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 1.102 by root, Sun Aug 21 01:03:09 2011 UTC

43 43
44#include "eio.h" 44#include "eio.h"
45#include "ecb.h" 45#include "ecb.h"
46 46
47#ifdef EIO_STACKSIZE 47#ifdef EIO_STACKSIZE
48# define XTHREAD_STACKSIZE EIO_STACKSIZE 48# define X_STACKSIZE EIO_STACKSIZE
49#endif 49#endif
50#include "xthread.h" 50#include "xthread.h"
51 51
52#include <errno.h> 52#include <errno.h>
53#include <stddef.h> 53#include <stddef.h>
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 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
111
112 #ifdef EIO_STRUCT_STATI64
113 /* look at perl's non-sloppy stat */
114 #define stat(path,buf) _stati64 (path,buf)
115 #define fstat(fd,buf) _fstati64 (fd,buf)
116 #endif
117 #define lstat(path,buf) stat (path,buf)
118 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
119 #define mkdir(path,mode) _mkdir (path)
120 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
121
122 #define chmod(path,mode) _chmod (path, mode)
123 #define dup(fd) _dup (fd)
124 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
125
126 #define fchmod(fd,mode) EIO_ENOSYS ()
127 #define chown(path,uid,gid) EIO_ENOSYS ()
128 #define fchown(fd,uid,gid) EIO_ENOSYS ()
129 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
130 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
131 #define mknod(path,mode,dev) EIO_ENOSYS ()
132 #define sync() EIO_ENOSYS ()
133 #define readlink(path,buf,s) EIO_ENOSYS ()
134 #define statvfs(path,buf) EIO_ENOSYS ()
135 #define fstatvfs(fd,buf) EIO_ENOSYS ()
136
137 /* rename() uses MoveFile, which fails to overwrite */
138 #define rename(old,neu) eio__rename (old, neu)
139
140 static int
141 eio__rename (const char *old, const char *neu)
142 {
143 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
144 return 0;
145
146 /* should steal _dosmaperr */
147 switch (GetLastError ())
148 {
149 case ERROR_FILE_NOT_FOUND:
150 case ERROR_PATH_NOT_FOUND:
151 case ERROR_INVALID_DRIVE:
152 case ERROR_NO_MORE_FILES:
153 case ERROR_BAD_NETPATH:
154 case ERROR_BAD_NET_NAME:
155 case ERROR_BAD_PATHNAME:
156 case ERROR_FILENAME_EXCED_RANGE:
157 errno = ENOENT;
158 break;
159
160 default:
161 errno = EACCES;
162 break;
163 }
164
165 return -1;
166 }
167
168 /* we could even stat and see if it exists */
169 static int
170 symlink (const char *old, const char *neu)
171 {
172 #if WINVER >= 0x0600
173 if (CreateSymbolicLink (neu, old, 1))
174 return 0;
175
176 if (CreateSymbolicLink (neu, old, 0))
177 return 0;
178 #endif
179
180 return EIO_ERRNO (ENOENT, -1);
181 }
182
183 /* POSIX API only */
184 #define CreateHardLink(neu,old,flags) 0
185 #define CreateSymbolicLink(neu,old,flags) 0
186
187 struct statvfs
188 {
189 int dummy;
190 };
191
192 #define DT_DIR EIO_DT_DIR
193 #define DT_REG EIO_DT_REG
194 #define D_NAME(entp) entp.cFileName
195 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
196
85#else 197#else
86 198
87# include <sys/time.h> 199 #include <sys/time.h>
88# include <sys/select.h> 200 #include <sys/select.h>
201 #include <sys/statvfs.h>
89# include <unistd.h> 202 #include <unistd.h>
203 #include <signal.h>
204 #include <dirent.h>
205
206 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
207 #include <sys/mman.h>
208 #endif
209
210 #define D_NAME(entp) entp->d_name
211
212 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
213 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
214 #define _DIRENT_HAVE_D_TYPE /* sigh */
215 #define D_INO(de) (de)->d_fileno
216 #define D_NAMLEN(de) (de)->d_namlen
217 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
218 #define D_INO(de) (de)->d_ino
219 #endif
220
221 #ifdef _D_EXACT_NAMLEN
222 #undef D_NAMLEN
223 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
224 #endif
225
226 #ifdef _DIRENT_HAVE_D_TYPE
227 #define D_TYPE(de) (de)->d_type
228 #endif
229
230 #ifndef EIO_STRUCT_DIRENT
231 #define EIO_STRUCT_DIRENT struct dirent
232 #endif
233
234#endif
235
236#if HAVE_UTIMES
90# include <utime.h> 237# 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 238#endif
106 239
107#ifdef _D_EXACT_NAMLEN 240#if HAVE_SYS_SYSCALL_H
108# undef D_NAMLEN 241# 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 242#endif
115 243
116# ifndef EIO_STRUCT_DIRENT 244#if HAVE_SYS_PRCTL_H
117# define EIO_STRUCT_DIRENT struct dirent 245# include <sys/prctl.h>
118# endif
119
120#endif 246#endif
121 247
122#if HAVE_SENDFILE 248#if HAVE_SENDFILE
123# if __linux 249# if __linux
124# include <sys/sendfile.h> 250# include <sys/sendfile.h>
139#endif 265#endif
140#ifndef D_INO 266#ifndef D_INO
141# define D_INO(de) 0 267# define D_INO(de) 0
142#endif 268#endif
143#ifndef D_NAMLEN 269#ifndef D_NAMLEN
144# define D_NAMLEN(de) strlen ((de)->d_name) 270# define D_NAMLEN(entp) strlen (D_NAME (entp))
145#endif 271#endif
146 272
147/* used for struct dirent, AIX doesn't provide it */ 273/* used for struct dirent, AIX doesn't provide it */
148#ifndef NAME_MAX 274#ifndef NAME_MAX
149# define NAME_MAX 4096 275# define NAME_MAX 4096
156 282
157/* buffer size for various temporary buffers */ 283/* buffer size for various temporary buffers */
158#define EIO_BUFSIZE 65536 284#define EIO_BUFSIZE 65536
159 285
160#define dBUF \ 286#define dBUF \
161 char *eio_buf; \
162 ETP_WORKER_LOCK (self); \
163 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 287 char *eio_buf = malloc (EIO_BUFSIZE); \
164 ETP_WORKER_UNLOCK (self); \
165 errno = ENOMEM; \ 288 errno = ENOMEM; \
166 if (!eio_buf) \ 289 if (!eio_buf) \
167 return -1; 290 return -1
291
292#define FUBd \
293 free (eio_buf)
168 294
169#define EIO_TICKS ((1000000 + 1023) >> 10) 295#define EIO_TICKS ((1000000 + 1023) >> 10)
170 296
171#define ETP_PRI_MIN EIO_PRI_MIN 297#define ETP_PRI_MIN EIO_PRI_MIN
172#define ETP_PRI_MAX EIO_PRI_MAX 298#define ETP_PRI_MAX EIO_PRI_MAX
178static int eio_finish (eio_req *req); 304static int eio_finish (eio_req *req);
179#define ETP_FINISH(req) eio_finish (req) 305#define ETP_FINISH(req) eio_finish (req)
180static void eio_execute (struct etp_worker *self, eio_req *req); 306static void eio_execute (struct etp_worker *self, eio_req *req);
181#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 307#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
182 308
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/*****************************************************************************/ 309/*****************************************************************************/
201 310
202#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 311#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
203 312
204/* calculate time difference in ~1/EIO_TICKS of a second */ 313/* calculate time difference in ~1/EIO_TICKS of a second */
205ECB_INLINE int 314ecb_inline int
206tvdiff (struct timeval *tv1, struct timeval *tv2) 315tvdiff (struct timeval *tv1, struct timeval *tv2)
207{ 316{
208 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 317 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
209 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 318 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
210} 319}
215static void (*done_poll_cb) (void); 324static void (*done_poll_cb) (void);
216 325
217static unsigned int max_poll_time; /* reslock */ 326static unsigned int max_poll_time; /* reslock */
218static unsigned int max_poll_reqs; /* reslock */ 327static unsigned int max_poll_reqs; /* reslock */
219 328
220static volatile unsigned int nreqs; /* reqlock */ 329static unsigned int nreqs; /* reqlock */
221static volatile unsigned int nready; /* reqlock */ 330static unsigned int nready; /* reqlock */
222static volatile unsigned int npending; /* reqlock */ 331static unsigned int npending; /* reqlock */
223static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */ 332static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
224static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */ 333static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
225 334
226static xmutex_t wrklock; 335static xmutex_t wrklock;
227static xmutex_t reslock; 336static xmutex_t reslock;
228static xmutex_t reqlock; 337static xmutex_t reqlock;
229static xcond_t reqwait; 338static xcond_t reqwait;
232/* 341/*
233 * make our pread/pwrite emulation safe against themselves, but not against 342 * make our pread/pwrite emulation safe against themselves, but not against
234 * normal read/write by using a mutex. slows down execution a lot, 343 * normal read/write by using a mutex. slows down execution a lot,
235 * but that's your problem, not mine. 344 * but that's your problem, not mine.
236 */ 345 */
237static xmutex_t preadwritelock = X_MUTEX_INIT; 346static xmutex_t preadwritelock;
238#endif 347#endif
239 348
240typedef struct etp_worker 349typedef struct etp_worker
241{ 350{
242 /* locked by wrklock */ 351 /* locked by wrklock */
245 xthread_t tid; 354 xthread_t tid;
246 355
247 /* locked by reslock, reqlock or wrklock */ 356 /* locked by reslock, reqlock or wrklock */
248 ETP_REQ *req; /* currently processed request */ 357 ETP_REQ *req; /* currently processed request */
249 358
359#ifdef ETP_WORKER_COMMON
250 ETP_WORKER_COMMON 360 ETP_WORKER_COMMON
361#endif
251} etp_worker; 362} etp_worker;
252 363
253static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 364static etp_worker wrk_first; /* NOT etp */
254 365
255#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 366#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
256#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 367#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
257 368
258/* worker threads management */ 369/* worker threads management */
259 370
260static void ecb_cold 371static void ecb_cold
261etp_worker_clear (etp_worker *wrk) 372etp_worker_clear (etp_worker *wrk)
262{ 373{
263 ETP_WORKER_CLEAR (wrk);
264} 374}
265 375
266static void ecb_cold 376static void ecb_cold
267etp_worker_free (etp_worker *wrk) 377etp_worker_free (etp_worker *wrk)
268{ 378{
329} etp_reqq; 439} etp_reqq;
330 440
331static etp_reqq req_queue; 441static etp_reqq req_queue;
332static etp_reqq res_queue; 442static etp_reqq res_queue;
333 443
444static void ecb_noinline ecb_cold
445reqq_init (etp_reqq *q)
446{
447 int pri;
448
449 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
450 q->qs[pri] = q->qe[pri] = 0;
451
452 q->size = 0;
453}
454
334static int ecb_noinline 455static int ecb_noinline
335reqq_push (etp_reqq *q, ETP_REQ *req) 456reqq_push (etp_reqq *q, ETP_REQ *req)
336{ 457{
337 int pri = req->pri; 458 int pri = req->pri;
338 req->next = 0; 459 req->next = 0;
372 } 493 }
373 494
374 abort (); 495 abort ();
375} 496}
376 497
377static void ecb_cold 498static int ecb_cold
378etp_thread_init (void) 499etp_init (void (*want_poll)(void), void (*done_poll)(void))
379{ 500{
380 X_MUTEX_CREATE (wrklock); 501 X_MUTEX_CREATE (wrklock);
381 X_MUTEX_CREATE (reslock); 502 X_MUTEX_CREATE (reslock);
382 X_MUTEX_CREATE (reqlock); 503 X_MUTEX_CREATE (reqlock);
383 X_COND_CREATE (reqwait); 504 X_COND_CREATE (reqwait);
384}
385 505
386static void ecb_cold 506 reqq_init (&req_queue);
387etp_atfork_prepare (void) 507 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}
396 508
397static void ecb_cold 509 wrk_first.next =
398etp_atfork_parent (void) 510 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 }
429 511
430 started = 0; 512 started = 0;
431 idle = 0; 513 idle = 0;
432 nreqs = 0; 514 nreqs = 0;
433 nready = 0; 515 nready = 0;
434 npending = 0; 516 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);
452 517
453 want_poll_cb = want_poll; 518 want_poll_cb = want_poll;
454 done_poll_cb = done_poll; 519 done_poll_cb = done_poll;
455 520
456 return 0; 521 return 0;
583} 648}
584 649
585static void 650static void
586etp_cancel (ETP_REQ *req) 651etp_cancel (ETP_REQ *req)
587{ 652{
588 X_LOCK (wrklock); 653 req->cancelled = 1;
589 req->flags |= EIO_FLAG_CANCELLED;
590 X_UNLOCK (wrklock);
591 654
592 eio_grp_cancel (req); 655 eio_grp_cancel (req);
593} 656}
594 657
595static void 658static void
711 return eio_finish (grp); 774 return eio_finish (grp);
712 else 775 else
713 return 0; 776 return 0;
714} 777}
715 778
716void 779static void
717eio_destroy (eio_req *req) 780eio_destroy (eio_req *req)
718{ 781{
719 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 782 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
720 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 783 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
721 784
739 if (grp->grp_first == req) 802 if (grp->grp_first == req)
740 grp->grp_first = req->grp_next; 803 grp->grp_first = req->grp_next;
741 804
742 res2 = grp_dec (grp); 805 res2 = grp_dec (grp);
743 806
744 if (!res && res2) 807 if (!res)
745 res = res2; 808 res = res2;
746 } 809 }
747 810
748 eio_destroy (req); 811 eio_destroy (req);
749 812
841# undef pread 904# undef pread
842# undef pwrite 905# undef pwrite
843# define pread eio__pread 906# define pread eio__pread
844# define pwrite eio__pwrite 907# define pwrite eio__pwrite
845 908
846static ssize_t 909static eio_ssize_t
847eio__pread (int fd, void *buf, size_t count, off_t offset) 910eio__pread (int fd, void *buf, size_t count, off_t offset)
848{ 911{
849 ssize_t res; 912 eio_ssize_t res;
850 off_t ooffset; 913 off_t ooffset;
851 914
852 X_LOCK (preadwritelock); 915 X_LOCK (preadwritelock);
853 ooffset = lseek (fd, 0, SEEK_CUR); 916 ooffset = lseek (fd, 0, SEEK_CUR);
854 lseek (fd, offset, SEEK_SET); 917 lseek (fd, offset, SEEK_SET);
857 X_UNLOCK (preadwritelock); 920 X_UNLOCK (preadwritelock);
858 921
859 return res; 922 return res;
860} 923}
861 924
862static ssize_t 925static eio_ssize_t
863eio__pwrite (int fd, void *buf, size_t count, off_t offset) 926eio__pwrite (int fd, void *buf, size_t count, off_t offset)
864{ 927{
865 ssize_t res; 928 eio_ssize_t res;
866 off_t ooffset; 929 off_t ooffset;
867 930
868 X_LOCK (preadwritelock); 931 X_LOCK (preadwritelock);
869 ooffset = lseek (fd, 0, SEEK_CUR); 932 ooffset = lseek (fd, 0, SEEK_CUR);
870 lseek (fd, offset, SEEK_SET); 933 lseek (fd, offset, SEEK_SET);
915 978
916#if !HAVE_FDATASYNC 979#if !HAVE_FDATASYNC
917# undef fdatasync 980# undef fdatasync
918# define fdatasync(fd) fsync (fd) 981# define fdatasync(fd) fsync (fd)
919#endif 982#endif
983
984static int
985eio__syncfs (int fd)
986{
987 int res;
988
989#if HAVE_SYS_SYNCFS
990 res = (int)syscall (__NR_syncfs, (int)(fd));
991#else
992 res = -1;
993 errno = ENOSYS;
994#endif
995
996 if (res < 0 && errno == ENOSYS && fd >= 0)
997 sync ();
998
999 return res;
1000}
920 1001
921/* sync_file_range always needs emulation */ 1002/* sync_file_range always needs emulation */
922static int 1003static int
923eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 1004eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
924{ 1005{
944 /* even though we could play tricks with the flags, it's better to always 1025 /* 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 */ 1026 * call fdatasync, as that matches the expectation of its users best */
946 return fdatasync (fd); 1027 return fdatasync (fd);
947} 1028}
948 1029
1030static int
1031eio__fallocate (int fd, int mode, off_t offset, size_t len)
1032{
1033#if HAVE_FALLOCATE
1034 return fallocate (fd, mode, offset, len);
1035#else
1036 errno = ENOSYS;
1037 return -1;
1038#endif
1039}
1040
949#if !HAVE_READAHEAD 1041#if !HAVE_READAHEAD
950# undef readahead 1042# undef readahead
951# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1043# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
952 1044
953static ssize_t 1045static eio_ssize_t
954eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 1046eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
955{ 1047{
956 size_t todo = count; 1048 size_t todo = count;
957 dBUF; 1049 dBUF;
958 1050
963 pread (fd, eio_buf, len, offset); 1055 pread (fd, eio_buf, len, offset);
964 offset += len; 1056 offset += len;
965 todo -= len; 1057 todo -= len;
966 } 1058 }
967 1059
1060 FUBd;
1061
968 errno = 0; 1062 errno = 0;
969 return count; 1063 return count;
970} 1064}
971 1065
972#endif 1066#endif
973 1067
974/* sendfile always needs emulation */ 1068/* sendfile always needs emulation */
975static ssize_t 1069static eio_ssize_t
976eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1070eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
977{ 1071{
978 ssize_t written = 0; 1072 eio_ssize_t written = 0;
979 ssize_t res; 1073 eio_ssize_t res;
980 1074
981 if (!count) 1075 if (!count)
982 return 0; 1076 return 0;
983 1077
984 for (;;) 1078 for (;;)
985 { 1079 {
1080#ifdef __APPLE__
1081# undef HAVE_SENDFILE /* broken, as everything on os x */
1082#endif
986#if HAVE_SENDFILE 1083#if HAVE_SENDFILE
987# if __linux 1084# if __linux
988 off_t soffset = offset; 1085 off_t soffset = offset;
989 res = sendfile (ofd, ifd, &soffset, count); 1086 res = sendfile (ofd, ifd, &soffset, count);
990 1087
1033 if (res < 0 && sbytes) 1130 if (res < 0 && sbytes)
1034 res = sbytes; 1131 res = sbytes;
1035 1132
1036# endif 1133# endif
1037 1134
1038#elif defined (_WIN32) 1135#elif defined (_WIN32) && 0
1039 /* does not work, just for documentation of what would need to be done */ 1136 /* 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, */ 1137 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1041 /* libeio guarantees that the file offset does not change, and windows */ 1138 /* 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 */ 1139 /* has no way to get an independent handle to the same file description */
1043 HANDLE h = TO_SOCKET (ifd); 1140 HANDLE h = TO_SOCKET (ifd);
1076 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1173 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1077 /* BSDs */ 1174 /* BSDs */
1078#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1175#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1079 || errno == ENOTSUP 1176 || errno == ENOTSUP
1080#endif 1177#endif
1178#ifdef EOPNOTSUPP /* windows */
1081 || errno == EOPNOTSUPP /* BSDs */ 1179 || errno == EOPNOTSUPP /* BSDs */
1180#endif
1082#if __solaris 1181#if __solaris
1083 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1182 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1084#endif 1183#endif
1085 ) 1184 )
1086 ) 1185 )
1090 1189
1091 res = 0; 1190 res = 0;
1092 1191
1093 while (count) 1192 while (count)
1094 { 1193 {
1095 ssize_t cnt; 1194 eio_ssize_t cnt;
1096 1195
1097 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1196 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1098 1197
1099 if (cnt <= 0) 1198 if (cnt <= 0)
1100 { 1199 {
1112 1211
1113 offset += cnt; 1212 offset += cnt;
1114 res += cnt; 1213 res += cnt;
1115 count -= cnt; 1214 count -= cnt;
1116 } 1215 }
1216
1217 FUBd;
1117 } 1218 }
1118 1219
1119 return res; 1220 return res;
1221}
1222
1223#ifdef PAGESIZE
1224# define eio_pagesize() PAGESIZE
1225#else
1226static intptr_t
1227eio_pagesize (void)
1228{
1229 static intptr_t page;
1230
1231 if (!page)
1232 page = sysconf (_SC_PAGESIZE);
1233
1234 return page;
1235}
1236#endif
1237
1238static void
1239eio_page_align (void **addr, size_t *length)
1240{
1241 intptr_t mask = eio_pagesize () - 1;
1242
1243 /* round down addr */
1244 intptr_t adj = mask & (intptr_t)*addr;
1245
1246 *addr = (void *)((intptr_t)*addr - adj);
1247 *length += adj;
1248
1249 /* round up length */
1250 *length = (*length + mask) & ~mask;
1251}
1252
1253#if !_POSIX_MEMLOCK
1254# define eio__mlockall(a) EIO_ENOSYS ()
1255#else
1256
1257static int
1258eio__mlockall (int flags)
1259{
1260 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1261 extern int mallopt (int, int);
1262 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1263 #endif
1264
1265 if (EIO_MCL_CURRENT != MCL_CURRENT
1266 || EIO_MCL_FUTURE != MCL_FUTURE)
1267 {
1268 flags = 0
1269 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1270 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1271 }
1272
1273 return mlockall (flags);
1274}
1275#endif
1276
1277#if !_POSIX_MEMLOCK_RANGE
1278# define eio__mlock(a,b) EIO_ENOSYS ()
1279#else
1280
1281static int
1282eio__mlock (void *addr, size_t length)
1283{
1284 eio_page_align (&addr, &length);
1285
1286 return mlock (addr, length);
1287}
1288
1289#endif
1290
1291#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1292# define eio__msync(a,b,c) EIO_ENOSYS ()
1293#else
1294
1295static int
1296eio__msync (void *mem, size_t len, int flags)
1297{
1298 eio_page_align (&mem, &len);
1299
1300 if (EIO_MS_ASYNC != MS_SYNC
1301 || EIO_MS_INVALIDATE != MS_INVALIDATE
1302 || EIO_MS_SYNC != MS_SYNC)
1303 {
1304 flags = 0
1305 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1306 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1307 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1308 }
1309
1310 return msync (mem, len, flags);
1311}
1312
1313#endif
1314
1315static int
1316eio__mtouch (eio_req *req)
1317{
1318 void *mem = req->ptr2;
1319 size_t len = req->size;
1320 int flags = req->int1;
1321
1322 eio_page_align (&mem, &len);
1323
1324 {
1325 intptr_t addr = (intptr_t)mem;
1326 intptr_t end = addr + len;
1327 intptr_t page = eio_pagesize ();
1328
1329 if (addr < end)
1330 if (flags & EIO_MT_MODIFY) /* modify */
1331 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1332 else
1333 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1334 }
1335
1336 return 0;
1337}
1338
1339/*****************************************************************************/
1340/* requests implemented outside eio_execute, because they are so large */
1341
1342static void
1343eio__realpath (eio_req *req, etp_worker *self)
1344{
1345 char *rel = req->ptr1;
1346 char *res;
1347 char *tmp1, *tmp2;
1348#if SYMLOOP_MAX > 32
1349 int symlinks = SYMLOOP_MAX;
1350#else
1351 int symlinks = 32;
1352#endif
1353
1354 req->result = -1;
1355
1356 errno = EINVAL;
1357 if (!rel)
1358 return;
1359
1360 errno = ENOENT;
1361 if (!*rel)
1362 return;
1363
1364 if (!req->ptr2)
1365 {
1366 X_LOCK (wrklock);
1367 req->flags |= EIO_FLAG_PTR2_FREE;
1368 X_UNLOCK (wrklock);
1369 req->ptr2 = malloc (PATH_MAX * 3);
1370
1371 errno = ENOMEM;
1372 if (!req->ptr2)
1373 return;
1374 }
1375
1376 res = req->ptr2;
1377 tmp1 = res + PATH_MAX;
1378 tmp2 = tmp1 + PATH_MAX;
1379
1380#if 0 /* disabled, the musl way to do things is just too racy */
1381#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1382 /* on linux we may be able to ask the kernel */
1383 {
1384 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1385
1386 if (fd >= 0)
1387 {
1388 sprintf (tmp1, "/proc/self/fd/%d", fd);
1389 req->result = readlink (tmp1, res, PATH_MAX);
1390 close (fd);
1391
1392 /* here we should probably stat the open file and the disk file, to make sure they still match */
1393
1394 if (req->result > 0)
1395 goto done;
1396 }
1397 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1398 return;
1399 }
1400#endif
1401#endif
1402
1403 if (*rel != '/')
1404 {
1405 if (!getcwd (res, PATH_MAX))
1406 return;
1407
1408 if (res [1]) /* only use if not / */
1409 res += strlen (res);
1410 }
1411
1412 while (*rel)
1413 {
1414 eio_ssize_t len, linklen;
1415 char *beg = rel;
1416
1417 while (*rel && *rel != '/')
1418 ++rel;
1419
1420 len = rel - beg;
1421
1422 if (!len) /* skip slashes */
1423 {
1424 ++rel;
1425 continue;
1426 }
1427
1428 if (beg [0] == '.')
1429 {
1430 if (len == 1)
1431 continue; /* . - nop */
1432
1433 if (beg [1] == '.' && len == 2)
1434 {
1435 /* .. - back up one component, if possible */
1436
1437 while (res != req->ptr2)
1438 if (*--res == '/')
1439 break;
1440
1441 continue;
1442 }
1443 }
1444
1445 errno = ENAMETOOLONG;
1446 if (res + 1 + len + 1 >= tmp1)
1447 return;
1448
1449 /* copy one component */
1450 *res = '/';
1451 memcpy (res + 1, beg, len);
1452
1453 /* zero-terminate, for readlink */
1454 res [len + 1] = 0;
1455
1456 /* now check if it's a symlink */
1457 linklen = readlink (req->ptr2, tmp1, PATH_MAX);
1458
1459 if (linklen < 0)
1460 {
1461 if (errno != EINVAL)
1462 return;
1463
1464 /* it's a normal directory. hopefully */
1465 res += len + 1;
1466 }
1467 else
1468 {
1469 /* yay, it was a symlink - build new path in tmp2 */
1470 int rellen = strlen (rel);
1471
1472 errno = ENAMETOOLONG;
1473 if (linklen + 1 + rellen >= PATH_MAX)
1474 return;
1475
1476 errno = ELOOP;
1477 if (!--symlinks)
1478 return;
1479
1480 if (*tmp1 == '/')
1481 res = req->ptr2; /* symlink resolves to an absolute path */
1482
1483 /* we need to be careful, as rel might point into tmp2 already */
1484 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1485 tmp2 [linklen] = '/';
1486 memcpy (tmp2, tmp1, linklen);
1487
1488 rel = tmp2;
1489 }
1490 }
1491
1492 /* special case for the lone root path */
1493 if (res == req->ptr2)
1494 *res++ = '/';
1495
1496 req->result = res - (char *)req->ptr2;
1497
1498done:
1499 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1120} 1500}
1121 1501
1122static signed char 1502static signed char
1123eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1503eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1124{ 1504{
1132 1512
1133#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1513#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 */ 1514#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1135 1515
1136static void 1516static void
1137eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1517eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1138{ 1518{
1139 unsigned char bits [9 + sizeof (ino_t) * 8]; 1519 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1140 unsigned char *bit = bits; 1520 unsigned char *bit = bits;
1141 1521
1142 assert (CHAR_BIT == 8); 1522 assert (CHAR_BIT == 8);
1143 assert (sizeof (eio_dirent) * 8 < 256); 1523 assert (sizeof (eio_dirent) * 8 < 256);
1144 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1524 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1146 1526
1147 if (size <= EIO_SORT_FAST) 1527 if (size <= EIO_SORT_FAST)
1148 return; 1528 return;
1149 1529
1150 /* first prepare an array of bits to test in our radix sort */ 1530 /* 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 */ 1531 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1152 /* inode_bits must contain all inodes ORed together */ 1532 /* inode_bits must contain all inodes ORed together */
1153 /* which is used to skip bits that are 0 everywhere, which is very common */ 1533 /* which is used to skip bits that are 0 everywhere, which is very common */
1154 { 1534 {
1155 ino_t endianness; 1535 eio_ino_t endianness;
1156 int i, j; 1536 int i, j;
1157 1537
1158 /* we store the byte offset of byte n into byte n of "endianness" */ 1538 /* we store the byte offset of byte n into byte n of "endianness" */
1159 for (i = 0; i < sizeof (ino_t); ++i) 1539 for (i = 0; i < sizeof (eio_ino_t); ++i)
1160 ((unsigned char *)&endianness)[i] = i; 1540 ((unsigned char *)&endianness)[i] = i;
1161 1541
1162 *bit++ = 0; 1542 *bit++ = 0;
1163 1543
1164 for (i = 0; i < sizeof (ino_t); ++i) 1544 for (i = 0; i < sizeof (eio_ino_t); ++i)
1165 { 1545 {
1166 /* shifting off the byte offsets out of "endianness" */ 1546 /* shifting off the byte offsets out of "endianness" */
1167 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1547 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1168 endianness >>= 8; 1548 endianness >>= 8;
1169 1549
1170 for (j = 0; j < 8; ++j) 1550 for (j = 0; j < 8; ++j)
1171 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1551 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1172 *bit++ = offs + j; 1552 *bit++ = offs + j;
1173 } 1553 }
1174 1554
1175 for (j = 0; j < 8; ++j) 1555 for (j = 0; j < 8; ++j)
1176 if (score_bits & (1 << j)) 1556 if (score_bits & (1 << j))
1177 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1557 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1178 } 1558 }
1179 1559
1180 /* now actually do the sorting (a variant of MSD radix sort) */ 1560 /* now actually do the sorting (a variant of MSD radix sort) */
1181 { 1561 {
1182 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1562 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1183 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1563 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1184 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1564 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1185 int stk_idx = 0; 1565 int stk_idx = 0;
1186 1566
1187 base_stk [stk_idx] = dents; 1567 base_stk [stk_idx] = dents;
1188 end_stk [stk_idx] = dents + size; 1568 end_stk [stk_idx] = dents + size;
1189 bit_stk [stk_idx] = bit - 1; 1569 bit_stk [stk_idx] = bit - 1;
1268 } 1648 }
1269 } 1649 }
1270} 1650}
1271 1651
1272static void 1652static void
1273eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1653eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1274{ 1654{
1275 if (size <= 1) 1655 if (size <= 1)
1276 return; /* our insertion sort relies on size > 0 */ 1656 return; /* our insertion sort relies on size > 0 */
1277 1657
1278 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1658 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1286 1666
1287/* read a full directory */ 1667/* read a full directory */
1288static void 1668static void
1289eio__scandir (eio_req *req, etp_worker *self) 1669eio__scandir (eio_req *req, etp_worker *self)
1290{ 1670{
1291 DIR *dirp;
1292 EIO_STRUCT_DIRENT *entp;
1293 char *name, *names; 1671 char *name, *names;
1294 int namesalloc = 4096; 1672 int namesalloc = 4096 - sizeof (void *) * 4;
1295 int namesoffs = 0; 1673 int namesoffs = 0;
1296 int flags = req->int1; 1674 int flags = req->int1;
1297 eio_dirent *dents = 0; 1675 eio_dirent *dents = 0;
1298 int dentalloc = 128; 1676 int dentalloc = 128;
1299 int dentoffs = 0; 1677 int dentoffs = 0;
1300 ino_t inode_bits = 0; 1678 eio_ino_t inode_bits = 0;
1679#ifdef _WIN32
1680 HANDLE dirp;
1681 WIN32_FIND_DATA entp;
1682#else
1683 DIR *dirp;
1684 EIO_STRUCT_DIRENT *entp;
1685#endif
1301 1686
1302 req->result = -1; 1687 req->result = -1;
1303 1688
1304 if (!(flags & EIO_READDIR_DENTS)) 1689 if (!(flags & EIO_READDIR_DENTS))
1305 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1690 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1306 1691
1307 X_LOCK (wrklock); 1692#ifdef _WIN32
1308 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1693 {
1694 int len = strlen ((const char *)req->ptr1);
1695 char *path = malloc (MAX_PATH);
1696 const char *fmt;
1697
1698 if (!len)
1699 fmt = "./*";
1700 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
1701 fmt = "%s*";
1702 else
1703 fmt = "%s/*";
1704
1705 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
1706 dirp = FindFirstFile (path, &entp);
1707 free (path);
1708
1709 if (dirp == INVALID_HANDLE_VALUE)
1710 {
1711 dirp = 0;
1712
1713 /* should steal _dosmaperr */
1714 switch (GetLastError ())
1715 {
1716 case ERROR_FILE_NOT_FOUND:
1717 req->result = 0;
1718 break;
1719
1720 case ERROR_INVALID_NAME:
1721 case ERROR_PATH_NOT_FOUND:
1722 case ERROR_NO_MORE_FILES:
1723 errno = ENOENT;
1724 break;
1725
1726 case ERROR_NOT_ENOUGH_MEMORY:
1727 errno = ENOMEM;
1728 break;
1729
1730 default:
1731 errno = EINVAL;
1732 break;
1733 }
1734 }
1735 }
1736#else
1309 self->dirp = dirp = opendir (req->ptr1); 1737 dirp = opendir (req->ptr1);
1738#endif
1739
1740 if (req->flags & EIO_FLAG_PTR1_FREE)
1741 free (req->ptr1);
1310 1742
1311 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1743 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1312 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1744 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1313 req->ptr2 = names = malloc (namesalloc); 1745 req->ptr2 = names = malloc (namesalloc);
1314 X_UNLOCK (wrklock);
1315 1746
1316 if (dirp && names && (!flags || dents)) 1747 if (dirp && names && (!flags || dents))
1317 for (;;) 1748 for (;;)
1318 { 1749 {
1750 int done;
1751
1752#ifdef _WIN32
1753 done = !dirp;
1754#else
1319 errno = 0; 1755 errno = 0;
1320 entp = readdir (dirp); 1756 entp = readdir (dirp);
1757 done = !entp;
1758#endif
1321 1759
1322 if (!entp) 1760 if (done)
1323 { 1761 {
1762#ifndef _WIN32
1763 int old_errno = errno;
1764 closedir (dirp);
1765 errno = old_errno;
1766
1324 if (errno) 1767 if (errno)
1325 break; 1768 break;
1769#endif
1326 1770
1327 /* sort etc. */ 1771 /* sort etc. */
1328 req->int1 = flags; 1772 req->int1 = flags;
1329 req->result = dentoffs; 1773 req->result = dentoffs;
1330 1774
1359 1803
1360 break; 1804 break;
1361 } 1805 }
1362 1806
1363 /* now add the entry to our list(s) */ 1807 /* now add the entry to our list(s) */
1364 name = entp->d_name; 1808 name = D_NAME (entp);
1365 1809
1366 /* skip . and .. entries */ 1810 /* skip . and .. entries */
1367 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1811 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1368 { 1812 {
1369 int len = D_NAMLEN (entp) + 1; 1813 int len = D_NAMLEN (entp) + 1;
1370 1814
1371 while (ecb_expect_false (namesoffs + len > namesalloc)) 1815 while (ecb_expect_false (namesoffs + len > namesalloc))
1372 { 1816 {
1373 namesalloc *= 2; 1817 namesalloc *= 2;
1374 X_LOCK (wrklock);
1375 req->ptr2 = names = realloc (names, namesalloc); 1818 req->ptr2 = names = realloc (names, namesalloc);
1376 X_UNLOCK (wrklock);
1377 1819
1378 if (!names) 1820 if (!names)
1379 break; 1821 break;
1380 } 1822 }
1381 1823
1386 struct eio_dirent *ent; 1828 struct eio_dirent *ent;
1387 1829
1388 if (ecb_expect_false (dentoffs == dentalloc)) 1830 if (ecb_expect_false (dentoffs == dentalloc))
1389 { 1831 {
1390 dentalloc *= 2; 1832 dentalloc *= 2;
1391 X_LOCK (wrklock);
1392 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1833 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1393 X_UNLOCK (wrklock);
1394 1834
1395 if (!dents) 1835 if (!dents)
1396 break; 1836 break;
1397 } 1837 }
1398 1838
1478 if (EIO_CANCELLED (req)) 1918 if (EIO_CANCELLED (req))
1479 { 1919 {
1480 errno = ECANCELED; 1920 errno = ECANCELED;
1481 break; 1921 break;
1482 } 1922 }
1923
1924#ifdef _WIN32
1925 if (!FindNextFile (dirp, &entp))
1926 {
1927 FindClose (dirp);
1928 dirp = 0;
1929 }
1930#endif
1483 } 1931 }
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} 1932}
1601 1933
1602/*****************************************************************************/ 1934/*****************************************************************************/
1603 1935
1604#define ALLOC(len) \ 1936#define ALLOC(len) \
1620{ 1952{
1621 ETP_REQ *req; 1953 ETP_REQ *req;
1622 struct timespec ts; 1954 struct timespec ts;
1623 etp_worker *self = (etp_worker *)thr_arg; 1955 etp_worker *self = (etp_worker *)thr_arg;
1624 1956
1957#if HAVE_PRCTL_SET_NAME
1958 prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
1959#endif
1960
1625 /* try to distribute timeouts somewhat randomly */ 1961 /* try to distribute timeouts somewhat evenly */
1626 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1962 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1627 1963
1628 for (;;) 1964 for (;;)
1629 { 1965 {
1966 ts.tv_sec = 0;
1967
1630 X_LOCK (reqlock); 1968 X_LOCK (reqlock);
1631 1969
1632 for (;;) 1970 for (;;)
1633 { 1971 {
1634 self->req = req = reqq_shift (&req_queue); 1972 self->req = req = reqq_shift (&req_queue);
1635 1973
1636 if (req) 1974 if (req)
1637 break; 1975 break;
1638 1976
1977 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1978 {
1979 X_UNLOCK (reqlock);
1980 X_LOCK (wrklock);
1981 --started;
1982 X_UNLOCK (wrklock);
1983 goto quit;
1984 }
1985
1639 ++idle; 1986 ++idle;
1640 1987
1641 ts.tv_sec = time (0) + idle_timeout; 1988 if (idle <= max_idle)
1642 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 1989 /* we are allowed to idle, so do so without any timeout */
1990 X_COND_WAIT (reqwait, reqlock);
1991 else
1643 { 1992 {
1644 if (idle > max_idle) 1993 /* initialise timeout once */
1645 { 1994 if (!ts.tv_sec)
1646 --idle; 1995 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 1996
1654 /* we are allowed to idle, so do so without any timeout */
1655 X_COND_WAIT (reqwait, reqlock); 1997 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1998 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1656 } 1999 }
1657 2000
1658 --idle; 2001 --idle;
1659 } 2002 }
1660 2003
1663 X_UNLOCK (reqlock); 2006 X_UNLOCK (reqlock);
1664 2007
1665 if (req->type < 0) 2008 if (req->type < 0)
1666 goto quit; 2009 goto quit;
1667 2010
1668 if (!EIO_CANCELLED (req))
1669 ETP_EXECUTE (self, req); 2011 ETP_EXECUTE (self, req);
1670 2012
1671 X_LOCK (reslock); 2013 X_LOCK (reslock);
1672 2014
1673 ++npending; 2015 ++npending;
1674 2016
1692/*****************************************************************************/ 2034/*****************************************************************************/
1693 2035
1694int ecb_cold 2036int ecb_cold
1695eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2037eio_init (void (*want_poll)(void), void (*done_poll)(void))
1696{ 2038{
2039#if !HAVE_PREADWRITE
2040 X_MUTEX_CREATE (preadwritelock);
2041#endif
2042
1697 return etp_init (want_poll, done_poll); 2043 return etp_init (want_poll, done_poll);
1698} 2044}
1699 2045
1700ECB_INLINE void 2046ecb_inline void
1701eio_api_destroy (eio_req *req) 2047eio_api_destroy (eio_req *req)
1702{ 2048{
1703 free (req); 2049 free (req);
1704} 2050}
1705 2051
1728 } 2074 }
1729 2075
1730static void 2076static void
1731eio_execute (etp_worker *self, eio_req *req) 2077eio_execute (etp_worker *self, eio_req *req)
1732{ 2078{
2079 if (ecb_expect_false (EIO_CANCELLED (req)))
2080 {
2081 req->result = -1;
2082 req->errorno = ECANCELED;
2083 return;
2084 }
2085
1733 switch (req->type) 2086 switch (req->type)
1734 { 2087 {
1735 case EIO_READ: ALLOC (req->size); 2088 case EIO_READ: ALLOC (req->size);
1736 req->result = req->offs >= 0 2089 req->result = req->offs >= 0
1737 ? pread (req->int1, req->ptr2, req->size, req->offs) 2090 ? pread (req->int1, req->ptr2, req->size, req->offs)
1739 case EIO_WRITE: req->result = req->offs >= 0 2092 case EIO_WRITE: req->result = req->offs >= 0
1740 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2093 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1741 : write (req->int1, req->ptr2, req->size); break; 2094 : write (req->int1, req->ptr2, req->size); break;
1742 2095
1743 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2096 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; 2097 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1745 2098
1746 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2099 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1747 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2100 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1748 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2101 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1749 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2102 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
1771 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; 2124 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1772 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; 2125 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1773 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; 2126 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; 2127 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1775 2128
2129 case EIO_REALPATH: eio__realpath (req, self); break;
2130
1776 case EIO_READLINK: ALLOC (PATH_MAX); 2131 case EIO_READLINK: ALLOC (PATH_MAX);
1777 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break; 2132 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1778 2133
1779 case EIO_SYNC: req->result = 0; sync (); break; 2134 case EIO_SYNC: req->result = 0; sync (); break;
1780 case EIO_FSYNC: req->result = fsync (req->int1); break; 2135 case EIO_FSYNC: req->result = fsync (req->int1); break;
1781 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2136 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2137 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2138 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; 2139 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1783 case EIO_MTOUCH: req->result = eio__mtouch (req); break; 2140 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1784 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 2141 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1785 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 2142 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; 2143 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1787 2144
1788 case EIO_READDIR: eio__scandir (req, self); break; 2145 case EIO_READDIR: eio__scandir (req, self); break;
1789 2146
1790 case EIO_BUSY: 2147 case EIO_BUSY:
1791#ifdef _WIN32 2148#ifdef _WIN32
1871eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2228eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1872{ 2229{
1873 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2230 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1874} 2231}
1875 2232
2233eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2234{
2235 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2236}
2237
2238eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2239{
2240 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2241}
2242
2243eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2244{
2245 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2246}
2247
1876eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2248eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1877{ 2249{
1878 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2250 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1879} 2251}
1880 2252
1886eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) 2258eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1887{ 2259{
1888 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2260 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1889} 2261}
1890 2262
1891eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2263eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
1892{ 2264{
1893 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2265 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} 2266}
1900 2267
1901eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2268eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1902{ 2269{
1903 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2270 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1941eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2308eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1942{ 2309{
1943 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2310 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1944} 2311}
1945 2312
1946eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2313eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1947{ 2314{
1948 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2315 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1949} 2316}
1950 2317
1951eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2318eio_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) 2338eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1972{ 2339{
1973 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2340 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1974} 2341}
1975 2342
1976eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2343eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1977{ 2344{
1978 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2345 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1979} 2346}
1980 2347
1981eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2348eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1995} 2362}
1996 2363
1997eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2364eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1998{ 2365{
1999 return eio__1path (EIO_READLINK, path, pri, cb, data); 2366 return eio__1path (EIO_READLINK, path, pri, cb, data);
2367}
2368
2369eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2370{
2371 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2000} 2372}
2001 2373
2002eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2374eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
2003{ 2375{
2004 return eio__1path (EIO_STAT, path, pri, cb, data); 2376 return eio__1path (EIO_STAT, path, pri, cb, data);
2123} 2495}
2124 2496
2125/*****************************************************************************/ 2497/*****************************************************************************/
2126/* misc garbage */ 2498/* misc garbage */
2127 2499
2128ssize_t 2500eio_ssize_t
2129eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2501eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2130{ 2502{
2131 etp_worker wrk;
2132 ssize_t ret;
2133
2134 wrk.dbuf = 0;
2135
2136 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2503 return eio__sendfile (ofd, ifd, offset, count);
2137
2138 if (wrk.dbuf)
2139 free (wrk.dbuf);
2140
2141 return ret;
2142} 2504}
2143 2505

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines