--- IO-AIO/AIO.xs 2006/10/30 23:30:00 1.86 +++ IO-AIO/AIO.xs 2009/07/17 00:56:54 1.155 @@ -1,12 +1,4 @@ -/* solaris */ -#define _POSIX_PTHREAD_SEMANTICS 1 - -#if __linux && !defined(_GNU_SOURCE) -# define _GNU_SOURCE -#endif - -/* just in case */ -#define _REENTRANT 1 +#include "libeio/xthread.h" #include @@ -14,316 +6,179 @@ #include "perl.h" #include "XSUB.h" -#include "autoconf/config.h" - -#include +#include "schmorp.h" #include +#include #include -#include -#include #include #include #include -#include #include -#include #include -#if HAVE_SENDFILE -# if __linux -# include -# elif __freebsd -# include -# include -# elif __hpux -# include -# elif __solaris /* not yet */ -# include -# else -# error sendfile support requested but not available -# endif -#endif +/* perl namespace pollution */ +#undef VERSION -/* number of seconds after which idle threads exit */ -#define IDLE_TIMEOUT 10 +#ifdef _WIN32 -/* used for struct dirent, AIX doesn't provide it */ -#ifndef NAME_MAX -# define NAME_MAX 4096 -#endif +# define EIO_STRUCT_DIRENT Direntry_t +# undef malloc +# undef free + +// perl overrides all those nice win32 functions +# undef open +# undef read +# undef write +# undef send +# undef recv +# undef stat +# undef fstat +# define lstat stat +# undef truncate +# undef ftruncate +# undef open +# undef close +# undef unlink +# undef rmdir +# undef rename +# undef lseek + +# define chown(a,b,c) (errno = ENOSYS, -1) +# define fchown(a,b,c) (errno = ENOSYS, -1) +# define fchmod(a,b) (errno = ENOSYS, -1) +# define symlink(a,b) (errno = ENOSYS, -1) +# define readlink(a,b,c) (errno = ENOSYS, -1) +# define mknod(a,b,c) (errno = ENOSYS, -1) +# define truncate(a,b) (errno = ENOSYS, -1) +# define ftruncate(fd,o) chsize ((fd), (o)) +# define fsync(fd) _commit (fd) +# define opendir(fd) (errno = ENOSYS, 0) +# define readdir(fd) (errno = ENOSYS, -1) +# define closedir(fd) (errno = ENOSYS, -1) +# define mkdir(a,b) mkdir (a) + +#else + +# include +# include +# include +# include +# include +# define EIO_STRUCT_DIRENT struct dirent -#ifndef PTHREAD_STACK_MIN -/* care for broken platforms, e.g. windows */ -# define PTHREAD_STACK_MIN 16384 #endif -#if __ia64 -# define STACKSIZE 65536 -#elif __i386 || __x86_64 /* 16k is unreasonably high :( */ -# define STACKSIZE PTHREAD_STACK_MIN +/* perl stupidly overrides readdir and maybe others */ +/* with thread-unsafe versions, imagine that :( */ +#undef readdir +#undef opendir +#undef closedir + +#define EIO_STRUCT_STAT Stat_t + +/* use NV for 32 bit perls as it allows larger offsets */ +#if IVSIZE >= 8 +# define VAL64 IV +# define SvVAL64 SvIV +# define newSVval64 newSViv #else -# define STACKSIZE 16384 +# define VAL64 NV +# define SvVAL64 SvNV +# define newSVval64 newSVnv #endif -/* wether word reads are potentially non-atomic. - * this is conservatice, likely most arches this runs - * on have atomic word read/writes. - */ -#ifndef WORDACCESS_UNSAFE -# if __i386 || __x86_64 -# define WORDACCESS_UNSAFE 0 -# else -# define WORDACCESS_UNSAFE 1 -# endif +/*****************************************************************************/ + +#if __GNUC__ >= 3 +# define expect(expr,value) __builtin_expect ((expr),(value)) +#else +# define expect(expr,value) (expr) #endif -/* buffer size for various temporary buffers */ -#define AIO_BUFSIZE 65536 +#define expect_false(expr) expect ((expr) != 0, 0) +#define expect_true(expr) expect ((expr) != 0, 1) -#define dBUF \ - char *aio_buf; \ - LOCK (wrklock); \ - self->dbuf = aio_buf = malloc (AIO_BUFSIZE); \ - UNLOCK (wrklock); \ - if (!aio_buf) \ - return -1; +/*****************************************************************************/ -enum { - REQ_QUIT, - REQ_OPEN, REQ_CLOSE, - REQ_READ, REQ_WRITE, REQ_READAHEAD, - REQ_SENDFILE, - REQ_STAT, REQ_LSTAT, REQ_FSTAT, - REQ_FSYNC, REQ_FDATASYNC, - REQ_UNLINK, REQ_RMDIR, REQ_RENAME, - REQ_MKNOD, REQ_READDIR, - REQ_LINK, REQ_SYMLINK, REQ_READLINK, - REQ_GROUP, REQ_NOP, - REQ_BUSY, -}; +static HV *stash; +typedef SV SV8; /* byte-sv, used for argument-checking */ +typedef int aio_rfd; /* read file desriptor */ +typedef int aio_wfd; /* write file descriptor */ #define AIO_REQ_KLASS "IO::AIO::REQ" #define AIO_GRP_KLASS "IO::AIO::GRP" -typedef struct aio_cb -{ - struct aio_cb *volatile next; - - SV *callback, *fh; - SV *sv1, *sv2; - void *ptr1, *ptr2; - Stat_t *statdata; - off_t offs; - size_t size; - ssize_t result; - - STRLEN stroffset; - int type; - int int1, int2; - int errorno; - mode_t mode; /* open */ - - 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, - FLAG_DATA_RO_OFF = 0x80, /* data was set readonly */ -}; - -typedef aio_cb *aio_req; -typedef aio_cb *aio_req_ornot; - -enum { - PRI_MIN = -4, - PRI_MAX = 4, - - DEFAULT_PRI = 0, - PRI_BIAS = -PRI_MIN, - NUM_PRI = PRI_MAX + PRI_BIAS + 1, -}; - -#define AIO_TICKS ((1000000 + 1023) >> 10) - -static unsigned int max_poll_time = 0; -static unsigned int max_poll_reqs = 0; - -/* calculcate time difference in ~1/AIO_TICKS of a second */ -static int tvdiff (struct timeval *tv1, struct timeval *tv2) -{ - return (tv2->tv_sec - tv1->tv_sec ) * AIO_TICKS - + ((tv2->tv_usec - tv1->tv_usec) >> 10); -} - -static int next_pri = DEFAULT_PRI + PRI_BIAS; - -static unsigned int started, idle, wanted; - -#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 +#define EIO_REQ_MEMBERS \ + SV *callback; \ + SV *sv1, *sv2; \ + STRLEN stroffset; \ + SV *self; + +#define EIO_NO_WRAPPERS 1 + +#include "libeio/eio.h" + +#ifndef POSIX_FADV_NORMAL +# define POSIX_FADV_NORMAL 0 +# define NO_FADVISE 1 #endif -#define LOCK(mutex) pthread_mutex_lock (&(mutex)) -#define UNLOCK(mutex) pthread_mutex_unlock (&(mutex)) - -/* worker threads management */ -static pthread_mutex_t wrklock = AIO_MUTEX_INIT; - -typedef struct worker { - /* locked by wrklock */ - struct worker *prev, *next; - - pthread_t tid; - - /* locked by reslock, reqlock or wrklock */ - aio_req req; /* currently processed request */ - void *dbuf; - DIR *dirp; -} worker; - -static worker wrk_first = { &wrk_first, &wrk_first, 0 }; - -static void worker_clear (worker *wrk) -{ - if (wrk->dirp) - { - closedir (wrk->dirp); - wrk->dirp = 0; - } - - if (wrk->dbuf) - { - free (wrk->dbuf); - wrk->dbuf = 0; - } -} - -static void worker_free (worker *wrk) -{ - wrk->next->prev = wrk->prev; - wrk->prev->next = wrk->next; - - free (wrk); -} - -static volatile unsigned int nreqs, nready, npending; -static volatile unsigned int max_idle = 4; -static volatile unsigned int max_outstanding = 0xffffffff; -static int respipe [2]; - -static pthread_mutex_t reslock = AIO_MUTEX_INIT; -static pthread_mutex_t reqlock = AIO_MUTEX_INIT; -static pthread_cond_t reqwait = PTHREAD_COND_INITIALIZER; - -#if WORDACCESS_UNSAFE - -static unsigned int get_nready () -{ - unsigned int retval; - - LOCK (reqlock); - retval = nready; - UNLOCK (reqlock); +#ifndef POSIX_FADV_SEQUENTIAL +# define POSIX_FADV_SEQUENTIAL 0 +#endif - return retval; -} +#ifndef POSIX_FADV_RANDOM +# define POSIX_FADV_RANDOM 0 +#endif -static unsigned int get_npending () -{ - unsigned int retval; +#ifndef POSIX_FADV_NOREUSE +# define POSIX_FADV_NOREUSE 0 +#endif - LOCK (reslock); - retval = npending; - UNLOCK (reslock); +#ifndef POSIX_FADV_WILLNEED +# define POSIX_FADV_WILLNEED 0 +#endif - return retval; -} +#ifndef POSIX_FADV_DONTNEED +# define POSIX_FADV_DONTNEED 0 +#endif -static unsigned int get_nthreads () -{ - unsigned int retval; +static int req_invoke (eio_req *req); +#define EIO_FINISH(req) req_invoke (req) +static void req_destroy (eio_req *grp); +#define EIO_DESTROY(req) req_destroy (req) - LOCK (wrklock); - retval = started; - UNLOCK (wrklock); +enum { + FLAG_SV2_RO_OFF = 0x40, /* data was set readonly */ +}; - return retval; -} +#include "libeio/eio.c" -#else +typedef eio_req *aio_req; +typedef eio_req *aio_req_ornot; -# define get_nready() nready -# define get_npending() npending -# define get_nthreads() started +static SV *on_next_submit; +static int next_pri = EIO_PRI_DEFAULT; +static int max_outstanding; -#endif +static s_epipe respipe; -/* - * a somewhat faster data structure might be nice, but - * with 8 priorities this actually needs <20 insns - * per shift, the most expensive operation. - */ -typedef struct { - aio_req qs[NUM_PRI], qe[NUM_PRI]; /* qstart, qend */ - int size; -} reqq; - -static reqq req_queue; -static reqq res_queue; +static void req_destroy (aio_req req); +static void req_cancel (aio_req req); -int reqq_push (reqq *q, aio_req req) +static void want_poll (void) { - int pri = req->pri; - req->next = 0; - - if (q->qe[pri]) - { - q->qe[pri]->next = req; - q->qe[pri] = req; - } - else - q->qe[pri] = q->qs[pri] = req; - - return q->size++; + /* write a dummy byte to the pipe so fh becomes ready */ + s_epipe_signal (&respipe); } -aio_req reqq_shift (reqq *q) +static void done_poll (void) { - int pri; - - if (!q->size) - return 0; - - --q->size; - - for (pri = NUM_PRI; pri--; ) - { - aio_req req = q->qs[pri]; - - if (req) - { - if (!(q->qs[pri] = req->next)) - q->qe[pri] = 0; - - return req; - } - } - - abort (); + /* read any signals sent by the worker threads */ + s_epipe_drain (&respipe); } -static int poll_cb (); -static void req_invoke (aio_req req); -static void req_free (aio_req req); -static void req_cancel (aio_req req); - /* must be called at most once */ static SV *req_sv (aio_req req, const char *klass) { @@ -350,65 +205,47 @@ static void aio_grp_feed (aio_req grp) { - while (grp->size < grp->int2 && !(grp->flags & FLAG_CANCELLED)) + if (grp->sv2 && SvOK (grp->sv2)) { - int old_len = grp->size; + dSP; - if (grp->sv2 && SvOK (grp->sv2)) - { - dSP; - - ENTER; - SAVETMPS; - PUSHMARK (SP); - XPUSHs (req_sv (grp, AIO_GRP_KLASS)); - PUTBACK; - call_sv (grp->sv2, G_VOID | G_EVAL | G_KEEPERR); - SPAGAIN; - FREETMPS; - LEAVE; - } - - /* stop if no progress has been made */ - if (old_len == grp->size) - { - SvREFCNT_dec (grp->sv2); - grp->sv2 = 0; - break; - } + ENTER; + SAVETMPS; + PUSHMARK (SP); + XPUSHs (req_sv (grp, AIO_GRP_KLASS)); + PUTBACK; + call_sv (grp->sv2, G_VOID | G_EVAL | G_KEEPERR); + SPAGAIN; + FREETMPS; + LEAVE; } } -static void aio_grp_dec (aio_req grp) +static void req_submit (eio_req *req) { - --grp->size; - - /* call feeder, if applicable */ - aio_grp_feed (grp); + eio_submit (req); - /* finish, if done */ - if (!grp->size && grp->int1) + if (expect_false (on_next_submit)) { - req_invoke (grp); - req_free (grp); + dSP; + SV *cb = sv_2mortal (on_next_submit); + + on_next_submit = 0; + + PUSHMARK (SP); + PUTBACK; + call_sv (cb, G_DISCARD | G_EVAL); } } -static void req_invoke (aio_req req) +static int req_invoke (eio_req *req) { dSP; - if (req->flags & FLAG_DATA_RO_OFF) - SvREADONLY_off (req->sv1); - - if (req->statdata) - { - PL_laststype = req->type == REQ_LSTAT ? OP_LSTAT : OP_STAT; - PL_laststatval = req->result; - PL_statcache = *(req->statdata); - } + if (req->flags & FLAG_SV2_RO_OFF) + SvREADONLY_off (req->sv2); - if (!(req->flags & FLAG_CANCELLED) && SvOK (req->callback)) + if (!EIO_CANCELLED (req) && req->callback) { ENTER; SAVETMPS; @@ -417,51 +254,99 @@ switch (req->type) { - case REQ_READDIR: + case EIO_READDIR: { SV *rv = &PL_sv_undef; if (req->result >= 0) { int i; - char *buf = req->ptr2; + char *names = (char *)req->ptr2; + eio_dirent *ent = (eio_dirent *)req->ptr1; /* might be 0 */ AV *av = newAV (); av_extend (av, req->result - 1); for (i = 0; i < req->result; ++i) { - SV *sv = newSVpv (buf, 0); - - av_store (av, i, sv); - buf += SvCUR (sv) + 1; + if (req->int1 & EIO_READDIR_DENTS) + { + SV *namesv = newSVpvn (names + ent->nameofs, ent->namelen); + + if (req->int1 & EIO_READDIR_CUSTOM2) + { + static SV *sv_type [EIO_DT_MAX + 1]; /* type sv cache */ + AV *avent = newAV (); + + av_extend (avent, 2); + + if (!sv_type [ent->type]) + { + sv_type [ent->type] = newSViv (ent->type); + SvREADONLY_on (sv_type [ent->type]); + } + + av_store (avent, 0, namesv); + av_store (avent, 1, SvREFCNT_inc (sv_type [ent->type])); + av_store (avent, 2, IVSIZE >= 8 ? newSVuv (ent->inode) : newSVnv (ent->inode)); + + av_store (av, i, newRV_noinc ((SV *)avent)); + } + else + av_store (av, i, namesv); + + ++ent; + } + else + { + SV *name = newSVpv (names, 0); + av_store (av, i, name); + names += SvCUR (name) + 1; + } } rv = sv_2mortal (newRV_noinc ((SV *)av)); } PUSHs (rv); + + if (req->int1 & EIO_READDIR_CUSTOM1) + XPUSHs (sv_2mortal (newSViv (req->int1 & ~(EIO_READDIR_CUSTOM1 | EIO_READDIR_CUSTOM2)))); } break; - case REQ_OPEN: + case EIO_OPEN: { /* convert fd to fh */ - SV *fh; + SV *fh = &PL_sv_undef; - PUSHs (sv_2mortal (newSViv (req->result))); - PUTBACK; - call_pv ("IO::AIO::_fd2fh", G_SCALAR | G_EVAL); - SPAGAIN; + if (req->result >= 0) + { + GV *gv = (GV *)sv_newmortal (); + int flags = req->int1 & (O_RDONLY | O_WRONLY | O_RDWR); + char sym [64]; + int symlen; + + symlen = snprintf (sym, sizeof (sym), "fd#%d", (int)req->result); + gv_init (gv, stash, sym, symlen, 0); + + symlen = snprintf ( + sym, + sizeof (sym), + "%s&=%d", + flags == O_RDONLY ? "<" : flags == O_WRONLY ? ">" : "+<", + (int)req->result + ); - fh = SvREFCNT_inc (POPs); + if (do_open (gv, sym, symlen, 0, 0, 0, 0)) + fh = (SV *)gv; + } - PUSHMARK (SP); - XPUSHs (sv_2mortal (fh)); + PUSHs (fh); } break; - case REQ_GROUP: + case EIO_GROUP: req->int1 = 2; /* mark group as finished */ if (req->sv1) @@ -475,23 +360,39 @@ } break; - case REQ_NOP: - case REQ_BUSY: + case EIO_NOP: + case EIO_BUSY: break; - case REQ_READLINK: + case EIO_READLINK: if (req->result > 0) - { - SvCUR_set (req->sv1, req->result); - *SvEND (req->sv1) = 0; - PUSHs (req->sv1); - } + PUSHs (sv_2mortal (newSVpvn (req->ptr2, req->result))); + break; + + case EIO_STAT: + case EIO_LSTAT: + case EIO_FSTAT: + PL_laststype = req->type == EIO_LSTAT ? OP_LSTAT : OP_STAT; + PL_laststatval = req->result; + PL_statcache = *(EIO_STRUCT_STAT *)(req->ptr2); + PUSHs (sv_2mortal (newSViv (req->result))); break; - case REQ_READ: - SvCUR_set (req->sv1, req->stroffset + (req->result > 0 ? req->result : 0)); - *SvEND (req->sv1) = 0; - /* fall through */ + case EIO_READ: + { + SvCUR_set (req->sv2, req->stroffset + (req->result > 0 ? req->result : 0)); + *SvEND (req->sv2) = 0; + SvPOK_only (req->sv2); + SvSETMAGIC (req->sv2); + PUSHs (sv_2mortal (newSViv (req->result))); + } + break; + + case EIO_DUP2: + if (req->result > 0) + req->result = 0; + /* FALLTHROUGH */ + default: PUSHs (sv_2mortal (newSViv (req->result))); break; @@ -500,35 +401,19 @@ errno = req->errorno; PUTBACK; - call_sv (req->callback, G_VOID | G_EVAL); + call_sv (req->callback, G_VOID | G_EVAL | G_DISCARD); SPAGAIN; FREETMPS; LEAVE; - } - 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); + PUTBACK; } - if (SvTRUE (ERRSV)) - { - req_free (req); - croak (0); - } + return !!SvTRUE (ERRSV); } -static void req_free (aio_req req) +static void req_destroy (aio_req req) { if (req->self) { @@ -536,14 +421,9 @@ SvREFCNT_dec (req->self); } - SvREFCNT_dec (req->fh); SvREFCNT_dec (req->sv1); SvREFCNT_dec (req->sv2); SvREFCNT_dec (req->callback); - Safefree (req->statdata); - - if (req->type == REQ_READDIR) - free (req->ptr2); Safefree (req); } @@ -552,857 +432,319 @@ { aio_req sub; - if (grp->type != REQ_GROUP) + if (grp->type != EIO_GROUP) return; SvREFCNT_dec (grp->sv2); grp->sv2 = 0; - for (sub = grp->grp_first; sub; sub = sub->grp_next) - req_cancel (sub); -} - -static void req_cancel (aio_req req) -{ - req->flags |= FLAG_CANCELLED; - - req_cancel_subs (req); -} - -static void *aio_proc(void *arg); - -static void start_thread (void) -{ - sigset_t fullsigset, oldsigset; - pthread_attr_t attr; - - worker *wrk = calloc (1, sizeof (worker)); - - if (!wrk) - croak ("unable to allocate worker thread data"); - - pthread_attr_init (&attr); - pthread_attr_setstacksize (&attr, STACKSIZE); - pthread_attr_setdetachstate (&attr, PTHREAD_CREATE_DETACHED); -#ifdef PTHREAD_SCOPE_PROCESS - pthread_attr_setscope (&attr, PTHREAD_SCOPE_PROCESS); -#endif - - sigfillset (&fullsigset); - - LOCK (wrklock); - sigprocmask (SIG_SETMASK, &fullsigset, &oldsigset); - - if (pthread_create (&wrk->tid, &attr, aio_proc, (void *)wrk) == 0) - { - wrk->prev = &wrk_first; - wrk->next = wrk_first.next; - wrk_first.next->prev = wrk; - wrk_first.next = wrk; - ++started; - } - else - free (wrk); - - sigprocmask (SIG_SETMASK, &oldsigset, 0); - UNLOCK (wrklock); -} - -static void maybe_start_thread () -{ - if (get_nthreads () >= wanted) - return; - - /* todo: maybe use idle here, but might be less exact */ - if (0 <= (int)get_nthreads () + (int)get_npending () - (int)nreqs) - return; - - start_thread (); + eio_grp_cancel (grp); } -static void req_send (aio_req req) +static void +create_respipe (void) { - ++nreqs; - - LOCK (reqlock); - ++nready; - reqq_push (&req_queue, req); - pthread_cond_signal (&reqwait); - UNLOCK (reqlock); - - maybe_start_thread (); -} - -static void end_thread (void) -{ - aio_req req; - - Newz (0, req, 1, aio_cb); - - req->type = REQ_QUIT; - req->pri = PRI_MAX + PRI_BIAS; - - LOCK (reqlock); - reqq_push (&req_queue, req); - pthread_cond_signal (&reqwait); - UNLOCK (reqlock); - - LOCK (wrklock); - --started; - UNLOCK (wrklock); -} - -static void set_max_idle (int nthreads) -{ - if (WORDACCESS_UNSAFE) LOCK (reqlock); - max_idle = nthreads <= 0 ? 1 : nthreads; - if (WORDACCESS_UNSAFE) UNLOCK (reqlock); -} - -static void min_parallel (int nthreads) -{ - if (wanted < nthreads) - wanted = nthreads; -} - -static void max_parallel (int nthreads) -{ - if (wanted > nthreads) - wanted = nthreads; - - while (started > wanted) - end_thread (); + if (s_epipe_renew (&respipe)) + croak ("IO::AIO: unable to initialize result pipe"); } -static void poll_wait () +static void poll_wait (void) { - fd_set rfd; - - while (nreqs) + while (eio_nreqs ()) { int size; - if (WORDACCESS_UNSAFE) LOCK (reslock); + + X_LOCK (reslock); size = res_queue.size; - if (WORDACCESS_UNSAFE) UNLOCK (reslock); + X_UNLOCK (reslock); if (size) return; - maybe_start_thread (); - - FD_ZERO(&rfd); - FD_SET(respipe [0], &rfd); + etp_maybe_start_thread (); - select (respipe [0] + 1, &rfd, 0, 0, 0); + s_epipe_wait (&respipe); } } -static int poll_cb () +static int poll_cb (void) { - dSP; - int count = 0; - int maxreqs = max_poll_reqs; - int do_croak = 0; - struct timeval tv_start, tv_now; - aio_req req; - - if (max_poll_time) - gettimeofday (&tv_start, 0); - for (;;) { - for (;;) - { - maybe_start_thread (); - - LOCK (reslock); - req = reqq_shift (&res_queue); - - if (req) - { - --npending; + int res = eio_poll (); - if (!res_queue.size) - { - /* read any signals sent by the worker threads */ - char buf [32]; - while (read (respipe [0], buf, 32) == 32) - ; - } - } + if (res > 0) + croak (0); - UNLOCK (reslock); - - if (!req) - break; - - --nreqs; - - if (req->type == REQ_GROUP && req->size) - { - req->int1 = 1; /* mark request as delayed */ - continue; - } - else - { - req_invoke (req); - - count++; - } - - req_free (req); - - if (maxreqs && !--maxreqs) - break; - - if (max_poll_time) - { - gettimeofday (&tv_now, 0); - - if (tvdiff (&tv_start, &tv_now) >= max_poll_time) - break; - } - } - - if (nreqs <= max_outstanding) - break; + if (!max_outstanding || max_outstanding > eio_nreqs ()) + return res; poll_wait (); - - ++maxreqs; - } - - return count; -} - -static void create_pipe () -{ - if (pipe (respipe)) - croak ("unable to initialize result pipe"); - - if (fcntl (respipe [0], F_SETFL, O_NONBLOCK)) - croak ("cannot set result pipe to nonblocking mode"); - - if (fcntl (respipe [1], F_SETFL, O_NONBLOCK)) - croak ("cannot set result pipe to nonblocking mode"); -} - -/*****************************************************************************/ -/* work around various missing functions */ - -#if !HAVE_PREADWRITE -# define pread aio_pread -# define pwrite aio_pwrite - -/* - * make our pread/pwrite safe against themselves, but not against - * normal read/write by using a mutex. slows down execution a lot, - * but that's your problem, not mine. - */ -static pthread_mutex_t preadwritelock = PTHREAD_MUTEX_INITIALIZER; - -static ssize_t pread (int fd, void *buf, size_t count, off_t offset) -{ - ssize_t res; - off_t ooffset; - - LOCK (preadwritelock); - ooffset = lseek (fd, 0, SEEK_CUR); - lseek (fd, offset, SEEK_SET); - res = read (fd, buf, count); - lseek (fd, ooffset, SEEK_SET); - UNLOCK (preadwritelock); - - return res; -} - -static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset) -{ - ssize_t res; - off_t ooffset; - - LOCK (preadwritelock); - ooffset = lseek (fd, 0, SEEK_CUR); - lseek (fd, offset, SEEK_SET); - res = write (fd, buf, count); - lseek (fd, offset, SEEK_SET); - UNLOCK (preadwritelock); - - return res; -} -#endif - -#if !HAVE_FDATASYNC -# define fdatasync fsync -#endif - -#if !HAVE_READAHEAD -# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) - -static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) -{ - dBUF; - - while (count > 0) - { - size_t len = count < AIO_BUFSIZE ? count : AIO_BUFSIZE; - - pread (fd, aio_buf, len, offset); - offset += len; - count -= len; - } - - errno = 0; -} - -#endif - -#if !HAVE_READDIR_R -# define readdir_r aio_readdir_r - -static pthread_mutex_t readdirlock = PTHREAD_MUTEX_INITIALIZER; - -static int readdir_r (DIR *dirp, struct dirent *ent, struct dirent **res) -{ - struct dirent *e; - int errorno; - - LOCK (readdirlock); - - e = readdir (dirp); - errorno = errno; - - if (e) - { - *res = ent; - strcpy (ent->d_name, e->d_name); - } - else - *res = 0; - - UNLOCK (readdirlock); - - errno = errorno; - return e ? 0 : -1; -} -#endif - -/* sendfile always needs emulation */ -static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) -{ - ssize_t res; - - if (!count) - return 0; - -#if HAVE_SENDFILE -# if __linux - res = sendfile (ofd, ifd, &offset, count); - -# elif __freebsd - /* - * Of course, the freebsd sendfile is a dire hack with no thoughts - * wasted on making it similar to other I/O functions. - */ - { - off_t sbytes; - res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); - - if (res < 0 && sbytes) - /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ - res = sbytes; - } - -# elif __hpux - res = sendfile (ofd, ifd, offset, count, 0, 0); - -# elif __solaris - { - struct sendfilevec vec; - size_t sbytes; - - vec.sfv_fd = ifd; - vec.sfv_flag = 0; - vec.sfv_off = offset; - vec.sfv_len = count; - - res = sendfilev (ofd, &vec, 1, &sbytes); - - if (res < 0 && sbytes) - res = sbytes; - } - -# endif -#else - res = -1; - errno = ENOSYS; -#endif - - if (res < 0 - && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK -#if __solaris - || errno == EAFNOSUPPORT || errno == EPROTOTYPE -#endif - ) - ) - { - /* emulate sendfile. this is a major pain in the ass */ - dBUF; - - res = 0; - - while (count) - { - ssize_t cnt; - - cnt = pread (ifd, aio_buf, count > AIO_BUFSIZE ? AIO_BUFSIZE : count, offset); - - if (cnt <= 0) - { - if (cnt && !res) res = -1; - break; - } - - cnt = write (ofd, aio_buf, cnt); - - if (cnt <= 0) - { - if (cnt && !res) res = -1; - break; - } - - offset += cnt; - res += cnt; - count -= cnt; - } - } - - return res; -} - -/* read a full directory */ -static void scandir_ (aio_req req, worker *self) -{ - DIR *dirp; - union - { - struct dirent d; - char b [offsetof (struct dirent, d_name) + NAME_MAX + 1]; - } *u; - struct dirent *entp; - char *name, *names; - int memlen = 4096; - int memofs = 0; - int res = 0; - int errorno; - - LOCK (wrklock); - self->dirp = dirp = opendir (req->ptr1); - self->dbuf = u = malloc (sizeof (*u)); - req->ptr2 = names = malloc (memlen); - UNLOCK (wrklock); - - if (dirp && u && names) - for (;;) - { - errno = 0; - readdir_r (dirp, &u->d, &entp); - - if (!entp) - break; - - name = entp->d_name; - - if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) - { - int len = strlen (name) + 1; - - res++; - - while (memofs + len > memlen) - { - memlen *= 2; - LOCK (wrklock); - req->ptr2 = names = realloc (names, memlen); - UNLOCK (wrklock); - - if (!names) - break; - } - - memcpy (names + memofs, name, len); - memofs += len; - } - } - - if (errno) - res = -1; - - req->result = res; -} - -/*****************************************************************************/ - -static void *aio_proc (void *thr_arg) -{ - aio_req req; - struct timespec ts; - worker *self = (worker *)thr_arg; - - /* try to distribute timeouts somewhat evenly */ - ts.tv_nsec = (((unsigned long)self + (unsigned long)ts.tv_sec) & 1023UL) - * (1000000000UL / 1024UL); - - for (;;) - { - ts.tv_sec = time (0) + IDLE_TIMEOUT; - - LOCK (reqlock); - - for (;;) - { - self->req = req = reqq_shift (&req_queue); - - if (req) - break; - - ++idle; - - if (pthread_cond_timedwait (&reqwait, &reqlock, &ts) - == ETIMEDOUT) - { - if (idle > max_idle) - { - --idle; - UNLOCK (reqlock); - LOCK (wrklock); - --started; - UNLOCK (wrklock); - goto quit; - } - - /* we are allowed to idle, so do so without any timeout */ - pthread_cond_wait (&reqwait, &reqlock); - ts.tv_sec = time (0) + IDLE_TIMEOUT; - } - - --idle; - } - - --nready; - - UNLOCK (reqlock); - - errno = 0; /* strictly unnecessary */ - - if (!(req->flags & FLAG_CANCELLED)) - switch (req->type) - { - case REQ_READ: req->result = pread (req->int1, req->ptr1, req->size, req->offs); break; - case REQ_WRITE: req->result = pwrite (req->int1, req->ptr1, req->size, req->offs); break; - - case REQ_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; - case REQ_SENDFILE: req->result = sendfile_ (req->int1, req->int2, req->offs, req->size, self); break; - - case REQ_STAT: req->result = stat (req->ptr1, req->statdata); break; - case REQ_LSTAT: req->result = lstat (req->ptr1, req->statdata); break; - case REQ_FSTAT: req->result = fstat (req->int1, req->statdata); break; - - case REQ_OPEN: req->result = open (req->ptr1, req->int1, req->mode); break; - case REQ_CLOSE: req->result = close (req->int1); break; - case REQ_UNLINK: req->result = unlink (req->ptr1); break; - case REQ_RMDIR: req->result = rmdir (req->ptr1); break; - case REQ_RENAME: req->result = rename (req->ptr2, req->ptr1); break; - case REQ_LINK: req->result = link (req->ptr2, req->ptr1); break; - case REQ_SYMLINK: req->result = symlink (req->ptr2, req->ptr1); break; - case REQ_MKNOD: req->result = mknod (req->ptr2, req->mode, (dev_t)req->offs); break; - case REQ_READLINK: req->result = readlink (req->ptr2, req->ptr1, NAME_MAX); break; - - case REQ_FDATASYNC: req->result = fdatasync (req->int1); break; - case REQ_FSYNC: req->result = fsync (req->int1); break; - case REQ_READDIR: scandir_ (req, self); break; - - case REQ_BUSY: - { - struct timeval tv; - - tv.tv_sec = req->int1; - tv.tv_usec = req->int2; - - req->result = select (0, 0, 0, 0, &tv); - } - - case REQ_GROUP: - case REQ_NOP: - break; - - case REQ_QUIT: - goto quit; - - default: - req->result = ENOSYS; - break; - } - - req->errorno = errno; - - LOCK (reslock); - - ++npending; - - if (!reqq_push (&res_queue, req)) - /* write a dummy byte to the pipe so fh becomes ready */ - write (respipe [1], &respipe, 1); - - self->req = 0; - worker_clear (self); - - UNLOCK (reslock); } - -quit: - LOCK (wrklock); - worker_free (self); - UNLOCK (wrklock); - - return 0; -} - -/*****************************************************************************/ - -static void atfork_prepare (void) -{ - LOCK (wrklock); - LOCK (reqlock); - LOCK (reslock); -#if !HAVE_PREADWRITE - LOCK (preadwritelock); -#endif -#if !HAVE_READDIR_R - LOCK (readdirlock); -#endif } -static void atfork_parent (void) +static void atfork_child (void) { -#if !HAVE_READDIR_R - UNLOCK (readdirlock); -#endif -#if !HAVE_PREADWRITE - UNLOCK (preadwritelock); -#endif - UNLOCK (reslock); - UNLOCK (reqlock); - UNLOCK (wrklock); + create_respipe (); } -static void atfork_child (void) +static SV * +get_cb (SV *cb_sv) { - aio_req prv; - - while (prv = reqq_shift (&req_queue)) - req_free (prv); - - while (prv = reqq_shift (&res_queue)) - req_free (prv); - - while (wrk_first.next != &wrk_first) - { - worker *wrk = wrk_first.next; - - if (wrk->req) - req_free (wrk->req); - - worker_clear (wrk); - worker_free (wrk); - } - - started = 0; - idle = 0; - nreqs = 0; - nready = 0; - npending = 0; - - close (respipe [0]); - close (respipe [1]); - create_pipe (); - - atfork_parent (); + SvGETMAGIC (cb_sv); + return SvOK (cb_sv) ? s_get_cv_croak (cb_sv) : 0; } #define dREQ \ + SV *cb_cv; \ aio_req req; \ int req_pri = next_pri; \ - next_pri = DEFAULT_PRI + PRI_BIAS; \ + next_pri = EIO_PRI_DEFAULT; \ \ - if (SvOK (callback) && !SvROK (callback)) \ - croak ("callback must be undef or of reference type"); \ + cb_cv = get_cb (callback); \ \ - Newz (0, req, 1, aio_cb); \ + Newz (0, req, 1, eio_req); \ if (!req) \ - croak ("out of memory during aio_req allocation"); \ + croak ("out of memory during eio_req allocation"); \ \ - req->callback = newSVsv (callback); \ + req->callback = SvREFCNT_inc (cb_cv); \ req->pri = req_pri #define REQ_SEND \ - req_send (req); \ + PUTBACK; \ + req_submit (req); \ + SPAGAIN; \ \ if (GIMME_V != G_VOID) \ XPUSHs (req_sv (req, AIO_REQ_KLASS)); - + MODULE = IO::AIO PACKAGE = IO::AIO PROTOTYPES: ENABLE BOOT: { - HV *stash = gv_stashpv ("IO::AIO", 1); + static const struct { + const char *name; + IV iv; + } *civ, const_iv[] = { +# define const_iv(name, value) { # name, (IV) value }, +# define const_eio(name) { # name, (IV) EIO_ ## name }, + const_iv (EXDEV , EXDEV) + const_iv (ENOSYS , ENOSYS) + const_iv (O_RDONLY, O_RDONLY) + const_iv (O_WRONLY, O_WRONLY) + const_iv (O_CREAT , O_CREAT) + const_iv (O_TRUNC , O_TRUNC) +#ifndef _WIN32 + const_iv (S_IFIFO , S_IFIFO) +#endif + const_iv (FADV_NORMAL , POSIX_FADV_NORMAL) + const_iv (FADV_SEQUENTIAL, POSIX_FADV_SEQUENTIAL) + const_iv (FADV_RANDOM , POSIX_FADV_RANDOM) + const_iv (FADV_NOREUSE , POSIX_FADV_NOREUSE) + const_iv (FADV_WILLNEED , POSIX_FADV_WILLNEED) + const_iv (FADV_DONTNEED , POSIX_FADV_DONTNEED) + + const_eio (SYNC_FILE_RANGE_WAIT_BEFORE) + const_eio (SYNC_FILE_RANGE_WRITE) + const_eio (SYNC_FILE_RANGE_WAIT_AFTER) + + const_eio (READDIR_DENTS) + const_eio (READDIR_DIRS_FIRST) + const_eio (READDIR_STAT_ORDER) + const_eio (READDIR_FOUND_UNKNOWN) + + const_eio (DT_UNKNOWN) + const_eio (DT_FIFO) + const_eio (DT_CHR) + const_eio (DT_DIR) + const_eio (DT_BLK) + const_eio (DT_REG) + const_eio (DT_LNK) + const_eio (DT_SOCK) + const_eio (DT_WHT) + }; + + stash = gv_stashpv ("IO::AIO", 1); + + for (civ = const_iv + sizeof (const_iv) / sizeof (const_iv [0]); civ-- > const_iv; ) + newCONSTSUB (stash, (char *)civ->name, newSViv (civ->iv)); + + create_respipe (); - newCONSTSUB (stash, "EXDEV", newSViv (EXDEV)); - newCONSTSUB (stash, "O_RDONLY", newSViv (O_RDONLY)); - newCONSTSUB (stash, "O_WRONLY", newSViv (O_WRONLY)); - newCONSTSUB (stash, "O_CREAT", newSViv (O_CREAT)); - newCONSTSUB (stash, "O_TRUNC", newSViv (O_TRUNC)); - newCONSTSUB (stash, "S_IFIFO", newSViv (S_IFIFO)); + if (eio_init (want_poll, done_poll) < 0) + croak ("IO::AIO: unable to initialise eio library"); - create_pipe (); - pthread_atfork (atfork_prepare, atfork_parent, atfork_child); + /* atfork child called in fifo order, so before eio's handler */ + X_THREAD_ATFORK (0, 0, atfork_child); } void max_poll_reqs (int nreqs) PROTOTYPE: $ CODE: - max_poll_reqs = nreqs; + eio_set_max_poll_reqs (nreqs); void max_poll_time (double nseconds) PROTOTYPE: $ CODE: - max_poll_time = nseconds * AIO_TICKS; + eio_set_max_poll_time (nseconds); void min_parallel (int nthreads) PROTOTYPE: $ + CODE: + eio_set_min_parallel (nthreads); void max_parallel (int nthreads) PROTOTYPE: $ + CODE: + eio_set_max_parallel (nthreads); void max_idle (int nthreads) PROTOTYPE: $ CODE: - set_max_idle (nthreads); + eio_set_max_idle (nthreads); -int +void max_outstanding (int maxreqs) PROTOTYPE: $ CODE: - RETVAL = max_outstanding; max_outstanding = maxreqs; - OUTPUT: - RETVAL void -aio_open (pathname,flags,mode,callback=&PL_sv_undef) - SV * pathname - int flags - int mode - SV * callback +aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef) PROTOTYPE: $$$;$ PPCODE: { dREQ; - req->type = REQ_OPEN; + req->type = EIO_OPEN; req->sv1 = newSVsv (pathname); - req->ptr1 = SvPVbyte_nolen (pathname); + req->ptr1 = SvPVbyte_nolen (req->sv1); req->int1 = flags; - req->mode = mode; + req->int2 = mode; REQ_SEND; } void -aio_close (fh,callback=&PL_sv_undef) - SV * fh - SV * callback +aio_fsync (SV *fh, SV *callback=&PL_sv_undef) PROTOTYPE: $;$ ALIAS: - aio_close = REQ_CLOSE - aio_fsync = REQ_FSYNC - aio_fdatasync = REQ_FDATASYNC + aio_fsync = EIO_FSYNC + aio_fdatasync = EIO_FDATASYNC PPCODE: { + int fd = s_fileno_croak (fh, 0); dREQ; req->type = ix; - req->fh = newSVsv (fh); - req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh))); + req->sv1 = newSVsv (fh); + req->int1 = fd; + + REQ_SEND (req); +} + +void +aio_sync_file_range (SV *fh, off_t offset, size_t nbytes, UV flags, SV *callback=&PL_sv_undef) + PROTOTYPE: $$$$;$ + PPCODE: +{ + int fd = s_fileno_croak (fh, 0); + dREQ; + + req->type = EIO_SYNC_FILE_RANGE; + req->sv1 = newSVsv (fh); + req->int1 = fd; + req->offs = offset; + req->size = nbytes; + req->int2 = flags; REQ_SEND (req); } void -aio_read (fh,offset,length,data,dataoffset,callback=&PL_sv_undef) - SV * fh - UV offset - UV length - SV * data - UV dataoffset - SV * callback +aio_close (SV *fh, SV *callback=&PL_sv_undef) + PROTOTYPE: $;$ + PPCODE: +{ + static int close_pipe = -1; /* dummy fd to close fds via dup2 */ + int fd = s_fileno_croak (fh, 0); + dREQ; + + if (close_pipe < 0) + { + int pipefd [2]; + + if (pipe (pipefd) < 0 + || close (pipefd [1]) < 0 + || fcntl (pipefd [0], F_SETFD, FD_CLOEXEC) < 0) + abort (); /*D*/ + + close_pipe = pipefd [0]; + } + + req->type = EIO_DUP2; + req->int1 = close_pipe; + req->sv2 = newSVsv (fh); + req->int2 = fd; + + REQ_SEND (req); +} + +void +aio_read (SV *fh, SV *offset, SV *length, SV8 *data, IV dataoffset, SV *callback=&PL_sv_undef) ALIAS: - aio_read = REQ_READ - aio_write = REQ_WRITE + aio_read = EIO_READ + aio_write = EIO_WRITE PROTOTYPE: $$$$$;$ PPCODE: { STRLEN svlen; + int fd = s_fileno_croak (fh, ix == EIO_WRITE); char *svptr = SvPVbyte (data, svlen); - - SvUPGRADE (data, SVt_PV); - SvPOK_on (data); + UV len = SvUV (length); if (dataoffset < 0) dataoffset += svlen; if (dataoffset < 0 || dataoffset > svlen) - croak ("data offset outside of string"); + croak ("dataoffset outside of data scalar"); - if (ix == REQ_WRITE) + if (ix == EIO_WRITE) { /* write: check length and adjust. */ - if (length < 0 || length + dataoffset > svlen) - length = svlen - dataoffset; + if (!SvOK (length) || len + dataoffset > svlen) + len = svlen - dataoffset; } else { - /* read: grow scalar as necessary */ - svptr = SvGROW (data, length + dataoffset); + /* read: check type and grow scalar as necessary */ + SvUPGRADE (data, SVt_PV); + svptr = SvGROW (data, len + dataoffset + 1); } - if (length < 0) - croak ("length must not be negative"); - { dREQ; req->type = ix; - req->fh = newSVsv (fh); - req->int1 = PerlIO_fileno (ix == REQ_READ ? IoIFP (sv_2io (fh)) - : IoOFP (sv_2io (fh))); - req->offs = offset; - req->size = length; - req->sv1 = SvREFCNT_inc (data); - req->ptr1 = (char *)svptr + dataoffset; + req->sv1 = newSVsv (fh); + req->int1 = fd; + req->offs = SvOK (offset) ? SvVAL64 (offset) : -1; + req->size = len; + req->sv2 = SvREFCNT_inc (data); + req->ptr2 = (char *)svptr + dataoffset; req->stroffset = dataoffset; if (!SvREADONLY (data)) { SvREADONLY_on (data); - req->flags |= FLAG_DATA_RO_OFF; + req->flags |= FLAG_SV2_RO_OFF; } REQ_SEND; @@ -1410,44 +752,34 @@ } void -aio_readlink (path,callback=&PL_sv_undef) - SV * path - SV * callback +aio_readlink (SV8 *path, SV *callback=&PL_sv_undef) PROTOTYPE: $$;$ PPCODE: { SV *data; dREQ; - data = newSV (NAME_MAX); - SvPOK_on (data); - - req->type = REQ_READLINK; - req->fh = newSVsv (path); - req->ptr2 = SvPVbyte_nolen (path); - req->sv1 = data; - req->ptr1 = SvPVbyte_nolen (data); + req->type = EIO_READLINK; + req->sv1 = newSVsv (path); + req->ptr1 = SvPVbyte_nolen (req->sv1); REQ_SEND; } void -aio_sendfile (out_fh,in_fh,in_offset,length,callback=&PL_sv_undef) - SV * out_fh - SV * in_fh - UV in_offset - UV length - SV * callback +aio_sendfile (SV *out_fh, SV *in_fh, off_t in_offset, size_t length, SV *callback=&PL_sv_undef) PROTOTYPE: $$$$;$ PPCODE: { + int ifd = s_fileno_croak (in_fh , 0); + int ofd = s_fileno_croak (out_fh, 1); dREQ; - req->type = REQ_SENDFILE; - req->fh = newSVsv (out_fh); - req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh))); + req->type = EIO_SENDFILE; + req->sv1 = newSVsv (out_fh); + req->int1 = ofd; req->sv2 = newSVsv (in_fh); - req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh))); + req->int2 = ifd; req->offs = in_offset; req->size = length; @@ -1455,19 +787,16 @@ } void -aio_readahead (fh,offset,length,callback=&PL_sv_undef) - SV * fh - UV offset - IV length - SV * callback +aio_readahead (SV *fh, off_t offset, size_t length, SV *callback=&PL_sv_undef) PROTOTYPE: $$$;$ PPCODE: { + int fd = s_fileno_croak (fh, 0); dREQ; - req->type = REQ_READAHEAD; - req->fh = newSVsv (fh); - req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh))); + req->type = EIO_READAHEAD; + req->sv1 = newSVsv (fh); + req->int1 = fd; req->offs = offset; req->size = length; @@ -1475,33 +804,97 @@ } void -aio_stat (fh_or_path,callback=&PL_sv_undef) - SV * fh_or_path - SV * callback +aio_stat (SV8 *fh_or_path, SV *callback=&PL_sv_undef) ALIAS: - aio_stat = REQ_STAT - aio_lstat = REQ_LSTAT + aio_stat = EIO_STAT + aio_lstat = EIO_LSTAT + PPCODE: +{ + dREQ; + + req->sv1 = newSVsv (fh_or_path); + + if (SvPOK (req->sv1)) + { + req->type = ix; + req->ptr1 = SvPVbyte_nolen (req->sv1); + } + else + { + req->type = EIO_FSTAT; + req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path))); + } + + REQ_SEND; +} + +void +aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef) + PPCODE: +{ + dREQ; + + req->nv1 = SvOK (atime) ? SvNV (atime) : -1.; + req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.; + req->sv1 = newSVsv (fh_or_path); + + if (SvPOK (req->sv1)) + { + req->type = EIO_UTIME; + req->ptr1 = SvPVbyte_nolen (req->sv1); + } + else + { + req->type = EIO_FUTIME; + req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path))); + } + + REQ_SEND; +} + +void +aio_truncate (SV8 *fh_or_path, SV *offset, SV *callback=&PL_sv_undef) PPCODE: { dREQ; - New (0, req->statdata, 1, Stat_t); - if (!req->statdata) + req->sv1 = newSVsv (fh_or_path); + req->offs = SvOK (offset) ? SvVAL64 (offset) : -1; + + if (SvPOK (req->sv1)) + { + req->type = EIO_TRUNCATE; + req->ptr1 = SvPVbyte_nolen (req->sv1); + } + else { - req_free (req); - croak ("out of memory during aio_req->statdata allocation"); + req->type = EIO_FTRUNCATE; + req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path))); } - if (SvPOK (fh_or_path)) + REQ_SEND; +} + +void +aio_chmod (SV8 *fh_or_path, int mode, SV *callback=&PL_sv_undef) + ALIAS: + aio_chmod = EIO_CHMOD + aio_mkdir = EIO_MKDIR + PPCODE: +{ + dREQ; + + req->int2 = mode; + req->sv1 = newSVsv (fh_or_path); + + if (SvPOK (req->sv1)) { req->type = ix; - req->sv1 = newSVsv (fh_or_path); - req->ptr1 = SvPVbyte_nolen (fh_or_path); + req->ptr1 = SvPVbyte_nolen (req->sv1); } else { - req->type = REQ_FSTAT; - req->fh = newSVsv (fh_or_path); + req->type = EIO_FCHMOD; req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path))); } @@ -1509,102 +902,132 @@ } void -aio_unlink (pathname,callback=&PL_sv_undef) - SV * pathname - SV * callback +aio_chown (SV8 *fh_or_path, SV *uid, SV *gid, SV *callback=&PL_sv_undef) + PPCODE: +{ + dREQ; + + req->int2 = SvOK (uid) ? SvIV (uid) : -1; + req->int3 = SvOK (gid) ? SvIV (gid) : -1; + req->sv1 = newSVsv (fh_or_path); + + if (SvPOK (req->sv1)) + { + req->type = EIO_CHOWN; + req->ptr1 = SvPVbyte_nolen (req->sv1); + } + else + { + req->type = EIO_FCHOWN; + req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path))); + } + + REQ_SEND; +} + +void +aio_readdirx (SV8 *pathname, IV flags, SV *callback=&PL_sv_undef) + PPCODE: +{ + dREQ; + + req->type = EIO_READDIR; + req->sv1 = newSVsv (pathname); + req->ptr1 = SvPVbyte_nolen (req->sv1); + req->int1 = flags | EIO_READDIR_DENTS | EIO_READDIR_CUSTOM1; + + if (flags & EIO_READDIR_DENTS) + req->int1 |= EIO_READDIR_CUSTOM2; + + REQ_SEND; +} + +void +aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef) ALIAS: - aio_unlink = REQ_UNLINK - aio_rmdir = REQ_RMDIR - aio_readdir = REQ_READDIR + aio_unlink = EIO_UNLINK + aio_rmdir = EIO_RMDIR + aio_readdir = EIO_READDIR PPCODE: { dREQ; req->type = ix; req->sv1 = newSVsv (pathname); - req->ptr1 = SvPVbyte_nolen (pathname); - + req->ptr1 = SvPVbyte_nolen (req->sv1); + REQ_SEND; } void -aio_link (oldpath,newpath,callback=&PL_sv_undef) - SV * oldpath - SV * newpath - SV * callback +aio_link (SV8 *oldpath, SV8 *newpath, SV *callback=&PL_sv_undef) ALIAS: - aio_link = REQ_LINK - aio_symlink = REQ_SYMLINK - aio_rename = REQ_RENAME + aio_link = EIO_LINK + aio_symlink = EIO_SYMLINK + aio_rename = EIO_RENAME PPCODE: { dREQ; req->type = ix; - req->fh = newSVsv (oldpath); - req->ptr2 = SvPVbyte_nolen (req->fh); - req->sv1 = newSVsv (newpath); - req->ptr1 = SvPVbyte_nolen (newpath); + req->sv1 = newSVsv (oldpath); + req->ptr1 = SvPVbyte_nolen (req->sv1); + req->sv2 = newSVsv (newpath); + req->ptr2 = SvPVbyte_nolen (req->sv2); REQ_SEND; } void -aio_mknod (pathname,mode,dev,callback=&PL_sv_undef) - SV * pathname - SV * callback - UV mode - UV dev +aio_mknod (SV8 *pathname, int mode, UV dev, SV *callback=&PL_sv_undef) PPCODE: { dREQ; - req->type = REQ_MKNOD; + req->type = EIO_MKNOD; req->sv1 = newSVsv (pathname); - req->ptr1 = SvPVbyte_nolen (pathname); - req->mode = (mode_t)mode; + req->ptr1 = SvPVbyte_nolen (req->sv1); + req->int2 = (mode_t)mode; req->offs = dev; REQ_SEND; } void -aio_busy (delay,callback=&PL_sv_undef) - double delay - SV * callback +aio_busy (double delay, SV *callback=&PL_sv_undef) PPCODE: { dREQ; - req->type = REQ_BUSY; - req->int1 = delay < 0. ? 0 : delay; - req->int2 = delay < 0. ? 0 : 1000. * (delay - req->int1); + req->type = EIO_BUSY; + req->nv1 = delay < 0. ? 0. : delay; REQ_SEND; } void -aio_group (callback=&PL_sv_undef) - SV * callback +aio_group (SV *callback=&PL_sv_undef) PROTOTYPE: ;$ PPCODE: { dREQ; - req->type = REQ_GROUP; + req->type = EIO_GROUP; - req_send (req); + req_submit (req); XPUSHs (req_sv (req, AIO_GRP_KLASS)); } void -aio_nop (callback=&PL_sv_undef) - SV * callback +aio_nop (SV *callback=&PL_sv_undef) + ALIAS: + aio_nop = EIO_NOP + aio_sync = EIO_SYNC PPCODE: { dREQ; - req->type = REQ_NOP; + req->type = ix; REQ_SEND; } @@ -1613,12 +1036,12 @@ aioreq_pri (int pri = 0) PROTOTYPE: ;$ CODE: - RETVAL = next_pri - PRI_BIAS; + RETVAL = next_pri; if (items > 0) { - if (pri < PRI_MIN) pri = PRI_MIN; - if (pri > PRI_MAX) pri = PRI_MAX; - next_pri = pri + PRI_BIAS; + if (pri < EIO_PRI_MIN) pri = EIO_PRI_MIN; + if (pri > EIO_PRI_MAX) pri = EIO_PRI_MAX; + next_pri = pri; } OUTPUT: RETVAL @@ -1627,35 +1050,34 @@ aioreq_nice (int nice = 0) CODE: nice = next_pri - nice; - if (nice < PRI_MIN) nice = PRI_MIN; - if (nice > PRI_MAX) nice = PRI_MAX; - next_pri = nice + PRI_BIAS; + if (nice < EIO_PRI_MIN) nice = EIO_PRI_MIN; + if (nice > EIO_PRI_MAX) nice = EIO_PRI_MAX; + next_pri = nice; void flush () PROTOTYPE: CODE: - while (nreqs) + while (eio_nreqs ()) { poll_wait (); - poll_cb (0); + poll_cb (); } -void +int poll() PROTOTYPE: CODE: - if (nreqs) - { - poll_wait (); - poll_cb (0); - } + poll_wait (); + RETVAL = poll_cb (); + OUTPUT: + RETVAL int poll_fileno() PROTOTYPE: CODE: - RETVAL = respipe [0]; + RETVAL = s_epipe_fd (&respipe); OUTPUT: RETVAL @@ -1671,14 +1093,13 @@ poll_wait() PROTOTYPE: CODE: - if (nreqs) - poll_wait (); + poll_wait (); int nreqs() PROTOTYPE: CODE: - RETVAL = nreqs; + RETVAL = eio_nreqs (); OUTPUT: RETVAL @@ -1686,7 +1107,7 @@ nready() PROTOTYPE: CODE: - RETVAL = get_nready (); + RETVAL = eio_nready (); OUTPUT: RETVAL @@ -1694,7 +1115,7 @@ npending() PROTOTYPE: CODE: - RETVAL = get_npending (); + RETVAL = eio_npending (); OUTPUT: RETVAL @@ -1702,12 +1123,33 @@ nthreads() PROTOTYPE: CODE: - if (WORDACCESS_UNSAFE) LOCK (wrklock); - RETVAL = started; - if (WORDACCESS_UNSAFE) UNLOCK (wrklock); + RETVAL = eio_nthreads (); OUTPUT: RETVAL +int +fadvise (aio_rfd fh, off_t offset, off_t length, IV advice) + PROTOTYPE: $$$$ + CODE: +#if _XOPEN_SOURCE >= 600 && !NO_FADVISE + RETVAL = posix_fadvise (fh, offset, length, advice); +#else + RETVAL = errno = ENOSYS; +#endif + OUTPUT: + RETVAL + +ssize_t +sendfile (aio_wfd ofh, aio_rfd ifh, off_t offset, size_t count) + PROTOTYPE: $$$$ + CODE: + eio_sendfile_sync (ofh, ifh, offset, count); + +void _on_next_submit (SV *cb) + CODE: + SvREFCNT_dec (on_next_submit); + on_next_submit = SvOK (cb) ? newSVsv (cb) : 0; + PROTOTYPES: DISABLE MODULE = IO::AIO PACKAGE = IO::AIO::REQ @@ -1715,13 +1157,23 @@ void cancel (aio_req_ornot req) CODE: - req_cancel (req); + eio_cancel (req); void cb (aio_req_ornot req, SV *callback=&PL_sv_undef) - CODE: - SvREFCNT_dec (req->callback); - req->callback = newSVsv (callback); + PPCODE: +{ + if (GIMME_V != G_VOID) + XPUSHs (req->callback ? sv_2mortal (newRV_inc (req->callback)) : &PL_sv_undef); + + if (items > 1) + { + SV *cb_cv =get_cb (callback); + + SvREFCNT_dec (req->callback); + req->callback = SvREFCNT_inc (cb_cv); + } +} MODULE = IO::AIO PACKAGE = IO::AIO::GRP @@ -1730,31 +1182,21 @@ PPCODE: { int i; - aio_req req; if (grp->int1 == 2) croak ("cannot add requests to IO::AIO::GRP after the group finished"); for (i = 1; i < items; ++i ) { + aio_req req; + if (GIMME_V != G_VOID) XPUSHs (sv_2mortal (newSVsv (ST (i)))); req = SvAIO_REQ (ST (i)); if (req) - { - ++grp->size; - 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; - } + eio_grp_add (grp, req); } } @@ -1773,6 +1215,7 @@ grp->errorno = errno; av = newAV (); + av_extend (av, items - 1); for (i = 1; i < items; ++i ) av_push (av, newSVsv (ST (i))); @@ -1789,19 +1232,19 @@ void limit (aio_req grp, int limit) CODE: - grp->int2 = limit; - aio_grp_feed (grp); + eio_grp_limit (grp, limit); void feed (aio_req grp, SV *callback=&PL_sv_undef) CODE: { SvREFCNT_dec (grp->sv2); - grp->sv2 = newSVsv (callback); + grp->sv2 = newSVsv (callback); + grp->feed = aio_grp_feed; if (grp->int2 <= 0) grp->int2 = 2; - aio_grp_feed (grp); + eio_grp_limit (grp, grp->int2); }