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

Comparing libeio/eio.c (file contents):
Revision 1.90 by root, Sat Jul 16 16:46:10 2011 UTC vs.
Revision 1.101 by root, Wed Aug 3 15:25:38 2011 UTC

71#endif 71#endif
72#ifndef ELOOP 72#ifndef ELOOP
73# define ELOOP EDOM 73# define ELOOP EDOM
74#endif 74#endif
75 75
76#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
77# define ENOTSOCK WSAENOTSOCK
78#endif
79
76static void eio_destroy (eio_req *req); 80static void eio_destroy (eio_req *req);
77 81
78#ifndef EIO_FINISH 82#ifndef EIO_FINISH
79# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 83# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
80#endif 84#endif
101#ifdef _WIN32 105#ifdef _WIN32
102 106
103 #undef PAGESIZE 107 #undef PAGESIZE
104 #define PAGESIZE 4096 /* GetSystemInfo? */ 108 #define PAGESIZE 4096 /* GetSystemInfo? */
105 109
110 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
111
106 #ifdef EIO_STRUCT_STATI64 112 #ifdef EIO_STRUCT_STATI64
113 /* look at perl's non-sloppy stat */
107 #define stat(path,buf) _stati64 (path,buf) 114 #define stat(path,buf) _stati64 (path,buf)
108 #define fstat(fd,buf) _fstati64 (path,buf) 115 #define fstat(fd,buf) _fstati64 (fd,buf)
109 #endif 116 #endif
110 #define lstat(path,buf) stat (path,buf) 117 #define lstat(path,buf) stat (path,buf)
111 #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))
112 #define mkdir(path,mode) _mkdir (path) 119 #define mkdir(path,mode) _mkdir (path)
113 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1)) 120 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
114 121
115 #define chmod(path,mode) _chmod (path, mode) 122 #define chmod(path,mode) _chmod (path, mode)
123 #define dup(fd) _dup (fd)
124 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
125
116 #define fchmod(fd,mode) EIO_ENOSYS () 126 #define fchmod(fd,mode) EIO_ENOSYS ()
117 #define chown(path,uid,gid) EIO_ENOSYS () 127 #define chown(path,uid,gid) EIO_ENOSYS ()
118 #define fchown(fd,uid,gid) EIO_ENOSYS () 128 #define fchown(fd,uid,gid) EIO_ENOSYS ()
119 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */ 129 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
120 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */ 130 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
122 #define sync() EIO_ENOSYS () 132 #define sync() EIO_ENOSYS ()
123 #define readlink(path,buf,s) EIO_ENOSYS () 133 #define readlink(path,buf,s) EIO_ENOSYS ()
124 #define statvfs(path,buf) EIO_ENOSYS () 134 #define statvfs(path,buf) EIO_ENOSYS ()
125 #define fstatvfs(fd,buf) EIO_ENOSYS () 135 #define fstatvfs(fd,buf) EIO_ENOSYS ()
126 136
137 /* rename() uses MoveFile, which fails to overwrite */
138 #define rename(old,neu) eio__rename (old, neu)
139
140 static int
141 eio__rename (const char *old, const char *neu)
142 {
143 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
144 return 0;
145
146 /* should steal _dosmaperr */
147 switch (GetLastError ())
148 {
149 case ERROR_FILE_NOT_FOUND:
150 case ERROR_PATH_NOT_FOUND:
151 case ERROR_INVALID_DRIVE:
152 case ERROR_NO_MORE_FILES:
153 case ERROR_BAD_NETPATH:
154 case ERROR_BAD_NET_NAME:
155 case ERROR_BAD_PATHNAME:
156 case ERROR_FILENAME_EXCED_RANGE:
157 errno = ENOENT;
158 break;
159
160 default:
161 errno = EACCES;
162 break;
163 }
164
165 return -1;
166 }
167
127 /* we could even stat and see if it exists */ 168 /* we could even stat and see if it exists */
128 static int 169 static int
129 symlink (const char *old, const char *neu) 170 symlink (const char *old, const char *neu)
130 { 171 {
131 #if WINVER >= 0x0600 172 #if WINVER >= 0x0600
138 179
139 return EIO_ERRNO (ENOENT, -1); 180 return EIO_ERRNO (ENOENT, -1);
140 } 181 }
141 182
142 /* POSIX API only */ 183 /* POSIX API only */
143 #define CreateHardLink(neu,old) 0 184 #define CreateHardLink(neu,old,flags) 0
144 #define CreateSymbolicLink(neu,old,flags) 0 185 #define CreateSymbolicLink(neu,old,flags) 0
145 186
146 struct statvfs 187 struct statvfs
147 { 188 {
148 int dummy; 189 int dummy;
149 }; 190 };
150 191
192 #define DT_DIR EIO_DT_DIR
193 #define DT_REG EIO_DT_REG
194 #define D_NAME(entp) entp.cFileName
195 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
196
151#else 197#else
152 198
153 #include <sys/time.h> 199 #include <sys/time.h>
154 #include <sys/select.h> 200 #include <sys/select.h>
155 #include <sys/statvfs.h> 201 #include <sys/statvfs.h>
156 #include <unistd.h> 202 #include <unistd.h>
157 #include <utime.h>
158 #include <signal.h> 203 #include <signal.h>
159 #include <dirent.h> 204 #include <dirent.h>
160 205
161 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES 206 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
162 #include <sys/mman.h> 207 #include <sys/mman.h>
163 #endif 208 #endif
209
210 #define D_NAME(entp) entp->d_name
164 211
165 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ 212 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
166 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ 213 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
167 #define _DIRENT_HAVE_D_TYPE /* sigh */ 214 #define _DIRENT_HAVE_D_TYPE /* sigh */
168 #define D_INO(de) (de)->d_fileno 215 #define D_INO(de) (de)->d_fileno
182 229
183 #ifndef EIO_STRUCT_DIRENT 230 #ifndef EIO_STRUCT_DIRENT
184 #define EIO_STRUCT_DIRENT struct dirent 231 #define EIO_STRUCT_DIRENT struct dirent
185 #endif 232 #endif
186 233
234#endif
235
236#if HAVE_UTIMES
237# include <utime.h>
238#endif
239
240#if HAVE_SYS_SYSCALL_H
241# include <sys/syscall.h>
242#endif
243
244#if HAVE_SYS_PRCTL_H
245# include <sys/prctl.h>
187#endif 246#endif
188 247
189#if HAVE_SENDFILE 248#if HAVE_SENDFILE
190# if __linux 249# if __linux
191# include <sys/sendfile.h> 250# include <sys/sendfile.h>
206#endif 265#endif
207#ifndef D_INO 266#ifndef D_INO
208# define D_INO(de) 0 267# define D_INO(de) 0
209#endif 268#endif
210#ifndef D_NAMLEN 269#ifndef D_NAMLEN
211# define D_NAMLEN(de) strlen ((de)->d_name) 270# define D_NAMLEN(entp) strlen (D_NAME (entp))
212#endif 271#endif
213 272
214/* used for struct dirent, AIX doesn't provide it */ 273/* used for struct dirent, AIX doesn't provide it */
215#ifndef NAME_MAX 274#ifndef NAME_MAX
216# define NAME_MAX 4096 275# define NAME_MAX 4096
223 282
224/* buffer size for various temporary buffers */ 283/* buffer size for various temporary buffers */
225#define EIO_BUFSIZE 65536 284#define EIO_BUFSIZE 65536
226 285
227#define dBUF \ 286#define dBUF \
228 char *eio_buf; \
229 ETP_WORKER_LOCK (self); \
230 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 287 char *eio_buf = malloc (EIO_BUFSIZE); \
231 ETP_WORKER_UNLOCK (self); \
232 errno = ENOMEM; \ 288 errno = ENOMEM; \
233 if (!eio_buf) \ 289 if (!eio_buf) \
234 return -1; 290 return -1
291
292#define FUBd \
293 free (eio_buf)
235 294
236#define EIO_TICKS ((1000000 + 1023) >> 10) 295#define EIO_TICKS ((1000000 + 1023) >> 10)
237 296
238#define ETP_PRI_MIN EIO_PRI_MIN 297#define ETP_PRI_MIN EIO_PRI_MIN
239#define ETP_PRI_MAX EIO_PRI_MAX 298#define ETP_PRI_MAX EIO_PRI_MAX
245static int eio_finish (eio_req *req); 304static int eio_finish (eio_req *req);
246#define ETP_FINISH(req) eio_finish (req) 305#define ETP_FINISH(req) eio_finish (req)
247static void eio_execute (struct etp_worker *self, eio_req *req); 306static void eio_execute (struct etp_worker *self, eio_req *req);
248#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 307#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
249 308
250#define ETP_WORKER_CLEAR(req) \
251 if (wrk->dbuf) \
252 { \
253 free (wrk->dbuf); \
254 wrk->dbuf = 0; \
255 } \
256 \
257 if (wrk->dirp) \
258 { \
259 closedir (wrk->dirp); \
260 wrk->dirp = 0; \
261 }
262
263#define ETP_WORKER_COMMON \
264 void *dbuf; \
265 DIR *dirp;
266
267/*****************************************************************************/ 309/*****************************************************************************/
268 310
269#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 311#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
270 312
271/* calculate time difference in ~1/EIO_TICKS of a second */ 313/* calculate time difference in ~1/EIO_TICKS of a second */
282static void (*done_poll_cb) (void); 324static void (*done_poll_cb) (void);
283 325
284static unsigned int max_poll_time; /* reslock */ 326static unsigned int max_poll_time; /* reslock */
285static unsigned int max_poll_reqs; /* reslock */ 327static unsigned int max_poll_reqs; /* reslock */
286 328
287static volatile unsigned int nreqs; /* reqlock */ 329static unsigned int nreqs; /* reqlock */
288static volatile unsigned int nready; /* reqlock */ 330static unsigned int nready; /* reqlock */
289static volatile unsigned int npending; /* reqlock */ 331static unsigned int npending; /* reqlock */
290static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */ 332static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
291static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */ 333static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
292 334
293static xmutex_t wrklock; 335static xmutex_t wrklock;
294static xmutex_t reslock; 336static xmutex_t reslock;
295static xmutex_t reqlock; 337static xmutex_t reqlock;
296static xcond_t reqwait; 338static xcond_t reqwait;
299/* 341/*
300 * make our pread/pwrite emulation safe against themselves, but not against 342 * make our pread/pwrite emulation safe against themselves, but not against
301 * normal read/write by using a mutex. slows down execution a lot, 343 * normal read/write by using a mutex. slows down execution a lot,
302 * but that's your problem, not mine. 344 * but that's your problem, not mine.
303 */ 345 */
304static xmutex_t preadwritelock = X_MUTEX_INIT; 346static xmutex_t preadwritelock;
305#endif 347#endif
306 348
307typedef struct etp_worker 349typedef struct etp_worker
308{ 350{
309 /* locked by wrklock */ 351 /* locked by wrklock */
312 xthread_t tid; 354 xthread_t tid;
313 355
314 /* locked by reslock, reqlock or wrklock */ 356 /* locked by reslock, reqlock or wrklock */
315 ETP_REQ *req; /* currently processed request */ 357 ETP_REQ *req; /* currently processed request */
316 358
359#ifdef ETP_WORKER_COMMON
317 ETP_WORKER_COMMON 360 ETP_WORKER_COMMON
361#endif
318} etp_worker; 362} etp_worker;
319 363
320static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 364static etp_worker wrk_first; /* NOT etp */
321 365
322#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 366#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
323#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 367#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
324 368
325/* worker threads management */ 369/* worker threads management */
326 370
327static void ecb_cold 371static void ecb_cold
328etp_worker_clear (etp_worker *wrk) 372etp_worker_clear (etp_worker *wrk)
329{ 373{
330 ETP_WORKER_CLEAR (wrk);
331} 374}
332 375
333static void ecb_cold 376static void ecb_cold
334etp_worker_free (etp_worker *wrk) 377etp_worker_free (etp_worker *wrk)
335{ 378{
396} etp_reqq; 439} etp_reqq;
397 440
398static etp_reqq req_queue; 441static etp_reqq req_queue;
399static etp_reqq res_queue; 442static etp_reqq res_queue;
400 443
444static void ecb_noinline ecb_cold
445reqq_init (etp_reqq *q)
446{
447 int pri;
448
449 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
450 q->qs[pri] = q->qe[pri] = 0;
451
452 q->size = 0;
453}
454
401static int ecb_noinline 455static int ecb_noinline
402reqq_push (etp_reqq *q, ETP_REQ *req) 456reqq_push (etp_reqq *q, ETP_REQ *req)
403{ 457{
404 int pri = req->pri; 458 int pri = req->pri;
405 req->next = 0; 459 req->next = 0;
439 } 493 }
440 494
441 abort (); 495 abort ();
442} 496}
443 497
444static void ecb_cold 498static int ecb_cold
445etp_thread_init (void) 499etp_init (void (*want_poll)(void), void (*done_poll)(void))
446{ 500{
447#if !HAVE_PREADWRITE
448 X_MUTEX_CREATE (preadwritelock);
449#endif
450 X_MUTEX_CREATE (wrklock); 501 X_MUTEX_CREATE (wrklock);
451 X_MUTEX_CREATE (reslock); 502 X_MUTEX_CREATE (reslock);
452 X_MUTEX_CREATE (reqlock); 503 X_MUTEX_CREATE (reqlock);
453 X_COND_CREATE (reqwait); 504 X_COND_CREATE (reqwait);
454}
455 505
456static void ecb_cold 506 reqq_init (&req_queue);
457etp_atfork_prepare (void) 507 reqq_init (&res_queue);
458{
459}
460 508
461static void ecb_cold 509 wrk_first.next =
462etp_atfork_parent (void) 510 wrk_first.prev = &wrk_first;
463{
464}
465
466static void ecb_cold
467etp_atfork_child (void)
468{
469 ETP_REQ *prv;
470
471 while ((prv = reqq_shift (&req_queue)))
472 ETP_DESTROY (prv);
473
474 while ((prv = reqq_shift (&res_queue)))
475 ETP_DESTROY (prv);
476
477 while (wrk_first.next != &wrk_first)
478 {
479 etp_worker *wrk = wrk_first.next;
480
481 if (wrk->req)
482 ETP_DESTROY (wrk->req);
483
484 etp_worker_clear (wrk);
485 etp_worker_free (wrk);
486 }
487 511
488 started = 0; 512 started = 0;
489 idle = 0; 513 idle = 0;
490 nreqs = 0; 514 nreqs = 0;
491 nready = 0; 515 nready = 0;
492 npending = 0; 516 npending = 0;
493
494 etp_thread_init ();
495}
496
497static void ecb_cold
498etp_once_init (void)
499{
500 etp_thread_init ();
501 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
502}
503
504static int ecb_cold
505etp_init (void (*want_poll)(void), void (*done_poll)(void))
506{
507 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
508
509 pthread_once (&doinit, etp_once_init);
510 517
511 want_poll_cb = want_poll; 518 want_poll_cb = want_poll;
512 done_poll_cb = done_poll; 519 done_poll_cb = done_poll;
513 520
514 return 0; 521 return 0;
972#if !HAVE_FDATASYNC 979#if !HAVE_FDATASYNC
973# undef fdatasync 980# undef fdatasync
974# define fdatasync(fd) fsync (fd) 981# define fdatasync(fd) fsync (fd)
975#endif 982#endif
976 983
984static int
985eio__syncfs (int fd)
986{
987 int res;
988
989#if HAVE_SYS_SYNCFS
990 res = (int)syscall (__NR_syncfs, (int)(fd));
991#else
992 res = -1;
993 errno = ENOSYS;
994#endif
995
996 if (res < 0 && errno == ENOSYS && fd >= 0)
997 sync ();
998
999 return res;
1000}
1001
977/* sync_file_range always needs emulation */ 1002/* sync_file_range always needs emulation */
978static int 1003static int
979eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 1004eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
980{ 1005{
981#if HAVE_SYNC_FILE_RANGE 1006#if HAVE_SYNC_FILE_RANGE
1030 pread (fd, eio_buf, len, offset); 1055 pread (fd, eio_buf, len, offset);
1031 offset += len; 1056 offset += len;
1032 todo -= len; 1057 todo -= len;
1033 } 1058 }
1034 1059
1060 FUBd;
1061
1035 errno = 0; 1062 errno = 0;
1036 return count; 1063 return count;
1037} 1064}
1038 1065
1039#endif 1066#endif
1040 1067
1041/* sendfile always needs emulation */ 1068/* sendfile always needs emulation */
1042static eio_ssize_t 1069static eio_ssize_t
1043eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1070eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
1044{ 1071{
1045 eio_ssize_t written = 0; 1072 eio_ssize_t written = 0;
1046 eio_ssize_t res; 1073 eio_ssize_t res;
1047 1074
1048 if (!count) 1075 if (!count)
1184 1211
1185 offset += cnt; 1212 offset += cnt;
1186 res += cnt; 1213 res += cnt;
1187 count -= cnt; 1214 count -= cnt;
1188 } 1215 }
1216
1217 FUBd;
1189 } 1218 }
1190 1219
1191 return res; 1220 return res;
1192} 1221}
1193 1222
1483 1512
1484#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1513#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1485#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1514#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1486 1515
1487static void 1516static void
1488eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1517eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1489{ 1518{
1490 unsigned char bits [9 + sizeof (ino_t) * 8]; 1519 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1491 unsigned char *bit = bits; 1520 unsigned char *bit = bits;
1492 1521
1493 assert (CHAR_BIT == 8); 1522 assert (CHAR_BIT == 8);
1494 assert (sizeof (eio_dirent) * 8 < 256); 1523 assert (sizeof (eio_dirent) * 8 < 256);
1495 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1524 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1497 1526
1498 if (size <= EIO_SORT_FAST) 1527 if (size <= EIO_SORT_FAST)
1499 return; 1528 return;
1500 1529
1501 /* first prepare an array of bits to test in our radix sort */ 1530 /* first prepare an array of bits to test in our radix sort */
1502 /* try to take endianness into account, as well as differences in ino_t sizes */ 1531 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1503 /* inode_bits must contain all inodes ORed together */ 1532 /* inode_bits must contain all inodes ORed together */
1504 /* which is used to skip bits that are 0 everywhere, which is very common */ 1533 /* which is used to skip bits that are 0 everywhere, which is very common */
1505 { 1534 {
1506 ino_t endianness; 1535 eio_ino_t endianness;
1507 int i, j; 1536 int i, j;
1508 1537
1509 /* we store the byte offset of byte n into byte n of "endianness" */ 1538 /* we store the byte offset of byte n into byte n of "endianness" */
1510 for (i = 0; i < sizeof (ino_t); ++i) 1539 for (i = 0; i < sizeof (eio_ino_t); ++i)
1511 ((unsigned char *)&endianness)[i] = i; 1540 ((unsigned char *)&endianness)[i] = i;
1512 1541
1513 *bit++ = 0; 1542 *bit++ = 0;
1514 1543
1515 for (i = 0; i < sizeof (ino_t); ++i) 1544 for (i = 0; i < sizeof (eio_ino_t); ++i)
1516 { 1545 {
1517 /* shifting off the byte offsets out of "endianness" */ 1546 /* shifting off the byte offsets out of "endianness" */
1518 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1547 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1519 endianness >>= 8; 1548 endianness >>= 8;
1520 1549
1521 for (j = 0; j < 8; ++j) 1550 for (j = 0; j < 8; ++j)
1522 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1551 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1523 *bit++ = offs + j; 1552 *bit++ = offs + j;
1524 } 1553 }
1525 1554
1526 for (j = 0; j < 8; ++j) 1555 for (j = 0; j < 8; ++j)
1527 if (score_bits & (1 << j)) 1556 if (score_bits & (1 << j))
1528 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1557 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1529 } 1558 }
1530 1559
1531 /* now actually do the sorting (a variant of MSD radix sort) */ 1560 /* now actually do the sorting (a variant of MSD radix sort) */
1532 { 1561 {
1533 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1562 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1534 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1563 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1535 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1564 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1536 int stk_idx = 0; 1565 int stk_idx = 0;
1537 1566
1538 base_stk [stk_idx] = dents; 1567 base_stk [stk_idx] = dents;
1539 end_stk [stk_idx] = dents + size; 1568 end_stk [stk_idx] = dents + size;
1540 bit_stk [stk_idx] = bit - 1; 1569 bit_stk [stk_idx] = bit - 1;
1619 } 1648 }
1620 } 1649 }
1621} 1650}
1622 1651
1623static void 1652static void
1624eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1653eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1625{ 1654{
1626 if (size <= 1) 1655 if (size <= 1)
1627 return; /* our insertion sort relies on size > 0 */ 1656 return; /* our insertion sort relies on size > 0 */
1628 1657
1629 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1658 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1637 1666
1638/* read a full directory */ 1667/* read a full directory */
1639static void 1668static void
1640eio__scandir (eio_req *req, etp_worker *self) 1669eio__scandir (eio_req *req, etp_worker *self)
1641{ 1670{
1642 DIR *dirp;
1643 EIO_STRUCT_DIRENT *entp;
1644 char *name, *names; 1671 char *name, *names;
1645 int namesalloc = 4096; 1672 int namesalloc = 4096 - sizeof (void *) * 4;
1646 int namesoffs = 0; 1673 int namesoffs = 0;
1647 int flags = req->int1; 1674 int flags = req->int1;
1648 eio_dirent *dents = 0; 1675 eio_dirent *dents = 0;
1649 int dentalloc = 128; 1676 int dentalloc = 128;
1650 int dentoffs = 0; 1677 int dentoffs = 0;
1651 ino_t inode_bits = 0; 1678 eio_ino_t inode_bits = 0;
1679#ifdef _WIN32
1680 HANDLE dirp;
1681 WIN32_FIND_DATA entp;
1682#else
1683 DIR *dirp;
1684 EIO_STRUCT_DIRENT *entp;
1685#endif
1652 1686
1653 req->result = -1; 1687 req->result = -1;
1654 1688
1655 if (!(flags & EIO_READDIR_DENTS)) 1689 if (!(flags & EIO_READDIR_DENTS))
1656 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1690 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1657 1691
1658 X_LOCK (wrklock); 1692#ifdef _WIN32
1659 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1693 {
1694 int len = strlen ((const char *)req->ptr1);
1695 char *path = malloc (MAX_PATH);
1696 const char *fmt;
1697
1698 if (!len)
1699 fmt = "./*";
1700 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
1701 fmt = "%s*";
1702 else
1703 fmt = "%s/*";
1704
1705 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
1706 dirp = FindFirstFile (path, &entp);
1707 free (path);
1708
1709 if (dirp == INVALID_HANDLE_VALUE)
1710 {
1711 dirp = 0;
1712
1713 /* should steal _dosmaperr */
1714 switch (GetLastError ())
1715 {
1716 case ERROR_FILE_NOT_FOUND:
1717 req->result = 0;
1718 break;
1719
1720 case ERROR_INVALID_NAME:
1721 case ERROR_PATH_NOT_FOUND:
1722 case ERROR_NO_MORE_FILES:
1723 errno = ENOENT;
1724 break;
1725
1726 case ERROR_NOT_ENOUGH_MEMORY:
1727 errno = ENOMEM;
1728 break;
1729
1730 default:
1731 errno = EINVAL;
1732 break;
1733 }
1734 }
1735 }
1736#else
1660 self->dirp = dirp = opendir (req->ptr1); 1737 dirp = opendir (req->ptr1);
1738#endif
1661 1739
1662 if (req->flags & EIO_FLAG_PTR1_FREE) 1740 if (req->flags & EIO_FLAG_PTR1_FREE)
1663 free (req->ptr1); 1741 free (req->ptr1);
1664 1742
1665 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1743 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1666 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1744 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1667 req->ptr2 = names = malloc (namesalloc); 1745 req->ptr2 = names = malloc (namesalloc);
1668 X_UNLOCK (wrklock);
1669 1746
1670 if (dirp && names && (!flags || dents)) 1747 if (dirp && names && (!flags || dents))
1671 for (;;) 1748 for (;;)
1672 { 1749 {
1750 int done;
1751
1752#ifdef _WIN32
1753 done = !dirp;
1754#else
1673 errno = 0; 1755 errno = 0;
1674 entp = readdir (dirp); 1756 entp = readdir (dirp);
1757 done = !entp;
1758#endif
1675 1759
1676 if (!entp) 1760 if (done)
1677 { 1761 {
1762#ifndef _WIN32
1763 int old_errno = errno;
1764 closedir (dirp);
1765 errno = old_errno;
1766
1678 if (errno) 1767 if (errno)
1679 break; 1768 break;
1769#endif
1680 1770
1681 /* sort etc. */ 1771 /* sort etc. */
1682 req->int1 = flags; 1772 req->int1 = flags;
1683 req->result = dentoffs; 1773 req->result = dentoffs;
1684 1774
1713 1803
1714 break; 1804 break;
1715 } 1805 }
1716 1806
1717 /* now add the entry to our list(s) */ 1807 /* now add the entry to our list(s) */
1718 name = entp->d_name; 1808 name = D_NAME (entp);
1719 1809
1720 /* skip . and .. entries */ 1810 /* skip . and .. entries */
1721 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1811 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1722 { 1812 {
1723 int len = D_NAMLEN (entp) + 1; 1813 int len = D_NAMLEN (entp) + 1;
1724 1814
1725 while (ecb_expect_false (namesoffs + len > namesalloc)) 1815 while (ecb_expect_false (namesoffs + len > namesalloc))
1726 { 1816 {
1727 namesalloc *= 2; 1817 namesalloc *= 2;
1728 X_LOCK (wrklock);
1729 req->ptr2 = names = realloc (names, namesalloc); 1818 req->ptr2 = names = realloc (names, namesalloc);
1730 X_UNLOCK (wrklock);
1731 1819
1732 if (!names) 1820 if (!names)
1733 break; 1821 break;
1734 } 1822 }
1735 1823
1740 struct eio_dirent *ent; 1828 struct eio_dirent *ent;
1741 1829
1742 if (ecb_expect_false (dentoffs == dentalloc)) 1830 if (ecb_expect_false (dentoffs == dentalloc))
1743 { 1831 {
1744 dentalloc *= 2; 1832 dentalloc *= 2;
1745 X_LOCK (wrklock);
1746 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); 1833 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1747 X_UNLOCK (wrklock);
1748 1834
1749 if (!dents) 1835 if (!dents)
1750 break; 1836 break;
1751 } 1837 }
1752 1838
1832 if (EIO_CANCELLED (req)) 1918 if (EIO_CANCELLED (req))
1833 { 1919 {
1834 errno = ECANCELED; 1920 errno = ECANCELED;
1835 break; 1921 break;
1836 } 1922 }
1923
1924#ifdef _WIN32
1925 if (!FindNextFile (dirp, &entp))
1926 {
1927 FindClose (dirp);
1928 dirp = 0;
1929 }
1930#endif
1837 } 1931 }
1838} 1932}
1839 1933
1840/*****************************************************************************/ 1934/*****************************************************************************/
1841 1935
1857X_THREAD_PROC (etp_proc) 1951X_THREAD_PROC (etp_proc)
1858{ 1952{
1859 ETP_REQ *req; 1953 ETP_REQ *req;
1860 struct timespec ts; 1954 struct timespec ts;
1861 etp_worker *self = (etp_worker *)thr_arg; 1955 etp_worker *self = (etp_worker *)thr_arg;
1862 int timeout; 1956
1957#if HAVE_PRCTL_SET_NAME
1958 prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
1959#endif
1863 1960
1864 /* try to distribute timeouts somewhat evenly */ 1961 /* try to distribute timeouts somewhat evenly */
1865 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1962 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1866 1963
1867 for (;;) 1964 for (;;)
1937/*****************************************************************************/ 2034/*****************************************************************************/
1938 2035
1939int ecb_cold 2036int ecb_cold
1940eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2037eio_init (void (*want_poll)(void), void (*done_poll)(void))
1941{ 2038{
2039#if !HAVE_PREADWRITE
2040 X_MUTEX_CREATE (preadwritelock);
2041#endif
2042
1942 return etp_init (want_poll, done_poll); 2043 return etp_init (want_poll, done_poll);
1943} 2044}
1944 2045
1945ecb_inline void 2046ecb_inline void
1946eio_api_destroy (eio_req *req) 2047eio_api_destroy (eio_req *req)
1991 case EIO_WRITE: req->result = req->offs >= 0 2092 case EIO_WRITE: req->result = req->offs >= 0
1992 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2093 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1993 : write (req->int1, req->ptr2, req->size); break; 2094 : write (req->int1, req->ptr2, req->size); break;
1994 2095
1995 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2096 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1996 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2097 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1997 2098
1998 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2099 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1999 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2100 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
2000 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2101 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2001 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2102 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
2031 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break; 2132 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
2032 2133
2033 case EIO_SYNC: req->result = 0; sync (); break; 2134 case EIO_SYNC: req->result = 0; sync (); break;
2034 case EIO_FSYNC: req->result = fsync (req->int1); break; 2135 case EIO_FSYNC: req->result = fsync (req->int1); break;
2035 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2136 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2137 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2138 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2036 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 2139 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2037 case EIO_MTOUCH: req->result = eio__mtouch (req); break; 2140 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2038 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 2141 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2039 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 2142 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2040 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2041 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break; 2143 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2042 2144
2043 case EIO_READDIR: eio__scandir (req, self); break; 2145 case EIO_READDIR: eio__scandir (req, self); break;
2044 2146
2045 case EIO_BUSY: 2147 case EIO_BUSY:
2126eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2228eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2127{ 2229{
2128 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2230 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2129} 2231}
2130 2232
2233eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2234{
2235 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2236}
2237
2238eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2239{
2240 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2241}
2242
2243eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2244{
2245 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2246}
2247
2131eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2248eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2132{ 2249{
2133 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2250 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2134} 2251}
2135 2252
2141eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) 2258eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2142{ 2259{
2143 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2260 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2144} 2261}
2145 2262
2146eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2147{
2148 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2149}
2150
2151eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data) 2263eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2152{ 2264{
2153 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND; 2265 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2154}
2155
2156eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2157{
2158 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2159} 2266}
2160 2267
2161eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2268eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2162{ 2269{
2163 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2270 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2391/* misc garbage */ 2498/* misc garbage */
2392 2499
2393eio_ssize_t 2500eio_ssize_t
2394eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2501eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2395{ 2502{
2396 etp_worker wrk;
2397 eio_ssize_t ret;
2398
2399 wrk.dbuf = 0;
2400
2401 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2503 return eio__sendfile (ofd, ifd, offset, count);
2402
2403 if (wrk.dbuf)
2404 free (wrk.dbuf);
2405
2406 return ret;
2407} 2504}
2408 2505

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines