malloc.h 36 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966
  1. /*
  2. This is a version (aka dlmalloc) of malloc/free/realloc written by
  3. Doug Lea and released to the public domain. Use, modify, and
  4. redistribute this code without permission or acknowledgement in any
  5. way you wish. Send questions, comments, complaints, performance
  6. data, etc to dl@cs.oswego.edu
  7. VERSION 2.7.2 Sat Aug 17 09:07:30 2002 Doug Lea (dl at gee)
  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. Hacked up for uClibc by Erik Andersen <andersen@codepoet.org>
  12. */
  13. #define mmap __mmap
  14. #define sysconf __sysconf
  15. #define sbrk __sbrk
  16. #include <features.h>
  17. #include <stddef.h>
  18. #include <unistd.h>
  19. #include <errno.h>
  20. #include <string.h>
  21. #include <malloc.h>
  22. #include <stdlib.h>
  23. #ifdef __UCLIBC_HAS_THREADS__
  24. # include <pthread.h>
  25. extern pthread_mutex_t __malloc_lock;
  26. #endif
  27. #define LOCK __pthread_mutex_lock(&__malloc_lock)
  28. #define UNLOCK __pthread_mutex_unlock(&__malloc_lock)
  29. /*
  30. MALLOC_ALIGNMENT is the minimum alignment for malloc'ed chunks.
  31. It must be a power of two at least 2 * (sizeof(size_t)), even on machines
  32. for which smaller alignments would suffice. It may be defined as
  33. larger than this though. Note however that code and data structures
  34. are optimized for the case of 8-byte alignment.
  35. */
  36. #ifndef MALLOC_ALIGNMENT
  37. #define MALLOC_ALIGNMENT (2 * (sizeof(size_t)))
  38. #endif
  39. /* The corresponding bit mask value */
  40. #define MALLOC_ALIGN_MASK (MALLOC_ALIGNMENT - 1)
  41. /*
  42. TRIM_FASTBINS controls whether free() of a very small chunk can
  43. immediately lead to trimming. Setting to true (1) can reduce memory
  44. footprint, but will almost always slow down programs that use a lot
  45. of small chunks.
  46. Define this only if you are willing to give up some speed to more
  47. aggressively reduce system-level memory footprint when releasing
  48. memory in programs that use many small chunks. You can get
  49. essentially the same effect by setting MXFAST to 0, but this can
  50. lead to even greater slowdowns in programs using many small chunks.
  51. TRIM_FASTBINS is an in-between compile-time option, that disables
  52. only those chunks bordering topmost memory from being placed in
  53. fastbins.
  54. */
  55. #ifndef TRIM_FASTBINS
  56. #define TRIM_FASTBINS 0
  57. #endif
  58. /*
  59. MORECORE-related declarations. By default, rely on sbrk
  60. */
  61. /*
  62. MORECORE is the name of the routine to call to obtain more memory
  63. from the system. See below for general guidance on writing
  64. alternative MORECORE functions, as well as a version for WIN32 and a
  65. sample version for pre-OSX macos.
  66. */
  67. #ifndef MORECORE
  68. #define MORECORE sbrk
  69. #endif
  70. /*
  71. MORECORE_FAILURE is the value returned upon failure of MORECORE
  72. as well as mmap. Since it cannot be an otherwise valid memory address,
  73. and must reflect values of standard sys calls, you probably ought not
  74. try to redefine it.
  75. */
  76. #ifndef MORECORE_FAILURE
  77. #define MORECORE_FAILURE (-1)
  78. #endif
  79. /*
  80. If MORECORE_CONTIGUOUS is true, take advantage of fact that
  81. consecutive calls to MORECORE with positive arguments always return
  82. contiguous increasing addresses. This is true of unix sbrk. Even
  83. if not defined, when regions happen to be contiguous, malloc will
  84. permit allocations spanning regions obtained from different
  85. calls. But defining this when applicable enables some stronger
  86. consistency checks and space efficiencies.
  87. */
  88. #ifndef MORECORE_CONTIGUOUS
  89. #define MORECORE_CONTIGUOUS 1
  90. #endif
  91. /*
  92. MMAP_AS_MORECORE_SIZE is the minimum mmap size argument to use if
  93. sbrk fails, and mmap is used as a backup (which is done only if
  94. HAVE_MMAP). The value must be a multiple of page size. This
  95. backup strategy generally applies only when systems have "holes" in
  96. address space, so sbrk cannot perform contiguous expansion, but
  97. there is still space available on system. On systems for which
  98. this is known to be useful (i.e. most linux kernels), this occurs
  99. only when programs allocate huge amounts of memory. Between this,
  100. and the fact that mmap regions tend to be limited, the size should
  101. be large, to avoid too many mmap calls and thus avoid running out
  102. of kernel resources.
