[ViewVC] Diff of: cvs/libeio/eio.c

Comparing libeio/eio.c (file contents):
Revision 1.24 by root, Tue Sep 30 13:17:32 2008 UTC vs.
Revision 1.53 by root, Sat Jan 9 10:03:09 2010 UTC

…		…
1	/*	1	/*
2	* libeio implementation	2	* libeio implementation
3	*	3	*
4	* Copyright (c) 2007,2008 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	*
36	* provisions above, a recipient may use your version of this file under	36	* provisions above, a recipient may use your version of this file under
37	* either the BSD or the GPL.	37	* either the BSD or the GPL.
38	*/	38	*/
39		39
40	#include "eio.h"	40	#include "eio.h"
		41
		42	#ifdef EIO_STACKSIZE
		43	# define XTHREAD_STACKSIZE EIO_STACKSIZE
		44	#endif
41	#include "xthread.h"	45	#include "xthread.h"
42		46
43	#include <errno.h>	47	#include <errno.h>
44	#include <stddef.h>	48	#include <stddef.h>
45	#include <stdlib.h>	49	#include <stdlib.h>
46	#include <string.h>	50	#include <string.h>
47	#include <errno.h>	51	#include <errno.h>
48	#include <sys/types.h>	52	#include <sys/types.h>
49	#include <sys/stat.h>	53	#include <sys/stat.h>
		54	#include <sys/statvfs.h>
50	#include <limits.h>	55	#include <limits.h>
51	#include <fcntl.h>	56	#include <fcntl.h>
52	#include <assert.h>	57	#include <assert.h>
53		58
54	#ifndef EIO_FINISH	59	#ifndef EIO_FINISH
…		…
64	#endif	69	#endif
65		70
66	#ifdef _WIN32	71	#ifdef _WIN32
67		72
68	/doh/	73	/doh/
69
70	#else	74	#else
71		75
72	# include "config.h"	76	# include "config.h"
73	# include <sys/time.h>	77	# include <sys/time.h>
74	# include <sys/select.h>	78	# include <sys/select.h>
		79	# include <sys/mman.h>
75	# include <unistd.h>	80	# include <unistd.h>
76	# include <utime.h>	81	# include <utime.h>
77	# include <signal.h>	82	# include <signal.h>
78	# include <dirent.h>	83	# include <dirent.h>
79		84
		85	/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
		86	# if __FreeBSD__ \|\| defined __NetBSD__ \|\| defined __OpenBSD__
		87	# define _DIRENT_HAVE_D_TYPE /* sigh */
		88	# define D_INO(de) (de)->d_fileno
		89	# define D_NAMLEN(de) (de)->d_namlen
		90	# elif __linux \|\| defined d_ino \|\| _XOPEN_SOURCE >= 600
		91	# define D_INO(de) (de)->d_ino
		92	# endif
		93
		94	#ifdef _D_EXACT_NAMLEN
		95	# undef D_NAMLEN
		96	# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
		97	#endif
		98
		99	# ifdef _DIRENT_HAVE_D_TYPE
		100	# define D_TYPE(de) (de)->d_type
		101	# endif
		102
80	# ifndef EIO_STRUCT_DIRENT	103	# ifndef EIO_STRUCT_DIRENT
81	# define EIO_STRUCT_DIRENT struct dirent	104	# define EIO_STRUCT_DIRENT struct dirent
82	# endif	105	# endif
83		106
84	#endif	107	#endif
85		108
86	#if HAVE_SENDFILE	109	#if HAVE_SENDFILE
87	# if __linux	110	# if __linux
88	# include <sys/sendfile.h>	111	# include <sys/sendfile.h>
89	# elif __freebsd	112	# elif __FreeBSD__ \|\| defined __APPLE__
90	# include <sys/socket.h>	113	# include <sys/socket.h>
91	# include <sys/uio.h>	114	# include <sys/uio.h>
92	# elif __hpux	115	# elif __hpux
93	# include <sys/socket.h>	116	# include <sys/socket.h>
94	# elif __solaris /* not yet */	117	# elif __solaris
95	# include <sys/sendfile.h>	118	# include <sys/sendfile.h>
96	# else	119	# else
97	# error sendfile support requested but not available	120	# error sendfile support requested but not available
98	# endif	121	# endif
		122	#endif
		123
		124	#ifndef D_TYPE
		125	# define D_TYPE(de) 0
		126	#endif
		127	#ifndef D_INO
		128	# define D_INO(de) 0
		129	#endif
		130	#ifndef D_NAMLEN
		131	# define D_NAMLEN(de) strlen ((de)->d_name)
99	#endif	132	#endif
100		133
101	/* number of seconds after which an idle threads exit */	134	/* number of seconds after which an idle threads exit */
102	#define IDLE_TIMEOUT 10	135	#define IDLE_TIMEOUT 10
103		136
…		…
155	if (wrk->dirp) \	188	if (wrk->dirp) \
156	{ \	189	{ \
157	closedir (wrk->dirp); \	190	closedir (wrk->dirp); \
158	wrk->dirp = 0; \	191	wrk->dirp = 0; \
159	}	192	}
		193
160	#define ETP_WORKER_COMMON \	194	#define ETP_WORKER_COMMON \
161	void *dbuf; \	195	void *dbuf; \
162	DIR *dirp;	196	DIR *dirp;
163		197
164	/*****************************************************************************/	198	/*****************************************************************************/
165		199
166	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)	200	#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
167		201
168	/* calculcate time difference in ~1/EIO_TICKS of a second */	202	/* calculate time difference in ~1/EIO_TICKS of a second */
169	static int tvdiff (struct timeval tv1, struct timeval tv2)	203	static int tvdiff (struct timeval tv1, struct timeval tv2)
170	{	204	{
171	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS	205	return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
172	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);	206	+ ((tv2->tv_usec - tv1->tv_usec) >> 10);
173	}	207	}
…		…
565	}	599	}
566		600
567	static void etp_set_max_poll_time (double nseconds)	601	static void etp_set_max_poll_time (double nseconds)
568	{	602	{
569	if (WORDACCESS_UNSAFE) X_LOCK (reslock);	603	if (WORDACCESS_UNSAFE) X_LOCK (reslock);
570	max_poll_time = nseconds;	604	max_poll_time = nseconds * EIO_TICKS;
571	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);	605	if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
572	}	606	}
573		607
574	static void etp_set_max_poll_reqs (unsigned int maxreqs)	608	static void etp_set_max_poll_reqs (unsigned int maxreqs)
575	{	609	{
…		…
604		638
605	static void grp_try_feed (eio_req *grp)	639	static void grp_try_feed (eio_req *grp)
606	{	640	{
607	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))	641	while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
608	{	642	{
609	int old_len = grp->size;	643	grp->flags &= ~EIO_FLAG_GROUPADD;
610		644
611	EIO_FEED (grp);	645	EIO_FEED (grp);
612		646
613	/* stop if no progress has been made */	647	/* stop if no progress has been made */
614	if (old_len == grp->size)	648	if (!(grp->flags & EIO_FLAG_GROUPADD))
615	{	649	{
616	grp->feed = 0;	650	grp->feed = 0;
617	break;	651	break;
618	}	652	}
619	}	653	}
…		…
736		770
737	/*****************************************************************************/	771	/*****************************************************************************/
738	/* work around various missing functions */	772	/* work around various missing functions */
739		773
740	#if !HAVE_PREADWRITE	774	#if !HAVE_PREADWRITE
		775	# undef pread
		776	# undef pwrite
741	# define pread eio__pread	777	# define pread eio__pread
742	# define pwrite eio__pwrite	778	# define pwrite eio__pwrite
743		779
744	static ssize_t	780	static ssize_t
745	eio__pread (int fd, void *buf, size_t count, off_t offset)	781	eio__pread (int fd, void *buf, size_t count, off_t offset)
…		…
765		801
766	X_LOCK (preadwritelock);	802	X_LOCK (preadwritelock);
767	ooffset = lseek (fd, 0, SEEK_CUR);	803	ooffset = lseek (fd, 0, SEEK_CUR);
768	lseek (fd, offset, SEEK_SET);	804	lseek (fd, offset, SEEK_SET);
769	res = write (fd, buf, count);	805	res = write (fd, buf, count);
770	lseek (fd, offset, SEEK_SET);	806	lseek (fd, ooffset, SEEK_SET);
771	X_UNLOCK (preadwritelock);	807	X_UNLOCK (preadwritelock);
772		808
773	return res;	809	return res;
774	}	810	}
775	#endif	811	#endif
776		812
777	#ifndef HAVE_FUTIMES	813	#ifndef HAVE_FUTIMES
778		814
		815	# undef utimes
		816	# undef futimes
779	# define utimes(path,times) eio__utimes (path, times)	817	# define utimes(path,times) eio__utimes (path, times)
780	# define futimes(fd,times) eio__futimes (fd, times)	818	# define futimes(fd,times) eio__futimes (fd, times)
781		819
782	static int	820	static int
783	eio__utimes (const char *filename, const struct timeval times[2])	821	eio__utimes (const char *filename, const struct timeval times[2])
…		…
802	}	840	}
803		841
804	#endif	842	#endif
805		843
806	#if !HAVE_FDATASYNC	844	#if !HAVE_FDATASYNC
		845	# undef fdatasync
807	# define fdatasync fsync	846	# define fdatasync(fd) fsync (fd)
808	#endif	847	#endif
		848
		849	/* sync_file_range always needs emulation */
		850	int
		851	eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
		852	{
		853	#if HAVE_SYNC_FILE_RANGE
		854	int res;
		855
		856	if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
		857	\|\| EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
		858	\|\| EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
		859	{
		860	flags = 0
		861	\| (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
		862	\| (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
		863	\| (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
		864	}
		865
		866	res = sync_file_range (fd, offset, nbytes, flags);
		867
		868	if (!res \|\| errno != ENOSYS)
		869	return res;
		870	#endif
		871
		872	/* even though we could play tricks with the flags, it's better to always
		873	* call fdatasync, as that matches the expectation of its users best */
		874	return fdatasync (fd);
		875	}
809		876
810	#if !HAVE_READAHEAD	877	#if !HAVE_READAHEAD
		878	# undef readahead
811	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)	879	# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
812		880
813	static ssize_t	881	static ssize_t
814	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)	882	eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
815	{	883	{
…		…
842		910
843	#if HAVE_SENDFILE	911	#if HAVE_SENDFILE
844	# if __linux	912	# if __linux
845	res = sendfile (ofd, ifd, &offset, count);	913	res = sendfile (ofd, ifd, &offset, count);
846		914
847	# elif __freebsd	915	# elif __FreeBSD__
848	/*	916	/*
849	* 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
850	* wasted on making it similar to other I/O functions.	918	* wasted on making it similar to other I/O functions.
851	*/	919	*/
852	{	920	{
853	off_t sbytes;	921	off_t sbytes;
854	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);	922	res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
855		923
856	if (res < 0 && sbytes)	924	#if 0 /* according to the manpage, this is correct, but broken behaviour */
857	/* maybe only on EAGAIN: as usual, the manpage leaves you guessing */	925	/* freebsd' sendfile will return 0 on success */
		926	/* freebsd 8 documents it as only setting sbytes on EINTR and EAGAIN, but /
		927	/* not on e.g. EIO or EPIPE - sounds broken */
		928	if ((res < 0 && (errno == EAGAIN \|\| errno == EINTR) && sbytes) \|\| res == 0)
		929	res = sbytes;
		930	#endif
		931
		932	/* according to source inspection, this is correct, and useful behaviour */
		933	if (sbytes)
		934	res = sbytes;
		935	}
		936
		937	# elif defined (__APPLE__)
		938
		939	{
		940	off_t sbytes = count;
		941	res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
		942
		943	/* according to the manpage, sbytes is always valid */
		944	if (sbytes)
858	res = sbytes;	945	res = sbytes;
859	}	946	}
860		947
861	# elif __hpux	948	# elif __hpux
862	res = sendfile (ofd, ifd, offset, count, 0, 0);	949	res = sendfile (ofd, ifd, offset, count, 0, 0);
…		…
876	if (res < 0 && sbytes)	963	if (res < 0 && sbytes)
877	res = sbytes;	964	res = sbytes;
878	}	965	}
879		966
880	# endif	967	# endif
		968
		969	#elif defined (_WIN32)
		970
		971	/* does not work, just for documentation of what would need to be done */
		972	{
		973	HANDLE h = TO_SOCKET (ifd);
		974	SetFilePointer (h, offset, 0, FILE_BEGIN);
		975	res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
		976	}
		977
881	#else	978	#else
882	res = -1;	979	res = -1;
883	errno = ENOSYS;	980	errno = ENOSYS;
884	#endif	981	#endif
885		982
886	if (res < 0	983	if (res < 0
887	&& (errno == ENOSYS \|\| errno == EINVAL \|\| errno == ENOTSOCK	984	&& (errno == ENOSYS \|\| errno == EINVAL \|\| errno == ENOTSOCK
		985	/* BSDs */
		986	#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
		987	\|\| errno == ENOTSUP
		988	#endif
		989	\|\| errno == EOPNOTSUPP /* BSDs */
888	#if __solaris	990	#if __solaris
889	\|\| errno == EAFNOSUPPORT \|\| errno == EPROTOTYPE	991	\|\| errno == EAFNOSUPPORT \|\| errno == EPROTOTYPE
890	#endif	992	#endif
891	)	993	)
892	)	994	)
…		…
923	}	1025	}
924		1026
925	return res;	1027	return res;
926	}	1028	}
927		1029
		1030	static signed char
		1031	eio_dent_cmp (const eio_dirent a, const eio_dirent b)
		1032	{
		1033	return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
		1034	: a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0;
		1035	}
		1036
		1037	#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
		1038
		1039	#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
		1040	#define EIO_SORT_FAST 60 /* when to only use insertion sort */
		1041
		1042	static void
		1043	eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1044	{
		1045	unsigned char bits [9 + sizeof (ino_t) * 8];
		1046	unsigned char *bit = bits;
		1047
		1048	assert (CHAR_BIT == 8);
		1049	assert (sizeof (eio_dirent) * 8 < 256);
		1050	assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */
		1051	assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */
		1052
		1053	if (size <= EIO_SORT_FAST)
		1054	return;
		1055
		1056	/* first prepare an array of bits to test in our radix sort */
		1057	/* try to take endianness into account, as well as differences in ino_t sizes */
		1058	/* inode_bits must contain all inodes ORed together */
		1059	/* which is used to skip bits that are 0 everywhere, which is very common */
		1060	{
		1061	ino_t endianness;
		1062	int i, j;
		1063
		1064	/* we store the byte offset of byte n into byte n of "endianness" */
		1065	for (i = 0; i < sizeof (ino_t); ++i)
		1066	((unsigned char *)&endianness)[i] = i;
		1067
		1068	*bit++ = 0;
		1069
		1070	for (i = 0; i < sizeof (ino_t); ++i)
		1071	{
		1072	/* shifting off the byte offsets out of "endianness" */
		1073	int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
		1074	endianness >>= 8;
		1075
		1076	for (j = 0; j < 8; ++j)
		1077	if (inode_bits & (((ino_t)1) << (i * 8 + j)))
		1078	*bit++ = offs + j;
		1079	}
		1080
		1081	for (j = 0; j < 8; ++j)
		1082	if (score_bits & (1 << j))
		1083	bit++ = offsetof (eio_dirent, score) 8 + j;
		1084	}
		1085
		1086	/* now actually do the sorting (a variant of MSD radix sort) */
		1087	{
		1088	eio_dirent base_stk [9 + sizeof (ino_t) 8], *base;
		1089	eio_dirent end_stk [9 + sizeof (ino_t) 8], *end;
		1090	unsigned char bit_stk [9 + sizeof (ino_t) 8];
		1091	int stk_idx = 0;
		1092
		1093	base_stk [stk_idx] = dents;
		1094	end_stk [stk_idx] = dents + size;
		1095	bit_stk [stk_idx] = bit - 1;
		1096
		1097	do
		1098	{
		1099	base = base_stk [stk_idx];
		1100	end = end_stk [stk_idx];
		1101	bit = bit_stk [stk_idx];
		1102
		1103	for (;;)
		1104	{
		1105	unsigned char O = *bit >> 3;
		1106	unsigned char M = 1 << (*bit & 7);
		1107
		1108	eio_dirent *a = base;
		1109	eio_dirent *b = end;
		1110
		1111	if (b - a < EIO_SORT_CUTOFF)
		1112	break;
		1113
		1114	/* now bit-partition the array on the bit */
		1115	/* this ugly asymmetric loop seems to perform much better than typical */
		1116	/* partition algos found in the literature */
		1117	do
		1118	if (!(((unsigned char *)a)[O] & M))
		1119	++a;
		1120	else if (!(((unsigned char *)--b)[O] & M))
		1121	{
		1122	eio_dirent tmp = a; a = b; b = tmp;
		1123	++a;
		1124	}
		1125	while (b > a);
		1126
		1127	/* next bit, or stop, if no bits left in this path */
		1128	if (!*--bit)
		1129	break;
		1130
		1131	base_stk [stk_idx] = a;
		1132	end_stk [stk_idx] = end;
		1133	bit_stk [stk_idx] = bit;
		1134	++stk_idx;
		1135
		1136	end = a;
		1137	}
		1138	}
		1139	while (stk_idx--);
		1140	}
		1141	}
		1142
		1143	static void
		1144	eio_dent_insertion_sort (eio_dirent *dents, int size)
		1145	{
		1146	/* first move the smallest element to the front, to act as a sentinel */
		1147	{
		1148	int i;
		1149	eio_dirent *min = dents;
		1150
		1151	/* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
		1152	for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
		1153	if (EIO_DENT_CMP (dents [i], <, *min))
		1154	min = &dents [i];
		1155
		1156	/* swap elements 0 and j (minimum) */
		1157	{
		1158	eio_dirent tmp = dents; dents = min; min = tmp;
		1159	}
		1160	}
		1161
		1162	/* then do standard insertion sort, assuming that all elements are >= dents [0] */
		1163	{
		1164	eio_dirent i, j;
		1165
		1166	for (i = dents + 1; i < dents + size; ++i)
		1167	{
		1168	eio_dirent value = *i;
		1169
		1170	for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
		1171	j [1] = j [0];
		1172
		1173	j [1] = value;
		1174	}
		1175	}
		1176	}
		1177
		1178	static void
		1179	eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits)
		1180	{
		1181	if (size <= 1)
		1182	return; /* our insertion sort relies on size > 0 */
		1183
		1184	/* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
		1185	/* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
		1186	eio_dent_radix_sort (dents, size, score_bits, inode_bits);
		1187
		1188	/* use an insertion sort at the end, or for small arrays, */
		1189	/* as insertion sort is more efficient for small partitions */
		1190	eio_dent_insertion_sort (dents, size);
		1191	}
		1192
928	/* read a full directory */	1193	/* read a full directory */
929	static void	1194	static void
930	eio__scandir (eio_req req, etp_worker self)	1195	eio__scandir (eio_req req, etp_worker self)
931	{	1196	{
932	DIR *dirp;	1197	DIR *dirp;
933	EIO_STRUCT_DIRENT *entp;	1198	EIO_STRUCT_DIRENT *entp;
934	char name, names;	1199	char name, names;
935	int memlen = 4096;	1200	int namesalloc = 4096;
936	int memofs = 0;	1201	int namesoffs = 0;
		1202	int flags = req->int1;
		1203	eio_dirent *dents = 0;
		1204	int dentalloc = 128;
937	int res = 0;	1205	int dentoffs = 0;
		1206	ino_t inode_bits = 0;
		1207
		1208	req->result = -1;
		1209
		1210	if (!(flags & EIO_READDIR_DENTS))
		1211	flags &= ~(EIO_READDIR_DIRS_FIRST \| EIO_READDIR_STAT_ORDER);
938		1212
939	X_LOCK (wrklock);	1213	X_LOCK (wrklock);
940	/* the corresponding closedir is in ETP_WORKER_CLEAR */	1214	/* the corresponding closedir is in ETP_WORKER_CLEAR */
941	self->dirp = dirp = opendir (req->ptr1);	1215	self->dirp = dirp = opendir (req->ptr1);
942	req->flags \|= EIO_FLAG_PTR2_FREE;	1216	req->flags \|= EIO_FLAG_PTR1_FREE \| EIO_FLAG_PTR2_FREE;
		1217	req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
943	req->ptr2 = names = malloc (memlen);	1218	req->ptr2 = names = malloc (namesalloc);
944	X_UNLOCK (wrklock);	1219	X_UNLOCK (wrklock);
945		1220
946	if (dirp && names)	1221	if (dirp && names && (!flags \|\| dents))
947	for (;;)	1222	for (;;)
948	{	1223	{
949	errno = 0;	1224	errno = 0;
950	entp = readdir (dirp);	1225	entp = readdir (dirp);
951		1226
952	if (!entp)	1227	if (!entp)
		1228	{
		1229	if (errno)
		1230	break;
		1231
		1232	/* sort etc. */
		1233	req->int1 = flags;
		1234	req->result = dentoffs;
		1235
		1236	if (flags & EIO_READDIR_STAT_ORDER)
		1237	eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */
		1238	else if (flags & EIO_READDIR_DIRS_FIRST)
		1239	if (flags & EIO_READDIR_FOUND_UNKNOWN)
		1240	eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
		1241	else
		1242	{
		1243	/* in this case, all is known, and we just put dirs first and sort them */
		1244	eio_dirent *oth = dents + dentoffs;
		1245	eio_dirent *dir = dents;
		1246
		1247	/* now partition dirs to the front, and non-dirs to the back */
		1248	/* by walking from both sides and swapping if necessary */
		1249	/* also clear score, so it doesn't influence sorting */
		1250	while (oth > dir)
		1251	{
		1252	if (dir->type == EIO_DT_DIR)
		1253	++dir;
		1254	else if ((--oth)->type == EIO_DT_DIR)
		1255	{
		1256	eio_dirent tmp = dir; dir = oth; oth = tmp;
		1257
		1258	++dir;
		1259	}
		1260	}
		1261
		1262	/* now sort the dirs only */
		1263	eio_dent_sort (dents, dir - dents, 0, inode_bits);
		1264	}
		1265
953	break;	1266	break;
		1267	}
954		1268
		1269	/* now add the entry to our list(s) */
955	name = entp->d_name;	1270	name = entp->d_name;
956		1271
		1272	/* skip . and .. entries */
957	if (name [0] != '.' \|\| (name [1] && (name [1] != '.' \|\| name [2])))	1273	if (name [0] != '.' \|\| (name [1] && (name [1] != '.' \|\| name [2])))
958	{	1274	{
959	int len = strlen (name) + 1;	1275	int len = D_NAMLEN (entp) + 1;
960		1276
961	res++;	1277	while (expect_false (namesoffs + len > namesalloc))
962
963	while (memofs + len > memlen)
964	{	1278	{
965	memlen *= 2;	1279	namesalloc *= 2;
966	X_LOCK (wrklock);	1280	X_LOCK (wrklock);
967	req->ptr2 = names = realloc (names, memlen);	1281	req->ptr2 = names = realloc (names, namesalloc);
968	X_UNLOCK (wrklock);	1282	X_UNLOCK (wrklock);
969		1283
970	if (!names)	1284	if (!names)
971	break;	1285	break;
972	}	1286	}
973		1287
974	memcpy (names + memofs, name, len);	1288	memcpy (names + namesoffs, name, len);
		1289
		1290	if (dents)
		1291	{
		1292	struct eio_dirent *ent;
		1293
		1294	if (expect_false (dentoffs == dentalloc))
		1295	{
		1296	dentalloc *= 2;
		1297	X_LOCK (wrklock);
		1298	req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
		1299	X_UNLOCK (wrklock);
		1300
		1301	if (!dents)
		1302	break;
		1303	}
		1304
		1305	ent = dents + dentoffs;
		1306
		1307	ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
		1308	ent->namelen = len - 1;
		1309	ent->inode = D_INO (entp);
		1310
		1311	inode_bits \|= ent->inode;
		1312
		1313	switch (D_TYPE (entp))
		1314	{
		1315	default:
		1316	ent->type = EIO_DT_UNKNOWN;
		1317	flags \|= EIO_READDIR_FOUND_UNKNOWN;
		1318	break;
		1319
		1320	#ifdef DT_FIFO
		1321	case DT_FIFO: ent->type = EIO_DT_FIFO; break;
		1322	#endif
		1323	#ifdef DT_CHR
		1324	case DT_CHR: ent->type = EIO_DT_CHR; break;
		1325	#endif
		1326	#ifdef DT_MPC
		1327	case DT_MPC: ent->type = EIO_DT_MPC; break;
		1328	#endif
		1329	#ifdef DT_DIR
		1330	case DT_DIR: ent->type = EIO_DT_DIR; break;
		1331	#endif
		1332	#ifdef DT_NAM
		1333	case DT_NAM: ent->type = EIO_DT_NAM; break;
		1334	#endif
		1335	#ifdef DT_BLK
		1336	case DT_BLK: ent->type = EIO_DT_BLK; break;
		1337	#endif
		1338	#ifdef DT_MPB
		1339	case DT_MPB: ent->type = EIO_DT_MPB; break;
		1340	#endif
		1341	#ifdef DT_REG
		1342	case DT_REG: ent->type = EIO_DT_REG; break;
		1343	#endif
		1344	#ifdef DT_NWK
		1345	case DT_NWK: ent->type = EIO_DT_NWK; break;
		1346	#endif
		1347	#ifdef DT_CMP
		1348	case DT_CMP: ent->type = EIO_DT_CMP; break;
		1349	#endif
		1350	#ifdef DT_LNK
		1351	case DT_LNK: ent->type = EIO_DT_LNK; break;
		1352	#endif
		1353	#ifdef DT_SOCK
		1354	case DT_SOCK: ent->type = EIO_DT_SOCK; break;
		1355	#endif
		1356	#ifdef DT_DOOR
		1357	case DT_DOOR: ent->type = EIO_DT_DOOR; break;
		1358	#endif
		1359	#ifdef DT_WHT
		1360	case DT_WHT: ent->type = EIO_DT_WHT; break;
		1361	#endif
		1362	}
		1363
		1364	ent->score = 7;
		1365
		1366	if (flags & EIO_READDIR_DIRS_FIRST)
		1367	{
		1368	if (ent->type == EIO_DT_UNKNOWN)
		1369	{
		1370	if (name == '.') / leading dots are likely directories, and, in any case, rare */
		1371	ent->score = 1;
		1372	else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
		1373	ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
		1374	}
		1375	else if (ent->type == EIO_DT_DIR)
		1376	ent->score = 0;
		1377	}
		1378	}
		1379
975	memofs += len;	1380	namesoffs += len;
		1381	++dentoffs;
		1382	}
		1383
		1384	if (EIO_CANCELLED (req))
		1385	{
		1386	errno = ECANCELED;
		1387	break;
976	}	1388	}
977	}	1389	}
978
979	if (errno)
980	res = -1;
981
982	req->result = res;
983	}	1390	}
984		1391
985	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)	1392	#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
986	# define msync(a,b,c) ENOSYS	1393	# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
		1394	#else
		1395
		1396	int
		1397	eio__msync (void *mem, size_t len, int flags)
		1398	{
		1399	if (EIO_MS_ASYNC != MS_SYNC
		1400	\|\| EIO_MS_INVALIDATE != MS_INVALIDATE
		1401	\|\| EIO_MS_SYNC != MS_SYNC)
		1402	{
		1403	flags = 0
		1404	\| (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
		1405	\| (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
		1406	\| (flags & EIO_MS_SYNC ? MS_SYNC : 0);
		1407	}
		1408
		1409	return msync (mem, len, flags);
		1410	}
		1411
987	#endif	1412	#endif
988		1413
989	int	1414	int
990	eio__mtouch (void *mem, size_t len, int flags)	1415	eio__mtouch (void *mem, size_t len, int flags)
991	{	1416	{
…		…
998		1423
999	if (!page)	1424	if (!page)
1000	page = sysconf (_SC_PAGESIZE);	1425	page = sysconf (_SC_PAGESIZE);
1001	#endif	1426	#endif
1002		1427
		1428	/* round down to start of page, although this is probably useless */
1003	addr &= ~(page - 1); /* assume page size is always a power of two */	1429	addr &= ~(page - 1); /* assume page size is always a power of two */
1004		1430
1005	if (addr < end)	1431	if (addr < end)
1006	if (flags) /* modify */	1432	if (flags & EIO_MT_MODIFY) /* modify */
1007	do { ((volatile sig_atomic_t )addr) \|= 0; } while ((addr += page) < len);	1433	do { ((volatile sig_atomic_t )addr) \|= 0; } while ((addr += page) < len);
1008	else	1434	else
1009	do { ((volatile sig_atomic_t )addr) ; } while ((addr += page) < len);	1435	do { ((volatile sig_atomic_t )addr) ; } while ((addr += page) < len);
1010		1436
1011	return 0;	1437	return 0;
…		…
1159	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1585	case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1160	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;	1586	req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1161	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));	1587	case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1162	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;	1588	req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1163		1589
		1590	case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1591	req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1592	case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
		1593	req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
		1594
1164	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;	1595	case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1165	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;	1596	case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1166	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;	1597	case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1167	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;	1598	case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1168	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;	1599	case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
…		…
1183	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;	1614	req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
1184		1615
1185	case EIO_SYNC: req->result = 0; sync (); break;	1616	case EIO_SYNC: req->result = 0; sync (); break;
1186	case EIO_FSYNC: req->result = fsync (req->int1); break;	1617	case EIO_FSYNC: req->result = fsync (req->int1); break;
1187	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;	1618	case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1188	case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break;	1619	case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1189	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;	1620	case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break;
		1621	case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1190		1622
1191	case EIO_READDIR: eio__scandir (req, self); break;	1623	case EIO_READDIR: eio__scandir (req, self); break;
1192		1624
1193	case EIO_BUSY:	1625	case EIO_BUSY:
1194	#ifdef _WIN32	1626	#ifdef _WIN32
…		…
1226		1658
1227	req->result = req->type == EIO_FUTIME	1659	req->result = req->type == EIO_FUTIME
1228	? futimes (req->int1, times)	1660	? futimes (req->int1, times)
1229	: utimes (req->ptr1, times);	1661	: utimes (req->ptr1, times);
1230	}	1662	}
		1663	break;
1231		1664
1232	case EIO_GROUP:	1665	case EIO_GROUP:
1233	abort (); /* handled in eio_request */	1666	abort (); /* handled in eio_request */
1234		1667
1235	case EIO_NOP:	1668	case EIO_NOP:
…		…
1278	eio_req eio_mtouch (void addr, size_t length, int flags, int pri, eio_cb cb, void *data)	1711	eio_req eio_mtouch (void addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1279	{	1712	{
1280	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;	1713	REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1281	}	1714	}
1282		1715
		1716	eio_req eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void data)
		1717	{
		1718	REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
		1719	}
		1720
1283	eio_req eio_fdatasync (int fd, int pri, eio_cb cb, void data)	1721	eio_req eio_fdatasync (int fd, int pri, eio_cb cb, void data)
1284	{	1722	{
1285	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;	1723	REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1286	}	1724	}
1287		1725
…		…
1306	}	1744	}
1307		1745
1308	eio_req eio_fstat (int fd, int pri, eio_cb cb, void data)	1746	eio_req eio_fstat (int fd, int pri, eio_cb cb, void data)
1309	{	1747	{
1310	REQ (EIO_FSTAT); req->int1 = fd; SEND;	1748	REQ (EIO_FSTAT); req->int1 = fd; SEND;
		1749	}
		1750
		1751	eio_req eio_fstatvfs (int fd, int pri, eio_cb cb, void data)
		1752	{
		1753	REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
1311	}	1754	}
1312		1755
1313	eio_req eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void data)	1756	eio_req eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void data)
1314	{	1757	{
1315	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;	1758	REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
…		…
1389	eio_req eio_lstat (const char path, int pri, eio_cb cb, void *data)	1832	eio_req eio_lstat (const char path, int pri, eio_cb cb, void *data)
1390	{	1833	{
1391	return eio__1path (EIO_LSTAT, path, pri, cb, data);	1834	return eio__1path (EIO_LSTAT, path, pri, cb, data);
1392	}	1835	}
1393		1836
		1837	eio_req eio_statvfs (const char path, int pri, eio_cb cb, void *data)
		1838	{
		1839	return eio__1path (EIO_STATVFS, path, pri, cb, data);
		1840	}
		1841
1394	eio_req eio_unlink (const char path, int pri, eio_cb cb, void *data)	1842	eio_req eio_unlink (const char path, int pri, eio_cb cb, void *data)
1395	{	1843	{
1396	return eio__1path (EIO_UNLINK, path, pri, cb, data);	1844	return eio__1path (EIO_UNLINK, path, pri, cb, data);
1397	}	1845	}
1398		1846
1399	eio_req eio_rmdir (const char path, int pri, eio_cb cb, void *data)	1847	eio_req eio_rmdir (const char path, int pri, eio_cb cb, void *data)
1400	{	1848	{
1401	return eio__1path (EIO_RMDIR, path, pri, cb, data);	1849	return eio__1path (EIO_RMDIR, path, pri, cb, data);
1402	}	1850	}
1403		1851
1404	eio_req eio_readdir (const char path, int pri, eio_cb cb, void *data)	1852	eio_req eio_readdir (const char path, int flags, int pri, eio_cb cb, void *data)
1405	{	1853	{
1406	return eio__1path (EIO_READDIR, path, pri, cb, data);	1854	REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1407	}	1855	}
1408		1856
1409	eio_req eio_mknod (const char path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)	1857	eio_req eio_mknod (const char path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1410	{	1858	{
1411	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;	1859	REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND;
…		…
1480		1928
1481	void eio_grp_add (eio_req grp, eio_req req)	1929	void eio_grp_add (eio_req grp, eio_req req)
1482	{	1930	{
1483	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));	1931	assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
1484		1932
		1933	grp->flags \|= EIO_FLAG_GROUPADD;
		1934
1485	++grp->size;	1935	++grp->size;
1486	req->grp = grp;	1936	req->grp = grp;
1487		1937
1488	req->grp_prev = 0;	1938	req->grp_prev = 0;
1489	req->grp_next = grp->grp_first;	1939	req->grp_next = grp->grp_first;
…		…
1498	/* misc garbage */	1948	/* misc garbage */
1499		1949
1500	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)	1950	ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1501	{	1951	{
1502	etp_worker wrk;	1952	etp_worker wrk;
		1953	ssize_t ret;
1503		1954
1504	wrk.dbuf = 0;	1955	wrk.dbuf = 0;
1505		1956
1506	eio__sendfile (ofd, ifd, offset, count, &wrk);	1957	ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
1507		1958
1508	if (wrk.dbuf)	1959	if (wrk.dbuf)
1509	free (wrk.dbuf);	1960	free (wrk.dbuf);
1510	}
1511		1961
		1962	return ret;
		1963	}
		1964

Diff Legend

-–
+Removed lines
-+
+Added lines
-<
+Changed lines
->
+Changed lines

Comparing libeio/eio.c (file contents): Revision 1.24 by root, Tue Sep 30 13:17:32 2008 UTC vs. Revision 1.53 by root, Sat Jan 9 10:03:09 2010 UTC

Diff Legend

Comparing libeio/eio.c (file contents):
Revision 1.24 by root, Tue Sep 30 13:17:32 2008 UTC vs.
Revision 1.53 by root, Sat Jan 9 10:03:09 2010 UTC