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

Comparing libeio/eio.c (file contents):
Revision 1.82 by root, Thu Jul 7 15:44:44 2011 UTC vs.
Revision 1.106 by root, Mon Sep 26 20:19:08 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>
62
63#if _POSIX_VERSION >= 200809L
64# define HAVE_AT 1
65#else
66# define HAVE_AT 0
67#endif
63 68
64/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */ 69/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
65/* intptr_t only comes from stdint.h, says idiot openbsd coder */ 70/* intptr_t only comes from stdint.h, says idiot openbsd coder */
66#if HAVE_STDINT_H 71#if HAVE_STDINT_H
67# include <stdint.h> 72# include <stdint.h>
68#endif 73#endif
69 74
70#ifndef ECANCELED 75#ifndef ECANCELED
71# define ECANCELED EDOM 76# define ECANCELED EDOM
72#endif 77#endif
78#ifndef ELOOP
79# define ELOOP EDOM
80#endif
81
82#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
83# define ENOTSOCK WSAENOTSOCK
84#endif
73 85
74static void eio_destroy (eio_req *req); 86static void eio_destroy (eio_req *req);
75 87
76#ifndef EIO_FINISH 88#ifndef EIO_FINISH
77# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 89# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
83 95
84#ifndef EIO_FEED 96#ifndef EIO_FEED
85# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 97# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
86#endif 98#endif
87 99
100#ifndef EIO_FD_TO_WIN32_HANDLE
101# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
102#endif
103#ifndef EIO_WIN32_HANDLE_TO_FD
104# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
105#endif
106
107#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
108
109#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
110
88#ifdef _WIN32 111#ifdef _WIN32
89 112
90 /*doh*/ 113 #undef PAGESIZE
114 #define PAGESIZE 4096 /* GetSystemInfo? */
115
116 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
117
118 #ifdef EIO_STRUCT_STATI64
119 /* look at perl's non-sloppy stat */
120 #define stat(path,buf) _stati64 (path,buf)
121 #define fstat(fd,buf) _fstati64 (fd,buf)
122 #endif
123 #define lstat(path,buf) stat (path,buf)
124 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
125 #define mkdir(path,mode) _mkdir (path)
126 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
127
128 #define chmod(path,mode) _chmod (path, mode)
129 #define dup(fd) _dup (fd)
130 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
131
132 #define fchmod(fd,mode) EIO_ENOSYS ()
133 #define chown(path,uid,gid) EIO_ENOSYS ()
134 #define fchown(fd,uid,gid) EIO_ENOSYS ()
135 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
136 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
137 #define mknod(path,mode,dev) EIO_ENOSYS ()
138 #define sync() EIO_ENOSYS ()
139 #define readlink(path,buf,s) EIO_ENOSYS ()
140 #define statvfs(path,buf) EIO_ENOSYS ()
141 #define fstatvfs(fd,buf) EIO_ENOSYS ()
142
143 /* rename() uses MoveFile, which fails to overwrite */
144 #define rename(old,neu) eio__rename (old, neu)
145
146 static int
147 eio__rename (const char *old, const char *neu)
148 {
149 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
150 return 0;
151
152 /* should steal _dosmaperr */
153 switch (GetLastError ())
154 {
155 case ERROR_FILE_NOT_FOUND:
156 case ERROR_PATH_NOT_FOUND:
157 case ERROR_INVALID_DRIVE:
158 case ERROR_NO_MORE_FILES:
159 case ERROR_BAD_NETPATH:
160 case ERROR_BAD_NET_NAME:
161 case ERROR_BAD_PATHNAME:
162 case ERROR_FILENAME_EXCED_RANGE:
163 errno = ENOENT;
164 break;
165
166 default:
167 errno = EACCES;
168 break;
169 }
170
171 return -1;
172 }
173
174 /* we could even stat and see if it exists */
175 static int
176 symlink (const char *old, const char *neu)
177 {
178 #if WINVER >= 0x0600
179 if (CreateSymbolicLink (neu, old, 1))
180 return 0;
181
182 if (CreateSymbolicLink (neu, old, 0))
183 return 0;
184 #endif
185
186 return EIO_ERRNO (ENOENT, -1);
187 }
188
189 /* POSIX API only */
190 #define CreateHardLink(neu,old,flags) 0
191 #define CreateSymbolicLink(neu,old,flags) 0
192
193 struct statvfs
194 {
195 int dummy;
196 };
197
198 #define DT_DIR EIO_DT_DIR
199 #define DT_REG EIO_DT_REG
200 #define D_NAME(entp) entp.cFileName
201 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
202
91#else 203#else
92 204
93# include <sys/time.h> 205 #include <sys/time.h>
94# include <sys/select.h> 206 #include <sys/select.h>
207 #include <sys/statvfs.h>
95# include <unistd.h> 208 #include <unistd.h>
209 #include <signal.h>
210 #include <dirent.h>
211
212 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
213 #include <sys/mman.h>
214 #endif
215
216 #define D_NAME(entp) entp->d_name
217
218 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
219 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
220 #define _DIRENT_HAVE_D_TYPE /* sigh */
221 #define D_INO(de) (de)->d_fileno
222 #define D_NAMLEN(de) (de)->d_namlen
223 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
224 #define D_INO(de) (de)->d_ino
225 #endif
226
227 #ifdef _D_EXACT_NAMLEN
228 #undef D_NAMLEN
229 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
230 #endif
231
232 #ifdef _DIRENT_HAVE_D_TYPE
233 #define D_TYPE(de) (de)->d_type
234 #endif
235
236 #ifndef EIO_STRUCT_DIRENT
237 #define EIO_STRUCT_DIRENT struct dirent
238 #endif
239
240#endif
241
242#if HAVE_UTIMES
96# include <utime.h> 243# 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 244#endif
112 245
113#ifdef _D_EXACT_NAMLEN 246#if HAVE_SYS_SYSCALL_H
114# undef D_NAMLEN 247# 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 248#endif
121 249
122# ifndef EIO_STRUCT_DIRENT 250#if HAVE_SYS_PRCTL_H
123# define EIO_STRUCT_DIRENT struct dirent 251# include <sys/prctl.h>
124# endif
125
126#endif 252#endif
127 253
128#if HAVE_SENDFILE 254#if HAVE_SENDFILE
129# if __linux 255# if __linux
130# include <sys/sendfile.h> 256# include <sys/sendfile.h>
145#endif 271#endif
146#ifndef D_INO 272#ifndef D_INO
147# define D_INO(de) 0 273# define D_INO(de) 0
148#endif 274#endif
149#ifndef D_NAMLEN 275#ifndef D_NAMLEN
150# define D_NAMLEN(de) strlen ((de)->d_name) 276# define D_NAMLEN(entp) strlen (D_NAME (entp))
151#endif 277#endif
152 278
153/* used for struct dirent, AIX doesn't provide it */ 279/* used for struct dirent, AIX doesn't provide it */
154#ifndef NAME_MAX 280#ifndef NAME_MAX
155# define NAME_MAX 4096 281# define NAME_MAX 4096
162 288
163/* buffer size for various temporary buffers */ 289/* buffer size for various temporary buffers */
164#define EIO_BUFSIZE 65536 290#define EIO_BUFSIZE 65536
165 291
166#define dBUF \ 292#define dBUF \
167 char *eio_buf; \
168 ETP_WORKER_LOCK (self); \
169 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 293 char *eio_buf = malloc (EIO_BUFSIZE); \
170 ETP_WORKER_UNLOCK (self); \
171 errno = ENOMEM; \ 294 errno = ENOMEM; \
172 if (!eio_buf) \ 295 if (!eio_buf) \
173 return -1; 296 return -1
297
298#define FUBd \
299 free (eio_buf)
174 300
175#define EIO_TICKS ((1000000 + 1023) >> 10) 301#define EIO_TICKS ((1000000 + 1023) >> 10)
302
303/*****************************************************************************/
304
305struct tmpbuf
306{
307 void *ptr;
308 int len;
309};
310
311static void *
312tmpbuf_get (struct tmpbuf *buf, int len)
313{
314 if (buf->len < len)
315 {
316 free (buf->ptr);
317 buf->ptr = malloc (buf->len = len);
318 }
319
320 return buf->ptr;
321}
322
323/*****************************************************************************/
176 324
177#define ETP_PRI_MIN EIO_PRI_MIN 325#define ETP_PRI_MIN EIO_PRI_MIN
178#define ETP_PRI_MAX EIO_PRI_MAX 326#define ETP_PRI_MAX EIO_PRI_MAX
179 327
180struct etp_worker; 328struct etp_worker;
184static int eio_finish (eio_req *req); 332static int eio_finish (eio_req *req);
185#define ETP_FINISH(req) eio_finish (req) 333#define ETP_FINISH(req) eio_finish (req)
186static void eio_execute (struct etp_worker *self, eio_req *req); 334static void eio_execute (struct etp_worker *self, eio_req *req);
187#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 335#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
188 336
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/*****************************************************************************/ 337/*****************************************************************************/
207 338
208#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 339#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
209 340
210/* calculate time difference in ~1/EIO_TICKS of a second */ 341/* calculate time difference in ~1/EIO_TICKS of a second */
221static void (*done_poll_cb) (void); 352static void (*done_poll_cb) (void);
222 353
223static unsigned int max_poll_time; /* reslock */ 354static unsigned int max_poll_time; /* reslock */
224static unsigned int max_poll_reqs; /* reslock */ 355static unsigned int max_poll_reqs; /* reslock */
225 356
226static volatile unsigned int nreqs; /* reqlock */ 357static unsigned int nreqs; /* reqlock */
227static volatile unsigned int nready; /* reqlock */ 358static unsigned int nready; /* reqlock */
228static volatile unsigned int npending; /* reqlock */ 359static unsigned int npending; /* reqlock */
229static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */ 360static 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 */ 361static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
231 362
232static xmutex_t wrklock; 363static xmutex_t wrklock;
233static xmutex_t reslock; 364static xmutex_t reslock;
234static xmutex_t reqlock; 365static xmutex_t reqlock;
235static xcond_t reqwait; 366static xcond_t reqwait;
238/* 369/*
239 * make our pread/pwrite emulation safe against themselves, but not against 370 * make our pread/pwrite emulation safe against themselves, but not against
240 * normal read/write by using a mutex. slows down execution a lot, 371 * normal read/write by using a mutex. slows down execution a lot,
241 * but that's your problem, not mine. 372 * but that's your problem, not mine.
242 */ 373 */
243static xmutex_t preadwritelock = X_MUTEX_INIT; 374static xmutex_t preadwritelock;
244#endif 375#endif
245 376
246typedef struct etp_worker 377typedef struct etp_worker
247{ 378{
379 struct tmpbuf tmpbuf;
380
248 /* locked by wrklock */ 381 /* locked by wrklock */
249 struct etp_worker *prev, *next; 382 struct etp_worker *prev, *next;
250 383
251 xthread_t tid; 384 xthread_t tid;
252 385
253 /* locked by reslock, reqlock or wrklock */ 386#ifdef ETP_WORKER_COMMON
254 ETP_REQ *req; /* currently processed request */
255
256 ETP_WORKER_COMMON 387 ETP_WORKER_COMMON
388#endif
257} etp_worker; 389} etp_worker;
258 390
259static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 391static etp_worker wrk_first; /* NOT etp */
260 392
261#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 393#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
262#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 394#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
263 395
264/* worker threads management */ 396/* worker threads management */
265 397
266static void ecb_cold 398static void ecb_cold
267etp_worker_clear (etp_worker *wrk) 399etp_worker_clear (etp_worker *wrk)
268{ 400{
269 ETP_WORKER_CLEAR (wrk);
270} 401}
271 402
272static void ecb_cold 403static void ecb_cold
273etp_worker_free (etp_worker *wrk) 404etp_worker_free (etp_worker *wrk)
274{ 405{
406 free (wrk->tmpbuf.ptr);
407
275 wrk->next->prev = wrk->prev; 408 wrk->next->prev = wrk->prev;
276 wrk->prev->next = wrk->next; 409 wrk->prev->next = wrk->next;
277 410
278 free (wrk); 411 free (wrk);
279} 412}
335} etp_reqq; 468} etp_reqq;
336 469
337static etp_reqq req_queue; 470static etp_reqq req_queue;
338static etp_reqq res_queue; 471static etp_reqq res_queue;
339 472
473static void ecb_noinline ecb_cold
474reqq_init (etp_reqq *q)
475{
476 int pri;
477
478 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
479 q->qs[pri] = q->qe[pri] = 0;
480
481 q->size = 0;
482}
483
340static int ecb_noinline 484static int ecb_noinline
341reqq_push (etp_reqq *q, ETP_REQ *req) 485reqq_push (etp_reqq *q, ETP_REQ *req)
342{ 486{
343 int pri = req->pri; 487 int pri = req->pri;
344 req->next = 0; 488 req->next = 0;
378 } 522 }
379 523
380 abort (); 524 abort ();
381} 525}
382 526
383static void ecb_cold 527static int ecb_cold
384etp_thread_init (void) 528etp_init (void (*want_poll)(void), void (*done_poll)(void))
385{ 529{
386 X_MUTEX_CREATE (wrklock); 530 X_MUTEX_CREATE (wrklock);
387 X_MUTEX_CREATE (reslock); 531 X_MUTEX_CREATE (reslock);
388 X_MUTEX_CREATE (reqlock); 532 X_MUTEX_CREATE (reqlock);
389 X_COND_CREATE (reqwait); 533 X_COND_CREATE (reqwait);
390}
391 534
392static void ecb_cold 535 reqq_init (&req_queue);
393etp_atfork_prepare (void) 536 reqq_init (&res_queue);
394{
395 X_LOCK (wrklock);
396 X_LOCK (reqlock);
397 X_LOCK (reslock);
398#if !HAVE_PREADWRITE
399 X_LOCK (preadwritelock);
400#endif
401}
402 537
403static void ecb_cold 538 wrk_first.next =
404etp_atfork_parent (void) 539 wrk_first.prev = &wrk_first;
405{
406#if !HAVE_PREADWRITE
407 X_UNLOCK (preadwritelock);
408#endif
409 X_UNLOCK (reslock);
410 X_UNLOCK (reqlock);
411 X_UNLOCK (wrklock);
412}
413
414static void ecb_cold
415etp_atfork_child (void)
416{
417 ETP_REQ *prv;
418
419 while ((prv = reqq_shift (&req_queue)))
420 ETP_DESTROY (prv);
421
422 while ((prv = reqq_shift (&res_queue)))
423 ETP_DESTROY (prv);
424
425 while (wrk_first.next != &wrk_first)
426 {
427 etp_worker *wrk = wrk_first.next;
428
429 if (wrk->req)
430 ETP_DESTROY (wrk->req);
431
432 etp_worker_clear (wrk);
433 etp_worker_free (wrk);
434 }
435 540
436 started = 0; 541 started = 0;
437 idle = 0; 542 idle = 0;
438 nreqs = 0; 543 nreqs = 0;
439 nready = 0; 544 nready = 0;
440 npending = 0; 545 npending = 0;
441
442 etp_thread_init ();
443}
444
445static void ecb_cold
446etp_once_init (void)
447{
448 etp_thread_init ();
449 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
450}
451
452static int ecb_cold
453etp_init (void (*want_poll)(void), void (*done_poll)(void))
454{
455 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
456
457 pthread_once (&doinit, etp_once_init);
458 546
459 want_poll_cb = want_poll; 547 want_poll_cb = want_poll;
460 done_poll_cb = done_poll; 548 done_poll_cb = done_poll;
461 549
462 return 0; 550 return 0;
502} 590}
503 591
504static void ecb_cold 592static void ecb_cold
505etp_end_thread (void) 593etp_end_thread (void)
506{ 594{
507 eio_req *req = calloc (1, sizeof (eio_req)); 595 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
508 596
509 req->type = -1; 597 req->type = -1;
510 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 598 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
511 599
512 X_LOCK (reqlock); 600 X_LOCK (reqlock);
845# undef pread 933# undef pread
846# undef pwrite 934# undef pwrite
847# define pread eio__pread 935# define pread eio__pread
848# define pwrite eio__pwrite 936# define pwrite eio__pwrite
849 937
850static ssize_t 938static eio_ssize_t
851eio__pread (int fd, void *buf, size_t count, off_t offset) 939eio__pread (int fd, void *buf, size_t count, off_t offset)