  103. */
  104. #ifndef MMAP_AS_MORECORE_SIZE
  105. #define MMAP_AS_MORECORE_SIZE (1024 * 1024)
  106. #endif
  107. /*
  108. The system page size. To the extent possible, this malloc manages
  109. memory from the system in page-size units. Note that this value is
  110. cached during initialization into a field of malloc_state. So even
  111. if malloc_getpagesize is a function, it is only called once.
  112. The following mechanics for getpagesize were adapted from bsd/gnu
  113. getpagesize.h. If none of the system-probes here apply, a value of
  114. 4096 is used, which should be OK: If they don't apply, then using
  115. the actual value probably doesn't impact performance.
  116. */
  117. #ifndef malloc_getpagesize
  118. # include <unistd.h>
  119. # define malloc_getpagesize sysconf(_SC_PAGESIZE)
  120. #else /* just guess */
  121. # define malloc_getpagesize (4096)
  122. #endif
  123. /* mallopt tuning options */
  124. /*
  125. M_MXFAST is the maximum request size used for "fastbins", special bins
  126. that hold returned chunks without consolidating their spaces. This
  127. enables future requests for chunks of the same size to be handled
  128. very quickly, but can increase fragmentation, and thus increase the
  129. overall memory footprint of a program.
  130. This malloc manages fastbins very conservatively yet still
  131. efficiently, so fragmentation is rarely a problem for values less
  132. than or equal to the default. The maximum supported value of MXFAST
  133. is 80. You wouldn't want it any higher than this anyway. Fastbins
  134. are designed especially for use with many small structs, objects or
  135. strings -- the default handles structs/objects/arrays with sizes up
  136. to 16 4byte fields, or small strings representing words, tokens,
  137. etc. Using fastbins for larger objects normally worsens
  138. fragmentation without improving speed.
  139. M_MXFAST is set in REQUEST size units. It is internally used in
  140. chunksize units, which adds padding and alignment. You can reduce
  141. M_MXFAST to 0 to disable all use of fastbins. This causes the malloc
  142. algorithm to be a closer approximation of fifo-best-fit in all cases,
  143. not just for larger requests, but will generally cause it to be
  144. slower.
  145. */
  146. /* M_MXFAST is a standard SVID/XPG tuning option, usually listed in malloc.h */
  147. #ifndef M_MXFAST
  148. #define M_MXFAST 1
  149. #endif
  150. #ifndef DEFAULT_MXFAST
  151. #define DEFAULT_MXFAST 64
  152. #endif
  153. /*
  154. M_TRIM_THRESHOLD is the maximum amount of unused top-most memory
  155. to keep before releasing via malloc_trim in free().
  156. Automatic trimming is mainly useful in long-lived programs.
  157. Because trimming via sbrk can be slow on some systems, and can
  158. sometimes be wasteful (in cases where programs immediately
  159. afterward allocate more large chunks) the value should be high
  160. enough so that your overall system performance would improve by
  161. releasing this much memory.
  162. The trim threshold and the mmap control parameters (see below)
  163. can be traded off with one another. Trimming and mmapping are
  164. two different ways of releasing unused memory back to the
  165. system. Between these two, it is often possible to keep
  166. system-level demands of a long-lived program down to a bare
  167. minimum. For example, in one test suite of sessions measuring
  168. the XF86 X server on Linux, using a trim threshold of 128K and a
  169. mmap threshold of 192K led to near-minimal long term resource
  170. consumption.
  171. If you are using this malloc in a long-lived program, it should
  172. pay to experiment with these values. As a rough guide, you
  173. might set to a value close to the average size of a process
  174. (program) running on your system. Releasing this much memory
  175. would allow such a process to run in memory. Generally, it's
  176. worth it to tune for trimming rather tham memory mapping when a
  177. program undergoes phases where several large chunks are
  178. allocated and released in ways that can reuse each other's
  179. storage, perhaps mixed with phases where there are no such
  180. chunks at all. And in well-behaved long-lived programs,
  181. controlling release of large blocks via trimming versus mapping
  182. is usually faster.
  183. However, in most programs, these parameters serve mainly as
  184. protection against the system-level effects of carrying around
  185. massive amounts of unneeded memory. Since frequent calls to
  186. sbrk, mmap, and munmap otherwise degrade performance, the default
  187. parameters are set to relatively high values that serve only as
  188. safeguards.
  189. The trim value must be greater than page size to have any useful
  190. effect. To disable trimming completely, you can set to
  191. (unsigned long)(-1)
  192. Trim settings interact with fastbin (MXFAST) settings: Unless
  193. TRIM_FASTBINS is defined, automatic trimming never takes place upon
  194. freeing a chunk with size less than or equal to MXFAST. Trimming is
  195. instead delayed until subsequent freeing of larger chunks. However,
  196. you can still force an attempted trim by calling malloc_trim.
  197. Also, trimming is not generally possible in cases where
  198. the main arena is obtained via mmap.
  199. Note that the trick some people use of mallocing a huge space and
  200. then freeing it at program startup, in an attempt to reserve system
  201. memory, doesn't have the intended effect under automatic trimming,
  202. since that memory will immediately be returned to the system.
  203. */
  204. #define M_TRIM_THRESHOLD -1
  205. #ifndef DEFAULT_TRIM_THRESHOLD
  206. #define DEFAULT_TRIM_THRESHOLD (256 * 1024)
  207. #endif
  208. /*
  209. M_TOP_PAD is the amount of extra `padding' space to allocate or
  210. retain whenever sbrk is called. It is used in two ways internally:
  211. * When sbrk is called to extend the top of the arena to satisfy
  212. a new malloc request, this much padding is added to the sbrk
  213. request.
  214. * When malloc_trim is called automatically from free(),
  215. it is used as the `pad' argument.
  216. In both cases, the actual amount of padding is rounded
  217. so that the end of the arena is always a system page boundary.
  218. The main reason for using padding is to avoid calling sbrk so
  219. often. Having even a small pad greatly reduces the likelihood
  220. that nearly every malloc request during program start-up (or
  221. after trimming) will invoke sbrk, which needlessly wastes
  222. time.
  223. Automatic rounding-up to page-size units is normally sufficient
  224. to avoid measurable overhead, so the default is 0. However, in
  225. systems where sbrk is relatively slow, it can pay to increase
  226. this value, at the expense of carrying around more memory than
  227. the program needs.
  228. */
  229. #define M_TOP_PAD -2
  230. #ifndef DEFAULT_TOP_PAD
  231. #define DEFAULT_TOP_PAD (0)
  232. #endif
  233. /*
  234. M_MMAP_THRESHOLD is the request size threshold for using mmap()
  235. to service a request. Requests of at least this size that cannot
  236. be allocated using already-existing space will be serviced via mmap.
  237. (If enough normal freed space already exists it is used instead.)
  238. Using mmap segregates relatively large chunks of memory so that
  239. they can be individually obtained and released from the host
  240. system. A request serviced through mmap is never reused by any
  241. other request (at least not directly; the system may just so
  242. happen to remap successive requests to the same locations).
  243. Segregating space in this way has the benefits that:
  244. 1. Mmapped space can ALWAYS be individually released back
  245. to the system, which helps keep the system level memory
  246. demands of a long-lived program low.
  247. 2. Mapped memory can never become `locked' between
  248. other chunks, as can happen with normally allocated chunks, which
  249. means that even trimming via malloc_trim would not release them.
  250. 3. On some systems with "holes" in address spaces, mmap can obtain
  251. memory that sbrk cannot.
  252. However, it has the disadvantages that:
  253. 1. The space cannot be reclaimed, consolidated, and then
  254. used to service later requests, as happens with normal chunks.
  255. 2. It can lead to more wastage because of mmap page alignment
  256. requirements
  257. 3. It causes malloc performance to be more dependent on host
  258. system memory management support routines which may vary in
  259. implementation quality and may impose arbitrary
  260. limitations. Generally, servicing a request via normal
  261. malloc steps is faster than going through a system's mmap.
  262. The advantages of mmap nearly always outweigh disadvantages for
  263. "large" chunks, but the value of "large" varies across systems. The
  264. default is an empirically derived value that works well in most
  265. systems.
  266. */
  267. #define M_MMAP_THRESHOLD -3
  268. #ifndef DEFAULT_MMAP_THRESHOLD
  269. #define DEFAULT_MMAP_THRESHOLD (256 * 1024)
  270. #endif
  271. /*
  272. M_MMAP_MAX is the maximum number of requests to simultaneously
  273. service using mmap. This parameter exists because
  274. . Some systems have a limited number of internal tables for
  275. use by mmap, and using more than a few of them may degrade
  276. performance.
  277. The default is set to a value that serves only as a safeguard.
  278. Setting to 0 disables use of mmap for servicing large requests. If
  279. HAVE_MMAP is not set, the default value is 0, and attempts to set it
  280. to non-zero values in mallopt will fail.
  281. */
  282. #define M_MMAP_MAX -4
  283. #ifndef DEFAULT_MMAP_MAX
  284. #define DEFAULT_MMAP_MAX (65536)
  285. #endif
  286. /* ------------------ MMAP support ------------------ */
  287. #include <fcntl.h>
  288. #include <sys/mman.h>
  289. #if !defined(MAP_ANONYMOUS) && defined(MAP_ANON)
  290. #define MAP_ANONYMOUS MAP_ANON
  291. #endif
  292. #ifdef __ARCH_HAS_MMU__
  293. #define MMAP(addr, size, prot) \
  294. (mmap((addr), (size), (prot), MAP_PRIVATE|MAP_ANONYMOUS, 0, 0))
  295. #else
  296. #define MMAP(addr, size, prot) \
  297. (mmap((addr), (size), (prot), MAP_SHARED|MAP_ANONYMOUS, 0, 0))
  298. #endif
  299. /* ----------------------- Chunk representations ----------------------- */
  300. /*
  301. This struct declaration is misleading (but accurate and necessary).
  302. It declares a "view" into memory allowing access to necessary
  303. fields at known offsets from a given base. See explanation below.
  304. */
  305. struct malloc_chunk {
  306. size_t prev_size; /* Size of previous chunk (if free). */
  307. size_t size; /* Size in bytes, including overhead. */
  308. struct malloc_chunk* fd; /* double links -- used only if free. */
  309. struct malloc_chunk* bk;
  310. };
  311. typedef struct malloc_chunk* mchunkptr;
  312. /*
  313. malloc_chunk details:
  314. (The following includes lightly edited explanations by Colin Plumb.)
  315. Chunks of memory are maintained using a `boundary tag' method as
  316. described in e.g., Knuth or Standish. (See the paper by Paul
  317. Wilson ftp://ftp.cs.utexas.edu/pub/garbage/allocsrv.ps for a
  318. survey of such techniques.) Sizes of free chunks are stored both
  319. in the front of each chunk and at the end. This makes
  320. consolidating fragmented chunks into bigger chunks very fast. The
  321. size fields also hold bits representing whether chunks are free or
  322. in use.
  323. An allocated chunk looks like this:
  324. chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  325. | Size of previous chunk, if allocated | |
  326. +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  327. | Size of chunk, in bytes |P|
  328. mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  329. | User data starts here... .
  330. . .
  331. . (malloc_usable_space() bytes) .
  332. . |
  333. nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  334. | Size of chunk |
  335. +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  336. Where "chunk" is the front of the chunk for the purpose of most of
  337. the malloc code, but "mem" is the pointer that is returned to the
  338. user. "Nextchunk" is the beginning of the next contiguous chunk.
  339. Chunks always begin on even word boundries, so the mem portion
  340. (which is returned to the user) is also on an even word boundary, and
  341. thus at least double-word aligned.
  342. Free chunks are stored in circular doubly-linked lists, and look like this:
  343. chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  344. | Size of previous chunk |
  345. +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  346. `head:' | Size of chunk, in bytes |P|
  347. mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  348. | Forward pointer to next chunk in list |
  349. +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  350. | Back pointer to previous chunk in list |
  351. +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  352. | Unused space (may be 0 bytes long) .
  353. . .
  354. . |
  355. nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  356. `foot:' | Size of chunk, in bytes |
  357. +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  358. The P (PREV_INUSE) bit, stored in the unused low-order bit of the
  359. chunk size (which is always a multiple of two words), is an in-use
  360. bit for the *previous* chunk. If that bit is *clear*, then the
  361. word before the current chunk size contains the previous chunk
  362. size, and can be used to find the front of the previous chunk.
  363. The very first chunk allocated always has this bit set,
  364. preventing access to non-existent (or non-owned) memory. If
  365. prev_inuse is set for any given chunk, then you CANNOT determine
  366. the size of the previous chunk, and might even get a memory
  367. addressing fault when trying to do so.
  368. Note that the `foot' of the current chunk is actually represented
  369. as the prev_size of the NEXT chunk. This makes it easier to
  370. deal with alignments etc but can be very confusing when trying
  371. to extend or adapt this code.
  372. The two exceptions to all this are
  373. 1. The special chunk `top' doesn't bother using the
  374. trailing size field since there is no next contiguous chunk
  375. that would have to index off it. After initialization, `top'
  376. is forced to always exist. If it would become less than
  377. MINSIZE bytes long, it is replenished.
  378. 2. Chunks allocated via mmap, which have the second-lowest-order
  379. bit (IS_MMAPPED) set in their size fields. Because they are
  380. allocated one-by-one, each must contain its own trailing size field.
  381. */
  382. /*
  383. ---------- Size and alignment checks and conversions ----------
  384. */
  385. /* conversion from malloc headers to user pointers, and back */
  386. #define chunk2mem(p) ((void*)((char*)(p) + 2*(sizeof(size_t))))
  387. #define mem2chunk(mem) ((mchunkptr)((char*)(mem) - 2*(sizeof(size_t))))
  388. /* The smallest possible chunk */
  389. #define MIN_CHUNK_SIZE (sizeof(struct malloc_chunk))
  390. /* The smallest size we can malloc is an aligned minimal chunk */
  391. #define MINSIZE \
  392. (unsigned long)(((MIN_CHUNK_SIZE+MALLOC_ALIGN_MASK) & ~MALLOC_ALIGN_MASK))
  393. /* Check if m has acceptable alignment */
  394. #define aligned_OK(m) (((unsigned long)((m)) & (MALLOC_ALIGN_MASK)) == 0)
  395. /* Check if a request is so large that it would wrap around zero when
  396. padded and aligned. To simplify some other code, the bound is made
  397. low enough so that adding MINSIZE will also not wrap around sero.
