--- libeio/eio.c 2010/01/08 09:29:56 1.52 +++ libeio/eio.c 2011/09/26 20:19:08 1.106 @@ -1,7 +1,7 @@ /* * libeio implementation * - * Copyright (c) 2007,2008,2009,2010 Marc Alexander Lehmann + * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann * All rights reserved. * * Redistribution and use in source and binary forms, with or without modifica- @@ -37,10 +37,15 @@ * either the BSD or the GPL. */ +#ifndef _WIN32 +# include "config.h" +#endif + #include "eio.h" +#include "ecb.h" #ifdef EIO_STACKSIZE -# define XTHREAD_STACKSIZE EIO_STACKSIZE +# define X_STACKSIZE EIO_STACKSIZE #endif #include "xthread.h" @@ -51,11 +56,35 @@ #include #include #include -#include #include #include #include +#if _POSIX_VERSION >= 200809L +# define HAVE_AT 1 +#else +# define HAVE_AT 0 +#endif + +/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */ +/* intptr_t only comes from stdint.h, says idiot openbsd coder */ +#if HAVE_STDINT_H +# include +#endif + +#ifndef ECANCELED +# define ECANCELED EDOM +#endif +#ifndef ELOOP +# define ELOOP EDOM +#endif + +#if !defined(ENOTSOCK) && defined(WSAENOTSOCK) +# define ENOTSOCK WSAENOTSOCK +#endif + +static void eio_destroy (eio_req *req); + #ifndef EIO_FINISH # define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 #endif @@ -68,42 +97,158 @@ # define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) #endif +#ifndef EIO_FD_TO_WIN32_HANDLE +# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd) +#endif +#ifndef EIO_WIN32_HANDLE_TO_FD +# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0) +#endif + +#define EIO_ERRNO(errval,retval) ((errno = errval), retval) + +#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1) + #ifdef _WIN32 - /*doh*/ -#else + #undef PAGESIZE + #define PAGESIZE 4096 /* GetSystemInfo? */ -# include "config.h" -# include -# include -# include -# include -# include -# include -# include + /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */ -/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ -# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ -# define _DIRENT_HAVE_D_TYPE /* sigh */ -# define D_INO(de) (de)->d_fileno -# define D_NAMLEN(de) (de)->d_namlen -# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 -# define D_INO(de) (de)->d_ino -# endif + #ifdef EIO_STRUCT_STATI64 + /* look at perl's non-sloppy stat */ + #define stat(path,buf) _stati64 (path,buf) + #define fstat(fd,buf) _fstati64 (fd,buf) + #endif + #define lstat(path,buf) stat (path,buf) + #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1)) + #define mkdir(path,mode) _mkdir (path) + #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1)) + + #define chmod(path,mode) _chmod (path, mode) + #define dup(fd) _dup (fd) + #define dup2(fd1,fd2) _dup2 (fd1, fd2) + + #define fchmod(fd,mode) EIO_ENOSYS () + #define chown(path,uid,gid) EIO_ENOSYS () + #define fchown(fd,uid,gid) EIO_ENOSYS () + #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */ + #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */ + #define mknod(path,mode,dev) EIO_ENOSYS () + #define sync() EIO_ENOSYS () + #define readlink(path,buf,s) EIO_ENOSYS () + #define statvfs(path,buf) EIO_ENOSYS () + #define fstatvfs(fd,buf) EIO_ENOSYS () + + /* rename() uses MoveFile, which fails to overwrite */ + #define rename(old,neu) eio__rename (old, neu) + + static int + eio__rename (const char *old, const char *neu) + { + if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING)) + return 0; + + /* should steal _dosmaperr */ + switch (GetLastError ()) + { + case ERROR_FILE_NOT_FOUND: + case ERROR_PATH_NOT_FOUND: + case ERROR_INVALID_DRIVE: + case ERROR_NO_MORE_FILES: + case ERROR_BAD_NETPATH: + case ERROR_BAD_NET_NAME: + case ERROR_BAD_PATHNAME: + case ERROR_FILENAME_EXCED_RANGE: + errno = ENOENT; + break; + + default: + errno = EACCES; + break; + } + + return -1; + } + + /* we could even stat and see if it exists */ + static int + symlink (const char *old, const char *neu) + { + #if WINVER >= 0x0600 + if (CreateSymbolicLink (neu, old, 1)) + return 0; + + if (CreateSymbolicLink (neu, old, 0)) + return 0; + #endif + + return EIO_ERRNO (ENOENT, -1); + } + + /* POSIX API only */ + #define CreateHardLink(neu,old,flags) 0 + #define CreateSymbolicLink(neu,old,flags) 0 + + struct statvfs + { + int dummy; + }; + + #define DT_DIR EIO_DT_DIR + #define DT_REG EIO_DT_REG + #define D_NAME(entp) entp.cFileName + #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG) + +#else + + #include + #include + #include + #include + #include + #include + + #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES + #include + #endif + + #define D_NAME(entp) entp->d_name + + /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ + #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ + #define _DIRENT_HAVE_D_TYPE /* sigh */ + #define D_INO(de) (de)->d_fileno + #define D_NAMLEN(de) (de)->d_namlen + #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 + #define D_INO(de) (de)->d_ino + #endif + + #ifdef _D_EXACT_NAMLEN + #undef D_NAMLEN + #define D_NAMLEN(de) _D_EXACT_NAMLEN (de) + #endif + + #ifdef _DIRENT_HAVE_D_TYPE + #define D_TYPE(de) (de)->d_type + #endif + + #ifndef EIO_STRUCT_DIRENT + #define EIO_STRUCT_DIRENT struct dirent + #endif -#ifdef _D_EXACT_NAMLEN -# undef D_NAMLEN -# define D_NAMLEN(de) _D_EXACT_NAMLEN (de) #endif -# ifdef _DIRENT_HAVE_D_TYPE -# define D_TYPE(de) (de)->d_type -# endif +#if HAVE_UTIMES +# include +#endif -# ifndef EIO_STRUCT_DIRENT -# define EIO_STRUCT_DIRENT struct dirent -# endif +#if HAVE_SYS_SYSCALL_H +# include +#endif +#if HAVE_SYS_PRCTL_H +# include #endif #if HAVE_SENDFILE @@ -128,41 +273,52 @@ # define D_INO(de) 0 #endif #ifndef D_NAMLEN -# define D_NAMLEN(de) strlen ((de)->d_name) +# define D_NAMLEN(entp) strlen (D_NAME (entp)) #endif -/* number of seconds after which an idle threads exit */ -#define IDLE_TIMEOUT 10 - /* used for struct dirent, AIX doesn't provide it */ #ifndef NAME_MAX # define NAME_MAX 4096 #endif +/* used for readlink etc. */ +#ifndef PATH_MAX +# define PATH_MAX 4096 +#endif + /* buffer size for various temporary buffers */ #define EIO_BUFSIZE 65536 #define dBUF \ - char *eio_buf; \ - ETP_WORKER_LOCK (self); \ - self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ - ETP_WORKER_UNLOCK (self); \ + char *eio_buf = malloc (EIO_BUFSIZE); \ errno = ENOMEM; \ if (!eio_buf) \ - return -1; + return -1 + +#define FUBd \ + free (eio_buf) #define EIO_TICKS ((1000000 + 1023) >> 10) /*****************************************************************************/ -#if __GNUC__ >= 3 -# define expect(expr,value) __builtin_expect ((expr),(value)) -#else -# define expect(expr,value) (expr) -#endif +struct tmpbuf +{ + void *ptr; + int len; +}; + +static void * +tmpbuf_get (struct tmpbuf *buf, int len) +{ + if (buf->len < len) + { + free (buf->ptr); + buf->ptr = malloc (buf->len = len); + } -#define expect_false(expr) expect ((expr) != 0, 0) -#define expect_true(expr) expect ((expr) != 0, 1) + return buf->ptr; +} /*****************************************************************************/ @@ -178,29 +334,13 @@ static void eio_execute (struct etp_worker *self, eio_req *req); #define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) -#define ETP_WORKER_CLEAR(req) \ - if (wrk->dbuf) \ - { \ - free (wrk->dbuf); \ - wrk->dbuf = 0; \ - } \ - \ - if (wrk->dirp) \ - { \ - closedir (wrk->dirp); \ - wrk->dirp = 0; \ - } - -#define ETP_WORKER_COMMON \ - void *dbuf; \ - DIR *dirp; - /*****************************************************************************/ #define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) /* calculate time difference in ~1/EIO_TICKS of a second */ -static int tvdiff (struct timeval *tv1, struct timeval *tv2) +ecb_inline int +tvdiff (struct timeval *tv1, struct timeval *tv2) { return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS + ((tv2->tv_usec - tv1->tv_usec) >> 10); @@ -214,15 +354,16 @@ static unsigned int max_poll_time; /* reslock */ static unsigned int max_poll_reqs; /* reslock */ -static volatile unsigned int nreqs; /* reqlock */ -static volatile unsigned int nready; /* reqlock */ -static volatile unsigned int npending; /* reqlock */ -static volatile unsigned int max_idle = 4; - -static mutex_t wrklock = X_MUTEX_INIT; -static mutex_t reslock = X_MUTEX_INIT; -static mutex_t reqlock = X_MUTEX_INIT; -static cond_t reqwait = X_COND_INIT; +static unsigned int nreqs; /* reqlock */ +static unsigned int nready; /* reqlock */ +static unsigned int npending; /* reqlock */ +static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */ +static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */ + +static xmutex_t wrklock; +static xmutex_t reslock; +static xmutex_t reqlock; +static xcond_t reqwait; #if !HAVE_PREADWRITE /* @@ -230,43 +371,48 @@ * 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 xmutex_t preadwritelock; #endif typedef struct etp_worker { + struct tmpbuf tmpbuf; + /* locked by wrklock */ struct etp_worker *prev, *next; - thread_t tid; - - /* locked by reslock, reqlock or wrklock */ - ETP_REQ *req; /* currently processed request */ + xthread_t tid; +#ifdef ETP_WORKER_COMMON ETP_WORKER_COMMON +#endif } etp_worker; -static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ +static etp_worker wrk_first; /* NOT etp */ #define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) #define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) /* worker threads management */ -static void etp_worker_clear (etp_worker *wrk) +static void ecb_cold +etp_worker_clear (etp_worker *wrk) { - ETP_WORKER_CLEAR (wrk); } -static void etp_worker_free (etp_worker *wrk) +static void ecb_cold +etp_worker_free (etp_worker *wrk) { + free (wrk->tmpbuf.ptr); + wrk->next->prev = wrk->prev; wrk->prev->next = wrk->next; free (wrk); } -static unsigned int etp_nreqs (void) +static unsigned int +etp_nreqs (void) { int retval; if (WORDACCESS_UNSAFE) X_LOCK (reqlock); @@ -275,7 +421,8 @@ return retval; } -static unsigned int etp_nready (void) +static unsigned int +etp_nready (void) { unsigned int retval; @@ -286,7 +433,8 @@ return retval; } -static unsigned int etp_npending (void) +static unsigned int +etp_npending (void) { unsigned int retval; @@ -297,7 +445,8 @@ return retval; } -static unsigned int etp_nthreads (void) +static unsigned int +etp_nthreads (void) { unsigned int retval; @@ -321,7 +470,19 @@ static etp_reqq req_queue; static etp_reqq res_queue; -static int reqq_push (etp_reqq *q, ETP_REQ *req) +static void ecb_noinline ecb_cold +reqq_init (etp_reqq *q) +{ + int pri; + + for (pri = 0; pri < ETP_NUM_PRI; ++pri) + q->qs[pri] = q->qe[pri] = 0; + + q->size = 0; +} + +static int ecb_noinline +reqq_push (etp_reqq *q, ETP_REQ *req) { int pri = req->pri; req->next = 0; @@ -337,7 +498,8 @@ return q->size++; } -static ETP_REQ *reqq_shift (etp_reqq *q) +static ETP_REQ * ecb_noinline +reqq_shift (etp_reqq *q) { int pri; @@ -362,46 +524,19 @@ abort (); } -static void etp_atfork_prepare (void) -{ - X_LOCK (wrklock); - X_LOCK (reqlock); - X_LOCK (reslock); -#if !HAVE_PREADWRITE - X_LOCK (preadwritelock); -#endif -} - -static void etp_atfork_parent (void) -{ -#if !HAVE_PREADWRITE - X_UNLOCK (preadwritelock); -#endif - X_UNLOCK (reslock); - X_UNLOCK (reqlock); - X_UNLOCK (wrklock); -} - -static void etp_atfork_child (void) +static int ecb_cold +etp_init (void (*want_poll)(void), void (*done_poll)(void)) { - ETP_REQ *prv; - - while ((prv = reqq_shift (&req_queue))) - ETP_DESTROY (prv); - - while ((prv = reqq_shift (&res_queue))) - ETP_DESTROY (prv); - - while (wrk_first.next != &wrk_first) - { - etp_worker *wrk = wrk_first.next; + X_MUTEX_CREATE (wrklock); + X_MUTEX_CREATE (reslock); + X_MUTEX_CREATE (reqlock); + X_COND_CREATE (reqwait); - if (wrk->req) - ETP_DESTROY (wrk->req); + reqq_init (&req_queue); + reqq_init (&res_queue); - etp_worker_clear (wrk); - etp_worker_free (wrk); - } + wrk_first.next = + wrk_first.prev = &wrk_first; started = 0; idle = 0; @@ -409,22 +544,6 @@ nready = 0; npending = 0; - etp_atfork_parent (); -} - -static void -etp_once_init (void) -{ - X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); -} - -static int -etp_init (void (*want_poll)(void), void (*done_poll)(void)) -{ - static pthread_once_t doinit = PTHREAD_ONCE_INIT; - - pthread_once (&doinit, etp_once_init); - want_poll_cb = want_poll; done_poll_cb = done_poll; @@ -433,7 +552,8 @@ X_THREAD_PROC (etp_proc); -static void etp_start_thread (void) +static void ecb_cold +etp_start_thread (void) { etp_worker *wrk = calloc (1, sizeof (etp_worker)); @@ -456,21 +576,23 @@ X_UNLOCK (wrklock); } -static void etp_maybe_start_thread (void) +static void +etp_maybe_start_thread (void) { - if (expect_true (etp_nthreads () >= wanted)) + if (ecb_expect_true (etp_nthreads () >= wanted)) return; /* todo: maybe use idle here, but might be less exact */ - if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) + if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) return; etp_start_thread (); } -static void etp_end_thread (void) +static void ecb_cold +etp_end_thread (void) { - eio_req *req = calloc (1, sizeof (eio_req)); + eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */ req->type = -1; req->pri = ETP_PRI_MAX - ETP_PRI_MIN; @@ -485,7 +607,8 @@ X_UNLOCK (wrklock); } -static int etp_poll (void) +static int +etp_poll (void) { unsigned int maxreqs; unsigned int maxtime; @@ -525,7 +648,7 @@ --nreqs; X_UNLOCK (reqlock); - if (expect_false (req->type == EIO_GROUP && req->size)) + if (ecb_expect_false (req->type == EIO_GROUP && req->size)) { req->int1 = 1; /* mark request as delayed */ continue; @@ -533,11 +656,11 @@ else { int res = ETP_FINISH (req); - if (expect_false (res)) + if (ecb_expect_false (res)) return res; } - if (expect_false (maxreqs && !--maxreqs)) + if (ecb_expect_false (maxreqs && !--maxreqs)) break; if (maxtime) @@ -553,23 +676,23 @@ return -1; } -static void etp_cancel (ETP_REQ *req) +static void +etp_cancel (ETP_REQ *req) { - X_LOCK (wrklock); - req->flags |= EIO_FLAG_CANCELLED; - X_UNLOCK (wrklock); + req->cancelled = 1; eio_grp_cancel (req); } -static void etp_submit (ETP_REQ *req) +static void +etp_submit (ETP_REQ *req) { req->pri -= ETP_PRI_MIN; - if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; - if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; + if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; + if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; - if (expect_false (req->type == EIO_GROUP)) + if (ecb_expect_false (req->type == EIO_GROUP)) { /* I hope this is worth it :/ */ X_LOCK (reqlock); @@ -598,34 +721,47 @@ } } -static void etp_set_max_poll_time (double nseconds) +static void ecb_cold +etp_set_max_poll_time (double nseconds) { if (WORDACCESS_UNSAFE) X_LOCK (reslock); max_poll_time = nseconds * EIO_TICKS; if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); } -static void etp_set_max_poll_reqs (unsigned int maxreqs) +static void ecb_cold +etp_set_max_poll_reqs (unsigned int maxreqs) { if (WORDACCESS_UNSAFE) X_LOCK (reslock); max_poll_reqs = maxreqs; if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); } -static void etp_set_max_idle (unsigned int nthreads) +static void ecb_cold +etp_set_max_idle (unsigned int nthreads) +{ + if (WORDACCESS_UNSAFE) X_LOCK (reqlock); + max_idle = nthreads; + if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); +} + +static void ecb_cold +etp_set_idle_timeout (unsigned int seconds) { if (WORDACCESS_UNSAFE) X_LOCK (reqlock); - max_idle = nthreads <= 0 ? 1 : nthreads; + idle_timeout = seconds; if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); } -static void etp_set_min_parallel (unsigned int nthreads) +static void ecb_cold +etp_set_min_parallel (unsigned int nthreads) { if (wanted < nthreads) wanted = nthreads; } -static void etp_set_max_parallel (unsigned int nthreads) +static void ecb_cold +etp_set_max_parallel (unsigned int nthreads) { if (wanted > nthreads) wanted = nthreads; @@ -636,7 +772,8 @@ /*****************************************************************************/ -static void grp_try_feed (eio_req *grp) +static void +grp_try_feed (eio_req *grp) { while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) { @@ -653,7 +790,8 @@ } } -static int grp_dec (eio_req *grp) +static int +grp_dec (eio_req *grp) { --grp->size; @@ -667,7 +805,8 @@ return 0; } -void eio_destroy (eio_req *req) +static void +eio_destroy (eio_req *req) { if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); @@ -675,7 +814,8 @@ EIO_DESTROY (req); } -static int eio_finish (eio_req *req) +static int +eio_finish (eio_req *req) { int res = EIO_FINISH (req); @@ -693,7 +833,7 @@ res2 = grp_dec (grp); - if (!res && res2) + if (!res) res = res2; } @@ -702,63 +842,81 @@ return res; } -void eio_grp_cancel (eio_req *grp) +void +eio_grp_cancel (eio_req *grp) { for (grp = grp->grp_first; grp; grp = grp->grp_next) eio_cancel (grp); } -void eio_cancel (eio_req *req) +void +eio_cancel (eio_req *req) { etp_cancel (req); } -void eio_submit (eio_req *req) +void +eio_submit (eio_req *req) { etp_submit (req); } -unsigned int eio_nreqs (void) +unsigned int +eio_nreqs (void) { return etp_nreqs (); } -unsigned int eio_nready (void) +unsigned int +eio_nready (void) { return etp_nready (); } -unsigned int eio_npending (void) +unsigned int +eio_npending (void) { return etp_npending (); } -unsigned int eio_nthreads (void) +unsigned int ecb_cold +eio_nthreads (void) { return etp_nthreads (); } -void eio_set_max_poll_time (double nseconds) +void ecb_cold +eio_set_max_poll_time (double nseconds) { etp_set_max_poll_time (nseconds); } -void eio_set_max_poll_reqs (unsigned int maxreqs) +void ecb_cold +eio_set_max_poll_reqs (unsigned int maxreqs) { etp_set_max_poll_reqs (maxreqs); } -void eio_set_max_idle (unsigned int nthreads) +void ecb_cold +eio_set_max_idle (unsigned int nthreads) { etp_set_max_idle (nthreads); } -void eio_set_min_parallel (unsigned int nthreads) +void ecb_cold +eio_set_idle_timeout (unsigned int seconds) +{ + etp_set_idle_timeout (seconds); +} + +void ecb_cold +eio_set_min_parallel (unsigned int nthreads) { etp_set_min_parallel (nthreads); } -void eio_set_max_parallel (unsigned int nthreads) +void ecb_cold +eio_set_max_parallel (unsigned int nthreads) { etp_set_max_parallel (nthreads); } @@ -777,10 +935,10 @@ # define pread eio__pread # define pwrite eio__pwrite -static ssize_t +static eio_ssize_t eio__pread (int fd, void *buf, size_t count, off_t offset) { - ssize_t res; + eio_ssize_t res; off_t ooffset; X_LOCK (preadwritelock); @@ -793,10 +951,10 @@ return res; } -static ssize_t +static eio_ssize_t eio__pwrite (int fd, void *buf, size_t count, off_t offset) { - ssize_t res; + eio_ssize_t res; off_t ooffset; X_LOCK (preadwritelock); @@ -810,12 +968,10 @@ } #endif -#ifndef HAVE_FUTIMES +#ifndef HAVE_UTIMES # undef utimes -# undef futimes -# define utimes(path,times) eio__utimes (path, times) -# define futimes(fd,times) eio__futimes (fd, times) +# define utimes(path,times) eio__utimes (path, times) static int eio__utimes (const char *filename, const struct timeval times[2]) @@ -833,7 +989,15 @@ return utime (filename, 0); } -static int eio__futimes (int fd, const struct timeval tv[2]) +#endif + +#ifndef HAVE_FUTIMES + +# undef futimes +# define futimes(fd,times) eio__futimes (fd, times) + +static int +eio__futimes (int fd, const struct timeval tv[2]) { errno = ENOSYS; return -1; @@ -846,8 +1010,26 @@ # define fdatasync(fd) fsync (fd) #endif +static int +eio__syncfs (int fd) +{ + int res; + +#if HAVE_SYS_SYNCFS + res = (int)syscall (__NR_syncfs, (int)(fd)); +#else + res = -1; + errno = ENOSYS; +#endif + + if (res < 0 && errno == ENOSYS && fd >= 0) + sync (); + + return res; +} + /* sync_file_range always needs emulation */ -int +static int eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) { #if HAVE_SYNC_FILE_RANGE @@ -874,11 +1056,22 @@ return fdatasync (fd); } +static int +eio__fallocate (int fd, int mode, off_t offset, size_t len) +{ +#if HAVE_FALLOCATE + return fallocate (fd, mode, offset, len); +#else + errno = ENOSYS; + return -1; +#endif +} + #if !HAVE_READAHEAD # undef readahead # define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) -static ssize_t +static eio_ssize_t eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) { size_t todo = count; @@ -893,6 +1086,8 @@ todo -= len; } + FUBd; + errno = 0; return count; } @@ -900,87 +1095,118 @@ #endif /* sendfile always needs emulation */ -static ssize_t -eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) +static eio_ssize_t +eio__sendfile (int ofd, int ifd, off_t offset, size_t count) { - ssize_t res; + eio_ssize_t written = 0; + eio_ssize_t res; if (!count) return 0; + for (;;) + { +#ifdef __APPLE__ +# undef HAVE_SENDFILE /* broken, as everything on os x */ +#endif #if HAVE_SENDFILE # if __linux - res = sendfile (ofd, ifd, &offset, count); + off_t soffset = offset; + res = sendfile (ofd, ifd, &soffset, 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); - - /* freebsd' sendfile will return 0 on success */ - /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ - /* not on e.g. EIO or EPIPE - sounds broken */ - if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) - res = sbytes; - } + /* + * 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 0 /* according to the manpage, this is correct, but broken behaviour */ + /* freebsd' sendfile will return 0 on success */ + /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ + /* not on e.g. EIO or EPIPE - sounds broken */ + if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) + res = sbytes; + #endif + + /* according to source inspection, this is correct, and useful behaviour */ + if (sbytes) + res = sbytes; # elif defined (__APPLE__) + off_t sbytes = count; + res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); - { - off_t sbytes = count; - res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); - - /* according to the manpage, sbytes is always valid */ - if (sbytes) - res = sbytes; - } + /* according to the manpage, sbytes is always valid */ + if (sbytes) + res = sbytes; # elif __hpux - res = sendfile (ofd, ifd, offset, count, 0, 0); + res = sendfile (ofd, ifd, offset, count, 0, 0); # elif __solaris - { - struct sendfilevec vec; - size_t sbytes; + struct sendfilevec vec; + size_t sbytes; - vec.sfv_fd = ifd; - vec.sfv_flag = 0; - vec.sfv_off = offset; - vec.sfv_len = count; + vec.sfv_fd = ifd; + vec.sfv_flag = 0; + vec.sfv_off = offset; + vec.sfv_len = count; - res = sendfilev (ofd, &vec, 1, &sbytes); + res = sendfilev (ofd, &vec, 1, &sbytes); - if (res < 0 && sbytes) - res = sbytes; - } + if (res < 0 && sbytes) + res = sbytes; # endif -#elif defined (_WIN32) - - /* does not work, just for documentation of what would need to be done */ - { - HANDLE h = TO_SOCKET (ifd); - SetFilePointer (h, offset, 0, FILE_BEGIN); - res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); - } +#elif defined (_WIN32) && 0 + /* does not work, just for documentation of what would need to be done */ + /* actually, cannot be done like this, as TransmitFile changes the file offset, */ + /* libeio guarantees that the file offset does not change, and windows */ + /* has no way to get an independent handle to the same file description */ + HANDLE h = TO_SOCKET (ifd); + SetFilePointer (h, offset, 0, FILE_BEGIN); + res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); #else - res = -1; - errno = ENOSYS; + res = -1; + errno = ENOSYS; #endif - if (res < 0 + /* we assume sendfile can copy at least 128mb in one go */ + if (res <= 128 * 1024 * 1024) + { + if (res > 0) + written += res; + + if (written) + return written; + + break; + } + else + { + /* if we requested more, then probably the kernel was lazy */ + written += res; + offset += res; + count -= res; + + if (!count) + return written; + } + } + + if (res < 0 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK /* BSDs */ #ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ || errno == ENOTSUP #endif +#ifdef EOPNOTSUPP /* windows */ || errno == EOPNOTSUPP /* BSDs */ +#endif #if __solaris || errno == EAFNOSUPPORT || errno == EPROTOTYPE #endif @@ -994,7 +1220,7 @@ while (count) { - ssize_t cnt; + eio_ssize_t cnt; cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); @@ -1016,16 +1242,300 @@ res += cnt; count -= cnt; } + + FUBd; } return res; } +#ifdef PAGESIZE +# define eio_pagesize() PAGESIZE +#else +static intptr_t +eio_pagesize (void) +{ + static intptr_t page; + + if (!page) + page = sysconf (_SC_PAGESIZE); + + return page; +} +#endif + +static void +eio_page_align (void **addr, size_t *length) +{ + intptr_t mask = eio_pagesize () - 1; + + /* round down addr */ + intptr_t adj = mask & (intptr_t)*addr; + + *addr = (void *)((intptr_t)*addr - adj); + *length += adj; + + /* round up length */ + *length = (*length + mask) & ~mask; +} + +#if !_POSIX_MEMLOCK +# define eio__mlockall(a) EIO_ENOSYS () +#else + +static int +eio__mlockall (int flags) +{ + #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7 + extern int mallopt (int, int); + mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */ + #endif + + if (EIO_MCL_CURRENT != MCL_CURRENT + || EIO_MCL_FUTURE != MCL_FUTURE) + { + flags = 0 + | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0) + | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0); + } + + return mlockall (flags); +} +#endif + +#if !_POSIX_MEMLOCK_RANGE +# define eio__mlock(a,b) EIO_ENOSYS () +#else + +static int +eio__mlock (void *addr, size_t length) +{ + eio_page_align (&addr, &length); + + return mlock (addr, length); +} + +#endif + +#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) +# define eio__msync(a,b,c) EIO_ENOSYS () +#else + +static int +eio__msync (void *mem, size_t len, int flags) +{ + eio_page_align (&mem, &len); + + if (EIO_MS_ASYNC != MS_SYNC + || EIO_MS_INVALIDATE != MS_INVALIDATE + || EIO_MS_SYNC != MS_SYNC) + { + flags = 0 + | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0) + | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0) + | (flags & EIO_MS_SYNC ? MS_SYNC : 0); + } + + return msync (mem, len, flags); +} + +#endif + +static int +eio__mtouch (eio_req *req) +{ + void *mem = req->ptr2; + size_t len = req->size; + int flags = req->int1; + + eio_page_align (&mem, &len); + + { + intptr_t addr = (intptr_t)mem; + intptr_t end = addr + len; + intptr_t page = eio_pagesize (); + + if (addr < end) + if (flags & EIO_MT_MODIFY) /* modify */ + do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req)); + else + do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req)); + } + + return 0; +} + +/*****************************************************************************/ +/* requests implemented outside eio_execute, because they are so large */ + +/* copies some absolute path to tmpbuf */ +static char * +eio__getwd (struct tmpbuf *tmpbuf, eio_wd wd) +{ + if (wd == EIO_CWD) + return getcwd (tmpbuf->ptr, PATH_MAX); + +#if HAVE_AT + abort (); /*TODO*/ +#else + strcpy (tmpbuf->ptr, wd); +#endif + return tmpbuf->ptr; +} + +/* result will always end up in tmpbuf, there is always space for adding a 0-byte */ +static int +eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path) +{ + const char *rel = path; + char *res; + char *tmp1, *tmp2; +#if SYMLOOP_MAX > 32 + int symlinks = SYMLOOP_MAX; +#else + int symlinks = 32; +#endif + + /*D*/ /*TODO: wd ignored */ + + errno = EINVAL; + if (!rel) + return -1; + + errno = ENOENT; + if (!*rel) + return -1; + + res = tmpbuf_get (tmpbuf, PATH_MAX * 3); + tmp1 = res + PATH_MAX; + tmp2 = tmp1 + PATH_MAX; + +#if 0 /* disabled, the musl way to do things is just too racy */ +#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME) + /* on linux we may be able to ask the kernel */ + { + int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME); + + if (fd >= 0) + { + sprintf (tmp1, "/proc/self/fd/%d", fd); + req->result = readlink (tmp1, res, PATH_MAX); + close (fd); + + /* here we should probably stat the open file and the disk file, to make sure they still match */ + + if (req->result > 0) + goto done; + } + else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO) + return; + } +#endif +#endif + + if (*rel != '/') + { + if (!eio__getwd (tmpbuf, wd)) + return -1; + + if (res [1]) /* only use if not / */ + res += strlen (res); + } + + while (*rel) + { + eio_ssize_t len, linklen; + const char *beg = rel; + + while (*rel && *rel != '/') + ++rel; + + len = rel - beg; + + if (!len) /* skip slashes */ + { + ++rel; + continue; + } + + if (beg [0] == '.') + { + if (len == 1) + continue; /* . - nop */ + + if (beg [1] == '.' && len == 2) + { + /* .. - back up one component, if possible */ + + while (res != tmpbuf->ptr) + if (*--res == '/') + break; + + continue; + } + } + + errno = ENAMETOOLONG; + if (res + 1 + len + 1 >= tmp1) + return; + + /* copy one component */ + *res = '/'; + memcpy (res + 1, beg, len); + + /* zero-terminate, for readlink */ + res [len + 1] = 0; + + /* now check if it's a symlink */ + linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX); + + if (linklen < 0) + { + if (errno != EINVAL) + return -1; + + /* it's a normal directory. hopefully */ + res += len + 1; + } + else + { + /* yay, it was a symlink - build new path in tmp2 */ + int rellen = strlen (rel); + + errno = ENAMETOOLONG; + if (linklen + 1 + rellen >= PATH_MAX) + return -1; + + errno = ELOOP; + if (!--symlinks) + return -1; + + if (*tmp1 == '/') + res = tmpbuf->ptr; /* symlink resolves to an absolute path */ + + /* we need to be careful, as rel might point into tmp2 already */ + memmove (tmp2 + linklen + 1, rel, rellen + 1); + tmp2 [linklen] = '/'; + memcpy (tmp2, tmp1, linklen); + + rel = tmp2; + } + } + + /* special case for the lone root path */ + if (res == tmpbuf->ptr) + *res++ = '/'; + + return res - (char *)tmpbuf->ptr; +} + static signed char eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) { - return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ - : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; + return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ + : a->inode < b->inode ? -1 + : a->inode > b->inode ? 1 + : 0; } #define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 @@ -1034,41 +1544,41 @@ #define EIO_SORT_FAST 60 /* when to only use insertion sort */ static void -eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) +eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits) { - unsigned char bits [9 + sizeof (ino_t) * 8]; + unsigned char bits [9 + sizeof (eio_ino_t) * 8]; unsigned char *bit = bits; assert (CHAR_BIT == 8); assert (sizeof (eio_dirent) * 8 < 256); - assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ - assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ + assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ + assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */ if (size <= EIO_SORT_FAST) return; /* first prepare an array of bits to test in our radix sort */ - /* try to take endianness into account, as well as differences in ino_t sizes */ + /* try to take endianness into account, as well as differences in eio_ino_t sizes */ /* inode_bits must contain all inodes ORed together */ /* which is used to skip bits that are 0 everywhere, which is very common */ { - ino_t endianness; + eio_ino_t endianness; int i, j; /* we store the byte offset of byte n into byte n of "endianness" */ - for (i = 0; i < sizeof (ino_t); ++i) + for (i = 0; i < sizeof (eio_ino_t); ++i) ((unsigned char *)&endianness)[i] = i; *bit++ = 0; - for (i = 0; i < sizeof (ino_t); ++i) + for (i = 0; i < sizeof (eio_ino_t); ++i) { /* shifting off the byte offsets out of "endianness" */ int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; endianness >>= 8; for (j = 0; j < 8; ++j) - if (inode_bits & (((ino_t)1) << (i * 8 + j))) + if (inode_bits & (((eio_ino_t)1) << (i * 8 + j))) *bit++ = offs + j; } @@ -1079,9 +1589,9 @@ /* now actually do the sorting (a variant of MSD radix sort) */ { - eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; - eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; - unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; + eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base; + eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end; + unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8]; int stk_idx = 0; base_stk [stk_idx] = dents; @@ -1170,7 +1680,7 @@ } static void -eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) +eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits) { if (size <= 1) return; /* our insertion sort relies on size > 0 */ @@ -1186,251 +1696,408 @@ /* read a full directory */ static void -eio__scandir (eio_req *req, etp_worker *self) +eio__scandir (eio_req *req) { - DIR *dirp; - EIO_STRUCT_DIRENT *entp; char *name, *names; - int namesalloc = 4096; + int namesalloc = 4096 - sizeof (void *) * 4; int namesoffs = 0; int flags = req->int1; eio_dirent *dents = 0; int dentalloc = 128; int dentoffs = 0; - ino_t inode_bits = 0; + eio_ino_t inode_bits = 0; +#ifdef _WIN32 + HANDLE dirp; + WIN32_FIND_DATA entp; +#else + DIR *dirp; + EIO_STRUCT_DIRENT *entp; +#endif req->result = -1; if (!(flags & EIO_READDIR_DENTS)) flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); - X_LOCK (wrklock); - /* the corresponding closedir is in ETP_WORKER_CLEAR */ - self->dirp = dirp = opendir (req->ptr1); +#ifdef _WIN32 + { + int len = strlen ((const char *)req->ptr1); + char *path = malloc (MAX_PATH); + const char *fmt; + + if (!len) + fmt = "./*"; + else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\') + fmt = "%s*"; + else + fmt = "%s/*"; + + _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1); + dirp = FindFirstFile (path, &entp); + free (path); + + if (dirp == INVALID_HANDLE_VALUE) + { + /* should steal _dosmaperr */ + switch (GetLastError ()) + { + case ERROR_FILE_NOT_FOUND: + req->result = 0; + break; + + case ERROR_INVALID_NAME: + case ERROR_PATH_NOT_FOUND: + case ERROR_NO_MORE_FILES: + errno = ENOENT; + break; + + case ERROR_NOT_ENOUGH_MEMORY: + errno = ENOMEM; + break; + + default: + errno = EINVAL; + break; + } + + return; + } + } +#else + dirp = opendir (req->ptr1); + + if (!dirp) + return; +#endif + + if (req->flags & EIO_FLAG_PTR1_FREE) + free (req->ptr1); + req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; req->ptr2 = names = malloc (namesalloc); - X_UNLOCK (wrklock); - if (dirp && names && (!flags || dents)) - for (;;) - { - errno = 0; - entp = readdir (dirp); + if (!names || (flags && !dents)) + return; - if (!entp) - { - if (errno) - break; + for (;;) + { + int done; + +#ifdef _WIN32 + done = !dirp; +#else + errno = 0; + entp = readdir (dirp); + done = !entp; +#endif + + if (done) + { +#ifndef _WIN32 + int old_errno = errno; + closedir (dirp); + errno = old_errno; + + if (errno) + break; +#endif + + /* sort etc. */ + req->int1 = flags; + req->result = dentoffs; + + if (flags & EIO_READDIR_STAT_ORDER) + eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); + else if (flags & EIO_READDIR_DIRS_FIRST) + if (flags & EIO_READDIR_FOUND_UNKNOWN) + eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ + else + { + /* in this case, all is known, and we just put dirs first and sort them */ + eio_dirent *oth = dents + dentoffs; + eio_dirent *dir = dents; + + /* now partition dirs to the front, and non-dirs to the back */ + /* by walking from both sides and swapping if necessary */ + while (oth > dir) + { + if (dir->type == EIO_DT_DIR) + ++dir; + else if ((--oth)->type == EIO_DT_DIR) + { + eio_dirent tmp = *dir; *dir = *oth; *oth = tmp; - /* sort etc. */ - req->int1 = flags; - req->result = dentoffs; - - if (flags & EIO_READDIR_STAT_ORDER) - eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ - else if (flags & EIO_READDIR_DIRS_FIRST) - if (flags & EIO_READDIR_FOUND_UNKNOWN) - eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ - else - { - /* in this case, all is known, and we just put dirs first and sort them */ - eio_dirent *oth = dents + dentoffs; - eio_dirent *dir = dents; - - /* now partition dirs to the front, and non-dirs to the back */ - /* by walking from both sides and swapping if necessary */ - /* also clear score, so it doesn't influence sorting */ - while (oth > dir) - { - if (dir->type == EIO_DT_DIR) ++dir; - else if ((--oth)->type == EIO_DT_DIR) - { - eio_dirent tmp = *dir; *dir = *oth; *oth = tmp; + } + } - ++dir; - } - } + /* now sort the dirs only (dirs all have the same score) */ + eio_dent_sort (dents, dir - dents, 0, inode_bits); + } - /* now sort the dirs only */ - eio_dent_sort (dents, dir - dents, 0, inode_bits); - } + break; + } - break; - } + /* now add the entry to our list(s) */ + name = D_NAME (entp); - /* now add the entry to our list(s) */ - name = entp->d_name; + /* skip . and .. entries */ + if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) + { + int len = D_NAMLEN (entp) + 1; - /* skip . and .. entries */ - if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) - { - int len = D_NAMLEN (entp) + 1; + while (ecb_expect_false (namesoffs + len > namesalloc)) + { + namesalloc *= 2; + req->ptr2 = names = realloc (names, namesalloc); - while (expect_false (namesoffs + len > namesalloc)) - { - namesalloc *= 2; - X_LOCK (wrklock); - req->ptr2 = names = realloc (names, namesalloc); - X_UNLOCK (wrklock); + if (!names) + break; + } - if (!names) - break; - } + memcpy (names + namesoffs, name, len); - memcpy (names + namesoffs, name, len); + if (dents) + { + struct eio_dirent *ent; - if (dents) - { - struct eio_dirent *ent; + if (ecb_expect_false (dentoffs == dentalloc)) + { + dentalloc *= 2; + req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); - if (expect_false (dentoffs == dentalloc)) - { - dentalloc *= 2; - X_LOCK (wrklock); - req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); - X_UNLOCK (wrklock); + if (!dents) + break; + } - if (!dents) - break; - } + ent = dents + dentoffs; - ent = dents + dentoffs; + ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */ + ent->namelen = len - 1; + ent->inode = D_INO (entp); - ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */ - ent->namelen = len - 1; - ent->inode = D_INO (entp); + inode_bits |= ent->inode; - inode_bits |= ent->inode; + switch (D_TYPE (entp)) + { + default: + ent->type = EIO_DT_UNKNOWN; + flags |= EIO_READDIR_FOUND_UNKNOWN; + break; + + #ifdef DT_FIFO + case DT_FIFO: ent->type = EIO_DT_FIFO; break; + #endif + #ifdef DT_CHR + case DT_CHR: ent->type = EIO_DT_CHR; break; + #endif + #ifdef DT_MPC + case DT_MPC: ent->type = EIO_DT_MPC; break; + #endif + #ifdef DT_DIR + case DT_DIR: ent->type = EIO_DT_DIR; break; + #endif + #ifdef DT_NAM + case DT_NAM: ent->type = EIO_DT_NAM; break; + #endif + #ifdef DT_BLK + case DT_BLK: ent->type = EIO_DT_BLK; break; + #endif + #ifdef DT_MPB + case DT_MPB: ent->type = EIO_DT_MPB; break; + #endif + #ifdef DT_REG + case DT_REG: ent->type = EIO_DT_REG; break; + #endif + #ifdef DT_NWK + case DT_NWK: ent->type = EIO_DT_NWK; break; + #endif + #ifdef DT_CMP + case DT_CMP: ent->type = EIO_DT_CMP; break; + #endif + #ifdef DT_LNK + case DT_LNK: ent->type = EIO_DT_LNK; break; + #endif + #ifdef DT_SOCK + case DT_SOCK: ent->type = EIO_DT_SOCK; break; + #endif + #ifdef DT_DOOR + case DT_DOOR: ent->type = EIO_DT_DOOR; break; + #endif + #ifdef DT_WHT + case DT_WHT: ent->type = EIO_DT_WHT; break; + #endif + } - switch (D_TYPE (entp)) - { - default: - ent->type = EIO_DT_UNKNOWN; - flags |= EIO_READDIR_FOUND_UNKNOWN; - break; - - #ifdef DT_FIFO - case DT_FIFO: ent->type = EIO_DT_FIFO; break; - #endif - #ifdef DT_CHR - case DT_CHR: ent->type = EIO_DT_CHR; break; - #endif - #ifdef DT_MPC - case DT_MPC: ent->type = EIO_DT_MPC; break; - #endif - #ifdef DT_DIR - case DT_DIR: ent->type = EIO_DT_DIR; break; - #endif - #ifdef DT_NAM - case DT_NAM: ent->type = EIO_DT_NAM; break; - #endif - #ifdef DT_BLK - case DT_BLK: ent->type = EIO_DT_BLK; break; - #endif - #ifdef DT_MPB - case DT_MPB: ent->type = EIO_DT_MPB; break; - #endif - #ifdef DT_REG - case DT_REG: ent->type = EIO_DT_REG; break; - #endif - #ifdef DT_NWK - case DT_NWK: ent->type = EIO_DT_NWK; break; - #endif - #ifdef DT_CMP - case DT_CMP: ent->type = EIO_DT_CMP; break; - #endif - #ifdef DT_LNK - case DT_LNK: ent->type = EIO_DT_LNK; break; - #endif - #ifdef DT_SOCK - case DT_SOCK: ent->type = EIO_DT_SOCK; break; - #endif - #ifdef DT_DOOR - case DT_DOOR: ent->type = EIO_DT_DOOR; break; - #endif - #ifdef DT_WHT - case DT_WHT: ent->type = EIO_DT_WHT; break; - #endif - } + ent->score = 7; - ent->score = 7; + if (flags & EIO_READDIR_DIRS_FIRST) + { + if (ent->type == EIO_DT_UNKNOWN) + { + if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ + ent->score = 1; + else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ + ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */ + } + else if (ent->type == EIO_DT_DIR) + ent->score = 0; + } + } - if (flags & EIO_READDIR_DIRS_FIRST) - { - if (ent->type == EIO_DT_UNKNOWN) - { - if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ - ent->score = 1; - else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ - ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */ - } - else if (ent->type == EIO_DT_DIR) - ent->score = 0; - } - } + namesoffs += len; + ++dentoffs; + } - namesoffs += len; - ++dentoffs; - } + if (EIO_CANCELLED (req)) + { + errno = ECANCELED; + break; + } - if (EIO_CANCELLED (req)) - { - errno = ECANCELED; - break; - } - } +#ifdef _WIN32 + if (!FindNextFile (dirp, &entp)) + { + FindClose (dirp); + dirp = 0; + } +#endif + } } -#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) -# define eio__msync(a,b,c) ((errno = ENOSYS), -1) -#else +/*****************************************************************************/ +/* working directory stuff */ -int -eio__msync (void *mem, size_t len, int flags) +#if HAVE_AT + +#define WD2FD(wd) ((wd) ? ((int)(long)(wd)) - 1 : AT_FDCWD) + +#ifndef O_SEARCH +# define O_SEARCH O_RDONLY +#endif + +eio_wd +eio_wd_open_sync (eio_wd wd, const char *path) { - if (EIO_MS_ASYNC != MS_SYNC - || EIO_MS_INVALIDATE != MS_INVALIDATE - || EIO_MS_SYNC != MS_SYNC) - { - flags = 0 - | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0) - | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0) - | (flags & EIO_MS_SYNC ? MS_SYNC : 0); - } + int fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY); - return msync (mem, len, flags); + return fd >= 0 ? (eio_wd)(long)(fd + 1) : EIO_INVALID_WD; } -#endif +static eio_wd +eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path) +{ + return eio_wd_open_sync (wd, path); +} -int -eio__mtouch (void *mem, size_t len, int flags) +void +eio_wd_close_sync (eio_wd wd) { - intptr_t addr = (intptr_t)mem; - intptr_t end = addr + len; -#ifdef PAGESIZE - const intptr_t page = PAGESIZE; + int fd = WD2FD (wd); + + if (fd >= 0) + close (fd); +} + +static int +eio__truncateat (int dirfd, const char *path, off_t length) +{ + int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC); + int res; + + if (fd < 0) + return fd; + + res = ftruncate (fd, length); + close (fd); + return res; +} + +static int +eio__statvfsat (int dirfd, const char *path, struct statvfs *buf) +{ + int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC); + int res; + + if (fd < 0) + return fd; + + res = fstatvfs (fd, buf); + close (fd); + return res; + +} + #else - static intptr_t page; - if (!page) - page = sysconf (_SC_PAGESIZE); -#endif +/* on legacy systems, we represent the working directories simply by their path strings */ - /* round down to start of page, although this is probably useless */ - addr &= ~(page - 1); /* assume page size is always a power of two */ +static const char * +wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path) +{ + if (!wd || *path == '/') + return path; - if (addr < end) - if (flags & EIO_MT_MODIFY) /* modify */ - do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); - else - do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); + { + int l1 = strlen ((const char *)wd); + int l2 = strlen (path); - return 0; + char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2); + + memcpy (res, wd, l1); + res [l1] = '/'; + memcpy (res + l1 + 1, path, l2 + 1); + + return res; + } +} + +static eio_wd +eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path) +{ + if (*path == '/') /* absolute paths ignore wd */ + path = strdup (path); + else if (path [0] == '.' && !path [1]) /* special case '.', as it is common */ + return wd; + else + { + int len = eio__realpath (tmpbuf, wd, path); + + path = EIO_INVALID_WD; + + if (len >= 0) + { + ((char *)tmpbuf->ptr)[len] = 0; + path = strdup (tmpbuf->ptr); + } + } + + if (!path) + path = EIO_INVALID_WD; + + return (eio_wd)path; +} + +eio_wd +eio_wd_open_sync (eio_wd wd, const char *path) +{ + struct tmpbuf tmpbuf = { 0 }; + wd = eio__wd_open_sync (&tmpbuf, wd, path); + free (tmpbuf.ptr); + + return wd; } +void +eio_wd_close_sync (eio_wd wd) +{ + if (wd != EIO_INVALID_WD) + free (wd); +} + +#endif + /*****************************************************************************/ #define ALLOC(len) \ @@ -1454,37 +2121,48 @@ struct timespec ts; etp_worker *self = (etp_worker *)thr_arg; - /* try to distribute timeouts somewhat randomly */ +#if HAVE_PRCTL_SET_NAME + prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0); +#endif + + /* try to distribute timeouts somewhat evenly */ ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); for (;;) { + ts.tv_sec = 0; + X_LOCK (reqlock); for (;;) { - self->req = req = reqq_shift (&req_queue); + req = reqq_shift (&req_queue); if (req) break; + if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */ + { + X_UNLOCK (reqlock); + X_LOCK (wrklock); + --started; + X_UNLOCK (wrklock); + goto quit; + } + ++idle; - ts.tv_sec = time (0) + IDLE_TIMEOUT; - if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) + if (idle <= max_idle) + /* we are allowed to idle, so do so without any timeout */ + X_COND_WAIT (reqwait, reqlock); + else { - if (idle > max_idle) - { - --idle; - X_UNLOCK (reqlock); - X_LOCK (wrklock); - --started; - X_UNLOCK (wrklock); - goto quit; - } + /* initialise timeout once */ + if (!ts.tv_sec) + ts.tv_sec = time (0) + idle_timeout; - /* we are allowed to idle, so do so without any timeout */ - X_COND_WAIT (reqwait, reqlock); + if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) + ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */ } --idle; @@ -1497,8 +2175,7 @@ if (req->type < 0) goto quit; - if (!EIO_CANCELLED (req)) - ETP_EXECUTE (self, req); + ETP_EXECUTE (self, req); X_LOCK (reslock); @@ -1507,13 +2184,14 @@ if (!reqq_push (&res_queue, req) && want_poll_cb) want_poll_cb (); - self->req = 0; etp_worker_clear (self); X_UNLOCK (reslock); } quit: + free (req); + X_LOCK (wrklock); etp_worker_free (self); X_UNLOCK (wrklock); @@ -1523,12 +2201,18 @@ /*****************************************************************************/ -int eio_init (void (*want_poll)(void), void (*done_poll)(void)) +int ecb_cold +eio_init (void (*want_poll)(void), void (*done_poll)(void)) { +#if !HAVE_PREADWRITE + X_MUTEX_CREATE (preadwritelock); +#endif + return etp_init (want_poll, done_poll); } -static void eio_api_destroy (eio_req *req) +ecb_inline void +eio_api_destroy (eio_req *req) { free (req); } @@ -1557,12 +2241,46 @@ return 0; \ } -static void eio_execute (etp_worker *self, eio_req *req) +static void +eio_execute (etp_worker *self, eio_req *req) { - errno = 0; +#if HAVE_AT + int dirfd; +#else + const char *path; +#endif + + if (ecb_expect_false (EIO_CANCELLED (req))) + { + req->result = -1; + req->errorno = ECANCELED; + return; + } + + if (ecb_expect_false (req->wd == EIO_INVALID_WD)) + { + req->result = -1; + req->errorno = ENOENT; + return; + } + + if (req->type >= EIO_OPEN) + { + #if HAVE_AT + dirfd = WD2FD (req->wd); + #else + path = wd_expand (&self->tmpbuf, req->wd, req->ptr1); + #endif + } switch (req->type) { + case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1); + req->result = req->wd == EIO_INVALID_WD ? -1 : 0; + break; + case EIO_WD_CLOSE: req->result = 0; + eio_wd_close_sync (req->wd); break; + case EIO_READ: ALLOC (req->size); req->result = req->offs >= 0 ? pread (req->int1, req->ptr2, req->size, req->offs) @@ -1572,90 +2290,150 @@ : write (req->int1, req->ptr2, req->size); break; case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; - case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; + case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break; + +#if HAVE_AT case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); - req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; + req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break; case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); - req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; + req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break; + case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break; + case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break; + case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break; + case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break; + + case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break; + case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break; + case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break; + case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD (req->int3), req->ptr2); break; + case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD (req->int3), req->ptr2, 0); break; + case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break; + case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break; + case EIO_READLINK: ALLOC (PATH_MAX); + req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break; + case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); + req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break; + case EIO_UTIME: + case EIO_FUTIME: + { + struct timespec ts[2]; + struct timespec *times; + + if (req->nv1 != -1. || req->nv2 != -1.) + { + ts[0].tv_sec = req->nv1; + ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9; + ts[1].tv_sec = req->nv2; + ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9; + + times = ts; + } + else + times = 0; + + req->result = req->type == EIO_FUTIME + ? futimens (req->int1, times) + : utimensat (dirfd, req->ptr1, times, 0); + } + break; + +#else + + case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); + req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break; + case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); + req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break; + case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break; + case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break; + case EIO_TRUNCATE: req->result = truncate (path , req->offs); break; + case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break; + + case EIO_UNLINK: req->result = unlink (path ); break; + case EIO_RMDIR: req->result = rmdir (path ); break; + case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break; + case EIO_RENAME: req->result = rename (path , req->ptr2); break; + case EIO_LINK: req->result = link (path , req->ptr2); break; + case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break; + case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break; + case EIO_READLINK: ALLOC (PATH_MAX); + req->result = readlink (path, req->ptr2, PATH_MAX); break; + case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); + req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break; + + case EIO_UTIME: + case EIO_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) * 1e6; + tv[1].tv_sec = req->nv2; + tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6; + + times = tv; + } + else + times = 0; + + req->result = req->type == EIO_FUTIME + ? futimes (req->int1, times) + : utimes (req->ptr1, times); + } + break; + +#endif + + case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1))) + { + ALLOC (req->result); + memcpy (req->ptr2, self->tmpbuf.ptr, req->result); + } + break; + case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; - case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); - req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break; case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; - case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; - case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; - case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; - case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break; case EIO_CLOSE: req->result = close (req->int1); break; case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; - case EIO_UNLINK: req->result = unlink (req->ptr1); break; - case EIO_RMDIR: req->result = rmdir (req->ptr1); break; - case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; - case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; - case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; - case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; - case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; - - case EIO_READLINK: ALLOC (NAME_MAX); - req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; - case EIO_SYNC: req->result = 0; sync (); break; case EIO_FSYNC: req->result = fsync (req->int1); break; case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; - case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; - case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; + case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break; case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; + case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; + case EIO_MTOUCH: req->result = eio__mtouch (req); break; + case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; + case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; + case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break; - case EIO_READDIR: eio__scandir (req, self); break; + case EIO_READDIR: eio__scandir (req); break; case EIO_BUSY: #ifdef _WIN32 - Sleep (req->nv1 * 1000.); + Sleep (req->nv1 * 1e3); #else { struct timeval tv; tv.tv_sec = req->nv1; - tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; + tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6; req->result = select (0, 0, 0, 0, &tv); } #endif break; - case EIO_UTIME: - case EIO_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 == EIO_FUTIME - ? futimes (req->int1, times) - : utimes (req->ptr1, times); - } - break; - case EIO_GROUP: abort (); /* handled in eio_request */ @@ -1664,10 +2442,11 @@ break; case EIO_CUSTOM: - ((void (*)(eio_req *))req->feed) (req); + req->feed (req); break; default: + errno = ENOSYS; req->result = -1; break; } @@ -1702,9 +2481,14 @@ REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; } -eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) +eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) { - REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; + REQ (EIO_FDATASYNC); req->int1 = fd; SEND; +} + +eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data) +{ + REQ (EIO_SYNCFS); req->int1 = fd; SEND; } eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) @@ -1712,9 +2496,24 @@ REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; } -eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) +eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) { - REQ (EIO_FDATASYNC); req->int1 = fd; SEND; + REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; +} + +eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data) +{ + REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND; +} + +eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) +{ + REQ (EIO_MLOCKALL); req->int1 = flags; SEND; +} + +eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data) +{ + REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND; } eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) @@ -1762,7 +2561,7 @@ REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; } -eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) +eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data) { REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; } @@ -1792,7 +2591,7 @@ REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; } -eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) +eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data) { REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; } @@ -1818,6 +2617,11 @@ return eio__1path (EIO_READLINK, path, pri, cb, data); } +eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data) +{ + return eio__1path (EIO_REALPATH, path, pri, cb, data); +} + eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) { return eio__1path (EIO_STAT, path, pri, cb, data); @@ -1850,7 +2654,7 @@ eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) { - REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; + REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND; } static eio_req * @@ -1884,9 +2688,9 @@ return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); } -eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) +eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data) { - REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; + REQ (EIO_CUSTOM); req->feed = execute; SEND; } #endif @@ -1905,7 +2709,8 @@ /*****************************************************************************/ /* grp functions */ -void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) +void +eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) { grp->int2 = limit; grp->feed = feed; @@ -1913,14 +2718,16 @@ grp_try_feed (grp); } -void eio_grp_limit (eio_req *grp, int limit) +void +eio_grp_limit (eio_req *grp, int limit) { grp->int2 = limit; grp_try_feed (grp); } -void eio_grp_add (eio_req *grp, eio_req *req) +void +eio_grp_add (eio_req *grp, eio_req *req) { assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); @@ -1941,18 +2748,9 @@ /*****************************************************************************/ /* misc garbage */ -ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) +eio_ssize_t +eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) { - etp_worker wrk; - ssize_t ret; - - wrk.dbuf = 0; - - ret = eio__sendfile (ofd, ifd, offset, count, &wrk); - - if (wrk.dbuf) - free (wrk.dbuf); - - return ret; + return eio__sendfile (ofd, ifd, offset, count); }