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

Comparing libeio/eio.c (file contents):
Revision 1.72 by root, Fri Jun 10 12:45:20 2011 UTC vs.
Revision 1.99 by root, Tue Jul 26 11:07:08 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
76#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
77# define ENOTSOCK WSAENOTSOCK
78#endif
79
80static void eio_destroy (eio_req *req);
81
70#ifndef EIO_FINISH 82#ifndef EIO_FINISH
71# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 83# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
72#endif 84#endif
73 85
74#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
77 89
78#ifndef EIO_FEED 90#ifndef EIO_FEED
79# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 91# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
80#endif 92#endif
81 93
94#ifndef EIO_FD_TO_WIN32_HANDLE
95# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
96#endif
97#ifndef EIO_WIN32_HANDLE_TO_FD
98# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
99#endif
100
101#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
102
103#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
104
82#ifdef _WIN32 105#ifdef _WIN32
83 106
84 /*doh*/ 107 #undef PAGESIZE
108 #define PAGESIZE 4096 /* GetSystemInfo? */
109
110 #ifdef EIO_STRUCT_STATI64
111 #define stat(path,buf) _stati64 (path,buf)
112 #define fstat(fd,buf) _fstati64 (fd,buf)
113 #endif
114 #define lstat(path,buf) stat (path,buf)
115 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
116 #define mkdir(path,mode) _mkdir (path)
117 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
118
119 #define chmod(path,mode) _chmod (path, mode)
120 #define dup(fd) _dup (fd)
121 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
122
123 #define fchmod(fd,mode) EIO_ENOSYS ()
124 #define chown(path,uid,gid) EIO_ENOSYS ()
125 #define fchown(fd,uid,gid) EIO_ENOSYS ()
126 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
127 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
128 #define mknod(path,mode,dev) EIO_ENOSYS ()
129 #define sync() EIO_ENOSYS ()
130 #define readlink(path,buf,s) EIO_ENOSYS ()
131 #define statvfs(path,buf) EIO_ENOSYS ()
132 #define fstatvfs(fd,buf) EIO_ENOSYS ()
133
134 /* rename() uses MoveFile, which fails to overwrite */
135 #define rename(old,neu) eio__rename (old, neu)
136
137 static int
138 eio__rename (const char *old, const char *neu)
139 {
140 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
141 return 0;
142
143 /* should steal _dosmaperr */
144 switch (GetLastError ())
145 {
146 case ERROR_FILE_NOT_FOUND:
147 case ERROR_PATH_NOT_FOUND:
148 case ERROR_INVALID_DRIVE:
149 case ERROR_NO_MORE_FILES:
150 case ERROR_BAD_NETPATH:
151 case ERROR_BAD_NET_NAME:
152 case ERROR_BAD_PATHNAME:
153 case ERROR_FILENAME_EXCED_RANGE:
154 errno = ENOENT;
155 break;
156
157 default:
158 errno = EACCES;
159 break;
160 }
161
162 return -1;
163 }
164
165 /* we could even stat and see if it exists */
166 static int
167 symlink (const char *old, const char *neu)
168 {
169 #if WINVER >= 0x0600
170 if (CreateSymbolicLink (neu, old, 1))
171 return 0;
172
173 if (CreateSymbolicLink (neu, old, 0))
174 return 0;
175 #endif
176
177 return EIO_ERRNO (ENOENT, -1);
178 }
179
180 /* POSIX API only */
181 #define CreateHardLink(neu,old,flags) 0
182 #define CreateSymbolicLink(neu,old,flags) 0
183
184 struct statvfs
185 {
186 int dummy;
187 };
188
189 #define DT_DIR EIO_DT_DIR
190 #define DT_REG EIO_DT_REG
191 #define D_NAME(entp) entp.cFileName
192 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
193
85#else 194#else
86 195
87# include <sys/time.h> 196 #include <sys/time.h>
88# include <sys/select.h> 197 #include <sys/select.h>
198 #include <sys/statvfs.h>
89# include <unistd.h> 199 #include <unistd.h>
200 #include <signal.h>
201 #include <dirent.h>
202
203 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
204 #include <sys/mman.h>
205 #endif
206
207 #define D_NAME(entp) entp->d_name
208
209 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
210 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
211 #define _DIRENT_HAVE_D_TYPE /* sigh */
212 #define D_INO(de) (de)->d_fileno
213 #define D_NAMLEN(de) (de)->d_namlen
214 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
215 #define D_INO(de) (de)->d_ino
216 #endif
217
218 #ifdef _D_EXACT_NAMLEN
219 #undef D_NAMLEN
220 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
221 #endif
222
223 #ifdef _DIRENT_HAVE_D_TYPE
224 #define D_TYPE(de) (de)->d_type
225 #endif
226
227 #ifndef EIO_STRUCT_DIRENT
228 #define EIO_STRUCT_DIRENT struct dirent
229 #endif
230
231#endif
232
233#if HAVE_UTIMES
90# include <utime.h> 234# include <utime.h>
91# include <signal.h>
92# include <dirent.h>
93
94#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
95# include <sys/mman.h>
96#endif
97
98/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
99# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
100# define _DIRENT_HAVE_D_TYPE /* sigh */
101# define D_INO(de) (de)->d_fileno
102# define D_NAMLEN(de) (de)->d_namlen
103# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
104# define D_INO(de) (de)->d_ino
105# endif 235#endif
106 236
107#ifdef _D_EXACT_NAMLEN 237#if HAVE_SYS_SYSCALL_H
108# undef D_NAMLEN 238# include <sys/syscall.h>
109# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
110#endif
111
112# ifdef _DIRENT_HAVE_D_TYPE
113# define D_TYPE(de) (de)->d_type
114# endif 239#endif
115 240
116# ifndef EIO_STRUCT_DIRENT 241#if HAVE_SYS_PRCTL_H
117# define EIO_STRUCT_DIRENT struct dirent 242# include <sys/prctl.h>
118# endif
119
120#endif 243#endif
121 244
122#if HAVE_SENDFILE 245#if HAVE_SENDFILE
123# if __linux 246# if __linux
124# include <sys/sendfile.h> 247# include <sys/sendfile.h>
139#endif 262#endif
140#ifndef D_INO 263#ifndef D_INO
141# define D_INO(de) 0 264# define D_INO(de) 0
142#endif 265#endif
143#ifndef D_NAMLEN 266#ifndef D_NAMLEN
144# define D_NAMLEN(de) strlen ((de)->d_name) 267# define D_NAMLEN(entp) strlen (D_NAME (entp))
