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

Comparing libeio/eio.c (file contents):
Revision 1.99 by root, Tue Jul 26 11:07:08 2011 UTC vs.
Revision 1.127 by root, Mon Feb 18 03:15:19 2013 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))
128 #define mknod(path,mode,dev) EIO_ENOSYS () 131 #define mknod(path,mode,dev) EIO_ENOSYS ()
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 ()
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 }
133 171
134 /* rename() uses MoveFile, which fails to overwrite */ 172 /* rename() uses MoveFile, which fails to overwrite */
135 #define rename(old,neu) eio__rename (old, neu) 173 #define rename(old,neu) eio__rename (old, neu)
136 174
137 static int 175 static int
193 231
194#else 232#else
195 233
196 #include <sys/time.h> 234 #include <sys/time.h>
197 #include <sys/select.h> 235 #include <sys/select.h>
198 #include <sys/statvfs.h>
199 #include <unistd.h> 236 #include <unistd.h>
200 #include <signal.h> 237 #include <signal.h>
201 #include <dirent.h> 238 #include <dirent.h>
202 239
240 #if ANDROID
241 #include <sys/vfs.h>
242 #define statvfs statfs
243 #define fstatvfs fstatfs
244 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
245 #else
246 #include <sys/statvfs.h>
247 #endif
248
203 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES 249 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
204 #include <sys/mman.h> 250 #include <sys/mman.h>
205 #endif 251 #endif
206 252
207 #define D_NAME(entp) entp->d_name 253 #define D_NAME(entp) entp->d_name
208 254
209 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ 255 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
210 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ 256 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
211 #define _DIRENT_HAVE_D_TYPE /* sigh */ 257 #define _DIRENT_HAVE_D_TYPE /* sigh */
212 #define D_INO(de) (de)->d_fileno 258 #define D_INO(de) (de)->d_fileno
213 #define D_NAMLEN(de) (de)->d_namlen 259 #define D_NAMLEN(de) (de)->d_namlen
214 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 260 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
215 #define D_INO(de) (de)->d_ino 261 #define D_INO(de) (de)->d_ino
278#endif 324#endif
279 325
280/* buffer size for various temporary buffers */ 326/* buffer size for various temporary buffers */
281#define EIO_BUFSIZE 65536 327#define EIO_BUFSIZE 65536
282 328
283#define dBUF \ 329#define dBUF \
284 char *eio_buf = malloc (EIO_BUFSIZE); \ 330 char *eio_buf = malloc (EIO_BUFSIZE); \
285 errno = ENOMEM; \ 331 errno = ENOMEM; \
286 if (!eio_buf) \ 332 if (!eio_buf) \
287 return -1 333 return -1
288 334
289#define FUBd \ 335#define FUBd \
290 free (eio_buf) 336 free (eio_buf)
291 337
292#define EIO_TICKS ((1000000 + 1023) >> 10) 338#define EIO_TICKS ((1000000 + 1023) >> 10)
339
340/*****************************************************************************/
341
342struct tmpbuf
343{
344 void *ptr;
345 int len;
346};
347
348static void *
349tmpbuf_get (struct tmpbuf *buf, int len)
350{
351 if (buf->len < len)
352 {
353 free (buf->ptr);
354 buf->ptr = malloc (buf->len = len);
355 }
356
357 return buf->ptr;
358}
359
360struct tmpbuf;
361
362#if _POSIX_VERSION >= 200809L
363 #define HAVE_AT 1
364 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
365 #ifndef O_SEARCH
366 #define O_SEARCH O_RDONLY
367 #endif
368#else
369 #define HAVE_AT 0
370 static const char *wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path);
371#endif
372
373struct eio_pwd
374{
375#if HAVE_AT
376 int fd;
377#endif
378 int len;
379 char str[1]; /* actually, a 0-terminated canonical path */
380};
381
382/*****************************************************************************/
293 383
294#define ETP_PRI_MIN EIO_PRI_MIN 384#define ETP_PRI_MIN EIO_PRI_MIN
295#define ETP_PRI_MAX EIO_PRI_MAX 385#define ETP_PRI_MAX EIO_PRI_MAX
296 386
297struct etp_worker; 387struct etp_worker;
321static void (*done_poll_cb) (void); 411static void (*done_poll_cb) (void);
322 412
323static unsigned int max_poll_time; /* reslock */ 413static unsigned int max_poll_time; /* reslock */
324static unsigned int max_poll_reqs; /* reslock */ 414static unsigned int max_poll_reqs; /* reslock */
325 415
326static volatile unsigned int nreqs; /* reqlock */ 416static unsigned int nreqs; /* reqlock */
327static volatile unsigned int nready; /* reqlock */ 417static unsigned int nready; /* reqlock */
328static volatile unsigned int npending; /* reqlock */ 418static unsigned int npending; /* reqlock */
329static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */ 419static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
330static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */ 420static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
331 421
332static xmutex_t wrklock; 422static xmutex_t wrklock;
333static xmutex_t reslock; 423static xmutex_t reslock;
334static xmutex_t reqlock; 424static xmutex_t reqlock;
335static xcond_t reqwait; 425static xcond_t reqwait;
336 426
337#if !HAVE_PREADWRITE
338/*
339 * make our pread/pwrite emulation safe against themselves, but not against
340 * normal read/write by using a mutex. slows down execution a lot,
341 * but that's your problem, not mine.
342 */
343static xmutex_t preadwritelock;
344#endif
345
346typedef struct etp_worker 427typedef struct etp_worker
347{ 428{
429 struct tmpbuf tmpbuf;
430
348 /* locked by wrklock */ 431 /* locked by wrklock */
349 struct etp_worker *prev, *next; 432 struct etp_worker *prev, *next;
350 433
351 xthread_t tid; 434 xthread_t tid;
352 435
353 /* locked by reslock, reqlock or wrklock */
354 ETP_REQ *req; /* currently processed request */
355
356#ifdef ETP_WORKER_COMMON 436#ifdef ETP_WORKER_COMMON
357 ETP_WORKER_COMMON 437 ETP_WORKER_COMMON
358#endif 438#endif
359} etp_worker; 439} etp_worker;
360 440
363#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 443#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
364#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 444#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
365 445
366/* worker threads management */ 446/* worker threads management */
367 447
368static void ecb_cold 448static void
369etp_worker_clear (etp_worker *wrk) 449etp_worker_clear (etp_worker *wrk)
370{ 450{
371} 451}
372 452
373static void ecb_cold 453static void ecb_cold
374etp_worker_free (etp_worker *wrk) 454etp_worker_free (etp_worker *wrk)
375{ 455{
456 free (wrk->tmpbuf.ptr);
457
376 wrk->next->prev = wrk->prev; 458 wrk->next->prev = wrk->prev;
377 wrk->prev->next = wrk->next; 459 wrk->prev->next = wrk->next;
378 460
379 free (wrk); 461 free (wrk);
380} 462}
528 /*TODO*/ 610 /*TODO*/
529 assert (("unable to allocate worker thread data", wrk)); 611 assert (("unable to allocate worker thread data", wrk));
530 612
531 X_LOCK (wrklock); 613 X_LOCK (wrklock);
532 614
533 if (thread_create (&wrk->tid, etp_proc, (void *)wrk)) 615 if (xthread_create (&wrk->tid, etp_proc, (void *)wrk))
534 { 616 {
535 wrk->prev = &wrk_first; 617 wrk->prev = &wrk_first;
536 wrk->next = wrk_first.next; 618 wrk->next = wrk_first.next;
537 wrk_first.next->prev = wrk; 619 wrk_first.next->prev = wrk;
538 wrk_first.next = wrk; 620 wrk_first.next = wrk;
558} 640}
559 641
560static void ecb_cold 642static void ecb_cold
561etp_end_thread (void) 643etp_end_thread (void)
562{ 644{
563 eio_req *req = calloc (1, sizeof (eio_req)); 645 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
564 646
565 req->type = -1; 647 req->type = -1;
566 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 648 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
567 649
568 X_LOCK (reqlock); 650 X_LOCK (reqlock);
895} 977}
896 978
897/*****************************************************************************/ 979/*****************************************************************************/
898/* work around various missing functions */ 980/* work around various missing functions */
899 981
900#if !HAVE_PREADWRITE
901# undef pread
902# undef pwrite
903# define pread eio__pread
904# define pwrite eio__pwrite
905
906static eio_ssize_t
907eio__pread (int fd, void *buf, size_t count, off_t offset)
908{
909 eio_ssize_t res;
910 off_t ooffset;
911
912 X_LOCK (preadwritelock);
913 ooffset = lseek (fd, 0, SEEK_CUR);
914 lseek (fd, offset, SEEK_SET);
915 res = read (fd, buf, count);
916 lseek (fd, ooffset, SEEK_SET);
917 X_UNLOCK (preadwritelock);
918
919 return res;
920}
921
922static eio_ssize_t
923eio__pwrite (int fd, void *buf, size_t count, off_t offset)
924{
925 eio_ssize_t res;
926 off_t ooffset;
927
928 X_LOCK (preadwritelock);
929 ooffset = lseek (fd, 0, SEEK_CUR);
930 lseek (fd, offset, SEEK_SET);
931 res = write (fd, buf, count);
932 lseek (fd, ooffset, SEEK_SET);
933 X_UNLOCK (preadwritelock);
934
935 return res;
936}
937#endif
938
939#ifndef HAVE_UTIMES 982#ifndef HAVE_UTIMES
940 983
941# undef utimes 984# undef utimes
942# define utimes(path,times) eio__utimes (path, times) 985# define utimes(path,times) eio__utimes (path, times)
943 986
984 int res; 1027 int res;
985 1028
986#if HAVE_SYS_SYNCFS 1029#if HAVE_SYS_SYNCFS
987 res = (int)syscall (__NR_syncfs, (int)(fd)); 1030 res = (int)syscall (__NR_syncfs, (int)(fd));
988#else 1031#else
989 res = -1; 1032 res = EIO_ENOSYS ();
990 errno = ENOSYS;
991#endif 1033#endif
992 1034
993 if (res < 0 && errno == ENOSYS && fd >= 0) 1035 if (res < 0 && errno == ENOSYS && fd >= 0)
994 sync (); 1036 sync ();
995 1037
1025} 1067}
1026 1068
1027static int 1069static int
1028eio__fallocate (int fd, int mode, off_t offset, size_t len) 1070eio__fallocate (int fd, int mode, off_t offset, size_t len)
1029{ 1071{
1030#if HAVE_FALLOCATE 1072#if HAVE_LINUX_FALLOCATE
1031 return fallocate (fd, mode, offset, len); 1073 return fallocate (fd, mode, offset, len);
1032#else 1074#else
1033 errno = ENOSYS; 1075 return EIO_ENOSYS ();
1034 return -1;
1035#endif 1076#endif
1036} 1077}
1037 1078
1038#if !HAVE_READAHEAD 1079#if !HAVE_READAHEAD
1039# undef readahead 1080# undef readahead
1054 todo -= len; 1095 todo -= len;
1055 } 1096 }
1056 1097
1057 FUBd; 1098 FUBd;
1058 1099
1059 errno = 0; 1100 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
1101 /* but not for e.g. EIO or eof, so we also never fail */
1060 return count; 1102 return 0;
1061} 1103}
1062 1104
1063#endif 1105#endif
1064 1106
1065/* sendfile always needs emulation */ 1107/* sendfile always needs emulation */
1100 1142
1101 /* according to source inspection, this is correct, and useful behaviour */ 1143 /* according to source inspection, this is correct, and useful behaviour */
1102 if (sbytes) 1144 if (sbytes)
1103 res = sbytes; 1145 res = sbytes;
1104 1146
1105# elif defined (__APPLE__) 1147# elif defined __APPLE__
1106 off_t sbytes = count; 1148 off_t sbytes = count;
1107 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 1149 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1108 1150
1109 /* according to the manpage, sbytes is always valid */ 1151 /* according to the manpage, sbytes is always valid */
1110 if (sbytes) 1152 if (sbytes)
1137 HANDLE h = TO_SOCKET (ifd); 1179 HANDLE h = TO_SOCKET (ifd);
1138 SetFilePointer (h, offset, 0, FILE_BEGIN); 1180 SetFilePointer (h, offset, 0, FILE_BEGIN);
1139 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 1181 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1140 1182
1141#else 1183#else
1142 res = -1; 1184 res = EIO_ENOSYS ();
1143 errno = ENOSYS;
1144#endif 1185#endif
1145 1186
1146 /* we assume sendfile can copy at least 128mb in one go */ 1187 /* we assume sendfile can copy at least 128mb in one go */
1147 if (res <= 128 * 1024 * 1024) 1188 if (res <= 128 * 1024 * 1024)
1148 { 1189 {
1335 1376
1336/*****************************************************************************/ 1377/*****************************************************************************/
1337/* requests implemented outside eio_execute, because they are so large */ 1378/* requests implemented outside eio_execute, because they are so large */
1338 1379
1339static void 1380static void
1340eio__realpath (eio_req *req, etp_worker *self) 1381eio__lseek (eio_req *req)
1341{ 1382{
1342 char *rel = req->ptr1; 1383 /* this usually gets optimised away completely, or your compiler sucks, */
1384 /* or the whence constants really are not 0, 1, 2 */
1385 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
1386 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
1387 : req->int2 == EIO_SEEK_END ? SEEK_END
1388 : req->int2;
1389
1390 req->offs = lseek (req->int1, req->offs, whence);
1391 req->result = req->offs == (off_t)-1 ? -1 : 0;
1392}
1393
1394/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1395static int
1396eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1397{
1398 const char *rel = path;
1343 char *res; 1399 char *res;
1344 char *tmp1, *tmp2; 1400 char *tmp1, *tmp2;
1345#if SYMLOOP_MAX > 32 1401#if SYMLOOP_MAX > 32
1346 int symlinks = SYMLOOP_MAX; 1402 int symlinks = SYMLOOP_MAX;
1347#else 1403#else
1348 int symlinks = 32; 1404 int symlinks = 32;
1349#endif 1405#endif
1350 1406
1351 req->result = -1;
1352
1353 errno = EINVAL; 1407 errno = EINVAL;
1354 if (!rel) 1408 if (!rel)
1355 return; 1409 return -1;
1356 1410
1357 errno = ENOENT; 1411 errno = ENOENT;
1358 if (!*rel) 1412 if (!*rel)
1359 return; 1413 return -1;
1360 1414
1361 if (!req->ptr2) 1415 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1362 {
1363 X_LOCK (wrklock);
1364 req->flags |= EIO_FLAG_PTR2_FREE;
1365 X_UNLOCK (wrklock);
1366 req->ptr2 = malloc (PATH_MAX * 3);
1367
1368 errno = ENOMEM;
1369 if (!req->ptr2)
1370 return;
1371 }
1372
1373 res = req->ptr2;
1374 tmp1 = res + PATH_MAX; 1416 tmp1 = res + PATH_MAX;
1375 tmp2 = tmp1 + PATH_MAX; 1417 tmp2 = tmp1 + PATH_MAX;
1376 1418
1377#if 0 /* disabled, the musl way to do things is just too racy */ 1419#if 0 /* disabled, the musl way to do things is just too racy */
1378#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME) 1420#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1382 1424
1383 if (fd >= 0) 1425 if (fd >= 0)
1384 { 1426 {
1385 sprintf (tmp1, "/proc/self/fd/%d", fd); 1427 sprintf (tmp1, "/proc/self/fd/%d", fd);
1386 req->result = readlink (tmp1, res, PATH_MAX); 1428 req->result = readlink (tmp1, res, PATH_MAX);
1429 /* here we should probably stat the open file and the disk file, to make sure they still match */
1387 close (fd); 1430 close (fd);
1388
1389 /* here we should probably stat the open file and the disk file, to make sure they still match */
1390 1431
1391 if (req->result > 0) 1432 if (req->result > 0)
1392 goto done; 1433 goto done;
1393 } 1434 }
1394 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO) 1435 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1395 return; 1436 return -1;
1396 } 1437 }
1397#endif 1438#endif
1398#endif 1439#endif
1399 1440
1400 if (*rel != '/') 1441 if (*rel != '/')
1401 { 1442 {
1443 int len;
1444
1445 errno = ENOENT;
1446 if (wd == EIO_INVALID_WD)
1447 return -1;
1448
1449 if (wd == EIO_CWD)
1450 {
1402 if (!getcwd (res, PATH_MAX)) 1451 if (!getcwd (res, PATH_MAX))
1403 return; 1452 return -1;
1453
1454 len = strlen (res);
1455 }
1456 else
1457 memcpy (res, wd->str, len = wd->len);
1404 1458
1405 if (res [1]) /* only use if not / */ 1459 if (res [1]) /* only use if not / */
1406 res += strlen (res); 1460 res += len;
1407 } 1461 }
1408 1462
1409 while (*rel) 1463 while (*rel)
1410 { 1464 {
1411 eio_ssize_t len, linklen; 1465 eio_ssize_t len, linklen;
1412 char *beg = rel; 1466 const char *beg = rel;
1413 1467
1414 while (*rel && *rel != '/') 1468 while (*rel && *rel != '/')
1415 ++rel; 1469 ++rel;
1416 1470
1417 len = rel - beg; 1471 len = rel - beg;
1429 1483
1430 if (beg [1] == '.' && len == 2) 1484 if (beg [1] == '.' && len == 2)
1431 { 1485 {
1432 /* .. - back up one component, if possible */ 1486 /* .. - back up one component, if possible */
1433 1487
1434 while (res != req->ptr2) 1488 while (res != tmpbuf->ptr)
1435 if (*--res == '/') 1489 if (*--res == '/')
1436 break; 1490 break;
1437 1491
1438 continue; 1492 continue;
1439 } 1493 }
1440 } 1494 }
1441 1495
1442 errno = ENAMETOOLONG; 1496 errno = ENAMETOOLONG;
1443 if (res + 1 + len + 1 >= tmp1) 1497 if (res + 1 + len + 1 >= tmp1)
1444 return; 1498 return -1;
1445 1499
1446 /* copy one component */ 1500 /* copy one component */
1447 *res = '/'; 1501 *res = '/';
1448 memcpy (res + 1, beg, len); 1502 memcpy (res + 1, beg, len);
1449 1503
1450 /* zero-terminate, for readlink */ 1504 /* zero-terminate, for readlink */
1451 res [len + 1] = 0; 1505 res [len + 1] = 0;
1452 1506
1453 /* now check if it's a symlink */ 1507 /* now check if it's a symlink */
1454 linklen = readlink (req->ptr2, tmp1, PATH_MAX); 1508 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1455 1509
1456 if (linklen < 0) 1510 if (linklen < 0)
1457 { 1511 {
1458 if (errno != EINVAL) 1512 if (errno != EINVAL)
1459 return; 1513 return -1;
1460 1514
1461 /* it's a normal directory. hopefully */ 1515 /* it's a normal directory. hopefully */
1462 res += len + 1; 1516 res += len + 1;
1463 } 1517 }
1464 else 1518 else
1466 /* yay, it was a symlink - build new path in tmp2 */ 1520 /* yay, it was a symlink - build new path in tmp2 */
1467 int rellen = strlen (rel); 1521 int rellen = strlen (rel);
1468 1522
1469 errno = ENAMETOOLONG; 1523 errno = ENAMETOOLONG;
1470 if (linklen + 1 + rellen >= PATH_MAX) 1524 if (linklen + 1 + rellen >= PATH_MAX)
1471 return; 1525 return -1;
1472 1526
1473 errno = ELOOP; 1527 errno = ELOOP;
1474 if (!--symlinks) 1528 if (!--symlinks)
1475 return; 1529 return -1;
1476 1530
1477 if (*tmp1 == '/') 1531 if (*tmp1 == '/')
1478 res = req->ptr2; /* symlink resolves to an absolute path */ 1532 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1479 1533
1480 /* we need to be careful, as rel might point into tmp2 already */ 1534 /* we need to be careful, as rel might point into tmp2 already */
1481 memmove (tmp2 + linklen + 1, rel, rellen + 1); 1535 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1482 tmp2 [linklen] = '/'; 1536 tmp2 [linklen] = '/';
1483 memcpy (tmp2, tmp1, linklen); 1537 memcpy (tmp2, tmp1, linklen);
1485 rel = tmp2; 1539 rel = tmp2;
1486 } 1540 }
1487 } 1541 }
1488 1542
1489 /* special case for the lone root path */ 1543 /* special case for the lone root path */
1490 if (res == req->ptr2) 1544 if (res == tmpbuf->ptr)
1491 *res++ = '/'; 1545 *res++ = '/';
1492 1546
1493 req->result = res - (char *)req->ptr2; 1547 return res - (char *)tmpbuf->ptr;
1494
1495done:
1496 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1497} 1548}
1498 1549
1499static signed char 1550static signed char
1500eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1551eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1501{ 1552{
1689#ifdef _WIN32 1740#ifdef _WIN32
1690 { 1741 {
1691 int len = strlen ((const char *)req->ptr1); 1742 int len = strlen ((const char *)req->ptr1);
1692 char *path = malloc (MAX_PATH); 1743 char *path = malloc (MAX_PATH);
1693 const char *fmt; 1744 const char *fmt;
1745 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1694 1746
1695 if (!len) 1747 if (!len)
1696 fmt = "./*"; 1748 fmt = "./*";
1697 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\') 1749 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1698 fmt = "%s*"; 1750 fmt = "%s*";
1699 else 1751 else
1700 fmt = "%s/*"; 1752 fmt = "%s/*";
1701 1753
1702 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1); 1754 _snprintf (path, MAX_PATH, fmt, reqpath);
1703 dirp = FindFirstFile (path, &entp); 1755 dirp = FindFirstFile (path, &entp);
1704 free (path); 1756 free (path);
1705 1757
1706 if (dirp == INVALID_HANDLE_VALUE) 1758 if (dirp == INVALID_HANDLE_VALUE)
1707 { 1759 {
1708 dirp = 0;
1709
1710 /* should steal _dosmaperr */ 1760 /* should steal _dosmaperr */
1711 switch (GetLastError ()) 1761 switch (GetLastError ())
1712 { 1762 {
1713 case ERROR_FILE_NOT_FOUND: 1763 case ERROR_FILE_NOT_FOUND:
1714 req->result = 0; 1764 req->result = 0;
1715 break; 1765 break;
1716 1766
1717 case ERROR_INVALID_NAME: 1767 case ERROR_INVALID_NAME:
1718 case ERROR_PATH_NOT_FOUND: 1768 case ERROR_PATH_NOT_FOUND:
1719 case ERROR_NO_MORE_FILES: 1769 case ERROR_NO_MORE_FILES:
1720 errno = ENOENT; 1770 errno = ENOENT;
1721 break; 1771 break;
1722 1772
1723 case ERROR_NOT_ENOUGH_MEMORY: 1773 case ERROR_NOT_ENOUGH_MEMORY:
1724 errno = ENOMEM; 1774 errno = ENOMEM;
1725 break; 1775 break;
1726 1776
1727 default: 1777 default:
1728 errno = EINVAL; 1778 errno = EINVAL;
1729 break; 1779 break;
1730 } 1780 }
1781
1782 return;
1731 } 1783 }
1732 } 1784 }
1733#else 1785#else
1786 #if HAVE_AT
1787 if (req->wd)
1788 {
1789 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1790
1791 if (fd < 0)
1792 return;
1793
1794 dirp = fdopendir (fd);
1795
1796 if (!dirp)
1797 close (fd);
1798 }
1799 else
1734 dirp = opendir (req->ptr1); 1800 dirp = opendir (req->ptr1);
1801 #else
1802 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1803 #endif
1804
1805 if (!dirp)
1806 return;
1735#endif 1807#endif
1736 1808
1737 if (req->flags & EIO_FLAG_PTR1_FREE) 1809 if (req->flags & EIO_FLAG_PTR1_FREE)
1738 free (req->ptr1); 1810 free (req->ptr1);
1739 1811
1740 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1812 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1741 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1813 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1742 req->ptr2 = names = malloc (namesalloc); 1814 req->ptr2 = names = malloc (namesalloc);
1743 1815
1744 if (dirp && names && (!flags || dents)) 1816 if (!names || (flags && !dents))
1817 return;
1818
1745 for (;;) 1819 for (;;)
1746 { 1820 {
1747 int done; 1821 int done;
1748 1822
1749#ifdef _WIN32 1823#ifdef _WIN32
1750 done = !dirp; 1824 done = !dirp;
1751#else 1825#else
1752 errno = 0; 1826 errno = 0;
1753 entp = readdir (dirp); 1827 entp = readdir (dirp);
1754 done = !entp; 1828 done = !entp;
1755#endif 1829#endif
1756 1830
1757 if (done) 1831 if (done)
1758 { 1832 {
1759#ifndef _WIN32 1833#ifndef _WIN32
1760 int old_errno = errno; 1834 int old_errno = errno;
1761 closedir (dirp); 1835 closedir (dirp);
1762 errno = old_errno; 1836 errno = old_errno;
1763 1837
1764 if (errno) 1838 if (errno)
1765 break; 1839 break;
1766#endif 1840#endif
1767 1841
1768 /* sort etc. */ 1842 /* sort etc. */
1769 req->int1 = flags; 1843 req->int1 = flags;
1770 req->result = dentoffs; 1844 req->result = dentoffs;
1771 1845
1772 if (flags & EIO_READDIR_STAT_ORDER) 1846 if (flags & EIO_READDIR_STAT_ORDER)
1773 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); 1847 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1774 else if (flags & EIO_READDIR_DIRS_FIRST) 1848 else if (flags & EIO_READDIR_DIRS_FIRST)
1775 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1849 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1776 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1850 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1777 else 1851 else
1852 {
1853 /* in this case, all is known, and we just put dirs first and sort them */
1854 eio_dirent *oth = dents + dentoffs;
1855 eio_dirent *dir = dents;
1856
1857 /* now partition dirs to the front, and non-dirs to the back */
1858 /* by walking from both sides and swapping if necessary */
1859 while (oth > dir)
1860 {
1861 if (dir->type == EIO_DT_DIR)
1862 ++dir;
1863 else if ((--oth)->type == EIO_DT_DIR)
1864 {
1865 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1866
1867 ++dir;
1868 }
1869 }
1870
1871 /* now sort the dirs only (dirs all have the same score) */
1872 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1873 }
1874
1875 break;
1876 }
1877
1878 /* now add the entry to our list(s) */
1879 name = D_NAME (entp);
1880
1881 /* skip . and .. entries */
1882 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1883 {
1884 int len = D_NAMLEN (entp) + 1;
1885
1886 while (ecb_expect_false (namesoffs + len > namesalloc))
1887 {
1888 namesalloc *= 2;
1889 req->ptr2 = names = realloc (names, namesalloc);
1890
1891 if (!names)
1892 break;
1893 }
1894
1895 memcpy (names + namesoffs, name, len);
1896
1897 if (dents)
1898 {
1899 struct eio_dirent *ent;
1900
1901 if (ecb_expect_false (dentoffs == dentalloc))
1778 { 1902 {
1779 /* in this case, all is known, and we just put dirs first and sort them */ 1903 dentalloc *= 2;
1904 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1905
1906 if (!dents)
1907 break;
1908 }
1909
1780 eio_dirent *oth = dents + dentoffs; 1910 ent = dents + dentoffs;
1781 eio_dirent *dir = dents;
1782 1911
1783 /* now partition dirs to the front, and non-dirs to the back */ 1912 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1784 /* by walking from both sides and swapping if necessary */ 1913 ent->namelen = len - 1;
1785 while (oth > dir) 1914 ent->inode = D_INO (entp);
1915
1916 inode_bits |= ent->inode;
1917
1918 switch (D_TYPE (entp))
1919 {
1920 default:
1921 ent->type = EIO_DT_UNKNOWN;
1922 flags |= EIO_READDIR_FOUND_UNKNOWN;
1923 break;
1924
1925 #ifdef DT_FIFO
1926 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1927 #endif
1928 #ifdef DT_CHR
1929 case DT_CHR: ent->type = EIO_DT_CHR; break;
1930 #endif
1931 #ifdef DT_MPC
1932 case DT_MPC: ent->type = EIO_DT_MPC; break;
1933 #endif
1934 #ifdef DT_DIR
1935 case DT_DIR: ent->type = EIO_DT_DIR; break;
1936 #endif
1937 #ifdef DT_NAM
1938 case DT_NAM: ent->type = EIO_DT_NAM; break;
1939 #endif
1940 #ifdef DT_BLK
1941 case DT_BLK: ent->type = EIO_DT_BLK; break;
1942 #endif
1943 #ifdef DT_MPB
1944 case DT_MPB: ent->type = EIO_DT_MPB; break;
1945 #endif
1946 #ifdef DT_REG
1947 case DT_REG: ent->type = EIO_DT_REG; break;
1948 #endif
1949 #ifdef DT_NWK
1950 case DT_NWK: ent->type = EIO_DT_NWK; break;
1951 #endif
1952 #ifdef DT_CMP
1953 case DT_CMP: ent->type = EIO_DT_CMP; break;
1954 #endif
1955 #ifdef DT_LNK
1956 case DT_LNK: ent->type = EIO_DT_LNK; break;
1957 #endif
1958 #ifdef DT_SOCK
1959 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1960 #endif
1961 #ifdef DT_DOOR
1962 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1963 #endif
1964 #ifdef DT_WHT
1965 case DT_WHT: ent->type = EIO_DT_WHT; break;
1966 #endif
1967 }
1968
1969 ent->score = 7;
1970
1971 if (flags & EIO_READDIR_DIRS_FIRST)
1972 {
1973 if (ent->type == EIO_DT_UNKNOWN)
1786 { 1974 {
1787 if (dir->type == EIO_DT_DIR) 1975 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1788 ++dir; 1976 ent->score = 1;
1789 else if ((--oth)->type == EIO_DT_DIR) 1977 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1790 { 1978 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1791 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1792
1793 ++dir;
1794 }
1795 } 1979 }
1796 1980 else if (ent->type == EIO_DT_DIR)
1797 /* now sort the dirs only (dirs all have the same score) */ 1981 ent->score = 0;
1798 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1799 } 1982 }
1800
1801 break;
1802 }
1803
1804 /* now add the entry to our list(s) */
1805 name = D_NAME (entp);
1806
1807 /* skip . and .. entries */
1808 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1809 {
1810 int len = D_NAMLEN (entp) + 1;
1811
1812 while (ecb_expect_false (namesoffs + len > namesalloc))
1813 {
1814 namesalloc *= 2;
1815 req->ptr2 = names = realloc (names, namesalloc);
1816
1817 if (!names)
1818 break;
1819 } 1983 }
1820 1984
1821 memcpy (names + namesoffs, name, len);
1822
1823 if (dents)
1824 {
1825 struct eio_dirent *ent;
1826
1827 if (ecb_expect_false (dentoffs == dentalloc))
1828 {
1829 dentalloc *= 2;
1830 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1831
1832 if (!dents)
1833 break;
1834 }
1835
1836 ent = dents + dentoffs;
1837
1838 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1839 ent->namelen = len - 1;
1840 ent->inode = D_INO (entp);
1841
1842 inode_bits |= ent->inode;
1843
1844 switch (D_TYPE (entp))
1845 {
1846 default:
1847 ent->type = EIO_DT_UNKNOWN;
1848 flags |= EIO_READDIR_FOUND_UNKNOWN;
1849 break;
1850
1851 #ifdef DT_FIFO
1852 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1853 #endif
1854 #ifdef DT_CHR
1855 case DT_CHR: ent->type = EIO_DT_CHR; break;
1856 #endif
1857 #ifdef DT_MPC
1858 case DT_MPC: ent->type = EIO_DT_MPC; break;
1859 #endif
1860 #ifdef DT_DIR
1861 case DT_DIR: ent->type = EIO_DT_DIR; break;
1862 #endif
1863 #ifdef DT_NAM
1864 case DT_NAM: ent->type = EIO_DT_NAM; break;
1865 #endif
1866 #ifdef DT_BLK
1867 case DT_BLK: ent->type = EIO_DT_BLK; break;
1868 #endif
1869 #ifdef DT_MPB
1870 case DT_MPB: ent->type = EIO_DT_MPB; break;
1871 #endif
1872 #ifdef DT_REG
1873 case DT_REG: ent->type = EIO_DT_REG; break;
1874 #endif
1875 #ifdef DT_NWK
1876 case DT_NWK: ent->type = EIO_DT_NWK; break;
1877 #endif
1878 #ifdef DT_CMP
1879 case DT_CMP: ent->type = EIO_DT_CMP; break;
1880 #endif
1881 #ifdef DT_LNK
1882 case DT_LNK: ent->type = EIO_DT_LNK; break;
1883 #endif
1884 #ifdef DT_SOCK
1885 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1886 #endif
1887 #ifdef DT_DOOR
1888 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1889 #endif
1890 #ifdef DT_WHT
1891 case DT_WHT: ent->type = EIO_DT_WHT; break;
1892 #endif
1893 }
1894
1895 ent->score = 7;
1896
1897 if (flags & EIO_READDIR_DIRS_FIRST)
1898 {
1899 if (ent->type == EIO_DT_UNKNOWN)
1900 {
1901 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1902 ent->score = 1;
1903 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1904 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1905 }
1906 else if (ent->type == EIO_DT_DIR)
1907 ent->score = 0;
1908 }
1909 }
1910
1911 namesoffs += len; 1985 namesoffs += len;
1912 ++dentoffs; 1986 ++dentoffs;
1913 } 1987 }
1914 1988
1915 if (EIO_CANCELLED (req)) 1989 if (EIO_CANCELLED (req))
1916 { 1990 {
1917 errno = ECANCELED; 1991 errno = ECANCELED;
1918 break; 1992 break;
1919 } 1993 }
1920 1994
1921#ifdef _WIN32 1995#ifdef _WIN32
1922 if (!FindNextFile (dirp, &entp)) 1996 if (!FindNextFile (dirp, &entp))
1923 { 1997 {
1924 FindClose (dirp); 1998 FindClose (dirp);
1925 dirp = 0; 1999 dirp = 0;
1926 } 2000 }
1927#endif 2001#endif
1928 } 2002 }
1929} 2003}
2004
2005/*****************************************************************************/
2006/* working directory stuff */
2007/* various deficiencies in the posix 2008 api force us to */
2008/* keep the absolute path in string form at all times */
2009/* fuck yeah. */
2010
2011#if !HAVE_AT
2012
2013/* a bit like realpath, but usually faster because it doesn'T have to return */
2014/* an absolute or canonical path */
2015static const char *
2016wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2017{
2018 if (!wd || *path == '/')
2019 return path;
2020
2021 if (path [0] == '.' && !path [1])
2022 return wd->str;
2023
2024 {
2025 int l1 = wd->len;
2026 int l2 = strlen (path);
2027
2028 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2029
2030 memcpy (res, wd->str, l1);
2031 res [l1] = '/';
2032 memcpy (res + l1 + 1, path, l2 + 1);
2033
2034 return res;
2035 }
2036}
2037
2038#endif
2039
2040static eio_wd
2041eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2042{
2043 int fd;
2044 eio_wd res;
2045 int len = eio__realpath (tmpbuf, wd, path);
2046
2047 if (len < 0)
2048 return EIO_INVALID_WD;
2049
2050#if HAVE_AT
2051 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
2052
2053 if (fd < 0)
2054 return EIO_INVALID_WD;
2055#endif
2056
2057 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
2058
2059#if HAVE_AT
2060 res->fd = fd;
2061#endif
2062
2063 res->len = len;
2064 memcpy (res->str, tmpbuf->ptr, len);
2065 res->str [len] = 0;
2066
2067 return res;
2068}
2069
2070eio_wd
2071eio_wd_open_sync (eio_wd wd, const char *path)
2072{
2073 struct tmpbuf tmpbuf = { 0 };
2074 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2075 free (tmpbuf.ptr);
2076
2077 return wd;
2078}
2079
2080void
2081eio_wd_close_sync (eio_wd wd)
2082{
2083 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
2084 {
2085 #if HAVE_AT
2086 close (wd->fd);
2087 #endif
2088 free (wd);
2089 }
2090}
2091
2092#if HAVE_AT
2093
2094/* they forgot these */
2095
2096static int
2097eio__truncateat (int dirfd, const char *path, off_t length)
2098{
2099 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
2100 int res;
2101
2102 if (fd < 0)
2103 return fd;
2104
2105 res = ftruncate (fd, length);
2106 close (fd);
2107 return res;
2108}
2109
2110static int
2111eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
2112{
2113 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2114 int res;
2115
2116 if (fd < 0)
2117 return fd;
2118
2119 res = fstatvfs (fd, buf);
2120 close (fd);
2121 return res;
2122
2123}
2124
2125#endif
1930 2126
1931/*****************************************************************************/ 2127/*****************************************************************************/
1932 2128
1933#define ALLOC(len) \ 2129#define ALLOC(len) \
1934 if (!req->ptr2) \ 2130 if (!req->ptr2) \
1943 req->result = -1; \ 2139 req->result = -1; \
1944 break; \ 2140 break; \
1945 } \ 2141 } \
1946 } 2142 }
1947 2143
2144static void ecb_noinline ecb_cold
2145etp_proc_init (void)
2146{
2147#if HAVE_PRCTL_SET_NAME
2148 /* provide a more sensible "thread name" */
2149 char name[16 + 1];
2150 const int namelen = sizeof (name) - 1;
2151 int len;
2152
2153 prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
2154 name [namelen] = 0;
2155 len = strlen (name);
2156 strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
2157 prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
2158#endif
2159}
2160
1948X_THREAD_PROC (etp_proc) 2161X_THREAD_PROC (etp_proc)
1949{ 2162{
1950 ETP_REQ *req; 2163 ETP_REQ *req;
1951 struct timespec ts; 2164 struct timespec ts;
1952 etp_worker *self = (etp_worker *)thr_arg; 2165 etp_worker *self = (etp_worker *)thr_arg;
1953 2166
1954#if HAVE_PRCTL_SET_NAME 2167 etp_proc_init ();
1955 prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
1956#endif
1957 2168
1958 /* try to distribute timeouts somewhat evenly */ 2169 /* try to distribute timeouts somewhat evenly */
1959 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2170 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1960 2171
1961 for (;;) 2172 for (;;)
1964 2175
1965 X_LOCK (reqlock); 2176 X_LOCK (reqlock);
1966 2177
1967 for (;;) 2178 for (;;)
1968 { 2179 {
1969 self->req = req = reqq_shift (&req_queue); 2180 req = reqq_shift (&req_queue);
1970 2181
1971 if (req) 2182 if (req)
1972 break; 2183 break;
1973 2184
1974 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */ 2185 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2012 ++npending; 2223 ++npending;
2013 2224
2014 if (!reqq_push (&res_queue, req) && want_poll_cb) 2225 if (!reqq_push (&res_queue, req) && want_poll_cb)
2015 want_poll_cb (); 2226 want_poll_cb ();
2016 2227
2017 self->req = 0;
2018 etp_worker_clear (self); 2228 etp_worker_clear (self);
2019 2229
2020 X_UNLOCK (reslock); 2230 X_UNLOCK (reslock);
2021 } 2231 }
2022 2232
2023quit: 2233quit:
2234 free (req);
2235
2024 X_LOCK (wrklock); 2236 X_LOCK (wrklock);
2025 etp_worker_free (self); 2237 etp_worker_free (self);
2026 X_UNLOCK (wrklock); 2238 X_UNLOCK (wrklock);
2027 2239
2028 return 0; 2240 return 0;
2031/*****************************************************************************/ 2243/*****************************************************************************/
2032 2244
2033int ecb_cold 2245int ecb_cold
2034eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2246eio_init (void (*want_poll)(void), void (*done_poll)(void))
2035{ 2247{
2036#if !HAVE_PREADWRITE
2037 X_MUTEX_CREATE (preadwritelock);
2038#endif
2039
2040 return etp_init (want_poll, done_poll); 2248 return etp_init (want_poll, done_poll);
2041} 2249}
2042 2250
2043ecb_inline void 2251ecb_inline void
2044eio_api_destroy (eio_req *req) 2252eio_api_destroy (eio_req *req)
2045{ 2253{
2046 free (req); 2254 free (req);
2047} 2255}
2048 2256
2049#define REQ(rtype) \ 2257#define REQ(rtype) \
2050 eio_req *req; \ 2258 eio_req *req; \
2051 \ 2259 \
2052 req = (eio_req *)calloc (1, sizeof *req); \ 2260 req = (eio_req *)calloc (1, sizeof *req); \
2053 if (!req) \ 2261 if (!req) \
2054 return 0; \ 2262 return 0; \
2068 { \ 2276 { \
2069 eio_api_destroy (req); \ 2277 eio_api_destroy (req); \
2070 return 0; \ 2278 return 0; \
2071 } 2279 }
2072 2280
2281#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
2282
2073static void 2283static void
2074eio_execute (etp_worker *self, eio_req *req) 2284eio_execute (etp_worker *self, eio_req *req)
2075{ 2285{
2286#if HAVE_AT
2287 int dirfd;
2288#else
2289 const char *path;
2290#endif
2291
2076 if (ecb_expect_false (EIO_CANCELLED (req))) 2292 if (ecb_expect_false (EIO_CANCELLED (req)))
2077 { 2293 {
2078 req->result = -1; 2294 req->result = -1;
2079 req->errorno = ECANCELED; 2295 req->errorno = ECANCELED;
2080 return; 2296 return;
2081 } 2297 }
2082 2298
2299 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2300 {
2301 req->result = -1;
2302 req->errorno = ENOENT;
2303 return;
2304 }
2305
2306 if (req->type >= EIO_OPEN)
2307 {
2308 #if HAVE_AT
2309 dirfd = WD2FD (req->wd);
2310 #else
2311 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2312 #endif
2313 }
2314
2083 switch (req->type) 2315 switch (req->type)
2084 { 2316 {
2317 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2318 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2319 break;
2320 case EIO_WD_CLOSE: req->result = 0;
2321 eio_wd_close_sync (req->wd); break;
2322
2323 case EIO_SEEK: eio__lseek (req); break;
2085 case EIO_READ: ALLOC (req->size); 2324 case EIO_READ: ALLOC (req->size);
2086 req->result = req->offs >= 0 2325 req->result = req->offs >= 0
2087 ? pread (req->int1, req->ptr2, req->size, req->offs) 2326 ? pread (req->int1, req->ptr2, req->size, req->offs)
2088 : read (req->int1, req->ptr2, req->size); break; 2327 : read (req->int1, req->ptr2, req->size); break;
2089 case EIO_WRITE: req->result = req->offs >= 0 2328 case EIO_WRITE: req->result = req->offs >= 0
2091 : write (req->int1, req->ptr2, req->size); break; 2330 : write (req->int1, req->ptr2, req->size); break;
2092 2331
2093 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2332 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
2094 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break; 2333 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2095 2334
2335#if HAVE_AT
2336
2096 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2337 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2097 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2338 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
2098 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2339 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2340 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
2341 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
2342 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
2343 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
2344 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
2345
2346 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
2347 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
2348 req->result = req->wd && SINGLEDOT (req->ptr1)
2349 ? rmdir (req->wd->str)
2350 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
2351 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
2352 case EIO_RENAME: /* complications arise because "." cannot be renamed, so we might have to expand */
2353 req->result = req->wd && SINGLEDOT (req->ptr1)
2354 ? rename (req->wd->str, req->ptr2)
2355 : renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
2356 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
2357 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
2358 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2359 case EIO_READLINK: ALLOC (PATH_MAX);
2360 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
2361 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2362 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2363 case EIO_UTIME:
2364 case EIO_FUTIME:
2365 {
2366 struct timespec ts[2];
2367 struct timespec *times;
2368
2369 if (req->nv1 != -1. || req->nv2 != -1.)
2370 {
2371 ts[0].tv_sec = req->nv1;
2372 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2373 ts[1].tv_sec = req->nv2;
2374 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2375
2376 times = ts;
2377 }
2378 else
2379 times = 0;
2380
2381 req->result = req->type == EIO_FUTIME
2382 ? futimens (req->int1, times)
2383 : utimensat (dirfd, req->ptr1, times, 0);
2384 }
2385 break;
2386
2387#else
2388
2389 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2390 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2391 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2099 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2392 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2393 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2394 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2395 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2396 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2397
2398 case EIO_UNLINK: req->result = unlink (path ); break;
2399 case EIO_RMDIR: req->result = rmdir (path ); break;
2400 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
2401 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2402 case EIO_LINK: req->result = link (path , req->ptr2); break;
2403 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2404 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2405 case EIO_READLINK: ALLOC (PATH_MAX);
2406 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2407 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2408 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2409
2410 case EIO_UTIME:
2411 case EIO_FUTIME:
2412 {
2413 struct timeval tv[2];
2414 struct timeval *times;
2415
2416 if (req->nv1 != -1. || req->nv2 != -1.)
2417 {
2418 tv[0].tv_sec = req->nv1;
2419 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
2420 tv[1].tv_sec = req->nv2;
2421 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
2422
2423 times = tv;
2424 }
2425 else
2426 times = 0;
2427
2428 req->result = req->type == EIO_FUTIME
2429 ? futimes (req->int1, times)
2430 : utimes (req->ptr1, times);
2431 }
2432 break;
2433
2434#endif
2435
2436 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2437 {
2438 ALLOC (req->result);
2439 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2440 }
2441 break;
2442
2100 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2443 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2101 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2444 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2102 2445
2103 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2104 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2105 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 2446 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2106 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 2447 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2107 2448
2108 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
2109 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2449 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2110 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
2111 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2450 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2112 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
2113 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 2451 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2114 2452
2115 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
2116 case EIO_CLOSE: req->result = close (req->int1); break; 2453 case EIO_CLOSE: req->result = close (req->int1); break;
2117 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 2454 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2118 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
2119 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
2120 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
2121 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
2122 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
2123 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
2124 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2125
2126 case EIO_REALPATH: eio__realpath (req, self); break;
2127
2128 case EIO_READLINK: ALLOC (PATH_MAX);
2129 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
2130
2131 case EIO_SYNC: req->result = 0; sync (); break; 2455 case EIO_SYNC: req->result = 0; sync (); break;
2132 case EIO_FSYNC: req->result = fsync (req->int1); break; 2456 case EIO_FSYNC: req->result = fsync (req->int1); break;
2133 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2457 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2134 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break; 2458 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2135 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 2459 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2154 req->result = select (0, 0, 0, 0, &tv); 2478 req->result = select (0, 0, 0, 0, &tv);
2155 } 2479 }
2156#endif 2480#endif
2157 break; 2481 break;
2158 2482
2159 case EIO_UTIME:
2160 case EIO_FUTIME:
2161 {
2162 struct timeval tv[2];
2163 struct timeval *times;
2164
2165 if (req->nv1 != -1. || req->nv2 != -1.)
2166 {
2167 tv[0].tv_sec = req->nv1;
2168 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
2169 tv[1].tv_sec = req->nv2;
2170 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
2171
2172 times = tv;
2173 }
2174 else
2175 times = 0;
2176
2177 req->result = req->type == EIO_FUTIME
2178 ? futimes (req->int1, times)
2179 : utimes (req->ptr1, times);
2180 }
2181 break;
2182
2183 case EIO_GROUP: 2483 case EIO_GROUP:
2184 abort (); /* handled in eio_request */ 2484 abort (); /* handled in eio_request */
2185 2485
2186 case EIO_NOP: 2486 case EIO_NOP:
2187 req->result = 0; 2487 req->result = 0;
2190 case EIO_CUSTOM: 2490 case EIO_CUSTOM:
2191 req->feed (req); 2491 req->feed (req);
2192 break; 2492 break;
2193 2493
2194 default: 2494 default:
2195 errno = ENOSYS;
2196 req->result = -1; 2495 req->result = EIO_ENOSYS ();
2197 break; 2496 break;
2198 } 2497 }
2199 2498
2200 req->errorno = errno; 2499 req->errorno = errno;
2201} 2500}
2202 2501
2203#ifndef EIO_NO_WRAPPERS 2502#ifndef EIO_NO_WRAPPERS
2204 2503
2504eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2505{
2506 REQ (EIO_WD_OPEN); PATH; SEND;
2507}
2508
2509eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2510{
2511 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2512}
2513
2205eio_req *eio_nop (int pri, eio_cb cb, void *data) 2514eio_req *eio_nop (int pri, eio_cb cb, void *data)
2206{ 2515{
2207 REQ (EIO_NOP); SEND; 2516 REQ (EIO_NOP); SEND;
2208} 2517}
2209 2518
2268} 2577}
2269 2578
2270eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2579eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
2271{ 2580{
2272 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2581 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
2582}
2583
2584eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2585{
2586 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2273} 2587}
2274 2588
2275eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2589eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2276{ 2590{
2277 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2591 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines