… | |
… | |
74 | #else |
74 | #else |
75 | |
75 | |
76 | # include "config.h" |
76 | # include "config.h" |
77 | # include <sys/time.h> |
77 | # include <sys/time.h> |
78 | # include <sys/select.h> |
78 | # include <sys/select.h> |
79 | # include <sys/mman.h> |
|
|
80 | # include <unistd.h> |
79 | # include <unistd.h> |
81 | # include <utime.h> |
80 | # include <utime.h> |
82 | # include <signal.h> |
81 | # include <signal.h> |
83 | # include <dirent.h> |
82 | # include <dirent.h> |
|
|
83 | |
|
|
84 | #if _POSIX_MEMLOCK || _POSIX_MAPPED_FILES |
|
|
85 | # include <sys/mman.h> |
|
|
86 | #endif |
84 | |
87 | |
85 | /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ |
88 | /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ |
86 | # if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ |
89 | # if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ |
87 | # define _DIRENT_HAVE_D_TYPE /* sigh */ |
90 | # define _DIRENT_HAVE_D_TYPE /* sigh */ |
88 | # define D_INO(de) (de)->d_fileno |
91 | # define D_INO(de) (de)->d_fileno |
… | |
… | |
137 | /* used for struct dirent, AIX doesn't provide it */ |
140 | /* used for struct dirent, AIX doesn't provide it */ |
138 | #ifndef NAME_MAX |
141 | #ifndef NAME_MAX |
139 | # define NAME_MAX 4096 |
142 | # define NAME_MAX 4096 |
140 | #endif |
143 | #endif |
141 | |
144 | |
|
|
145 | /* used for readlink etc. */ |
|
|
146 | #ifndef PATH_MAX |
|
|
147 | # define PATH_MAX 4096 |
|
|
148 | #endif |
|
|
149 | |
142 | /* buffer size for various temporary buffers */ |
150 | /* buffer size for various temporary buffers */ |
143 | #define EIO_BUFSIZE 65536 |
151 | #define EIO_BUFSIZE 65536 |
144 | |
152 | |
145 | #define dBUF \ |
153 | #define dBUF \ |
146 | char *eio_buf; \ |
154 | char *eio_buf; \ |
… | |
… | |
217 | static volatile unsigned int nreqs; /* reqlock */ |
225 | static volatile unsigned int nreqs; /* reqlock */ |
218 | static volatile unsigned int nready; /* reqlock */ |
226 | static volatile unsigned int nready; /* reqlock */ |
219 | static volatile unsigned int npending; /* reqlock */ |
227 | static volatile unsigned int npending; /* reqlock */ |
220 | static volatile unsigned int max_idle = 4; |
228 | static volatile unsigned int max_idle = 4; |
221 | |
229 | |
222 | static mutex_t wrklock = X_MUTEX_INIT; |
230 | static xmutex_t wrklock = X_MUTEX_INIT; |
223 | static mutex_t reslock = X_MUTEX_INIT; |
231 | static xmutex_t reslock = X_MUTEX_INIT; |
224 | static mutex_t reqlock = X_MUTEX_INIT; |
232 | static xmutex_t reqlock = X_MUTEX_INIT; |
225 | static cond_t reqwait = X_COND_INIT; |
233 | static xcond_t reqwait = X_COND_INIT; |
226 | |
234 | |
227 | #if !HAVE_PREADWRITE |
235 | #if !HAVE_PREADWRITE |
228 | /* |
236 | /* |
229 | * make our pread/pwrite emulation safe against themselves, but not against |
237 | * make our pread/pwrite emulation safe against themselves, but not against |
230 | * normal read/write by using a mutex. slows down execution a lot, |
238 | * normal read/write by using a mutex. slows down execution a lot, |
231 | * but that's your problem, not mine. |
239 | * but that's your problem, not mine. |
232 | */ |
240 | */ |
233 | static mutex_t preadwritelock = X_MUTEX_INIT; |
241 | static xmutex_t preadwritelock = X_MUTEX_INIT; |
234 | #endif |
242 | #endif |
235 | |
243 | |
236 | typedef struct etp_worker |
244 | typedef struct etp_worker |
237 | { |
245 | { |
238 | /* locked by wrklock */ |
246 | /* locked by wrklock */ |
239 | struct etp_worker *prev, *next; |
247 | struct etp_worker *prev, *next; |
240 | |
248 | |
241 | thread_t tid; |
249 | xthread_t tid; |
242 | |
250 | |
243 | /* locked by reslock, reqlock or wrklock */ |
251 | /* locked by reslock, reqlock or wrklock */ |
244 | ETP_REQ *req; /* currently processed request */ |
252 | ETP_REQ *req; /* currently processed request */ |
245 | |
253 | |
246 | ETP_WORKER_COMMON |
254 | ETP_WORKER_COMMON |
… | |
… | |
919 | */ |
927 | */ |
920 | { |
928 | { |
921 | off_t sbytes; |
929 | off_t sbytes; |
922 | res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
930 | res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
923 | |
931 | |
|
|
932 | #if 0 /* according to the manpage, this is correct, but broken behaviour */ |
924 | /* freebsd' sendfile will return 0 on success */ |
933 | /* freebsd' sendfile will return 0 on success */ |
925 | /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ |
934 | /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ |
926 | /* not on e.g. EIO or EPIPE - sounds broken */ |
935 | /* not on e.g. EIO or EPIPE - sounds broken */ |
927 | if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) |
936 | if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) |
|
|
937 | res = sbytes; |
|
|
938 | #endif |
|
|
939 | |
|
|
940 | /* according to source inspection, this is correct, and useful behaviour */ |
|
|
941 | if (sbytes) |
928 | res = sbytes; |
942 | res = sbytes; |
929 | } |
943 | } |
930 | |
944 | |
931 | # elif defined (__APPLE__) |
945 | # elif defined (__APPLE__) |
932 | |
946 | |
… | |
… | |
1381 | break; |
1395 | break; |
1382 | } |
1396 | } |
1383 | } |
1397 | } |
1384 | } |
1398 | } |
1385 | |
1399 | |
|
|
1400 | #ifdef PAGESIZE |
|
|
1401 | # define eio_pagesize() PAGESIZE |
|
|
1402 | #else |
|
|
1403 | static intptr_t |
|
|
1404 | eio_pagesize (void) |
|
|
1405 | { |
|
|
1406 | static intptr_t page; |
|
|
1407 | |
|
|
1408 | if (!page) |
|
|
1409 | page = sysconf (_SC_PAGESIZE); |
|
|
1410 | |
|
|
1411 | return page; |
|
|
1412 | } |
|
|
1413 | #endif |
|
|
1414 | |
|
|
1415 | static void |
|
|
1416 | eio_page_align (void **addr, size_t *length) |
|
|
1417 | { |
|
|
1418 | intptr_t mask = eio_pagesize () - 1; |
|
|
1419 | |
|
|
1420 | /* round down addr */ |
|
|
1421 | intptr_t adj = mask & (intptr_t)*addr; |
|
|
1422 | |
|
|
1423 | *addr = (void *)((intptr_t)*addr - adj); |
|
|
1424 | *length += adj; |
|
|
1425 | |
|
|
1426 | /* round up length */ |
|
|
1427 | *length = (*length + mask) & ~mask; |
|
|
1428 | } |
|
|
1429 | |
|
|
1430 | #if !_POSIX_MEMLOCK |
|
|
1431 | # define eio__mlock(a,b) ((errno = ENOSYS), -1) |
|
|
1432 | # define eio__mlockall(a) ((errno = ENOSYS), -1) |
|
|
1433 | #else |
|
|
1434 | |
|
|
1435 | static int |
|
|
1436 | eio__mlock (void *addr, size_t length) |
|
|
1437 | { |
|
|
1438 | eio_page_align (&addr, &length); |
|
|
1439 | |
|
|
1440 | return mlock (addr, length); |
|
|
1441 | } |
|
|
1442 | |
|
|
1443 | static int |
|
|
1444 | eio__mlockall (int flags) |
|
|
1445 | { |
|
|
1446 | #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7 |
|
|
1447 | extern int mallopt (int, int); |
|
|
1448 | mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */ |
|
|
1449 | #endif |
|
|
1450 | |
|
|
1451 | if (EIO_MCL_CURRENT != MCL_CURRENT |
|
|
1452 | || EIO_MCL_FUTURE != MCL_FUTURE) |
|
|
1453 | { |
|
|
1454 | flags = 0 |
|
|
1455 | | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0) |
|
|
1456 | | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0); |
|
|
1457 | } |
|
|
1458 | |
|
|
1459 | return mlockall (flags); |
|
|
1460 | } |
|
|
1461 | #endif |
|
|
1462 | |
1386 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
1463 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
1387 | # define eio__msync(a,b,c) ((errno = ENOSYS), -1) |
1464 | # define eio__msync(a,b,c) ((errno = ENOSYS), -1) |
1388 | #else |
1465 | #else |
1389 | |
1466 | |
1390 | int |
1467 | int |
1391 | eio__msync (void *mem, size_t len, int flags) |
1468 | eio__msync (void *mem, size_t len, int flags) |
1392 | { |
1469 | { |
|
|
1470 | eio_page_align (&mem, &len); |
|
|
1471 | |
1393 | if (EIO_MS_ASYNC != MS_SYNC |
1472 | if (EIO_MS_ASYNC != MS_SYNC |
1394 | || EIO_MS_INVALIDATE != MS_INVALIDATE |
1473 | || EIO_MS_INVALIDATE != MS_INVALIDATE |
1395 | || EIO_MS_SYNC != MS_SYNC) |
1474 | || EIO_MS_SYNC != MS_SYNC) |
1396 | { |
1475 | { |
1397 | flags = 0 |
1476 | flags = 0 |
… | |
… | |
1406 | #endif |
1485 | #endif |
1407 | |
1486 | |
1408 | int |
1487 | int |
1409 | eio__mtouch (void *mem, size_t len, int flags) |
1488 | eio__mtouch (void *mem, size_t len, int flags) |
1410 | { |
1489 | { |
|
|
1490 | eio_page_align (&mem, &len); |
|
|
1491 | |
|
|
1492 | { |
1411 | intptr_t addr = (intptr_t)mem; |
1493 | intptr_t addr = (intptr_t)mem; |
1412 | intptr_t end = addr + len; |
1494 | intptr_t end = addr + len; |
1413 | #ifdef PAGESIZE |
1495 | intptr_t page = eio_pagesize (); |
1414 | const intptr_t page = PAGESIZE; |
|
|
1415 | #else |
|
|
1416 | static intptr_t page; |
|
|
1417 | |
1496 | |
1418 | if (!page) |
|
|
1419 | page = sysconf (_SC_PAGESIZE); |
|
|
1420 | #endif |
|
|
1421 | |
|
|
1422 | /* round down to start of page, although this is probably useless */ |
|
|
1423 | addr &= ~(page - 1); /* assume page size is always a power of two */ |
|
|
1424 | |
|
|
1425 | if (addr < end) |
1497 | if (addr < end) |
1426 | if (flags & EIO_MT_MODIFY) /* modify */ |
1498 | if (flags & EIO_MT_MODIFY) /* modify */ |
1427 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
1499 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
1428 | else |
1500 | else |
1429 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
1501 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
|
|
1502 | } |
1430 | |
1503 | |
1431 | return 0; |
1504 | return 0; |
1432 | } |
1505 | } |
1433 | |
1506 | |
1434 | /*****************************************************************************/ |
1507 | /*****************************************************************************/ |
… | |
… | |
1602 | case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
1675 | case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
1603 | case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
1676 | case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
1604 | case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
1677 | case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
1605 | case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; |
1678 | case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; |
1606 | |
1679 | |
1607 | case EIO_READLINK: ALLOC (NAME_MAX); |
1680 | case EIO_READLINK: ALLOC (PATH_MAX); |
1608 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
1681 | req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break; |
1609 | |
1682 | |
1610 | case EIO_SYNC: req->result = 0; sync (); break; |
1683 | case EIO_SYNC: req->result = 0; sync (); break; |
1611 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1684 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1612 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1685 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1613 | case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; |
1686 | case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; |
1614 | case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; |
1687 | case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; |
|
|
1688 | case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; |
|
|
1689 | case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; |
1615 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
1690 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
1616 | |
1691 | |
1617 | case EIO_READDIR: eio__scandir (req, self); break; |
1692 | case EIO_READDIR: eio__scandir (req, self); break; |
1618 | |
1693 | |
1619 | case EIO_BUSY: |
1694 | case EIO_BUSY: |
1620 | #ifdef _WIN32 |
1695 | #ifdef _WIN32 |
1621 | Sleep (req->nv1 * 1000.); |
1696 | Sleep (req->nv1 * 1e3); |
1622 | #else |
1697 | #else |
1623 | { |
1698 | { |
1624 | struct timeval tv; |
1699 | struct timeval tv; |
1625 | |
1700 | |
1626 | tv.tv_sec = req->nv1; |
1701 | tv.tv_sec = req->nv1; |
1627 | tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; |
1702 | tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6; |
1628 | |
1703 | |
1629 | req->result = select (0, 0, 0, 0, &tv); |
1704 | req->result = select (0, 0, 0, 0, &tv); |
1630 | } |
1705 | } |
1631 | #endif |
1706 | #endif |
1632 | break; |
1707 | break; |
… | |
… | |
1705 | eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
1780 | eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
1706 | { |
1781 | { |
1707 | REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
1782 | REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
1708 | } |
1783 | } |
1709 | |
1784 | |
|
|
1785 | eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data) |
|
|
1786 | { |
|
|
1787 | REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND; |
|
|
1788 | } |
|
|
1789 | |
|
|
1790 | eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) |
|
|
1791 | { |
|
|
1792 | REQ (EIO_MLOCKALL); req->int1 = flags; SEND; |
|
|
1793 | } |
|
|
1794 | |
1710 | eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) |
1795 | eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) |
1711 | { |
1796 | { |
1712 | REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; |
1797 | REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; |
1713 | } |
1798 | } |
1714 | |
1799 | |