */
1160 DLMALLOC_EXPORT void** dlindependent_calloc(size_t, size_t, void**);
1161 
1162 /*
1163  independent_comalloc(size_t n_elements, size_t sizes[], void* chunks[]);
1164 
1165  independent_comalloc allocates, all at once, a set of n_elements
1166  chunks with sizes indicated in the "sizes" array. It returns
1167  an array of pointers to these elements, each of which can be
1168  independently freed, realloc'ed etc. The elements are guaranteed to
1169  be adjacently allocated (this is not guaranteed to occur with
1170  multiple callocs or mallocs), which may also improve cache locality
1171  in some applications.
1172 
1173  The "chunks" argument is optional (i.e., may be null). If it is null
1174  the returned array is itself dynamically allocated and should also
1175  be freed when it is no longer needed. Otherwise, the chunks array
1176  must be of at least n_elements in length. It is filled in with the
1177  pointers to the chunks.
1178 
1179  In either case, independent_comalloc returns this pointer array, or
1180  null if the allocation failed. If n_elements is zero and chunks is
1181  null, it returns a chunk representing an array with zero elements
1182  (which should be freed if not wanted).
1183 
1184  Each element must be freed when it is no longer needed. This can be
1185  done all at once using bulk_free.
1186 
1187  independent_comallac differs from independent_calloc in that each
1188  element may have a different size, and also that it does not
1189  automatically clear elements.
1190 
1191  independent_comalloc can be used to speed up allocation in cases
1192  where several structs or objects must always be allocated at the
1193  same time. For example:
1194 
1195  struct Head { ... }
1196  struct Foot { ... }
1197 
1198  void send_message(char* msg) {
1199  int msglen = strlen(msg);
1200  size_t sizes[3] = { sizeof(struct Head), msglen, sizeof(struct Foot) };
1201  void* chunks[3];
1202  if (independent_comalloc(3, sizes, chunks) == 0)
1203  die();
1204  struct Head* head = (struct Head*)(chunks[0]);
1205  char* body = (char*)(chunks[1]);
1206  struct Foot* foot = (struct Foot*)(chunks[2]);
1207  // ...
1208  }
1209 
1210  In general though, independent_comalloc is worth using only for
1211  larger values of n_elements. For small values, you probably won't
1212  detect enough difference from series of malloc calls to bother.
1213 
1214  Overuse of independent_comalloc can increase overall memory usage,
1215  since it cannot reuse existing noncontiguous small chunks that
1216  might be available for some of the elements.
1217 */
1218 DLMALLOC_EXPORT void** dlindependent_comalloc(size_t, size_t*, void**);
1219 
1220 /*
1221  bulk_free(void* array[], size_t n_elements)
1222  Frees and clears (sets to null) each non-null pointer in the given
1223  array. This is likely to be faster than freeing them one-by-one.
1224  If footers are used, pointers that have been allocated in different
1225  mspaces are not freed or cleared, and the count of all such pointers
1226  is returned. For large arrays of pointers with poor locality, it
1227  may be worthwhile to sort this array before calling bulk_free.
1228 */
1229 DLMALLOC_EXPORT size_t dlbulk_free(void**, size_t n_elements);
1230 
1231 /*
1232  pvalloc(size_t n);
1233  Equivalent to valloc(minimum-page-that-holds(n)), that is,
1234  round up n to nearest pagesize.
1235  */
1236 DLMALLOC_EXPORT void* dlpvalloc(size_t);
1237 
1238 /*
1239  malloc_trim(size_t pad);
1240 
1241  If possible, gives memory back to the system (via negative arguments
1242  to sbrk) if there is unused memory at the `high' end of the malloc
1243  pool or in unused MMAP segments. You can call this after freeing
1244  large blocks of memory to potentially reduce the system-level memory
1245  requirements of a program. However, it cannot guarantee to reduce
1246  memory. Under some allocation patterns, some large free blocks of
1247  memory will be locked between two used chunks, so they cannot be
1248  given back to the system.
1249 
1250  The `pad' argument to malloc_trim represents the amount of free
1251  trailing space to leave untrimmed. If this argument is zero, only
1252  the minimum amount of memory to maintain internal data structures
1253  will be left. Non-zero arguments can be supplied to maintain enough
1254  trailing space to service future expected allocations without having
1255  to re-obtain memory from the system.
1256 
1257  Malloc_trim returns 1 if it actually released any memory, else 0.
1258 */
1259 DLMALLOC_EXPORT int dlmalloc_trim(size_t);
1260 
1261 /*
1262  malloc_stats();
1263  Prints on stderr the amount of space obtained from the system (both
1264  via sbrk and mmap), the maximum amount (which may be more than
1265  current if malloc_trim and/or munmap got called), and the current
1266  number of bytes allocated via malloc (or realloc, etc) but not yet
1267  freed. Note that this is the number of bytes allocated, not the
1268  number requested. It will be larger than the number requested
1269  because of alignment and bookkeeping overhead. Because it includes
1270  alignment wastage as being in use, this figure may be greater than
1271  zero even when no user-level chunks are allocated.
1272 
1273  The reported current and maximum system memory can be inaccurate if
1274  a program makes other calls to system memory allocation functions
1275  (normally sbrk) outside of malloc.
1276 
1277  malloc_stats prints only the most commonly interesting statistics.
1278  More information can be obtained by calling mallinfo.
1279 */
1280 DLMALLOC_EXPORT void dlmalloc_stats(void);
1281 
1282 /*
1283  malloc_usable_size(void* p);
1284 
1285  Returns the number of bytes you can actually use in
1286  an allocated chunk, which may be more than you requested (although
1287  often not) due to alignment and minimum size constraints.
1288  You can use this many bytes without worrying about
1289  overwriting other allocated objects. This is not a particularly great
1290  programming practice. malloc_usable_size can be more useful in
1291  debugging and assertions, for example:
1292 
1293  p = malloc(n);
1294  assert(malloc_usable_size(p) >= 256);
1295 */
1296 size_t dlmalloc_usable_size(void*);
1297 
1298 #endif /* ONLY_MSPACES */
1299 
1300 #if MSPACES
1301 
1302 /*
1303  mspace is an opaque type representing an independent
1304  region of space that supports mspace_malloc, etc.
1305 */
1306 typedef void* mspace;
1307 
1308 /*
1309  create_mspace creates and returns a new independent space with the
1310  given initial capacity, or, if 0, the default granularity size. It
1311  returns null if there is no system memory available to create the
1312  space. If argument locked is non-zero, the space uses a separate
1313  lock to control access. The capacity of the space will grow
1314  dynamically as needed to service mspace_malloc requests. You can
1315  control the sizes of incremental increases of this space by
1316  compiling with a different DEFAULT_GRANULARITY or dynamically
1317  setting with mallopt(M_GRANULARITY, value).
1318 */
1319 DLMALLOC_EXPORT mspace create_mspace(size_t capacity, int locked);
1320 
1321 /*
1322  destroy_mspace destroys the given space, and attempts to return all
1323  of its memory back to the system, returning the total number of
1324  bytes freed. After destruction, the results of access to all memory
1325  used by the space become undefined.
1326 */
1327 DLMALLOC_EXPORT size_t destroy_mspace(mspace msp);
1328 
1329 /*
1330  create_mspace_with_base uses the memory supplied as the initial base
1331  of a new mspace. Part (less than 128*sizeof(size_t) bytes) of this
1332  space is used for bookkeeping, so the capacity must be at least this
1333  large. (Otherwise 0 is returned.) When this initial space is
1334  exhausted, additional memory will be obtained from the system.
1335  Destroying this space will deallocate all additionally allocated
1336  space (if possible) but not the initial base.
1337 */
1338 DLMALLOC_EXPORT mspace create_mspace_with_base(void* base, size_t capacity, int locked);
1339 
1340 /*
1341  mspace_track_large_chunks controls whether requests for large chunks
1342  are allocated in their own untracked mmapped regions, separate from
1343  others in this mspace. By default large chunks are not tracked,
1344  which reduces fragmentation. However, such chunks are not
1345  necessarily released to the system upon destroy_mspace. Enabling
1346  tracking by setting to true may increase fragmentation, but avoids
1347  leakage when relying on destroy_mspace to release all memory
1348  allocated using this space. The function returns the previous
1349  setting.
1350 */
1351 DLMALLOC_EXPORT int mspace_track_large_chunks(mspace msp, int enable);
1352 
1353 
1354 /*
1355  mspace_malloc behaves as malloc, but operates within
1356  the given space.
1357 */
1358 DLMALLOC_EXPORT void* mspace_malloc(mspace msp, size_t bytes);
1359 
1360 /*
1361  mspace_free behaves as free, but operates within
1362  the given space.
1363 
1364  If compiled with FOOTERS==1, mspace_free is not actually needed.
1365  free may be called instead of mspace_free because freed chunks from
1366  any space are handled by their originating spaces.
1367 */
1368 DLMALLOC_EXPORT void mspace_free(mspace msp, void* mem);
1369 
1370 /*
1371  mspace_realloc behaves as realloc, but operates within
1372  the given space.
1373 
1374  If compiled with FOOTERS==1, mspace_realloc is not actually
1375  needed. realloc may be called instead of mspace_realloc because
1376  realloced chunks from any space are handled by their originating
1377  spaces.
1378 */
1379 DLMALLOC_EXPORT void* mspace_realloc(mspace msp, void* mem, size_t newsize);
1380 
1381 /*
1382  mspace_calloc behaves as calloc, but operates within
1383  the given space.
1384 */
1385 DLMALLOC_EXPORT void* mspace_calloc(mspace msp, size_t n_elements, size_t elem_size);
1386 
1387 /*
1388  mspace_memalign behaves as memalign, but operates within
1389  the given space.
1390 */
1391 DLMALLOC_EXPORT void* mspace_memalign(mspace msp, size_t alignment, size_t bytes);
1392 
1393 /*
1394  mspace_independent_calloc behaves as independent_calloc, but
1395  operates within the given space.
1396 */
1397 DLMALLOC_EXPORT void** mspace_independent_calloc(mspace msp, size_t n_elements,
1398  size_t elem_size, void* chunks[]);
1399 
1400 /*
1401  mspace_independent_comalloc behaves as independent_comalloc, but
1402  operates within the given space.
1403 */
1404 DLMALLOC_EXPORT void** mspace_independent_comalloc(mspace msp, size_t n_elements,
1405  size_t sizes[], void* chunks[]);
1406 
1407 /*
1408  mspace_footprint() returns the number of bytes obtained from the
1409  system for this space.
1410 */
1411 DLMALLOC_EXPORT size_t mspace_footprint(mspace msp);
1412 
1413 /*
1414  mspace_max_footprint() returns the peak number of bytes obtained from the
1415  system for this space.
1416 */
1417 DLMALLOC_EXPORT size_t mspace_max_footprint(mspace msp);
1418 
1419 
1420 #if !NO_MALLINFO
1421 /*
1422  mspace_mallinfo behaves as mallinfo, but reports properties of
1423  the given space.
1424 */
1425 DLMALLOC_EXPORT struct dlmallinfo mspace_mallinfo(mspace msp);
1426 #endif /* NO_MALLINFO */
1427 
1428 /*
1429  malloc_usable_size(void* p) behaves the same as malloc_usable_size;
1430 */
1431 DLMALLOC_EXPORT size_t mspace_usable_size(const void* mem);
1432 
1433 /*
1434  mspace_malloc_stats behaves as malloc_stats, but reports
1435  properties of the given space.
1436 */
1437 DLMALLOC_EXPORT void mspace_malloc_stats(mspace msp);
1438 
1439 /*
1440  mspace_trim behaves as malloc_trim, but
1441  operates within the given space.
1442 */
1443 DLMALLOC_EXPORT int mspace_trim(mspace msp, size_t pad);
1444 
1445 /*
1446  An alias for mallopt.
1447 */
1448 DLMALLOC_EXPORT int mspace_mallopt(int, int);
1449 
1450 DLMALLOC_EXPORT void* mspace_get_aligned (mspace msp,
1451  unsigned long n_user_data_bytes,
1452  unsigned long align,
1453  unsigned long align_offset);
1454 
1455 DLMALLOC_EXPORT int mspace_is_heap_object (mspace msp, void *p);
1456 
1457 DLMALLOC_EXPORT void mspace_get_address_and_size (mspace msp, char **addrp, size_t *sizep);
1458 DLMALLOC_EXPORT void mspace_put (mspace msp, void *p);
1459 DLMALLOC_EXPORT void mspace_put_no_offset (mspace msp, void *p);
1460 DLMALLOC_EXPORT size_t mspace_usable_size_with_delta (const void *p);
1461 DLMALLOC_EXPORT void mspace_disable_expand (mspace msp);
1462 DLMALLOC_EXPORT void *mspace_least_addr (mspace msp);
1465 DLMALLOC_EXPORT int mspace_enable_disable_trace (mspace msp, int enable);
1466 DLMALLOC_EXPORT int mspace_is_traced (mspace msp);
1467 
1468 #endif /* MSPACES */
1469 
1470 #ifdef __cplusplus
1471 } /* end of extern "C" */
1472 #endif /* __cplusplus */
1473 
1474 /*
1475  ========================================================================
1476  To make a fully customizable malloc.h header file, cut everything
1477  above this line, put into file malloc.h, edit to suit, and #include it
1478  on the next line, as well as in programs that use this malloc.
1479  ========================================================================
1480 */
#define DLMALLOC_EXPORT
Definition: dlmalloc.h:545
size_t dlmalloc_footprint(void)
Definition: dlmalloc.c:3941
void ** dlindependent_calloc(size_t n_elements, size_t elem_size, void *chunks[])
Definition: dlmalloc.c:3902
DLMALLOC_EXPORT int mspace_trim(mspace msp, size_t pad)
u8 pad[3]
log2 (size of the packing page block)
Definition: bihash_doc.h:61
DLMALLOC_EXPORT void * mspace_malloc(mspace msp, size_t bytes)
void * dlmemalign(size_t alignment, size_t bytes)
Definition: dlmalloc.c:3858
void * mspace
Definition: dlmalloc.h:1306
DLMALLOC_EXPORT void ** mspace_independent_calloc(mspace msp, size_t n_elements, size_t elem_size, void *chunks[])
size_t dlbulk_free(void *array[], size_t nelem)
Definition: dlmalloc.c:3913
DLMALLOC_EXPORT void * mspace_realloc(mspace msp, void *mem, size_t newsize)
void * dlpvalloc(size_t bytes)
Definition: dlmalloc.c:3895
DLMALLOC_EXPORT struct dlmallinfo mspace_mallinfo(mspace msp)
MALLINFO_FIELD_TYPE hblkhd
Definition: dlmalloc.h:804
void dlmalloc_stats()
Definition: dlmalloc.c:3972
DLMALLOC_EXPORT mspace create_mspace_with_base(void *base, size_t capacity, int locked)
DLMALLOC_EXPORT int mspace_is_heap_object(mspace msp, void *p)
MALLINFO_FIELD_TYPE uordblks
Definition: dlmalloc.h:807
DLMALLOC_EXPORT void * mspace_calloc(mspace msp, size_t n_elements, size_t elem_size)
MALLINFO_FIELD_TYPE arena
Definition: dlmalloc.h:800
size_t dlmalloc_usable_size(void *mem)
Definition: dlmalloc.c:3981
DLMALLOC_EXPORT void mspace_put_no_offset(mspace msp, void *p)
MALLINFO_FIELD_TYPE fsmblks
Definition: dlmalloc.h:806
MALLINFO_FIELD_TYPE ordblks
Definition: dlmalloc.h:801
void * dlmalloc(size_t bytes)
Definition: dlmalloc.c:3141
void * dlvalloc(size_t bytes)
Definition: dlmalloc.c:3888
MALLINFO_FIELD_TYPE hblks
Definition: dlmalloc.h:803
int dlmalloc_trim(size_t pad)
Definition: dlmalloc.c:3931
DLMALLOC_EXPORT void * mspace_memalign(mspace msp, size_t alignment, size_t bytes)
DLMALLOC_EXPORT void mspace_malloc_stats(mspace msp)
uword size
DLMALLOC_EXPORT void mheap_put_trace(uword offset, uword size)
Definition: mem_dlmalloc.c:150
MALLINFO_FIELD_TYPE usmblks
Definition: dlmalloc.h:805
DLMALLOC_EXPORT void mspace_disable_expand(mspace msp)
void * dlrealloc(void *oldmem, size_t bytes)
Definition: dlmalloc.c:3782
DLMALLOC_EXPORT int mspace_is_traced(mspace msp)
DLMALLOC_EXPORT mspace create_mspace(size_t capacity, int locked)
DLMALLOC_EXPORT void mspace_get_address_and_size(mspace msp, char **addrp, size_t *sizep)
DLMALLOC_EXPORT int mspace_mallopt(int, int)
DLMALLOC_EXPORT void ** mspace_independent_comalloc(mspace msp, size_t n_elements, size_t sizes[], void *chunks[])
size_t dlmalloc_max_footprint(void)
Definition: dlmalloc.c:3945
MALLINFO_FIELD_TYPE keepcost
Definition: dlmalloc.h:809
DLMALLOC_EXPORT int mspace_enable_disable_trace(mspace msp, int enable)
MALLINFO_FIELD_TYPE smblks
Definition: dlmalloc.h:802
#define align_offset(A)
Definition: dlmalloc.c:199
DLMALLOC_EXPORT void mspace_free(mspace msp, void *mem)
template key/value backing page structure
Definition: bihash_doc.h:44
DLMALLOC_EXPORT size_t mspace_usable_size(const void *mem)
u64 uword
Definition: types.h:112
void * dlrealloc_in_place(void *oldmem, size_t bytes)
Definition: dlmalloc.c:3827
DLMALLOC_EXPORT void * mspace_get_aligned(mspace msp, unsigned long n_user_data_bytes, unsigned long align, unsigned long align_offset)
#define MALLINFO_FIELD_TYPE
Definition: dlmalloc.h:745
void * mem
DLMALLOC_EXPORT size_t destroy_mspace(mspace msp)
DLMALLOC_EXPORT size_t mspace_footprint(mspace msp)
DLMALLOC_EXPORT void mheap_get_trace(uword offset, uword size)
Definition: mem_dlmalloc.c:68
DLMALLOC_EXPORT void mspace_put(mspace msp, void *p)
MALLINFO_FIELD_TYPE fordblks
Definition: dlmalloc.h:808
DLMALLOC_EXPORT size_t mspace_usable_size_with_delta(const void *p)
void ** dlindependent_comalloc(size_t n_elements, size_t sizes[], void *chunks[])
Definition: dlmalloc.c:3908
int dlmallopt(int param_number, int value)
Definition: dlmalloc.c:3977
size_t dlmalloc_set_footprint_limit(size_t bytes)
Definition: dlmalloc.c:3954
DLMALLOC_EXPORT int mspace_track_large_chunks(mspace msp, int enable)
int dlposix_memalign(void **pp, size_t alignment, size_t bytes)
Definition: dlmalloc.c:3865
DLMALLOC_EXPORT void * mspace_least_addr(mspace msp)
size_t dlmalloc_footprint_limit(void)
Definition: dlmalloc.c:3949
DLMALLOC_EXPORT size_t mspace_max_footprint(mspace msp)
void * dlcalloc(size_t n_elements, size_t elem_size)
Definition: dlmalloc.c:3388
void dlfree(void *mem)
Definition: dlmalloc.c:3279