852{ 940{
853 ssize_t res; 941 eio_ssize_t res;
854 off_t ooffset; 942 off_t ooffset;
855 943
856 X_LOCK (preadwritelock); 944 X_LOCK (preadwritelock);
857 ooffset = lseek (fd, 0, SEEK_CUR); 945 ooffset = lseek (fd, 0, SEEK_CUR);
858 lseek (fd, offset, SEEK_SET); 946 lseek (fd, offset, SEEK_SET);
861 X_UNLOCK (preadwritelock); 949 X_UNLOCK (preadwritelock);
862 950
863 return res; 951 return res;
864} 952}
865 953
866static ssize_t 954static eio_ssize_t
867eio__pwrite (int fd, void *buf, size_t count, off_t offset) 955eio__pwrite (int fd, void *buf, size_t count, off_t offset)
868{ 956{
869 ssize_t res; 957 eio_ssize_t res;
870 off_t ooffset; 958 off_t ooffset;
871 959
872 X_LOCK (preadwritelock); 960 X_LOCK (preadwritelock);
873 ooffset = lseek (fd, 0, SEEK_CUR); 961 ooffset = lseek (fd, 0, SEEK_CUR);
874 lseek (fd, offset, SEEK_SET); 962 lseek (fd, offset, SEEK_SET);
919 1007
920#if !HAVE_FDATASYNC 1008#if !HAVE_FDATASYNC
921# undef fdatasync 1009# undef fdatasync
922# define fdatasync(fd) fsync (fd) 1010# define fdatasync(fd) fsync (fd)
923#endif 1011#endif
1012
1013static int
1014eio__syncfs (int fd)
1015{
1016 int res;
1017
1018#if HAVE_SYS_SYNCFS
1019 res = (int)syscall (__NR_syncfs, (int)(fd));
1020#else
1021 res = -1;
1022 errno = ENOSYS;
1023#endif
1024
1025 if (res < 0 && errno == ENOSYS && fd >= 0)
1026 sync ();
1027
1028 return res;
1029}
924 1030
925/* sync_file_range always needs emulation */ 1031/* sync_file_range always needs emulation */
926static int 1032static int
927eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 1033eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
928{ 1034{
948 /* even though we could play tricks with the flags, it's better to always 1054 /* even though we could play tricks with the flags, it's better to always
949 * call fdatasync, as that matches the expectation of its users best */ 1055 * call fdatasync, as that matches the expectation of its users best */
950 return fdatasync (fd); 1056 return fdatasync (fd);
951} 1057}
952 1058
1059static int
1060eio__fallocate (int fd, int mode, off_t offset, size_t len)
1061{
1062#if HAVE_FALLOCATE
1063 return fallocate (fd, mode, offset, len);
1064#else
1065 errno = ENOSYS;
1066 return -1;
1067#endif
1068}
1069
953#if !HAVE_READAHEAD 1070#if !HAVE_READAHEAD
954# undef readahead 1071# undef readahead
955# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1072# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
956 1073
957static ssize_t 1074static eio_ssize_t
958eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 1075eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
959{ 1076{
960 size_t todo = count; 1077 size_t todo = count;
961 dBUF; 1078 dBUF;
962 1079
967 pread (fd, eio_buf, len, offset); 1084 pread (fd, eio_buf, len, offset);
968 offset += len; 1085 offset += len;
969 todo -= len; 1086 todo -= len;
970 } 1087 }
971 1088
1089 FUBd;
1090
972 errno = 0; 1091 errno = 0;
973 return count; 1092 return count;
974} 1093}
975 1094
976#endif 1095#endif
977 1096
978/* sendfile always needs emulation */ 1097/* sendfile always needs emulation */
979static ssize_t 1098static eio_ssize_t
980eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1099eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
981{ 1100{
982 ssize_t written = 0; 1101 eio_ssize_t written = 0;
983 ssize_t res; 1102 eio_ssize_t res;
984 1103
985 if (!count) 1104 if (!count)
986 return 0; 1105 return 0;
987 1106
988 for (;;) 1107 for (;;)
1040 if (res < 0 && sbytes) 1159 if (res < 0 && sbytes)
1041 res = sbytes; 1160 res = sbytes;
1042 1161
1043# endif 1162# endif
1044 1163
1045#elif defined (_WIN32) 1164#elif defined (_WIN32) && 0
1046 /* does not work, just for documentation of what would need to be done */ 1165 /* does not work, just for documentation of what would need to be done */
1047 /* actually, cannot be done like this, as TransmitFile changes the file offset, */ 1166 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1048 /* libeio guarantees that the file offset does not change, and windows */ 1167 /* libeio guarantees that the file offset does not change, and windows */
1049 /* has no way to get an independent handle to the same file description */ 1168 /* has no way to get an independent handle to the same file description */
1050 HANDLE h = TO_SOCKET (ifd); 1169 HANDLE h = TO_SOCKET (ifd);
1083 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1202 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1084 /* BSDs */ 1203 /* BSDs */
1085#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1204#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1086 || errno == ENOTSUP 1205 || errno == ENOTSUP
1087#endif 1206#endif
1207#ifdef EOPNOTSUPP /* windows */
1088 || errno == EOPNOTSUPP /* BSDs */ 1208 || errno == EOPNOTSUPP /* BSDs */
1209#endif
1089#if __solaris 1210#if __solaris
1090 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1211 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1091#endif 1212#endif
1092 ) 1213 )
1093 ) 1214 )
1097 1218
1098 res = 0; 1219 res = 0;
1099 1220
1100 while (count) 1221 while (count)
1101 { 1222 {
1102 ssize_t cnt; 1223 eio_ssize_t cnt;
1103 1224
1104 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1225 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1105 1226
1106 if (cnt <= 0) 1227 if (cnt <= 0)
1107 { 1228 {
1119 1240
1120 offset += cnt; 1241 offset += cnt;
1121 res += cnt; 1242 res += cnt;
1122 count -= cnt; 1243 count -= cnt;
1123 } 1244 }
1245
1246 FUBd;
1124 } 1247 }
1125 1248
1126 return res; 1249 return res;
1127} 1250}
1128 1251
1155 /* round up length */ 1278 /* round up length */
1156 *length = (*length + mask) & ~mask; 1279 *length = (*length + mask) & ~mask;
1157} 1280}
1158 1281
1159#if !_POSIX_MEMLOCK 1282#if !_POSIX_MEMLOCK
1160# define eio__mlockall(a) ((errno = ENOSYS), -1) 1283# define eio__mlockall(a) EIO_ENOSYS ()
1161#else 1284#else
1162 1285
1163static int 1286static int
1164eio__mlockall (int flags) 1287eio__mlockall (int flags)
1165{ 1288{
1179 return mlockall (flags); 1302 return mlockall (flags);
1180} 1303}
1181#endif 1304#endif
1182 1305
1183#if !_POSIX_MEMLOCK_RANGE 1306#if !_POSIX_MEMLOCK_RANGE
1184# define eio__mlock(a,b) ((errno = ENOSYS), -1) 1307# define eio__mlock(a,b) EIO_ENOSYS ()
1185#else 1308#else
1186 1309
1187static int 1310static int
1188eio__mlock (void *addr, size_t length) 1311eio__mlock (void *addr, size_t length)
1189{ 1312{
1193} 1316}
1194 1317
1195#endif 1318#endif
1196 1319
1197#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1320#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1198# define eio__msync(a,b,c) ((errno = ENOSYS), -1) 1321# define eio__msync(a,b,c) EIO_ENOSYS ()
1199#else 1322#else
1200 1323
1201static int 1324static int
1202eio__msync (void *mem, size_t len, int flags) 1325eio__msync (void *mem, size_t len, int flags)
1203{ 1326{
1243} 1366}
1244 1367
1245/*****************************************************************************/ 1368/*****************************************************************************/
1246/* requests implemented outside eio_execute, because they are so large */ 1369/* requests implemented outside eio_execute, because they are so large */
1247 1370
1371/* copies some absolute path to tmpbuf */
1372static char *
1373eio__getwd (struct tmpbuf *tmpbuf, eio_wd wd)
1374{
1375 if (wd == EIO_CWD)
1376 return getcwd (tmpbuf->ptr, PATH_MAX);
1377
1378#if HAVE_AT
1379 abort (); /*TODO*/
1380#else
1381 strcpy (tmpbuf->ptr, wd);
1382#endif
1383 return tmpbuf->ptr;
1384}
1385
1386/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1248static void 1387static int
1249eio__realpath (eio_req *req, etp_worker *self) 1388eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1250{ 1389{
1251 char *rel = req->ptr1; 1390 const char *rel = path;
1252 char *res; 1391 char *res;
1253 char *tmp1, *tmp2; 1392 char *tmp1, *tmp2;
1254#if SYMLOOP_MAX > 32 1393#if SYMLOOP_MAX > 32
1255 int symlinks = SYMLOOP_MAX; 1394 int symlinks = SYMLOOP_MAX;
1256#else 1395#else
1257 int symlinks = 32; 1396 int symlinks = 32;
1258#endif 1397#endif
1259 1398
1260 req->result = -1; 1399 /*D*/ /*TODO: wd ignored */
1261 1400
1262 errno = EINVAL; 1401 errno = EINVAL;
1263 if (!rel) 1402 if (!rel)
1264 return; 1403 return -1;
1265 1404
1266 errno = ENOENT; 1405 errno = ENOENT;
1267 if (!*rel) 1406 if (!*rel)
1268 return; 1407 return -1;
1269 1408
1270 if (!req->ptr2) 1409 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1271 {
1272 X_LOCK (wrklock);
1273 req->flags |= EIO_FLAG_PTR2_FREE;
1274 X_UNLOCK (wrklock);
1275 req->ptr2 = malloc (PATH_MAX * 3);
1276
1277 errno = ENOMEM;
1278 if (!req->ptr2)
1279 return;
1280 }
1281
1282 res = req->ptr2;
1283 tmp1 = res + PATH_MAX; 1410 tmp1 = res + PATH_MAX;
1284 tmp2 = tmp1 + PATH_MAX; 1411 tmp2 = tmp1 + PATH_MAX;
1285 1412
1286#if 0 /* disabled, the musl way to do things is just too racy */ 1413#if 0 /* disabled, the musl way to do things is just too racy */
1287#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME) 1414#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1306#endif 1433#endif
1307#endif 1434#endif
1308 1435
1309 if (*rel != '/') 1436 if (*rel != '/')
1310 { 1437 {
1311 if (!getcwd (res, PATH_MAX)) 1438 if (!eio__getwd (tmpbuf, wd))
1312 return; 1439 return -1;
1313 1440
1314 if (res [1]) /* only use if not / */ 1441 if (res [1]) /* only use if not / */
1315 res += strlen (res); 1442 res += strlen (res);
1316 } 1443 }
1317 1444
1318 while (*rel) 1445 while (*rel)
1319 { 1446 {
1320 ssize_t len, linklen; 1447 eio_ssize_t len, linklen;
1321 char *beg = rel; 1448 const char *beg = rel;
1322 1449
1323 while (*rel && *rel != '/') 1450 while (*rel && *rel != '/')
1324 ++rel; 1451 ++rel;
1325 1452
1326 len = rel - beg; 1453 len = rel - beg;
1338 1465
1339 if (beg [1] == '.' && len == 2) 1466 if (beg [1] == '.' && len == 2)
1340 { 1467 {
1341 /* .. - back up one component, if possible */ 1468 /* .. - back up one component, if possible */
1342 1469
1343 while (res != req->ptr2) 1470 while (res != tmpbuf->ptr)
1344 if (*--res == '/') 1471 if (*--res == '/')
1345 break; 1472 break;
1346 1473
1347 continue; 1474 continue;
1348 } 1475 }
1358 1485
1359 /* zero-terminate, for readlink */ 1486 /* zero-terminate, for readlink */
1360 res [len + 1] = 0; 1487 res [len + 1] = 0;
1361 1488
1362 /* now check if it's a symlink */ 1489 /* now check if it's a symlink */
1363 linklen = readlink (req->ptr2, tmp1, PATH_MAX); 1490 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1364 1491
1365 if (linklen < 0) 1492 if (linklen < 0)
1366 { 1493 {
1367 if (errno != EINVAL) 1494 if (errno != EINVAL)
1368 return; 1495 return -1;
1369 1496
1370 /* it's a normal directory. hopefully */ 1497 /* it's a normal directory. hopefully */
1371 res += len + 1; 1498 res += len + 1;
1372 } 1499 }
1373 else 1500 else
1375 /* yay, it was a symlink - build new path in tmp2 */ 1502 /* yay, it was a symlink - build new path in tmp2 */
1376 int rellen = strlen (rel); 1503 int rellen = strlen (rel);
1377 1504
1378 errno = ENAMETOOLONG; 1505 errno = ENAMETOOLONG;
1379 if (linklen + 1 + rellen >= PATH_MAX) 1506 if (linklen + 1 + rellen >= PATH_MAX)
1380 return; 1507 return -1;
1381 1508
1382 errno = ELOOP; 1509 errno = ELOOP;
1383 if (!--symlinks) 1510 if (!--symlinks)
1384 return; 1511 return -1;
1385 1512
1386 if (*tmp1 == '/') 1513 if (*tmp1 == '/')
1387 res = req->ptr2; /* symlink resolves to an absolute path */ 1514 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1388 1515
1389 /* we need to be careful, as rel might point into tmp2 already */ 1516 /* we need to be careful, as rel might point into tmp2 already */
1390 memmove (tmp2 + linklen + 1, rel, rellen + 1); 1517 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1391 tmp2 [linklen] = '/'; 1518 tmp2 [linklen] = '/';
1392 memcpy (tmp2, tmp1, linklen); 1519 memcpy (tmp2, tmp1, linklen);
1394 rel = tmp2; 1521 rel = tmp2;
1395 } 1522 }
1396 } 1523 }
1397 1524
1398 /* special case for the lone root path */ 1525 /* special case for the lone root path */
1399 if (res == req->ptr2) 1526 if (res == tmpbuf->ptr)
1400 *res++ = '/'; 1527 *res++ = '/';
1401 1528
1402 req->result = res - (char *)req->ptr2; 1529 return res - (char *)tmpbuf->ptr;
1403
1404done:
1405 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1406} 1530}
1407 1531
1408static signed char 1532static signed char
1409eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1533eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1410{ 1534{
1418 1542
1419#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1543#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1420#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1544#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1421 1545
1422static void 1546static void
1423eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1547eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1424{ 1548{
1425 unsigned char bits [9 + sizeof (ino_t) * 8]; 1549 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1426 unsigned char *bit = bits; 1550 unsigned char *bit = bits;
1427 1551
1428 assert (CHAR_BIT == 8); 1552 assert (CHAR_BIT == 8);
1429 assert (sizeof (eio_dirent) * 8 < 256); 1553 assert (sizeof (eio_dirent) * 8 < 256);
1430 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1554 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1432 1556
1433 if (size <= EIO_SORT_FAST) 1557 if (size <= EIO_SORT_FAST)
1434 return; 1558 return;
1435 1559
1436 /* first prepare an array of bits to test in our radix sort */ 1560 /* first prepare an array of bits to test in our radix sort */
1437 /* try to take endianness into account, as well as differences in ino_t sizes */ 1561 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1438 /* inode_bits must contain all inodes ORed together */ 1562 /* inode_bits must contain all inodes ORed together */
1439 /* which is used to skip bits that are 0 everywhere, which is very common */ 1563 /* which is used to skip bits that are 0 everywhere, which is very common */
1440 { 1564 {
1441 ino_t endianness; 1565 eio_ino_t endianness;
1442 int i, j; 1566 int i, j;
1443 1567
1444 /* we store the byte offset of byte n into byte n of "endianness" */ 1568 /* we store the byte offset of byte n into byte n of "endianness" */
1445 for (i = 0; i < sizeof (ino_t); ++i) 1569 for (i = 0; i < sizeof (eio_ino_t); ++i)
1446 ((unsigned char *)&endianness)[i] = i; 1570 ((unsigned char *)&endianness)[i] = i;
1447 1571
1448 *bit++ = 0; 1572 *bit++ = 0;
1449 1573
1450 for (i = 0; i < sizeof (ino_t); ++i) 1574 for (i = 0; i < sizeof (eio_ino_t); ++i)
1451 { 1575 {
1452 /* shifting off the byte offsets out of "endianness" */ 1576 /* shifting off the byte offsets out of "endianness" */
1453 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1577 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1454 endianness >>= 8; 1578 endianness >>= 8;
1455 1579
1456 for (j = 0; j < 8; ++j) 1580 for (j = 0; j < 8; ++j)
1457 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1581 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1458 *bit++ = offs + j; 1582 *bit++ = offs + j;
1459 } 1583 }
1460 1584
1461 for (j = 0; j < 8; ++j) 1585 for (j = 0; j < 8; ++j)
1462 if (score_bits & (1 << j)) 1586 if (score_bits & (1 << j))
1463 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1587 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1464 } 1588 }
1465 1589
1466 /* now actually do the sorting (a variant of MSD radix sort) */ 1590 /* now actually do the sorting (a variant of MSD radix sort) */
1467 { 1591 {
1468 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1592 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1469 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1593 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1470 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1594 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1471 int stk_idx = 0; 1595 int stk_idx = 0;
1472 1596
1473 base_stk [stk_idx] = dents; 1597 base_stk [stk_idx] = dents;
1474 end_stk [stk_idx] = dents + size; 1598 end_stk [stk_idx] = dents + size;
1475 bit_stk [stk_idx] = bit - 1; 1599 bit_stk [stk_idx] = bit - 1;
1554 } 1678 }
1555 } 1679 }
1556} 1680}
1557 1681
1558static void 1682static void
1559eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1683eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1560{ 1684{
1561 if (size <= 1) 1685 if (size <= 1)
1562 return; /* our insertion sort relies on size > 0 */ 1686 return; /* our insertion sort relies on size > 0 */
1563 1687
1564 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1688 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1570 eio_dent_insertion_sort (dents, size); 1694 eio_dent_insertion_sort (dents, size);
1571} 1695}
1572 1696
1573/* read a full directory */ 1697/* read a full directory */
1574static void 1698static void
1575eio__scandir (eio_req *req, etp_worker *self) 1699eio__scandir (eio_req *req)
1576{ 1700{
1577 DIR *dirp;
1578 EIO_STRUCT_DIRENT *entp;
1579 char *name, *names; 1701 char *name, *names;
1580 int namesalloc = 4096; 1702 int namesalloc = 4096 - sizeof (void *) * 4;
1581 int namesoffs = 0; 1703 int namesoffs = 0;
1582 int flags = req->int1; 1704 int flags = req->int1;
1583 eio_dirent *dents = 0; 1705 eio_dirent *dents = 0;
1584 int dentalloc = 128; 1706 int dentalloc = 128;
1585 int dentoffs = 0; 1707 int dentoffs = 0;
1586 ino_t inode_bits = 0; 1708 eio_ino_t inode_bits = 0;
1709#ifdef _WIN32
1710 HANDLE dirp;
1711 WIN32_FIND_DATA entp;
1712#else
1713 DIR *dirp;
1714 EIO_STRUCT_DIRENT *entp;
1715#endif
1587 1716
1588 req->result = -1; 1717 req->result = -1;
1589 1718
1590 if (!(flags & EIO_READDIR_DENTS)) 1719 if (!(flags & EIO_READDIR_DENTS))
1591 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1720 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1592 1721
1593 X_LOCK (wrklock); 1722#ifdef _WIN32
1594 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1723 {
1724 int len = strlen ((const char *)req->ptr1);
1725 char *path = malloc (MAX_PATH);
1726 const char *fmt;
1727
1728 if (!len)
1729 fmt = "./*";
1730 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
1731 fmt = "%s*";
1732 else
1733 fmt = "%s/*";
1734
1735 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
1736 dirp = FindFirstFile (path, &entp);
1737 free (path);
1738
1739 if (dirp == INVALID_HANDLE_VALUE)
1740 {
1741 /* should steal _dosmaperr */
1742 switch (GetLastError ())
1743 {
1744 case ERROR_FILE_NOT_FOUND:
1745 req->result = 0;
1746 break;
1747
1748 case ERROR_INVALID_NAME:
1749 case ERROR_PATH_NOT_FOUND:
1750 case ERROR_NO_MORE_FILES:
1751 errno = ENOENT;
1752 break;
1753
1754 case ERROR_NOT_ENOUGH_MEMORY:
1755 errno = ENOMEM;
1756 break;
1757
1758 default:
1759 errno = EINVAL;
1760 break;
1761 }
1762
1763 return;
1764 }
1765 }
1766#else
1595 self->dirp = dirp = opendir (req->ptr1); 1767 dirp = opendir (req->ptr1);
1768
1769 if (!dirp)
1770 return;
1771#endif
1596 1772
1597 if (req->flags & EIO_FLAG_PTR1_FREE) 1773 if (req->flags & EIO_FLAG_PTR1_FREE)
1598 free (req->ptr1); 1774 free (req->ptr1);
1599 1775
1600 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1776 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1601 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1777 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1602 req->ptr2 = names = malloc (namesalloc); 1778 req->ptr2 = names = malloc (namesalloc);
1603 X_UNLOCK (wrklock);
1604 1779
1605 if (dirp && names && (!flags || dents)) 1780 if (!names || (flags && !dents))
1781 return;
1782
1606 for (;;) 1783 for (;;)
1607 { 1784 {
1785 int done;
1786
1787#ifdef _WIN32
1788 done = !dirp;
1789#else
1608 errno = 0; 1790 errno = 0;
1609 entp = readdir (dirp); 1791 entp = readdir (dirp);
1792 done = !entp;
1793#endif
1610 1794
1611 if (!entp) 1795 if (done)
1612 { 1796 {
1797#ifndef _WIN32
1798 int old_errno = errno;
1799 closedir (dirp);
1800 errno = old_errno;
1801
1613 if (errno) 1802 if (errno)
1614 break; 1803 break;
1804#endif
1615 1805
1616 /* sort etc. */ 1806 /* sort etc. */
1617 req->int1 = flags; 1807 req->int1 = flags;
1618 req->result = dentoffs; 1808 req->result = dentoffs;
1619 1809
1620 if (flags & EIO_READDIR_STAT_ORDER) 1810 if (flags & EIO_READDIR_STAT_ORDER)
1621 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); 1811 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1622 else if (flags & EIO_READDIR_DIRS_FIRST) 1812 else if (flags & EIO_READDIR_DIRS_FIRST)
1623 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1813 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1624 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1814 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1625 else 1815 else
1816 {
1817 /* in this case, all is known, and we just put dirs first and sort them */
1818 eio_dirent *oth = dents + dentoffs;
1819 eio_dirent *dir = dents;
1820
1821 /* now partition dirs to the front, and non-dirs to the back */
1822 /* by walking from both sides and swapping if necessary */
1823 while (oth > dir)
1824 {
1825 if (dir->type == EIO_DT_DIR)
1826 ++dir;
1827 else if ((--oth)->type == EIO_DT_DIR)
1828 {
1829 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1830
1831 ++dir;
1832 }
1833 }
1834
1835 /* now sort the dirs only (dirs all have the same score) */
1836 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1837 }
1838
1839 break;
1840 }
1841
1842 /* now add the entry to our list(s) */
1843 name = D_NAME (entp);
1844
1845 /* skip . and .. entries */
1846 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1847 {
1848 int len = D_NAMLEN (entp) + 1;
1849
1850 while (ecb_expect_false (namesoffs + len > namesalloc))
1851 {
1852 namesalloc *= 2;
1853 req->ptr2 = names = realloc (names, namesalloc);
1854
1855 if (!names)
1856 break;
1857 }
1858
1859 memcpy (names + namesoffs, name, len);
1860
1861 if (dents)
1862 {
1863 struct eio_dirent *ent;
1864
1865 if (ecb_expect_false (dentoffs == dentalloc))
1626 { 1866 {
1627 /* in this case, all is known, and we just put dirs first and sort them */ 1867 dentalloc *= 2;
1868 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1869
1870 if (!dents)
1871 break;
1872 }
1873
1628 eio_dirent *oth = dents + dentoffs; 1874 ent = dents + dentoffs;
1629 eio_dirent *dir = dents;
1630 1875
1631 /* now partition dirs to the front, and non-dirs to the back */ 1876 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1632 /* by walking from both sides and swapping if necessary */ 1877 ent->namelen = len - 1;
1633 while (oth > dir) 1878 ent->inode = D_INO (entp);
1879
1880 inode_bits |= ent->inode;
1881
1882 switch (D_TYPE (entp))
1883 {
1884 default:
1885 ent->type = EIO_DT_UNKNOWN;
1886 flags |= EIO_READDIR_FOUND_UNKNOWN;
1887 break;
1888
1889 #ifdef DT_FIFO
1890 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1891 #endif
1892 #ifdef DT_CHR
1893 case DT_CHR: ent->type = EIO_DT_CHR; break;
1894 #endif
1895 #ifdef DT_MPC
1896 case DT_MPC: ent->type = EIO_DT_MPC; break;
1897 #endif
1898 #ifdef DT_DIR
1899 case DT_DIR: ent->type = EIO_DT_DIR; break;
1900 #endif
1901 #ifdef DT_NAM
1902 case DT_NAM: ent->type = EIO_DT_NAM; break;
1903 #endif
1904 #ifdef DT_BLK
1905 case DT_BLK: ent->type = EIO_DT_BLK; break;
1906 #endif
1907 #ifdef DT_MPB
1908 case DT_MPB: ent->type = EIO_DT_MPB; break;
1909 #endif
1910 #ifdef DT_REG
1911 case DT_REG: ent->type = EIO_DT_REG; break;
1912 #endif
1913 #ifdef DT_NWK
1914 case DT_NWK: ent->type = EIO_DT_NWK; break;
1915 #endif
1916 #ifdef DT_CMP
1917 case DT_CMP: ent->type = EIO_DT_CMP; break;
1918 #endif
1919 #ifdef DT_LNK
1920 case DT_LNK: ent->type = EIO_DT_LNK; break;
1921 #endif
1922 #ifdef DT_SOCK
1923 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1924 #endif
1925 #ifdef DT_DOOR
1926 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1927 #endif
1928 #ifdef DT_WHT
1929 case DT_WHT: ent->type = EIO_DT_WHT; break;
1930 #endif
1931 }
1932
1933 ent->score = 7;
1934
1935 if (flags & EIO_READDIR_DIRS_FIRST)
1936 {
1937 if (ent->type == EIO_DT_UNKNOWN)
1634 { 1938 {
1635 if (dir->type == EIO_DT_DIR) 1939 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1636 ++dir; 1940 ent->score = 1;
1637 else if ((--oth)->type == EIO_DT_DIR) 1941 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1638 { 1942 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1639 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1640
1641 ++dir;
1642 }
1643 } 1943 }
1644 1944 else if (ent->type == EIO_DT_DIR)
1645 /* now sort the dirs only (dirs all have the same score) */ 1945 ent->score = 0;
1646 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1647 } 1946 }
1648
1649 break;
1650 }
1651
1652 /* now add the entry to our list(s) */
1653 name = entp->d_name;
1654
1655 /* skip . and .. entries */
1656 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1657 {
1658 int len = D_NAMLEN (entp) + 1;
1659
1660 while (ecb_expect_false (namesoffs + len > namesalloc))
1661 {
1662 namesalloc *= 2;
1663 X_LOCK (wrklock);
1664 req->ptr2 = names = realloc (names, namesalloc);
1665 X_UNLOCK (wrklock);
1666
1667 if (!names)
1668 break;
1669 } 1947 }
1670 1948
1671 memcpy (names + namesoffs, name, len);
1672
1673 if (dents)
1674 {
1675 struct eio_dirent *ent;
1676
1677 if (ecb_expect_false (dentoffs == dentalloc))
1678 {
1679 dentalloc *= 2;
1680 X_LOCK (wrklock);
1681 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1682 X_UNLOCK (wrklock);
1683
1684 if (!dents)
1685 break;
1686 }
1687
1688 ent = dents + dentoffs;
1689
1690 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1691 ent->namelen = len - 1;
1692 ent->inode = D_INO (entp);
1693
1694 inode_bits |= ent->inode;
1695
1696 switch (D_TYPE (entp))
1697 {
1698 default:
1699 ent->type = EIO_DT_UNKNOWN;
1700 flags |= EIO_READDIR_FOUND_UNKNOWN;
1701 break;
1702
1703 #ifdef DT_FIFO
1704 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1705 #endif
1706 #ifdef DT_CHR
1707 case DT_CHR: ent->type = EIO_DT_CHR; break;
1708 #endif
1709 #ifdef DT_MPC
1710 case DT_MPC: ent->type = EIO_DT_MPC; break;
1711 #endif
1712 #ifdef DT_DIR
1713 case DT_DIR: ent->type = EIO_DT_DIR; break;
1714 #endif
1715 #ifdef DT_NAM
1716 case DT_NAM: ent->type = EIO_DT_NAM; break;
1717 #endif
1718 #ifdef DT_BLK
1719 case DT_BLK: ent->type = EIO_DT_BLK; break;
1720 #endif
1721 #ifdef DT_MPB
1722 case DT_MPB: ent->type = EIO_DT_MPB; break;
1723 #endif
1724 #ifdef DT_REG
1725 case DT_REG: ent->type = EIO_DT_REG; break;
1726 #endif
1727 #ifdef DT_NWK
1728 case DT_NWK: ent->type = EIO_DT_NWK; break;
1729 #endif
1730 #ifdef DT_CMP
1731 case DT_CMP: ent->type = EIO_DT_CMP; break;
1732 #endif
1733 #ifdef DT_LNK
1734 case DT_LNK: ent->type = EIO_DT_LNK; break;
1735 #endif
1736 #ifdef DT_SOCK
1737 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1738 #endif
1739 #ifdef DT_DOOR
1740 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1741 #endif
1742 #ifdef DT_WHT
1743 case DT_WHT: ent->type = EIO_DT_WHT; break;
1744 #endif
1745 }
1746
1747 ent->score = 7;
1748
1749 if (flags & EIO_READDIR_DIRS_FIRST)
1750 {
1751 if (ent->type == EIO_DT_UNKNOWN)
1752 {
1753 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1754 ent->score = 1;
1755 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1756 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1757 }
1758 else if (ent->type == EIO_DT_DIR)
1759 ent->score = 0;
1760 }
1761 }
1762
1763 namesoffs += len; 1949 namesoffs += len;
1764 ++dentoffs; 1950 ++dentoffs;
1765 } 1951 }
1766 1952
1767 if (EIO_CANCELLED (req)) 1953 if (EIO_CANCELLED (req))
1768 { 1954 {
1769 errno = ECANCELED; 1955 errno = ECANCELED;
1770 break; 1956 break;
1771 } 1957 }
1958
1959#ifdef _WIN32
1960 if (!FindNextFile (dirp, &entp))
1961 {
1962 FindClose (dirp);
1963 dirp = 0;
1772 } 1964 }
1965#endif
1966 }
1773} 1967}
1968
1969/*****************************************************************************/
1970/* working directory stuff */
1971
1972#if HAVE_AT
1973
1974#define WD2FD(wd) ((wd) ? ((int)(long)(wd)) - 1 : AT_FDCWD)
1975
1976#ifndef O_SEARCH
1977# define O_SEARCH O_RDONLY
1978#endif
1979
1980eio_wd
1981eio_wd_open_sync (eio_wd wd, const char *path)
1982{
1983 int fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1984
1985 return fd >= 0 ? (eio_wd)(long)(fd + 1) : EIO_INVALID_WD;
1986}
1987
1988static eio_wd
1989eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1990{
1991 return eio_wd_open_sync (wd, path);
1992}
1993
1994void
1995eio_wd_close_sync (eio_wd wd)
1996{
1997 int fd = WD2FD (wd);
1998
1999 if (fd >= 0)
2000 close (fd);
2001}
2002
2003static int
2004eio__truncateat (int dirfd, const char *path, off_t length)
2005{
2006 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
2007 int res;
2008
2009 if (fd < 0)
2010 return fd;
2011
2012 res = ftruncate (fd, length);
2013 close (fd);
2014 return res;
2015}
2016
2017static int
2018eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
2019{
2020 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2021 int res;
2022
2023 if (fd < 0)
2024 return fd;
2025
2026 res = fstatvfs (fd, buf);
2027 close (fd);
2028 return res;
2029
2030}
2031
2032#else
2033
2034/* on legacy systems, we represent the working directories simply by their path strings */
2035
2036static const char *
2037wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2038{
2039 if (!wd || *path == '/')
2040 return path;
2041
2042 {
2043 int l1 = strlen ((const char *)wd);
2044 int l2 = strlen (path);
2045
2046 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2047
2048 memcpy (res, wd, l1);
2049 res [l1] = '/';
2050 memcpy (res + l1 + 1, path, l2 + 1);
2051
2052 return res;
2053 }
2054}
2055
2056static eio_wd
2057eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2058{
2059 if (*path == '/') /* absolute paths ignore wd */
2060 path = strdup (path);
2061 else if (path [0] == '.' && !path [1]) /* special case '.', as it is common */
2062 return wd;
2063 else
2064 {
2065 int len = eio__realpath (tmpbuf, wd, path);
2066
2067 path = EIO_INVALID_WD;
2068
2069 if (len >= 0)
2070 {
2071 ((char *)tmpbuf->ptr)[len] = 0;
2072 path = strdup (tmpbuf->ptr);
2073 }
2074 }
2075
2076 if (!path)
2077 path = EIO_INVALID_WD;
2078
2079 return (eio_wd)path;
2080}
2081
2082eio_wd
2083eio_wd_open_sync (eio_wd wd, const char *path)
2084{
2085 struct tmpbuf tmpbuf = { 0 };
2086 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2087 free (tmpbuf.ptr);
2088
2089 return wd;
2090}
2091
2092void
2093eio_wd_close_sync (eio_wd wd)
2094{
2095 if (wd != EIO_INVALID_WD)
2096 free (wd);
2097}
2098
2099#endif
1774 2100
1775/*****************************************************************************/ 2101/*****************************************************************************/
1776 2102
1777#define ALLOC(len) \ 2103#define ALLOC(len) \
1778 if (!req->ptr2) \ 2104 if (!req->ptr2) \
1793{ 2119{
1794 ETP_REQ *req; 2120 ETP_REQ *req;
1795 struct timespec ts; 2121 struct timespec ts;
1796 etp_worker *self = (etp_worker *)thr_arg; 2122 etp_worker *self = (etp_worker *)thr_arg;
1797 2123
2124#if HAVE_PRCTL_SET_NAME
2125 prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
2126#endif
2127
1798 /* try to distribute timeouts somewhat randomly */ 2128 /* try to distribute timeouts somewhat evenly */
1799 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2129 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1800 2130
1801 for (;;) 2131 for (;;)
1802 { 2132 {
2133 ts.tv_sec = 0;
2134
1803 X_LOCK (reqlock); 2135 X_LOCK (reqlock);
1804 2136
1805 for (;;) 2137 for (;;)
1806 { 2138 {
1807 self->req = req = reqq_shift (&req_queue); 2139 req = reqq_shift (&req_queue);
1808 2140
1809 if (req) 2141 if (req)
1810 break; 2142 break;
1811 2143
2144 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2145 {
2146 X_UNLOCK (reqlock);
2147 X_LOCK (wrklock);
2148 --started;
2149 X_UNLOCK (wrklock);
2150 goto quit;
2151 }
2152
1812 ++idle; 2153 ++idle;
1813 2154
1814 ts.tv_sec = time (0) + idle_timeout; 2155 if (idle <= max_idle)
1815 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 2156 /* we are allowed to idle, so do so without any timeout */
2157 X_COND_WAIT (reqwait, reqlock);
2158 else
1816 { 2159 {
1817 if (idle > max_idle) 2160 /* initialise timeout once */
1818 { 2161 if (!ts.tv_sec)
1819 --idle; 2162 ts.tv_sec = time (0) + idle_timeout;
1820 X_UNLOCK (reqlock);
1821 X_LOCK (wrklock);
1822 --started;
1823 X_UNLOCK (wrklock);
1824 goto quit;
1825 }
1826 2163
1827 /* we are allowed to idle, so do so without any timeout */
1828 X_COND_WAIT (reqwait, reqlock); 2164 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
2165 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1829 } 2166 }
1830 2167
1831 --idle; 2168 --idle;
1832 } 2169 }
1833 2170
1845 ++npending; 2182 ++npending;
1846 2183
1847 if (!reqq_push (&res_queue, req) && want_poll_cb) 2184 if (!reqq_push (&res_queue, req) && want_poll_cb)
1848 want_poll_cb (); 2185 want_poll_cb ();
1849 2186
1850 self->req = 0;
1851 etp_worker_clear (self); 2187 etp_worker_clear (self);
1852 2188
1853 X_UNLOCK (reslock); 2189 X_UNLOCK (reslock);
1854 } 2190 }
1855 2191
1856quit: 2192quit:
2193 free (req);
2194
1857 X_LOCK (wrklock); 2195 X_LOCK (wrklock);
1858 etp_worker_free (self); 2196 etp_worker_free (self);
1859 X_UNLOCK (wrklock); 2197 X_UNLOCK (wrklock);
1860 2198
1861 return 0; 2199 return 0;
1864/*****************************************************************************/ 2202/*****************************************************************************/
1865 2203
1866int ecb_cold 2204int ecb_cold
1867eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2205eio_init (void (*want_poll)(void), void (*done_poll)(void))
1868{ 2206{
2207#if !HAVE_PREADWRITE
2208 X_MUTEX_CREATE (preadwritelock);
2209#endif
2210
1869 return etp_init (want_poll, done_poll); 2211 return etp_init (want_poll, done_poll);
1870} 2212}
1871 2213
1872ecb_inline void 2214ecb_inline void
1873eio_api_destroy (eio_req *req) 2215eio_api_destroy (eio_req *req)
1900 } 2242 }
1901 2243
1902static void 2244static void
1903eio_execute (etp_worker *self, eio_req *req) 2245eio_execute (etp_worker *self, eio_req *req)
1904{ 2246{
2247#if HAVE_AT
2248 int dirfd;
2249#else
2250 const char *path;
2251#endif
2252
1905 if (ecb_expect_false (EIO_CANCELLED (req))) 2253 if (ecb_expect_false (EIO_CANCELLED (req)))
1906 { 2254 {
1907 req->result = -1; 2255 req->result = -1;
1908 req->errorno = ECANCELED; 2256 req->errorno = ECANCELED;
1909 return; 2257 return;
1910 } 2258 }
1911 2259
2260 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2261 {
2262 req->result = -1;
2263 req->errorno = ENOENT;
2264 return;
2265 }
2266
2267 if (req->type >= EIO_OPEN)
2268 {
2269 #if HAVE_AT
2270 dirfd = WD2FD (req->wd);
2271 #else
2272 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2273 #endif
2274 }
2275
1912 switch (req->type) 2276 switch (req->type)
1913 { 2277 {
2278 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2279 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2280 break;
2281 case EIO_WD_CLOSE: req->result = 0;
2282 eio_wd_close_sync (req->wd); break;
2283
1914 case EIO_READ: ALLOC (req->size); 2284 case EIO_READ: ALLOC (req->size);
1915 req->result = req->offs >= 0 2285 req->result = req->offs >= 0
1916 ? pread (req->int1, req->ptr2, req->size, req->offs) 2286 ? pread (req->int1, req->ptr2, req->size, req->offs)
1917 : read (req->int1, req->ptr2, req->size); break; 2287 : read (req->int1, req->ptr2, req->size); break;
1918 case EIO_WRITE: req->result = req->offs >= 0 2288 case EIO_WRITE: req->result = req->offs >= 0
1919 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2289 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1920 : write (req->int1, req->ptr2, req->size); break; 2290 : write (req->int1, req->ptr2, req->size); break;
1921 2291
1922 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2292 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1923 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2293 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2294
2295#if HAVE_AT
1924 2296
1925 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2297 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1926 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2298 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1927 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2299 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2300 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
2301 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
2302 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
2303 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
2304 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
2305
2306 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
2307 case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
2308 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
2309 case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD (req->int3), req->ptr2); break;
2310 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD (req->int3), req->ptr2, 0); break;
2311 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
2312 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2313 case EIO_READLINK: ALLOC (PATH_MAX);
2314 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
2315 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2316 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2317 case EIO_UTIME:
2318 case EIO_FUTIME:
2319 {
2320 struct timespec ts[2];
2321 struct timespec *times;
2322
2323 if (req->nv1 != -1. || req->nv2 != -1.)
2324 {
2325 ts[0].tv_sec = req->nv1;
2326 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2327 ts[1].tv_sec = req->nv2;
2328 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2329
2330 times = ts;
2331 }
2332 else
2333 times = 0;
2334
2335 req->result = req->type == EIO_FUTIME
2336 ? futimens (req->int1, times)
2337 : utimensat (dirfd, req->ptr1, times, 0);
2338 }
2339 break;
2340
2341#else
2342
2343 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2344 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2345 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1928 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2346 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2347 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2348 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2349 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2350 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2351
2352 case EIO_UNLINK: req->result = unlink (path ); break;
2353 case EIO_RMDIR: req->result = rmdir (path ); break;
2354 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
2355 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2356 case EIO_LINK: req->result = link (path , req->ptr2); break;
2357 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2358 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2359 case EIO_READLINK: ALLOC (PATH_MAX);
2360 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2361 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2362 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2363
2364 case EIO_UTIME:
2365 case EIO_FUTIME:
2366 {
2367 struct timeval tv[2];
2368 struct timeval *times;
2369
2370 if (req->nv1 != -1. || req->nv2 != -1.)
2371 {
2372 tv[0].tv_sec = req->nv1;
2373 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
2374 tv[1].tv_sec = req->nv2;
2375 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
2376
2377 times = tv;
2378 }
2379 else
2380 times = 0;
2381
2382 req->result = req->type == EIO_FUTIME
2383 ? futimes (req->int1, times)
2384 : utimes (req->ptr1, times);
2385 }
2386 break;
2387
2388#endif
2389
2390 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2391 {
2392 ALLOC (req->result);
2393 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2394 }
2395 break;
2396
1929 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2397 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1930 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2398 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1931 2399
1932 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1933 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1934 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 2400 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1935 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 2401 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1936 2402
1937 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1938 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2403 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1939 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1940 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2404 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1941 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1942 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 2405 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1943 2406
1944 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1945 case EIO_CLOSE: req->result = close (req->int1); break; 2407 case EIO_CLOSE: req->result = close (req->int1); break;
1946 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 2408 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1947 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1948 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1949 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1950 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1951 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1952 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1953 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1954
1955 case EIO_REALPATH: eio__realpath (req, self); break;
1956
1957 case EIO_READLINK: ALLOC (PATH_MAX);
1958 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1959
1960 case EIO_SYNC: req->result = 0; sync (); break; 2409 case EIO_SYNC: req->result = 0; sync (); break;
1961 case EIO_FSYNC: req->result = fsync (req->int1); break; 2410 case EIO_FSYNC: req->result = fsync (req->int1); break;
1962 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2411 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2412 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2413 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1963 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 2414 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1964 case EIO_MTOUCH: req->result = eio__mtouch (req); break; 2415 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1965 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 2416 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1966 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 2417 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1967 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 2418 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1968 2419
1969 case EIO_READDIR: eio__scandir (req, self); break; 2420 case EIO_READDIR: eio__scandir (req); break;
1970 2421
1971 case EIO_BUSY: 2422 case EIO_BUSY:
1972#ifdef _WIN32 2423#ifdef _WIN32
1973 Sleep (req->nv1 * 1e3); 2424 Sleep (req->nv1 * 1e3);
1974#else 2425#else
1981 req->result = select (0, 0, 0, 0, &tv); 2432 req->result = select (0, 0, 0, 0, &tv);
1982 } 2433 }
1983#endif 2434#endif
1984 break; 2435 break;
1985 2436
1986 case EIO_UTIME:
1987 case EIO_FUTIME:
1988 {
1989 struct timeval tv[2];
1990 struct timeval *times;
1991
1992 if (req->nv1 != -1. || req->nv2 != -1.)
1993 {
1994 tv[0].tv_sec = req->nv1;
1995 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1996 tv[1].tv_sec = req->nv2;
1997 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1998
1999 times = tv;
2000 }
2001 else
2002 times = 0;
2003
2004 req->result = req->type == EIO_FUTIME
2005 ? futimes (req->int1, times)
2006 : utimes (req->ptr1, times);
2007 }
2008 break;
2009
2010 case EIO_GROUP: 2437 case EIO_GROUP:
2011 abort (); /* handled in eio_request */ 2438 abort (); /* handled in eio_request */
2012 2439
2013 case EIO_NOP: 2440 case EIO_NOP:
2014 req->result = 0; 2441 req->result = 0;
2052eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2479eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2053{ 2480{
2054 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2481 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2055} 2482}
2056 2483
2484eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2485{
2486 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2487}
2488
2489eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2490{
2491 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2492}
2493
2494eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2495{
2496 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2497}
2498
2057eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2499eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2058{ 2500{
2059 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2501 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2060} 2502}
2061 2503
2067eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) 2509eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2068{ 2510{
2069 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2511 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2070} 2512}
2071 2513
2072eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2514eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2073{ 2515{
2074 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2516 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2075}
2076
2077eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2078{
2079 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2080} 2517}
2081 2518
2082eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2519eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2083{ 2520{
2084 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2521 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2122eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2559eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
2123{ 2560{
2124 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2561 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
2125} 2562}
2126 2563
2127eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2564eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2128{ 2565{
2129 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2566 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2130} 2567}
2131 2568
2132eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2569eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
2152eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2589eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
2153{ 2590{
2154 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2591 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
2155} 2592}
2156 2593
2157eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2594eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2158{ 2595{
2159 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2596 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2160} 2597}
2161 2598
2162eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2599eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2309} 2746}
2310 2747
2311/*****************************************************************************/ 2748/*****************************************************************************/
2312/* misc garbage */ 2749/* misc garbage */
2313 2750
2314ssize_t 2751eio_ssize_t
2315eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2752eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2316{ 2753{
2317 etp_worker wrk;
2318 ssize_t ret;
2319
2320 wrk.dbuf = 0;
2321
2322 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2754 return eio__sendfile (ofd, ifd, offset, count);
2323
2324 if (wrk.dbuf)
2325 free (wrk.dbuf);
2326
2327 return ret;
2328} 2755}
2329 2756

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines