… | |
… | |
120 | |
120 | |
121 | #define EIO_TICKS ((1000000 + 1023) >> 10) |
121 | #define EIO_TICKS ((1000000 + 1023) >> 10) |
122 | |
122 | |
123 | /*****************************************************************************/ |
123 | /*****************************************************************************/ |
124 | |
124 | |
|
|
125 | #if __GNUC__ >= 3 |
|
|
126 | # define expect(expr,value) __builtin_expect ((expr),(value)) |
|
|
127 | #else |
|
|
128 | # define expect(expr,value) (expr) |
|
|
129 | #endif |
|
|
130 | |
|
|
131 | #define expect_false(expr) expect ((expr) != 0, 0) |
|
|
132 | #define expect_true(expr) expect ((expr) != 0, 1) |
|
|
133 | |
|
|
134 | /*****************************************************************************/ |
|
|
135 | |
125 | #define ETP_PRI_MIN EIO_PRI_MIN |
136 | #define ETP_PRI_MIN EIO_PRI_MIN |
126 | #define ETP_PRI_MAX EIO_PRI_MAX |
137 | #define ETP_PRI_MAX EIO_PRI_MAX |
127 | |
138 | |
128 | struct etp_worker; |
139 | struct etp_worker; |
129 | |
140 | |
… | |
… | |
177 | static mutex_t wrklock = X_MUTEX_INIT; |
188 | static mutex_t wrklock = X_MUTEX_INIT; |
178 | static mutex_t reslock = X_MUTEX_INIT; |
189 | static mutex_t reslock = X_MUTEX_INIT; |
179 | static mutex_t reqlock = X_MUTEX_INIT; |
190 | static mutex_t reqlock = X_MUTEX_INIT; |
180 | static cond_t reqwait = X_COND_INIT; |
191 | static cond_t reqwait = X_COND_INIT; |
181 | |
192 | |
|
|
193 | #if !HAVE_PREADWRITE |
|
|
194 | /* |
|
|
195 | * make our pread/pwrite emulation safe against themselves, but not against |
|
|
196 | * normal read/write by using a mutex. slows down execution a lot, |
|
|
197 | * but that's your problem, not mine. |
|
|
198 | */ |
|
|
199 | static mutex_t preadwritelock = X_MUTEX_INIT; |
|
|
200 | #endif |
|
|
201 | |
182 | typedef struct etp_worker |
202 | typedef struct etp_worker |
183 | { |
203 | { |
184 | /* locked by wrklock */ |
204 | /* locked by wrklock */ |
185 | struct etp_worker *prev, *next; |
205 | struct etp_worker *prev, *next; |
186 | |
206 | |
… | |
… | |
330 | |
350 | |
331 | static void etp_atfork_child (void) |
351 | static void etp_atfork_child (void) |
332 | { |
352 | { |
333 | ETP_REQ *prv; |
353 | ETP_REQ *prv; |
334 | |
354 | |
335 | while (prv = reqq_shift (&req_queue)) |
355 | while ((prv = reqq_shift (&req_queue))) |
336 | ETP_DESTROY (prv); |
356 | ETP_DESTROY (prv); |
337 | |
357 | |
338 | while (prv = reqq_shift (&res_queue)) |
358 | while ((prv = reqq_shift (&res_queue))) |
339 | ETP_DESTROY (prv); |
359 | ETP_DESTROY (prv); |
340 | |
360 | |
341 | while (wrk_first.next != &wrk_first) |
361 | while (wrk_first.next != &wrk_first) |
342 | { |
362 | { |
343 | etp_worker *wrk = wrk_first.next; |
363 | etp_worker *wrk = wrk_first.next; |
… | |
… | |
371 | |
391 | |
372 | pthread_once (&doinit, etp_once_init); |
392 | pthread_once (&doinit, etp_once_init); |
373 | |
393 | |
374 | want_poll_cb = want_poll; |
394 | want_poll_cb = want_poll; |
375 | done_poll_cb = done_poll; |
395 | done_poll_cb = done_poll; |
|
|
396 | |
|
|
397 | return 0; |
376 | } |
398 | } |
377 | |
399 | |
378 | X_THREAD_PROC (etp_proc); |
400 | X_THREAD_PROC (etp_proc); |
379 | |
401 | |
380 | static void etp_start_thread (void) |
402 | static void etp_start_thread (void) |
… | |
… | |
400 | X_UNLOCK (wrklock); |
422 | X_UNLOCK (wrklock); |
401 | } |
423 | } |
402 | |
424 | |
403 | static void etp_maybe_start_thread (void) |
425 | static void etp_maybe_start_thread (void) |
