--- IO-AIO/AIO.xs 2007/07/08 11:05:36 1.104 +++ IO-AIO/AIO.xs 2014/04/11 05:19:40 1.229 @@ -1,4 +1,4 @@ -#include "xthread.h" +#include "libeio/xthread.h" #include @@ -6,6 +6,8 @@ #include "perl.h" #include "XSUB.h" +#include "schmorp.h" + #include #include #include @@ -15,367 +17,286 @@ #include #include -#ifdef _WIN32 - -# define SIGIO 0 - typedef Direntry_t X_DIRENT; -#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 "autoconf/config.h" -# include -# include -# include -# include -# include - typedef struct dirent X_DIRENT; - +#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES +# include #endif -#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 +/* the incompetent fool that created musl keeps __linux__, refuses + * to implement any linux standard apis, and also has no way to test + * for his broken iplementation. on't complain if this fails for you. + */ +#if __linux__ && (defined __GLIBC__ || defined __UCLIBC__) +# include +# ifdef FS_IOC_FIEMAP +# include +# include +# define HAVE_FIEMAP 1 # endif #endif -/* number of seconds after which idle threads exit */ -#define IDLE_TIMEOUT 10 - -/* used for struct dirent, AIX doesn't provide it */ -#ifndef NAME_MAX -# define NAME_MAX 4096 -#endif +/* perl namespace pollution */ +#undef VERSION -/* buffer size for various temporary buffers */ -#define AIO_BUFSIZE 65536 +/* perl stupidly overrides readdir and maybe others */ +/* with thread-unsafe versions, imagine that :( */ +#undef readdir +#undef opendir +#undef closedir -/* use NV for 32 bit perls as it allows larger offsets */ -#if IVSIZE >= 8 -# define SvVAL64 SvIV -#else -# define SvVAL64 SvNV -#endif - -#define dBUF \ - char *aio_buf; \ - X_LOCK (wrklock); \ - self->dbuf = aio_buf = malloc (AIO_BUFSIZE); \ - X_UNLOCK (wrklock); \ - if (!aio_buf) \ - return -1; +#ifdef _WIN32 -typedef SV SV8; /* byte-sv, used for argument-checking */ + // perl overrides all those nice libc functions -enum { - REQ_QUIT, - REQ_OPEN, REQ_CLOSE, - REQ_READ, REQ_WRITE, - REQ_READAHEAD, REQ_SENDFILE, - REQ_STAT, REQ_LSTAT, REQ_FSTAT, - REQ_TRUNCATE, REQ_FTRUNCATE, - REQ_UTIME, REQ_FUTIME, - REQ_CHMOD, REQ_FCHMOD, - REQ_CHOWN, REQ_FCHOWN, - REQ_FSYNC, REQ_FDATASYNC, - REQ_UNLINK, REQ_RMDIR, REQ_MKDIR, REQ_RENAME, - REQ_MKNOD, REQ_READDIR, - REQ_LINK, REQ_SYMLINK, REQ_READLINK, - REQ_GROUP, REQ_NOP, - REQ_BUSY, -}; + #undef malloc + #undef free + #undef open + #undef read + #undef write + #undef send + #undef recv + #undef stat + #undef lstat + #undef fstat + #undef truncate + #undef ftruncate + #undef open + #undef link + #undef close + #undef unlink + #undef mkdir + #undef rmdir + #undef rename + #undef lseek + #undef opendir + #undef readdir + #undef closedir + #undef chmod + #undef fchmod + #undef dup + #undef dup2 + #undef abort + #undef pipe -#define AIO_REQ_KLASS "IO::AIO::REQ" -#define AIO_GRP_KLASS "IO::AIO::GRP" + #define EIO_STRUCT_STAT struct _stati64 + #define EIO_STRUCT_STATI64 -typedef struct aio_cb -{ - struct aio_cb *volatile next; +#else - SV *callback; - SV *sv1, *sv2; - void *ptr1, *ptr2; - off_t offs; - size_t size; - ssize_t result; - double nv1, nv2; - - STRLEN stroffset; - int type; - int int1, int2, int3; - 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; + #include + #include + #include + #include + #include -enum { - FLAG_CANCELLED = 0x01, /* request was cancelled */ - FLAG_SV2_RO_OFF = 0x40, /* data was set readonly */ - FLAG_PTR2_FREE = 0x80, /* need to free(ptr2) */ -}; + #define EIO_STRUCT_STAT Stat_t -typedef aio_cb *aio_req; -typedef aio_cb *aio_req_ornot; +#endif -enum { - PRI_MIN = -4, - PRI_MAX = 4, +/*****************************************************************************/ - DEFAULT_PRI = 0, - PRI_BIAS = -PRI_MIN, - NUM_PRI = PRI_MAX + PRI_BIAS + 1, -}; +#if __GNUC__ >= 3 +# define expect(expr,value) __builtin_expect ((expr),(value)) +#else +# define expect(expr,value) (expr) +#endif -#define AIO_TICKS ((1000000 + 1023) >> 10) +#define expect_false(expr) expect ((expr) != 0, 0) +#define expect_true(expr) expect ((expr) != 0, 1) -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); -} +typedef SV SV8; /* byte-sv, used for argument-checking */ +typedef int aio_rfd; /* read file desriptor */ +typedef int aio_wfd; /* write file descriptor */ -static thread_t main_tid; -static int main_sig; -static int block_sig_level; +static HV *aio_stash, *aio_req_stash, *aio_grp_stash, *aio_wd_stash; -void block_sig () -{ - sigset_t ss; +#define EIO_REQ_MEMBERS \ + SV *callback; \ + SV *sv1, *sv2; \ + SV *sv3, *sv4; \ + STRLEN stroffset; \ + SV *self; - if (block_sig_level++) - return; +#define EIO_NO_WRAPPERS 1 - if (!main_sig) - return; +#include "libeio/config.h" +#include "libeio/eio.h" - sigemptyset (&ss); - sigaddset (&ss, main_sig); - pthread_sigmask (SIG_BLOCK, &ss, 0); -} +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) -void unblock_sig () -{ - sigset_t ss; +#include "libeio/eio.c" - if (--block_sig_level) - return; +#if !HAVE_POSIX_FADVISE +# define posix_fadvise(a,b,c,d) errno = ENOSYS /* also return ENOSYS */ +#endif - if (!main_sig) - return; +#if !HAVE_POSIX_MADVISE +# define posix_madvise(a,b,c) errno = ENOSYS /* also return ENOSYS */ +#endif - sigemptyset (&ss); - sigaddset (&ss, main_sig); - pthread_sigmask (SIG_UNBLOCK, &ss, 0); -} +#ifndef MAP_ANONYMOUS +# ifdef MAP_ANON +# define MAP_ANONYMOUS MAP_ANON +# else +# define MAP_ANONYMOUS MAP_FIXED /* and hope this fails */ +# endif +#endif -static int next_pri = DEFAULT_PRI + PRI_BIAS; +/* defines all sorts of constants to 0 unless they are already defined */ +#include "def0.h" -static unsigned int started, idle, wanted; +#ifndef makedev +# define makedev(maj,min) (((maj) << 8) | (min)) +#endif +#ifndef major +# define major(dev) ((dev) >> 8) +#endif +#ifndef minor +# define minor(dev) ((dev) & 0xff) +#endif -/* worker threads management */ -static mutex_t wrklock = X_MUTEX_INIT; +#if PAGESIZE <= 0 +# define PAGESIZE sysconf (_SC_PAGESIZE) +#endif -typedef struct worker { - /* locked by wrklock */ - struct worker *prev, *next; +/*****************************************************************************/ - thread_t tid; +static void +fiemap (eio_req *req) +{ + req->result = -1; - /* locked by reslock, reqlock or wrklock */ - aio_req req; /* currently processed request */ - void *dbuf; - DIR *dirp; -} worker; +#if HAVE_FIEMAP + /* assume some c99 */ + struct fiemap *fiemap = 0; + size_t end_offset; + int count = req->int3; + + req->flags |= EIO_FLAG_PTR1_FREE; + + /* heuristic: start with 512 bytes (8 extents), and if that isn't enough, */ + /* increase in 3.5kb steps */ + if (count < 0) + count = 8; + + fiemap = malloc (sizeof (*fiemap) + sizeof (struct fiemap_extent) * count); + errno = ENOMEM; + if (!fiemap) + return; -static worker wrk_first = { &wrk_first, &wrk_first, 0 }; + req->ptr1 = fiemap; -static void worker_clear (worker *wrk) -{ - if (wrk->dirp) - { - closedir (wrk->dirp); - wrk->dirp = 0; - } + fiemap->fm_start = req->offs; + fiemap->fm_length = req->size; + fiemap->fm_flags = req->int2; + fiemap->fm_extent_count = count; - if (wrk->dbuf) - { - free (wrk->dbuf); - wrk->dbuf = 0; - } -} + if (ioctl (req->int1, FS_IOC_FIEMAP, fiemap) < 0) + return; -static void worker_free (worker *wrk) -{ - wrk->next->prev = wrk->prev; - wrk->prev->next = wrk->next; + if (req->int3 >= 0 /* not autosizing */ + || !fiemap->fm_mapped_extents /* no more extents */ + || fiemap->fm_extents [fiemap->fm_mapped_extents - 1].fe_flags & FIEMAP_EXTENT_LAST /* hit eof */) + goto done; + + /* else we have to loop - + * it would be tempting (actually I tried that first) to just query the + * number of extents needed, but linux often feels like not returning all + * extents, without telling us it left any out. this complicates + * this quite a bit. + */ - free (wrk); -} + end_offset = fiemap->fm_length + (fiemap->fm_length == FIEMAP_MAX_OFFSET ? 0 : fiemap->fm_start); -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], respipe_osf [2]; + for (;;) + { + /* we go in 54 extent steps - 3kb, in the hope that this fits nicely on the eio stack (normally 16+ kb) */ + char scratch[3072]; + struct fiemap *incmap = (struct fiemap *)scratch; + + incmap->fm_start = fiemap->fm_extents [fiemap->fm_mapped_extents - 1].fe_logical + + fiemap->fm_extents [fiemap->fm_mapped_extents - 1].fe_length; + incmap->fm_length = fiemap->fm_length - (incmap->fm_start - fiemap->fm_start); + incmap->fm_flags = fiemap->fm_flags; + incmap->fm_extent_count = (sizeof (scratch) - sizeof (struct fiemap)) / sizeof (struct fiemap_extent); -static mutex_t reslock = X_MUTEX_INIT; -static mutex_t reqlock = X_MUTEX_INIT; -static cond_t reqwait = X_COND_INIT; + if (ioctl (req->int1, FS_IOC_FIEMAP, incmap) < 0) + return; -#if WORDACCESS_UNSAFE + if (!incmap->fm_mapped_extents) + goto done; -static unsigned int get_nready () -{ - unsigned int retval; + count = fiemap->fm_mapped_extents + incmap->fm_mapped_extents; + fiemap = realloc (fiemap, sizeof (*fiemap) + sizeof (struct fiemap_extent) * count); + errno = ENOMEM; + if (!fiemap) + return; - X_LOCK (reqlock); - retval = nready; - X_UNLOCK (reqlock); + req->ptr1 = fiemap; - return retval; -} + for (count = 0; count < incmap->fm_mapped_extents; ++count) + { + struct fiemap_extent *e = incmap->fm_extents + count; -static unsigned int get_npending () -{ - unsigned int retval; + if (e->fe_logical + e->fe_length >= end_offset) + goto done; - X_LOCK (reslock); - retval = npending; - X_UNLOCK (reslock); + fiemap->fm_extents [fiemap->fm_mapped_extents++] = *e; - return retval; -} + if (e->fe_flags & FIEMAP_EXTENT_LAST) + goto done; -static unsigned int get_nthreads () -{ - unsigned int retval; + } + } - X_LOCK (wrklock); - retval = started; - X_UNLOCK (wrklock); +done: + req->result = 0; - return retval; +#else + errno = ENOSYS; +#endif } -#else +/*****************************************************************************/ -# define get_nready() nready -# define get_npending() npending -# define get_nthreads() started +enum { + FLAG_SV2_RO_OFF = 0x40, /* data was set readonly */ +}; -#endif +typedef eio_req *aio_req; +typedef eio_req *aio_req_ornot; +typedef eio_wd aio_wd; + +static SV *on_next_submit; +static int next_pri = EIO_PRI_DEFAULT; +static int max_outstanding; -/* - * 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 s_epipe respipe; -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 int req_invoke (aio_req req); -static void req_destroy (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) +static SV * +req_sv (aio_req req, HV *stash) { if (!req->self) { @@ -383,140 +304,246 @@ sv_magic (req->self, 0, PERL_MAGIC_ext, (char *)req, 0); } - return sv_2mortal (sv_bless (newRV_inc (req->self), gv_stashpv (klass, 1))); + return sv_2mortal (sv_bless (newRV_inc (req->self), stash)); +} + +static SV * +newSVaio_wd (aio_wd wd) +{ + return sv_bless (newRV_noinc (newSViv ((IV)wd)), aio_wd_stash); } -static aio_req SvAIO_REQ (SV *sv) +static aio_req +SvAIO_REQ (SV *sv) { MAGIC *mg; - if (!sv_derived_from (sv, AIO_REQ_KLASS) || !SvROK (sv)) - croak ("object of class " AIO_REQ_KLASS " expected"); + if (!SvROK (sv) + /* for speed reasons, we do not verify that SvROK actually has a stash ptr */ + || (SvSTASH (SvRV (sv)) != aio_grp_stash + && SvSTASH (SvRV (sv)) != aio_req_stash + && !sv_derived_from (sv, "IO::AIO::REQ"))) + croak ("object of class IO::AIO::REQ expected"); mg = mg_find (SvRV (sv), PERL_MAGIC_ext); return mg ? (aio_req)mg->mg_ptr : 0; } -static void aio_grp_feed (aio_req grp) +static aio_wd +SvAIO_WD (SV *sv) { - block_sig (); - - while (grp->size < grp->int2 && !(grp->flags & FLAG_CANCELLED)) - { - int old_len = grp->size; + if (!SvROK (sv) + || SvTYPE (SvRV (sv)) != SVt_PVMG + || SvSTASH (SvRV (sv)) != aio_wd_stash) + croak ("IO::AIO: expected a working directory object as returned by aio_wd"); - if (grp->sv2 && SvOK (grp->sv2)) - { - dSP; + return (aio_wd)(long)SvIVX (SvRV (sv)); +} - 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_feed (aio_req grp) +{ + if (grp->sv2 && SvOK (grp->sv2)) + { + dSP; - /* 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_stash)); + PUTBACK; + call_sv (grp->sv2, G_VOID | G_EVAL | G_KEEPERR); + SPAGAIN; + FREETMPS; + LEAVE; } - - unblock_sig (); } -static void aio_grp_dec (aio_req grp) +static void +req_submit (eio_req *req) { - --grp->size; + eio_submit (req); - /* call feeder, if applicable */ - aio_grp_feed (grp); - - /* finish, if done */ - if (!grp->size && grp->int1) + if (expect_false (on_next_submit)) { - block_sig (); + dSP; + SV *cb = sv_2mortal (on_next_submit); - if (!req_invoke (grp)) - { - req_destroy (grp); - unblock_sig (); - croak (0); - } + on_next_submit = 0; - req_destroy (grp); - unblock_sig (); + PUSHMARK (SP); + PUTBACK; + call_sv (cb, G_DISCARD | G_EVAL); } } -static int req_invoke (aio_req req) +static int +req_invoke (eio_req *req) { - dSP; - 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) { + dSP; + static SV *sv_result_cache; /* caches the result integer SV */ + SV *sv_result; + ENTER; SAVETMPS; PUSHMARK (SP); EXTEND (SP, 1); + /* do not recreate the result IV from scratch each time */ + if (expect_true (sv_result_cache)) + { + sv_result = sv_result_cache; sv_result_cache = 0; + SvIV_set (sv_result, req->result); + SvIOK_only (sv_result); + } + else + { + sv_result = newSViv (req->result); + SvREADONLY_on (sv_result); + } + switch (req->type) { - case REQ_READDIR: + case EIO_WD_OPEN: + PUSHs (req->result ? &PL_sv_undef : sv_2mortal (newSVaio_wd (req->wd))); + break; + + 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; + + 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, aio_stash, sym, symlen, 0); + + symlen = snprintf ( + sym, + sizeof (sym), + "%s&=%d", + flags == O_RDONLY ? "<" : flags == O_WRONLY ? ">" : "+<", + (int)req->result + ); + + if (do_open (gv, sym, symlen, 0, 0, 0, 0)) + fh = (SV *)gv; + } + + PUSHs (fh); + } + break; + + case EIO_STATVFS: + case EIO_FSTATVFS: + { + SV *rv = &PL_sv_undef; + +#ifndef _WIN32 + if (req->result >= 0) + { + EIO_STRUCT_STATVFS *f = EIO_STATVFS_BUF (req); + HV *hv = newHV (); + /* POSIX requires fsid to be unsigned long, but AIX in its infinite wisdom + * chooses to make it a struct. + */ + unsigned long fsid = 0; + memcpy (&fsid, &f->f_fsid, sizeof (unsigned long) < sizeof (f->f_fsid) ? sizeof (unsigned long) : sizeof (f->f_fsid)); + + rv = sv_2mortal (newRV_noinc ((SV *)hv)); + + hv_store (hv, "bsize" , sizeof ("bsize" ) - 1, newSVval64 (f->f_bsize ), 0); + hv_store (hv, "frsize" , sizeof ("frsize" ) - 1, newSVval64 (f->f_frsize ), 0); + hv_store (hv, "blocks" , sizeof ("blocks" ) - 1, newSVval64 (f->f_blocks ), 0); + hv_store (hv, "bfree" , sizeof ("bfree" ) - 1, newSVval64 (f->f_bfree ), 0); + hv_store (hv, "bavail" , sizeof ("bavail" ) - 1, newSVval64 (f->f_bavail ), 0); + hv_store (hv, "files" , sizeof ("files" ) - 1, newSVval64 (f->f_files ), 0); + hv_store (hv, "ffree" , sizeof ("ffree" ) - 1, newSVval64 (f->f_ffree ), 0); + hv_store (hv, "favail" , sizeof ("favail" ) - 1, newSVval64 (f->f_favail ), 0); + hv_store (hv, "fsid" , sizeof ("fsid" ) - 1, newSVval64 (fsid ), 0); + hv_store (hv, "flag" , sizeof ("flag" ) - 1, newSVval64 (f->f_flag ), 0); + hv_store (hv, "namemax", sizeof ("namemax") - 1, newSVval64 (f->f_namemax), 0); + } +#endif - PUSHs (sv_2mortal (newSViv (req->result))); - PUTBACK; - call_pv ("IO::AIO::_fd2fh", G_SCALAR | G_EVAL); - SPAGAIN; - - fh = POPs; - PUSHMARK (SP); - XPUSHs (fh); + PUSHs (rv); } + break; - case REQ_GROUP: + case EIO_GROUP: req->int1 = 2; /* mark group as finished */ if (req->sv1) @@ -530,67 +557,114 @@ } break; - case REQ_NOP: - case REQ_BUSY: + case EIO_NOP: + case EIO_BUSY: break; - case REQ_READLINK: + case EIO_READLINK: + case EIO_REALPATH: if (req->result > 0) - { - SvCUR_set (req->sv2, req->result); - *SvEND (req->sv2) = 0; - PUSHs (req->sv2); - } + 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; + + if (!(PL_laststatval = req->result)) + /* if compilation fails here then perl's Stat_t is not struct _stati64 */ + PL_statcache = *(EIO_STRUCT_STAT *)(req->ptr2); + + PUSHs (sv_result); + break; + + case EIO_SEEK: + PUSHs (req->result ? sv_result : sv_2mortal (newSVval64 (req->offs))); break; - case REQ_STAT: - case REQ_LSTAT: - case REQ_FSTAT: - PL_laststype = req->type == REQ_LSTAT ? OP_LSTAT : OP_STAT; - PL_laststatval = req->result; - PL_statcache = *(Stat_t *)(req->ptr2); - PUSHs (sv_2mortal (newSViv (req->result))); + 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_result); + } break; - case REQ_READ: - SvCUR_set (req->sv2, req->stroffset + (req->result > 0 ? req->result : 0)); - *SvEND (req->sv2) = 0; - PUSHs (sv_2mortal (newSViv (req->result))); + case EIO_CUSTOM: + if (req->feed == fiemap) + { +#if HAVE_FIEMAP + if (!req->result) + { + struct fiemap *fiemap = (struct fiemap *)req->ptr1; + + if (fiemap->fm_extent_count) + { + AV *av = newAV (); + int i; + + while (fiemap->fm_mapped_extents) + { + struct fiemap_extent *extent = &fiemap->fm_extents [--fiemap->fm_mapped_extents]; + AV *ext_av = newAV (); + + av_store (ext_av, 3, newSVuv (extent->fe_flags)); + av_store (ext_av, 2, newSVval64 (extent->fe_length)); + av_store (ext_av, 1, newSVval64 (extent->fe_physical)); + av_store (ext_av, 0, newSVval64 (extent->fe_logical)); + + av_store (av, fiemap->fm_mapped_extents, newRV_noinc ((SV *)ext_av)); + } + + PUSHs (sv_2mortal (newRV_noinc ((SV *)av))); + } + else + { + SvIV_set (sv_result, fiemap->fm_mapped_extents); + PUSHs (sv_result); + } + } +#endif + } + else + PUSHs (sv_result); break; + case EIO_DUP2: /* EIO_DUP2 actually means aio_close(), so fudge result value */ + if (req->result > 0) + SvIV_set (sv_result, 0); + /* FALLTHROUGH */ + default: - PUSHs (sv_2mortal (newSViv (req->result))); + PUSHs (sv_result); break; } errno = req->errorno; PUTBACK; - call_sv (req->callback, G_VOID | G_EVAL); + call_sv (req->callback, G_VOID | G_EVAL | G_DISCARD); SPAGAIN; + if (expect_false (SvREFCNT (sv_result) != 1 || sv_result_cache)) + SvREFCNT_dec (sv_result); + else + sv_result_cache = sv_result; + 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; } - return !SvTRUE (ERRSV); + return !!SvTRUE (ERRSV); } -static void req_destroy (aio_req req) +static void +req_destroy (aio_req req) { if (req->self) { @@ -600,930 +674,590 @@ SvREFCNT_dec (req->sv1); SvREFCNT_dec (req->sv2); + SvREFCNT_dec (req->sv3); + SvREFCNT_dec (req->sv4); SvREFCNT_dec (req->callback); - if (req->flags & FLAG_PTR2_FREE) - free (req->ptr2); - - Safefree (req); + free (req); } -static void req_cancel_subs (aio_req grp) +static void +req_cancel_subs (aio_req grp) { - 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); + eio_grp_cancel (grp); } -static void req_cancel (aio_req req) +static void ecb_cold +create_respipe (void) { - req->flags |= FLAG_CANCELLED; - - req_cancel_subs (req); + if (s_epipe_renew (&respipe)) + croak ("IO::AIO: unable to initialize result pipe"); } -#ifdef USE_SOCKETS_AS_HANDLES -# define TO_SOCKET(x) (win32_get_osfhandle (x)) -#else -# define TO_SOCKET(x) (x) -#endif - static void -create_pipe (int fd[2]) -{ -#ifdef _WIN32 - int arg = 1; - if (PerlSock_socketpair (AF_UNIX, SOCK_STREAM, 0, fd) - || ioctlsocket (TO_SOCKET (fd [0]), FIONBIO, &arg) - || ioctlsocket (TO_SOCKET (fd [1]), FIONBIO, &arg)) -#else - if (pipe (fd) - || fcntl (fd [0], F_SETFL, O_NONBLOCK) - || fcntl (fd [1], F_SETFL, O_NONBLOCK)) -#endif - croak ("unable to initialize result pipe"); - - respipe_osf [0] = TO_SOCKET (respipe [0]); - respipe_osf [1] = TO_SOCKET (respipe [1]); -} - -X_THREAD_PROC (aio_proc); - -static void start_thread (void) +poll_wait (void) { - worker *wrk = calloc (1, sizeof (worker)); - - if (!wrk) - croak ("unable to allocate worker thread data"); - - X_LOCK (wrklock); - - if (thread_create (&wrk->tid, aio_proc, (void *)wrk)) - { - wrk->prev = &wrk_first; - wrk->next = wrk_first.next; - wrk_first.next->prev = wrk; - wrk_first.next = wrk; - ++started; - } - else - free (wrk); - - X_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 (); -} - -static void req_send (aio_req req) -{ - block_sig (); - - ++nreqs; - - X_LOCK (reqlock); - ++nready; - reqq_push (&req_queue, req); - X_COND_SIGNAL (reqwait); - X_UNLOCK (reqlock); - - unblock_sig (); - - 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; - - X_LOCK (reqlock); - reqq_push (&req_queue, req); - X_COND_SIGNAL (reqwait); - X_UNLOCK (reqlock); - - X_LOCK (wrklock); - --started; - X_UNLOCK (wrklock); -} - -static void set_max_idle (int nthreads) -{ - if (WORDACCESS_UNSAFE) X_LOCK (reqlock); - max_idle = nthreads <= 0 ? 1 : nthreads; - if (WORDACCESS_UNSAFE) X_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 (); -} - -static void poll_wait () -{ - fd_set rfd; - - while (nreqs) + while (eio_nreqs ()) { int size; - if (WORDACCESS_UNSAFE) X_LOCK (reslock); + + X_LOCK (reslock); size = res_queue.size; - if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); + X_UNLOCK (reslock); if (size) return; - maybe_start_thread (); - - FD_ZERO (&rfd); - FD_SET (respipe [0], &rfd); + etp_maybe_start_thread (); - PerlSock_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); - - block_sig (); - for (;;) { - for (;;) - { - maybe_start_thread (); - - X_LOCK (reslock); - req = reqq_shift (&res_queue); - - if (req) - { - --npending; - - if (!res_queue.size) - { - /* read any signals sent by the worker threads */ - char buf [4]; - while (PerlSock_recv (respipe [0], buf, 4, 0) == 4) - ; - } - } - - X_UNLOCK (reslock); - - if (!req) - break; - - --nreqs; - - if (req->type == REQ_GROUP && req->size) - { - req->int1 = 1; /* mark request as delayed */ - continue; - } - else - { - if (!req_invoke (req)) - { - req_destroy (req); - unblock_sig (); - croak (0); - } - - count++; - } - - req_destroy (req); - - if (maxreqs && !--maxreqs) - break; + int res = eio_poll (); - if (max_poll_time) - { - gettimeofday (&tv_now, 0); - - if (tvdiff (&tv_start, &tv_now) >= max_poll_time) - break; - } - } + if (res > 0) + croak (0); - if (nreqs <= max_outstanding) - break; + if (!max_outstanding || max_outstanding > eio_nreqs ()) + return res; poll_wait (); - - ++maxreqs; } - - unblock_sig (); - return count; } -/*****************************************************************************/ -/* 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 mutex_t preadwritelock = X_MUTEX_INIT; - -static ssize_t pread (int fd, void *buf, size_t count, off_t offset) +static void ecb_cold +reinit (void) { - ssize_t res; - off_t ooffset; - - X_LOCK (preadwritelock); - ooffset = lseek (fd, 0, SEEK_CUR); - lseek (fd, offset, SEEK_SET); - res = read (fd, buf, count); - lseek (fd, ooffset, SEEK_SET); - X_UNLOCK (preadwritelock); + create_respipe (); - return res; + if (eio_init (want_poll, done_poll) < 0) + croak ("IO::AIO: unable to initialise eio library"); } -static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset) -{ - ssize_t res; - off_t ooffset; - - X_LOCK (preadwritelock); - ooffset = lseek (fd, 0, SEEK_CUR); - lseek (fd, offset, SEEK_SET); - res = write (fd, buf, count); - lseek (fd, offset, SEEK_SET); - X_UNLOCK (preadwritelock); +/*****************************************************************************/ - return res; -} +#if !_POSIX_MAPPED_FILES +# define mmap(addr,length,prot,flags,fd,offs) EIO_ENOSYS () +# define munmap(addr,length) EIO_ENOSYS () #endif -#ifndef HAVE_FUTIMES - -# define utimes(path,times) aio_utimes (path, times) -# define futimes(fd,times) aio_futimes (fd, times) - -int aio_utimes (const char *filename, const struct timeval times[2]) -{ - if (times) - { - struct utimbuf buf; - - buf.actime = times[0].tv_sec; - buf.modtime = times[1].tv_sec; +#if !_POSIX_MEMORY_PROTECTION +# define mprotect(addr,len,prot) EIO_ENOSYS () +# define PROT_NONE 0 +# define PROT_WRITE 0 +# define MAP_PRIVATE 0 +# define MAP_SHARED 0 +# define MAP_FIXED 0 +#endif - return utime (filename, &buf); - } - else - return utime (filename, 0); -} +#define MMAP_MAGIC PERL_MAGIC_ext -int aio_futimes (int fd, const struct timeval tv[2]) +static int ecb_cold +mmap_free (pTHX_ SV *sv, MAGIC *mg) { - errno = ENOSYS; - return -1; -} - -#endif - -#if !HAVE_FDATASYNC -# define fdatasync fsync -#endif + int old_errno = errno; + munmap (mg->mg_ptr, (size_t)mg->mg_obj); + errno = old_errno; -#if !HAVE_READAHEAD -# define readahead(fd,offset,count) aio_readahead (fd, offset, count, self) + mg->mg_obj = 0; /* just in case */ -static ssize_t aio_readahead (int fd, off_t offset, size_t count, worker *self) -{ - size_t todo = count; - dBUF; + SvREADONLY_off (sv); - while (todo > 0) - { - size_t len = todo < AIO_BUFSIZE ? todo : AIO_BUFSIZE; + if (SvPVX (sv) != mg->mg_ptr) + croak ("ERROR: IO::AIO::mmap-mapped scalar changed location, detected"); - pread (fd, aio_buf, len, offset); - offset += len; - todo -= len; - } + SvCUR_set (sv, 0); + SvPVX (sv) = 0; + SvOK_off (sv); - errno = 0; - return count; + return 0; } -#endif +static MGVTBL mmap_vtbl = { + 0, 0, 0, 0, mmap_free +}; -#if !HAVE_READDIR_R -# define readdir_r aio_readdir_r +/*****************************************************************************/ -static mutex_t readdirlock = X_MUTEX_INIT; - -static int readdir_r (DIR *dirp, X_DIRENT *ent, X_DIRENT **res) +static SV * +get_cb (SV *cb_sv) { - X_DIRENT *e; - int errorno; - - X_LOCK (readdirlock); - - e = readdir (dirp); - errorno = errno; - - if (e) - { - *res = ent; - strcpy (ent->d_name, e->d_name); - } - else - *res = 0; - - X_UNLOCK (readdirlock); - - errno = errorno; - return e ? 0 : -1; + SvGETMAGIC (cb_sv); + return SvOK (cb_sv) ? s_get_cv_croak (cb_sv) : 0; } -#endif -/* sendfile always needs emulation */ -static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count, worker *self) +static aio_req ecb_noinline +dreq (SV *callback) { - 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; - } + SV *cb_cv; + aio_req req; + int req_pri = next_pri; + next_pri = EIO_PRI_DEFAULT; - cnt = write (ofd, aio_buf, cnt); + cb_cv = get_cb (callback); - if (cnt <= 0) - { - if (cnt && !res) res = -1; - break; - } + req = calloc (sizeof (*req), 1); + if (!req) + croak ("out of memory during eio_req allocation"); - offset += cnt; - res += cnt; - count -= cnt; - } - } + req->callback = SvREFCNT_inc (cb_cv); + req->pri = req_pri; - return res; + return req; } -/* read a full directory */ -static void scandir_ (aio_req req, worker *self) -{ - DIR *dirp; - union - { - X_DIRENT d; - char b [offsetof (X_DIRENT, d_name) + NAME_MAX + 1]; - } *u; - X_DIRENT *entp; - char *name, *names; - int memlen = 4096; - int memofs = 0; - int res = 0; - - X_LOCK (wrklock); - self->dirp = dirp = opendir (req->ptr1); - self->dbuf = u = malloc (sizeof (*u)); - req->flags |= FLAG_PTR2_FREE; - req->ptr2 = names = malloc (memlen); - X_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; - X_LOCK (wrklock); - req->ptr2 = names = realloc (names, memlen); - X_UNLOCK (wrklock); - - if (!names) - break; - } - - memcpy (names + memofs, name, len); - memofs += len; - } - } - - if (errno) - res = -1; - - req->result = res; -} +#define dREQ \ + aio_req req = dreq (callback); \ -/*****************************************************************************/ +#define REQ_SEND \ + PUTBACK; \ + req_submit (req); \ + SPAGAIN; \ + \ + if (GIMME_V != G_VOID) \ + XPUSHs (req_sv (req, aio_req_stash)); -X_THREAD_PROC (aio_proc) +ecb_inline void +req_set_path (aio_req req, SV *path, SV **wdsv, SV **pathsv, eio_wd *wd, void **ptr) { - {//D - aio_req req; - struct timespec ts; - worker *self = (worker *)thr_arg; - - /* try to distribute timeouts somewhat randomly */ - ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); - - for (;;) + if (expect_false (SvROK (path))) { - ts.tv_sec = time (0) + IDLE_TIMEOUT; - - X_LOCK (reqlock); + SV *rv = SvRV (path); + SV *wdob; - for (;;) + if (SvTYPE (rv) == SVt_PVAV && AvFILLp (rv) == 1) { - self->req = req = reqq_shift (&req_queue); - - if (req) - break; - - ++idle; + path = AvARRAY (rv)[1]; + wdob = AvARRAY (rv)[0]; - if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) - == ETIMEDOUT) + if (SvOK (wdob)) { - if (idle > max_idle) - { - --idle; - X_UNLOCK (reqlock); - X_LOCK (wrklock); - --started; - X_UNLOCK (wrklock); - goto quit; - } - - /* we are allowed to idle, so do so without any timeout */ - X_COND_WAIT (reqwait, reqlock); - ts.tv_sec = time (0) + IDLE_TIMEOUT; + *wd = SvAIO_WD (wdob); + *wdsv = SvREFCNT_inc_NN (SvRV (wdob)); } - - --idle; + else + *wd = EIO_INVALID_WD; } - - --nready; - - X_UNLOCK (reqlock); - - errno = 0; /* strictly unnecessary */ - - if (!(req->flags & FLAG_CANCELLED)) - switch (req->type) - { - case REQ_READ: req->result = req->offs >= 0 - ? pread (req->int1, req->ptr1, req->size, req->offs) - : read (req->int1, req->ptr1, req->size); break; - case REQ_WRITE: req->result = req->offs >= 0 - ? pwrite (req->int1, req->ptr1, req->size, req->offs) - : write (req->int1, req->ptr1, req->size); 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, (Stat_t *)req->ptr2); break; - case REQ_LSTAT: req->result = lstat (req->ptr1, (Stat_t *)req->ptr2); break; - case REQ_FSTAT: req->result = fstat (req->int1, (Stat_t *)req->ptr2); break; - - case REQ_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; - case REQ_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; - case REQ_CHMOD: req->result = chmod (req->ptr1, req->mode); break; - case REQ_FCHMOD: req->result = fchmod (req->int1, req->mode); break; - case REQ_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; - case REQ_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); 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_MKDIR: req->result = mkdir (req->ptr1, req->mode); 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: -#ifdef _WIN32 - Sleep (req->nv1 * 1000.); -#else - { - struct timeval tv; - - tv.tv_sec = req->nv1; - tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; - - req->result = select (0, 0, 0, 0, &tv); - } -#endif - break; - - case REQ_UTIME: - case REQ_FUTIME: - { - struct timeval tv[2]; - struct timeval *times; - - if (req->nv1 != -1. || req->nv2 != -1.) - { - tv[0].tv_sec = req->nv1; - tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; - tv[1].tv_sec = req->nv2; - tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; - - times = tv; - } - else - times = 0; - - - req->result = req->type == REQ_FUTIME - ? futimes (req->int1, times) - : utimes (req->ptr1, times); - } - - case REQ_GROUP: - case REQ_NOP: - break; - - case REQ_QUIT: - goto quit; - - default: - req->result = -1; - break; - } - - req->errorno = errno; - - X_LOCK (reslock); - - ++npending; - - if (!reqq_push (&res_queue, req)) + else if (SvTYPE (rv) == SVt_PVMG && SvSTASH (rv) == aio_wd_stash) { - /* write a dummy byte to the pipe so fh becomes ready */ - send (respipe_osf [1], (const void *)&respipe_osf, 1, 0); - - /* optionally signal the main thread asynchronously */ - if (main_sig) - pthread_kill (main_tid, main_sig); + *wd = (aio_wd)(long)SvIVX (rv); + *wdsv = SvREFCNT_inc_NN (rv); + *ptr = "."; + return; /* path set to "." */ } - - self->req = 0; - worker_clear (self); - - X_UNLOCK (reslock); + else + croak ("IO::AIO: pathname arguments must be specified as a string, an IO::AIO::WD object or a [IO::AIO::WD, path] pair"); } -quit: - X_LOCK (wrklock); - worker_free (self); - X_UNLOCK (wrklock); - - return 0; - }//D + *pathsv = newSVsv (path); + *ptr = SvPVbyte_nolen (*pathsv); } -/*****************************************************************************/ - -static void atfork_prepare (void) +static void ecb_noinline +req_set_path1 (aio_req req, SV *path) { - X_LOCK (wrklock); - X_LOCK (reqlock); - X_LOCK (reslock); -#if !HAVE_PREADWRITE - X_LOCK (preadwritelock); -#endif -#if !HAVE_READDIR_R - X_LOCK (readdirlock); -#endif + req_set_path (req, path, &req->sv1, &req->sv3, &req->wd, &req->ptr1); } -static void atfork_parent (void) +static void ecb_noinline +req_set_fh_or_path (aio_req req, int type_path, int type_fh, SV *fh_or_path) { -#if !HAVE_READDIR_R - X_UNLOCK (readdirlock); -#endif -#if !HAVE_PREADWRITE - X_UNLOCK (preadwritelock); -#endif - X_UNLOCK (reslock); - X_UNLOCK (reqlock); - X_UNLOCK (wrklock); -} - -static void atfork_child (void) -{ - aio_req prv; + SV *rv = SvROK (fh_or_path) ? SvRV (fh_or_path) : fh_or_path; - while (prv = reqq_shift (&req_queue)) - req_destroy (prv); - - while (prv = reqq_shift (&res_queue)) - req_destroy (prv); - - while (wrk_first.next != &wrk_first) + switch (SvTYPE (rv)) { - worker *wrk = wrk_first.next; - - if (wrk->req) - req_destroy (wrk->req); + case SVt_PVIO: + case SVt_PVLV: + case SVt_PVGV: + req->type = type_fh; + req->sv1 = newSVsv (fh_or_path); + req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path))); + break; - worker_clear (wrk); - worker_free (wrk); + default: + req->type = type_path; + req_set_path1 (req, fh_or_path); + break; } - - started = 0; - idle = 0; - nreqs = 0; - nready = 0; - npending = 0; - - PerlSock_closesocket (respipe [0]); - PerlSock_closesocket (respipe [1]); - - create_pipe (respipe); - - atfork_parent (); } -#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"); \ - \ - Newz (0, req, 1, aio_cb); \ - if (!req) \ - croak ("out of memory during aio_req allocation"); \ - \ - req->callback = newSVsv (callback); \ - req->pri = req_pri +XS(boot_IO__AIO) ecb_cold; -#define REQ_SEND \ - req_send (req); \ - \ - 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_niv(name, value) { # name, (IV) value }, +# define const_iv(name) { # name, (IV) name }, +# define const_eio(name) { # name, (IV) EIO_ ## name }, + + /* you have to re-run ./gendef0 after adding/Removing any constants here */ + + const_iv (ENOSYS) + const_iv (EXDEV) + const_iv (EBADR) + + const_iv (O_RDONLY) + const_iv (O_WRONLY) + const_iv (O_RDWR) + const_iv (O_CREAT) + const_iv (O_TRUNC) + const_iv (O_EXCL) + const_iv (O_APPEND) + + const_iv (O_ASYNC) + const_iv (O_DIRECT) + const_iv (O_NOATIME) + + const_iv (O_CLOEXEC) + const_iv (O_NOCTTY) + const_iv (O_NOFOLLOW) + const_iv (O_NONBLOCK) + const_iv (O_EXEC) + const_iv (O_SEARCH) + const_iv (O_DIRECTORY) + const_iv (O_DSYNC) + const_iv (O_RSYNC) + const_iv (O_SYNC) + const_iv (O_TTY_INIT) + + const_iv (S_IFIFO) + const_iv (S_IFCHR) + const_iv (S_IFBLK) + const_iv (S_IFLNK) + const_iv (S_IFREG) + const_iv (S_IFDIR) + const_iv (S_IFWHT) + const_iv (S_IFSOCK) + const_iv (S_IFMT) + + const_niv (FADV_NORMAL , POSIX_FADV_NORMAL) + const_niv (FADV_SEQUENTIAL, POSIX_FADV_SEQUENTIAL) + const_niv (FADV_RANDOM , POSIX_FADV_RANDOM) + const_niv (FADV_NOREUSE , POSIX_FADV_NOREUSE) + const_niv (FADV_WILLNEED , POSIX_FADV_WILLNEED) + const_niv (FADV_DONTNEED , POSIX_FADV_DONTNEED) + + const_niv (MADV_NORMAL , POSIX_MADV_NORMAL) + const_niv (MADV_SEQUENTIAL, POSIX_MADV_SEQUENTIAL) + const_niv (MADV_RANDOM , POSIX_MADV_RANDOM) + const_niv (MADV_WILLNEED , POSIX_MADV_WILLNEED) + const_niv (MADV_DONTNEED , POSIX_MADV_DONTNEED) + + const_iv (ST_RDONLY) + const_iv (ST_NOSUID) + const_iv (ST_NODEV) + const_iv (ST_NOEXEC) + const_iv (ST_SYNCHRONOUS) + const_iv (ST_MANDLOCK) + const_iv (ST_WRITE) + const_iv (ST_APPEND) + const_iv (ST_IMMUTABLE) + const_iv (ST_NOATIME) + const_iv (ST_NODIRATIME) + const_iv (ST_RELATIME) + + const_iv (PROT_NONE) + const_iv (PROT_EXEC) + const_iv (PROT_READ) + const_iv (PROT_WRITE) + + /*const_iv (MAP_FIXED)*/ + const_iv (MAP_PRIVATE) + const_iv (MAP_SHARED) + const_iv (MAP_ANONYMOUS) + + /* linuxish */ + const_iv (MAP_HUGETLB) + const_iv (MAP_LOCKED) + const_iv (MAP_NORESERVE) + const_iv (MAP_POPULATE) + const_iv (MAP_NONBLOCK) + + const_iv (FIEMAP_FLAG_SYNC) + const_iv (FIEMAP_FLAG_XATTR) + const_iv (FIEMAP_FLAGS_COMPAT) + const_iv (FIEMAP_EXTENT_LAST) + const_iv (FIEMAP_EXTENT_UNKNOWN) + const_iv (FIEMAP_EXTENT_DELALLOC) + const_iv (FIEMAP_EXTENT_ENCODED) + const_iv (FIEMAP_EXTENT_DATA_ENCRYPTED) + const_iv (FIEMAP_EXTENT_NOT_ALIGNED) + const_iv (FIEMAP_EXTENT_DATA_INLINE) + const_iv (FIEMAP_EXTENT_DATA_TAIL) + const_iv (FIEMAP_EXTENT_UNWRITTEN) + const_iv (FIEMAP_EXTENT_MERGED) + const_iv (FIEMAP_EXTENT_SHARED) + + const_iv (SPLICE_F_MOVE) + const_iv (SPLICE_F_NONBLOCK) + const_iv (SPLICE_F_MORE) + const_iv (SPLICE_F_GIFT) + + const_iv (SEEK_DATA) + const_iv (SEEK_HOLE) + + /* libeio constants */ + const_eio (SEEK_SET) + const_eio (SEEK_CUR) + const_eio (SEEK_END) + + const_eio (MCL_FUTURE) + const_eio (MCL_CURRENT) + + const_eio (MS_ASYNC) + const_eio (MS_INVALIDATE) + const_eio (MS_SYNC) + + const_eio (MT_MODIFY) + + const_eio (SYNC_FILE_RANGE_WAIT_BEFORE) + const_eio (SYNC_FILE_RANGE_WRITE) + const_eio (SYNC_FILE_RANGE_WAIT_AFTER) + + const_eio (FALLOC_FL_KEEP_SIZE) + const_eio (FALLOC_FL_PUNCH_HOLE) + + 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) + }; + + aio_stash = gv_stashpv ("IO::AIO" , 1); + aio_req_stash = gv_stashpv ("IO::AIO::REQ", 1); + aio_grp_stash = gv_stashpv ("IO::AIO::GRP", 1); + aio_wd_stash = gv_stashpv ("IO::AIO::WD" , 1); + + for (civ = const_iv + sizeof (const_iv) / sizeof (const_iv [0]); civ > const_iv; civ--) + newCONSTSUB (aio_stash, (char *)civ[-1].name, newSViv (civ[-1].iv)); - 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)); -#ifdef _WIN32 - X_MUTEX_CHECK (wrklock); - X_MUTEX_CHECK (reslock); - X_MUTEX_CHECK (reqlock); - X_MUTEX_CHECK (reqwait); - X_MUTEX_CHECK (preadwritelock); - X_MUTEX_CHECK (readdirlock); + newCONSTSUB (aio_stash, "PAGESIZE", newSViv (PAGESIZE)); - X_COND_CHECK (reqwait); -#else - newCONSTSUB (stash, "S_IFIFO", newSViv (S_IFIFO)); - newCONSTSUB (stash, "SIGIO", newSViv (SIGIO)); -#endif - - create_pipe (respipe); - - X_THREAD_ATFORK (atfork_prepare, atfork_parent, atfork_child); + reinit (); } void -max_poll_reqs (int nreqs) +reinit () + PROTOTYPE: + +void +max_poll_reqs (unsigned 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 (unsigned int nthreads) + PROTOTYPE: $ + CODE: + eio_set_min_parallel (nthreads); void -min_parallel (int nthreads) +max_parallel (unsigned int nthreads) PROTOTYPE: $ + CODE: + eio_set_max_parallel (nthreads); void -max_parallel (int nthreads) +max_idle (unsigned int nthreads) PROTOTYPE: $ + CODE: + eio_set_max_idle (nthreads); void -max_idle (int nthreads) +idle_timeout (unsigned int seconds) PROTOTYPE: $ CODE: - set_max_idle (nthreads); + eio_set_idle_timeout (seconds); -int -max_outstanding (int maxreqs) +void +max_outstanding (unsigned int maxreqs) PROTOTYPE: $ CODE: - RETVAL = max_outstanding; max_outstanding = maxreqs; - OUTPUT: - RETVAL + +void +aio_wd (SV8 *pathname, SV *callback=&PL_sv_undef) + PPCODE: +{ + dREQ; + + req->type = EIO_WD_OPEN; + req_set_path1 (req, pathname); + + REQ_SEND; +} void aio_open (SV8 *pathname, int flags, int mode, SV *callback=&PL_sv_undef) - PROTOTYPE: $$$;$ PPCODE: { dREQ; - req->type = REQ_OPEN; - req->sv1 = newSVsv (pathname); - req->ptr1 = SvPVbyte_nolen (req->sv1); + req->type = EIO_OPEN; + req_set_path1 (req, pathname); req->int1 = flags; - req->mode = mode; + req->int2 = mode; REQ_SEND; } void -aio_close (SV *fh, SV *callback=&PL_sv_undef) - PROTOTYPE: $;$ +aio_fsync (SV *fh, SV *callback=&PL_sv_undef) ALIAS: - aio_close = REQ_CLOSE - aio_fsync = REQ_FSYNC - aio_fdatasync = REQ_FDATASYNC + aio_fsync = EIO_FSYNC + aio_fdatasync = EIO_FDATASYNC + aio_syncfs = EIO_SYNCFS PPCODE: { + int fd = s_fileno_croak (fh, 0); dREQ; req->type = ix; req->sv1 = newSVsv (fh); - req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh))); + req->int1 = fd; + + REQ_SEND; +} + +void +aio_sync_file_range (SV *fh, off_t offset, size_t nbytes, UV flags, SV *callback=&PL_sv_undef) + 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); + REQ_SEND; +} + +void +aio_allocate (SV *fh, int mode, off_t offset, size_t len, SV *callback=&PL_sv_undef) + PPCODE: +{ + int fd = s_fileno_croak (fh, 0); + dREQ; + + req->type = EIO_FALLOCATE; + req->sv1 = newSVsv (fh); + req->int1 = fd; + req->int2 = mode; + req->offs = offset; + req->size = len; + + REQ_SEND; +} + +void +aio_close (SV *fh, SV *callback=&PL_sv_undef) + PPCODE: +{ + static int close_fd = -1; /* dummy fd to close fds via dup2 */ + int fd = s_fileno_croak (fh, 0); + dREQ; + + if (expect_false (close_fd < 0)) + { + int pipefd [2]; + + if ( +#ifdef _WIN32 + _pipe (pipefd, 1, _O_BINARY) < 0 +#else + pipe (pipefd) < 0 + || fcntl (pipefd [0], F_SETFD, FD_CLOEXEC) < 0 +#endif + || close (pipefd [1]) < 0 + ) + abort (); /*D*/ + + close_fd = pipefd [0]; + } + + req->type = EIO_DUP2; + req->int1 = close_fd; + req->sv2 = newSVsv (fh); + req->int2 = fd; + + REQ_SEND; +} + +void +aio_seek (SV *fh, SV *offset, int whence, SV *callback=&PL_sv_undef) + PPCODE: +{ + int fd = s_fileno_croak (fh, 0); + dREQ; + + req->type = EIO_SEEK; + req->sv1 = newSVsv (fh); + req->int1 = fd; + req->offs = SvVAL64 (offset); + req->int2 = whence; + + REQ_SEND; } 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 - PROTOTYPE: $$$$$;$ + aio_read = EIO_READ + aio_write = EIO_WRITE PPCODE: { STRLEN svlen; + int fd = s_fileno_croak (fh, ix == EIO_WRITE); char *svptr = SvPVbyte (data, svlen); UV len = SvUV (length); - SvUPGRADE (data, SVt_PV); - SvPOK_on (data); - if (dataoffset < 0) dataoffset += svlen; if (dataoffset < 0 || dataoffset > svlen) croak ("dataoffset outside of data scalar"); - if (ix == REQ_WRITE) + if (ix == EIO_WRITE) { /* write: check length and adjust. */ if (!SvOK (length) || len + dataoffset > svlen) @@ -1531,24 +1265,24 @@ } else { - /* read: grow scalar as necessary */ - svptr = SvGROW (data, len + dataoffset + 1); + /* read: check type and grow scalar as necessary */ + SvUPGRADE (data, SVt_PV); + if (SvLEN (data) >= SvCUR (data)) + svptr = SvGROW (data, len + dataoffset + 1); + else if (SvCUR (data) < len + dataoffset) + croak ("length + dataoffset outside of scalar, and cannot grow"); } - if (len < 0) - croak ("length must not be negative"); - { dREQ; req->type = ix; req->sv1 = newSVsv (fh); - req->int1 = PerlIO_fileno (ix == REQ_READ ? IoIFP (sv_2io (fh)) - : IoOFP (sv_2io (fh))); + req->int1 = fd; req->offs = SvOK (offset) ? SvVAL64 (offset) : -1; req->size = len; req->sv2 = SvREFCNT_inc (data); - req->ptr1 = (char *)svptr + dataoffset; + req->ptr2 = (char *)svptr + dataoffset; req->stroffset = dataoffset; if (!SvREADONLY (data)) @@ -1562,54 +1296,50 @@ } void -aio_readlink (SV8 *path, SV *callback=&PL_sv_undef) - PROTOTYPE: $$;$ +aio_readlink (SV8 *pathname, SV *callback=&PL_sv_undef) + ALIAS: + aio_readlink = EIO_READLINK + aio_realpath = EIO_REALPATH PPCODE: { - SV *data; dREQ; - data = newSV (NAME_MAX); - SvPOK_on (data); - - req->type = REQ_READLINK; - req->sv1 = newSVsv (path); - req->ptr2 = SvPVbyte_nolen (req->sv1); - req->sv2 = data; - req->ptr1 = SvPVbyte_nolen (data); + req->type = ix; + req_set_path1 (req, pathname); REQ_SEND; } void -aio_sendfile (SV *out_fh, SV *in_fh, SV *in_offset, UV length, SV *callback=&PL_sv_undef) - PROTOTYPE: $$$$;$ +aio_sendfile (SV *out_fh, SV *in_fh, off_t in_offset, size_t length, SV *callback=&PL_sv_undef) PPCODE: { + int ifd = s_fileno_croak (in_fh , 0); + int ofd = s_fileno_croak (out_fh, 1); dREQ; - req->type = REQ_SENDFILE; + req->type = EIO_SENDFILE; req->sv1 = newSVsv (out_fh); - req->int1 = PerlIO_fileno (IoIFP (sv_2io (out_fh))); + req->int1 = ofd; req->sv2 = newSVsv (in_fh); - req->int2 = PerlIO_fileno (IoIFP (sv_2io (in_fh))); - req->offs = SvVAL64 (in_offset); + req->int2 = ifd; + req->offs = in_offset; req->size = length; REQ_SEND; } void -aio_readahead (SV *fh, SV *offset, IV length, SV *callback=&PL_sv_undef) - PROTOTYPE: $$$;$ +aio_readahead (SV *fh, off_t offset, size_t length, SV *callback=&PL_sv_undef) PPCODE: { + int fd = s_fileno_croak (fh, 0); dREQ; - req->type = REQ_READAHEAD; + req->type = EIO_READAHEAD; req->sv1 = newSVsv (fh); - req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh))); - req->offs = SvVAL64 (offset); + req->int1 = fd; + req->offs = offset; req->size = length; REQ_SEND; @@ -1618,36 +1348,34 @@ void 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 + aio_statvfs = EIO_STATVFS PPCODE: { dREQ; - req->ptr2 = malloc (sizeof (Stat_t)); - if (!req->ptr2) - { - req_destroy (req); - croak ("out of memory during aio_stat statdata allocation"); - } - - req->flags |= FLAG_PTR2_FREE; - req->sv1 = newSVsv (fh_or_path); - - if (SvPOK (fh_or_path)) - { - req->type = ix; - req->ptr1 = SvPVbyte_nolen (req->sv1); - } - else - { - req->type = REQ_FSTAT; - req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path))); - } + req_set_fh_or_path (req, ix, ix == EIO_STATVFS ? EIO_FSTATVFS : EIO_FSTAT, fh_or_path); REQ_SEND; } +UV +major (UV dev) + ALIAS: + minor = 1 + CODE: + RETVAL = ix ? major (dev) : minor (dev); + OUTPUT: + RETVAL + +UV +makedev (UV maj, UV min) + CODE: + RETVAL = makedev (maj, min); + OUTPUT: + RETVAL + void aio_utime (SV8 *fh_or_path, SV *atime, SV *mtime, SV *callback=&PL_sv_undef) PPCODE: @@ -1656,18 +1384,7 @@ req->nv1 = SvOK (atime) ? SvNV (atime) : -1.; req->nv2 = SvOK (mtime) ? SvNV (mtime) : -1.; - req->sv1 = newSVsv (fh_or_path); - - if (SvPOK (fh_or_path)) - { - req->type = REQ_UTIME; - req->ptr1 = SvPVbyte_nolen (req->sv1); - } - else - { - req->type = REQ_FUTIME; - req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path))); - } + req_set_fh_or_path (req, EIO_UTIME, EIO_FUTIME, fh_or_path); REQ_SEND; } @@ -1678,19 +1395,8 @@ { dREQ; - req->sv1 = newSVsv (fh_or_path); req->offs = SvOK (offset) ? SvVAL64 (offset) : -1; - - if (SvPOK (fh_or_path)) - { - req->type = REQ_TRUNCATE; - req->ptr1 = SvPVbyte_nolen (req->sv1); - } - else - { - req->type = REQ_FTRUNCATE; - req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path))); - } + req_set_fh_or_path (req, EIO_TRUNCATE, EIO_FTRUNCATE, fh_or_path); REQ_SEND; } @@ -1701,19 +1407,8 @@ { dREQ; - req->mode = mode; - req->sv1 = newSVsv (fh_or_path); - - if (SvPOK (fh_or_path)) - { - req->type = REQ_CHMOD; - req->ptr1 = SvPVbyte_nolen (req->sv1); - } - else - { - req->type = REQ_FCHMOD; - req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path))); - } + req->int2 = mode; + req_set_fh_or_path (req, EIO_CHMOD, EIO_FCHMOD, fh_or_path); REQ_SEND; } @@ -1726,35 +1421,24 @@ req->int2 = SvOK (uid) ? SvIV (uid) : -1; req->int3 = SvOK (gid) ? SvIV (gid) : -1; - req->sv1 = newSVsv (fh_or_path); - - if (SvPOK (fh_or_path)) - { - req->type = REQ_CHOWN; - req->ptr1 = SvPVbyte_nolen (req->sv1); - } - else - { - req->type = REQ_FCHOWN; - req->int1 = PerlIO_fileno (IoIFP (sv_2io (fh_or_path))); - } + req_set_fh_or_path (req, EIO_CHOWN, EIO_FCHOWN, fh_or_path); 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_readdirx (SV8 *pathname, IV flags, SV *callback=&PL_sv_undef) PPCODE: { dREQ; - req->type = ix; - req->sv1 = newSVsv (pathname); - req->ptr1 = SvPVbyte_nolen (req->sv1); + req->type = EIO_READDIR; + req->int1 = flags | EIO_READDIR_DENTS | EIO_READDIR_CUSTOM1; + + if (flags & EIO_READDIR_DENTS) + req->int1 |= EIO_READDIR_CUSTOM2; + + req_set_path1 (req, pathname); REQ_SEND; } @@ -1764,11 +1448,26 @@ PPCODE: { dREQ; + + req->type = EIO_MKDIR; + req->int2 = mode; + req_set_path1 (req, pathname); + + REQ_SEND; +} + +void +aio_unlink (SV8 *pathname, SV *callback=&PL_sv_undef) + ALIAS: + aio_unlink = EIO_UNLINK + aio_rmdir = EIO_RMDIR + aio_readdir = EIO_READDIR + PPCODE: +{ + dREQ; - req->type = REQ_MKDIR; - req->sv1 = newSVsv (pathname); - req->ptr1 = SvPVbyte_nolen (req->sv1); - req->mode = mode; + req->type = ix; + req_set_path1 (req, pathname); REQ_SEND; } @@ -1776,18 +1475,18 @@ void 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: { + eio_wd wd2 = 0; dREQ; req->type = ix; - req->sv2 = newSVsv (oldpath); - req->ptr2 = SvPVbyte_nolen (req->sv2); - req->sv1 = newSVsv (newpath); - req->ptr1 = SvPVbyte_nolen (req->sv1); + req_set_path1 (req, oldpath); + req_set_path (req, newpath, &req->sv2, &req->sv4, &wd2, &req->ptr2); + req->int3 = (long)wd2; REQ_SEND; } @@ -1798,22 +1497,118 @@ { dREQ; - req->type = REQ_MKNOD; - req->sv1 = newSVsv (pathname); - req->ptr1 = SvPVbyte_nolen (req->sv1); - req->mode = (mode_t)mode; + req->type = EIO_MKNOD; + req->int2 = (mode_t)mode; req->offs = dev; + req_set_path1 (req, pathname); REQ_SEND; } void +aio_mtouch (SV8 *data, IV offset = 0, SV *length = &PL_sv_undef, int flags = 0, SV *callback=&PL_sv_undef) + ALIAS: + aio_mtouch = EIO_MTOUCH + aio_msync = EIO_MSYNC + PPCODE: +{ + STRLEN svlen; + char *svptr = SvPVbyte (data, svlen); + UV len = SvUV (length); + + if (offset < 0) + offset += svlen; + + if (offset < 0 || offset > svlen) + croak ("offset outside of scalar"); + + if (!SvOK (length) || len + offset > svlen) + len = svlen - offset; + + { + dREQ; + + req->type = ix; + req->sv2 = SvREFCNT_inc (data); + req->ptr2 = (char *)svptr + offset; + req->size = len; + req->int1 = flags; + + REQ_SEND; + } +} + +void +aio_mlock (SV8 *data, IV offset = 0, SV *length = &PL_sv_undef, SV *callback=&PL_sv_undef) + PPCODE: +{ + STRLEN svlen; + char *svptr = SvPVbyte (data, svlen); + UV len = SvUV (length); + + if (offset < 0) + offset += svlen; + + if (offset < 0 || offset > svlen) + croak ("offset outside of scalar"); + + if (!SvOK (length) || len + offset > svlen) + len = svlen - offset; + + { + dREQ; + + req->type = EIO_MLOCK; + req->sv2 = SvREFCNT_inc (data); + req->ptr2 = (char *)svptr + offset; + req->size = len; + + REQ_SEND; + } +} + +void +aio_mlockall (IV flags, SV *callback=&PL_sv_undef) + PPCODE: +{ + dREQ; + + req->type = EIO_MLOCKALL; + req->int1 = flags; + + REQ_SEND; +} + +void +aio_fiemap (SV *fh, off_t start, SV *length, U32 flags, SV *count, SV *callback=&PL_sv_undef) + PPCODE: +{ + int fd = s_fileno_croak (fh, 0); + dREQ; + + req->type = EIO_CUSTOM; + req->sv1 = newSVsv (fh); + req->int1 = fd; + + req->feed = fiemap; +#if HAVE_FIEMAP + /* keep our fingers crossed that the next two types are 64 bit */ + req->offs = start; + req->size = SvOK (length) ? SvVAL64 (length) : ~0ULL; + req->int2 = flags; + req->int3 = SvOK (count) ? SvIV (count) : -1; +#endif + + REQ_SEND; +} + +void aio_busy (double delay, SV *callback=&PL_sv_undef) PPCODE: { dREQ; - req->type = REQ_BUSY; + req->type = EIO_BUSY; req->nv1 = delay < 0. ? 0. : delay; REQ_SEND; @@ -1821,38 +1616,42 @@ void aio_group (SV *callback=&PL_sv_undef) - PROTOTYPE: ;$ PPCODE: { dREQ; - req->type = REQ_GROUP; + req->type = EIO_GROUP; + + PUTBACK; + req_submit (req); + SPAGAIN; - req_send (req); - XPUSHs (req_sv (req, AIO_GRP_KLASS)); + XPUSHs (req_sv (req, aio_grp_stash)); } void 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; } int 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 @@ -1861,23 +1660,21 @@ 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 (); } int -poll() - PROTOTYPE: +poll () CODE: poll_wait (); RETVAL = poll_cb (); @@ -1885,15 +1682,14 @@ RETVAL int -poll_fileno() - PROTOTYPE: +poll_fileno () CODE: - RETVAL = respipe [0]; + RETVAL = s_epipe_fd (&respipe); OUTPUT: RETVAL int -poll_cb(...) +poll_cb (...) PROTOTYPE: CODE: RETVAL = poll_cb (); @@ -1901,96 +1697,261 @@ RETVAL void -poll_wait() - PROTOTYPE: +poll_wait () CODE: poll_wait (); -void -setsig (int signum = SIGIO) - PROTOTYPE: ;$ +int +nreqs () + CODE: + RETVAL = eio_nreqs (); + OUTPUT: + RETVAL + +int +nready () + CODE: + RETVAL = eio_nready (); + OUTPUT: + RETVAL + +int +npending () + CODE: + RETVAL = eio_npending (); + OUTPUT: + RETVAL + +int +nthreads () + CODE: + RETVAL = eio_nthreads (); + OUTPUT: + RETVAL + +int +fadvise (aio_rfd fh, off_t offset, off_t length, IV advice) + CODE: + RETVAL = posix_fadvise (fh, offset, length, advice); + OUTPUT: + RETVAL + +IV +sendfile (aio_wfd ofh, aio_rfd ifh, off_t offset, size_t count) CODE: + RETVAL = eio_sendfile_sync (ofh, ifh, offset, count); + OUTPUT: + RETVAL + +void +mmap (SV *scalar, size_t length, int prot, int flags, SV *fh = &PL_sv_undef, off_t offset = 0) + PPCODE: + sv_unmagic (scalar, MMAP_MAGIC); { - if (block_sig_level) - croak ("cannot call IO::AIO::setsig from within aio_block/callback"); + int fd = SvOK (fh) ? s_fileno_croak (fh, flags & PROT_WRITE) : -1; + void *addr = (void *)mmap (0, length, prot, flags, fd, offset); + if (addr == (void *)-1) + XSRETURN_NO; - X_LOCK (reslock); - main_tid = pthread_self (); - main_sig = signum; - X_UNLOCK (reslock); + sv_force_normal (scalar); - if (main_sig && npending) - pthread_kill (main_tid, main_sig); + /* we store the length in mg_obj, as namlen is I32 :/ */ + sv_magicext (scalar, 0, MMAP_MAGIC, &mmap_vtbl, (char *)addr, 0) + ->mg_obj = (SV *)length; + + SvUPGRADE (scalar, SVt_PV); /* nop... */ + + if (!(prot & PROT_WRITE)) + SvREADONLY_on (scalar); + + if (SvLEN (scalar)) + Safefree (SvPVX (scalar)); + + SvPVX (scalar) = (char *)addr; + SvCUR_set (scalar, length); + SvLEN_set (scalar, 0); + SvPOK_only (scalar); + + XSRETURN_YES; } void -aio_block (SV *cb) - PROTOTYPE: & - PPCODE: +munmap (SV *scalar) + CODE: + sv_unmagic (scalar, MMAP_MAGIC); + +int +madvise (SV *scalar, off_t offset = 0, SV *length = &PL_sv_undef, IV advice_or_prot) + ALIAS: + mprotect = 1 + CODE: { - int count; + STRLEN svlen; + void *addr = SvPVbyte (scalar, svlen); + size_t len = SvUV (length); - block_sig (); - PUSHMARK (SP); - PUTBACK; - count = call_sv (cb, GIMME_V | G_NOARGS | G_EVAL); - SPAGAIN; - unblock_sig (); + if (offset < 0) + offset += svlen; - if (SvTRUE (ERRSV)) - croak (0); + if (offset < 0 || offset > svlen) + croak ("offset outside of scalar"); + + if (!SvOK (length) || len + offset > svlen) + len = svlen - offset; + + addr = (void *)(((intptr_t)addr) + offset); + eio_page_align (&addr, &len); + + switch (ix) + { + case 0: RETVAL = posix_madvise (addr, len, advice_or_prot); break; + case 1: RETVAL = mprotect (addr, len, advice_or_prot); break; + } +} + OUTPUT: + RETVAL - XSRETURN (count); +int +munlock (SV *scalar, off_t offset = 0, SV *length = &PL_sv_undef) + CODE: +{ + STRLEN svlen; + void *addr = SvPVbyte (scalar, svlen); + size_t len = SvUV (length); + + if (offset < 0) + offset += svlen; + + if (offset < 0 || offset > svlen) + croak ("offset outside of scalar"); + + if (!SvOK (length) || len + offset > svlen) + len = svlen - offset; + + addr = (void *)(((intptr_t)addr) + offset); + eio_page_align (&addr, &len); +#if _POSIX_MEMLOCK_RANGE + RETVAL = munlock (addr, len); +#else + RETVAL = EIO_ENOSYS (); +#endif } + OUTPUT: + RETVAL int -nreqs() - PROTOTYPE: - CODE: - RETVAL = nreqs; - OUTPUT: - RETVAL +munlockall () + CODE: +#if _POSIX_MEMLOCK + munlockall (); +#else + RETVAL = EIO_ENOSYS (); +#endif + OUTPUT: + RETVAL int -nready() - PROTOTYPE: - CODE: - RETVAL = get_nready (); +splice (aio_rfd rfh, SV *off_in, aio_wfd wfh, SV *off_out, size_t length, unsigned int flags) + CODE: +{ +#if HAVE_LINUX_SPLICE + loff_t off_in_, off_out_; + RETVAL = splice ( + rfh, SvOK (off_in ) ? (off_in_ = SvVAL64 (off_in )), &off_in_ : 0, + wfh, SvOK (off_out) ? (off_out_ = SvVAL64 (off_out)), &off_out_ : 0, + length, flags + ); +#else + RETVAL = EIO_ENOSYS (); +#endif +} OUTPUT: - RETVAL + RETVAL int -npending() - PROTOTYPE: - CODE: - RETVAL = get_npending (); +tee (aio_rfd rfh, aio_wfd wfh, size_t length, unsigned int flags) + CODE: +#if HAVE_LINUX_SPLICE + RETVAL = tee (rfh, wfh, length, flags); +#else + RETVAL = EIO_ENOSYS (); +#endif OUTPUT: - RETVAL + RETVAL int -nthreads() - PROTOTYPE: +pipesize (aio_rfd rfh, int new_size = -1) + PROTOTYPE: $;$ + CODE: +#if defined(F_SETPIPE_SZ) && defined(F_GETPIPE_SZ) + if (new_size >= 0) + RETVAL = fcntl (rfh, F_SETPIPE_SZ, new_size); + else + RETVAL = fcntl (rfh, F_GETPIPE_SZ); +#else + errno = ENOSYS; + RETVAL = -1; +#endif + OUTPUT: + RETVAL + +void _on_next_submit (SV *cb) CODE: - if (WORDACCESS_UNSAFE) X_LOCK (wrklock); - RETVAL = started; - if (WORDACCESS_UNSAFE) X_UNLOCK (wrklock); - OUTPUT: - RETVAL + SvREFCNT_dec (on_next_submit); + on_next_submit = SvOK (cb) ? newSVsv (cb) : 0; PROTOTYPES: DISABLE +MODULE = IO::AIO PACKAGE = IO::AIO::WD + +BOOT: +{ + newCONSTSUB (aio_stash, "CWD" , newSVaio_wd (EIO_CWD )); + newCONSTSUB (aio_stash, "INVALID_WD", newSVaio_wd (EIO_INVALID_WD)); +} + +void +DESTROY (SV *self) + CODE: +{ + aio_wd wd = SvAIO_WD (self); +#if HAVE_AT + { + SV *callback = &PL_sv_undef; + dREQ; /* clobbers next_pri :/ */ + next_pri = req->pri; /* restore next_pri */ + req->pri = EIO_PRI_MAX; /* better use max. priority to conserve fds */ + req->type = EIO_WD_CLOSE; + req->wd = wd; + REQ_SEND; + } +#else + eio_wd_close_sync (wd); +#endif +} + MODULE = IO::AIO PACKAGE = IO::AIO::REQ 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 @@ -1999,34 +1960,21 @@ PPCODE: { int i; - aio_req req; - - if (main_sig && !block_sig_level) - croak ("aio_group->add called outside aio_block/callback context while IO::AIO::setsig is in use"); 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); } } @@ -2045,6 +1993,7 @@ grp->errorno = errno; av = newAV (); + av_extend (av, items - 1); for (i = 1; i < items; ++i ) av_push (av, newSVsv (ST (i))); @@ -2061,19 +2010,19 @@ 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); }