  398. */
  399. #define REQUEST_OUT_OF_RANGE(req) \
  400. ((unsigned long)(req) >= \
  401. (unsigned long)(size_t)(-2 * MINSIZE))
  402. /* pad request bytes into a usable size -- internal version */
  403. #define request2size(req) \
  404. (((req) + (sizeof(size_t)) + MALLOC_ALIGN_MASK < MINSIZE) ? \
  405. MINSIZE : \
  406. ((req) + (sizeof(size_t)) + MALLOC_ALIGN_MASK) & ~MALLOC_ALIGN_MASK)
  407. /* Same, except also perform argument check */
  408. #define checked_request2size(req, sz) \
  409. if (REQUEST_OUT_OF_RANGE(req)) { \
  410. errno = ENOMEM; \
  411. return 0; \
  412. } \
  413. (sz) = request2size(req);
  414. /*
  415. --------------- Physical chunk operations ---------------
  416. */
  417. /* size field is or'ed with PREV_INUSE when previous adjacent chunk in use */
  418. #define PREV_INUSE 0x1
  419. /* extract inuse bit of previous chunk */
  420. #define prev_inuse(p) ((p)->size & PREV_INUSE)
  421. /* size field is or'ed with IS_MMAPPED if the chunk was obtained with mmap() */
  422. #define IS_MMAPPED 0x2
  423. /* check for mmap()'ed chunk */
  424. #define chunk_is_mmapped(p) ((p)->size & IS_MMAPPED)
  425. /* Bits to mask off when extracting size
  426. Note: IS_MMAPPED is intentionally not masked off from size field in
  427. macros for which mmapped chunks should never be seen. This should
  428. cause helpful core dumps to occur if it is tried by accident by
  429. people extending or adapting this malloc.
  430. */
  431. #define SIZE_BITS (PREV_INUSE|IS_MMAPPED)
  432. /* Get size, ignoring use bits */
  433. #define chunksize(p) ((p)->size & ~(SIZE_BITS))
  434. /* Ptr to next physical malloc_chunk. */
  435. #define next_chunk(p) ((mchunkptr)( ((char*)(p)) + ((p)->size & ~PREV_INUSE) ))
  436. /* Ptr to previous physical malloc_chunk */
  437. #define prev_chunk(p) ((mchunkptr)( ((char*)(p)) - ((p)->prev_size) ))
  438. /* Treat space at ptr + offset as a chunk */
  439. #define chunk_at_offset(p, s) ((mchunkptr)(((char*)(p)) + (s)))
  440. /* extract p's inuse bit */
  441. #define inuse(p)\
  442. ((((mchunkptr)(((char*)(p))+((p)->size & ~PREV_INUSE)))->size) & PREV_INUSE)
  443. /* set/clear chunk as being inuse without otherwise disturbing */
  444. #define set_inuse(p)\
  445. ((mchunkptr)(((char*)(p)) + ((p)->size & ~PREV_INUSE)))->size |= PREV_INUSE
  446. #define clear_inuse(p)\
  447. ((mchunkptr)(((char*)(p)) + ((p)->size & ~PREV_INUSE)))->size &= ~(PREV_INUSE)
  448. /* check/set/clear inuse bits in known places */
  449. #define inuse_bit_at_offset(p, s)\
  450. (((mchunkptr)(((char*)(p)) + (s)))->size & PREV_INUSE)
  451. #define set_inuse_bit_at_offset(p, s)\
  452. (((mchunkptr)(((char*)(p)) + (s)))->size |= PREV_INUSE)
  453. #define clear_inuse_bit_at_offset(p, s)\
  454. (((mchunkptr)(((char*)(p)) + (s)))->size &= ~(PREV_INUSE))
  455. /* Set size at head, without disturbing its use bit */
  456. #define set_head_size(p, s) ((p)->size = (((p)->size & PREV_INUSE) | (s)))
  457. /* Set size/use field */
  458. #define set_head(p, s) ((p)->size = (s))
  459. /* Set size at footer (only when chunk is not in use) */
  460. #define set_foot(p, s) (((mchunkptr)((char*)(p) + (s)))->prev_size = (s))
  461. /* -------------------- Internal data structures -------------------- */
  462. /*
  463. Bins
  464. An array of bin headers for free chunks. Each bin is doubly
  465. linked. The bins are approximately proportionally (log) spaced.
  466. There are a lot of these bins (128). This may look excessive, but
  467. works very well in practice. Most bins hold sizes that are
  468. unusual as malloc request sizes, but are more usual for fragments
  469. and consolidated sets of chunks, which is what these bins hold, so
  470. they can be found quickly. All procedures maintain the invariant
  471. that no consolidated chunk physically borders another one, so each
  472. chunk in a list is known to be preceeded and followed by either
  473. inuse chunks or the ends of memory.
  474. Chunks in bins are kept in size order, with ties going to the
  475. approximately least recently used chunk. Ordering isn't needed
  476. for the small bins, which all contain the same-sized chunks, but
  477. facilitates best-fit allocation for larger chunks. These lists
  478. are just sequential. Keeping them in order almost never requires
  479. enough traversal to warrant using fancier ordered data
  480. structures.
  481. Chunks of the same size are linked with the most
  482. recently freed at the front, and allocations are taken from the
  483. back. This results in LRU (FIFO) allocation order, which tends
  484. to give each chunk an equal opportunity to be consolidated with
  485. adjacent freed chunks, resulting in larger free chunks and less
  486. fragmentation.
  487. To simplify use in double-linked lists, each bin header acts
  488. as a malloc_chunk. This avoids special-casing for headers.
  489. But to conserve space and improve locality, we allocate
  490. only the fd/bk pointers of bins, and then use repositioning tricks
  491. to treat these as the fields of a malloc_chunk*.
  492. */
  493. typedef struct malloc_chunk* mbinptr;
  494. /* addressing -- note that bin_at(0) does not exist */
  495. #define bin_at(m, i) ((mbinptr)((char*)&((m)->bins[(i)<<1]) - ((sizeof(size_t))<<1)))
  496. /* analog of ++bin */
  497. #define next_bin(b) ((mbinptr)((char*)(b) + (sizeof(mchunkptr)<<1)))
  498. /* Reminders about list directionality within bins */
  499. #define first(b) ((b)->fd)
  500. #define last(b) ((b)->bk)
  501. /* Take a chunk off a bin list */
  502. #define unlink(P, BK, FD) { \
  503. FD = P->fd; \
  504. BK = P->bk; \
  505. if (FD->bk != P || BK->fd != P) \
  506. abort(); \
  507. FD->bk = BK; \
  508. BK->fd = FD; \
  509. }
  510. /*
  511. Indexing
  512. Bins for sizes < 512 bytes contain chunks of all the same size, spaced
  513. 8 bytes apart. Larger bins are approximately logarithmically spaced:
  514. 64 bins of size 8
  515. 32 bins of size 64
  516. 16 bins of size 512
  517. 8 bins of size 4096
  518. 4 bins of size 32768
  519. 2 bins of size 262144
  520. 1 bin of size what's left
  521. The bins top out around 1MB because we expect to service large
  522. requests via mmap.
  523. */
  524. #define NBINS 96
  525. #define NSMALLBINS 32
  526. #define SMALLBIN_WIDTH 8
  527. #define MIN_LARGE_SIZE 256
  528. #define in_smallbin_range(sz) \
  529. ((unsigned long)(sz) < (unsigned long)MIN_LARGE_SIZE)
  530. #define smallbin_index(sz) (((unsigned)(sz)) >> 3)
  531. #define bin_index(sz) \
  532. ((in_smallbin_range(sz)) ? smallbin_index(sz) : __malloc_largebin_index(sz))
  533. /*
  534. FIRST_SORTED_BIN_SIZE is the chunk size corresponding to the
  535. first bin that is maintained in sorted order. This must
  536. be the smallest size corresponding to a given bin.
  537. Normally, this should be MIN_LARGE_SIZE. But you can weaken
  538. best fit guarantees to sometimes speed up malloc by increasing value.
  539. Doing this means that malloc may choose a chunk that is
  540. non-best-fitting by up to the width of the bin.
  541. Some useful cutoff values:
  542. 512 - all bins sorted
  543. 2560 - leaves bins <= 64 bytes wide unsorted
  544. 12288 - leaves bins <= 512 bytes wide unsorted
  545. 65536 - leaves bins <= 4096 bytes wide unsorted
  546. 262144 - leaves bins <= 32768 bytes wide unsorted
  547. -1 - no bins sorted (not recommended!)
  548. */
  549. #define FIRST_SORTED_BIN_SIZE MIN_LARGE_SIZE
  550. /* #define FIRST_SORTED_BIN_SIZE 65536 */
  551. /*
  552. Unsorted chunks
  553. All remainders from chunk splits, as well as all returned chunks,
  554. are first placed in the "unsorted" bin. They are then placed
  555. in regular bins after malloc gives them ONE chance to be used before
  556. binning. So, basically, the unsorted_chunks list acts as a queue,
  557. with chunks being placed on it in free (and __malloc_consolidate),
  558. and taken off (to be either used or placed in bins) in malloc.
  559. */
  560. /* The otherwise unindexable 1-bin is used to hold unsorted chunks. */
  561. #define unsorted_chunks(M) (bin_at(M, 1))
  562. /*
  563. Top
  564. The top-most available chunk (i.e., the one bordering the end of
  565. available memory) is treated specially. It is never included in
  566. any bin, is used only if no other chunk is available, and is
  567. released back to the system if it is very large (see
  568. M_TRIM_THRESHOLD). Because top initially
  569. points to its own bin with initial zero size, thus forcing
  570. extension on the first malloc request, we avoid having any special
  571. code in malloc to check whether it even exists yet. But we still
  572. need to do so when getting memory from system, so we make
  573. initial_top treat the bin as a legal but unusable chunk during the
  574. interval between initialization and the first call to
  575. __malloc_alloc. (This is somewhat delicate, since it relies on
  576. the 2 preceding words to be zero during this interval as well.)
  577. */
  578. /* Conveniently, the unsorted bin can be used as dummy top on first call */
  579. #define initial_top(M) (unsorted_chunks(M))
  580. /*
  581. Binmap
  582. To help compensate for the large number of bins, a one-level index
  583. structure is used for bin-by-bin searching. `binmap' is a
  584. bitvector recording whether bins are definitely empty so they can
  585. be skipped over during during traversals. The bits are NOT always
  586. cleared as soon as bins are empty, but instead only
  587. when they are noticed to be empty during traversal in malloc.
  588. */
  589. /* Conservatively use 32 bits per map word, even if on 64bit system */
  590. #define BINMAPSHIFT 5
  591. #define BITSPERMAP (1U << BINMAPSHIFT)
  592. #define BINMAPSIZE (NBINS / BITSPERMAP)
  593. #define idx2block(i) ((i) >> BINMAPSHIFT)
  594. #define idx2bit(i) ((1U << ((i) & ((1U << BINMAPSHIFT)-1))))
  595. #define mark_bin(m,i) ((m)->binmap[idx2block(i)] |= idx2bit(i))
  596. #define unmark_bin(m,i) ((m)->binmap[idx2block(i)] &= ~(idx2bit(i)))
  597. #define get_binmap(m,i) ((m)->binmap[idx2block(i)] & idx2bit(i))
  598. /*
  599. Fastbins
  600. An array of lists holding recently freed small chunks. Fastbins
  601. are not doubly linked. It is faster to single-link them, and
  602. since chunks are never removed from the middles of these lists,
  603. double linking is not necessary. Also, unlike regular bins, they
  604. are not even processed in FIFO order (they use faster LIFO) since
  605. ordering doesn't much matter in the transient contexts in which
  606. fastbins are normally used.
  607. Chunks in fastbins keep their inuse bit set, so they cannot
  608. be consolidated with other free chunks. __malloc_consolidate
  609. releases all chunks in fastbins and consolidates them with
  610. other free chunks.
  611. */
  612. typedef struct malloc_chunk* mfastbinptr;
  613. /* offset 2 to use otherwise unindexable first 2 bins */
  614. #define fastbin_index(sz) ((((unsigned int)(sz)) >> 3) - 2)
  615. /* The maximum fastbin request size we support */
  616. #define MAX_FAST_SIZE 80
  617. #define NFASTBINS (fastbin_index(request2size(MAX_FAST_SIZE))+1)
  618. /*
  619. FASTBIN_CONSOLIDATION_THRESHOLD is the size of a chunk in free()
  620. that triggers automatic consolidation of possibly-surrounding
  621. fastbin chunks. This is a heuristic, so the exact value should not
  622. matter too much. It is defined at half the default trim threshold as a
  623. compromise heuristic to only attempt consolidation if it is likely
  624. to lead to trimming. However, it is not dynamically tunable, since
  625. consolidation reduces fragmentation surrounding loarge chunks even
  626. if trimming is not used.
  627. */
  628. #define FASTBIN_CONSOLIDATION_THRESHOLD \
  629. ((unsigned long)(DEFAULT_TRIM_THRESHOLD) >> 1)
  630. /*
  631. Since the lowest 2 bits in max_fast don't matter in size comparisons,
  632. they are used as flags.
  633. */
  634. /*
  635. ANYCHUNKS_BIT held in max_fast indicates that there may be any
  636. freed chunks at all. It is set true when entering a chunk into any
  637. bin.
  638. */
  639. #define ANYCHUNKS_BIT (1U)
  640. #define have_anychunks(M) (((M)->max_fast & ANYCHUNKS_BIT))
  641. #define set_anychunks(M) ((M)->max_fast |= ANYCHUNKS_BIT)
  642. #define clear_anychunks(M) ((M)->max_fast &= ~ANYCHUNKS_BIT)
  643. /*
  644. FASTCHUNKS_BIT held in max_fast indicates that there are probably
  645. some fastbin chunks. It is set true on entering a chunk into any
  646. fastbin, and cleared only in __malloc_consolidate.