404 | { |
426 | { |
405 | if (etp_nthreads () >= wanted) |
427 | if (expect_true (etp_nthreads () >= wanted)) |
406 | return; |
428 | return; |
407 | |
429 | |
408 | /* todo: maybe use idle here, but might be less exact */ |
430 | /* todo: maybe use idle here, but might be less exact */ |
409 | if (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()) |
431 | if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) |
410 | return; |
432 | return; |
411 | |
433 | |
412 | etp_start_thread (); |
434 | etp_start_thread (); |
413 | } |
435 | } |
414 | |
436 | |
… | |
… | |
467 | |
489 | |
468 | X_LOCK (reqlock); |
490 | X_LOCK (reqlock); |
469 | --nreqs; |
491 | --nreqs; |
470 | X_UNLOCK (reqlock); |
492 | X_UNLOCK (reqlock); |
471 | |
493 | |
472 | if (req->type == EIO_GROUP && req->size) |
494 | if (expect_false (req->type == EIO_GROUP && req->size)) |
473 | { |
495 | { |
474 | req->int1 = 1; /* mark request as delayed */ |
496 | req->int1 = 1; /* mark request as delayed */ |
475 | continue; |
497 | continue; |
476 | } |
498 | } |
477 | else |
499 | else |
478 | { |
500 | { |
479 | int res = ETP_FINISH (req); |
501 | int res = ETP_FINISH (req); |
480 | if (res) |
502 | if (expect_false (res)) |
481 | return res; |
503 | return res; |
482 | } |
504 | } |
483 | |
505 | |
484 | if (maxreqs && !--maxreqs) |
506 | if (expect_false (maxreqs && !--maxreqs)) |
485 | break; |
507 | break; |
486 | |
508 | |
487 | if (maxtime) |
509 | if (maxtime) |
488 | { |
510 | { |
489 | gettimeofday (&tv_now, 0); |
511 | gettimeofday (&tv_now, 0); |
… | |
… | |
508 | |
530 | |
509 | static void etp_submit (ETP_REQ *req) |
531 | static void etp_submit (ETP_REQ *req) |
510 | { |
532 | { |
511 | req->pri -= ETP_PRI_MIN; |
533 | req->pri -= ETP_PRI_MIN; |
512 | |
534 | |
513 | if (req->pri < ETP_PRI_MIN - ETP_PRI_MIN) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; |
535 | if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; |
514 | if (req->pri > ETP_PRI_MAX - ETP_PRI_MIN) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
536 | if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
515 | |
537 | |
|
|
538 | if (expect_false (req->type == EIO_GROUP)) |
|
|
539 | { |
|
|
540 | /* I hope this is worth it :/ */ |
516 | X_LOCK (reqlock); |
541 | X_LOCK (reqlock); |
517 | ++nreqs; |
542 | ++nreqs; |
|
|
543 | X_UNLOCK (reqlock); |
|
|
544 | |
|
|
545 | X_LOCK (reslock); |
|
|
546 | |
|
|
547 | ++npending; |
|
|
548 | |
|
|
549 | if (!reqq_push (&res_queue, req) && want_poll_cb) |
|
|
550 | want_poll_cb (); |
|
|
551 | |
|
|
552 | X_UNLOCK (reslock); |
|
|
553 | } |
|
|
554 | else |
|
|
555 | { |
|
|
556 | X_LOCK (reqlock); |
|
|
557 | ++nreqs; |
518 | ++nready; |
558 | ++nready; |
519 | reqq_push (&req_queue, req); |
559 | reqq_push (&req_queue, req); |
520 | X_COND_SIGNAL (reqwait); |
560 | X_COND_SIGNAL (reqwait); |
521 | X_UNLOCK (reqlock); |
561 | X_UNLOCK (reqlock); |
522 | |
562 | |
523 | etp_maybe_start_thread (); |
563 | etp_maybe_start_thread (); |
|
|
564 | } |
524 | } |
565 | } |
525 | |
566 | |
526 | static void etp_set_max_poll_time (double nseconds) |
567 | static void etp_set_max_poll_time (double nseconds) |
527 | { |
568 | { |
528 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
569 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
… | |
… | |
563 | |
604 | |
564 | static void grp_try_feed (eio_req *grp) |
605 | static void grp_try_feed (eio_req *grp) |
565 | { |
606 | { |
566 | while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) |
607 | while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) |
567 | { |
608 | { |
568 | int old_len = grp->size; |
609 | grp->flags &= ~EIO_FLAG_GROUPADD; |
569 | |
610 | |
570 | EIO_FEED (grp); |
611 | EIO_FEED (grp); |
571 | |
612 | |
572 | /* stop if no progress has been made */ |
613 | /* stop if no progress has been made */ |
573 | if (old_len == grp->size) |
614 | if (!(grp->flags & EIO_FLAG_GROUPADD)) |
574 | { |
615 | { |
575 | grp->feed = 0; |
616 | grp->feed = 0; |
576 | break; |
617 | break; |
577 | } |
618 | } |
578 | } |
619 | } |
… | |
… | |
697 | /* work around various missing functions */ |
738 | /* work around various missing functions */ |
698 | |
739 | |
699 | #if !HAVE_PREADWRITE |
740 | #if !HAVE_PREADWRITE |
700 | # define pread eio__pread |
741 | # define pread eio__pread |
701 | # define pwrite eio__pwrite |
742 | # define pwrite eio__pwrite |
702 | |
|
|
703 | /* |
|
|
704 | * make our pread/pwrite safe against themselves, but not against |
|
|
705 | * normal read/write by using a mutex. slows down execution a lot, |
|
|
706 | * but that's your problem, not mine. |
|
|
707 | */ |
|
|
708 | static mutex_t preadwritelock = X_MUTEX_INIT; |
|
|
709 | |
743 | |
710 | static ssize_t |
744 | static ssize_t |
711 | eio__pread (int fd, void *buf, size_t count, off_t offset) |
745 | eio__pread (int fd, void *buf, size_t count, off_t offset) |
712 | { |
746 | { |
713 | ssize_t res; |
747 | ssize_t res; |
… | |
… | |
775 | |
809 | |
776 | #if !HAVE_READAHEAD |
810 | #if !HAVE_READAHEAD |
777 | # define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) |
811 | # define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) |
778 | |
812 | |
779 | static ssize_t |
813 | static ssize_t |
780 | eio__readahead (int fd, off_t offset, size_t count, worker *self) |
814 | eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) |
781 | { |
815 | { |
782 | size_t todo = count; |
816 | size_t todo = count; |
783 | dBUF; |
817 | dBUF; |
784 | |
818 | |
785 | while (todo > 0) |
819 | while (todo > 0) |
… | |
… | |
901 | int memlen = 4096; |
935 | int memlen = 4096; |
902 | int memofs = 0; |
936 | int memofs = 0; |
903 | int res = 0; |
937 | int res = 0; |
904 | |
938 | |
905 | X_LOCK (wrklock); |
939 | X_LOCK (wrklock); |
|
|
940 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
906 | self->dirp = dirp = opendir (req->ptr1); |
941 | self->dirp = dirp = opendir (req->ptr1); |
907 | req->flags |= EIO_FLAG_PTR2_FREE; |
942 | req->flags |= EIO_FLAG_PTR2_FREE; |
908 | req->ptr2 = names = malloc (memlen); |
943 | req->ptr2 = names = malloc (memlen); |
909 | X_UNLOCK (wrklock); |
944 | X_UNLOCK (wrklock); |
910 | |
945 | |
… | |
… | |
943 | |
978 | |
944 | if (errno) |
979 | if (errno) |
945 | res = -1; |
980 | res = -1; |
946 | |
981 | |
947 | req->result = res; |
982 | req->result = res; |
|
|
983 | } |
|
|
984 | |
|
|
985 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
|
|
986 | # define msync(a,b,c) ENOSYS |
|
|
987 | #endif |
|
|
988 | |
|
|
989 | int |
|
|
990 | eio__mtouch (void *mem, size_t len, int flags) |
|
|
991 | { |
|
|
992 | intptr_t addr = (intptr_t)mem; |
|
|
993 | intptr_t end = addr + len; |
|
|
994 | #ifdef PAGESIZE |
|
|
995 | const intptr_t page = PAGESIZE; |
|
|
996 | #else |
|
|
997 | static intptr_t page; |
|
|
998 | |
|
|
999 | if (!page) |
|
|
1000 | page = sysconf (_SC_PAGESIZE); |
|
|
1001 | #endif |
|
|
1002 | |
|
|
1003 | addr &= ~(page - 1); /* assume page size is always a power of two */ |
|
|
1004 | |
|
|
1005 | if (addr < end) |
|
|
1006 | if (flags) /* modify */ |
|
|
1007 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
|
|
1008 | else |
|
|
1009 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
|
|
1010 | |
|
|
1011 | return 0; |
948 | } |
1012 | } |
949 | |
1013 | |
950 | /*****************************************************************************/ |
1014 | /*****************************************************************************/ |
951 | |
1015 | |
952 | #define ALLOC(len) \ |
1016 | #define ALLOC(len) \ |
… | |
… | |
1039 | |
1103 | |
1040 | /*****************************************************************************/ |
1104 | /*****************************************************************************/ |
1041 | |
1105 | |
1042 | int eio_init (void (*want_poll)(void), void (*done_poll)(void)) |
1106 | int eio_init (void (*want_poll)(void), void (*done_poll)(void)) |
1043 | { |
1107 | { |
1044 | etp_init (want_poll, done_poll); |
1108 | return etp_init (want_poll, done_poll); |
1045 | } |
1109 | } |
1046 | |
1110 | |
1047 | static void eio_api_destroy (eio_req *req) |
1111 | static void eio_api_destroy (eio_req *req) |
1048 | { |
1112 | { |
1049 | free (req); |
1113 | free (req); |
… | |
… | |
1111 | case EIO_RMDIR: req->result = rmdir (req->ptr1); break; |
1175 | case EIO_RMDIR: req->result = rmdir (req->ptr1); break; |
1112 | case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; |
1176 | case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; |
1113 | case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
1177 | case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
1114 | case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
1178 | case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
1115 | case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
1179 | case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
1116 | case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break; |
1180 | case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; |
1117 | |
1181 | |
1118 | case EIO_READLINK: ALLOC (NAME_MAX); |
1182 | case EIO_READLINK: ALLOC (NAME_MAX); |
1119 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
1183 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
1120 | |
1184 | |
1121 | case EIO_SYNC: req->result = 0; sync (); break; |
1185 | case EIO_SYNC: req->result = 0; sync (); break; |
1122 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1186 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1123 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1187 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
|
|
1188 | case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; |
|
|
1189 | case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; |
1124 | |
1190 | |
1125 | case EIO_READDIR: eio__scandir (req, self); break; |
1191 | case EIO_READDIR: eio__scandir (req, self); break; |
1126 | |
1192 | |
1127 | case EIO_BUSY: |
1193 | case EIO_BUSY: |
1128 | #ifdef _WIN32 |
1194 | #ifdef _WIN32 |
… | |
… | |
1160 | |
1226 | |
1161 | req->result = req->type == EIO_FUTIME |
1227 | req->result = req->type == EIO_FUTIME |
1162 | ? futimes (req->int1, times) |
1228 | ? futimes (req->int1, times) |
1163 | : utimes (req->ptr1, times); |
1229 | : utimes (req->ptr1, times); |
1164 | } |
1230 | } |
|
|
1231 | break; |
1165 | |
1232 | |
1166 | case EIO_GROUP: |
1233 | case EIO_GROUP: |
|
|
1234 | abort (); /* handled in eio_request */ |
|
|
1235 | |
1167 | case EIO_NOP: |
1236 | case EIO_NOP: |
1168 | req->result = 0; |
1237 | req->result = 0; |
1169 | break; |
1238 | break; |
1170 | |
1239 | |
1171 | case EIO_CUSTOM: |
1240 | case EIO_CUSTOM: |
1172 | req->feed (req); |
1241 | ((void (*)(eio_req *))req->feed) (req); |
1173 | break; |
1242 | break; |
1174 | |
1243 | |
1175 | default: |
1244 | default: |
1176 | req->result = -1; |
1245 | req->result = -1; |
1177 | break; |
1246 | break; |
… | |
… | |
1200 | eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) |
1269 | eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) |
1201 | { |
1270 | { |
1202 | REQ (EIO_FSYNC); req->int1 = fd; SEND; |
1271 | REQ (EIO_FSYNC); req->int1 = fd; SEND; |
1203 | } |
1272 | } |
1204 | |
1273 | |
|
|
1274 | eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
|
|
1275 | { |
|
|
1276 | REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
|
|
1277 | } |
|
|
1278 | |
|
|
1279 | eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
|
|
1280 | { |
|
|
1281 | REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
|
|
1282 | } |
|
|
1283 | |
1205 | eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) |
1284 | eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) |
1206 | { |
1285 | { |
1207 | REQ (EIO_FDATASYNC); req->int1 = fd; SEND; |
1286 | REQ (EIO_FDATASYNC); req->int1 = fd; SEND; |
1208 | } |
1287 | } |
1209 | |
1288 | |
… | |
… | |
1328 | return eio__1path (EIO_READDIR, path, pri, cb, data); |
1407 | return eio__1path (EIO_READDIR, path, pri, cb, data); |
1329 | } |
1408 | } |
1330 | |
1409 | |
1331 | eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
1410 | eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
1332 | { |
1411 | { |
1333 | REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; |
1412 | REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; |
1334 | } |
1413 | } |
1335 | |
1414 | |
1336 | static eio_req * |
1415 | static eio_req * |
1337 | eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1416 | eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1338 | { |
1417 | { |
… | |
… | |
1364 | return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); |
1443 | return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); |
1365 | } |
1444 | } |
1366 | |
1445 | |
1367 | eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) |
1446 | eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) |
1368 | { |
1447 | { |
1369 | REQ (EIO_CUSTOM); req->feed = execute; SEND; |
1448 | REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; |
1370 | } |
1449 | } |
1371 | |
1450 | |
1372 | #endif |
1451 | #endif |
1373 | |
1452 | |
1374 | eio_req *eio_grp (eio_cb cb, void *data) |
1453 | eio_req *eio_grp (eio_cb cb, void *data) |
… | |
… | |
1402 | |
1481 | |
1403 | void eio_grp_add (eio_req *grp, eio_req *req) |
1482 | void eio_grp_add (eio_req *grp, eio_req *req) |
1404 | { |
1483 | { |
1405 | assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); |
1484 | assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); |
1406 | |
1485 | |
|
|
1486 | grp->flags |= EIO_FLAG_GROUPADD; |
|
|
1487 | |
1407 | ++grp->size; |
1488 | ++grp->size; |
1408 | req->grp = grp; |
1489 | req->grp = grp; |
1409 | |
1490 | |
1410 | req->grp_prev = 0; |
1491 | req->grp_prev = 0; |
1411 | req->grp_next = grp->grp_first; |
1492 | req->grp_next = grp->grp_first; |