145#endif 268#endif
146 269
147/* used for struct dirent, AIX doesn't provide it */ 270/* used for struct dirent, AIX doesn't provide it */
148#ifndef NAME_MAX 271#ifndef NAME_MAX
149# define NAME_MAX 4096 272# define NAME_MAX 4096
156 279
157/* buffer size for various temporary buffers */ 280/* buffer size for various temporary buffers */
158#define EIO_BUFSIZE 65536 281#define EIO_BUFSIZE 65536
159 282
160#define dBUF \ 283#define dBUF \
161 char *eio_buf; \
162 ETP_WORKER_LOCK (self); \
163 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 284 char *eio_buf = malloc (EIO_BUFSIZE); \
164 ETP_WORKER_UNLOCK (self); \
165 errno = ENOMEM; \ 285 errno = ENOMEM; \
166 if (!eio_buf) \ 286 if (!eio_buf) \
167 return -1; 287 return -1
288
289#define FUBd \
290 free (eio_buf)
168 291
169#define EIO_TICKS ((1000000 + 1023) >> 10) 292#define EIO_TICKS ((1000000 + 1023) >> 10)
170 293
171#define ETP_PRI_MIN EIO_PRI_MIN 294#define ETP_PRI_MIN EIO_PRI_MIN
172#define ETP_PRI_MAX EIO_PRI_MAX 295#define ETP_PRI_MAX EIO_PRI_MAX
178static int eio_finish (eio_req *req); 301static int eio_finish (eio_req *req);
179#define ETP_FINISH(req) eio_finish (req) 302#define ETP_FINISH(req) eio_finish (req)
180static void eio_execute (struct etp_worker *self, eio_req *req); 303static void eio_execute (struct etp_worker *self, eio_req *req);
181#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 304#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
182 305
183#define ETP_WORKER_CLEAR(req) \
184 if (wrk->dbuf) \
185 { \
186 free (wrk->dbuf); \
187 wrk->dbuf = 0; \
188 } \
189 \
190 if (wrk->dirp) \
191 { \
192 closedir (wrk->dirp); \
193 wrk->dirp = 0; \
194 }
195
196#define ETP_WORKER_COMMON \
197 void *dbuf; \
198 DIR *dirp;
199
200/*****************************************************************************/ 306/*****************************************************************************/
201 307
202#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 308#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
203 309
204/* calculate time difference in ~1/EIO_TICKS of a second */ 310/* calculate time difference in ~1/EIO_TICKS of a second */
205ECB_INLINE int 311ecb_inline int
206tvdiff (struct timeval *tv1, struct timeval *tv2) 312tvdiff (struct timeval *tv1, struct timeval *tv2)
207{ 313{
208 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 314 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
209 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 315 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
210} 316}
232/* 338/*
233 * make our pread/pwrite emulation safe against themselves, but not against 339 * make our pread/pwrite emulation safe against themselves, but not against
234 * normal read/write by using a mutex. slows down execution a lot, 340 * normal read/write by using a mutex. slows down execution a lot,
235 * but that's your problem, not mine. 341 * but that's your problem, not mine.
236 */ 342 */
237static xmutex_t preadwritelock = X_MUTEX_INIT; 343static xmutex_t preadwritelock;
238#endif 344#endif
239 345
240typedef struct etp_worker 346typedef struct etp_worker
241{ 347{
242 /* locked by wrklock */ 348 /* locked by wrklock */
245 xthread_t tid; 351 xthread_t tid;
246 352
247 /* locked by reslock, reqlock or wrklock */ 353 /* locked by reslock, reqlock or wrklock */
248 ETP_REQ *req; /* currently processed request */ 354 ETP_REQ *req; /* currently processed request */
249 355
356#ifdef ETP_WORKER_COMMON
250 ETP_WORKER_COMMON 357 ETP_WORKER_COMMON
358#endif
251} etp_worker; 359} etp_worker;
252 360
253static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 361static etp_worker wrk_first; /* NOT etp */
254 362
255#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 363#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
256#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 364#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
257 365
258/* worker threads management */ 366/* worker threads management */
259 367
260static void ecb_cold 368static void ecb_cold
261etp_worker_clear (etp_worker *wrk) 369etp_worker_clear (etp_worker *wrk)
262{ 370{
263 ETP_WORKER_CLEAR (wrk);
264} 371}
265 372
266static void ecb_cold 373static void ecb_cold
267etp_worker_free (etp_worker *wrk) 374etp_worker_free (etp_worker *wrk)
268{ 375{
329} etp_reqq; 436} etp_reqq;
330 437
331static etp_reqq req_queue; 438static etp_reqq req_queue;
332static etp_reqq res_queue; 439static etp_reqq res_queue;
333 440
441static void ecb_noinline ecb_cold
442reqq_init (etp_reqq *q)
443{
444 int pri;
445
446 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
447 q->qs[pri] = q->qe[pri] = 0;
448
449 q->size = 0;
450}
451
334static int ecb_noinline 452static int ecb_noinline
335reqq_push (etp_reqq *q, ETP_REQ *req) 453reqq_push (etp_reqq *q, ETP_REQ *req)
336{ 454{
337 int pri = req->pri; 455 int pri = req->pri;
338 req->next = 0; 456 req->next = 0;
372 } 490 }
373 491
374 abort (); 492 abort ();
375} 493}
376 494
377static void ecb_cold 495static int ecb_cold
378etp_thread_init (void) 496etp_init (void (*want_poll)(void), void (*done_poll)(void))
379{ 497{
380 X_MUTEX_CREATE (wrklock); 498 X_MUTEX_CREATE (wrklock);
381 X_MUTEX_CREATE (reslock); 499 X_MUTEX_CREATE (reslock);
382 X_MUTEX_CREATE (reqlock); 500 X_MUTEX_CREATE (reqlock);
383 X_COND_CREATE (reqwait); 501 X_COND_CREATE (reqwait);
384}
385 502
386static void ecb_cold 503 reqq_init (&req_queue);
387etp_atfork_prepare (void) 504 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}
396 505
397static void ecb_cold 506 wrk_first.next =
398etp_atfork_parent (void) 507 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 }
429 508
430 started = 0; 509 started = 0;
431 idle = 0; 510 idle = 0;
432 nreqs = 0; 511 nreqs = 0;
433 nready = 0; 512 nready = 0;
434 npending = 0; 513 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);
452 514
453 want_poll_cb = want_poll; 515 want_poll_cb = want_poll;
454 done_poll_cb = done_poll; 516 done_poll_cb = done_poll;
455 517
456 return 0; 518 return 0;
583} 645}
584 646
585static void 647static void
586etp_cancel (ETP_REQ *req) 648etp_cancel (ETP_REQ *req)
587{ 649{
588 X_LOCK (wrklock); 650 req->cancelled = 1;
589 req->flags |= EIO_FLAG_CANCELLED;
590 X_UNLOCK (wrklock);
591 651
592 eio_grp_cancel (req); 652 eio_grp_cancel (req);
593} 653}
594 654
595static void 655static void
711 return eio_finish (grp); 771 return eio_finish (grp);
712 else 772 else
713 return 0; 773 return 0;
714} 774}
715 775
716void 776static void
717eio_destroy (eio_req *req) 777eio_destroy (eio_req *req)
718{ 778{
719 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 779 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
720 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 780 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
721 781
739 if (grp->grp_first == req) 799 if (grp->grp_first == req)
740 grp->grp_first = req->grp_next; 800 grp->grp_first = req->grp_next;
741 801
742 res2 = grp_dec (grp); 802 res2 = grp_dec (grp);
743 803
744 if (!res && res2) 804 if (!res)
745 res = res2; 805 res = res2;
746 } 806 }
747 807
748 eio_destroy (req); 808 eio_destroy (req);
749 809
841# undef pread 901# undef pread
842# undef pwrite 902# undef pwrite
843# define pread eio__pread 903# define pread eio__pread
844# define pwrite eio__pwrite 904# define pwrite eio__pwrite
845 905
846static ssize_t 906static eio_ssize_t
847eio__pread (int fd, void *buf, size_t count, off_t offset) 907eio__pread (int fd, void *buf, size_t count, off_t offset)
848{ 908{
849 ssize_t res; 909 eio_ssize_t res;
850 off_t ooffset; 910 off_t ooffset;
851 911
852 X_LOCK (preadwritelock); 912 X_LOCK (preadwritelock);
853 ooffset = lseek (fd, 0, SEEK_CUR); 913 ooffset = lseek (fd, 0, SEEK_CUR);
854 lseek (fd, offset, SEEK_SET); 914 lseek (fd, offset, SEEK_SET);
857 X_UNLOCK (preadwritelock); 917 X_UNLOCK (preadwritelock);
858 918
859 return res; 919 return res;
860} 920}
861 921
862static ssize_t 922static eio_ssize_t
863eio__pwrite (int fd, void *buf, size_t count, off_t offset) 923eio__pwrite (int fd, void *buf, size_t count, off_t offset)
864{ 924{
865 ssize_t res; 925 eio_ssize_t res;
866 off_t ooffset; 926 off_t ooffset;
867 927
868 X_LOCK (preadwritelock); 928 X_LOCK (preadwritelock);
869 ooffset = lseek (fd, 0, SEEK_CUR); 929 ooffset = lseek (fd, 0, SEEK_CUR);
870 lseek (fd, offset, SEEK_SET); 930 lseek (fd, offset, SEEK_SET);
915 975
916#if !HAVE_FDATASYNC 976#if !HAVE_FDATASYNC
917# undef fdatasync 977# undef fdatasync
918# define fdatasync(fd) fsync (fd) 978# define fdatasync(fd) fsync (fd)
919#endif 979#endif
980
981static int
982eio__syncfs (int fd)
983{
984 int res;
985
986#if HAVE_SYS_SYNCFS
987 res = (int)syscall (__NR_syncfs, (int)(fd));
988#else
989 res = -1;
990 errno = ENOSYS;
991#endif
992
993 if (res < 0 && errno == ENOSYS && fd >= 0)
994 sync ();
995
996 return res;
997}
920 998
921/* sync_file_range always needs emulation */ 999/* sync_file_range always needs emulation */
922static int 1000static int
923eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 1001eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
924{ 1002{
944 /* even though we could play tricks with the flags, it's better to always 1022 /* even though we could play tricks with the flags, it's better to always
945 * call fdatasync, as that matches the expectation of its users best */ 1023 * call fdatasync, as that matches the expectation of its users best */
946 return fdatasync (fd); 1024 return fdatasync (fd);
947} 1025}
948 1026
1027static int
1028eio__fallocate (int fd, int mode, off_t offset, size_t len)
1029{
1030#if HAVE_FALLOCATE
1031 return fallocate (fd, mode, offset, len);
1032#else
1033 errno = ENOSYS;
1034 return -1;
1035#endif
1036}
1037
949#if !HAVE_READAHEAD 1038#if !HAVE_READAHEAD
950# undef readahead 1039# undef readahead
951# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1040# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
952 1041
953static ssize_t 1042static eio_ssize_t
954eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 1043eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
955{ 1044{
956 size_t todo = count; 1045 size_t todo = count;
957 dBUF; 1046 dBUF;
958 1047
963 pread (fd, eio_buf, len, offset); 1052 pread (fd, eio_buf, len, offset);
964 offset += len; 1053 offset += len;
965 todo -= len; 1054 todo -= len;
966 } 1055 }
967 1056
1057 FUBd;
1058
968 errno = 0; 1059 errno = 0;
969 return count; 1060 return count;
970} 1061}
971 1062
972#endif 1063#endif
973 1064
974/* sendfile always needs emulation */ 1065/* sendfile always needs emulation */
975static ssize_t 1066static eio_ssize_t
976eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1067eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
977{ 1068{
978 ssize_t written = 0; 1069 eio_ssize_t written = 0;
979 ssize_t res; 1070 eio_ssize_t res;
980 1071
981 if (!count) 1072 if (!count)
982 return 0; 1073 return 0;
983 1074
984 for (;;) 1075 for (;;)
985 { 1076 {
1077#ifdef __APPLE__
1078# undef HAVE_SENDFILE /* broken, as everything on os x */
1079#endif
986#if HAVE_SENDFILE 1080#if HAVE_SENDFILE
987# if __linux 1081# if __linux
988 off_t soffset = offset; 1082 off_t soffset = offset;
989 res = sendfile (ofd, ifd, &soffset, count); 1083 res = sendfile (ofd, ifd, &soffset, count);
990 1084
1033 if (res < 0 && sbytes) 1127 if (res < 0 && sbytes)
1034 res = sbytes; 1128 res = sbytes;
1035 1129
1036# endif 1130# endif
1037 1131
1038#elif defined (_WIN32) 1132#elif defined (_WIN32) && 0
1039 /* does not work, just for documentation of what would need to be done */ 1133 /* does not work, just for documentation of what would need to be done */
1040 /* actually, cannot be done like this, as TransmitFile changes the file offset, */ 1134 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1041 /* libeio guarantees that the file offset does not change, and windows */ 1135 /* libeio guarantees that the file offset does not change, and windows */
1042 /* has no way to get an independent handle to the same file description */ 1136 /* has no way to get an independent handle to the same file description */
1043 HANDLE h = TO_SOCKET (ifd); 1137 HANDLE h = TO_SOCKET (ifd);
1076 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1170 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1077 /* BSDs */ 1171 /* BSDs */
1078#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1172#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1079 || errno == ENOTSUP 1173 || errno == ENOTSUP
1080#endif 1174#endif
1175#ifdef EOPNOTSUPP /* windows */
1081 || errno == EOPNOTSUPP /* BSDs */ 1176 || errno == EOPNOTSUPP /* BSDs */
1177#endif
1082#if __solaris 1178#if __solaris
1083 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1179 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1084#endif 1180#endif
1085 ) 1181 )
1086 ) 1182 )
1090 1186
1091 res = 0; 1187 res = 0;
1092 1188
1093 while (count) 1189 while (count)
1094 { 1190 {
1095 ssize_t cnt; 1191 eio_ssize_t cnt;
1096 1192
1097 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1193 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1098 1194
1099 if (cnt <= 0) 1195 if (cnt <= 0)
1100 { 1196 {
1112 1208
1113 offset += cnt; 1209 offset += cnt;
1114 res += cnt; 1210 res += cnt;
1115 count -= cnt; 1211 count -= cnt;
1116 } 1212 }
1213
1214 FUBd;
1117 } 1215 }
1118 1216
1119 return res; 1217 return res;
1218}
1219
1220#ifdef PAGESIZE
1221# define eio_pagesize() PAGESIZE
1222#else
1223static intptr_t
1224eio_pagesize (void)
1225{
1226 static intptr_t page;
1227
1228 if (!page)
1229 page = sysconf (_SC_PAGESIZE);
1230
1231 return page;
1232}
1233#endif
1234
1235static void
1236eio_page_align (void **addr, size_t *length)
1237{
1238 intptr_t mask = eio_pagesize () - 1;
1239
1240 /* round down addr */
1241 intptr_t adj = mask & (intptr_t)*addr;
1242
1243 *addr = (void *)((intptr_t)*addr - adj);
1244 *length += adj;
1245
1246 /* round up length */
1247 *length = (*length + mask) & ~mask;
1248}
1249
1250#if !_POSIX_MEMLOCK
1251# define eio__mlockall(a) EIO_ENOSYS ()
1252#else
1253
1254static int
1255eio__mlockall (int flags)
1256{
1257 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1258 extern int mallopt (int, int);
1259 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1260 #endif
1261
1262 if (EIO_MCL_CURRENT != MCL_CURRENT
1263 || EIO_MCL_FUTURE != MCL_FUTURE)
1264 {
1265 flags = 0
1266 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1267 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1268 }
1269
1270 return mlockall (flags);
1271}
1272#endif
1273
1274#if !_POSIX_MEMLOCK_RANGE
1275# define eio__mlock(a,b) EIO_ENOSYS ()
1276#else
1277
1278static int
1279eio__mlock (void *addr, size_t length)
1280{
1281 eio_page_align (&addr, &length);
1282
1283 return mlock (addr, length);
1284}
1285
1286#endif
1287
1288#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1289# define eio__msync(a,b,c) EIO_ENOSYS ()
1290#else
1291
1292static int
1293eio__msync (void *mem, size_t len, int flags)
1294{
1295 eio_page_align (&mem, &len);
1296
1297 if (EIO_MS_ASYNC != MS_SYNC
1298 || EIO_MS_INVALIDATE != MS_INVALIDATE
1299 || EIO_MS_SYNC != MS_SYNC)
1300 {
1301 flags = 0
1302 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1303 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1304 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1305 }
1306
1307 return msync (mem, len, flags);
1308}
1309
1310#endif
1311
1312static int
1313eio__mtouch (eio_req *req)
1314{
1315 void *mem = req->ptr2;
1316 size_t len = req->size;
1317 int flags = req->int1;
1318
1319 eio_page_align (&mem, &len);
1320
1321 {
1322 intptr_t addr = (intptr_t)mem;
1323 intptr_t end = addr + len;
1324 intptr_t page = eio_pagesize ();
1325
1326 if (addr < end)
1327 if (flags & EIO_MT_MODIFY) /* modify */
1328 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1329 else
1330 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1331 }
1332
1333 return 0;
1334}
1335
1336/*****************************************************************************/
1337/* requests implemented outside eio_execute, because they are so large */
1338
1339static void
1340eio__realpath (eio_req *req, etp_worker *self)
1341{
1342 char *rel = req->ptr1;
1343 char *res;
1344 char *tmp1, *tmp2;
1345#if SYMLOOP_MAX > 32
1346 int symlinks = SYMLOOP_MAX;
1347#else
1348 int symlinks = 32;
1349#endif
1350
1351 req->result = -1;
1352
1353 errno = EINVAL;
1354 if (!rel)
1355 return;
1356
1357 errno = ENOENT;
1358 if (!*rel)
1359 return;
1360
1361 if (!req->ptr2)
1362 {
1363 X_LOCK (wrklock);
1364 req->flags |= EIO_FLAG_PTR2_FREE;
1365 X_UNLOCK (wrklock);
1366 req->ptr2 = malloc (PATH_MAX * 3);
1367
1368 errno = ENOMEM;
1369 if (!req->ptr2)
1370 return;
1371 }
1372
1373 res = req->ptr2;
1374 tmp1 = res + PATH_MAX;
1375 tmp2 = tmp1 + PATH_MAX;
1376
1377#if 0 /* disabled, the musl way to do things is just too racy */
1378#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1379 /* on linux we may be able to ask the kernel */
1380 {
1381 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1382
1383 if (fd >= 0)
1384 {
1385 sprintf (tmp1, "/proc/self/fd/%d", fd);
1386 req->result = readlink (tmp1, res, PATH_MAX);
1387 close (fd);
1388
1389 /* here we should probably stat the open file and the disk file, to make sure they still match */
1390
1391 if (req->result > 0)
1392 goto done;
1393 }
1394 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1395 return;
1396 }
1397#endif
1398#endif
1399
1400 if (*rel != '/')
1401 {
1402 if (!getcwd (res, PATH_MAX))
1403 return;
1404
1405 if (res [1]) /* only use if not / */
1406 res += strlen (res);
1407 }
1408
1409 while (*rel)
1410 {
1411 eio_ssize_t len, linklen;
1412 char *beg = rel;
1413
1414 while (*rel && *rel != '/')
1415 ++rel;
1416
1417 len = rel - beg;
1418
1419 if (!len) /* skip slashes */
1420 {
1421 ++rel;
1422 continue;
1423 }
1424
1425 if (beg [0] == '.')
1426 {
1427 if (len == 1)
1428 continue; /* . - nop */
1429
1430 if (beg [1] == '.' && len == 2)
1431 {
1432 /* .. - back up one component, if possible */
1433
1434 while (res != req->ptr2)
1435 if (*--res == '/')
1436 break;
1437
1438 continue;
1439 }
1440 }
1441
1442 errno = ENAMETOOLONG;
1443 if (res + 1 + len + 1 >= tmp1)
1444 return;
1445
1446 /* copy one component */
1447 *res = '/';
1448 memcpy (res + 1, beg, len);
1449
1450 /* zero-terminate, for readlink */
1451 res [len + 1] = 0;
1452
1453 /* now check if it's a symlink */
1454 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1455
1456 if (linklen < 0)
1457 {
1458 if (errno != EINVAL)
1459 return;
1460
1461 /* it's a normal directory. hopefully */
1462 res += len + 1;
1463 }
1464 else
1465 {
1466 /* yay, it was a symlink - build new path in tmp2 */
1467 int rellen = strlen (rel);
1468
1469 errno = ENAMETOOLONG;
1470 if (linklen + 1 + rellen >= PATH_MAX)
1471 return;
1472
1473 errno = ELOOP;
1474 if (!--symlinks)
1475 return;
1476
1477 if (*tmp1 == '/')
1478 res = req->ptr2; /* symlink resolves to an absolute path */
1479
1480 /* we need to be careful, as rel might point into tmp2 already */
1481 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1482 tmp2 [linklen] = '/';
1483 memcpy (tmp2, tmp1, linklen);
1484
1485 rel = tmp2;
1486 }
1487 }
1488
1489 /* special case for the lone root path */
1490 if (res == req->ptr2)
1491 *res++ = '/';
1492
1493 req->result = res - (char *)req->ptr2;
1494
1495done:
1496 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1120} 1497}
1121 1498
1122static signed char 1499static signed char
1123eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1500eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1124{ 1501{
1132 1509
1133#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1510#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1134#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1511#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1135 1512
1136static void 1513static void
1137eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1514eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1138{ 1515{
1139 unsigned char bits [9 + sizeof (ino_t) * 8]; 1516 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1140 unsigned char *bit = bits; 1517 unsigned char *bit = bits;
1141 1518
1142 assert (CHAR_BIT == 8); 1519 assert (CHAR_BIT == 8);
1143 assert (sizeof (eio_dirent) * 8 < 256); 1520 assert (sizeof (eio_dirent) * 8 < 256);
1144 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1521 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1146 1523
1147 if (size <= EIO_SORT_FAST) 1524 if (size <= EIO_SORT_FAST)
1148 return; 1525 return;
1149 1526
1150 /* first prepare an array of bits to test in our radix sort */ 1527 /* first prepare an array of bits to test in our radix sort */
1151 /* try to take endianness into account, as well as differences in ino_t sizes */ 1528 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1152 /* inode_bits must contain all inodes ORed together */ 1529 /* inode_bits must contain all inodes ORed together */
1153 /* which is used to skip bits that are 0 everywhere, which is very common */ 1530 /* which is used to skip bits that are 0 everywhere, which is very common */
1154 { 1531 {
1155 ino_t endianness; 1532 eio_ino_t endianness;
1156 int i, j; 1533 int i, j;
1157 1534
1158 /* we store the byte offset of byte n into byte n of "endianness" */ 1535 /* we store the byte offset of byte n into byte n of "endianness" */
1159 for (i = 0; i < sizeof (ino_t); ++i) 1536 for (i = 0; i < sizeof (eio_ino_t); ++i)
1160 ((unsigned char *)&endianness)[i] = i; 1537 ((unsigned char *)&endianness)[i] = i;
1161 1538
1162 *bit++ = 0; 1539 *bit++ = 0;
1163 1540
1164 for (i = 0; i < sizeof (ino_t); ++i) 1541 for (i = 0; i < sizeof (eio_ino_t); ++i)
1165 { 1542 {
1166 /* shifting off the byte offsets out of "endianness" */ 1543 /* shifting off the byte offsets out of "endianness" */
1167 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1544 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1168 endianness >>= 8; 1545 endianness >>= 8;
1169 1546
1170 for (j = 0; j < 8; ++j) 1547 for (j = 0; j < 8; ++j)
1171 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1548 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1172 *bit++ = offs + j; 1549 *bit++ = offs + j;
1173 } 1550 }
1174 1551
1175 for (j = 0; j < 8; ++j) 1552 for (j = 0; j < 8; ++j)
1176 if (score_bits & (1 << j)) 1553 if (score_bits & (1 << j))
1177 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1554 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1178 } 1555 }
1179 1556
1180 /* now actually do the sorting (a variant of MSD radix sort) */ 1557 /* now actually do the sorting (a variant of MSD radix sort) */
1181 { 1558 {
1182 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1559 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1183 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1560 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1184 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1561 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1185 int stk_idx = 0; 1562 int stk_idx = 0;
1186 1563
1187 base_stk [stk_idx] = dents; 1564 base_stk [stk_idx] = dents;
1188 end_stk [stk_idx] = dents + size; 1565 end_stk [stk_idx] = dents + size;
1189 bit_stk [stk_idx] = bit - 1; 1566 bit_stk [stk_idx] = bit - 1;
1268 } 1645 }
1269 } 1646 }
1270} 1647}
1271 1648
1272static void 1649static void
1273eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1650eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1274{ 1651{
1275 if (size <= 1) 1652 if (size <= 1)
1276 return; /* our insertion sort relies on size > 0 */ 1653 return; /* our insertion sort relies on size > 0 */
1277 1654
1278 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1655 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1286 1663
1287/* read a full directory */ 1664/* read a full directory */
1288static void 1665static void
1289eio__scandir (eio_req *req, etp_worker *self) 1666eio__scandir (eio_req *req, etp_worker *self)
1290{ 1667{
1291 DIR *dirp;
1292 EIO_STRUCT_DIRENT *entp;
1293 char *name, *names; 1668 char *name, *names;
1294 int namesalloc = 4096; 1669 int namesalloc = 4096 - sizeof (void *) * 4;
1295 int namesoffs = 0; 1670 int namesoffs = 0;
1296 int flags = req->int1; 1671 int flags = req->int1;
1297 eio_dirent *dents = 0; 1672 eio_dirent *dents = 0;
1298 int dentalloc = 128; 1673 int dentalloc = 128;
1299 int dentoffs = 0; 1674 int dentoffs = 0;
1300 ino_t inode_bits = 0; 1675 eio_ino_t inode_bits = 0;
1676#ifdef _WIN32
1677 HANDLE dirp;
1678 WIN32_FIND_DATA entp;
1679#else
1680 DIR *dirp;
1681 EIO_STRUCT_DIRENT *entp;
1682#endif
1301 1683
1302 req->result = -1; 1684 req->result = -1;
1303 1685
1304 if (!(flags & EIO_READDIR_DENTS)) 1686 if (!(flags & EIO_READDIR_DENTS))
1305 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1687 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1306 1688
1307 X_LOCK (wrklock); 1689#ifdef _WIN32
1308 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1690 {
1691 int len = strlen ((const char *)req->ptr1);
1692 char *path = malloc (MAX_PATH);
1693 const char *fmt;
1694
1695 if (!len)
1696 fmt = "./*";
1697 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
1698 fmt = "%s*";
1699 else
1700 fmt = "%s/*";
1701
1702 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
1703 dirp = FindFirstFile (path, &entp);
1704 free (path);
1705
1706 if (dirp == INVALID_HANDLE_VALUE)
1707 {
1708 dirp = 0;
1709
1710 /* should steal _dosmaperr */
1711 switch (GetLastError ())
1712 {
1713 case ERROR_FILE_NOT_FOUND:
1714 req->result = 0;
1715 break;
1716
1717 case ERROR_INVALID_NAME:
1718 case ERROR_PATH_NOT_FOUND:
1719 case ERROR_NO_MORE_FILES:
1720 errno = ENOENT;
1721 break;
1722
1723 case ERROR_NOT_ENOUGH_MEMORY:
1724 errno = ENOMEM;
1725 break;
1726
1727 default:
1728 errno = EINVAL;
1729 break;
1730 }
1731 }
1732 }
1733#else
1309 self->dirp = dirp = opendir (req->ptr1); 1734 dirp = opendir (req->ptr1);
1735#endif
1736
1737 if (req->flags & EIO_FLAG_PTR1_FREE)
1738 free (req->ptr1);
1310 1739
1311 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1740 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1312 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1741 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1313 req->ptr2 = names = malloc (namesalloc); 1742 req->ptr2 = names = malloc (namesalloc);
1314 X_UNLOCK (wrklock);
1315 1743
1316 if (dirp && names && (!flags || dents)) 1744 if (dirp && names && (!flags || dents))
1317 for (;;) 1745 for (;;)
1318 { 1746 {
1747 int done;
1748
1749#ifdef _WIN32
1750 done = !dirp;
1751#else
1319 errno = 0; 1752 errno = 0;
1320 entp = readdir (dirp); 1753 entp = readdir (dirp);
1754 done = !entp;
1755#endif
1321 1756
1322 if (!entp) 1757 if (done)
1323 { 1758 {
1759#ifndef _WIN32
1760 int old_errno = errno;
1761 closedir (dirp);
1762 errno = old_errno;
1763
1324 if (errno) 1764 if (errno)
1325 break; 1765 break;
1766#endif
1326 1767
1327 /* sort etc. */ 1768 /* sort etc. */
1328 req->int1 = flags; 1769 req->int1 = flags;
1329 req->result = dentoffs; 1770 req->result = dentoffs;
1330 1771
1359 1800
1360 break; 1801 break;
1361 } 1802 }
1362 1803
1363 /* now add the entry to our list(s) */ 1804 /* now add the entry to our list(s) */
1364 name = entp->d_name; 1805 name = D_NAME (entp);
1365 1806
1366 /* skip . and .. entries */ 1807 /* skip . and .. entries */
1367 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1808 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1368 { 1809 {
1369 int len = D_NAMLEN (entp) + 1; 1810 int len = D_NAMLEN (entp) + 1;
1370 1811
1371 while (ecb_expect_false (namesoffs + len > namesalloc)) 1812 while (ecb_expect_false (namesoffs + len > namesalloc))
1372 { 1813 {
1373 namesalloc *= 2; 1814 namesalloc *= 2;
1374 X_LOCK (wrklock);
1375 req->ptr2 = names = realloc (names, namesalloc); 1815 req->ptr2 = names = realloc (names, namesalloc);
1376 X_UNLOCK (wrklock);
1377 1816
1378 if (!names) 1817 if (!names)
1379 break; 1818 break;
1380 } 1819 }
1381 1820
1386 struct eio_dirent *ent; 1825 struct eio_dirent *ent;
1387 1826
1388 if (ecb_expect_false (dentoffs == dentalloc)) 1827 if (ecb_expect_false (dentoffs == dentalloc))
1389 { 1828 {
1390 dentalloc *= 2; 1829 dentalloc *= 2;
1391 X_LOCK (wrklock);
1392 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1830 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1393 X_UNLOCK (wrklock);
1394 1831
1395 if (!dents) 1832 if (!dents)
1396 break; 1833 break;
1397 } 1834 }
1398 1835
1478 if (EIO_CANCELLED (req)) 1915 if (EIO_CANCELLED (req))
1479 { 1916 {
1480 errno = ECANCELED; 1917 errno = ECANCELED;
1481 break; 1918 break;
1482 } 1919 }
1920
1921#ifdef _WIN32
1922 if (!FindNextFile (dirp, &entp))
1923 {
1924 FindClose (dirp);
1925 dirp = 0;
1926 }
1927#endif
1483 } 1928 }
1484}
1485
1486#ifdef PAGESIZE
1487# define eio_pagesize() PAGESIZE
1488#else
1489static intptr_t
1490eio_pagesize (void)
1491{
1492 static intptr_t page;
1493
1494 if (!page)
1495 page = sysconf (_SC_PAGESIZE);
1496
1497 return page;
1498}
1499#endif
1500
1501static void
1502eio_page_align (void **addr, size_t *length)
1503{
1504 intptr_t mask = eio_pagesize () - 1;
1505
1506 /* round down addr */
1507 intptr_t adj = mask & (intptr_t)*addr;
1508
1509 *addr = (void *)((intptr_t)*addr - adj);
1510 *length += adj;
1511
1512 /* round up length */
1513 *length = (*length + mask) & ~mask;
1514}
1515
1516#if !_POSIX_MEMLOCK
1517# define eio__mlockall(a) ((errno = ENOSYS), -1)
1518#else
1519
1520static int
1521eio__mlockall (int flags)
1522{
1523 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1524 extern int mallopt (int, int);
1525 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1526 #endif
1527
1528 if (EIO_MCL_CURRENT != MCL_CURRENT
1529 || EIO_MCL_FUTURE != MCL_FUTURE)
1530 {
1531 flags = 0
1532 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1533 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1534 }
1535
1536 return mlockall (flags);
1537}
1538#endif
1539
1540#if !_POSIX_MEMLOCK_RANGE
1541# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1542#else
1543
1544static int
1545eio__mlock (void *addr, size_t length)
1546{
1547 eio_page_align (&addr, &length);
1548
1549 return mlock (addr, length);
1550}
1551
1552#endif
1553
1554#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1555# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1556#else
1557
1558static int
1559eio__msync (void *mem, size_t len, int flags)
1560{
1561 eio_page_align (&mem, &len);
1562
1563 if (EIO_MS_ASYNC != MS_SYNC
1564 || EIO_MS_INVALIDATE != MS_INVALIDATE
1565 || EIO_MS_SYNC != MS_SYNC)
1566 {
1567 flags = 0
1568 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1569 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1570 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1571 }
1572
1573 return msync (mem, len, flags);
1574}
1575
1576#endif
1577
1578static int
1579eio__mtouch (eio_req *req)
1580{
1581 void *mem = req->ptr2;
1582 size_t len = req->size;
1583 int flags = req->int1;
1584
1585 eio_page_align (&mem, &len);
1586
1587 {
1588 intptr_t addr = (intptr_t)mem;
1589 intptr_t end = addr + len;
1590 intptr_t page = eio_pagesize ();
1591
1592 if (addr < end)
1593 if (flags & EIO_MT_MODIFY) /* modify */
1594 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1595 else
1596 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1597 }
1598
1599 return 0;
1600} 1929}
1601 1930
1602/*****************************************************************************/ 1931/*****************************************************************************/
1603 1932
1604#define ALLOC(len) \ 1933#define ALLOC(len) \
1620{ 1949{
1621 ETP_REQ *req; 1950 ETP_REQ *req;
1622 struct timespec ts; 1951 struct timespec ts;
1623 etp_worker *self = (etp_worker *)thr_arg; 1952 etp_worker *self = (etp_worker *)thr_arg;
1624 1953
1954#if HAVE_PRCTL_SET_NAME
1955 prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
1956#endif
1957
1625 /* try to distribute timeouts somewhat randomly */ 1958 /* try to distribute timeouts somewhat evenly */
1626 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1959 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1627 1960
1628 for (;;) 1961 for (;;)
1629 { 1962 {
1963 ts.tv_sec = 0;
1964
1630 X_LOCK (reqlock); 1965 X_LOCK (reqlock);
1631 1966
1632 for (;;) 1967 for (;;)
1633 { 1968 {
1634 self->req = req = reqq_shift (&req_queue); 1969 self->req = req = reqq_shift (&req_queue);
1635 1970
1636 if (req) 1971 if (req)
1637 break; 1972 break;
1638 1973
1974 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1975 {
1976 X_UNLOCK (reqlock);
1977 X_LOCK (wrklock);
1978 --started;
1979 X_UNLOCK (wrklock);
1980 goto quit;
1981 }
1982
1639 ++idle; 1983 ++idle;
1640 1984
1641 ts.tv_sec = time (0) + idle_timeout; 1985 if (idle <= max_idle)
1642 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1986 /* we are allowed to idle, so do so without any timeout */
1987 X_COND_WAIT (reqwait, reqlock);
1988 else
1643 { 1989 {
1644 if (idle > max_idle) 1990 /* initialise timeout once */
1645 { 1991 if (!ts.tv_sec)
1646 --idle; 1992 ts.tv_sec = time (0) + idle_timeout;
1647 X_UNLOCK (reqlock);
1648 X_LOCK (wrklock);
1649 --started;
1650 X_UNLOCK (wrklock);
1651 goto quit;
1652 }
1653 1993
1654 /* we are allowed to idle, so do so without any timeout */
1655 X_COND_WAIT (reqwait, reqlock); 1994 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1995 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1656 } 1996 }
1657 1997
1658 --idle; 1998 --idle;
1659 } 1999 }
1660 2000
1663 X_UNLOCK (reqlock); 2003 X_UNLOCK (reqlock);
1664 2004
1665 if (req->type < 0) 2005 if (req->type < 0)
1666 goto quit; 2006 goto quit;
1667 2007
1668 if (!EIO_CANCELLED (req))
1669 ETP_EXECUTE (self, req); 2008 ETP_EXECUTE (self, req);
1670 2009
1671 X_LOCK (reslock); 2010 X_LOCK (reslock);
1672 2011
1673 ++npending; 2012 ++npending;
1674 2013
1692/*****************************************************************************/ 2031/*****************************************************************************/
1693 2032
1694int ecb_cold 2033int ecb_cold
1695eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2034eio_init (void (*want_poll)(void), void (*done_poll)(void))
1696{ 2035{
2036#if !HAVE_PREADWRITE
2037 X_MUTEX_CREATE (preadwritelock);
2038#endif
2039
1697 return etp_init (want_poll, done_poll); 2040 return etp_init (want_poll, done_poll);
1698} 2041}
1699 2042
1700ECB_INLINE void 2043ecb_inline void
1701eio_api_destroy (eio_req *req) 2044eio_api_destroy (eio_req *req)
1702{ 2045{
1703 free (req); 2046 free (req);
1704} 2047}
1705 2048
1728 } 2071 }
1729 2072
1730static void 2073static void
1731eio_execute (etp_worker *self, eio_req *req) 2074eio_execute (etp_worker *self, eio_req *req)
1732{ 2075{
2076 if (ecb_expect_false (EIO_CANCELLED (req)))
2077 {
2078 req->result = -1;
2079 req->errorno = ECANCELED;
2080 return;
2081 }
2082
1733 switch (req->type) 2083 switch (req->type)
1734 { 2084 {
1735 case EIO_READ: ALLOC (req->size); 2085 case EIO_READ: ALLOC (req->size);
1736 req->result = req->offs >= 0 2086 req->result = req->offs >= 0
1737 ? pread (req->int1, req->ptr2, req->size, req->offs) 2087 ? pread (req->int1, req->ptr2, req->size, req->offs)
1739 case EIO_WRITE: req->result = req->offs >= 0 2089 case EIO_WRITE: req->result = req->offs >= 0
1740 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2090 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1741 : write (req->int1, req->ptr2, req->size); break; 2091 : write (req->int1, req->ptr2, req->size); break;
1742 2092
1743 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2093 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1744 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2094 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1745 2095
1746 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2096 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1747 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2097 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1748 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2098 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1749 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2099 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1771 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2121 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1772 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2122 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1773 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2123 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1774 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break; 2124 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1775 2125
2126 case EIO_REALPATH: eio__realpath (req, self); break;
2127
1776 case EIO_READLINK: ALLOC (PATH_MAX); 2128 case EIO_READLINK: ALLOC (PATH_MAX);
1777 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break; 2129 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1778 2130
1779 case EIO_SYNC: req->result = 0; sync (); break; 2131 case EIO_SYNC: req->result = 0; sync (); break;
1780 case EIO_FSYNC: req->result = fsync (req->int1); break; 2132 case EIO_FSYNC: req->result = fsync (req->int1); break;
1781 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2133 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2134 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2135 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1782 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 2136 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1783 case EIO_MTOUCH: req->result = eio__mtouch (req); break; 2137 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1784 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 2138 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1785 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 2139 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1786 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 2140 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1787 2141
1788 case EIO_READDIR: eio__scandir (req, self); break; 2142 case EIO_READDIR: eio__scandir (req, self); break;
1789 2143
1790 case EIO_BUSY: 2144 case EIO_BUSY:
1791#ifdef _WIN32 2145#ifdef _WIN32
1871eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2225eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1872{ 2226{
1873 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2227 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1874} 2228}
1875 2229
2230eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2231{
2232 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2233}
2234
2235eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2236{
2237 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2238}
2239
2240eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2241{
2242 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2243}
2244
1876eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2245eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1877{ 2246{
1878 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2247 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1879} 2248}
1880 2249
1886eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) 2255eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1887{ 2256{
1888 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2257 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1889} 2258}
1890 2259
1891eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2260eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
1892{ 2261{
1893 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2262 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1894}
1895
1896eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1897{
1898 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1899} 2263}
1900 2264
1901eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2265eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1902{ 2266{
1903 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2267 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1941eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2305eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1942{ 2306{
1943 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2307 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1944} 2308}
1945 2309
1946eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2310eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1947{ 2311{
1948 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2312 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1949} 2313}
1950 2314
1951eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2315eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1971eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2335eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1972{ 2336{
1973 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2337 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1974} 2338}
1975 2339
1976eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2340eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1977{ 2341{
1978 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2342 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1979} 2343}
1980 2344
1981eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2345eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1995} 2359}
1996 2360
1997eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2361eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1998{ 2362{
1999 return eio__1path (EIO_READLINK, path, pri, cb, data); 2363 return eio__1path (EIO_READLINK, path, pri, cb, data);
2364}
2365
2366eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2367{
2368 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2000} 2369}
2001 2370
2002eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2371eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
2003{ 2372{
2004 return eio__1path (EIO_STAT, path, pri, cb, data); 2373 return eio__1path (EIO_STAT, path, pri, cb, data);
2123} 2492}
2124 2493
2125/*****************************************************************************/ 2494/*****************************************************************************/
2126/* misc garbage */ 2495/* misc garbage */
2127 2496
2128ssize_t 2497eio_ssize_t
2129eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2498eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2130{ 2499{
2131 etp_worker wrk;
2132 ssize_t ret;
2133
2134 wrk.dbuf = 0;
2135
2136 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2500 return eio__sendfile (ofd, ifd, offset, count);
2137
2138 if (wrk.dbuf)
2139 free (wrk.dbuf);
2140
2141 return ret;
2142} 2501}
2143 2502

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines