--- IO-AIO/AIO.xs 2006/10/21 23:20:29 1.44 +++ IO-AIO/AIO.xs 2006/10/24 00:34:47 1.66 @@ -1,4 +1,9 @@ +#if __linux +# define _GNU_SOURCE +#endif + #define _REENTRANT 1 + #include #include "EXTERN.h" @@ -11,6 +16,8 @@ #include #include +#include +#include #include #include #include @@ -41,10 +48,22 @@ #if __ia64 # define STACKSIZE 65536 +#elif __i386 || __x86_64 /* 16k is unreasonably high :( */ +# define STACKSIZE PTHREAD_STACK_MIN #else -# define STACKSIZE 8192 +# define STACKSIZE 16384 #endif +/* buffer size for various temporary buffers */ +#define AIO_BUFSIZE 65536 + +#define dBUF \ + char *aio_buf = malloc (AIO_BUFSIZE); \ + if (!aio_buf) \ + return -1; + +#define fBUF free (aio_buf) + enum { REQ_QUIT, REQ_OPEN, REQ_CLOSE, @@ -55,7 +74,8 @@ REQ_UNLINK, REQ_RMDIR, REQ_RENAME, REQ_READDIR, REQ_LINK, REQ_SYMLINK, - REQ_GROUP, + REQ_GROUP, REQ_NOP, + REQ_SLEEP, }; #define AIO_REQ_KLASS "IO::AIO::REQ" @@ -63,12 +83,8 @@ typedef struct aio_cb { - struct aio_cb *grp, *grp_prev, *grp_next; - struct aio_cb *volatile next; - SV *self; /* the perl counterpart of this request, if any */ - SV *data, *callback; SV *fh, *fh2; void *dataptr, *data2ptr; @@ -77,71 +93,278 @@ size_t length; ssize_t result; + STRLEN dataoffset; int type; int fd, fd2; int errorno; - STRLEN dataoffset; mode_t mode; /* open */ - unsigned char cancelled; + + unsigned char flags; + unsigned char pri; + + SV *self; /* the perl counterpart of this request, if any */ + struct aio_cb *grp, *grp_prev, *grp_next, *grp_first; } aio_cb; +enum { + FLAG_CANCELLED = 0x01, +}; + typedef aio_cb *aio_req; typedef aio_cb *aio_req_ornot; -typedef aio_cb *aio_group; + +enum { + PRI_MIN = -4, + PRI_MAX = 4, + + DEFAULT_PRI = 0, + PRI_BIAS = -PRI_MIN, +}; + +static int next_pri = DEFAULT_PRI + PRI_BIAS; static int started, wanted; static volatile int nreqs; static int max_outstanding = 1<<30; static int respipe [2]; -static pthread_mutex_t reslock = PTHREAD_MUTEX_INITIALIZER; -static pthread_mutex_t reqlock = PTHREAD_MUTEX_INITIALIZER; +#if __linux && defined (PTHREAD_ADAPTIVE_MUTEX_INITIALIZER_NP) +# define AIO_MUTEX_INIT PTHREAD_ADAPTIVE_MUTEX_INITIALIZER_NP +#else +# define AIO_MUTEX_INIT PTHREAD_MUTEX_INITIALIZER +#endif + +static pthread_mutex_t reslock = AIO_MUTEX_INIT; +static pthread_mutex_t reqlock = AIO_MUTEX_INIT; static pthread_cond_t reqwait = PTHREAD_COND_INITIALIZER; static volatile aio_req reqs, reqe; /* queue start, queue end */ static volatile aio_req ress, rese; /* queue start, queue end */ +static void req_invoke (aio_req req); +static void req_free (aio_req req); + /* must be called at most once */ static SV *req_sv (aio_req req, const char *klass) { - req->self = (SV *)newHV (); - sv_magic (req->self, 0, PERL_MAGIC_ext, (char *)req, 0); + if (!req->self) + { + req->self = (SV *)newHV (); + sv_magic (req->self, 0, PERL_MAGIC_ext, (char *)req, 0); + } - return sv_2mortal (sv_bless (newRV_noinc (req->self), gv_stashpv (klass, 1))); + return sv_2mortal (sv_bless (newRV_inc (req->self), gv_stashpv (klass, 1))); } -static aio_req SvAIO_REQ (SV *sv, const char *klass) +static aio_req SvAIO_REQ (SV *sv) { - if (!sv_derived_from (sv, klass) || !SvROK (sv)) - croak ("object of class %s expected", klass); + MAGIC *mg; + + if (!sv_derived_from (sv, AIO_REQ_KLASS) || !SvROK (sv)) + croak ("object of class " AIO_REQ_KLASS " expected"); - MAGIC *mg = mg_find (SvRV (sv), PERL_MAGIC_ext); + mg = mg_find (SvRV (sv), PERL_MAGIC_ext); return mg ? (aio_req)mg->mg_ptr : 0; } -static void req_free (aio_req req) +static void aio_grp_feed (aio_req grp) { - if (req->self) + while (grp->length < grp->fd2 && !(grp->flags & FLAG_CANCELLED)) { - sv_unmagic (req->self, PERL_MAGIC_ext); - SvREFCNT_dec (req->self); + int old_len = grp->length; + + if (grp->fh2 && SvOK (grp->fh2)) + { + dSP; + + ENTER; + SAVETMPS; + PUSHMARK (SP); + XPUSHs (req_sv (grp, AIO_GRP_KLASS)); + PUTBACK; + call_sv (grp->fh2, G_VOID | G_EVAL); + SPAGAIN; + FREETMPS; + LEAVE; + } + + /* stop if no progress has been made */ + if (old_len == grp->length) + { + SvREFCNT_dec (grp->fh2); + grp->fh2 = 0; + break; + } } +} - if (req->data) - SvREFCNT_dec (req->data); +static void aio_grp_dec (aio_req grp) +{ + --grp->length; - if (req->fh) - SvREFCNT_dec (req->fh); + /* call feeder, if applicable */ + aio_grp_feed (grp); - if (req->fh2) - SvREFCNT_dec (req->fh2); + /* finish, if done */ + if (!grp->length && grp->fd) + { + req_invoke (grp); + req_free (grp); + } +} + +static void poll_wait () +{ + fd_set rfd; + + while (nreqs) + { + aio_req req; +#if !(__i386 || __x86_64) /* safe without sempahore on this archs */ + pthread_mutex_lock (&reslock); +#endif + req = ress; +#if !(__i386 || __x86_64) /* safe without sempahore on this archs */ + pthread_mutex_unlock (&reslock); +#endif + + if (req) + return; + + FD_ZERO(&rfd); + FD_SET(respipe [0], &rfd); + + select (respipe [0] + 1, &rfd, 0, 0, 0); + } +} + +static void req_invoke (aio_req req) +{ + dSP; + int errorno = errno; + + if (req->flags & FLAG_CANCELLED || !SvOK (req->callback)) + return; + + errno = req->errorno; + + ENTER; + SAVETMPS; + PUSHMARK (SP); + EXTEND (SP, 1); + + switch (req->type) + { + case REQ_READDIR: + { + SV *rv = &PL_sv_undef; + + if (req->result >= 0) + { + char *buf = req->data2ptr; + AV *av = newAV (); + + while (req->result) + { + SV *sv = newSVpv (buf, 0); + + av_push (av, sv); + buf += SvCUR (sv) + 1; + req->result--; + } + + rv = sv_2mortal (newRV_noinc ((SV *)av)); + } + + PUSHs (rv); + } + break; + + case REQ_OPEN: + { + /* convert fd to fh */ + SV *fh; + + PUSHs (sv_2mortal (newSViv (req->result))); + PUTBACK; + call_pv ("IO::AIO::_fd2fh", G_SCALAR | G_EVAL); + SPAGAIN; + + fh = SvREFCNT_inc (POPs); + + PUSHMARK (SP); + XPUSHs (sv_2mortal (fh)); + } + break; + + case REQ_GROUP: + req->fd = 2; /* mark group as finished */ + + if (req->data) + { + int i; + AV *av = (AV *)req->data; + + EXTEND (SP, AvFILL (av) + 1); + for (i = 0; i <= AvFILL (av); ++i) + PUSHs (*av_fetch (av, i, 0)); + } + break; + + case REQ_NOP: + case REQ_SLEEP: + break; + + default: + PUSHs (sv_2mortal (newSViv (req->result))); + break; + } + + + PUTBACK; + call_sv (req->callback, G_VOID | G_EVAL); + SPAGAIN; + + FREETMPS; + LEAVE; + + errno = errorno; + + if (SvTRUE (ERRSV)) + { + req_free (req); + croak (0); + } +} + +static void req_free (aio_req req) +{ + if (req->grp) + { + aio_req grp = req->grp; + + /* unlink request */ + if (req->grp_next) req->grp_next->grp_prev = req->grp_prev; + if (req->grp_prev) req->grp_prev->grp_next = req->grp_next; + + if (grp->grp_first == req) + grp->grp_first = req->grp_next; + + aio_grp_dec (grp); + } - if (req->statdata) - Safefree (req->statdata); + if (req->self) + { + sv_unmagic (req->self, PERL_MAGIC_ext); + SvREFCNT_dec (req->self); + } - if (req->callback) - SvREFCNT_dec (req->callback); + SvREFCNT_dec (req->data); + SvREFCNT_dec (req->fh); + SvREFCNT_dec (req->fh2); + SvREFCNT_dec (req->callback); + Safefree (req->statdata); if (req->type == REQ_READDIR && req->result >= 0) free (req->data2ptr); @@ -149,21 +372,20 @@ Safefree (req); } -static void -poll_wait () +static void req_cancel (aio_req req) { - if (nreqs && !ress) + req->flags |= FLAG_CANCELLED; + + if (req->type == REQ_GROUP) { - fd_set rfd; - FD_ZERO(&rfd); - FD_SET(respipe [0], &rfd); + aio_req sub; - select (respipe [0] + 1, &rfd, 0, 0, 0); + for (sub = req->grp_first; sub; sub = sub->grp_next) + req_cancel (sub); } } -static int -poll_cb () +static int poll_cb () { dSP; int count = 0; @@ -195,15 +417,17 @@ if (!req) break; - nreqs--; + --nreqs; if (req->type == REQ_QUIT) started--; + else if (req->type == REQ_GROUP && req->length) + { + req->fd = 1; /* mark request as delayed */ + continue; + } else { - int errorno = errno; - errno = req->errorno; - if (req->type == REQ_READ) SvCUR_set (req->data, req->dataoffset + (req->result > 0 ? req->result : 0)); @@ -217,68 +441,8 @@ PL_statcache = *(req->statdata); } - ENTER; - PUSHMARK (SP); - - if (req->type == REQ_READDIR) - { - SV *rv = &PL_sv_undef; - - if (req->result >= 0) - { - char *buf = req->data2ptr; - AV *av = newAV (); - - while (req->result) - { - SV *sv = newSVpv (buf, 0); - - av_push (av, sv); - buf += SvCUR (sv) + 1; - req->result--; - } - - rv = sv_2mortal (newRV_noinc ((SV *)av)); - } - - XPUSHs (rv); - } - else - { - XPUSHs (sv_2mortal (newSViv (req->result))); - - if (req->type == REQ_OPEN) - { - /* convert fd to fh */ - SV *fh; - - PUTBACK; - call_pv ("IO::AIO::_fd2fh", G_SCALAR | G_EVAL); - SPAGAIN; - - fh = SvREFCNT_inc (POPs); + req_invoke (req); - PUSHMARK (SP); - XPUSHs (sv_2mortal (fh)); - } - } - - if (SvOK (req->callback) && !req->cancelled) - { - PUTBACK; - call_sv (req->callback, G_VOID | G_EVAL); - SPAGAIN; - - if (SvTRUE (ERRSV)) - { - req_free (req); - croak (0); - } - } - - LEAVE; - - errno = errorno; count++; } @@ -290,8 +454,7 @@ static void *aio_proc(void *arg); -static void -start_thread (void) +static void start_thread (void) { sigset_t fullsigset, oldsigset; pthread_t tid; @@ -310,13 +473,12 @@ sigprocmask (SIG_SETMASK, &oldsigset, 0); } -static void -req_send (aio_req req) +static void req_send (aio_req req) { while (started < wanted && nreqs >= started) start_thread (); - nreqs++; + ++nreqs; pthread_mutex_lock (&reqlock); @@ -345,8 +507,7 @@ } } -static void -end_thread (void) +static void end_thread (void) { aio_req req; Newz (0, req, 1, aio_cb); @@ -407,8 +568,7 @@ */ static pthread_mutex_t preadwritelock = PTHREAD_MUTEX_INITIALIZER; -static ssize_t -pread (int fd, void *buf, size_t count, off_t offset) +static ssize_t pread (int fd, void *buf, size_t count, off_t offset) { ssize_t res; off_t ooffset; @@ -423,8 +583,7 @@ return res; } -static ssize_t -pwrite (int fd, void *buf, size_t count, off_t offset) +static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset) { ssize_t res; off_t ooffset; @@ -447,20 +606,21 @@ #if !HAVE_READAHEAD # define readahead aio_readahead -static ssize_t -readahead (int fd, off_t offset, size_t count) +static ssize_t readahead (int fd, off_t offset, size_t count) { - char readahead_buf[4096]; + dBUF; while (count > 0) { - size_t len = count < sizeof (readahead_buf) ? count : sizeof (readahead_buf); + size_t len = count < AIO_BUFSIZE ? count : AIO_BUFSIZE; - pread (fd, readahead_buf, len, offset); + pread (fd, aio_buf, len, offset); offset += len; count -= len; } + fBUF; + errno = 0; } #endif @@ -470,8 +630,7 @@ static pthread_mutex_t readdirlock = PTHREAD_MUTEX_INITIALIZER; -static int -readdir_r (DIR *dirp, struct dirent *ent, struct dirent **res) +static int readdir_r (DIR *dirp, struct dirent *ent, struct dirent **res) { struct dirent *e; int errorno; @@ -497,8 +656,7 @@ #endif /* sendfile always needs emulation */ -static ssize_t -sendfile_ (int ofd, int ifd, off_t offset, size_t count) +static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count) { ssize_t res; @@ -557,14 +715,15 @@ ) { /* emulate sendfile. this is a major pain in the ass */ - char buf[4096]; + dBUF; + res = 0; while (count) { ssize_t cnt; - cnt = pread (ifd, buf, count > 4096 ? 4096 : count, offset); + cnt = pread (ifd, aio_buf, count > AIO_BUFSIZE ? AIO_BUFSIZE : count, offset); if (cnt <= 0) { @@ -572,7 +731,7 @@ break; } - cnt = write (ofd, buf, cnt); + cnt = write (ofd, aio_buf, cnt); if (cnt <= 0) { @@ -584,21 +743,22 @@ res += cnt; count -= cnt; } + + fBUF; } return res; } /* read a full directory */ -static int -scandir_ (const char *path, void **namesp) +static int scandir_ (const char *path, void **namesp) { - DIR *dirp = opendir (path); + DIR *dirp; union { struct dirent d; char b [offsetof (struct dirent, d_name) + NAME_MAX + 1]; - } u; + } *u; struct dirent *entp; char *name, *names; int memlen = 4096; @@ -606,40 +766,45 @@ int res = 0; int errorno; + dirp = opendir (path); if (!dirp) return -1; + u = malloc (sizeof (*u)); names = malloc (memlen); - for (;;) - { - errno = 0, readdir_r (dirp, &u.d, &entp); + if (u && names) + for (;;) + { + errno = 0; + readdir_r (dirp, &u->d, &entp); - if (!entp) - break; + if (!entp) + break; - name = entp->d_name; + name = entp->d_name; - if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) - { - int len = strlen (name) + 1; + if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) + { + int len = strlen (name) + 1; - res++; + res++; - while (memofs + len > memlen) - { - memlen *= 2; - names = realloc (names, memlen); - if (!names) - break; - } + while (memofs + len > memlen) + { + memlen *= 2; + names = realloc (names, memlen); + if (!names) + break; + } - memcpy (names + memofs, name, len); - memofs += len; - } - } + memcpy (names + memofs, name, len); + memofs += len; + } + } errorno = errno; + free (u); closedir (dirp); if (errorno) @@ -655,8 +820,7 @@ /*****************************************************************************/ -static void * -aio_proc (void *thr_arg) +static void *aio_proc (void *thr_arg) { aio_req req; int type; @@ -684,9 +848,10 @@ pthread_mutex_unlock (&reqlock); errno = 0; /* strictly unnecessary */ + type = req->type; /* remember type for QUIT check */ - if (!req->cancelled) - switch (req->type) + if (!(req->flags & FLAG_CANCELLED)) + switch (type) { case REQ_READ: req->result = pread (req->fd, req->dataptr, req->length, req->offset); break; case REQ_WRITE: req->result = pwrite (req->fd, req->dataptr, req->length, req->offset); break; @@ -710,6 +875,18 @@ case REQ_FSYNC: req->result = fsync (req->fd); break; case REQ_READDIR: req->result = scandir_ (req->dataptr, &req->data2ptr); break; + case REQ_SLEEP: + { + struct timeval tv; + + tv.tv_sec = req->fd; + tv.tv_usec = req->fd2; + + req->result = select (0, 0, 0, 0, &tv); + } + + case REQ_GROUP: + case REQ_NOP: case REQ_QUIT: break; @@ -803,6 +980,8 @@ #define dREQ \ aio_req req; \ + int req_pri = next_pri; \ + next_pri = DEFAULT_PRI + PRI_BIAS; \ \ if (SvOK (callback) && !SvROK (callback)) \ croak ("callback must be undef or of reference type"); \ @@ -811,7 +990,8 @@ if (!req) \ croak ("out of memory during aio_req allocation"); \ \ - req->callback = newSVsv (callback) + req->callback = newSVsv (callback); \ + req->pri = req_pri #define REQ_SEND \ req_send (req); \ @@ -1075,17 +1255,58 @@ } void +aio_sleep (delay,callback=&PL_sv_undef) + double delay + SV * callback + PPCODE: +{ + dREQ; + + req->type = REQ_SLEEP; + req->fd = delay < 0. ? 0 : delay; + req->fd2 = delay < 0. ? 0 : 1000. * (delay - req->fd); + + REQ_SEND; +} + +void aio_group (callback=&PL_sv_undef) SV * callback - PROTOTYPE: ;& + PROTOTYPE: ;$ PPCODE: { dREQ; + req->type = REQ_GROUP; + req_send (req); + XPUSHs (req_sv (req, AIO_GRP_KLASS)); } void +aio_nop (callback=&PL_sv_undef) + SV * callback + PPCODE: +{ + dREQ; + + req->type = REQ_NOP; + + REQ_SEND; +} + +#if 0 + +void +aio_pri (int pri = DEFAULT_PRI) + CODE: + if (pri < PRI_MIN) pri = PRI_MIN; + if (pri > PRI_MAX) pri = PRI_MAX; + next_pri = pri + PRI_BIAS; + +#endif + +void flush () PROTOTYPE: CODE: @@ -1136,11 +1357,86 @@ OUTPUT: RETVAL +PROTOTYPES: DISABLE + MODULE = IO::AIO PACKAGE = IO::AIO::REQ void cancel (aio_req_ornot req) - PROTOTYPE: CODE: - req->cancelled = 1; + req_cancel (req); + +void +cb (aio_req_ornot req, SV *callback=&PL_sv_undef) + CODE: + SvREFCNT_dec (req->callback); + req->callback = newSVsv (callback); + +MODULE = IO::AIO PACKAGE = IO::AIO::GRP + +void +add (aio_req grp, ...) + PPCODE: +{ + int i; + aio_req req; + + if (grp->fd == 2) + croak ("cannot add requests to IO::AIO::GRP after the group finished"); + + for (i = 1; i < items; ++i ) + { + if (GIMME_V != G_VOID) + XPUSHs (sv_2mortal (newSVsv (ST (i)))); + + req = SvAIO_REQ (ST (i)); + + if (req) + { + ++grp->length; + req->grp = grp; + + req->grp_prev = 0; + req->grp_next = grp->grp_first; + + if (grp->grp_first) + grp->grp_first->grp_prev = req; + + grp->grp_first = req; + } + } +} + +void +result (aio_req grp, ...) + CODE: +{ + int i; + AV *av = newAV (); + + for (i = 1; i < items; ++i ) + av_push (av, newSVsv (ST (i))); + + SvREFCNT_dec (grp->data); + grp->data = (SV *)av; +} + +void +feed_limit (aio_req grp, int limit) + CODE: + grp->fd2 = limit; + aio_grp_feed (grp); + +void +feed (aio_req grp, SV *callback=&PL_sv_undef) + CODE: +{ + SvREFCNT_dec (grp->fh2); + grp->fh2 = newSVsv (callback); + + if (grp->fd2 <= 0) + grp->fd2 = 2; + + aio_grp_feed (grp); +}