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

Comparing libeio/eio.c (file contents):
Revision 1.85 by root, Wed Jul 13 00:53:03 2011 UTC vs.
Revision 1.120 by root, Tue Apr 24 18:47:50 2012 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011,2012 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved. 5 * All rights reserved.
6 * 6 *
7 * Redistribution and use in source and binary forms, with or without modifica- 7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met: 8 * tion, are permitted provided that the following conditions are met:
9 * 9 *
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 */
68#endif 67#endif
69 68
70#ifndef ECANCELED 69#ifndef ECANCELED
71# define ECANCELED EDOM 70# define ECANCELED EDOM
72#endif 71#endif
72#ifndef ELOOP
73# define ELOOP EDOM
74#endif
75
76#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
77# define ENOTSOCK WSAENOTSOCK
78#endif
73 79
74static void eio_destroy (eio_req *req); 80static void eio_destroy (eio_req *req);
75 81
76#ifndef EIO_FINISH 82#ifndef EIO_FINISH
77# 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
83 89
84#ifndef EIO_FEED 90#ifndef EIO_FEED
85# 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)
86#endif 92#endif
87 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
88#ifdef _WIN32 105#ifdef _WIN32
89 106
90 /*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
91#else 197#else
92 198
93# include <sys/time.h> 199 #include <sys/time.h>
94# include <sys/select.h> 200 #include <sys/select.h>
201 #include <sys/statvfs.h>
95# 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__ || __NetBSD__ || __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
96# include <utime.h> 237# include <utime.h>
97# include <signal.h>
98# include <dirent.h>
99
100#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
101# include <sys/mman.h>
102#endif
103
104/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
105# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
106# define _DIRENT_HAVE_D_TYPE /* sigh */
107# define D_INO(de) (de)->d_fileno
108# define D_NAMLEN(de) (de)->d_namlen
109# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
110# define D_INO(de) (de)->d_ino
111# endif 238#endif
112 239
113#ifdef _D_EXACT_NAMLEN 240#if HAVE_SYS_SYSCALL_H
114# undef D_NAMLEN 241# include <sys/syscall.h>
115# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
116#endif
117
118# ifdef _DIRENT_HAVE_D_TYPE
119# define D_TYPE(de) (de)->d_type
120# endif 242#endif
121 243
122# ifndef EIO_STRUCT_DIRENT 244#if HAVE_SYS_PRCTL_H
123# define EIO_STRUCT_DIRENT struct dirent 245# include <sys/prctl.h>
124# endif
125
126#endif 246#endif
127 247
128#if HAVE_SENDFILE 248#if HAVE_SENDFILE
129# if __linux 249# if __linux
130# include <sys/sendfile.h> 250# include <sys/sendfile.h>
145#endif 265#endif
146#ifndef D_INO 266#ifndef D_INO
147# define D_INO(de) 0 267# define D_INO(de) 0
148#endif 268#endif
149#ifndef D_NAMLEN 269#ifndef D_NAMLEN
150# define D_NAMLEN(de) strlen ((de)->d_name) 270# define D_NAMLEN(entp) strlen (D_NAME (entp))
151#endif 271#endif
152 272
153/* used for struct dirent, AIX doesn't provide it */ 273/* used for struct dirent, AIX doesn't provide it */
154#ifndef NAME_MAX 274#ifndef NAME_MAX
155# define NAME_MAX 4096 275# define NAME_MAX 4096
162 282
163/* buffer size for various temporary buffers */ 283/* buffer size for various temporary buffers */
164#define EIO_BUFSIZE 65536 284#define EIO_BUFSIZE 65536
165 285
166#define dBUF \ 286#define dBUF \
167 char *eio_buf; \
168 ETP_WORKER_LOCK (self); \
169 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 287 char *eio_buf = malloc (EIO_BUFSIZE); \
170 ETP_WORKER_UNLOCK (self); \
171 errno = ENOMEM; \ 288 errno = ENOMEM; \
172 if (!eio_buf) \ 289 if (!eio_buf) \
173 return -1; 290 return -1
291
292#define FUBd \
293 free (eio_buf)
174 294
175#define EIO_TICKS ((1000000 + 1023) >> 10) 295#define EIO_TICKS ((1000000 + 1023) >> 10)
296
297/*****************************************************************************/
298
299struct tmpbuf
300{
301 void *ptr;
302 int len;
303};
304
305static void *
306tmpbuf_get (struct tmpbuf *buf, int len)
307{
308 if (buf->len < len)
309 {
310 free (buf->ptr);
311 buf->ptr = malloc (buf->len = len);
312 }
313
314 return buf->ptr;
315}
316
317struct tmpbuf;
318
319#if _POSIX_VERSION >= 200809L
320 #define HAVE_AT 1
321 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
322 #ifndef O_SEARCH
323 #define O_SEARCH O_RDONLY
324 #endif
325#else
326 #define HAVE_AT 0
327 static const char *wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path);
328#endif
329
330struct eio_pwd
331{
332#if HAVE_AT
333 int fd;
334#endif
335 int len;
336 char str[1]; /* actually, a 0-terminated canonical path */
337};
338
339/*****************************************************************************/
176 340
177#define ETP_PRI_MIN EIO_PRI_MIN 341#define ETP_PRI_MIN EIO_PRI_MIN
178#define ETP_PRI_MAX EIO_PRI_MAX 342#define ETP_PRI_MAX EIO_PRI_MAX
179 343
180struct etp_worker; 344struct etp_worker;
184static int eio_finish (eio_req *req); 348static int eio_finish (eio_req *req);
185#define ETP_FINISH(req) eio_finish (req) 349#define ETP_FINISH(req) eio_finish (req)
186static void eio_execute (struct etp_worker *self, eio_req *req); 350static void eio_execute (struct etp_worker *self, eio_req *req);
187#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 351#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
188 352
189#define ETP_WORKER_CLEAR(req) \
190 if (wrk->dbuf) \
191 { \
192 free (wrk->dbuf); \
193 wrk->dbuf = 0; \
194 } \
195 \
196 if (wrk->dirp) \
197 { \
198 closedir (wrk->dirp); \
199 wrk->dirp = 0; \
200 }
201
202#define ETP_WORKER_COMMON \
203 void *dbuf; \
204 DIR *dirp;
205
206/*****************************************************************************/ 353/*****************************************************************************/
207 354
208#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 355#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
209 356
210/* calculate time difference in ~1/EIO_TICKS of a second */ 357/* calculate time difference in ~1/EIO_TICKS of a second */
221static void (*done_poll_cb) (void); 368static void (*done_poll_cb) (void);
222 369
223static unsigned int max_poll_time; /* reslock */ 370static unsigned int max_poll_time; /* reslock */
224static unsigned int max_poll_reqs; /* reslock */ 371static unsigned int max_poll_reqs; /* reslock */
225 372
226static volatile unsigned int nreqs; /* reqlock */ 373static unsigned int nreqs; /* reqlock */
227static volatile unsigned int nready; /* reqlock */ 374static unsigned int nready; /* reqlock */
228static volatile unsigned int npending; /* reqlock */ 375static unsigned int npending; /* reqlock */
229static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */ 376static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
230static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */ 377static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
231 378
232static xmutex_t wrklock; 379static xmutex_t wrklock;
233static xmutex_t reslock; 380static xmutex_t reslock;
234static xmutex_t reqlock; 381static xmutex_t reqlock;
235static xcond_t reqwait; 382static xcond_t reqwait;
238/* 385/*
239 * make our pread/pwrite emulation safe against themselves, but not against 386 * make our pread/pwrite emulation safe against themselves, but not against
240 * normal read/write by using a mutex. slows down execution a lot, 387 * normal read/write by using a mutex. slows down execution a lot,
241 * but that's your problem, not mine. 388 * but that's your problem, not mine.
242 */ 389 */
243static xmutex_t preadwritelock = X_MUTEX_INIT; 390static xmutex_t preadwritelock;
244#endif 391#endif
245 392
246typedef struct etp_worker 393typedef struct etp_worker
247{ 394{
395 struct tmpbuf tmpbuf;
396
248 /* locked by wrklock */ 397 /* locked by wrklock */
249 struct etp_worker *prev, *next; 398 struct etp_worker *prev, *next;
250 399
251 xthread_t tid; 400 xthread_t tid;
252 401
253 /* locked by reslock, reqlock or wrklock */ 402#ifdef ETP_WORKER_COMMON
254 ETP_REQ *req; /* currently processed request */
255
256 ETP_WORKER_COMMON 403 ETP_WORKER_COMMON
404#endif
257} etp_worker; 405} etp_worker;
258 406
259static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 407static etp_worker wrk_first; /* NOT etp */
260 408
261#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 409#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
262#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 410#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
263 411
264/* worker threads management */ 412/* worker threads management */
265 413
266static void ecb_cold 414static void
267etp_worker_clear (etp_worker *wrk) 415etp_worker_clear (etp_worker *wrk)
268{ 416{
269 ETP_WORKER_CLEAR (wrk);
270} 417}
271 418
272static void ecb_cold 419static void ecb_cold
273etp_worker_free (etp_worker *wrk) 420etp_worker_free (etp_worker *wrk)
274{ 421{
422 free (wrk->tmpbuf.ptr);
423
275 wrk->next->prev = wrk->prev; 424 wrk->next->prev = wrk->prev;
276 wrk->prev->next = wrk->next; 425 wrk->prev->next = wrk->next;
277 426
278 free (wrk); 427 free (wrk);
279} 428}
335} etp_reqq; 484} etp_reqq;
336 485
337static etp_reqq req_queue; 486static etp_reqq req_queue;
338static etp_reqq res_queue; 487static etp_reqq res_queue;
339 488
489static void ecb_noinline ecb_cold
490reqq_init (etp_reqq *q)
491{
492 int pri;
493
494 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
495 q->qs[pri] = q->qe[pri] = 0;
496
497 q->size = 0;
498}
499
340static int ecb_noinline 500static int ecb_noinline
341reqq_push (etp_reqq *q, ETP_REQ *req) 501reqq_push (etp_reqq *q, ETP_REQ *req)
342{ 502{
343 int pri = req->pri; 503 int pri = req->pri;
344 req->next = 0; 504 req->next = 0;
378 } 538 }
379 539
380 abort (); 540 abort ();
381} 541}
382 542
383static void ecb_cold 543static int ecb_cold
384etp_thread_init (void) 544etp_init (void (*want_poll)(void), void (*done_poll)(void))
385{ 545{
386#if !HAVE_PREADWRITE
387 X_MUTEX_CREATE (preadwritelock);
388#endif
389 X_MUTEX_CREATE (wrklock); 546 X_MUTEX_CREATE (wrklock);
390 X_MUTEX_CREATE (reslock); 547 X_MUTEX_CREATE (reslock);
391 X_MUTEX_CREATE (reqlock); 548 X_MUTEX_CREATE (reqlock);
392 X_COND_CREATE (reqwait); 549 X_COND_CREATE (reqwait);
393}
394 550
395static void ecb_cold 551 reqq_init (&req_queue);
396etp_atfork_prepare (void) 552 reqq_init (&res_queue);
397{
398}
399 553
400static void ecb_cold 554 wrk_first.next =
401etp_atfork_parent (void) 555 wrk_first.prev = &wrk_first;
402{
403}
404
405static void ecb_cold
406etp_atfork_child (void)
407{
408 ETP_REQ *prv;
409
410 while ((prv = reqq_shift (&req_queue)))
411 ETP_DESTROY (prv);
412
413 while ((prv = reqq_shift (&res_queue)))
414 ETP_DESTROY (prv);
415
416 while (wrk_first.next != &wrk_first)
417 {
418 etp_worker *wrk = wrk_first.next;
419
420 if (wrk->req)
421 ETP_DESTROY (wrk->req);
422
423 etp_worker_clear (wrk);
424 etp_worker_free (wrk);
425 }
426 556
427 started = 0; 557 started = 0;
428 idle = 0; 558 idle = 0;
429 nreqs = 0; 559 nreqs = 0;
430 nready = 0; 560 nready = 0;
431 npending = 0; 561 npending = 0;
432
433 etp_thread_init ();
434}
435
436static void ecb_cold
437etp_once_init (void)
438{
439 etp_thread_init ();
440 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
441}
442
443static int ecb_cold
444etp_init (void (*want_poll)(void), void (*done_poll)(void))
445{
446 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
447
448 pthread_once (&doinit, etp_once_init);
449 562
450 want_poll_cb = want_poll; 563 want_poll_cb = want_poll;
451 done_poll_cb = done_poll; 564 done_poll_cb = done_poll;
452 565
453 return 0; 566 return 0;
493} 606}
494 607
495static void ecb_cold 608static void ecb_cold
496etp_end_thread (void) 609etp_end_thread (void)
497{ 610{
498 eio_req *req = calloc (1, sizeof (eio_req)); 611 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
499 612
500 req->type = -1; 613 req->type = -1;
501 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 614 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
502 615
503 X_LOCK (reqlock); 616 X_LOCK (reqlock);
836# undef pread 949# undef pread
837# undef pwrite 950# undef pwrite
838# define pread eio__pread 951# define pread eio__pread
839# define pwrite eio__pwrite 952# define pwrite eio__pwrite
840 953
841static ssize_t 954static eio_ssize_t
842eio__pread (int fd, void *buf, size_t count, off_t offset) 955eio__pread (int fd, void *buf, size_t count, off_t offset)
843{ 956{
844 ssize_t res; 957 eio_ssize_t res;
845 off_t ooffset; 958 off_t ooffset;
846 959
847 X_LOCK (preadwritelock); 960 X_LOCK (preadwritelock);
848 ooffset = lseek (fd, 0, SEEK_CUR); 961 ooffset = lseek (fd, 0, SEEK_CUR);
849 lseek (fd, offset, SEEK_SET); 962 lseek (fd, offset, SEEK_SET);
852 X_UNLOCK (preadwritelock); 965 X_UNLOCK (preadwritelock);
853 966
854 return res; 967 return res;
855} 968}
856 969
857static ssize_t 970static eio_ssize_t
858eio__pwrite (int fd, void *buf, size_t count, off_t offset) 971eio__pwrite (int fd, void *buf, size_t count, off_t offset)
859{ 972{
860 ssize_t res; 973 eio_ssize_t res;
861 off_t ooffset; 974 off_t ooffset;
862 975
863 X_LOCK (preadwritelock); 976 X_LOCK (preadwritelock);
864 ooffset = lseek (fd, 0, SEEK_CUR); 977 ooffset = lseek (fd, 0, SEEK_CUR);
865 lseek (fd, offset, SEEK_SET); 978 lseek (fd, offset, SEEK_SET);
910 1023
911#if !HAVE_FDATASYNC 1024#if !HAVE_FDATASYNC
912# undef fdatasync 1025# undef fdatasync
913# define fdatasync(fd) fsync (fd) 1026# define fdatasync(fd) fsync (fd)
914#endif 1027#endif
1028
1029static int
1030eio__syncfs (int fd)
1031{
1032 int res;
1033
1034#if HAVE_SYS_SYNCFS
1035 res = (int)syscall (__NR_syncfs, (int)(fd));
1036#else
1037 res = EIO_ENOSYS ();
1038#endif
1039
1040 if (res < 0 && errno == ENOSYS && fd >= 0)
1041 sync ();
1042
1043 return res;
1044}
915 1045
916/* sync_file_range always needs emulation */ 1046/* sync_file_range always needs emulation */
917static int 1047static int
918eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 1048eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
919{ 1049{
942} 1072}
943 1073
944static int 1074static int
945eio__fallocate (int fd, int mode, off_t offset, size_t len) 1075eio__fallocate (int fd, int mode, off_t offset, size_t len)
946{ 1076{
947#if HAVE_FALLOCATE 1077#if HAVE_LINUX_FALLOCATE
948 return fallocate (fd, mode, offset, len); 1078 return fallocate (fd, mode, offset, len);
949#else 1079#else
950 errno = ENOSYS; 1080 return EIO_ENOSYS ();
951 return -1;
952#endif 1081#endif
953} 1082}
954 1083
955#if !HAVE_READAHEAD 1084#if !HAVE_READAHEAD
956# undef readahead 1085# undef readahead
957# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1086# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
958 1087
959static ssize_t 1088static eio_ssize_t
960eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 1089eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
961{ 1090{
962 size_t todo = count; 1091 size_t todo = count;
963 dBUF; 1092 dBUF;
964 1093
969 pread (fd, eio_buf, len, offset); 1098 pread (fd, eio_buf, len, offset);
970 offset += len; 1099 offset += len;
971 todo -= len; 1100 todo -= len;
972 } 1101 }
973 1102
974 errno = 0; 1103 FUBd;
1104
1105 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
1106 /* but not for e.g. EIO or eof, so we also never fail */
975 return count; 1107 return 0;
976} 1108}
977 1109
978#endif 1110#endif
979 1111
980/* sendfile always needs emulation */ 1112/* sendfile always needs emulation */
981static ssize_t 1113static eio_ssize_t
982eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1114eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
983{ 1115{
984 ssize_t written = 0; 1116 eio_ssize_t written = 0;
985 ssize_t res; 1117 eio_ssize_t res;
986 1118
987 if (!count) 1119 if (!count)
988 return 0; 1120 return 0;
989 1121
990 for (;;) 1122 for (;;)
1015 1147
1016 /* according to source inspection, this is correct, and useful behaviour */ 1148 /* according to source inspection, this is correct, and useful behaviour */
1017 if (sbytes) 1149 if (sbytes)
1018 res = sbytes; 1150 res = sbytes;
1019 1151
1020# elif defined (__APPLE__) 1152# elif defined __APPLE__
1021 off_t sbytes = count; 1153 off_t sbytes = count;
1022 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 1154 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1023 1155
1024 /* according to the manpage, sbytes is always valid */ 1156 /* according to the manpage, sbytes is always valid */
1025 if (sbytes) 1157 if (sbytes)
1042 if (res < 0 && sbytes) 1174 if (res < 0 && sbytes)
1043 res = sbytes; 1175 res = sbytes;
1044 1176
1045# endif 1177# endif
1046 1178
1047#elif defined (_WIN32) 1179#elif defined (_WIN32) && 0
1048 /* does not work, just for documentation of what would need to be done */ 1180 /* does not work, just for documentation of what would need to be done */
1049 /* actually, cannot be done like this, as TransmitFile changes the file offset, */ 1181 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1050 /* libeio guarantees that the file offset does not change, and windows */ 1182 /* libeio guarantees that the file offset does not change, and windows */
1051 /* has no way to get an independent handle to the same file description */ 1183 /* has no way to get an independent handle to the same file description */
1052 HANDLE h = TO_SOCKET (ifd); 1184 HANDLE h = TO_SOCKET (ifd);
1053 SetFilePointer (h, offset, 0, FILE_BEGIN); 1185 SetFilePointer (h, offset, 0, FILE_BEGIN);
1054 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 1186 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1055 1187
1056#else 1188#else
1057 res = -1; 1189 res = EIO_ENOSYS ();
1058 errno = ENOSYS;
1059#endif 1190#endif
1060 1191
1061 /* we assume sendfile can copy at least 128mb in one go */ 1192 /* we assume sendfile can copy at least 128mb in one go */
1062 if (res <= 128 * 1024 * 1024) 1193 if (res <= 128 * 1024 * 1024)
1063 { 1194 {
1085 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1216 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1086 /* BSDs */ 1217 /* BSDs */
1087#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1218#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1088 || errno == ENOTSUP 1219 || errno == ENOTSUP
1089#endif 1220#endif
1221#ifdef EOPNOTSUPP /* windows */
1090 || errno == EOPNOTSUPP /* BSDs */ 1222 || errno == EOPNOTSUPP /* BSDs */
1223#endif
1091#if __solaris 1224#if __solaris
1092 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1225 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1093#endif 1226#endif
1094 ) 1227 )
1095 ) 1228 )
1099 1232
1100 res = 0; 1233 res = 0;
1101 1234
1102 while (count) 1235 while (count)
1103 { 1236 {
1104 ssize_t cnt; 1237 eio_ssize_t cnt;
1105 1238
1106 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1239 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1107 1240
1108 if (cnt <= 0) 1241 if (cnt <= 0)
1109 { 1242 {
1121 1254
1122 offset += cnt; 1255 offset += cnt;
1123 res += cnt; 1256 res += cnt;
1124 count -= cnt; 1257 count -= cnt;
1125 } 1258 }
1259
1260 FUBd;
1126 } 1261 }
1127 1262
1128 return res; 1263 return res;
1129} 1264}
1130 1265
1157 /* round up length */ 1292 /* round up length */
1158 *length = (*length + mask) & ~mask; 1293 *length = (*length + mask) & ~mask;
1159} 1294}
1160 1295
1161#if !_POSIX_MEMLOCK 1296#if !_POSIX_MEMLOCK
1162# define eio__mlockall(a) ((errno = ENOSYS), -1) 1297# define eio__mlockall(a) EIO_ENOSYS ()
1163#else 1298#else
1164 1299
1165static int 1300static int
1166eio__mlockall (int flags) 1301eio__mlockall (int flags)
1167{ 1302{
1181 return mlockall (flags); 1316 return mlockall (flags);
1182} 1317}
1183#endif 1318#endif
1184 1319
1185#if !_POSIX_MEMLOCK_RANGE 1320#if !_POSIX_MEMLOCK_RANGE
1186# define eio__mlock(a,b) ((errno = ENOSYS), -1) 1321# define eio__mlock(a,b) EIO_ENOSYS ()
1187#else 1322#else
1188 1323
1189static int 1324static int
1190eio__mlock (void *addr, size_t length) 1325eio__mlock (void *addr, size_t length)
1191{ 1326{
1195} 1330}
1196 1331
1197#endif 1332#endif
1198 1333
1199#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1334#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1200# define eio__msync(a,b,c) ((errno = ENOSYS), -1) 1335# define eio__msync(a,b,c) EIO_ENOSYS ()
1201#else 1336#else
1202 1337
1203static int 1338static int
1204eio__msync (void *mem, size_t len, int flags) 1339eio__msync (void *mem, size_t len, int flags)
1205{ 1340{
1246 1381
1247/*****************************************************************************/ 1382/*****************************************************************************/
1248/* requests implemented outside eio_execute, because they are so large */ 1383/* requests implemented outside eio_execute, because they are so large */
1249 1384
1250static void 1385static void
1251eio__realpath (eio_req *req, etp_worker *self) 1386eio__lseek (eio_req *req)
1252{ 1387{
1253 char *rel = req->ptr1; 1388 /* this usually gets optimised away completely, or your compiler sucks, */
1389 /* or the whence constants really are not 0, 1, 2 */
1390 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
1391 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
1392 : req->int2 == EIO_SEEK_END ? SEEK_END
1393 : req->int2;
1394
1395 req->offs = lseek (req->int1, req->offs, whence);
1396 req->result = req->offs == (off_t)-1 ? -1 : 0;
1397}
1398
1399/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1400static int
1401eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1402{
1403 const char *rel = path;
1254 char *res; 1404 char *res;
1255 char *tmp1, *tmp2; 1405 char *tmp1, *tmp2;
1256#if SYMLOOP_MAX > 32 1406#if SYMLOOP_MAX > 32
1257 int symlinks = SYMLOOP_MAX; 1407 int symlinks = SYMLOOP_MAX;
1258#else 1408#else
1259 int symlinks = 32; 1409 int symlinks = 32;
1260#endif 1410#endif
1261 1411
1262 req->result = -1;
1263
1264 errno = EINVAL; 1412 errno = EINVAL;
1265 if (!rel) 1413 if (!rel)
1266 return; 1414 return -1;
1267 1415
1268 errno = ENOENT; 1416 errno = ENOENT;
1269 if (!*rel) 1417 if (!*rel)
1270 return; 1418 return -1;
1271 1419
1272 if (!req->ptr2) 1420 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1273 {
1274 X_LOCK (wrklock);
1275 req->flags |= EIO_FLAG_PTR2_FREE;
1276 X_UNLOCK (wrklock);
1277 req->ptr2 = malloc (PATH_MAX * 3);
1278
1279 errno = ENOMEM;
1280 if (!req->ptr2)
1281 return;
1282 }
1283
1284 res = req->ptr2;
1285 tmp1 = res + PATH_MAX; 1421 tmp1 = res + PATH_MAX;
1286 tmp2 = tmp1 + PATH_MAX; 1422 tmp2 = tmp1 + PATH_MAX;
1287 1423
1288#if 0 /* disabled, the musl way to do things is just too racy */ 1424#if 0 /* disabled, the musl way to do things is just too racy */
1289#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME) 1425#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1308#endif 1444#endif
1309#endif 1445#endif
1310 1446
1311 if (*rel != '/') 1447 if (*rel != '/')
1312 { 1448 {
1449 int len;
1450
1451 errno = ENOENT;
1452 if (wd == EIO_INVALID_WD)
1453 return -1;
1454
1455 if (wd == EIO_CWD)
1456 {
1313 if (!getcwd (res, PATH_MAX)) 1457 if (!getcwd (res, PATH_MAX))
1314 return; 1458 return -1;
1459
1460 len = strlen (res);
1461 }
1462 else
1463 memcpy (res, wd->str, len = wd->len);
1315 1464
1316 if (res [1]) /* only use if not / */ 1465 if (res [1]) /* only use if not / */
1317 res += strlen (res); 1466 res += len;
1318 } 1467 }
1319 1468
1320 while (*rel) 1469 while (*rel)
1321 { 1470 {
1322 ssize_t len, linklen; 1471 eio_ssize_t len, linklen;
1323 char *beg = rel; 1472 const char *beg = rel;
1324 1473
1325 while (*rel && *rel != '/') 1474 while (*rel && *rel != '/')
1326 ++rel; 1475 ++rel;
1327 1476
1328 len = rel - beg; 1477 len = rel - beg;
1340 1489
1341 if (beg [1] == '.' && len == 2) 1490 if (beg [1] == '.' && len == 2)
1342 { 1491 {
1343 /* .. - back up one component, if possible */ 1492 /* .. - back up one component, if possible */
1344 1493
1345 while (res != req->ptr2) 1494 while (res != tmpbuf->ptr)
1346 if (*--res == '/') 1495 if (*--res == '/')
1347 break; 1496 break;
1348 1497
1349 continue; 1498 continue;
1350 } 1499 }
1351 } 1500 }
1352 1501
1353 errno = ENAMETOOLONG; 1502 errno = ENAMETOOLONG;
1354 if (res + 1 + len + 1 >= tmp1) 1503 if (res + 1 + len + 1 >= tmp1)
1355 return; 1504 return -1;
1356 1505
1357 /* copy one component */ 1506 /* copy one component */
1358 *res = '/'; 1507 *res = '/';
1359 memcpy (res + 1, beg, len); 1508 memcpy (res + 1, beg, len);
1360 1509
1361 /* zero-terminate, for readlink */ 1510 /* zero-terminate, for readlink */
1362 res [len + 1] = 0; 1511 res [len + 1] = 0;
1363 1512
1364 /* now check if it's a symlink */ 1513 /* now check if it's a symlink */
1365 linklen = readlink (req->ptr2, tmp1, PATH_MAX); 1514 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1366 1515
1367 if (linklen < 0) 1516 if (linklen < 0)
1368 { 1517 {
1369 if (errno != EINVAL) 1518 if (errno != EINVAL)
1370 return; 1519 return -1;
1371 1520
1372 /* it's a normal directory. hopefully */ 1521 /* it's a normal directory. hopefully */
1373 res += len + 1; 1522 res += len + 1;
1374 } 1523 }
1375 else 1524 else
1377 /* yay, it was a symlink - build new path in tmp2 */ 1526 /* yay, it was a symlink - build new path in tmp2 */
1378 int rellen = strlen (rel); 1527 int rellen = strlen (rel);
1379 1528
1380 errno = ENAMETOOLONG; 1529 errno = ENAMETOOLONG;
1381 if (linklen + 1 + rellen >= PATH_MAX) 1530 if (linklen + 1 + rellen >= PATH_MAX)
1382 return; 1531 return -1;
1383 1532
1384 errno = ELOOP; 1533 errno = ELOOP;
1385 if (!--symlinks) 1534 if (!--symlinks)
1386 return; 1535 return -1;
1387 1536
1388 if (*tmp1 == '/') 1537 if (*tmp1 == '/')
1389 res = req->ptr2; /* symlink resolves to an absolute path */ 1538 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1390 1539
1391 /* we need to be careful, as rel might point into tmp2 already */ 1540 /* we need to be careful, as rel might point into tmp2 already */
1392 memmove (tmp2 + linklen + 1, rel, rellen + 1); 1541 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1393 tmp2 [linklen] = '/'; 1542 tmp2 [linklen] = '/';
1394 memcpy (tmp2, tmp1, linklen); 1543 memcpy (tmp2, tmp1, linklen);
1396 rel = tmp2; 1545 rel = tmp2;
1397 } 1546 }
1398 } 1547 }
1399 1548
1400 /* special case for the lone root path */ 1549 /* special case for the lone root path */
1401 if (res == req->ptr2) 1550 if (res == tmpbuf->ptr)
1402 *res++ = '/'; 1551 *res++ = '/';
1403 1552
1404 req->result = res - (char *)req->ptr2; 1553 return res - (char *)tmpbuf->ptr;
1405
1406done:
1407 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1408} 1554}
1409 1555
1410static signed char 1556static signed char
1411eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1557eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1412{ 1558{
1420 1566
1421#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1567#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1422#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1568#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1423 1569
1424static void 1570static void
1425eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1571eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1426{ 1572{
1427 unsigned char bits [9 + sizeof (ino_t) * 8]; 1573 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1428 unsigned char *bit = bits; 1574 unsigned char *bit = bits;
1429 1575
1430 assert (CHAR_BIT == 8); 1576 assert (CHAR_BIT == 8);
1431 assert (sizeof (eio_dirent) * 8 < 256); 1577 assert (sizeof (eio_dirent) * 8 < 256);
1432 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1578 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1434 1580
1435 if (size <= EIO_SORT_FAST) 1581 if (size <= EIO_SORT_FAST)
1436 return; 1582 return;
1437 1583
1438 /* first prepare an array of bits to test in our radix sort */ 1584 /* first prepare an array of bits to test in our radix sort */
1439 /* try to take endianness into account, as well as differences in ino_t sizes */ 1585 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1440 /* inode_bits must contain all inodes ORed together */ 1586 /* inode_bits must contain all inodes ORed together */
1441 /* which is used to skip bits that are 0 everywhere, which is very common */ 1587 /* which is used to skip bits that are 0 everywhere, which is very common */
1442 { 1588 {
1443 ino_t endianness; 1589 eio_ino_t endianness;
1444 int i, j; 1590 int i, j;
1445 1591
1446 /* we store the byte offset of byte n into byte n of "endianness" */ 1592 /* we store the byte offset of byte n into byte n of "endianness" */
1447 for (i = 0; i < sizeof (ino_t); ++i) 1593 for (i = 0; i < sizeof (eio_ino_t); ++i)
1448 ((unsigned char *)&endianness)[i] = i; 1594 ((unsigned char *)&endianness)[i] = i;
1449 1595
1450 *bit++ = 0; 1596 *bit++ = 0;
1451 1597
1452 for (i = 0; i < sizeof (ino_t); ++i) 1598 for (i = 0; i < sizeof (eio_ino_t); ++i)
1453 { 1599 {
1454 /* shifting off the byte offsets out of "endianness" */ 1600 /* shifting off the byte offsets out of "endianness" */
1455 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1601 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1456 endianness >>= 8; 1602 endianness >>= 8;
1457 1603
1458 for (j = 0; j < 8; ++j) 1604 for (j = 0; j < 8; ++j)
1459 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1605 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1460 *bit++ = offs + j; 1606 *bit++ = offs + j;
1461 } 1607 }
1462 1608
1463 for (j = 0; j < 8; ++j) 1609 for (j = 0; j < 8; ++j)
1464 if (score_bits & (1 << j)) 1610 if (score_bits & (1 << j))
1465 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1611 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1466 } 1612 }
1467 1613
1468 /* now actually do the sorting (a variant of MSD radix sort) */ 1614 /* now actually do the sorting (a variant of MSD radix sort) */
1469 { 1615 {
1470 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1616 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1471 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1617 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1472 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1618 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1473 int stk_idx = 0; 1619 int stk_idx = 0;
1474 1620
1475 base_stk [stk_idx] = dents; 1621 base_stk [stk_idx] = dents;
1476 end_stk [stk_idx] = dents + size; 1622 end_stk [stk_idx] = dents + size;
1477 bit_stk [stk_idx] = bit - 1; 1623 bit_stk [stk_idx] = bit - 1;
1556 } 1702 }
1557 } 1703 }
1558} 1704}
1559 1705
1560static void 1706static void
1561eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1707eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1562{ 1708{
1563 if (size <= 1) 1709 if (size <= 1)
1564 return; /* our insertion sort relies on size > 0 */ 1710 return; /* our insertion sort relies on size > 0 */
1565 1711
1566 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1712 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1574 1720
1575/* read a full directory */ 1721/* read a full directory */
1576static void 1722static void
1577eio__scandir (eio_req *req, etp_worker *self) 1723eio__scandir (eio_req *req, etp_worker *self)
1578{ 1724{
1579 DIR *dirp;
1580 EIO_STRUCT_DIRENT *entp;
1581 char *name, *names; 1725 char *name, *names;
1582 int namesalloc = 4096; 1726 int namesalloc = 4096 - sizeof (void *) * 4;
1583 int namesoffs = 0; 1727 int namesoffs = 0;
1584 int flags = req->int1; 1728 int flags = req->int1;
1585 eio_dirent *dents = 0; 1729 eio_dirent *dents = 0;
1586 int dentalloc = 128; 1730 int dentalloc = 128;
1587 int dentoffs = 0; 1731 int dentoffs = 0;
1588 ino_t inode_bits = 0; 1732 eio_ino_t inode_bits = 0;
1733#ifdef _WIN32
1734 HANDLE dirp;
1735 WIN32_FIND_DATA entp;
1736#else
1737 DIR *dirp;
1738 EIO_STRUCT_DIRENT *entp;
1739#endif
1589 1740
1590 req->result = -1; 1741 req->result = -1;
1591 1742
1592 if (!(flags & EIO_READDIR_DENTS)) 1743 if (!(flags & EIO_READDIR_DENTS))
1593 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1744 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1594 1745
1595 X_LOCK (wrklock); 1746#ifdef _WIN32
1596 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1747 {
1748 int len = strlen ((const char *)req->ptr1);
1749 char *path = malloc (MAX_PATH);
1750 const char *fmt;
1751 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1752
1753 if (!len)
1754 fmt = "./*";
1755 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1756 fmt = "%s*";
1757 else
1758 fmt = "%s/*";
1759
1760 _snprintf (path, MAX_PATH, fmt, reqpath);
1761 dirp = FindFirstFile (path, &entp);
1762 free (path);
1763
1764 if (dirp == INVALID_HANDLE_VALUE)
1765 {
1766 /* should steal _dosmaperr */
1767 switch (GetLastError ())
1768 {
1769 case ERROR_FILE_NOT_FOUND:
1770 req->result = 0;
1771 break;
1772
1773 case ERROR_INVALID_NAME:
1774 case ERROR_PATH_NOT_FOUND:
1775 case ERROR_NO_MORE_FILES:
1776 errno = ENOENT;
1777 break;
1778
1779 case ERROR_NOT_ENOUGH_MEMORY:
1780 errno = ENOMEM;
1781 break;
1782
1783 default:
1784 errno = EINVAL;
1785 break;
1786 }
1787
1788 return;
1789 }
1790 }
1791#else
1792 #if HAVE_AT
1793 if (req->wd)
1794 {
1795 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1796
1797 if (fd < 0)
1798 return;
1799
1800 dirp = fdopendir (fd);
1801
1802 if (!dirp)
1803 close (fd);
1804 }
1805 else
1597 self->dirp = dirp = opendir (req->ptr1); 1806 dirp = opendir (req->ptr1);
1807 #else
1808 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1809 #endif
1810
1811 if (!dirp)
1812 return;
1813#endif
1598 1814
1599 if (req->flags & EIO_FLAG_PTR1_FREE) 1815 if (req->flags & EIO_FLAG_PTR1_FREE)
1600 free (req->ptr1); 1816 free (req->ptr1);
1601 1817
1602 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1818 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1603 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1819 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1604 req->ptr2 = names = malloc (namesalloc); 1820 req->ptr2 = names = malloc (namesalloc);
1605 X_UNLOCK (wrklock);
1606 1821
1607 if (dirp && names && (!flags || dents)) 1822 if (!names || (flags && !dents))
1823 return;
1824
1608 for (;;) 1825 for (;;)
1609 { 1826 {
1827 int done;
1828
1829#ifdef _WIN32
1830 done = !dirp;
1831#else
1610 errno = 0; 1832 errno = 0;
1611 entp = readdir (dirp); 1833 entp = readdir (dirp);
1834 done = !entp;
1835#endif
1612 1836
1613 if (!entp) 1837 if (done)
1614 { 1838 {
1839#ifndef _WIN32
1840 int old_errno = errno;
1841 closedir (dirp);
1842 errno = old_errno;
1843
1615 if (errno) 1844 if (errno)
1616 break; 1845 break;
1846#endif
1617 1847
1618 /* sort etc. */ 1848 /* sort etc. */
1619 req->int1 = flags; 1849 req->int1 = flags;
1620 req->result = dentoffs; 1850 req->result = dentoffs;
1621 1851
1622 if (flags & EIO_READDIR_STAT_ORDER) 1852 if (flags & EIO_READDIR_STAT_ORDER)
1623 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); 1853 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1624 else if (flags & EIO_READDIR_DIRS_FIRST) 1854 else if (flags & EIO_READDIR_DIRS_FIRST)
1625 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1855 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1626 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1856 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1627 else 1857 else
1858 {
1859 /* in this case, all is known, and we just put dirs first and sort them */
1860 eio_dirent *oth = dents + dentoffs;
1861 eio_dirent *dir = dents;
1862
1863 /* now partition dirs to the front, and non-dirs to the back */
1864 /* by walking from both sides and swapping if necessary */
1865 while (oth > dir)
1866 {
1867 if (dir->type == EIO_DT_DIR)
1868 ++dir;
1869 else if ((--oth)->type == EIO_DT_DIR)
1870 {
1871 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1872
1873 ++dir;
1874 }
1875 }
1876
1877 /* now sort the dirs only (dirs all have the same score) */
1878 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1879 }
1880
1881 break;
1882 }
1883
1884 /* now add the entry to our list(s) */
1885 name = D_NAME (entp);
1886
1887 /* skip . and .. entries */
1888 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1889 {
1890 int len = D_NAMLEN (entp) + 1;
1891
1892 while (ecb_expect_false (namesoffs + len > namesalloc))
1893 {
1894 namesalloc *= 2;
1895 req->ptr2 = names = realloc (names, namesalloc);
1896
1897 if (!names)
1898 break;
1899 }
1900
1901 memcpy (names + namesoffs, name, len);
1902
1903 if (dents)
1904 {
1905 struct eio_dirent *ent;
1906
1907 if (ecb_expect_false (dentoffs == dentalloc))
1628 { 1908 {
1629 /* in this case, all is known, and we just put dirs first and sort them */ 1909 dentalloc *= 2;
1910 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1911
1912 if (!dents)
1913 break;
1914 }
1915
1630 eio_dirent *oth = dents + dentoffs; 1916 ent = dents + dentoffs;
1631 eio_dirent *dir = dents;
1632 1917
1633 /* now partition dirs to the front, and non-dirs to the back */ 1918 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1634 /* by walking from both sides and swapping if necessary */ 1919 ent->namelen = len - 1;
1635 while (oth > dir) 1920 ent->inode = D_INO (entp);
1921
1922 inode_bits |= ent->inode;
1923
1924 switch (D_TYPE (entp))
1925 {
1926 default:
1927 ent->type = EIO_DT_UNKNOWN;
1928 flags |= EIO_READDIR_FOUND_UNKNOWN;
1929 break;
1930
1931 #ifdef DT_FIFO
1932 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1933 #endif
1934 #ifdef DT_CHR
1935 case DT_CHR: ent->type = EIO_DT_CHR; break;
1936 #endif
1937 #ifdef DT_MPC
1938 case DT_MPC: ent->type = EIO_DT_MPC; break;
1939 #endif
1940 #ifdef DT_DIR
1941 case DT_DIR: ent->type = EIO_DT_DIR; break;
1942 #endif
1943 #ifdef DT_NAM
1944 case DT_NAM: ent->type = EIO_DT_NAM; break;
1945 #endif
1946 #ifdef DT_BLK
1947 case DT_BLK: ent->type = EIO_DT_BLK; break;
1948 #endif
1949 #ifdef DT_MPB
1950 case DT_MPB: ent->type = EIO_DT_MPB; break;
1951 #endif
1952 #ifdef DT_REG
1953 case DT_REG: ent->type = EIO_DT_REG; break;
1954 #endif
1955 #ifdef DT_NWK
1956 case DT_NWK: ent->type = EIO_DT_NWK; break;
1957 #endif
1958 #ifdef DT_CMP
1959 case DT_CMP: ent->type = EIO_DT_CMP; break;
1960 #endif
1961 #ifdef DT_LNK
1962 case DT_LNK: ent->type = EIO_DT_LNK; break;
1963 #endif
1964 #ifdef DT_SOCK
1965 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1966 #endif
1967 #ifdef DT_DOOR
1968 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1969 #endif
1970 #ifdef DT_WHT
1971 case DT_WHT: ent->type = EIO_DT_WHT; break;
1972 #endif
1973 }
1974
1975 ent->score = 7;
1976
1977 if (flags & EIO_READDIR_DIRS_FIRST)
1978 {
1979 if (ent->type == EIO_DT_UNKNOWN)
1636 { 1980 {
1637 if (dir->type == EIO_DT_DIR) 1981 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1638 ++dir; 1982 ent->score = 1;
1639 else if ((--oth)->type == EIO_DT_DIR) 1983 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1640 { 1984 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1641 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1642
1643 ++dir;
1644 }
1645 } 1985 }
1646 1986 else if (ent->type == EIO_DT_DIR)
1647 /* now sort the dirs only (dirs all have the same score) */ 1987 ent->score = 0;
1648 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1649 } 1988 }
1650
1651 break;
1652 }
1653
1654 /* now add the entry to our list(s) */
1655 name = entp->d_name;
1656
1657 /* skip . and .. entries */
1658 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1659 {
1660 int len = D_NAMLEN (entp) + 1;
1661
1662 while (ecb_expect_false (namesoffs + len > namesalloc))
1663 {
1664 namesalloc *= 2;
1665 X_LOCK (wrklock);
1666 req->ptr2 = names = realloc (names, namesalloc);
1667 X_UNLOCK (wrklock);
1668
1669 if (!names)
1670 break;
1671 } 1989 }
1672 1990
1673 memcpy (names + namesoffs, name, len);
1674
1675 if (dents)
1676 {
1677 struct eio_dirent *ent;
1678
1679 if (ecb_expect_false (dentoffs == dentalloc))
1680 {
1681 dentalloc *= 2;
1682 X_LOCK (wrklock);
1683 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1684 X_UNLOCK (wrklock);
1685
1686 if (!dents)
1687 break;
1688 }
1689
1690 ent = dents + dentoffs;
1691
1692 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1693 ent->namelen = len - 1;
1694 ent->inode = D_INO (entp);
1695
1696 inode_bits |= ent->inode;
1697
1698 switch (D_TYPE (entp))
1699 {
1700 default:
1701 ent->type = EIO_DT_UNKNOWN;
1702 flags |= EIO_READDIR_FOUND_UNKNOWN;
1703 break;
1704
1705 #ifdef DT_FIFO
1706 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1707 #endif
1708 #ifdef DT_CHR
1709 case DT_CHR: ent->type = EIO_DT_CHR; break;
1710 #endif
1711 #ifdef DT_MPC
1712 case DT_MPC: ent->type = EIO_DT_MPC; break;
1713 #endif
1714 #ifdef DT_DIR
1715 case DT_DIR: ent->type = EIO_DT_DIR; break;
1716 #endif
1717 #ifdef DT_NAM
1718 case DT_NAM: ent->type = EIO_DT_NAM; break;
1719 #endif
1720 #ifdef DT_BLK
1721 case DT_BLK: ent->type = EIO_DT_BLK; break;
1722 #endif
1723 #ifdef DT_MPB
1724 case DT_MPB: ent->type = EIO_DT_MPB; break;
1725 #endif
1726 #ifdef DT_REG
1727 case DT_REG: ent->type = EIO_DT_REG; break;
1728 #endif
1729 #ifdef DT_NWK
1730 case DT_NWK: ent->type = EIO_DT_NWK; break;
1731 #endif
1732 #ifdef DT_CMP
1733 case DT_CMP: ent->type = EIO_DT_CMP; break;
1734 #endif
1735 #ifdef DT_LNK
1736 case DT_LNK: ent->type = EIO_DT_LNK; break;
1737 #endif
1738 #ifdef DT_SOCK
1739 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1740 #endif
1741 #ifdef DT_DOOR
1742 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1743 #endif
1744 #ifdef DT_WHT
1745 case DT_WHT: ent->type = EIO_DT_WHT; break;
1746 #endif
1747 }
1748
1749 ent->score = 7;
1750
1751 if (flags & EIO_READDIR_DIRS_FIRST)
1752 {
1753 if (ent->type == EIO_DT_UNKNOWN)
1754 {
1755 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1756 ent->score = 1;
1757 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1758 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1759 }
1760 else if (ent->type == EIO_DT_DIR)
1761 ent->score = 0;
1762 }
1763 }
1764
1765 namesoffs += len; 1991 namesoffs += len;
1766 ++dentoffs; 1992 ++dentoffs;
1767 } 1993 }
1768 1994
1769 if (EIO_CANCELLED (req)) 1995 if (EIO_CANCELLED (req))
1770 { 1996 {
1771 errno = ECANCELED; 1997 errno = ECANCELED;
1772 break; 1998 break;
1773 } 1999 }
2000
2001#ifdef _WIN32
2002 if (!FindNextFile (dirp, &entp))
2003 {
2004 FindClose (dirp);
2005 dirp = 0;
1774 } 2006 }
2007#endif
2008 }
1775} 2009}
2010
2011/*****************************************************************************/
2012/* working directory stuff */
2013/* various deficiencies in the posix 2008 api force us to */
2014/* keep the absolute path in string form at all times */
2015/* fuck yeah. */
2016
2017#if !HAVE_AT
2018
2019/* a bit like realpath, but usually faster because it doesn'T have to return */
2020/* an absolute or canonical path */
2021static const char *
2022wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2023{
2024 if (!wd || *path == '/')
2025 return path;
2026
2027 if (path [0] == '.' && !path [1])
2028 return wd->str;
2029
2030 {
2031 int l1 = wd->len;
2032 int l2 = strlen (path);
2033
2034 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2035
2036 memcpy (res, wd->str, l1);
2037 res [l1] = '/';
2038 memcpy (res + l1 + 1, path, l2 + 1);
2039
2040 return res;
2041 }
2042}
2043
2044#endif
2045
2046static eio_wd
2047eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2048{
2049 int fd;
2050 eio_wd res;
2051 int len = eio__realpath (tmpbuf, wd, path);
2052
2053 if (len < 0)
2054 return EIO_INVALID_WD;
2055
2056#if HAVE_AT
2057 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
2058
2059 if (fd < 0)
2060 return EIO_INVALID_WD;
2061#endif
2062
2063 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
2064
2065#if HAVE_AT
2066 res->fd = fd;
2067#endif
2068
2069 res->len = len;
2070 memcpy (res->str, tmpbuf->ptr, len);
2071 res->str [len] = 0;
2072
2073 return res;
2074}
2075
2076eio_wd
2077eio_wd_open_sync (eio_wd wd, const char *path)
2078{
2079 struct tmpbuf tmpbuf = { 0 };
2080 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2081 free (tmpbuf.ptr);
2082
2083 return wd;
2084}
2085
2086void
2087eio_wd_close_sync (eio_wd wd)
2088{
2089 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
2090 {
2091 #if HAVE_AT
2092 close (wd->fd);
2093 #endif
2094 free (wd);
2095 }
2096}
2097
2098#if HAVE_AT
2099
2100/* they forgot these */
2101
2102static int
2103eio__truncateat (int dirfd, const char *path, off_t length)
2104{
2105 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
2106 int res;
2107
2108 if (fd < 0)
2109 return fd;
2110
2111 res = ftruncate (fd, length);
2112 close (fd);
2113 return res;
2114}
2115
2116static int
2117eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
2118{
2119 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2120 int res;
2121
2122 if (fd < 0)
2123 return fd;
2124
2125 res = fstatvfs (fd, buf);
2126 close (fd);
2127 return res;
2128
2129}
2130
2131#endif
1776 2132
1777/*****************************************************************************/ 2133/*****************************************************************************/
1778 2134
1779#define ALLOC(len) \ 2135#define ALLOC(len) \
1780 if (!req->ptr2) \ 2136 if (!req->ptr2) \
1789 req->result = -1; \ 2145 req->result = -1; \
1790 break; \ 2146 break; \
1791 } \ 2147 } \
1792 } 2148 }
1793 2149
2150static void ecb_noinline ecb_cold
2151etp_proc_init (void)
2152{
2153#if HAVE_PRCTL_SET_NAME
2154 /* provide a more sensible "thread name" */
2155 char name[16 + 1];
2156 const int namelen = sizeof (name) - 1;
2157 int len;
2158
2159 prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
2160 name [namelen] = 0;
2161 len = strlen (name);
2162 strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
2163 prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
2164#endif
2165}
2166
1794X_THREAD_PROC (etp_proc) 2167X_THREAD_PROC (etp_proc)
1795{ 2168{
1796 ETP_REQ *req; 2169 ETP_REQ *req;
1797 struct timespec ts; 2170 struct timespec ts;
1798 etp_worker *self = (etp_worker *)thr_arg; 2171 etp_worker *self = (etp_worker *)thr_arg;
1799 int timeout; 2172
2173 etp_proc_init ();
1800 2174
1801 /* try to distribute timeouts somewhat evenly */ 2175 /* try to distribute timeouts somewhat evenly */
1802 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2176 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1803 2177
1804 for (;;) 2178 for (;;)
1807 2181
1808 X_LOCK (reqlock); 2182 X_LOCK (reqlock);
1809 2183
1810 for (;;) 2184 for (;;)
1811 { 2185 {
1812 self->req = req = reqq_shift (&req_queue); 2186 req = reqq_shift (&req_queue);
1813 2187
1814 if (req) 2188 if (req)
1815 break; 2189 break;
1816 2190
1817 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */ 2191 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1855 ++npending; 2229 ++npending;
1856 2230
1857 if (!reqq_push (&res_queue, req) && want_poll_cb) 2231 if (!reqq_push (&res_queue, req) && want_poll_cb)
1858 want_poll_cb (); 2232 want_poll_cb ();
1859 2233
1860 self->req = 0;
1861 etp_worker_clear (self); 2234 etp_worker_clear (self);
1862 2235
1863 X_UNLOCK (reslock); 2236 X_UNLOCK (reslock);
1864 } 2237 }
1865 2238
1866quit: 2239quit:
2240 free (req);
2241
1867 X_LOCK (wrklock); 2242 X_LOCK (wrklock);
1868 etp_worker_free (self); 2243 etp_worker_free (self);
1869 X_UNLOCK (wrklock); 2244 X_UNLOCK (wrklock);
1870
1871 return 0;
1872} 2245}
1873 2246
1874/*****************************************************************************/ 2247/*****************************************************************************/
1875 2248
1876int ecb_cold 2249int ecb_cold
1877eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2250eio_init (void (*want_poll)(void), void (*done_poll)(void))
1878{ 2251{
2252#if !HAVE_PREADWRITE
2253 X_MUTEX_CREATE (preadwritelock);
2254#endif
2255
1879 return etp_init (want_poll, done_poll); 2256 return etp_init (want_poll, done_poll);
1880} 2257}
1881 2258
1882ecb_inline void 2259ecb_inline void
1883eio_api_destroy (eio_req *req) 2260eio_api_destroy (eio_req *req)
1910 } 2287 }
1911 2288
1912static void 2289static void
1913eio_execute (etp_worker *self, eio_req *req) 2290eio_execute (etp_worker *self, eio_req *req)
1914{ 2291{
2292#if HAVE_AT
2293 int dirfd;
2294#else
2295 const char *path;
2296#endif
2297
1915 if (ecb_expect_false (EIO_CANCELLED (req))) 2298 if (ecb_expect_false (EIO_CANCELLED (req)))
1916 { 2299 {
1917 req->result = -1; 2300 req->result = -1;
1918 req->errorno = ECANCELED; 2301 req->errorno = ECANCELED;
1919 return; 2302 return;
1920 } 2303 }
1921 2304
2305 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2306 {
2307 req->result = -1;
2308 req->errorno = ENOENT;
2309 return;
2310 }
2311
2312 if (req->type >= EIO_OPEN)
2313 {
2314 #if HAVE_AT
2315 dirfd = WD2FD (req->wd);
2316 #else
2317 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2318 #endif
2319 }
2320
1922 switch (req->type) 2321 switch (req->type)
1923 { 2322 {
2323 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2324 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2325 break;
2326 case EIO_WD_CLOSE: req->result = 0;
2327 eio_wd_close_sync (req->wd); break;
2328
2329 case EIO_SEEK: eio__lseek (req); break;
1924 case EIO_READ: ALLOC (req->size); 2330 case EIO_READ: ALLOC (req->size);
1925 req->result = req->offs >= 0 2331 req->result = req->offs >= 0
1926 ? pread (req->int1, req->ptr2, req->size, req->offs) 2332 ? pread (req->int1, req->ptr2, req->size, req->offs)
1927 : read (req->int1, req->ptr2, req->size); break; 2333 : read (req->int1, req->ptr2, req->size); break;
1928 case EIO_WRITE: req->result = req->offs >= 0 2334 case EIO_WRITE: req->result = req->offs >= 0
1929 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2335 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1930 : write (req->int1, req->ptr2, req->size); break; 2336 : write (req->int1, req->ptr2, req->size); break;
1931 2337
1932 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2338 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1933 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2339 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2340
2341#if HAVE_AT
1934 2342
1935 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2343 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1936 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2344 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1937 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2345 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2346 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
2347 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
2348 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
2349 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
2350 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
2351
2352 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
2353 case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
2354 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
2355 case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
2356 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
2357 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
2358 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2359 case EIO_READLINK: ALLOC (PATH_MAX);
2360 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
2361 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2362 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2363 case EIO_UTIME:
2364 case EIO_FUTIME:
2365 {
2366 struct timespec ts[2];
2367 struct timespec *times;
2368
2369 if (req->nv1 != -1. || req->nv2 != -1.)
2370 {
2371 ts[0].tv_sec = req->nv1;
2372 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2373 ts[1].tv_sec = req->nv2;
2374 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2375
2376 times = ts;
2377 }
2378 else
2379 times = 0;
2380
2381 req->result = req->type == EIO_FUTIME
2382 ? futimens (req->int1, times)
2383 : utimensat (dirfd, req->ptr1, times, 0);
2384 }
2385 break;
2386
2387#else
2388
2389 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2390 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2391 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1938 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2392 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2393 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2394 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2395 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2396 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2397
2398 case EIO_UNLINK: req->result = unlink (path ); break;
2399 case EIO_RMDIR: req->result = rmdir (path ); break;
2400 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
2401 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2402 case EIO_LINK: req->result = link (path , req->ptr2); break;
2403 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2404 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2405 case EIO_READLINK: ALLOC (PATH_MAX);
2406 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2407 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2408 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2409
2410 case EIO_UTIME:
2411 case EIO_FUTIME:
2412 {
2413 struct timeval tv[2];
2414 struct timeval *times;
2415
2416 if (req->nv1 != -1. || req->nv2 != -1.)
2417 {
2418 tv[0].tv_sec = req->nv1;
2419 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
2420 tv[1].tv_sec = req->nv2;
2421 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
2422
2423 times = tv;
2424 }
2425 else
2426 times = 0;
2427
2428 req->result = req->type == EIO_FUTIME
2429 ? futimes (req->int1, times)
2430 : utimes (req->ptr1, times);
2431 }
2432 break;
2433
2434#endif
2435
2436 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2437 {
2438 ALLOC (req->result);
2439 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2440 }
2441 break;
2442
1939 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2443 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1940 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2444 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1941 2445
1942 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1943 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1944 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 2446 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1945 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 2447 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1946 2448
1947 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1948 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2449 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1949 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1950 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2450 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1951 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1952 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 2451 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1953 2452
1954 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1955 case EIO_CLOSE: req->result = close (req->int1); break; 2453 case EIO_CLOSE: req->result = close (req->int1); break;
1956 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 2454 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1957 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1958 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1959 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1960 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1961 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1962 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1963 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1964
1965 case EIO_REALPATH: eio__realpath (req, self); break;
1966
1967 case EIO_READLINK: ALLOC (PATH_MAX);
1968 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1969
1970 case EIO_SYNC: req->result = 0; sync (); break; 2455 case EIO_SYNC: req->result = 0; sync (); break;
1971 case EIO_FSYNC: req->result = fsync (req->int1); break; 2456 case EIO_FSYNC: req->result = fsync (req->int1); break;
1972 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2457 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2458 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2459 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1973 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 2460 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1974 case EIO_MTOUCH: req->result = eio__mtouch (req); break; 2461 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1975 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 2462 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1976 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 2463 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1977 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1978 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break; 2464 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1979 2465
1980 case EIO_READDIR: eio__scandir (req, self); break; 2466 case EIO_READDIR: eio__scandir (req, self); break;
1981 2467
1982 case EIO_BUSY: 2468 case EIO_BUSY:
1992 req->result = select (0, 0, 0, 0, &tv); 2478 req->result = select (0, 0, 0, 0, &tv);
1993 } 2479 }
1994#endif 2480#endif
1995 break; 2481 break;
1996 2482
1997 case EIO_UTIME:
1998 case EIO_FUTIME:
1999 {
2000 struct timeval tv[2];
2001 struct timeval *times;
2002
2003 if (req->nv1 != -1. || req->nv2 != -1.)
2004 {
2005 tv[0].tv_sec = req->nv1;
2006 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
2007 tv[1].tv_sec = req->nv2;
2008 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
2009
2010 times = tv;
2011 }
2012 else
2013 times = 0;
2014
2015 req->result = req->type == EIO_FUTIME
2016 ? futimes (req->int1, times)
2017 : utimes (req->ptr1, times);
2018 }
2019 break;
2020
2021 case EIO_GROUP: 2483 case EIO_GROUP:
2022 abort (); /* handled in eio_request */ 2484 abort (); /* handled in eio_request */
2023 2485
2024 case EIO_NOP: 2486 case EIO_NOP:
2025 req->result = 0; 2487 req->result = 0;
2028 case EIO_CUSTOM: 2490 case EIO_CUSTOM:
2029 req->feed (req); 2491 req->feed (req);
2030 break; 2492 break;
2031 2493
2032 default: 2494 default:
2033 errno = ENOSYS;
2034 req->result = -1; 2495 req->result = EIO_ENOSYS ();
2035 break; 2496 break;
2036 } 2497 }
2037 2498
2038 req->errorno = errno; 2499 req->errorno = errno;
2039} 2500}
2040 2501
2041#ifndef EIO_NO_WRAPPERS 2502#ifndef EIO_NO_WRAPPERS
2042 2503
2504eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2505{
2506 REQ (EIO_WD_OPEN); PATH; SEND;
2507}
2508
2509eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2510{
2511 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2512}
2513
2043eio_req *eio_nop (int pri, eio_cb cb, void *data) 2514eio_req *eio_nop (int pri, eio_cb cb, void *data)
2044{ 2515{
2045 REQ (EIO_NOP); SEND; 2516 REQ (EIO_NOP); SEND;
2046} 2517}
2047 2518
2063eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2534eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2064{ 2535{
2065 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2536 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2066} 2537}
2067 2538
2539eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2540{
2541 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2542}
2543
2544eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2545{
2546 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2547}
2548
2549eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2550{
2551 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2552}
2553
2068eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2554eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2069{ 2555{
2070 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2556 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2071} 2557}
2072 2558
2078eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) 2564eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2079{ 2565{
2080 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2566 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2081} 2567}
2082 2568
2083eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2084{
2085 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2086}
2087
2088eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data) 2569eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2089{ 2570{
2090 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND; 2571 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2091} 2572}
2092 2573
2093eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2094{
2095 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2096}
2097
2098eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2574eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2099{ 2575{
2100 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2576 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2101} 2577}
2102 2578
2103eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2579eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
2104{ 2580{
2105 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2581 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
2106} 2582}
2107 2583
2584eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2585{
2586 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2587}
2588
2108eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2589eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2109{ 2590{
2110 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2591 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2111} 2592}
2112 2593
2138eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2619eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
2139{ 2620{
2140 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2621 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
2141} 2622}
2142 2623
2143eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2624eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2144{ 2625{
2145 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2626 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2146} 2627}
2147 2628
2148eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2629eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
2168eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2649eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
2169{ 2650{
2170 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2651 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
2171} 2652}
2172 2653
2173eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2654eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2174{ 2655{
2175 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2656 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2176} 2657}
2177 2658
2178eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2659eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2325} 2806}
2326 2807
2327/*****************************************************************************/ 2808/*****************************************************************************/
2328/* misc garbage */ 2809/* misc garbage */
2329 2810
2330ssize_t 2811eio_ssize_t
2331eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2812eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2332{ 2813{
2333 etp_worker wrk;
2334 ssize_t ret;
2335
2336 wrk.dbuf = 0;
2337
2338 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2814 return eio__sendfile (ofd, ifd, offset, count);
2339
2340 if (wrk.dbuf)
2341 free (wrk.dbuf);
2342
2343 return ret;
2344} 2815}
2345 2816

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines