1 | /* |
1 | /* |
2 | * libeio implementation |
2 | * libeio implementation |
3 | * |
3 | * |
4 | * Copyright (c) 2007,2008,2009 Marc Alexander Lehmann <libeio@schmorp.de> |
4 | * Copyright (c) 2007,2008,2009,2010 Marc Alexander Lehmann <libeio@schmorp.de> |
5 | * All rights reserved. |
5 | * All rights reserved. |
6 | * |
6 | * |
7 | * Redistribution and use in source and binary forms, with or without modifica- |
7 | * Redistribution and use in source and binary forms, with or without modifica- |
8 | * tion, are permitted provided that the following conditions are met: |
8 | * tion, are permitted provided that the following conditions are met: |
9 | * |
9 | * |
… | |
… | |
49 | #include <stdlib.h> |
49 | #include <stdlib.h> |
50 | #include <string.h> |
50 | #include <string.h> |
51 | #include <errno.h> |
51 | #include <errno.h> |
52 | #include <sys/types.h> |
52 | #include <sys/types.h> |
53 | #include <sys/stat.h> |
53 | #include <sys/stat.h> |
|
|
54 | #include <sys/statvfs.h> |
54 | #include <limits.h> |
55 | #include <limits.h> |
55 | #include <fcntl.h> |
56 | #include <fcntl.h> |
56 | #include <assert.h> |
57 | #include <assert.h> |
57 | |
58 | |
58 | #ifndef EIO_FINISH |
59 | #ifndef EIO_FINISH |
… | |
… | |
80 | # include <utime.h> |
81 | # include <utime.h> |
81 | # include <signal.h> |
82 | # include <signal.h> |
82 | # include <dirent.h> |
83 | # include <dirent.h> |
83 | |
84 | |
84 | /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ |
85 | /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ |
85 | # if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) |
86 | # if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ |
86 | # define _DIRENT_HAVE_D_TYPE /* sigh */ |
87 | # define _DIRENT_HAVE_D_TYPE /* sigh */ |
87 | # define D_INO(de) (de)->d_fileno |
88 | # define D_INO(de) (de)->d_fileno |
88 | # define D_NAMLEN(de) (de)->d_namlen |
89 | # define D_NAMLEN(de) (de)->d_namlen |
89 | # elif defined(__linux) || defined(d_ino) || _XOPEN_SOURCE >= 600 |
90 | # elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 |
90 | # define D_INO(de) (de)->d_ino |
91 | # define D_INO(de) (de)->d_ino |
91 | # endif |
92 | # endif |
92 | |
93 | |
93 | #ifdef _D_EXACT_NAMLEN |
94 | #ifdef _D_EXACT_NAMLEN |
94 | # undef D_NAMLEN |
95 | # undef D_NAMLEN |
… | |
… | |
106 | #endif |
107 | #endif |
107 | |
108 | |
108 | #if HAVE_SENDFILE |
109 | #if HAVE_SENDFILE |
109 | # if __linux |
110 | # if __linux |
110 | # include <sys/sendfile.h> |
111 | # include <sys/sendfile.h> |
111 | # elif __freebsd |
112 | # elif __FreeBSD__ || defined __APPLE__ |
112 | # include <sys/socket.h> |
113 | # include <sys/socket.h> |
113 | # include <sys/uio.h> |
114 | # include <sys/uio.h> |
114 | # elif __hpux |
115 | # elif __hpux |
115 | # include <sys/socket.h> |
116 | # include <sys/socket.h> |
116 | # elif __solaris /* not yet */ |
117 | # elif __solaris |
117 | # include <sys/sendfile.h> |
118 | # include <sys/sendfile.h> |
118 | # else |
119 | # else |
119 | # error sendfile support requested but not available |
120 | # error sendfile support requested but not available |
120 | # endif |
121 | # endif |
121 | #endif |
122 | #endif |
… | |
… | |
196 | |
197 | |
197 | /*****************************************************************************/ |
198 | /*****************************************************************************/ |
198 | |
199 | |
199 | #define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
200 | #define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
200 | |
201 | |
201 | /* calculcate time difference in ~1/EIO_TICKS of a second */ |
202 | /* calculate time difference in ~1/EIO_TICKS of a second */ |
202 | static int tvdiff (struct timeval *tv1, struct timeval *tv2) |
203 | static int tvdiff (struct timeval *tv1, struct timeval *tv2) |
203 | { |
204 | { |
204 | return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
205 | return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
205 | + ((tv2->tv_usec - tv1->tv_usec) >> 10); |
206 | + ((tv2->tv_usec - tv1->tv_usec) >> 10); |
206 | } |
207 | } |
… | |
… | |
598 | } |
599 | } |
599 | |
600 | |
600 | static void etp_set_max_poll_time (double nseconds) |
601 | static void etp_set_max_poll_time (double nseconds) |
601 | { |
602 | { |
602 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
603 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
603 | max_poll_time = nseconds; |
604 | max_poll_time = nseconds * EIO_TICKS; |
604 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
605 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
605 | } |
606 | } |
606 | |
607 | |
607 | static void etp_set_max_poll_reqs (unsigned int maxreqs) |
608 | static void etp_set_max_poll_reqs (unsigned int maxreqs) |
608 | { |
609 | { |
… | |
… | |
867 | if (!res || errno != ENOSYS) |
868 | if (!res || errno != ENOSYS) |
868 | return res; |
869 | return res; |
869 | #endif |
870 | #endif |
870 | |
871 | |
871 | /* even though we could play tricks with the flags, it's better to always |
872 | /* even though we could play tricks with the flags, it's better to always |
872 | * call fdatasync, as thta matches the expectation of it's users best */ |
873 | * call fdatasync, as that matches the expectation of its users best */ |
873 | return fdatasync (fd); |
874 | return fdatasync (fd); |
874 | } |
875 | } |
875 | |
876 | |
876 | #if !HAVE_READAHEAD |
877 | #if !HAVE_READAHEAD |
877 | # undef readahead |
878 | # undef readahead |
… | |
… | |
909 | |
910 | |
910 | #if HAVE_SENDFILE |
911 | #if HAVE_SENDFILE |
911 | # if __linux |
912 | # if __linux |
912 | res = sendfile (ofd, ifd, &offset, count); |
913 | res = sendfile (ofd, ifd, &offset, count); |
913 | |
914 | |
914 | # elif __freebsd |
915 | # elif __FreeBSD__ |
915 | /* |
916 | /* |
916 | * Of course, the freebsd sendfile is a dire hack with no thoughts |
917 | * Of course, the freebsd sendfile is a dire hack with no thoughts |
917 | * wasted on making it similar to other I/O functions. |
918 | * wasted on making it similar to other I/O functions. |
918 | */ |
919 | */ |
919 | { |
920 | { |
920 | off_t sbytes; |
921 | off_t sbytes; |
921 | res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
922 | res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
922 | |
923 | |
923 | if (res < 0 && sbytes) |
924 | /* freebsd' sendfile will return 0 on success */ |
924 | /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ |
925 | /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ |
|
|
926 | /* not on e.g. EIO or EPIPE - sounds broken */ |
|
|
927 | if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) |
|
|
928 | res = sbytes; |
|
|
929 | } |
|
|
930 | |
|
|
931 | # elif defined (__APPLE__) |
|
|
932 | |
|
|
933 | { |
|
|
934 | off_t sbytes = count; |
|
|
935 | res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); |
|
|
936 | |
|
|
937 | /* according to the manpage, sbytes is always valid */ |
|
|
938 | if (sbytes) |
925 | res = sbytes; |
939 | res = sbytes; |
926 | } |
940 | } |
927 | |
941 | |
928 | # elif __hpux |
942 | # elif __hpux |
929 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
943 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
… | |
… | |
943 | if (res < 0 && sbytes) |
957 | if (res < 0 && sbytes) |
944 | res = sbytes; |
958 | res = sbytes; |
945 | } |
959 | } |
946 | |
960 | |
947 | # endif |
961 | # endif |
|
|
962 | |
|
|
963 | #elif defined (_WIN32) |
|
|
964 | |
|
|
965 | /* does not work, just for documentation of what would need to be done */ |
|
|
966 | { |
|
|
967 | HANDLE h = TO_SOCKET (ifd); |
|
|
968 | SetFilePointer (h, offset, 0, FILE_BEGIN); |
|
|
969 | res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); |
|
|
970 | } |
|
|
971 | |
948 | #else |
972 | #else |
949 | res = -1; |
973 | res = -1; |
950 | errno = ENOSYS; |
974 | errno = ENOSYS; |
951 | #endif |
975 | #endif |
952 | |
976 | |
953 | if (res < 0 |
977 | if (res < 0 |
954 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
978 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
|
|
979 | /* BSDs */ |
|
|
980 | #ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ |
|
|
981 | || errno == ENOTSUP |
|
|
982 | #endif |
|
|
983 | || errno == EOPNOTSUPP /* BSDs */ |
955 | #if __solaris |
984 | #if __solaris |
956 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
985 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
957 | #endif |
986 | #endif |
958 | ) |
987 | ) |
959 | ) |
988 | ) |
… | |
… | |
1353 | } |
1382 | } |
1354 | } |
1383 | } |
1355 | } |
1384 | } |
1356 | |
1385 | |
1357 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
1386 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
1358 | # undef msync |
|
|
1359 | # define msync(a,b,c) ((errno = ENOSYS), -1) |
1387 | # define eio__msync(a,b,c) ((errno = ENOSYS), -1) |
|
|
1388 | #else |
|
|
1389 | |
|
|
1390 | int |
|
|
1391 | eio__msync (void *mem, size_t len, int flags) |
|
|
1392 | { |
|
|
1393 | if (EIO_MS_ASYNC != MS_SYNC |
|
|
1394 | || EIO_MS_INVALIDATE != MS_INVALIDATE |
|
|
1395 | || EIO_MS_SYNC != MS_SYNC) |
|
|
1396 | { |
|
|
1397 | flags = 0 |
|
|
1398 | | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0) |
|
|
1399 | | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0) |
|
|
1400 | | (flags & EIO_MS_SYNC ? MS_SYNC : 0); |
|
|
1401 | } |
|
|
1402 | |
|
|
1403 | return msync (mem, len, flags); |
|
|
1404 | } |
|
|
1405 | |
1360 | #endif |
1406 | #endif |
1361 | |
1407 | |
1362 | int |
1408 | int |
1363 | eio__mtouch (void *mem, size_t len, int flags) |
1409 | eio__mtouch (void *mem, size_t len, int flags) |
1364 | { |
1410 | { |
… | |
… | |
1371 | |
1417 | |
1372 | if (!page) |
1418 | if (!page) |
1373 | page = sysconf (_SC_PAGESIZE); |
1419 | page = sysconf (_SC_PAGESIZE); |
1374 | #endif |
1420 | #endif |
1375 | |
1421 | |
|
|
1422 | /* round down to start of page, although this is probably useless */ |
1376 | addr &= ~(page - 1); /* assume page size is always a power of two */ |
1423 | addr &= ~(page - 1); /* assume page size is always a power of two */ |
1377 | |
1424 | |
1378 | if (addr < end) |
1425 | if (addr < end) |
1379 | if (flags) /* modify */ |
1426 | if (flags & EIO_MT_MODIFY) /* modify */ |
1380 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
1427 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
1381 | else |
1428 | else |
1382 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
1429 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
1383 | |
1430 | |
1384 | return 0; |
1431 | return 0; |
… | |
… | |
1532 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1579 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1533 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1580 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1534 | case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1581 | case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1535 | req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1582 | req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1536 | |
1583 | |
|
|
1584 | case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); |
|
|
1585 | req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break; |
|
|
1586 | case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); |
|
|
1587 | req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; |
|
|
1588 | |
1537 | case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; |
1589 | case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; |
1538 | case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; |
1590 | case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; |
1539 | case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; |
1591 | case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; |
1540 | case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; |
1592 | case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; |
1541 | case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; |
1593 | case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; |
… | |
… | |
1556 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
1608 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
1557 | |
1609 | |
1558 | case EIO_SYNC: req->result = 0; sync (); break; |
1610 | case EIO_SYNC: req->result = 0; sync (); break; |
1559 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1611 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1560 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1612 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1561 | case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; |
1613 | case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; |
1562 | case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; |
1614 | case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; |
1563 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
1615 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
1564 | |
1616 | |
1565 | case EIO_READDIR: eio__scandir (req, self); break; |
1617 | case EIO_READDIR: eio__scandir (req, self); break; |
1566 | |
1618 | |
… | |
… | |
1688 | eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) |
1740 | eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) |
1689 | { |
1741 | { |
1690 | REQ (EIO_FSTAT); req->int1 = fd; SEND; |
1742 | REQ (EIO_FSTAT); req->int1 = fd; SEND; |
1691 | } |
1743 | } |
1692 | |
1744 | |
|
|
1745 | eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data) |
|
|
1746 | { |
|
|
1747 | REQ (EIO_FSTATVFS); req->int1 = fd; SEND; |
|
|
1748 | } |
|
|
1749 | |
1693 | eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) |
1750 | eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) |
1694 | { |
1751 | { |
1695 | REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; |
1752 | REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; |
1696 | } |
1753 | } |
1697 | |
1754 | |
… | |
… | |
1767 | } |
1824 | } |
1768 | |
1825 | |
1769 | eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) |
1826 | eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) |
1770 | { |
1827 | { |
1771 | return eio__1path (EIO_LSTAT, path, pri, cb, data); |
1828 | return eio__1path (EIO_LSTAT, path, pri, cb, data); |
|
|
1829 | } |
|
|
1830 | |
|
|
1831 | eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data) |
|
|
1832 | { |
|
|
1833 | return eio__1path (EIO_STATVFS, path, pri, cb, data); |
1772 | } |
1834 | } |
1773 | |
1835 | |
1774 | eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) |
1836 | eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) |
1775 | { |
1837 | { |
1776 | return eio__1path (EIO_UNLINK, path, pri, cb, data); |
1838 | return eio__1path (EIO_UNLINK, path, pri, cb, data); |