ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libeio/eio.c
(Generate patch)

Comparing libeio/eio.c (file contents):
Revision 1.76 by root, Tue Jul 5 09:24:12 2011 UTC vs.
Revision 1.93 by root, Mon Jul 18 01:27:03 2011 UTC

… …
54#include <stdlib.h> 54#include <stdlib.h>
55#include <string.h> 55#include <string.h>
56#include <errno.h> 56#include <errno.h>
57#include <sys/types.h> 57#include <sys/types.h>
58#include <sys/stat.h> 58#include <sys/stat.h>
59#include <sys/statvfs.h>
60#include <limits.h> 59#include <limits.h>
61#include <fcntl.h> 60#include <fcntl.h>
62#include <assert.h> 61#include <assert.h>
63 62
64/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */ 63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
65/* intptr_t only comes form stdint.h, says idiot openbsd coder */ 64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
66#if HAVE_STDINT_H 65#if HAVE_STDINT_H
67# include <stdint.h> 66# include <stdint.h>
68#endif 67#endif
69 68
69#ifndef ECANCELED
70# define ECANCELED EDOM
71#endif
72#ifndef ELOOP
73# define ELOOP EDOM
74#endif
75
76static void eio_destroy (eio_req *req);
77
70#ifndef EIO_FINISH 78#ifndef EIO_FINISH
71# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 79# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
72#endif 80#endif
73 81
74#ifndef EIO_DESTROY 82#ifndef EIO_DESTROY
… …
77 85
78#ifndef EIO_FEED 86#ifndef EIO_FEED
79# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 87# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
80#endif 88#endif
81 89
90#ifndef EIO_FD_TO_WIN32_HANDLE
91# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
92#endif
93#ifndef EIO_WIN32_HANDLE_TO_FD
94# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
95#endif
96
97#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
98
99#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
100
82#ifdef _WIN32 101#ifdef _WIN32
83 102
84 /*doh*/ 103 #undef PAGESIZE
104 #define PAGESIZE 4096 /* GetSystemInfo? */
105
106 #ifdef EIO_STRUCT_STATI64
107 #define stat(path,buf) _stati64 (path,buf)
108 #define fstat(fd,buf) _fstati64 (path,buf)
109 #endif
110 #define lstat(path,buf) stat (path,buf)
111 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
112 #define mkdir(path,mode) _mkdir (path)
113 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
114
115 #define chmod(path,mode) _chmod (path, mode)
116 #define dup(fd) _dup (fd)
117 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
118
119 #define fchmod(fd,mode) EIO_ENOSYS ()
120 #define chown(path,uid,gid) EIO_ENOSYS ()
121 #define fchown(fd,uid,gid) EIO_ENOSYS ()
122 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
123 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
124 #define mknod(path,mode,dev) EIO_ENOSYS ()
125 #define sync() EIO_ENOSYS ()
126 #define readlink(path,buf,s) EIO_ENOSYS ()
127 #define statvfs(path,buf) EIO_ENOSYS ()
128 #define fstatvfs(fd,buf) EIO_ENOSYS ()
129
130 /* we could even stat and see if it exists */
131 static int
132 symlink (const char *old, const char *neu)
133 {
134 #if WINVER >= 0x0600
135 if (CreateSymbolicLink (neu, old, 1))
136 return 0;
137
138 if (CreateSymbolicLink (neu, old, 0))
139 return 0;
140 #endif
141
142 return EIO_ERRNO (ENOENT, -1);
143 }
144
145 /* POSIX API only */
146 #define CreateHardLink(neu,old,flags) 0
147 #define CreateSymbolicLink(neu,old,flags) 0
148
149 struct statvfs
150 {
151 int dummy;
152 };
153
154 #define DT_DIR EIO_DT_DIR
155 #define DT_REG EIO_DT_REG
156 #define D_NAME(entp) entp.cFileName
157 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
158
85#else 159#else
86 160
87# include <sys/time.h> 161 #include <sys/time.h>
88# include <sys/select.h> 162 #include <sys/select.h>
163 #include <sys/statvfs.h>
89# include <unistd.h> 164 #include <unistd.h>
90# include <utime.h> 165 #include <utime.h>
91# include <signal.h> 166 #include <signal.h>
92# include <dirent.h> 167 #include <dirent.h>
93 168
94#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES 169 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
95# include <sys/mman.h> 170 #include <sys/mman.h>
96#endif 171 #endif
97 172
173 #define D_NAME(entp) entp->d_name
174
98/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ 175 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
99# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ 176 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
100# define _DIRENT_HAVE_D_TYPE /* sigh */ 177 #define _DIRENT_HAVE_D_TYPE /* sigh */
101# define D_INO(de) (de)->d_fileno 178 #define D_INO(de) (de)->d_fileno
102# define D_NAMLEN(de) (de)->d_namlen 179 #define D_NAMLEN(de) (de)->d_namlen
103# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 180 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
104# define D_INO(de) (de)->d_ino 181 #define D_INO(de) (de)->d_ino
105# endif 182 #endif
106 183
107#ifdef _D_EXACT_NAMLEN 184 #ifdef _D_EXACT_NAMLEN
108# undef D_NAMLEN 185 #undef D_NAMLEN
109# define D_NAMLEN(de) _D_EXACT_NAMLEN (de) 186 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
110#endif 187 #endif
111 188
112# ifdef _DIRENT_HAVE_D_TYPE 189 #ifdef _DIRENT_HAVE_D_TYPE
113# define D_TYPE(de) (de)->d_type 190 #define D_TYPE(de) (de)->d_type
114# endif 191 #endif
115 192
116# ifndef EIO_STRUCT_DIRENT 193 #ifndef EIO_STRUCT_DIRENT
117# define EIO_STRUCT_DIRENT struct dirent 194 #define EIO_STRUCT_DIRENT struct dirent
118# endif 195 #endif
119 196
120#endif 197#endif
121 198
122#if HAVE_SENDFILE 199#if HAVE_SENDFILE
123# if __linux 200# if __linux
… …
139#endif 216#endif
140#ifndef D_INO 217#ifndef D_INO
141# define D_INO(de) 0 218# define D_INO(de) 0
142#endif 219#endif
143#ifndef D_NAMLEN 220#ifndef D_NAMLEN
144# define D_NAMLEN(de) strlen ((de)->d_name) 221# define D_NAMLEN(entp) strlen (D_NAME (entp))
145#endif 222#endif
146 223
147/* used for struct dirent, AIX doesn't provide it */ 224/* used for struct dirent, AIX doesn't provide it */
148#ifndef NAME_MAX 225#ifndef NAME_MAX
149# define NAME_MAX 4096 226# define NAME_MAX 4096
… …
183#define ETP_WORKER_CLEAR(req) \ 260#define ETP_WORKER_CLEAR(req) \
184 if (wrk->dbuf) \ 261 if (wrk->dbuf) \
185 { \ 262 { \
186 free (wrk->dbuf); \ 263 free (wrk->dbuf); \
187 wrk->dbuf = 0; \ 264 wrk->dbuf = 0; \
188 } \
189 \
190 if (wrk->dirp) \
191 { \
192 closedir (wrk->dirp); \
193 wrk->dirp = 0; \
194 } 265 }
195 266
196#define ETP_WORKER_COMMON \ 267#define ETP_WORKER_COMMON \
197 void *dbuf; \ 268 void *dbuf;
198 DIR *dirp;
199 269
200/*****************************************************************************/ 270/*****************************************************************************/
201 271
202#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 272#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
203 273
… …
232/* 302/*
233 * make our pread/pwrite emulation safe against themselves, but not against 303 * make our pread/pwrite emulation safe against themselves, but not against
234 * normal read/write by using a mutex. slows down execution a lot, 304 * normal read/write by using a mutex. slows down execution a lot,
235 * but that's your problem, not mine. 305 * but that's your problem, not mine.
236 */ 306 */
237static xmutex_t preadwritelock = X_MUTEX_INIT; 307static xmutex_t preadwritelock;
238#endif 308#endif
239 309
240typedef struct etp_worker 310typedef struct etp_worker
241{ 311{
242 /* locked by wrklock */ 312 /* locked by wrklock */
… …
248 ETP_REQ *req; /* currently processed request */ 318 ETP_REQ *req; /* currently processed request */
249 319
250 ETP_WORKER_COMMON 320 ETP_WORKER_COMMON
251} etp_worker; 321} etp_worker;
252 322
253static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 323static etp_worker wrk_first; /* NOT etp */
254 324
255#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 325#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
256#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 326#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
257 327
258/* worker threads management */ 328/* worker threads management */
… …
329} etp_reqq; 399} etp_reqq;
330 400
331static etp_reqq req_queue; 401static etp_reqq req_queue;
332static etp_reqq res_queue; 402static etp_reqq res_queue;
333 403
404static void ecb_noinline ecb_cold
405reqq_init (etp_reqq *q)
406{
407 int pri;
408
409 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
410 q->qs[pri] = q->qe[pri] = 0;
411
412 q->size = 0;
413}
414
334static int ecb_noinline 415static int ecb_noinline
335reqq_push (etp_reqq *q, ETP_REQ *req) 416reqq_push (etp_reqq *q, ETP_REQ *req)
336{ 417{
337 int pri = req->pri; 418 int pri = req->pri;
338 req->next = 0; 419 req->next = 0;
… …
372 } 453 }
373 454
374 abort (); 455 abort ();
375} 456}
376 457
377static void ecb_cold 458static int ecb_cold
378etp_thread_init (void) 459etp_init (void (*want_poll)(void), void (*done_poll)(void))
379{ 460{
380 X_MUTEX_CREATE (wrklock); 461 X_MUTEX_CREATE (wrklock);
381 X_MUTEX_CREATE (reslock); 462 X_MUTEX_CREATE (reslock);
382 X_MUTEX_CREATE (reqlock); 463 X_MUTEX_CREATE (reqlock);
383 X_COND_CREATE (reqwait); 464 X_COND_CREATE (reqwait);
384}
385 465
386static void ecb_cold 466 reqq_init (&req_queue);
387etp_atfork_prepare (void) 467 reqq_init (&res_queue);
388{
389 X_LOCK (wrklock);
390 X_LOCK (reqlock);
391 X_LOCK (reslock);
392#if !HAVE_PREADWRITE
393 X_LOCK (preadwritelock);
394#endif
395}
396 468
397static void ecb_cold 469 wrk_first.next =
398etp_atfork_parent (void) 470 wrk_first.prev = &wrk_first;
399{
400#if !HAVE_PREADWRITE
401 X_UNLOCK (preadwritelock);
402#endif
403 X_UNLOCK (reslock);
404 X_UNLOCK (reqlock);
405 X_UNLOCK (wrklock);
406}
407
408static void ecb_cold
409etp_atfork_child (void)
410{
411 ETP_REQ *prv;
412
413 while ((prv = reqq_shift (&req_queue)))
414 ETP_DESTROY (prv);
415
416 while ((prv = reqq_shift (&res_queue)))
417 ETP_DESTROY (prv);
418
419 while (wrk_first.next != &wrk_first)
420 {
421 etp_worker *wrk = wrk_first.next;
422
423 if (wrk->req)
424 ETP_DESTROY (wrk->req);
425
426 etp_worker_clear (wrk);
427 etp_worker_free (wrk);
428 }
429 471
430 started = 0; 472 started = 0;
431 idle = 0; 473 idle = 0;
432 nreqs = 0; 474 nreqs = 0;
433 nready = 0; 475 nready = 0;
434 npending = 0; 476 npending = 0;
435
436 etp_thread_init ();
437}
438
439static void ecb_cold
440etp_once_init (void)
441{
442 etp_thread_init ();
443 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
444}
445
446static int ecb_cold
447etp_init (void (*want_poll)(void), void (*done_poll)(void))
448{
449 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
450
451 pthread_once (&doinit, etp_once_init);
452 477
453 want_poll_cb = want_poll; 478 want_poll_cb = want_poll;
454 done_poll_cb = done_poll; 479 done_poll_cb = done_poll;
455 480
456 return 0; 481 return 0;
… …
583} 608}
584 609
585static void 610static void
586etp_cancel (ETP_REQ *req) 611etp_cancel (ETP_REQ *req)
587{ 612{
588 X_LOCK (wrklock); 613 req->cancelled = 1;
589 req->flags |= EIO_FLAG_CANCELLED;
590 X_UNLOCK (wrklock);
591 614
592 eio_grp_cancel (req); 615 eio_grp_cancel (req);
593} 616}
594 617
595static void 618static void
… …
711 return eio_finish (grp); 734 return eio_finish (grp);
712 else 735 else
713 return 0; 736 return 0;
714} 737}
715 738
716void 739static void
717eio_destroy (eio_req *req) 740eio_destroy (eio_req *req)
718{ 741{
719 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 742 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
720 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 743 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
721 744
… …
739 if (grp->grp_first == req) 762 if (grp->grp_first == req)
740 grp->grp_first = req->grp_next; 763 grp->grp_first = req->grp_next;
741 764
742 res2 = grp_dec (grp); 765 res2 = grp_dec (grp);
743 766
744 if (!res && res2) 767 if (!res)
745 res = res2; 768 res = res2;
746 } 769 }
747 770
748 eio_destroy (req); 771 eio_destroy (req);
749 772
… …
834 return etp_poll (); 857 return etp_poll ();
835} 858}
836 859
837/*****************************************************************************/ 860/*****************************************************************************/
838/* work around various missing functions */ 861/* work around various missing functions */
839
840#if _POSIX_VERSION < 200809L
841# define realpath(path,resolved_path) (errno = ENOSYS, 0)
842#endif
843 862
844#if !HAVE_PREADWRITE 863#if !HAVE_PREADWRITE
845# undef pread 864# undef pread
846# undef pwrite 865# undef pwrite
847# define pread eio__pread 866# define pread eio__pread
848# define pwrite eio__pwrite 867# define pwrite eio__pwrite
849 868
850static ssize_t 869static eio_ssize_t
851eio__pread (int fd, void *buf, size_t count, off_t offset) 870eio__pread (int fd, void *buf, size_t count, off_t offset)
852{ 871{
853 ssize_t res; 872 eio_ssize_t res;
854 off_t ooffset; 873 off_t ooffset;
855 874
856 X_LOCK (preadwritelock); 875 X_LOCK (preadwritelock);
857 ooffset = lseek (fd, 0, SEEK_CUR); 876 ooffset = lseek (fd, 0, SEEK_CUR);
858 lseek (fd, offset, SEEK_SET); 877 lseek (fd, offset, SEEK_SET);
… …
861 X_UNLOCK (preadwritelock); 880 X_UNLOCK (preadwritelock);
862 881
863 return res; 882 return res;
864} 883}
865 884
866static ssize_t 885static eio_ssize_t
867eio__pwrite (int fd, void *buf, size_t count, off_t offset) 886eio__pwrite (int fd, void *buf, size_t count, off_t offset)
868{ 887{
869 ssize_t res; 888 eio_ssize_t res;
870 off_t ooffset; 889 off_t ooffset;
871 890
872 X_LOCK (preadwritelock); 891 X_LOCK (preadwritelock);
873 ooffset = lseek (fd, 0, SEEK_CUR); 892 ooffset = lseek (fd, 0, SEEK_CUR);
874 lseek (fd, offset, SEEK_SET); 893 lseek (fd, offset, SEEK_SET);
… …
948 /* even though we could play tricks with the flags, it's better to always 967 /* even though we could play tricks with the flags, it's better to always
949 * call fdatasync, as that matches the expectation of its users best */ 968 * call fdatasync, as that matches the expectation of its users best */
950 return fdatasync (fd); 969 return fdatasync (fd);
951} 970}
952 971
972static int
973eio__fallocate (int fd, int mode, off_t offset, size_t len)
974{
975#if HAVE_FALLOCATE
976 return fallocate (fd, mode, offset, len);
977#else
978 errno = ENOSYS;
979 return -1;
980#endif
981}
982
953#if !HAVE_READAHEAD 983#if !HAVE_READAHEAD
954# undef readahead 984# undef readahead
955# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 985# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
956 986
957static ssize_t 987static eio_ssize_t
958eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 988eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
959{ 989{
960 size_t todo = count; 990 size_t todo = count;
961 dBUF; 991 dBUF;
962 992
… …
974} 1004}
975 1005
976#endif 1006#endif
977 1007
978/* sendfile always needs emulation */ 1008/* sendfile always needs emulation */
979static ssize_t 1009static eio_ssize_t
980eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1010eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self)
981{ 1011{
982 ssize_t written = 0; 1012 eio_ssize_t written = 0;
983 ssize_t res; 1013 eio_ssize_t res;
984 1014
985 if (!count) 1015 if (!count)
986 return 0; 1016 return 0;
987 1017
988 for (;;) 1018 for (;;)
989 { 1019 {
1020#ifdef __APPLE__
1021# undef HAVE_SENDFILE /* broken, as everything on os x */
1022#endif
990#if HAVE_SENDFILE 1023#if HAVE_SENDFILE
991# if __linux 1024# if __linux
992 off_t soffset = offset; 1025 off_t soffset = offset;
993 res = sendfile (ofd, ifd, &soffset, count); 1026 res = sendfile (ofd, ifd, &soffset, count);
994 1027
… …
1010 1043
1011 /* according to source inspection, this is correct, and useful behaviour */ 1044 /* according to source inspection, this is correct, and useful behaviour */
1012 if (sbytes) 1045 if (sbytes)
1013 res = sbytes; 1046 res = sbytes;
1014 1047
1015# elif defined (__APPLE__) && 0 /* broken, as everything on os x */ 1048# elif defined (__APPLE__)
1016 off_t sbytes = count; 1049 off_t sbytes = count;
1017 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 1050 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1018 1051
1019 /* according to the manpage, sbytes is always valid */ 1052 /* according to the manpage, sbytes is always valid */
1020 if (sbytes) 1053 if (sbytes)
… …
1037 if (res < 0 && sbytes) 1070 if (res < 0 && sbytes)
1038 res = sbytes; 1071 res = sbytes;
1039 1072
1040# endif 1073# endif
1041 1074
1042#elif defined (_WIN32) 1075#elif defined (_WIN32) && 0
1043 /* does not work, just for documentation of what would need to be done */ 1076 /* does not work, just for documentation of what would need to be done */
1044 /* actually, cannot be done like this, as TransmitFile changes the file offset, */ 1077 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1045 /* libeio guarantees that the file offset does not change, and windows */ 1078 /* libeio guarantees that the file offset does not change, and windows */
1046 /* has no way to get an independent handle to the same file description */ 1079 /* has no way to get an independent handle to the same file description */
1047 HANDLE h = TO_SOCKET (ifd); 1080 HANDLE h = TO_SOCKET (ifd);
… …
1080 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1113 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1081 /* BSDs */ 1114 /* BSDs */
1082#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1115#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1083 || errno == ENOTSUP 1116 || errno == ENOTSUP
1084#endif 1117#endif
1118#ifdef EOPNOTSUPP /* windows */
1085 || errno == EOPNOTSUPP /* BSDs */ 1119 || errno == EOPNOTSUPP /* BSDs */
1120#endif
1086#if __solaris 1121#if __solaris
1087 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1122 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1088#endif 1123#endif
1089 ) 1124 )
1090 ) 1125 )
… …
1094 1129
1095 res = 0; 1130 res = 0;
1096 1131
1097 while (count) 1132 while (count)
1098 { 1133 {
1099 ssize_t cnt; 1134 eio_ssize_t cnt;
1100 1135
1101 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1136 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1102 1137
1103 if (cnt <= 0) 1138 if (cnt <= 0)
1104 { 1139 {
… …
1121 } 1156 }
1122 1157
1123 return res; 1158 return res;
1124} 1159}
1125 1160
1161#ifdef PAGESIZE
1162# define eio_pagesize() PAGESIZE
1163#else
1164static intptr_t
1165eio_pagesize (void)
1166{
1167 static intptr_t page;
1168
1169 if (!page)
1170 page = sysconf (_SC_PAGESIZE);
1171
1172 return page;
1173}
1174#endif
1175
1176static void
1177eio_page_align (void **addr, size_t *length)
1178{
1179 intptr_t mask = eio_pagesize () - 1;
1180
1181 /* round down addr */
1182 intptr_t adj = mask & (intptr_t)*addr;
1183
1184 *addr = (void *)((intptr_t)*addr - adj);
1185 *length += adj;
1186
1187 /* round up length */
1188 *length = (*length + mask) & ~mask;
1189}
1190
1191#if !_POSIX_MEMLOCK
1192# define eio__mlockall(a) EIO_ENOSYS ()
1193#else
1194
1195static int
1196eio__mlockall (int flags)
1197{
1198 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1199 extern int mallopt (int, int);
1200 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1201 #endif
1202
1203 if (EIO_MCL_CURRENT != MCL_CURRENT
1204 || EIO_MCL_FUTURE != MCL_FUTURE)
1205 {
1206 flags = 0
1207 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1208 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1209 }
1210
1211 return mlockall (flags);
1212}
1213#endif
1214
1215#if !_POSIX_MEMLOCK_RANGE
1216# define eio__mlock(a,b) EIO_ENOSYS ()
1217#else
1218
1219static int
1220eio__mlock (void *addr, size_t length)
1221{
1222 eio_page_align (&addr, &length);
1223
1224 return mlock (addr, length);
1225}
1226
1227#endif
1228
1229#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1230# define eio__msync(a,b,c) EIO_ENOSYS ()
1231#else
1232
1233static int
1234eio__msync (void *mem, size_t len, int flags)
1235{
1236 eio_page_align (&mem, &len);
1237
1238 if (EIO_MS_ASYNC != MS_SYNC
1239 || EIO_MS_INVALIDATE != MS_INVALIDATE
1240 || EIO_MS_SYNC != MS_SYNC)
1241 {
1242 flags = 0
1243 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1244 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1245 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1246 }
1247
1248 return msync (mem, len, flags);
1249}
1250
1251#endif
1252
1253static int
1254eio__mtouch (eio_req *req)
1255{
1256 void *mem = req->ptr2;
1257 size_t len = req->size;
1258 int flags = req->int1;
1259
1260 eio_page_align (&mem, &len);
1261
1262 {
1263 intptr_t addr = (intptr_t)mem;
1264 intptr_t end = addr + len;
1265 intptr_t page = eio_pagesize ();
1266
1267 if (addr < end)
1268 if (flags & EIO_MT_MODIFY) /* modify */
1269 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1270 else
1271 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1272 }
1273
1274 return 0;
1275}
1276
1277/*****************************************************************************/
1278/* requests implemented outside eio_execute, because they are so large */
1279
1280static void
1281eio__realpath (eio_req *req, etp_worker *self)
1282{
1283 char *rel = req->ptr1;
1284 char *res;
1285 char *tmp1, *tmp2;
1286#if SYMLOOP_MAX > 32
1287 int symlinks = SYMLOOP_MAX;
1288#else
1289 int symlinks = 32;
1290#endif
1291
1292 req->result = -1;
1293
1294 errno = EINVAL;
1295 if (!rel)
1296 return;
1297
1298 errno = ENOENT;
1299 if (!*rel)
1300 return;
1301
1302 if (!req->ptr2)
1303 {
1304 X_LOCK (wrklock);
1305 req->flags |= EIO_FLAG_PTR2_FREE;
1306 X_UNLOCK (wrklock);
1307 req->ptr2 = malloc (PATH_MAX * 3);
1308
1309 errno = ENOMEM;
1310 if (!req->ptr2)
1311 return;
1312 }
1313
1314 res = req->ptr2;
1315 tmp1 = res + PATH_MAX;
1316 tmp2 = tmp1 + PATH_MAX;
1317
1318#if 0 /* disabled, the musl way to do things is just too racy */
1319#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1320 /* on linux we may be able to ask the kernel */
1321 {
1322 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1323
1324 if (fd >= 0)
1325 {
1326 sprintf (tmp1, "/proc/self/fd/%d", fd);
1327 req->result = readlink (tmp1, res, PATH_MAX);
1328 close (fd);
1329
1330 /* here we should probably stat the open file and the disk file, to make sure they still match */
1331
1332 if (req->result > 0)
1333 goto done;
1334 }
1335 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1336 return;
1337 }
1338#endif
1339#endif
1340
1341 if (*rel != '/')
1342 {
1343 if (!getcwd (res, PATH_MAX))
1344 return;
1345
1346 if (res [1]) /* only use if not / */
1347 res += strlen (res);
1348 }
1349
1350 while (*rel)
1351 {
1352 eio_ssize_t len, linklen;
1353 char *beg = rel;
1354
1355 while (*rel && *rel != '/')
1356 ++rel;
1357
1358 len = rel - beg;
1359
1360 if (!len) /* skip slashes */
1361 {
1362 ++rel;
1363 continue;
1364 }
1365
1366 if (beg [0] == '.')
1367 {
1368 if (len == 1)
1369 continue; /* . - nop */
1370
1371 if (beg [1] == '.' && len == 2)
1372 {
1373 /* .. - back up one component, if possible */
1374
1375 while (res != req->ptr2)
1376 if (*--res == '/')
1377 break;
1378
1379 continue;
1380 }
1381 }
1382
1383 errno = ENAMETOOLONG;
1384 if (res + 1 + len + 1 >= tmp1)
1385 return;
1386
1387 /* copy one component */
1388 *res = '/';
1389 memcpy (res + 1, beg, len);
1390
1391 /* zero-terminate, for readlink */
1392 res [len + 1] = 0;
1393
1394 /* now check if it's a symlink */
1395 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1396
1397 if (linklen < 0)
1398 {
1399 if (errno != EINVAL)
1400 return;
1401
1402 /* it's a normal directory. hopefully */
1403 res += len + 1;
1404 }
1405 else
1406 {
1407 /* yay, it was a symlink - build new path in tmp2 */
1408 int rellen = strlen (rel);
1409
1410 errno = ENAMETOOLONG;
1411 if (linklen + 1 + rellen >= PATH_MAX)
1412 return;
1413
1414 errno = ELOOP;
1415 if (!--symlinks)
1416 return;
1417
1418 if (*tmp1 == '/')
1419 res = req->ptr2; /* symlink resolves to an absolute path */
1420
1421 /* we need to be careful, as rel might point into tmp2 already */
1422 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1423 tmp2 [linklen] = '/';
1424 memcpy (tmp2, tmp1, linklen);
1425
1426 rel = tmp2;
1427 }
1428 }
1429
1430 /* special case for the lone root path */
1431 if (res == req->ptr2)
1432 *res++ = '/';
1433
1434 req->result = res - (char *)req->ptr2;
1435
1436done:
1437 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1438}
1439
1126static signed char 1440static signed char
1127eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1441eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1128{ 1442{
1129 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ 1443 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1130 : a->inode < b->inode ? -1 1444 : a->inode < b->inode ? -1
… …
1136 1450
1137#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1451#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1138#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1452#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1139 1453
1140static void 1454static void
1141eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1455eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1142{ 1456{
1143 unsigned char bits [9 + sizeof (ino_t) * 8]; 1457 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1144 unsigned char *bit = bits; 1458 unsigned char *bit = bits;
1145 1459
1146 assert (CHAR_BIT == 8); 1460 assert (CHAR_BIT == 8);
1147 assert (sizeof (eio_dirent) * 8 < 256); 1461 assert (sizeof (eio_dirent) * 8 < 256);
1148 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1462 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
… …
1150 1464
1151 if (size <= EIO_SORT_FAST) 1465 if (size <= EIO_SORT_FAST)
1152 return; 1466 return;
1153 1467
1154 /* first prepare an array of bits to test in our radix sort */ 1468 /* first prepare an array of bits to test in our radix sort */
1155 /* try to take endianness into account, as well as differences in ino_t sizes */ 1469 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1156 /* inode_bits must contain all inodes ORed together */ 1470 /* inode_bits must contain all inodes ORed together */
1157 /* which is used to skip bits that are 0 everywhere, which is very common */ 1471 /* which is used to skip bits that are 0 everywhere, which is very common */
1158 { 1472 {
1159 ino_t endianness; 1473 eio_ino_t endianness;
1160 int i, j; 1474 int i, j;
1161 1475
1162 /* we store the byte offset of byte n into byte n of "endianness" */ 1476 /* we store the byte offset of byte n into byte n of "endianness" */
1163 for (i = 0; i < sizeof (ino_t); ++i) 1477 for (i = 0; i < sizeof (eio_ino_t); ++i)
1164 ((unsigned char *)&endianness)[i] = i; 1478 ((unsigned char *)&endianness)[i] = i;
1165 1479
1166 *bit++ = 0; 1480 *bit++ = 0;
1167 1481
1168 for (i = 0; i < sizeof (ino_t); ++i) 1482 for (i = 0; i < sizeof (eio_ino_t); ++i)
1169 { 1483 {
1170 /* shifting off the byte offsets out of "endianness" */ 1484 /* shifting off the byte offsets out of "endianness" */
1171 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1485 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1172 endianness >>= 8; 1486 endianness >>= 8;
1173 1487
1174 for (j = 0; j < 8; ++j) 1488 for (j = 0; j < 8; ++j)
1175 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1489 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1176 *bit++ = offs + j; 1490 *bit++ = offs + j;
1177 } 1491 }
1178 1492
1179 for (j = 0; j < 8; ++j) 1493 for (j = 0; j < 8; ++j)
1180 if (score_bits & (1 << j)) 1494 if (score_bits & (1 << j))
1181 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1495 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1182 } 1496 }
1183 1497
1184 /* now actually do the sorting (a variant of MSD radix sort) */ 1498 /* now actually do the sorting (a variant of MSD radix sort) */
1185 { 1499 {
1186 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1500 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1187 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1501 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1188 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1502 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1189 int stk_idx = 0; 1503 int stk_idx = 0;
1190 1504
1191 base_stk [stk_idx] = dents; 1505 base_stk [stk_idx] = dents;
1192 end_stk [stk_idx] = dents + size; 1506 end_stk [stk_idx] = dents + size;
1193 bit_stk [stk_idx] = bit - 1; 1507 bit_stk [stk_idx] = bit - 1;
… …
1272 } 1586 }
1273 } 1587 }
1274} 1588}
1275 1589
1276static void 1590static void
1277eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1591eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1278{ 1592{
1279 if (size <= 1) 1593 if (size <= 1)
1280 return; /* our insertion sort relies on size > 0 */ 1594 return; /* our insertion sort relies on size > 0 */
1281 1595
1282 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1596 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
… …
1290 1604
1291/* read a full directory */ 1605/* read a full directory */
1292static void 1606static void
1293eio__scandir (eio_req *req, etp_worker *self) 1607eio__scandir (eio_req *req, etp_worker *self)
1294{ 1608{
1295 DIR *dirp;
1296 EIO_STRUCT_DIRENT *entp;
1297 char *name, *names; 1609 char *name, *names;
1298 int namesalloc = 4096; 1610 int namesalloc = 4096 - sizeof (void *) * 4;
1299 int namesoffs = 0; 1611 int namesoffs = 0;
1300 int flags = req->int1; 1612 int flags = req->int1;
1301 eio_dirent *dents = 0; 1613 eio_dirent *dents = 0;
1302 int dentalloc = 128; 1614 int dentalloc = 128;
1303 int dentoffs = 0; 1615 int dentoffs = 0;
1304 ino_t inode_bits = 0; 1616 eio_ino_t inode_bits = 0;
1617#ifdef _WIN32
1618 HANDLE dirp;
1619 WIN32_FIND_DATA entp;
1620#else
1621 DIR *dirp;
1622 EIO_STRUCT_DIRENT *entp;
1623#endif
1305 1624
1306 req->result = -1; 1625 req->result = -1;
1307 1626
1308 if (!(flags & EIO_READDIR_DENTS)) 1627 if (!(flags & EIO_READDIR_DENTS))
1309 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1628 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1310 1629
1311 X_LOCK (wrklock); 1630#ifdef _WIN32
1312 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1631 {
1632 int len = strlen ((const char *)req->ptr1);
1633 char *path = malloc (MAX_PATH);
1634 const char *fmt;
1635
1636 if (!len)
1637 fmt = "./*";
1638 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
1639 fmt = "%s*";
1640 else
1641 fmt = "%s/*";
1642
1643 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
1644 dirp = FindFirstFile (path, &entp);
1645 free (path);
1646
1647 if (dirp == INVALID_HANDLE_VALUE)
1648 {
1649 dirp = 0;
1650
1651 switch (GetLastError ())
1652 {
1653 case ERROR_FILE_NOT_FOUND:
1654 req->result = 0;
1655 break;
1656
1657 case ERROR_INVALID_NAME:
1658 case ERROR_PATH_NOT_FOUND:
1659 case ERROR_NO_MORE_FILES:
1660 errno = ENOENT;
1661 break;
1662
1663 case ERROR_NOT_ENOUGH_MEMORY:
1664 errno = ENOMEM;
1665 break;
1666
1667 default:
1668 errno = EINVAL;
1669 break;
1670 }
1671 }
1672 }
1673#else
1313 self->dirp = dirp = opendir (req->ptr1); 1674 dirp = opendir (req->ptr1);
1675#endif
1314 1676
1315 if (req->flags & EIO_FLAG_PTR1_FREE) 1677 if (req->flags & EIO_FLAG_PTR1_FREE)
1316 free (req->ptr1); 1678 free (req->ptr1);
1317 1679
1318 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1680 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1319 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1681 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1320 req->ptr2 = names = malloc (namesalloc); 1682 req->ptr2 = names = malloc (namesalloc);
1321 X_UNLOCK (wrklock);
1322 1683
1323 if (dirp && names && (!flags || dents)) 1684 if (dirp && names && (!flags || dents))
1324 for (;;) 1685 for (;;)
1325 { 1686 {
1687 int done;
1688
1689#ifdef _WIN32
1690 done = !dirp;
1691#else
1326 errno = 0; 1692 errno = 0;
1327 entp = readdir (dirp); 1693 entp = readdir (dirp);
1694 done = !entp;
1695#endif
1328 1696
1329 if (!entp) 1697 if (done)
1330 { 1698 {
1699#ifndef _WIN32
1700 int old_errno = errno;
1701 closedir (dirp);
1702 errno = old_errno;
1703
1331 if (errno) 1704 if (errno)
1332 break; 1705 break;
1706#endif
1333 1707
1334 /* sort etc. */ 1708 /* sort etc. */
1335 req->int1 = flags; 1709 req->int1 = flags;
1336 req->result = dentoffs; 1710 req->result = dentoffs;
1337 1711
… …
1366 1740
1367 break; 1741 break;
1368 } 1742 }
1369 1743
1370 /* now add the entry to our list(s) */ 1744 /* now add the entry to our list(s) */
1371 name = entp->d_name; 1745 name = D_NAME (entp);
1372 1746
1373 /* skip . and .. entries */ 1747 /* skip . and .. entries */
1374 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1748 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1375 { 1749 {
1376 int len = D_NAMLEN (entp) + 1; 1750 int len = D_NAMLEN (entp) + 1;
1377 1751
1378 while (ecb_expect_false (namesoffs + len > namesalloc)) 1752 while (ecb_expect_false (namesoffs + len > namesalloc))
1379 { 1753 {
1380 namesalloc *= 2; 1754 namesalloc *= 2;
1381 X_LOCK (wrklock);
1382 req->ptr2 = names = realloc (names, namesalloc); 1755 req->ptr2 = names = realloc (names, namesalloc);
1383 X_UNLOCK (wrklock);
1384 1756
1385 if (!names) 1757 if (!names)
1386 break; 1758 break;
1387 } 1759 }
1388 1760
… …
1393 struct eio_dirent *ent; 1765 struct eio_dirent *ent;
1394 1766
1395 if (ecb_expect_false (dentoffs == dentalloc)) 1767 if (ecb_expect_false (dentoffs == dentalloc))
1396 { 1768 {
1397 dentalloc *= 2; 1769 dentalloc *= 2;
1398 X_LOCK (wrklock);
1399 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1770 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1400 X_UNLOCK (wrklock);
1401 1771
1402 if (!dents) 1772 if (!dents)
1403 break; 1773 break;
1404 } 1774 }
1405 1775
… …
1485 if (EIO_CANCELLED (req)) 1855 if (EIO_CANCELLED (req))
1486 { 1856 {
1487 errno = ECANCELED; 1857 errno = ECANCELED;
1488 break; 1858 break;
1489 } 1859 }
1860
1861#ifdef _WIN32
1862 if (!FindNextFile (dirp, &entp))
1863 {
1864 FindClose (dirp);
1865 dirp = 0;
1866 }
1867#endif
1490 } 1868 }
1491}
1492
1493#ifdef PAGESIZE
1494# define eio_pagesize() PAGESIZE
1495#else
1496static intptr_t
1497eio_pagesize (void)
1498{
1499 static intptr_t page;
1500
1501 if (!page)
1502 page = sysconf (_SC_PAGESIZE);
1503
1504 return page;
1505}
1506#endif
1507
1508static void
1509eio_page_align (void **addr, size_t *length)
1510{
1511 intptr_t mask = eio_pagesize () - 1;
1512
1513 /* round down addr */
1514 intptr_t adj = mask & (intptr_t)*addr;
1515
1516 *addr = (void *)((intptr_t)*addr - adj);
1517 *length += adj;
1518
1519 /* round up length */
1520 *length = (*length + mask) & ~mask;
1521}
1522
1523#if !_POSIX_MEMLOCK
1524# define eio__mlockall(a) ((errno = ENOSYS), -1)
1525#else
1526
1527static int
1528eio__mlockall (int flags)
1529{
1530 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1531 extern int mallopt (int, int);
1532 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1533 #endif
1534
1535 if (EIO_MCL_CURRENT != MCL_CURRENT
1536 || EIO_MCL_FUTURE != MCL_FUTURE)
1537 {
1538 flags = 0
1539 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1540 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1541 }
1542
1543 return mlockall (flags);
1544}
1545#endif
1546
1547#if !_POSIX_MEMLOCK_RANGE
1548# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1549#else
1550
1551static int
1552eio__mlock (void *addr, size_t length)
1553{
1554 eio_page_align (&addr, &length);
1555
1556 return mlock (addr, length);
1557}
1558
1559#endif
1560
1561#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1562# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1563#else
1564
1565static int
1566eio__msync (void *mem, size_t len, int flags)
1567{
1568 eio_page_align (&mem, &len);
1569
1570 if (EIO_MS_ASYNC != MS_SYNC
1571 || EIO_MS_INVALIDATE != MS_INVALIDATE
1572 || EIO_MS_SYNC != MS_SYNC)
1573 {
1574 flags = 0
1575 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1576 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1577 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1578 }
1579
1580 return msync (mem, len, flags);
1581}
1582
1583#endif
1584
1585static int
1586eio__mtouch (eio_req *req)
1587{
1588 void *mem = req->ptr2;
1589 size_t len = req->size;
1590 int flags = req->int1;
1591
1592 eio_page_align (&mem, &len);
1593
1594 {
1595 intptr_t addr = (intptr_t)mem;
1596 intptr_t end = addr + len;
1597 intptr_t page = eio_pagesize ();
1598
1599 if (addr < end)
1600 if (flags & EIO_MT_MODIFY) /* modify */
1601 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1602 else
1603 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1604 }
1605
1606 return 0;
1607} 1869}
1608 1870
1609/*****************************************************************************/ 1871/*****************************************************************************/
1610 1872
1611#define ALLOC(len) \ 1873#define ALLOC(len) \
… …
1627{ 1889{
1628 ETP_REQ *req; 1890 ETP_REQ *req;
1629 struct timespec ts; 1891 struct timespec ts;
1630 etp_worker *self = (etp_worker *)thr_arg; 1892 etp_worker *self = (etp_worker *)thr_arg;
1631 1893
1632 /* try to distribute timeouts somewhat randomly */ 1894 /* try to distribute timeouts somewhat evenly */
1633 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1895 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1634 1896
1635 for (;;) 1897 for (;;)
1636 { 1898 {
1899 ts.tv_sec = 0;
1900
1637 X_LOCK (reqlock); 1901 X_LOCK (reqlock);
1638 1902
1639 for (;;) 1903 for (;;)
1640 { 1904 {
1641 self->req = req = reqq_shift (&req_queue); 1905 self->req = req = reqq_shift (&req_queue);
1642 1906
1643 if (req) 1907 if (req)
1644 break; 1908 break;
1645 1909
1910 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1911 {
1912 X_UNLOCK (reqlock);
1913 X_LOCK (wrklock);
1914 --started;
1915 X_UNLOCK (wrklock);
1916 goto quit;
1917 }
1918
1646 ++idle; 1919 ++idle;
1647 1920
1648 ts.tv_sec = time (0) + idle_timeout; 1921 if (idle <= max_idle)
1649 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1922 /* we are allowed to idle, so do so without any timeout */
1923 X_COND_WAIT (reqwait, reqlock);
1924 else
1650 { 1925 {
1651 if (idle > max_idle) 1926 /* initialise timeout once */
1652 { 1927 if (!ts.tv_sec)
1653 --idle; 1928 ts.tv_sec = time (0) + idle_timeout;
1654 X_UNLOCK (reqlock);
1655 X_LOCK (wrklock);
1656 --started;
1657 X_UNLOCK (wrklock);
1658 goto quit;
1659 }
1660 1929
1661 /* we are allowed to idle, so do so without any timeout */
1662 X_COND_WAIT (reqwait, reqlock); 1930 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1931 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1663 } 1932 }
1664 1933
1665 --idle; 1934 --idle;
1666 } 1935 }
1667 1936
… …
1670 X_UNLOCK (reqlock); 1939 X_UNLOCK (reqlock);
1671 1940
1672 if (req->type < 0) 1941 if (req->type < 0)
1673 goto quit; 1942 goto quit;
1674 1943
1675 if (!EIO_CANCELLED (req))
1676 ETP_EXECUTE (self, req); 1944 ETP_EXECUTE (self, req);
1677 1945
1678 X_LOCK (reslock); 1946 X_LOCK (reslock);
1679 1947
1680 ++npending; 1948 ++npending;
1681 1949
… …
1699/*****************************************************************************/ 1967/*****************************************************************************/
1700 1968
1701int ecb_cold 1969int ecb_cold
1702eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1970eio_init (void (*want_poll)(void), void (*done_poll)(void))
1703{ 1971{
1972#if !HAVE_PREADWRITE
1973 X_MUTEX_CREATE (preadwritelock);
1974#endif
1975
1704 return etp_init (want_poll, done_poll); 1976 return etp_init (want_poll, done_poll);
1705} 1977}
1706 1978
1707ecb_inline void 1979ecb_inline void
1708eio_api_destroy (eio_req *req) 1980eio_api_destroy (eio_req *req)
… …
1735 } 2007 }
1736 2008
1737static void 2009static void
1738eio_execute (etp_worker *self, eio_req *req) 2010eio_execute (etp_worker *self, eio_req *req)
1739{ 2011{
2012 if (ecb_expect_false (EIO_CANCELLED (req)))
2013 {
2014 req->result = -1;
2015 req->errorno = ECANCELED;
2016 return;
2017 }
2018
1740 switch (req->type) 2019 switch (req->type)
1741 { 2020 {
1742 case EIO_READ: ALLOC (req->size); 2021 case EIO_READ: ALLOC (req->size);
1743 req->result = req->offs >= 0 2022 req->result = req->offs >= 0
1744 ? pread (req->int1, req->ptr2, req->size, req->offs) 2023 ? pread (req->int1, req->ptr2, req->size, req->offs)
… …
1778 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2057 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1779 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2058 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1780 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2059 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1781 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break; 2060 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1782 2061
1783 case EIO_REALPATH: req->flags |= EIO_FLAG_PTR2_FREE; 2062 case EIO_REALPATH: eio__realpath (req, self); break;
1784 req->ptr2 = realpath (req->ptr1, 0);
1785 req->result = req->ptr2 ? strlen (req->ptr2) : -1;
1786 break;
1787 2063
1788 case EIO_READLINK: ALLOC (PATH_MAX); 2064 case EIO_READLINK: ALLOC (PATH_MAX);
1789 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break; 2065 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1790 2066
1791 case EIO_SYNC: req->result = 0; sync (); break; 2067 case EIO_SYNC: req->result = 0; sync (); break;
… …
1794 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 2070 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1795 case EIO_MTOUCH: req->result = eio__mtouch (req); break; 2071 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1796 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 2072 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1797 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 2073 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1798 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 2074 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2075 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1799 2076
1800 case EIO_READDIR: eio__scandir (req, self); break; 2077 case EIO_READDIR: eio__scandir (req, self); break;
1801 2078
1802 case EIO_BUSY: 2079 case EIO_BUSY:
1803#ifdef _WIN32 2080#ifdef _WIN32
… …
1903eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2180eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
1904{ 2181{
1905 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2182 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
1906} 2183}
1907 2184
2185eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2186{
2187 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2188}
2189
1908eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2190eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1909{ 2191{
1910 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2192 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1911} 2193}
1912 2194
… …
1953eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2235eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1954{ 2236{
1955 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2237 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1956} 2238}
1957 2239
1958eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2240eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1959{ 2241{
1960 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2242 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1961} 2243}
1962 2244
1963eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2245eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1983eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2265eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1984{ 2266{
1985 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2267 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1986} 2268}
1987 2269
1988eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2270eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1989{ 2271{
1990 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2272 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1991} 2273}
1992 2274
1993eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2275eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
2140} 2422}
2141 2423
2142/*****************************************************************************/ 2424/*****************************************************************************/
2143/* misc garbage */ 2425/* misc garbage */
2144 2426
2145ssize_t 2427eio_ssize_t
2146eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2428eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2147{ 2429{
2148 etp_worker wrk; 2430 etp_worker wrk;
2149 ssize_t ret; 2431 eio_ssize_t ret;
2150 2432
2151 wrk.dbuf = 0; 2433 wrk.dbuf = 0;
2152 2434
2153 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2435 ret = eio__sendfile (ofd, ifd, offset, count, &wrk);
2154 2436

Diff Legend

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