malloc.h 35 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962
  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. #include <features.h>
  14. #include <stddef.h>
  15. #include <unistd.h>
  16. #include <errno.h>
  17. #include <string.h>
  18. #include <malloc.h>
  19. #ifdef __UCLIBC_HAS_THREADS__
  20. #include <pthread.h>
  21. extern pthread_mutex_t __malloc_lock;
  22. # define LOCK __pthread_mutex_lock(&__malloc_lock)
  23. # define UNLOCK __pthread_mutex_unlock(&__malloc_lock);
  24. #else
  25. # define LOCK
  26. # define UNLOCK
  27. #endif
  28. /*
  29. MALLOC_ALIGNMENT is the minimum alignment for malloc'ed chunks.
  30. It must be a power of two at least 2 * (sizeof(size_t)), even on machines
  31. for which smaller alignments would suffice. It may be defined as
  32. larger than this though. Note however that code and data structures
  33. are optimized for the case of 8-byte alignment.
  34. */
  35. #ifndef MALLOC_ALIGNMENT
  36. #define MALLOC_ALIGNMENT (2 * (sizeof(size_t)))
  37. #endif
  38. /* The corresponding bit mask value */
  39. #define MALLOC_ALIGN_MASK (MALLOC_ALIGNMENT - 1)
  40. /*
  41. TRIM_FASTBINS controls whether free() of a very small chunk can
  42. immediately lead to trimming. Setting to true (1) can reduce memory
  43. footprint, but will almost always slow down programs that use a lot
  44. of small chunks.
  45. Define this only if you are willing to give up some speed to more
  46. aggressively reduce system-level memory footprint when releasing
  47. memory in programs that use many small chunks. You can get
  48. essentially the same effect by setting MXFAST to 0, but this can
  49. lead to even greater slowdowns in programs using many small chunks.
  50. TRIM_FASTBINS is an in-between compile-time option, that disables
  51. only those chunks bordering topmost memory from being placed in
  52. fastbins.
  53. */
  54. #ifndef TRIM_FASTBINS
  55. #define TRIM_FASTBINS 0
  56. #endif
  57. /*
  58. MORECORE-related declarations. By default, rely on sbrk
  59. */
  60. /*
  61. MORECORE is the name of the routine to call to obtain more memory
  62. from the system. See below for general guidance on writing
  63. alternative MORECORE functions, as well as a version for WIN32 and a
  64. sample version for pre-OSX macos.
  65. */
  66. #ifndef MORECORE
  67. #define MORECORE sbrk
  68. #endif
  69. /*
  70. MORECORE_FAILURE is the value returned upon failure of MORECORE
  71. as well as mmap. Since it cannot be an otherwise valid memory address,
  72. and must reflect values of standard sys calls, you probably ought not
  73. try to redefine it.
  74. */
  75. #ifndef MORECORE_FAILURE
  76. #define MORECORE_FAILURE (-1)
  77. #endif
  78. /*
  79. If MORECORE_CONTIGUOUS is true, take advantage of fact that
  80. consecutive calls to MORECORE with positive arguments always return
  81. contiguous increasing addresses. This is true of unix sbrk. Even
  82. if not defined, when regions happen to be contiguous, malloc will
  83. permit allocations spanning regions obtained from different
  84. calls. But defining this when applicable enables some stronger
  85. consistency checks and space efficiencies.
  86. */
  87. #ifndef MORECORE_CONTIGUOUS
  88. #define MORECORE_CONTIGUOUS 1
  89. #endif
  90. /*
  91. MMAP_AS_MORECORE_SIZE is the minimum mmap size argument to use if
  92. sbrk fails, and mmap is used as a backup (which is done only if
  93. HAVE_MMAP). The value must be a multiple of page size. This
  94. backup strategy generally applies only when systems have "holes" in
  95. address space, so sbrk cannot perform contiguous expansion, but
  96. there is still space available on system. On systems for which
  97. this is known to be useful (i.e. most linux kernels), this occurs
  98. only when programs allocate huge amounts of memory. Between this,
  99. and the fact that mmap regions tend to be limited, the size should
  100. be large, to avoid too many mmap calls and thus avoid running out
  101. of kernel resources.
  102. */
  103. #ifndef MMAP_AS_MORECORE_SIZE
  104. #define MMAP_AS_MORECORE_SIZE (1024 * 1024)
  105. #endif
  106. /*
  107. The system page size. To the extent possible, this malloc manages
  108. memory from the system in page-size units. Note that this value is
  109. cached during initialization into a field of malloc_state. So even
  110. if malloc_getpagesize is a function, it is only called once.
  111. The following mechanics for getpagesize were adapted from bsd/gnu
  112. getpagesize.h. If none of the system-probes here apply, a value of
  113. 4096 is used, which should be OK: If they don't apply, then using
  114. the actual value probably doesn't impact performance.
  115. */
  116. #ifndef malloc_getpagesize
  117. # include <unistd.h>
  118. # define malloc_getpagesize sysconf(_SC_PAGESIZE)
  119. #else /* just guess */
  120. # define malloc_getpagesize (4096)
  121. #endif
  122. /* mallopt tuning options */
  123. /*
  124. M_MXFAST is the maximum request size used for "fastbins", special bins
  125. that hold returned chunks without consolidating their spaces. This
  126. enables future requests for chunks of the same size to be handled
  127. very quickly, but can increase fragmentation, and thus increase the
  128. overall memory footprint of a program.
  129. This malloc manages fastbins very conservatively yet still
  130. efficiently, so fragmentation is rarely a problem for values less
  131. than or equal to the default. The maximum supported value of MXFAST
  132. is 80. You wouldn't want it any higher than this anyway. Fastbins
  133. are designed especially for use with many small structs, objects or
  134. strings -- the default handles structs/objects/arrays with sizes up
  135. to 16 4byte fields, or small strings representing words, tokens,
  136. etc. Using fastbins for larger objects normally worsens
  137. fragmentation without improving speed.
  138. M_MXFAST is set in REQUEST size units. It is internally used in
  139. chunksize units, which adds padding and alignment. You can reduce
  140. M_MXFAST to 0 to disable all use of fastbins. This causes the malloc
  141. algorithm to be a closer approximation of fifo-best-fit in all cases,
  142. not just for larger requests, but will generally cause it to be
  143. slower.
  144. */
  145. /* M_MXFAST is a standard SVID/XPG tuning option, usually listed in malloc.h */
  146. #ifndef M_MXFAST
  147. #define M_MXFAST 1
  148. #endif
  149. #ifndef DEFAULT_MXFAST
  150. #define DEFAULT_MXFAST 64
  151. #endif
  152. /*
  153. M_TRIM_THRESHOLD is the maximum amount of unused top-most memory
  154. to keep before releasing via malloc_trim in free().
  155. Automatic trimming is mainly useful in long-lived programs.
  156. Because trimming via sbrk can be slow on some systems, and can
  157. sometimes be wasteful (in cases where programs immediately
  158. afterward allocate more large chunks) the value should be high
  159. enough so that your overall system performance would improve by
  160. releasing this much memory.
  161. The trim threshold and the mmap control parameters (see below)
  162. can be traded off with one another. Trimming and mmapping are
  163. two different ways of releasing unused memory back to the
  164. system. Between these two, it is often possible to keep
  165. system-level demands of a long-lived program down to a bare
  166. minimum. For example, in one test suite of sessions measuring
  167. the XF86 X server on Linux, using a trim threshold of 128K and a
  168. mmap threshold of 192K led to near-minimal long term resource
  169. consumption.
  170. If you are using this malloc in a long-lived program, it should
  171. pay to experiment with these values. As a rough guide, you
  172. might set to a value close to the average size of a process
  173. (program) running on your system. Releasing this much memory
  174. would allow such a process to run in memory. Generally, it's
  175. worth it to tune for trimming rather tham memory mapping when a
  176. program undergoes phases where several large chunks are
  177. allocated and released in ways that can reuse each other's
  178. storage, perhaps mixed with phases where there are no such
  179. chunks at all. And in well-behaved long-lived programs,
  180. controlling release of large blocks via trimming versus mapping
  181. is usually faster.
  182. However, in most programs, these parameters serve mainly as
  183. protection against the system-level effects of carrying around
  184. massive amounts of unneeded memory. Since frequent calls to
  185. sbrk, mmap, and munmap otherwise degrade performance, the default
  186. parameters are set to relatively high values that serve only as
  187. safeguards.
  188. The trim value must be greater than page size to have any useful
  189. effect. To disable trimming completely, you can set to
  190. (unsigned long)(-1)
  191. Trim settings interact with fastbin (MXFAST) settings: Unless
  192. TRIM_FASTBINS is defined, automatic trimming never takes place upon
  193. freeing a chunk with size less than or equal to MXFAST. Trimming is
  194. instead delayed until subsequent freeing of larger chunks. However,
  195. you can still force an attempted trim by calling malloc_trim.
  196. Also, trimming is not generally possible in cases where
  197. the main arena is obtained via mmap.
  198. Note that the trick some people use of mallocing a huge space and
  199. then freeing it at program startup, in an attempt to reserve system
  200. memory, doesn't have the intended effect under automatic trimming,
  201. since that memory will immediately be returned to the system.
  202. */
  203. #define M_TRIM_THRESHOLD -1
  204. #ifndef DEFAULT_TRIM_THRESHOLD
  205. #define DEFAULT_TRIM_THRESHOLD (256 * 1024)
  206. #endif
  207. /*
  208. M_TOP_PAD is the amount of extra `padding' space to allocate or
  209. retain whenever sbrk is called. It is used in two ways internally:
  210. * When sbrk is called to extend the top of the arena to satisfy
  211. a new malloc request, this much padding is added to the sbrk
  212. request.
  213. * When malloc_trim is called automatically from free(),
  214. it is used as the `pad' argument.
  215. In both cases, the actual amount of padding is rounded
  216. so that the end of the arena is always a system page boundary.
  217. The main reason for using padding is to avoid calling sbrk so
  218. often. Having even a small pad greatly reduces the likelihood
  219. that nearly every malloc request during program start-up (or
  220. after trimming) will invoke sbrk, which needlessly wastes
  221. time.
  222. Automatic rounding-up to page-size units is normally sufficient
  223. to avoid measurable overhead, so the default is 0. However, in
  224. systems where sbrk is relatively slow, it can pay to increase
  225. this value, at the expense of carrying around more memory than
  226. the program needs.
  227. */
  228. #define M_TOP_PAD -2
  229. #ifndef DEFAULT_TOP_PAD
  230. #define DEFAULT_TOP_PAD (0)
  231. #endif
  232. /*
  233. M_MMAP_THRESHOLD is the request size threshold for using mmap()
  234. to service a request. Requests of at least this size that cannot
  235. be allocated using already-existing space will be serviced via mmap.
  236. (If enough normal freed space already exists it is used instead.)
  237. Using mmap segregates relatively large chunks of memory so that
  238. they can be individually obtained and released from the host
  239. system. A request serviced through mmap is never reused by any
  240. other request (at least not directly; the system may just so
  241. happen to remap successive requests to the same locations).
  242. Segregating space in this way has the benefits that:
  243. 1. Mmapped space can ALWAYS be individually released back
  244. to the system, which helps keep the system level memory
  245. demands of a long-lived program low.
  246. 2. Mapped memory can never become `locked' between
  247. other chunks, as can happen with normally allocated chunks, which
  248. means that even trimming via malloc_trim would not release them.
  249. 3. On some systems with "holes" in address spaces, mmap can obtain
  250. memory that sbrk cannot.
  251. However, it has the disadvantages that:
  252. 1. The space cannot be reclaimed, consolidated, and then
  253. used to service later requests, as happens with normal chunks.
  254. 2. It can lead to more wastage because of mmap page alignment
  255. requirements
  256. 3. It causes malloc performance to be more dependent on host
  257. system memory management support routines which may vary in
  258. implementation quality and may impose arbitrary
  259. limitations. Generally, servicing a request via normal
  260. malloc steps is faster than going through a system's mmap.
  261. The advantages of mmap nearly always outweigh disadvantages for
  262. "large" chunks, but the value of "large" varies across systems. The
  263. default is an empirically derived value that works well in most
  264. systems.
  265. */
  266. #define M_MMAP_THRESHOLD -3
  267. #ifndef DEFAULT_MMAP_THRESHOLD
  268. #define DEFAULT_MMAP_THRESHOLD (256 * 1024)
  269. #endif
  270. /*
  271. M_MMAP_MAX is the maximum number of requests to simultaneously
  272. service using mmap. This parameter exists because
  273. . Some systems have a limited number of internal tables for
  274. use by mmap, and using more than a few of them may degrade
  275. performance.
  276. The default is set to a value that serves only as a safeguard.
  277. Setting to 0 disables use of mmap for servicing large requests. If
  278. HAVE_MMAP is not set, the default value is 0, and attempts to set it
  279. to non-zero values in mallopt will fail.
  280. */
  281. #define M_MMAP_MAX -4
  282. #ifndef DEFAULT_MMAP_MAX
  283. #define DEFAULT_MMAP_MAX (65536)
  284. #endif
  285. /* ------------------ MMAP support ------------------ */
  286. #include <fcntl.h>
  287. #include <sys/mman.h>
  288. #if !defined(MAP_ANONYMOUS) && defined(MAP_ANON)
  289. #define MAP_ANONYMOUS MAP_ANON
  290. #endif
  291. #ifdef __ARCH_HAS_MMU__
  292. #define MMAP(addr, size, prot) \
  293. (mmap((addr), (size), (prot), MAP_PRIVATE|MAP_ANONYMOUS, 0, 0))
  294. #else
  295. #define MMAP(addr, size, prot) \
  296. (mmap((addr), (size), (prot), MAP_SHARED|MAP_ANONYMOUS, 0, 0))
  297. #endif
  298. /* ----------------------- Chunk representations ----------------------- */
  299. /*
  300. This struct declaration is misleading (but accurate and necessary).
  301. It declares a "view" into memory allowing access to necessary
  302. fields at known offsets from a given base. See explanation below.
  303. */
  304. struct malloc_chunk {
  305. size_t prev_size; /* Size of previous chunk (if free). */
  306. size_t size; /* Size in bytes, including overhead. */
  307. struct malloc_chunk* fd; /* double links -- used only if free. */
  308. struct malloc_chunk* bk;
  309. };
  310. typedef struct malloc_chunk* mchunkptr;
  311. /*
  312. malloc_chunk details:
  313. (The following includes lightly edited explanations by Colin Plumb.)
  314. Chunks of memory are maintained using a `boundary tag' method as
  315. described in e.g., Knuth or Standish. (See the paper by Paul
  316. Wilson ftp://ftp.cs.utexas.edu/pub/garbage/allocsrv.ps for a
  317. survey of such techniques.) Sizes of free chunks are stored both
  318. in the front of each chunk and at the end. This makes
  319. consolidating fragmented chunks into bigger chunks very fast. The
  320. size fields also hold bits representing whether chunks are free or
  321. in use.
  322. An allocated chunk looks like this:
  323. chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  324. | Size of previous chunk, if allocated | |
  325. +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  326. | Size of chunk, in bytes |P|
  327. mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  328. | User data starts here... .
  329. . .
  330. . (malloc_usable_space() bytes) .
  331. . |
  332. nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  333. | Size of chunk |
  334. +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  335. Where "chunk" is the front of the chunk for the purpose of most of
  336. the malloc code, but "mem" is the pointer that is returned to the
  337. user. "Nextchunk" is the beginning of the next contiguous chunk.
  338. Chunks always begin on even word boundries, so the mem portion
  339. (which is returned to the user) is also on an even word boundary, and
  340. thus at least double-word aligned.
  341. Free chunks are stored in circular doubly-linked lists, and look like this:
  342. chunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  343. | Size of previous chunk |
  344. +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  345. `head:' | Size of chunk, in bytes |P|
  346. mem-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  347. | Forward pointer to next chunk in list |
  348. +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  349. | Back pointer to previous chunk in list |
  350. +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  351. | Unused space (may be 0 bytes long) .
  352. . .
  353. . |
  354. nextchunk-> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  355. `foot:' | Size of chunk, in bytes |
  356. +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  357. The P (PREV_INUSE) bit, stored in the unused low-order bit of the
  358. chunk size (which is always a multiple of two words), is an in-use
  359. bit for the *previous* chunk. If that bit is *clear*, then the
  360. word before the current chunk size contains the previous chunk
  361. size, and can be used to find the front of the previous chunk.
  362. The very first chunk allocated always has this bit set,
  363. preventing access to non-existent (or non-owned) memory. If
  364. prev_inuse is set for any given chunk, then you CANNOT determine
  365. the size of the previous chunk, and might even get a memory
  366. addressing fault when trying to do so.
  367. Note that the `foot' of the current chunk is actually represented
  368. as the prev_size of the NEXT chunk. This makes it easier to
  369. deal with alignments etc but can be very confusing when trying
  370. to extend or adapt this code.
  371. The two exceptions to all this are
  372. 1. The special chunk `top' doesn't bother using the
  373. trailing size field since there is no next contiguous chunk
  374. that would have to index off it. After initialization, `top'
  375. is forced to always exist. If it would become less than
  376. MINSIZE bytes long, it is replenished.
  377. 2. Chunks allocated via mmap, which have the second-lowest-order
  378. bit (IS_MMAPPED) set in their size fields. Because they are
  379. allocated one-by-one, each must contain its own trailing size field.
  380. */
  381. /*
  382. ---------- Size and alignment checks and conversions ----------
  383. */
  384. /* conversion from malloc headers to user pointers, and back */
  385. #define chunk2mem(p) ((void*)((char*)(p) + 2*(sizeof(size_t))))
  386. #define mem2chunk(mem) ((mchunkptr)((char*)(mem) - 2*(sizeof(size_t))))
  387. /* The smallest possible chunk */
  388. #define MIN_CHUNK_SIZE (sizeof(struct malloc_chunk))
  389. /* The smallest size we can malloc is an aligned minimal chunk */
  390. #define MINSIZE \
  391. (unsigned long)(((MIN_CHUNK_SIZE+MALLOC_ALIGN_MASK) & ~MALLOC_ALIGN_MASK))
  392. /* Check if m has acceptable alignment */
  393. #define aligned_OK(m) (((unsigned long)((m)) & (MALLOC_ALIGN_MASK)) == 0)
  394. /* Check if a request is so large that it would wrap around zero when
  395. padded and aligned. To simplify some other code, the bound is made
  396. low enough so that adding MINSIZE will also not wrap around sero.
  397. */
  398. #define REQUEST_OUT_OF_RANGE(req) \
  399. ((unsigned long)(req) >= \
  400. (unsigned long)(size_t)(-2 * MINSIZE))
  401. /* pad request bytes into a usable size -- internal version */
  402. #define request2size(req) \
  403. (((req) + (sizeof(size_t)) + MALLOC_ALIGN_MASK < MINSIZE) ? \
  404. MINSIZE : \
  405. ((req) + (sizeof(size_t)) + MALLOC_ALIGN_MASK) & ~MALLOC_ALIGN_MASK)
  406. /* Same, except also perform argument check */
  407. #define checked_request2size(req, sz) \
  408. if (REQUEST_OUT_OF_RANGE(req)) { \
  409. errno = ENOMEM; \
  410. return 0; \
  411. } \
  412. (sz) = request2size(req);
  413. /*
  414. --------------- Physical chunk operations ---------------
  415. */
  416. /* size field is or'ed with PREV_INUSE when previous adjacent chunk in use */
  417. #define PREV_INUSE 0x1
  418. /* extract inuse bit of previous chunk */
  419. #define prev_inuse(p) ((p)->size & PREV_INUSE)
  420. /* size field is or'ed with IS_MMAPPED if the chunk was obtained with mmap() */
  421. #define IS_MMAPPED 0x2
  422. /* check for mmap()'ed chunk */
  423. #define chunk_is_mmapped(p) ((p)->size & IS_MMAPPED)
  424. /* Bits to mask off when extracting size
  425. Note: IS_MMAPPED is intentionally not masked off from size field in
  426. macros for which mmapped chunks should never be seen. This should
  427. cause helpful core dumps to occur if it is tried by accident by
  428. people extending or adapting this malloc.
  429. */
  430. #define SIZE_BITS (PREV_INUSE|IS_MMAPPED)
  431. /* Get size, ignoring use bits */
  432. #define chunksize(p) ((p)->size & ~(SIZE_BITS))
  433. /* Ptr to next physical malloc_chunk. */
  434. #define next_chunk(p) ((mchunkptr)( ((char*)(p)) + ((p)->size & ~PREV_INUSE) ))
  435. /* Ptr to previous physical malloc_chunk */
  436. #define prev_chunk(p) ((mchunkptr)( ((char*)(p)) - ((p)->prev_size) ))
  437. /* Treat space at ptr + offset as a chunk */
  438. #define chunk_at_offset(p, s) ((mchunkptr)(((char*)(p)) + (s)))
  439. /* extract p's inuse bit */
  440. #define inuse(p)\
  441. ((((mchunkptr)(((char*)(p))+((p)->size & ~PREV_INUSE)))->size) & PREV_INUSE)
  442. /* set/clear chunk as being inuse without otherwise disturbing */
  443. #define set_inuse(p)\
  444. ((mchunkptr)(((char*)(p)) + ((p)->size & ~PREV_INUSE)))->size |= PREV_INUSE
  445. #define clear_inuse(p)\
  446. ((mchunkptr)(((char*)(p)) + ((p)->size & ~PREV_INUSE)))->size &= ~(PREV_INUSE)
  447. /* check/set/clear inuse bits in known places */
  448. #define inuse_bit_at_offset(p, s)\
  449. (((mchunkptr)(((char*)(p)) + (s)))->size & PREV_INUSE)
  450. #define set_inuse_bit_at_offset(p, s)\
  451. (((mchunkptr)(((char*)(p)) + (s)))->size |= PREV_INUSE)
  452. #define clear_inuse_bit_at_offset(p, s)\
  453. (((mchunkptr)(((char*)(p)) + (s)))->size &= ~(PREV_INUSE))
  454. /* Set size at head, without disturbing its use bit */
  455. #define set_head_size(p, s) ((p)->size = (((p)->size & PREV_INUSE) | (s)))
  456. /* Set size/use field */
  457. #define set_head(p, s) ((p)->size = (s))
  458. /* Set size at footer (only when chunk is not in use) */
  459. #define set_foot(p, s) (((mchunkptr)((char*)(p) + (s)))->prev_size = (s))
  460. /* -------------------- Internal data structures -------------------- */
  461. /*
  462. Bins
  463. An array of bin headers for free chunks. Each bin is doubly
  464. linked. The bins are approximately proportionally (log) spaced.
  465. There are a lot of these bins (128). This may look excessive, but
  466. works very well in practice. Most bins hold sizes that are
  467. unusual as malloc request sizes, but are more usual for fragments
  468. and consolidated sets of chunks, which is what these bins hold, so
  469. they can be found quickly. All procedures maintain the invariant
  470. that no consolidated chunk physically borders another one, so each
  471. chunk in a list is known to be preceeded and followed by either
  472. inuse chunks or the ends of memory.
  473. Chunks in bins are kept in size order, with ties going to the
  474. approximately least recently used chunk. Ordering isn't needed
  475. for the small bins, which all contain the same-sized chunks, but
  476. facilitates best-fit allocation for larger chunks. These lists
  477. are just sequential. Keeping them in order almost never requires
  478. enough traversal to warrant using fancier ordered data
  479. structures.
  480. Chunks of the same size are linked with the most
  481. recently freed at the front, and allocations are taken from the
  482. back. This results in LRU (FIFO) allocation order, which tends
  483. to give each chunk an equal opportunity to be consolidated with
  484. adjacent freed chunks, resulting in larger free chunks and less
  485. fragmentation.
  486. To simplify use in double-linked lists, each bin header acts
  487. as a malloc_chunk. This avoids special-casing for headers.
  488. But to conserve space and improve locality, we allocate
  489. only the fd/bk pointers of bins, and then use repositioning tricks
  490. to treat these as the fields of a malloc_chunk*.
  491. */
  492. typedef struct malloc_chunk* mbinptr;
  493. /* addressing -- note that bin_at(0) does not exist */
  494. #define bin_at(m, i) ((mbinptr)((char*)&((m)->bins[(i)<<1]) - ((sizeof(size_t))<<1)))
  495. /* analog of ++bin */
  496. #define next_bin(b) ((mbinptr)((char*)(b) + (sizeof(mchunkptr)<<1)))
  497. /* Reminders about list directionality within bins */
  498. #define first(b) ((b)->fd)
  499. #define last(b) ((b)->bk)
  500. /* Take a chunk off a bin list */
  501. #define unlink(P, BK, FD) { \
  502. FD = P->fd; \
  503. BK = P->bk; \
  504. FD->bk = BK; \
  505. BK->fd = FD; \
  506. }
  507. /*
  508. Indexing
  509. Bins for sizes < 512 bytes contain chunks of all the same size, spaced
  510. 8 bytes apart. Larger bins are approximately logarithmically spaced:
  511. 64 bins of size 8
  512. 32 bins of size 64
  513. 16 bins of size 512
  514. 8 bins of size 4096
  515. 4 bins of size 32768
  516. 2 bins of size 262144
  517. 1 bin of size what's left
  518. The bins top out around 1MB because we expect to service large
  519. requests via mmap.
  520. */
  521. #define NBINS 96
  522. #define NSMALLBINS 32
  523. #define SMALLBIN_WIDTH 8
  524. #define MIN_LARGE_SIZE 256
  525. #define in_smallbin_range(sz) \
  526. ((unsigned long)(sz) < (unsigned long)MIN_LARGE_SIZE)
  527. #define smallbin_index(sz) (((unsigned)(sz)) >> 3)
  528. #define bin_index(sz) \
  529. ((in_smallbin_range(sz)) ? smallbin_index(sz) : __malloc_largebin_index(sz))
  530. /*
  531. FIRST_SORTED_BIN_SIZE is the chunk size corresponding to the
  532. first bin that is maintained in sorted order. This must
  533. be the smallest size corresponding to a given bin.
  534. Normally, this should be MIN_LARGE_SIZE. But you can weaken
  535. best fit guarantees to sometimes speed up malloc by increasing value.
  536. Doing this means that malloc may choose a chunk that is
  537. non-best-fitting by up to the width of the bin.
  538. Some useful cutoff values:
  539. 512 - all bins sorted
  540. 2560 - leaves bins <= 64 bytes wide unsorted
  541. 12288 - leaves bins <= 512 bytes wide unsorted
  542. 65536 - leaves bins <= 4096 bytes wide unsorted
  543. 262144 - leaves bins <= 32768 bytes wide unsorted
  544. -1 - no bins sorted (not recommended!)
  545. */
  546. #define FIRST_SORTED_BIN_SIZE MIN_LARGE_SIZE
  547. /* #define FIRST_SORTED_BIN_SIZE 65536 */
  548. /*
  549. Unsorted chunks
  550. All remainders from chunk splits, as well as all returned chunks,
  551. are first placed in the "unsorted" bin. They are then placed
  552. in regular bins after malloc gives them ONE chance to be used before
  553. binning. So, basically, the unsorted_chunks list acts as a queue,
  554. with chunks being placed on it in free (and __malloc_consolidate),
  555. and taken off (to be either used or placed in bins) in malloc.
  556. */
  557. /* The otherwise unindexable 1-bin is used to hold unsorted chunks. */
  558. #define unsorted_chunks(M) (bin_at(M, 1))
  559. /*
  560. Top
  561. The top-most available chunk (i.e., the one bordering the end of
  562. available memory) is treated specially. It is never included in
  563. any bin, is used only if no other chunk is available, and is
  564. released back to the system if it is very large (see
  565. M_TRIM_THRESHOLD). Because top initially
  566. points to its own bin with initial zero size, thus forcing
  567. extension on the first malloc request, we avoid having any special
  568. code in malloc to check whether it even exists yet. But we still
  569. need to do so when getting memory from system, so we make
  570. initial_top treat the bin as a legal but unusable chunk during the
  571. interval between initialization and the first call to
  572. __malloc_alloc. (This is somewhat delicate, since it relies on
  573. the 2 preceding words to be zero during this interval as well.)
  574. */
  575. /* Conveniently, the unsorted bin can be used as dummy top on first call */
  576. #define initial_top(M) (unsorted_chunks(M))
  577. /*
  578. Binmap
  579. To help compensate for the large number of bins, a one-level index
  580. structure is used for bin-by-bin searching. `binmap' is a
  581. bitvector recording whether bins are definitely empty so they can
  582. be skipped over during during traversals. The bits are NOT always
  583. cleared as soon as bins are empty, but instead only
  584. when they are noticed to be empty during traversal in malloc.
  585. */
  586. /* Conservatively use 32 bits per map word, even if on 64bit system */
  587. #define BINMAPSHIFT 5
  588. #define BITSPERMAP (1U << BINMAPSHIFT)
  589. #define BINMAPSIZE (NBINS / BITSPERMAP)
  590. #define idx2block(i) ((i) >> BINMAPSHIFT)
  591. #define idx2bit(i) ((1U << ((i) & ((1U << BINMAPSHIFT)-1))))
  592. #define mark_bin(m,i) ((m)->binmap[idx2block(i)] |= idx2bit(i))
  593. #define unmark_bin(m,i) ((m)->binmap[idx2block(i)] &= ~(idx2bit(i)))
  594. #define get_binmap(m,i) ((m)->binmap[idx2block(i)] & idx2bit(i))
  595. /*
  596. Fastbins
  597. An array of lists holding recently freed small chunks. Fastbins
  598. are not doubly linked. It is faster to single-link them, and
  599. since chunks are never removed from the middles of these lists,
  600. double linking is not necessary. Also, unlike regular bins, they
  601. are not even processed in FIFO order (they use faster LIFO) since
  602. ordering doesn't much matter in the transient contexts in which
  603. fastbins are normally used.
  604. Chunks in fastbins keep their inuse bit set, so they cannot
  605. be consolidated with other free chunks. __malloc_consolidate
  606. releases all chunks in fastbins and consolidates them with
  607. other free chunks.
  608. */
  609. typedef struct malloc_chunk* mfastbinptr;
  610. /* offset 2 to use otherwise unindexable first 2 bins */
  611. #define fastbin_index(sz) ((((unsigned int)(sz)) >> 3) - 2)
  612. /* The maximum fastbin request size we support */
  613. #define MAX_FAST_SIZE 80
  614. #define NFASTBINS (fastbin_index(request2size(MAX_FAST_SIZE))+1)
  615. /*
  616. FASTBIN_CONSOLIDATION_THRESHOLD is the size of a chunk in free()
  617. that triggers automatic consolidation of possibly-surrounding
  618. fastbin chunks. This is a heuristic, so the exact value should not
  619. matter too much. It is defined at half the default trim threshold as a
  620. compromise heuristic to only attempt consolidation if it is likely
  621. to lead to trimming. However, it is not dynamically tunable, since
  622. consolidation reduces fragmentation surrounding loarge chunks even
  623. if trimming is not used.
  624. */
  625. #define FASTBIN_CONSOLIDATION_THRESHOLD \
  626. ((unsigned long)(DEFAULT_TRIM_THRESHOLD) >> 1)
  627. /*
  628. Since the lowest 2 bits in max_fast don't matter in size comparisons,
  629. they are used as flags.
  630. */
  631. /*
  632. ANYCHUNKS_BIT held in max_fast indicates that there may be any
  633. freed chunks at all. It is set true when entering a chunk into any
  634. bin.
  635. */
  636. #define ANYCHUNKS_BIT (1U)
  637. #define have_anychunks(M) (((M)->max_fast & ANYCHUNKS_BIT))
  638. #define set_anychunks(M) ((M)->max_fast |= ANYCHUNKS_BIT)
  639. #define clear_anychunks(M) ((M)->max_fast &= ~ANYCHUNKS_BIT)
  640. /*
  641. FASTCHUNKS_BIT held in max_fast indicates that there are probably
  642. some fastbin chunks. It is set true on entering a chunk into any
  643. fastbin, and cleared only in __malloc_consolidate.
  644. */
  645. #define FASTCHUNKS_BIT (2U)
  646. #define have_fastchunks(M) (((M)->max_fast & FASTCHUNKS_BIT))
  647. #define set_fastchunks(M) ((M)->max_fast |= (FASTCHUNKS_BIT|ANYCHUNKS_BIT))
  648. #define clear_fastchunks(M) ((M)->max_fast &= ~(FASTCHUNKS_BIT))
  649. /* Set value of max_fast. Use impossibly small value if 0. */
  650. #define set_max_fast(M, s) \
  651. (M)->max_fast = (((s) == 0)? SMALLBIN_WIDTH: request2size(s)) | \
  652. ((M)->max_fast & (FASTCHUNKS_BIT|ANYCHUNKS_BIT))
  653. #define get_max_fast(M) \
  654. ((M)->max_fast & ~(FASTCHUNKS_BIT | ANYCHUNKS_BIT))
  655. /*
  656. morecore_properties is a status word holding dynamically discovered
  657. or controlled properties of the morecore function
  658. */
  659. #define MORECORE_CONTIGUOUS_BIT (1U)
  660. #define contiguous(M) \
  661. (((M)->morecore_properties & MORECORE_CONTIGUOUS_BIT))
  662. #define noncontiguous(M) \
  663. (((M)->morecore_properties & MORECORE_CONTIGUOUS_BIT) == 0)
  664. #define set_contiguous(M) \
  665. ((M)->morecore_properties |= MORECORE_CONTIGUOUS_BIT)
  666. #define set_noncontiguous(M) \
  667. ((M)->morecore_properties &= ~MORECORE_CONTIGUOUS_BIT)
  668. /*
  669. ----------- Internal state representation and initialization -----------
  670. */
  671. struct malloc_state {
  672. /* The maximum chunk size to be eligible for fastbin */
  673. size_t max_fast; /* low 2 bits used as flags */
  674. /* Fastbins */
  675. mfastbinptr fastbins[NFASTBINS];
  676. /* Base of the topmost chunk -- not otherwise kept in a bin */
  677. mchunkptr top;
  678. /* The remainder from the most recent split of a small request */
  679. mchunkptr last_remainder;
  680. /* Normal bins packed as described above */
  681. mchunkptr bins[NBINS * 2];
  682. /* Bitmap of bins. Trailing zero map handles cases of largest binned size */
  683. unsigned int binmap[BINMAPSIZE+1];
  684. /* Tunable parameters */
  685. unsigned long trim_threshold;
  686. size_t top_pad;
  687. size_t mmap_threshold;
  688. /* Memory map support */
  689. int n_mmaps;
  690. int n_mmaps_max;
  691. int max_n_mmaps;
  692. /* Cache malloc_getpagesize */
  693. unsigned int pagesize;
  694. /* Track properties of MORECORE */
  695. unsigned int morecore_properties;
  696. /* Statistics */
  697. size_t mmapped_mem;
  698. size_t sbrked_mem;
  699. size_t max_sbrked_mem;
  700. size_t max_mmapped_mem;
  701. size_t max_total_mem;
  702. };
  703. typedef struct malloc_state *mstate;
  704. /*
  705. There is exactly one instance of this struct in this malloc.
  706. If you are adapting this malloc in a way that does NOT use a static
  707. malloc_state, you MUST explicitly zero-fill it before using. This
  708. malloc relies on the property that malloc_state is initialized to
  709. all zeroes (as is true of C statics).
  710. */
  711. extern struct malloc_state __malloc_state; /* never directly referenced */
  712. /*
  713. All uses of av_ are via get_malloc_state().
  714. At most one "call" to get_malloc_state is made per invocation of
  715. the public versions of malloc and free, but other routines
  716. that in turn invoke malloc and/or free may call more then once.
  717. Also, it is called in check* routines if __MALLOC_DEBUGGING is set.
  718. */
  719. #define get_malloc_state() (&(__malloc_state))
  720. /* External internal utilities operating on mstates */
  721. void __malloc_consolidate(mstate);
  722. /* Debugging support */
  723. #if ! __MALLOC_DEBUGGING
  724. #define check_chunk(P)
  725. #define check_free_chunk(P)
  726. #define check_inuse_chunk(P)
  727. #define check_remalloced_chunk(P,N)
  728. #define check_malloced_chunk(P,N)
  729. #define check_malloc_state()
  730. #define assert(x) ((void)0)
  731. #else
  732. #define check_chunk(P) __do_check_chunk(P)
  733. #define check_free_chunk(P) __do_check_free_chunk(P)
  734. #define check_inuse_chunk(P) __do_check_inuse_chunk(P)
  735. #define check_remalloced_chunk(P,N) __do_check_remalloced_chunk(P,N)
  736. #define check_malloced_chunk(P,N) __do_check_malloced_chunk(P,N)
  737. #define check_malloc_state() __do_check_malloc_state()
  738. extern void __do_check_chunk(mchunkptr p);
  739. extern void __do_check_free_chunk(mchunkptr p);
  740. extern void __do_check_inuse_chunk(mchunkptr p);
  741. extern void __do_check_remalloced_chunk(mchunkptr p, size_t s);
  742. extern void __do_check_malloced_chunk(mchunkptr p, size_t s);
  743. extern void __do_check_malloc_state(void);
  744. #include <assert.h>
  745. #endif