  647. */
  648. #define FASTCHUNKS_BIT (2U)
  649. #define have_fastchunks(M) (((M)->max_fast & FASTCHUNKS_BIT))
  650. #define set_fastchunks(M) ((M)->max_fast |= (FASTCHUNKS_BIT|ANYCHUNKS_BIT))
  651. #define clear_fastchunks(M) ((M)->max_fast &= ~(FASTCHUNKS_BIT))
  652. /* Set value of max_fast. Use impossibly small value if 0. */
  653. #define set_max_fast(M, s) \
  654. (M)->max_fast = (((s) == 0)? SMALLBIN_WIDTH: request2size(s)) | \
  655. ((M)->max_fast & (FASTCHUNKS_BIT|ANYCHUNKS_BIT))
  656. #define get_max_fast(M) \
  657. ((M)->max_fast & ~(FASTCHUNKS_BIT | ANYCHUNKS_BIT))
  658. /*
  659. morecore_properties is a status word holding dynamically discovered
  660. or controlled properties of the morecore function
  661. */
  662. #define MORECORE_CONTIGUOUS_BIT (1U)
  663. #define contiguous(M) \
  664. (((M)->morecore_properties & MORECORE_CONTIGUOUS_BIT))
  665. #define noncontiguous(M) \
  666. (((M)->morecore_properties & MORECORE_CONTIGUOUS_BIT) == 0)
  667. #define set_contiguous(M) \
  668. ((M)->morecore_properties |= MORECORE_CONTIGUOUS_BIT)
  669. #define set_noncontiguous(M) \
  670. ((M)->morecore_properties &= ~MORECORE_CONTIGUOUS_BIT)
  671. /*
  672. ----------- Internal state representation and initialization -----------
  673. */
  674. struct malloc_state {
  675. /* The maximum chunk size to be eligible for fastbin */
  676. size_t max_fast; /* low 2 bits used as flags */
  677. /* Fastbins */
  678. mfastbinptr fastbins[NFASTBINS];
  679. /* Base of the topmost chunk -- not otherwise kept in a bin */
  680. mchunkptr top;
  681. /* The remainder from the most recent split of a small request */
  682. mchunkptr last_remainder;
  683. /* Normal bins packed as described above */
  684. mchunkptr bins[NBINS * 2];
  685. /* Bitmap of bins. Trailing zero map handles cases of largest binned size */
  686. unsigned int binmap[BINMAPSIZE+1];
  687. /* Tunable parameters */
  688. unsigned long trim_threshold;
  689. size_t top_pad;
  690. size_t mmap_threshold;
  691. /* Memory map support */
  692. int n_mmaps;
  693. int n_mmaps_max;
  694. int max_n_mmaps;
  695. /* Cache malloc_getpagesize */
  696. unsigned int pagesize;
  697. /* Track properties of MORECORE */
  698. unsigned int morecore_properties;
  699. /* Statistics */
  700. size_t mmapped_mem;
  701. size_t sbrked_mem;
  702. size_t max_sbrked_mem;
  703. size_t max_mmapped_mem;
  704. size_t max_total_mem;
  705. };
  706. typedef struct malloc_state *mstate;
  707. /*
  708. There is exactly one instance of this struct in this malloc.
  709. If you are adapting this malloc in a way that does NOT use a static
  710. malloc_state, you MUST explicitly zero-fill it before using. This
  711. malloc relies on the property that malloc_state is initialized to
  712. all zeroes (as is true of C statics).
  713. */
  714. extern struct malloc_state __malloc_state; /* never directly referenced */
  715. /*
  716. All uses of av_ are via get_malloc_state().
  717. At most one "call" to get_malloc_state is made per invocation of
  718. the public versions of malloc and free, but other routines
  719. that in turn invoke malloc and/or free may call more then once.
  720. Also, it is called in check* routines if __MALLOC_DEBUGGING is set.
  721. */
  722. #define get_malloc_state() (&(__malloc_state))
  723. /* External internal utilities operating on mstates */
  724. void __malloc_consolidate(mstate) attribute_hidden;
  725. /* Debugging support */
  726. #if ! __MALLOC_DEBUGGING
  727. #define check_chunk(P)
  728. #define check_free_chunk(P)
  729. #define check_inuse_chunk(P)
  730. #define check_remalloced_chunk(P,N)
  731. #define check_malloced_chunk(P,N)
  732. #define check_malloc_state()
  733. #define assert(x) ((void)0)
  734. #else
  735. #define check_chunk(P) __do_check_chunk(P)
  736. #define check_free_chunk(P) __do_check_free_chunk(P)
  737. #define check_inuse_chunk(P) __do_check_inuse_chunk(P)
  738. #define check_remalloced_chunk(P,N) __do_check_remalloced_chunk(P,N)
  739. #define check_malloced_chunk(P,N) __do_check_malloced_chunk(P,N)
  740. #define check_malloc_state() __do_check_malloc_state()
  741. extern void __do_check_chunk(mchunkptr p);
  742. extern void __do_check_free_chunk(mchunkptr p);
  743. extern void __do_check_inuse_chunk(mchunkptr p);
  744. extern void __do_check_remalloced_chunk(mchunkptr p, size_t s);
  745. extern void __do_check_malloced_chunk(mchunkptr p, size_t s);
  746. extern void __do_check_malloc_state(void);
  747. #include <assert.h>
  748. #endif