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

Comparing libeio/eio.c (file contents):
Revision 1.98 by root, Sun Jul 24 05:53:34 2011 UTC vs.
Revision 1.126 by root, Fri Dec 28 07:33:41 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 *
10 * 1. Redistributions of source code must retain the above copyright notice, 10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer. 11 * this list of conditions and the following disclaimer.
12 * 12 *
13 * 2. Redistributions in binary form must reproduce the above copyright 13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the 14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution. 15 * documentation and/or other materials provided with the distribution.
16 * 16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER- 18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO 19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE- 20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
… …
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>
… …
105#ifdef _WIN32 105#ifdef _WIN32
106 106
107 #undef PAGESIZE 107 #undef PAGESIZE
108 #define PAGESIZE 4096 /* GetSystemInfo? */ 108 #define PAGESIZE 4096 /* GetSystemInfo? */
109 109
110 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
111
110 #ifdef EIO_STRUCT_STATI64 112 #ifdef EIO_STRUCT_STATI64
113 /* look at perl's non-sloppy stat */
111 #define stat(path,buf) _stati64 (path,buf) 114 #define stat(path,buf) _stati64 (path,buf)
112 #define fstat(fd,buf) _fstati64 (fd,buf) 115 #define fstat(fd,buf) _fstati64 (fd,buf)
113 #endif 116 #endif
114 #define lstat(path,buf) stat (path,buf) 117 #define lstat(path,buf) stat (path,buf)
115 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1)) 118 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
… …
129 #define sync() EIO_ENOSYS () 132 #define sync() EIO_ENOSYS ()
130 #define readlink(path,buf,s) EIO_ENOSYS () 133 #define readlink(path,buf,s) EIO_ENOSYS ()
131 #define statvfs(path,buf) EIO_ENOSYS () 134 #define statvfs(path,buf) EIO_ENOSYS ()
132 #define fstatvfs(fd,buf) EIO_ENOSYS () 135 #define fstatvfs(fd,buf) EIO_ENOSYS ()
133 136
137 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
138 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
139
140 #if __GNUC__
141 typedef long long eio_off_t; /* signed for compatibility to msvc */
142 #else
143 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
144 #endif
145
146 static eio_ssize_t
147 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
148 {
149 OVERLAPPED o = { 0 };
150 DWORD got;
151
152 o.Offset = offset;
153 o.OffsetHigh = offset >> 32;
154
155 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
156 ? got : -1;
157 }
158
159 static eio_ssize_t
160 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
161 {
162 OVERLAPPED o = { 0 };
163 DWORD got;
164
165 o.Offset = offset;
166 o.OffsetHigh = offset >> 32;
167
168 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
169 ? got : -1;
170 }
171
172 /* rename() uses MoveFile, which fails to overwrite */
173 #define rename(old,neu) eio__rename (old, neu)
174
175 static int
176 eio__rename (const char *old, const char *neu)
177 {
178 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
179 return 0;
180
181 /* should steal _dosmaperr */
182 switch (GetLastError ())
183 {
184 case ERROR_FILE_NOT_FOUND:
185 case ERROR_PATH_NOT_FOUND:
186 case ERROR_INVALID_DRIVE:
187 case ERROR_NO_MORE_FILES:
188 case ERROR_BAD_NETPATH:
189 case ERROR_BAD_NET_NAME:
190 case ERROR_BAD_PATHNAME:
191 case ERROR_FILENAME_EXCED_RANGE:
192 errno = ENOENT;
193 break;
194
195 default:
196 errno = EACCES;
197 break;
198 }
199
200 return -1;
201 }
202
134 /* we could even stat and see if it exists */ 203 /* we could even stat and see if it exists */
135 static int 204 static int
136 symlink (const char *old, const char *neu) 205 symlink (const char *old, const char *neu)
137 { 206 {
138 #if WINVER >= 0x0600 207 #if WINVER >= 0x0600
… …
174 #endif 243 #endif
175 244
176 #define D_NAME(entp) entp->d_name 245 #define D_NAME(entp) entp->d_name
177 246
178 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ 247 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
179 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ 248 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
180 #define _DIRENT_HAVE_D_TYPE /* sigh */ 249 #define _DIRENT_HAVE_D_TYPE /* sigh */
181 #define D_INO(de) (de)->d_fileno 250 #define D_INO(de) (de)->d_fileno
182 #define D_NAMLEN(de) (de)->d_namlen 251 #define D_NAMLEN(de) (de)->d_namlen
183 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 252 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
184 #define D_INO(de) (de)->d_ino 253 #define D_INO(de) (de)->d_ino
… …
247#endif 316#endif
248 317
249/* buffer size for various temporary buffers */ 318/* buffer size for various temporary buffers */
250#define EIO_BUFSIZE 65536 319#define EIO_BUFSIZE 65536
251 320
252#define dBUF \ 321#define dBUF \
253 char *eio_buf = malloc (EIO_BUFSIZE); \ 322 char *eio_buf = malloc (EIO_BUFSIZE); \
254 errno = ENOMEM; \ 323 errno = ENOMEM; \
255 if (!eio_buf) \ 324 if (!eio_buf) \
256 return -1 325 return -1
257 326
258#define FUBd \ 327#define FUBd \
259 free (eio_buf) 328 free (eio_buf)
260 329
261#define EIO_TICKS ((1000000 + 1023) >> 10) 330#define EIO_TICKS ((1000000 + 1023) >> 10)
331
332/*****************************************************************************/
333
334struct tmpbuf
335{
336 void *ptr;
337 int len;
338};
339
340static void *
341tmpbuf_get (struct tmpbuf *buf, int len)
342{
343 if (buf->len < len)
344 {
345 free (buf->ptr);
346 buf->ptr = malloc (buf->len = len);
347 }
348
349 return buf->ptr;
350}
351
352struct tmpbuf;
353
354#if _POSIX_VERSION >= 200809L
355 #define HAVE_AT 1
356 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
357 #ifndef O_SEARCH
358 #define O_SEARCH O_RDONLY
359 #endif
360#else
361 #define HAVE_AT 0
362 static const char *wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path);
363#endif
364
365struct eio_pwd
366{
367#if HAVE_AT
368 int fd;
369#endif
370 int len;
371 char str[1]; /* actually, a 0-terminated canonical path */
372};
373
374/*****************************************************************************/
262 375
263#define ETP_PRI_MIN EIO_PRI_MIN 376#define ETP_PRI_MIN EIO_PRI_MIN
264#define ETP_PRI_MAX EIO_PRI_MAX 377#define ETP_PRI_MAX EIO_PRI_MAX
265 378
266struct etp_worker; 379struct etp_worker;
… …
290static void (*done_poll_cb) (void); 403static void (*done_poll_cb) (void);
291 404
292static unsigned int max_poll_time; /* reslock */ 405static unsigned int max_poll_time; /* reslock */
293static unsigned int max_poll_reqs; /* reslock */ 406static unsigned int max_poll_reqs; /* reslock */
294 407
295static volatile unsigned int nreqs; /* reqlock */ 408static unsigned int nreqs; /* reqlock */
296static volatile unsigned int nready; /* reqlock */ 409static unsigned int nready; /* reqlock */
297static volatile unsigned int npending; /* reqlock */ 410static unsigned int npending; /* reqlock */
298static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */ 411static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
299static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */ 412static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
300 413
301static xmutex_t wrklock; 414static xmutex_t wrklock;
302static xmutex_t reslock; 415static xmutex_t reslock;
303static xmutex_t reqlock; 416static xmutex_t reqlock;
304static xcond_t reqwait; 417static xcond_t reqwait;
305 418
306#if !HAVE_PREADWRITE
307/*
308 * make our pread/pwrite emulation safe against themselves, but not against
309 * normal read/write by using a mutex. slows down execution a lot,
310 * but that's your problem, not mine.
311 */
312static xmutex_t preadwritelock;
313#endif
314
315typedef struct etp_worker 419typedef struct etp_worker
316{ 420{
421 struct tmpbuf tmpbuf;
422
317 /* locked by wrklock */ 423 /* locked by wrklock */
318 struct etp_worker *prev, *next; 424 struct etp_worker *prev, *next;
319 425
320 xthread_t tid; 426 xthread_t tid;
321 427
322 /* locked by reslock, reqlock or wrklock */
323 ETP_REQ *req; /* currently processed request */
324
325#ifdef ETP_WORKER_COMMON 428#ifdef ETP_WORKER_COMMON
326 ETP_WORKER_COMMON 429 ETP_WORKER_COMMON
327#endif 430#endif
328} etp_worker; 431} etp_worker;
329 432
… …
332#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 435#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
333#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 436#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
334 437
335/* worker threads management */ 438/* worker threads management */
336 439
337static void ecb_cold 440static void
338etp_worker_clear (etp_worker *wrk) 441etp_worker_clear (etp_worker *wrk)
339{ 442{
340} 443}
341 444
342static void ecb_cold 445static void ecb_cold
343etp_worker_free (etp_worker *wrk) 446etp_worker_free (etp_worker *wrk)
344{ 447{
448 free (wrk->tmpbuf.ptr);
449
345 wrk->next->prev = wrk->prev; 450 wrk->next->prev = wrk->prev;
346 wrk->prev->next = wrk->next; 451 wrk->prev->next = wrk->next;
347 452
348 free (wrk); 453 free (wrk);
349} 454}
… …
497 /*TODO*/ 602 /*TODO*/
498 assert (("unable to allocate worker thread data", wrk)); 603 assert (("unable to allocate worker thread data", wrk));
499 604
500 X_LOCK (wrklock); 605 X_LOCK (wrklock);
501 606
502 if (thread_create (&wrk->tid, etp_proc, (void *)wrk)) 607 if (xthread_create (&wrk->tid, etp_proc, (void *)wrk))
503 { 608 {
504 wrk->prev = &wrk_first; 609 wrk->prev = &wrk_first;
505 wrk->next = wrk_first.next; 610 wrk->next = wrk_first.next;
506 wrk_first.next->prev = wrk; 611 wrk_first.next->prev = wrk;
507 wrk_first.next = wrk; 612 wrk_first.next = wrk;
… …
527} 632}
528 633
529static void ecb_cold 634static void ecb_cold
530etp_end_thread (void) 635etp_end_thread (void)
531{ 636{
532 eio_req *req = calloc (1, sizeof (eio_req)); 637 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
533 638
534 req->type = -1; 639 req->type = -1;
535 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 640 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
536 641
537 X_LOCK (reqlock); 642 X_LOCK (reqlock);
… …
864} 969}
865 970
866/*****************************************************************************/ 971/*****************************************************************************/
867/* work around various missing functions */ 972/* work around various missing functions */
868 973
869#if !HAVE_PREADWRITE
870# undef pread
871# undef pwrite
872# define pread eio__pread
873# define pwrite eio__pwrite
874
875static eio_ssize_t
876eio__pread (int fd, void *buf, size_t count, off_t offset)
877{
878 eio_ssize_t res;
879 off_t ooffset;
880
881 X_LOCK (preadwritelock);
882 ooffset = lseek (fd, 0, SEEK_CUR);
883 lseek (fd, offset, SEEK_SET);
884 res = read (fd, buf, count);
885 lseek (fd, ooffset, SEEK_SET);
886 X_UNLOCK (preadwritelock);
887
888 return res;
889}
890
891static eio_ssize_t
892eio__pwrite (int fd, void *buf, size_t count, off_t offset)
893{
894 eio_ssize_t res;
895 off_t ooffset;
896
897 X_LOCK (preadwritelock);
898 ooffset = lseek (fd, 0, SEEK_CUR);
899 lseek (fd, offset, SEEK_SET);
900 res = write (fd, buf, count);
901 lseek (fd, ooffset, SEEK_SET);
902 X_UNLOCK (preadwritelock);
903
904 return res;
905}
906#endif
907
908#ifndef HAVE_UTIMES 974#ifndef HAVE_UTIMES
909 975
910# undef utimes 976# undef utimes
911# define utimes(path,times) eio__utimes (path, times) 977# define utimes(path,times) eio__utimes (path, times)
912 978
… …
953 int res; 1019 int res;
954 1020
955#if HAVE_SYS_SYNCFS 1021#if HAVE_SYS_SYNCFS
956 res = (int)syscall (__NR_syncfs, (int)(fd)); 1022 res = (int)syscall (__NR_syncfs, (int)(fd));
957#else 1023#else
958 res = -1; 1024 res = EIO_ENOSYS ();
959 errno = ENOSYS;
960#endif 1025#endif
961 1026
962 if (res < 0 && errno == ENOSYS && fd >= 0) 1027 if (res < 0 && errno == ENOSYS && fd >= 0)
963 sync (); 1028 sync ();
964 1029
… …
994} 1059}
995 1060
996static int 1061static int
997eio__fallocate (int fd, int mode, off_t offset, size_t len) 1062eio__fallocate (int fd, int mode, off_t offset, size_t len)
998{ 1063{
999#if HAVE_FALLOCATE 1064#if HAVE_LINUX_FALLOCATE
1000 return fallocate (fd, mode, offset, len); 1065 return fallocate (fd, mode, offset, len);
1001#else 1066#else
1002 errno = ENOSYS; 1067 return EIO_ENOSYS ();
1003 return -1;
1004#endif 1068#endif
1005} 1069}
1006 1070
1007#if !HAVE_READAHEAD 1071#if !HAVE_READAHEAD
1008# undef readahead 1072# undef readahead
… …
1023 todo -= len; 1087 todo -= len;
1024 } 1088 }
1025 1089
1026 FUBd; 1090 FUBd;
1027 1091
1028 errno = 0; 1092 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
1093 /* but not for e.g. EIO or eof, so we also never fail */
1029 return count; 1094 return 0;
1030} 1095}
1031 1096
1032#endif 1097#endif
1033 1098
1034/* sendfile always needs emulation */ 1099/* sendfile always needs emulation */
… …
1069 1134
1070 /* according to source inspection, this is correct, and useful behaviour */ 1135 /* according to source inspection, this is correct, and useful behaviour */
1071 if (sbytes) 1136 if (sbytes)
1072 res = sbytes; 1137 res = sbytes;
1073 1138
1074# elif defined (__APPLE__) 1139# elif defined __APPLE__
1075 off_t sbytes = count; 1140 off_t sbytes = count;
1076 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 1141 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1077 1142
1078 /* according to the manpage, sbytes is always valid */ 1143 /* according to the manpage, sbytes is always valid */
1079 if (sbytes) 1144 if (sbytes)
… …
1106 HANDLE h = TO_SOCKET (ifd); 1171 HANDLE h = TO_SOCKET (ifd);
1107 SetFilePointer (h, offset, 0, FILE_BEGIN); 1172 SetFilePointer (h, offset, 0, FILE_BEGIN);
1108 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 1173 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1109 1174
1110#else 1175#else
1111 res = -1; 1176 res = EIO_ENOSYS ();
1112 errno = ENOSYS;
1113#endif 1177#endif
1114 1178
1115 /* we assume sendfile can copy at least 128mb in one go */ 1179 /* we assume sendfile can copy at least 128mb in one go */
1116 if (res <= 128 * 1024 * 1024) 1180 if (res <= 128 * 1024 * 1024)
1117 { 1181 {
… …
1304 1368
1305/*****************************************************************************/ 1369/*****************************************************************************/
1306/* requests implemented outside eio_execute, because they are so large */ 1370/* requests implemented outside eio_execute, because they are so large */
1307 1371
1308static void 1372static void
1309eio__realpath (eio_req *req, etp_worker *self) 1373eio__lseek (eio_req *req)
1310{ 1374{
1311 char *rel = req->ptr1; 1375 /* this usually gets optimised away completely, or your compiler sucks, */
1376 /* or the whence constants really are not 0, 1, 2 */
1377 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
1378 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
1379 : req->int2 == EIO_SEEK_END ? SEEK_END
1380 : req->int2;
1381
1382 req->offs = lseek (req->int1, req->offs, whence);
1383 req->result = req->offs == (off_t)-1 ? -1 : 0;
1384}
1385
1386/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1387static int
1388eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1389{
1390 const char *rel = path;
1312 char *res; 1391 char *res;
1313 char *tmp1, *tmp2; 1392 char *tmp1, *tmp2;
1314#if SYMLOOP_MAX > 32 1393#if SYMLOOP_MAX > 32
1315 int symlinks = SYMLOOP_MAX; 1394 int symlinks = SYMLOOP_MAX;
1316#else 1395#else
1317 int symlinks = 32; 1396 int symlinks = 32;
1318#endif 1397#endif
1319 1398
1320 req->result = -1;
1321
1322 errno = EINVAL; 1399 errno = EINVAL;
1323 if (!rel) 1400 if (!rel)
1324 return; 1401 return -1;
1325 1402
1326 errno = ENOENT; 1403 errno = ENOENT;
1327 if (!*rel) 1404 if (!*rel)
1328 return; 1405 return -1;
1329 1406
1330 if (!req->ptr2) 1407 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1331 {
1332 X_LOCK (wrklock);
1333 req->flags |= EIO_FLAG_PTR2_FREE;
1334 X_UNLOCK (wrklock);
1335 req->ptr2 = malloc (PATH_MAX * 3);
1336
1337 errno = ENOMEM;
1338 if (!req->ptr2)
1339 return;
1340 }
1341
1342 res = req->ptr2;
1343 tmp1 = res + PATH_MAX; 1408 tmp1 = res + PATH_MAX;
1344 tmp2 = tmp1 + PATH_MAX; 1409 tmp2 = tmp1 + PATH_MAX;
1345 1410
1346#if 0 /* disabled, the musl way to do things is just too racy */ 1411#if 0 /* disabled, the musl way to do things is just too racy */
1347#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME) 1412#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
… …
1351 1416
1352 if (fd >= 0) 1417 if (fd >= 0)
1353 { 1418 {
1354 sprintf (tmp1, "/proc/self/fd/%d", fd); 1419 sprintf (tmp1, "/proc/self/fd/%d", fd);
1355 req->result = readlink (tmp1, res, PATH_MAX); 1420 req->result = readlink (tmp1, res, PATH_MAX);
1421 /* here we should probably stat the open file and the disk file, to make sure they still match */
1356 close (fd); 1422 close (fd);
1357
1358 /* here we should probably stat the open file and the disk file, to make sure they still match */
1359 1423
1360 if (req->result > 0) 1424 if (req->result > 0)
1361 goto done; 1425 goto done;
1362 } 1426 }
1363 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO) 1427 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1364 return; 1428 return -1;
1365 } 1429 }
1366#endif 1430#endif
1367#endif 1431#endif
1368 1432
1369 if (*rel != '/') 1433 if (*rel != '/')
1370 { 1434 {
1435 int len;
1436
1437 errno = ENOENT;
1438 if (wd == EIO_INVALID_WD)
1439 return -1;
1440
1441 if (wd == EIO_CWD)
1442 {
1371 if (!getcwd (res, PATH_MAX)) 1443 if (!getcwd (res, PATH_MAX))
1372 return; 1444 return -1;
1445
1446 len = strlen (res);
1447 }
1448 else
1449 memcpy (res, wd->str, len = wd->len);
1373 1450
1374 if (res [1]) /* only use if not / */ 1451 if (res [1]) /* only use if not / */
1375 res += strlen (res); 1452 res += len;
1376 } 1453 }
1377 1454
1378 while (*rel) 1455 while (*rel)
1379 { 1456 {
1380 eio_ssize_t len, linklen; 1457 eio_ssize_t len, linklen;
1381 char *beg = rel; 1458 const char *beg = rel;
1382 1459
1383 while (*rel && *rel != '/') 1460 while (*rel && *rel != '/')
1384 ++rel; 1461 ++rel;
1385 1462
1386 len = rel - beg; 1463 len = rel - beg;
… …
1398 1475
1399 if (beg [1] == '.' && len == 2) 1476 if (beg [1] == '.' && len == 2)
1400 { 1477 {
1401 /* .. - back up one component, if possible */ 1478 /* .. - back up one component, if possible */
1402 1479
1403 while (res != req->ptr2) 1480 while (res != tmpbuf->ptr)
1404 if (*--res == '/') 1481 if (*--res == '/')
1405 break; 1482 break;
1406 1483
1407 continue; 1484 continue;
1408 } 1485 }
1409 } 1486 }
1410 1487
1411 errno = ENAMETOOLONG; 1488 errno = ENAMETOOLONG;
1412 if (res + 1 + len + 1 >= tmp1) 1489 if (res + 1 + len + 1 >= tmp1)
1413 return; 1490 return -1;
1414 1491
1415 /* copy one component */ 1492 /* copy one component */
1416 *res = '/'; 1493 *res = '/';
1417 memcpy (res + 1, beg, len); 1494 memcpy (res + 1, beg, len);
1418 1495
1419 /* zero-terminate, for readlink */ 1496 /* zero-terminate, for readlink */
1420 res [len + 1] = 0; 1497 res [len + 1] = 0;
1421 1498
1422 /* now check if it's a symlink */ 1499 /* now check if it's a symlink */
1423 linklen = readlink (req->ptr2, tmp1, PATH_MAX); 1500 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1424 1501
1425 if (linklen < 0) 1502 if (linklen < 0)
1426 { 1503 {
1427 if (errno != EINVAL) 1504 if (errno != EINVAL)
1428 return; 1505 return -1;
1429 1506
1430 /* it's a normal directory. hopefully */ 1507 /* it's a normal directory. hopefully */
1431 res += len + 1; 1508 res += len + 1;
1432 } 1509 }
1433 else 1510 else
… …
1435 /* yay, it was a symlink - build new path in tmp2 */ 1512 /* yay, it was a symlink - build new path in tmp2 */
1436 int rellen = strlen (rel); 1513 int rellen = strlen (rel);
1437 1514
1438 errno = ENAMETOOLONG; 1515 errno = ENAMETOOLONG;
1439 if (linklen + 1 + rellen >= PATH_MAX) 1516 if (linklen + 1 + rellen >= PATH_MAX)
1440 return; 1517 return -1;
1441 1518
1442 errno = ELOOP; 1519 errno = ELOOP;
1443 if (!--symlinks) 1520 if (!--symlinks)
1444 return; 1521 return -1;
1445 1522
1446 if (*tmp1 == '/') 1523 if (*tmp1 == '/')
1447 res = req->ptr2; /* symlink resolves to an absolute path */ 1524 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1448 1525
1449 /* we need to be careful, as rel might point into tmp2 already */ 1526 /* we need to be careful, as rel might point into tmp2 already */
1450 memmove (tmp2 + linklen + 1, rel, rellen + 1); 1527 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1451 tmp2 [linklen] = '/'; 1528 tmp2 [linklen] = '/';
1452 memcpy (tmp2, tmp1, linklen); 1529 memcpy (tmp2, tmp1, linklen);
… …
1454 rel = tmp2; 1531 rel = tmp2;
1455 } 1532 }
1456 } 1533 }
1457 1534
1458 /* special case for the lone root path */ 1535 /* special case for the lone root path */
1459 if (res == req->ptr2) 1536 if (res == tmpbuf->ptr)
1460 *res++ = '/'; 1537 *res++ = '/';
1461 1538
1462 req->result = res - (char *)req->ptr2; 1539 return res - (char *)tmpbuf->ptr;
1463
1464done:
1465 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1466} 1540}
1467 1541
1468static signed char 1542static signed char
1469eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1543eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1470{ 1544{
… …
1658#ifdef _WIN32 1732#ifdef _WIN32
1659 { 1733 {
1660 int len = strlen ((const char *)req->ptr1); 1734 int len = strlen ((const char *)req->ptr1);
1661 char *path = malloc (MAX_PATH); 1735 char *path = malloc (MAX_PATH);
1662 const char *fmt; 1736 const char *fmt;
1737 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1663 1738
1664 if (!len) 1739 if (!len)
1665 fmt = "./*"; 1740 fmt = "./*";
1666 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\') 1741 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1667 fmt = "%s*"; 1742 fmt = "%s*";
1668 else 1743 else
1669 fmt = "%s/*"; 1744 fmt = "%s/*";
1670 1745
1671 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1); 1746 _snprintf (path, MAX_PATH, fmt, reqpath);
1672 dirp = FindFirstFile (path, &entp); 1747 dirp = FindFirstFile (path, &entp);
1673 free (path); 1748 free (path);
1674 1749
1675 if (dirp == INVALID_HANDLE_VALUE) 1750 if (dirp == INVALID_HANDLE_VALUE)
1676 { 1751 {
1677 dirp = 0; 1752 /* should steal _dosmaperr */
1678
1679 switch (GetLastError ()) 1753 switch (GetLastError ())
1680 { 1754 {
1681 case ERROR_FILE_NOT_FOUND: 1755 case ERROR_FILE_NOT_FOUND:
1682 req->result = 0; 1756 req->result = 0;
1683 break; 1757 break;
1684 1758
1685 case ERROR_INVALID_NAME: 1759 case ERROR_INVALID_NAME:
1686 case ERROR_PATH_NOT_FOUND: 1760 case ERROR_PATH_NOT_FOUND:
1687 case ERROR_NO_MORE_FILES: 1761 case ERROR_NO_MORE_FILES:
1688 errno = ENOENT; 1762 errno = ENOENT;
1689 break; 1763 break;
1690 1764
1691 case ERROR_NOT_ENOUGH_MEMORY: 1765 case ERROR_NOT_ENOUGH_MEMORY:
1692 errno = ENOMEM; 1766 errno = ENOMEM;
1693 break; 1767 break;
1694 1768
1695 default: 1769 default:
1696 errno = EINVAL; 1770 errno = EINVAL;
1697 break; 1771 break;
1698 } 1772 }
1773
1774 return;
1699 } 1775 }
1700 } 1776 }
1701#else 1777#else
1778 #if HAVE_AT
1779 if (req->wd)
1780 {
1781 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1782
1783 if (fd < 0)
1784 return;
1785
1786 dirp = fdopendir (fd);
1787
1788 if (!dirp)
1789 close (fd);
1790 }
1791 else
1702 dirp = opendir (req->ptr1); 1792 dirp = opendir (req->ptr1);
1793 #else
1794 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1795 #endif
1796
1797 if (!dirp)
1798 return;
1703#endif 1799#endif
1704 1800
1705 if (req->flags & EIO_FLAG_PTR1_FREE) 1801 if (req->flags & EIO_FLAG_PTR1_FREE)
1706 free (req->ptr1); 1802 free (req->ptr1);
1707 1803
1708 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1804 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1709 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1805 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1710 req->ptr2 = names = malloc (namesalloc); 1806 req->ptr2 = names = malloc (namesalloc);
1711 1807
1712 if (dirp && names && (!flags || dents)) 1808 if (!names || (flags && !dents))
1809 return;
1810
1713 for (;;) 1811 for (;;)
1714 { 1812 {
1715 int done; 1813 int done;
1716 1814
1717#ifdef _WIN32 1815#ifdef _WIN32
1718 done = !dirp; 1816 done = !dirp;
1719#else 1817#else
1720 errno = 0; 1818 errno = 0;
1721 entp = readdir (dirp); 1819 entp = readdir (dirp);
1722 done = !entp; 1820 done = !entp;
1723#endif 1821#endif
1724 1822
1725 if (done) 1823 if (done)
1726 { 1824 {
1727#ifndef _WIN32 1825#ifndef _WIN32
1728 int old_errno = errno; 1826 int old_errno = errno;
1729 closedir (dirp); 1827 closedir (dirp);
1730 errno = old_errno; 1828 errno = old_errno;
1731 1829
1732 if (errno) 1830 if (errno)
1733 break; 1831 break;
1734#endif 1832#endif
1735 1833
1736 /* sort etc. */ 1834 /* sort etc. */
1737 req->int1 = flags; 1835 req->int1 = flags;
1738 req->result = dentoffs; 1836 req->result = dentoffs;
1739 1837
1740 if (flags & EIO_READDIR_STAT_ORDER) 1838 if (flags & EIO_READDIR_STAT_ORDER)
1741 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); 1839 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1742 else if (flags & EIO_READDIR_DIRS_FIRST) 1840 else if (flags & EIO_READDIR_DIRS_FIRST)
1743 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1841 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1744 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1842 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1745 else 1843 else
1844 {
1845 /* in this case, all is known, and we just put dirs first and sort them */
1846 eio_dirent *oth = dents + dentoffs;
1847 eio_dirent *dir = dents;
1848
1849 /* now partition dirs to the front, and non-dirs to the back */
1850 /* by walking from both sides and swapping if necessary */
1851 while (oth > dir)
1852 {
1853 if (dir->type == EIO_DT_DIR)
1854 ++dir;
1855 else if ((--oth)->type == EIO_DT_DIR)
1856 {
1857 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1858
1859 ++dir;
1860 }
1861 }
1862
1863 /* now sort the dirs only (dirs all have the same score) */
1864 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1865 }
1866
1867 break;
1868 }
1869
1870 /* now add the entry to our list(s) */
1871 name = D_NAME (entp);
1872
1873 /* skip . and .. entries */
1874 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1875 {
1876 int len = D_NAMLEN (entp) + 1;
1877
1878 while (ecb_expect_false (namesoffs + len > namesalloc))
1879 {
1880 namesalloc *= 2;
1881 req->ptr2 = names = realloc (names, namesalloc);
1882
1883 if (!names)
1884 break;
1885 }
1886
1887 memcpy (names + namesoffs, name, len);
1888
1889 if (dents)
1890 {
1891 struct eio_dirent *ent;
1892
1893 if (ecb_expect_false (dentoffs == dentalloc))
1746 { 1894 {
1747 /* in this case, all is known, and we just put dirs first and sort them */ 1895 dentalloc *= 2;
1896 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1897
1898 if (!dents)
1899 break;
1900 }
1901
1748 eio_dirent *oth = dents + dentoffs; 1902 ent = dents + dentoffs;
1749 eio_dirent *dir = dents;
1750 1903
1751 /* now partition dirs to the front, and non-dirs to the back */ 1904 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1752 /* by walking from both sides and swapping if necessary */ 1905 ent->namelen = len - 1;
1753 while (oth > dir) 1906 ent->inode = D_INO (entp);
1907
1908 inode_bits |= ent->inode;
1909
1910 switch (D_TYPE (entp))
1911 {
1912 default:
1913 ent->type = EIO_DT_UNKNOWN;
1914 flags |= EIO_READDIR_FOUND_UNKNOWN;
1915 break;
1916
1917 #ifdef DT_FIFO
1918 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1919 #endif
1920 #ifdef DT_CHR
1921 case DT_CHR: ent->type = EIO_DT_CHR; break;
1922 #endif
1923 #ifdef DT_MPC
1924 case DT_MPC: ent->type = EIO_DT_MPC; break;
1925 #endif
1926 #ifdef DT_DIR
1927 case DT_DIR: ent->type = EIO_DT_DIR; break;
1928 #endif
1929 #ifdef DT_NAM
1930 case DT_NAM: ent->type = EIO_DT_NAM; break;
1931 #endif
1932 #ifdef DT_BLK
1933 case DT_BLK: ent->type = EIO_DT_BLK; break;
1934 #endif
1935 #ifdef DT_MPB
1936 case DT_MPB: ent->type = EIO_DT_MPB; break;
1937 #endif
1938 #ifdef DT_REG
1939 case DT_REG: ent->type = EIO_DT_REG; break;
1940 #endif
1941 #ifdef DT_NWK
1942 case DT_NWK: ent->type = EIO_DT_NWK; break;
1943 #endif
1944 #ifdef DT_CMP
1945 case DT_CMP: ent->type = EIO_DT_CMP; break;
1946 #endif
1947 #ifdef DT_LNK
1948 case DT_LNK: ent->type = EIO_DT_LNK; break;
1949 #endif
1950 #ifdef DT_SOCK
1951 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1952 #endif
1953 #ifdef DT_DOOR
1954 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1955 #endif
1956 #ifdef DT_WHT
1957 case DT_WHT: ent->type = EIO_DT_WHT; break;
1958 #endif
1959 }
1960
1961 ent->score = 7;
1962
1963 if (flags & EIO_READDIR_DIRS_FIRST)
1964 {
1965 if (ent->type == EIO_DT_UNKNOWN)
1754 { 1966 {
1755 if (dir->type == EIO_DT_DIR) 1967 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1756 ++dir; 1968 ent->score = 1;
1757 else if ((--oth)->type == EIO_DT_DIR) 1969 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1758 { 1970 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1759 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1760
1761 ++dir;
1762 }
1763 } 1971 }
1764 1972 else if (ent->type == EIO_DT_DIR)
1765 /* now sort the dirs only (dirs all have the same score) */ 1973 ent->score = 0;
1766 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1767 } 1974 }
1768
1769 break;
1770 }
1771
1772 /* now add the entry to our list(s) */
1773 name = D_NAME (entp);
1774
1775 /* skip . and .. entries */
1776 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1777 {
1778 int len = D_NAMLEN (entp) + 1;
1779
1780 while (ecb_expect_false (namesoffs + len > namesalloc))
1781 {
1782 namesalloc *= 2;
1783 req->ptr2 = names = realloc (names, namesalloc);
1784
1785 if (!names)
1786 break;
1787 } 1975 }
1788 1976
1789 memcpy (names + namesoffs, name, len);
1790
1791 if (dents)
1792 {
1793 struct eio_dirent *ent;
1794
1795 if (ecb_expect_false (dentoffs == dentalloc))
1796 {
1797 dentalloc *= 2;
1798 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1799
1800 if (!dents)
1801 break;
1802 }
1803
1804 ent = dents + dentoffs;
1805
1806 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1807 ent->namelen = len - 1;
1808 ent->inode = D_INO (entp);
1809
1810 inode_bits |= ent->inode;
1811
1812 switch (D_TYPE (entp))
1813 {
1814 default:
1815 ent->type = EIO_DT_UNKNOWN;
1816 flags |= EIO_READDIR_FOUND_UNKNOWN;
1817 break;
1818
1819 #ifdef DT_FIFO
1820 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1821 #endif
1822 #ifdef DT_CHR
1823 case DT_CHR: ent->type = EIO_DT_CHR; break;
1824 #endif
1825 #ifdef DT_MPC
1826 case DT_MPC: ent->type = EIO_DT_MPC; break;
1827 #endif
1828 #ifdef DT_DIR
1829 case DT_DIR: ent->type = EIO_DT_DIR; break;
1830 #endif
1831 #ifdef DT_NAM
1832 case DT_NAM: ent->type = EIO_DT_NAM; break;
1833 #endif
1834 #ifdef DT_BLK
1835 case DT_BLK: ent->type = EIO_DT_BLK; break;
1836 #endif
1837 #ifdef DT_MPB
1838 case DT_MPB: ent->type = EIO_DT_MPB; break;
1839 #endif
1840 #ifdef DT_REG
1841 case DT_REG: ent->type = EIO_DT_REG; break;
1842 #endif
1843 #ifdef DT_NWK
1844 case DT_NWK: ent->type = EIO_DT_NWK; break;
1845 #endif
1846 #ifdef DT_CMP
1847 case DT_CMP: ent->type = EIO_DT_CMP; break;
1848 #endif
1849 #ifdef DT_LNK
1850 case DT_LNK: ent->type = EIO_DT_LNK; break;
1851 #endif
1852 #ifdef DT_SOCK
1853 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1854 #endif
1855 #ifdef DT_DOOR
1856 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1857 #endif
1858 #ifdef DT_WHT
1859 case DT_WHT: ent->type = EIO_DT_WHT; break;
1860 #endif
1861 }
1862
1863 ent->score = 7;
1864
1865 if (flags & EIO_READDIR_DIRS_FIRST)
1866 {
1867 if (ent->type == EIO_DT_UNKNOWN)
1868 {
1869 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1870 ent->score = 1;
1871 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1872 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1873 }
1874 else if (ent->type == EIO_DT_DIR)
1875 ent->score = 0;
1876 }
1877 }
1878
1879 namesoffs += len; 1977 namesoffs += len;
1880 ++dentoffs; 1978 ++dentoffs;
1881 } 1979 }
1882 1980
1883 if (EIO_CANCELLED (req)) 1981 if (EIO_CANCELLED (req))
1884 { 1982 {
1885 errno = ECANCELED; 1983 errno = ECANCELED;
1886 break; 1984 break;
1887 } 1985 }
1888 1986
1889#ifdef _WIN32 1987#ifdef _WIN32
1890 if (!FindNextFile (dirp, &entp)) 1988 if (!FindNextFile (dirp, &entp))
1891 { 1989 {
1892 FindClose (dirp); 1990 FindClose (dirp);
1893 dirp = 0; 1991 dirp = 0;
1894 } 1992 }
1895#endif 1993#endif
1896 } 1994 }
1897} 1995}
1996
1997/*****************************************************************************/
1998/* working directory stuff */
1999/* various deficiencies in the posix 2008 api force us to */
2000/* keep the absolute path in string form at all times */
2001/* fuck yeah. */
2002
2003#if !HAVE_AT
2004
2005/* a bit like realpath, but usually faster because it doesn'T have to return */
2006/* an absolute or canonical path */
2007static const char *
2008wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2009{
2010 if (!wd || *path == '/')
2011 return path;
2012
2013 if (path [0] == '.' && !path [1])
2014 return wd->str;
2015
2016 {
2017 int l1 = wd->len;
2018 int l2 = strlen (path);
2019
2020 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2021
2022 memcpy (res, wd->str, l1);
2023 res [l1] = '/';
2024 memcpy (res + l1 + 1, path, l2 + 1);
2025
2026 return res;
2027 }
2028}
2029
2030#endif
2031
2032static eio_wd
2033eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2034{
2035 int fd;
2036 eio_wd res;
2037 int len = eio__realpath (tmpbuf, wd, path);
2038
2039 if (len < 0)
2040 return EIO_INVALID_WD;
2041
2042#if HAVE_AT
2043 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
2044
2045 if (fd < 0)
2046 return EIO_INVALID_WD;
2047#endif
2048
2049 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
2050
2051#if HAVE_AT
2052 res->fd = fd;
2053#endif
2054
2055 res->len = len;
2056 memcpy (res->str, tmpbuf->ptr, len);
2057 res->str [len] = 0;
2058
2059 return res;
2060}
2061
2062eio_wd
2063eio_wd_open_sync (eio_wd wd, const char *path)
2064{
2065 struct tmpbuf tmpbuf = { 0 };
2066 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2067 free (tmpbuf.ptr);
2068
2069 return wd;
2070}
2071
2072void
2073eio_wd_close_sync (eio_wd wd)
2074{
2075 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
2076 {
2077 #if HAVE_AT
2078 close (wd->fd);
2079 #endif
2080 free (wd);
2081 }
2082}
2083
2084#if HAVE_AT
2085
2086/* they forgot these */
2087
2088static int
2089eio__truncateat (int dirfd, const char *path, off_t length)
2090{
2091 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
2092 int res;
2093
2094 if (fd < 0)
2095 return fd;
2096
2097 res = ftruncate (fd, length);
2098 close (fd);
2099 return res;
2100}
2101
2102static int
2103eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
2104{
2105 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2106 int res;
2107
2108 if (fd < 0)
2109 return fd;
2110
2111 res = fstatvfs (fd, buf);
2112 close (fd);
2113 return res;
2114
2115}
2116
2117#endif
1898 2118
1899/*****************************************************************************/ 2119/*****************************************************************************/
1900 2120
1901#define ALLOC(len) \ 2121#define ALLOC(len) \
1902 if (!req->ptr2) \ 2122 if (!req->ptr2) \
… …
1911 req->result = -1; \ 2131 req->result = -1; \
1912 break; \ 2132 break; \
1913 } \ 2133 } \
1914 } 2134 }
1915 2135
2136static void ecb_noinline ecb_cold
2137etp_proc_init (void)
2138{
2139#if HAVE_PRCTL_SET_NAME
2140 /* provide a more sensible "thread name" */
2141 char name[16 + 1];
2142 const int namelen = sizeof (name) - 1;
2143 int len;
2144
2145 prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
2146 name [namelen] = 0;
2147 len = strlen (name);
2148 strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
2149 prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
2150#endif
2151}
2152
1916X_THREAD_PROC (etp_proc) 2153X_THREAD_PROC (etp_proc)
1917{ 2154{
1918 ETP_REQ *req; 2155 ETP_REQ *req;
1919 struct timespec ts; 2156 struct timespec ts;
1920 etp_worker *self = (etp_worker *)thr_arg; 2157 etp_worker *self = (etp_worker *)thr_arg;
1921 2158
1922#if HAVE_PRCTL_SET_NAME 2159 etp_proc_init ();
1923 prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
1924#endif
1925 2160
1926 /* try to distribute timeouts somewhat evenly */ 2161 /* try to distribute timeouts somewhat evenly */
1927 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2162 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1928 2163
1929 for (;;) 2164 for (;;)
… …
1932 2167
1933 X_LOCK (reqlock); 2168 X_LOCK (reqlock);
1934 2169
1935 for (;;) 2170 for (;;)
1936 { 2171 {
1937 self->req = req = reqq_shift (&req_queue); 2172 req = reqq_shift (&req_queue);
1938 2173
1939 if (req) 2174 if (req)
1940 break; 2175 break;
1941 2176
1942 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */ 2177 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
… …
1980 ++npending; 2215 ++npending;
1981 2216
1982 if (!reqq_push (&res_queue, req) && want_poll_cb) 2217 if (!reqq_push (&res_queue, req) && want_poll_cb)
1983 want_poll_cb (); 2218 want_poll_cb ();
1984 2219
1985 self->req = 0;
1986 etp_worker_clear (self); 2220 etp_worker_clear (self);
1987 2221
1988 X_UNLOCK (reslock); 2222 X_UNLOCK (reslock);
1989 } 2223 }
1990 2224
1991quit: 2225quit:
2226 free (req);
2227
1992 X_LOCK (wrklock); 2228 X_LOCK (wrklock);
1993 etp_worker_free (self); 2229 etp_worker_free (self);
1994 X_UNLOCK (wrklock); 2230 X_UNLOCK (wrklock);
1995 2231
1996 return 0; 2232 return 0;
… …
1999/*****************************************************************************/ 2235/*****************************************************************************/
2000 2236
2001int ecb_cold 2237int ecb_cold
2002eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2238eio_init (void (*want_poll)(void), void (*done_poll)(void))
2003{ 2239{
2004#if !HAVE_PREADWRITE
2005 X_MUTEX_CREATE (preadwritelock);
2006#endif
2007
2008 return etp_init (want_poll, done_poll); 2240 return etp_init (want_poll, done_poll);
2009} 2241}
2010 2242
2011ecb_inline void 2243ecb_inline void
2012eio_api_destroy (eio_req *req) 2244eio_api_destroy (eio_req *req)
2013{ 2245{
2014 free (req); 2246 free (req);
2015} 2247}
2016 2248
2017#define REQ(rtype) \ 2249#define REQ(rtype) \
2018 eio_req *req; \ 2250 eio_req *req; \
2019 \ 2251 \
2020 req = (eio_req *)calloc (1, sizeof *req); \ 2252 req = (eio_req *)calloc (1, sizeof *req); \
2021 if (!req) \ 2253 if (!req) \
2022 return 0; \ 2254 return 0; \
… …
2036 { \ 2268 { \
2037 eio_api_destroy (req); \ 2269 eio_api_destroy (req); \
2038 return 0; \ 2270 return 0; \
2039 } 2271 }
2040 2272
2273#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
2274
2041static void 2275static void
2042eio_execute (etp_worker *self, eio_req *req) 2276eio_execute (etp_worker *self, eio_req *req)
2043{ 2277{
2278#if HAVE_AT
2279 int dirfd;
2280#else
2281 const char *path;
2282#endif
2283
2044 if (ecb_expect_false (EIO_CANCELLED (req))) 2284 if (ecb_expect_false (EIO_CANCELLED (req)))
2045 { 2285 {
2046 req->result = -1; 2286 req->result = -1;
2047 req->errorno = ECANCELED; 2287 req->errorno = ECANCELED;
2048 return; 2288 return;
2049 } 2289 }
2050 2290
2291 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2292 {
2293 req->result = -1;
2294 req->errorno = ENOENT;
2295 return;
2296 }
2297
2298 if (req->type >= EIO_OPEN)
2299 {
2300 #if HAVE_AT
2301 dirfd = WD2FD (req->wd);
2302 #else
2303 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2304 #endif
2305 }
2306
2051 switch (req->type) 2307 switch (req->type)
2052 { 2308 {
2309 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2310 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2311 break;
2312 case EIO_WD_CLOSE: req->result = 0;
2313 eio_wd_close_sync (req->wd); break;
2314
2315 case EIO_SEEK: eio__lseek (req); break;
2053 case EIO_READ: ALLOC (req->size); 2316 case EIO_READ: ALLOC (req->size);
2054 req->result = req->offs >= 0 2317 req->result = req->offs >= 0
2055 ? pread (req->int1, req->ptr2, req->size, req->offs) 2318 ? pread (req->int1, req->ptr2, req->size, req->offs)
2056 : read (req->int1, req->ptr2, req->size); break; 2319 : read (req->int1, req->ptr2, req->size); break;
2057 case EIO_WRITE: req->result = req->offs >= 0 2320 case EIO_WRITE: req->result = req->offs >= 0
… …
2059 : write (req->int1, req->ptr2, req->size); break; 2322 : write (req->int1, req->ptr2, req->size); break;
2060 2323
2061 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2324 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
2062 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break; 2325 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2063 2326
2327#if HAVE_AT
2328
2064 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2329 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2065 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2330 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
2066 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2331 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2332 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
2333 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
2334 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
2335 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
2336 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
2337
2338 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
2339 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
2340 req->result = req->wd && SINGLEDOT (req->ptr1)
2341 ? rmdir (req->wd->str)
2342 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
2343 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
2344 case EIO_RENAME: /* complications arise because "." cannot be renamed, so we might have to expand */
2345 req->result = req->wd && SINGLEDOT (req->ptr1)
2346 ? rename (req->wd->str, req->ptr2)
2347 : renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
2348 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
2349 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
2350 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2351 case EIO_READLINK: ALLOC (PATH_MAX);
2352 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
2353 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2354 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2355 case EIO_UTIME:
2356 case EIO_FUTIME:
2357 {
2358 struct timespec ts[2];
2359 struct timespec *times;
2360
2361 if (req->nv1 != -1. || req->nv2 != -1.)
2362 {
2363 ts[0].tv_sec = req->nv1;
2364 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2365 ts[1].tv_sec = req->nv2;
2366 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2367
2368 times = ts;
2369 }
2370 else
2371 times = 0;
2372
2373 req->result = req->type == EIO_FUTIME
2374 ? futimens (req->int1, times)
2375 : utimensat (dirfd, req->ptr1, times, 0);
2376 }
2377 break;
2378
2379#else
2380
2381 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2382 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2383 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2067 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2384 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2385 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2386 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2387 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2388 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2389
2390 case EIO_UNLINK: req->result = unlink (path ); break;
2391 case EIO_RMDIR: req->result = rmdir (path ); break;
2392 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
2393 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2394 case EIO_LINK: req->result = link (path , req->ptr2); break;
2395 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2396 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2397 case EIO_READLINK: ALLOC (PATH_MAX);
2398 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2399 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2400 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2401
2402 case EIO_UTIME:
2403 case EIO_FUTIME:
2404 {
2405 struct timeval tv[2];
2406 struct timeval *times;
2407
2408 if (req->nv1 != -1. || req->nv2 != -1.)
2409 {
2410 tv[0].tv_sec = req->nv1;
2411 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
2412 tv[1].tv_sec = req->nv2;
2413 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
2414
2415 times = tv;
2416 }
2417 else
2418 times = 0;
2419
2420 req->result = req->type == EIO_FUTIME
2421 ? futimes (req->int1, times)
2422 : utimes (req->ptr1, times);
2423 }
2424 break;
2425
2426#endif
2427
2428 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2429 {
2430 ALLOC (req->result);
2431 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2432 }
2433 break;
2434
2068 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2435 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2069 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2436 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2070 2437
2071 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2072 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2073 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 2438 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2074 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 2439 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2075 2440
2076 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
2077 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2441 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2078 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
2079 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2442 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2080 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
2081 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 2443 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2082 2444
2083 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
2084 case EIO_CLOSE: req->result = close (req->int1); break; 2445 case EIO_CLOSE: req->result = close (req->int1); break;
2085 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 2446 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2086 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
2087 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
2088 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
2089 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
2090 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
2091 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
2092 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2093
2094 case EIO_REALPATH: eio__realpath (req, self); break;
2095
2096 case EIO_READLINK: ALLOC (PATH_MAX);
2097 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
2098
2099 case EIO_SYNC: req->result = 0; sync (); break; 2447 case EIO_SYNC: req->result = 0; sync (); break;
2100 case EIO_FSYNC: req->result = fsync (req->int1); break; 2448 case EIO_FSYNC: req->result = fsync (req->int1); break;
2101 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2449 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2102 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break; 2450 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2103 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 2451 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
… …
2122 req->result = select (0, 0, 0, 0, &tv); 2470 req->result = select (0, 0, 0, 0, &tv);
2123 } 2471 }
2124#endif 2472#endif
2125 break; 2473 break;
2126 2474
2127 case EIO_UTIME:
2128 case EIO_FUTIME:
2129 {
2130 struct timeval tv[2];
2131 struct timeval *times;
2132
2133 if (req->nv1 != -1. || req->nv2 != -1.)
2134 {
2135 tv[0].tv_sec = req->nv1;
2136 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
2137 tv[1].tv_sec = req->nv2;
2138 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
2139
2140 times = tv;
2141 }
2142 else
2143 times = 0;
2144
2145 req->result = req->type == EIO_FUTIME
2146 ? futimes (req->int1, times)
2147 : utimes (req->ptr1, times);
2148 }
2149 break;
2150
2151 case EIO_GROUP: 2475 case EIO_GROUP:
2152 abort (); /* handled in eio_request */ 2476 abort (); /* handled in eio_request */
2153 2477
2154 case EIO_NOP: 2478 case EIO_NOP:
2155 req->result = 0; 2479 req->result = 0;
… …
2158 case EIO_CUSTOM: 2482 case EIO_CUSTOM:
2159 req->feed (req); 2483 req->feed (req);
2160 break; 2484 break;
2161 2485
2162 default: 2486 default:
2163 errno = ENOSYS;
2164 req->result = -1; 2487 req->result = EIO_ENOSYS ();
2165 break; 2488 break;
2166 } 2489 }
2167 2490
2168 req->errorno = errno; 2491 req->errorno = errno;
2169} 2492}
2170 2493
2171#ifndef EIO_NO_WRAPPERS 2494#ifndef EIO_NO_WRAPPERS
2172 2495
2496eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2497{
2498 REQ (EIO_WD_OPEN); PATH; SEND;
2499}
2500
2501eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2502{
2503 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2504}
2505
2173eio_req *eio_nop (int pri, eio_cb cb, void *data) 2506eio_req *eio_nop (int pri, eio_cb cb, void *data)
2174{ 2507{
2175 REQ (EIO_NOP); SEND; 2508 REQ (EIO_NOP); SEND;
2176} 2509}
2177 2510
… …
2236} 2569}
2237 2570
2238eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2571eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
2239{ 2572{
2240 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2573 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
2574}
2575
2576eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2577{
2578 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2241} 2579}
2242 2580
2243eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2581eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2244{ 2582{
2245 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2583 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;

Diff Legend

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