1 /* This file contains functions which implement those POSIX and Linux functions
2 * that MinGW and Microsoft don't provide. The implementations contain just enough
3 * functionality to support fio.
7 #include <netinet/in.h>
16 #include <semaphore.h>
20 #include <sys/resource.h>
23 #include "../os-windows.h"
24 #include "../../lib/hweight.h"
26 extern unsigned long mtime_since_now(struct timeval *);
27 extern void fio_gettime(struct timeval *, void *);
29 /* These aren't defined in the MinGW headers */
30 HRESULT WINAPI StringCchCopyA(
35 HRESULT WINAPI StringCchPrintfA(
38 const char *pszFormat,
43 size_t numberOfElements,
47 int win_to_posix_error(DWORD winerr)
51 case ERROR_FILE_NOT_FOUND: return ENOENT;
52 case ERROR_PATH_NOT_FOUND: return ENOENT;
53 case ERROR_ACCESS_DENIED: return EACCES;
54 case ERROR_INVALID_HANDLE: return EBADF;
55 case ERROR_NOT_ENOUGH_MEMORY: return ENOMEM;
56 case ERROR_INVALID_DATA: return EINVAL;
57 case ERROR_OUTOFMEMORY: return ENOMEM;
58 case ERROR_INVALID_DRIVE: return ENODEV;
59 case ERROR_NOT_SAME_DEVICE: return EXDEV;
60 case ERROR_WRITE_PROTECT: return EROFS;
61 case ERROR_BAD_UNIT: return ENODEV;
62 case ERROR_SHARING_VIOLATION: return EACCES;
63 case ERROR_LOCK_VIOLATION: return EACCES;
64 case ERROR_SHARING_BUFFER_EXCEEDED: return ENOLCK;
65 case ERROR_HANDLE_DISK_FULL: return ENOSPC;
66 case ERROR_NOT_SUPPORTED: return ENOSYS;
67 case ERROR_FILE_EXISTS: return EEXIST;
68 case ERROR_CANNOT_MAKE: return EPERM;
69 case ERROR_INVALID_PARAMETER: return EINVAL;
70 case ERROR_NO_PROC_SLOTS: return EAGAIN;
71 case ERROR_BROKEN_PIPE: return EPIPE;
72 case ERROR_OPEN_FAILED: return EIO;
73 case ERROR_NO_MORE_SEARCH_HANDLES: return ENFILE;
74 case ERROR_CALL_NOT_IMPLEMENTED: return ENOSYS;
75 case ERROR_INVALID_NAME: return ENOENT;
76 case ERROR_WAIT_NO_CHILDREN: return ECHILD;
77 case ERROR_CHILD_NOT_COMPLETE: return EBUSY;
78 case ERROR_DIR_NOT_EMPTY: return ENOTEMPTY;
79 case ERROR_SIGNAL_REFUSED: return EIO;
80 case ERROR_BAD_PATHNAME: return ENOENT;
81 case ERROR_SIGNAL_PENDING: return EBUSY;
82 case ERROR_MAX_THRDS_REACHED: return EAGAIN;
83 case ERROR_BUSY: return EBUSY;
84 case ERROR_ALREADY_EXISTS: return EEXIST;
85 case ERROR_NO_SIGNAL_SENT: return EIO;
86 case ERROR_FILENAME_EXCED_RANGE: return EINVAL;
87 case ERROR_META_EXPANSION_TOO_LONG: return EINVAL;
88 case ERROR_INVALID_SIGNAL_NUMBER: return EINVAL;
89 case ERROR_THREAD_1_INACTIVE: return EINVAL;
90 case ERROR_BAD_PIPE: return EINVAL;
91 case ERROR_PIPE_BUSY: return EBUSY;
92 case ERROR_NO_DATA: return EPIPE;
93 case ERROR_MORE_DATA: return EAGAIN;
94 case ERROR_DIRECTORY: return ENOTDIR;
95 case ERROR_PIPE_CONNECTED: return EBUSY;
96 case ERROR_NO_TOKEN: return EINVAL;
97 case ERROR_PROCESS_ABORTED: return EFAULT;
98 case ERROR_BAD_DEVICE: return ENODEV;
99 case ERROR_BAD_USERNAME: return EINVAL;
100 case ERROR_OPEN_FILES: return EAGAIN;
101 case ERROR_ACTIVE_CONNECTIONS: return EAGAIN;
102 case ERROR_DEVICE_IN_USE: return EAGAIN;
103 case ERROR_INVALID_AT_INTERRUPT_TIME: return EINTR;
104 case ERROR_IO_DEVICE: return EIO;
105 case ERROR_NOT_OWNER: return EPERM;
106 case ERROR_END_OF_MEDIA: return ENOSPC;
107 case ERROR_EOM_OVERFLOW: return ENOSPC;
108 case ERROR_BEGINNING_OF_MEDIA: return ESPIPE;
109 case ERROR_SETMARK_DETECTED: return ESPIPE;
110 case ERROR_NO_DATA_DETECTED: return ENOSPC;
111 case ERROR_POSSIBLE_DEADLOCK: return EDEADLOCK;
112 case ERROR_CRC: return EIO;
113 case ERROR_NEGATIVE_SEEK: return EINVAL;
114 case ERROR_DISK_FULL: return ENOSPC;
115 case ERROR_NOACCESS: return EFAULT;
116 case ERROR_FILE_INVALID: return ENXIO;
122 int GetNumLogicalProcessors(void)
124 SYSTEM_LOGICAL_PROCESSOR_INFORMATION *processor_info = NULL;
126 DWORD num_processors = 0;
130 while (!GetLogicalProcessorInformation(processor_info, &len)) {
131 error = GetLastError();
132 if (error == ERROR_INSUFFICIENT_BUFFER)
133 processor_info = malloc(len);
135 log_err("Error: GetLogicalProcessorInformation failed: %d\n", error);
139 if (processor_info == NULL) {
140 log_err("Error: failed to allocate memory for GetLogicalProcessorInformation");
145 for (i = 0; i < len / sizeof(SYSTEM_LOGICAL_PROCESSOR_INFORMATION); i++)
147 if (processor_info[i].Relationship == RelationProcessorCore)
148 num_processors += hweight64(processor_info[i].ProcessorMask);
151 free(processor_info);
152 return num_processors;
155 long sysconf(int name)
160 MEMORYSTATUSEX status;
164 case _SC_NPROCESSORS_ONLN:
165 val = GetNumLogicalProcessors();
167 log_err("sysconf(_SC_NPROCESSORS_ONLN) failed\n");
172 GetSystemInfo(&sysInfo);
173 val = sysInfo.dwPageSize;
177 status.dwLength = sizeof(status);
178 val2 = sysconf(_SC_PAGESIZE);
179 if (GlobalMemoryStatusEx(&status) && val2 != -1)
180 val = status.ullTotalPhys / val2;
182 log_err("sysconf(_SC_PHYS_PAGES) failed\n");
185 log_err("sysconf(%d) is not implemented\n", name);
192 char *dl_error = NULL;
194 int dlclose(void *handle)
196 return !FreeLibrary((HMODULE)handle);
199 void *dlopen(const char *file, int mode)
203 hMod = LoadLibrary(file);
204 if (hMod == INVALID_HANDLE_VALUE)
205 dl_error = (char*)"LoadLibrary failed";
212 void *dlsym(void *handle, const char *name)
216 fnPtr = GetProcAddress((HMODULE)handle, name);
218 dl_error = (char*)"GetProcAddress failed";
230 int gettimeofday(struct timeval *restrict tp, void *restrict tzp)
233 uint64_t unix_time, windows_time;
234 const uint64_t MILLISECONDS_BETWEEN_1601_AND_1970 = 11644473600000;
236 /* Ignore the timezone parameter */
240 * Windows time is stored as the number 100 ns intervals since January 1 1601.
241 * Conversion details from http://www.informit.com/articles/article.aspx?p=102236&seqNum=3
242 * Its precision is 100 ns but accuracy is only one clock tick, or normally around 15 ms.
244 GetSystemTimeAsFileTime(&fileTime);
245 windows_time = ((uint64_t)fileTime.dwHighDateTime << 32) + fileTime.dwLowDateTime;
246 /* Divide by 10,000 to convert to ms and subtract the time between 1601 and 1970 */
247 unix_time = (((windows_time)/10000) - MILLISECONDS_BETWEEN_1601_AND_1970);
248 /* unix_time is now the number of milliseconds since 1970 (the Unix epoch) */
249 tp->tv_sec = unix_time / 1000;
250 tp->tv_usec = (unix_time % 1000) * 1000;
254 int sigaction(int sig, const struct sigaction *act,
255 struct sigaction *oact)
258 void (*prev_handler)(int);
260 prev_handler = signal(sig, act->sa_handler);
262 oact->sa_handler = prev_handler;
264 if (prev_handler == SIG_ERR)
270 int lstat(const char * path, struct stat * buf)
272 return stat(path, buf);
275 void *mmap(void *addr, size_t len, int prot, int flags,
276 int fildes, off_t off)
279 void* allocAddr = NULL;
281 if (prot & PROT_NONE)
282 vaProt |= PAGE_NOACCESS;
284 if ((prot & PROT_READ) && !(prot & PROT_WRITE))
285 vaProt |= PAGE_READONLY;
287 if (prot & PROT_WRITE)
288 vaProt |= PAGE_READWRITE;
290 if ((flags & MAP_ANON) | (flags & MAP_ANONYMOUS))
292 allocAddr = VirtualAlloc(addr, len, MEM_COMMIT, vaProt);
293 if (allocAddr == NULL)
294 errno = win_to_posix_error(GetLastError());
300 int munmap(void *addr, size_t len)
302 if (!VirtualFree(addr, 0, MEM_RELEASE)) {
303 errno = win_to_posix_error(GetLastError());
312 log_err("%s is not implemented\n", __func__);
319 log_err("%s is not implemented\n", __func__);
324 static HANDLE log_file = INVALID_HANDLE_VALUE;
326 void openlog(const char *ident, int logopt, int facility)
328 if (log_file == INVALID_HANDLE_VALUE)
329 log_file = CreateFileA("syslog.txt", GENERIC_WRITE, FILE_SHARE_READ | FILE_SHARE_WRITE, NULL, OPEN_ALWAYS, 0, NULL);
334 CloseHandle(log_file);
335 log_file = INVALID_HANDLE_VALUE;
338 void syslog(int priority, const char *message, ... /* argument */)
345 if (log_file == INVALID_HANDLE_VALUE) {
346 log_file = CreateFileA("syslog.txt", GENERIC_WRITE, FILE_SHARE_READ | FILE_SHARE_WRITE, NULL, OPEN_ALWAYS, 0, NULL);
349 if (log_file == INVALID_HANDLE_VALUE) {
350 log_err("syslog: failed to open log file\n");
354 va_start(v, message);
355 len = _vscprintf(message, v);
356 output = malloc(len + sizeof(char));
357 vsprintf(output, message, v);
358 WriteFile(log_file, output, len, &bytes_written, NULL);
363 int kill(pid_t pid, int sig)
370 * This is assumed to be used only by the network code,
371 * and so doesn't try and handle any of the other cases
373 int fcntl(int fildes, int cmd, ...)
376 * non-blocking mode doesn't work the same as in BSD sockets,
381 int val, opt, status;
385 else if (cmd != F_SETFL) {
392 opt = va_arg(ap, int);
393 if (opt & O_NONBLOCK)
398 status = ioctlsocket((SOCKET)fildes, opt, &val);
400 if (status == SOCKET_ERROR) {
413 * Get the value of a local clock source.
414 * This implementation supports 2 clocks: CLOCK_MONOTONIC provides high-accuracy
415 * relative time, while CLOCK_REALTIME provides a low-accuracy wall time.
417 int clock_gettime(clockid_t clock_id, struct timespec *tp)
421 if (clock_id == CLOCK_MONOTONIC)
423 static LARGE_INTEGER freq = {{0,0}};
424 LARGE_INTEGER counts;
427 QueryPerformanceCounter(&counts);
428 if (freq.QuadPart == 0)
429 QueryPerformanceFrequency(&freq);
431 tp->tv_sec = counts.QuadPart / freq.QuadPart;
432 /* Get the difference between the number of ns stored
433 * in 'tv_sec' and that stored in 'counts' */
434 t = tp->tv_sec * freq.QuadPart;
435 t = counts.QuadPart - t;
436 /* 't' now contains the number of cycles since the last second.
437 * We want the number of nanoseconds, so multiply out by 1,000,000,000
438 * and then divide by the frequency. */
440 tp->tv_nsec = t / freq.QuadPart;
442 else if (clock_id == CLOCK_REALTIME)
444 /* clock_gettime(CLOCK_REALTIME,...) is just an alias for gettimeofday with a
445 * higher-precision field. */
447 gettimeofday(&tv, NULL);
448 tp->tv_sec = tv.tv_sec;
449 tp->tv_nsec = tv.tv_usec * 1000;
458 int mlock(const void * addr, size_t len)
462 HANDLE process = GetCurrentProcess();
464 success = GetProcessWorkingSetSize(process, &min, &max);
466 errno = win_to_posix_error(GetLastError());
472 success = SetProcessWorkingSetSize(process, min, max);
474 errno = win_to_posix_error(GetLastError());
478 success = VirtualLock((LPVOID)addr, len);
480 errno = win_to_posix_error(GetLastError());
487 int munlock(const void * addr, size_t len)
489 BOOL success = VirtualUnlock((LPVOID)addr, len);
491 errno = win_to_posix_error(GetLastError());
498 pid_t waitpid(pid_t pid, int *stat_loc, int options)
500 log_err("%s is not implemented\n", __func__);
505 int usleep(useconds_t useconds)
507 Sleep(useconds / 1000);
511 char *basename(char *path)
513 static char name[MAX_PATH];
516 if (path == NULL || strlen(path) == 0)
519 i = strlen(path) - 1;
521 while (path[i] != '\\' && path[i] != '/' && i >= 0)
524 strncpy(name, path + i + 1, MAX_PATH);
529 int fsync(int fildes)
531 HANDLE hFile = (HANDLE)_get_osfhandle(fildes);
532 if (!FlushFileBuffers(hFile)) {
533 errno = win_to_posix_error(GetLastError());
540 int nFileMappings = 0;
541 HANDLE fileMappings[1024];
543 int shmget(key_t key, size_t size, int shmflg)
546 uint32_t size_low = size & 0xFFFFFFFF;
547 uint32_t size_high = ((uint64_t)size) >> 32;
548 HANDLE hMapping = CreateFileMapping(INVALID_HANDLE_VALUE, NULL, (PAGE_EXECUTE_READWRITE | SEC_RESERVE), size_high, size_low, NULL);
549 if (hMapping != NULL) {
550 fileMappings[nFileMappings] = hMapping;
551 mapid = nFileMappings;
560 void *shmat(int shmid, const void *shmaddr, int shmflg)
563 MEMORY_BASIC_INFORMATION memInfo;
564 mapAddr = MapViewOfFile(fileMappings[shmid], FILE_MAP_ALL_ACCESS, 0, 0, 0);
565 if (mapAddr == NULL) {
566 errno = win_to_posix_error(GetLastError());
570 if (VirtualQuery(mapAddr, &memInfo, sizeof(memInfo)) == 0) {
571 errno = win_to_posix_error(GetLastError());
575 mapAddr = VirtualAlloc(mapAddr, memInfo.RegionSize, MEM_COMMIT, PAGE_READWRITE);
576 if (mapAddr == NULL) {
577 errno = win_to_posix_error(GetLastError());
584 int shmdt(const void *shmaddr)
586 if (!UnmapViewOfFile(shmaddr)) {
587 errno = win_to_posix_error(GetLastError());
594 int shmctl(int shmid, int cmd, struct shmid_ds *buf)
596 if (cmd == IPC_RMID) {
597 fileMappings[shmid] = INVALID_HANDLE_VALUE;
600 log_err("%s is not implemented\n", __func__);
606 int setuid(uid_t uid)
608 log_err("%s is not implemented\n", __func__);
613 int setgid(gid_t gid)
615 log_err("%s is not implemented\n", __func__);
630 int getrusage(int who, struct rusage *r_usage)
632 const uint64_t SECONDS_BETWEEN_1601_AND_1970 = 11644473600;
633 FILETIME cTime, eTime, kTime, uTime;
637 memset(r_usage, 0, sizeof(*r_usage));
639 if (who == RUSAGE_SELF) {
640 h = GetCurrentProcess();
641 GetProcessTimes(h, &cTime, &eTime, &kTime, &uTime);
642 } else if (who == RUSAGE_THREAD) {
643 h = GetCurrentThread();
644 GetThreadTimes(h, &cTime, &eTime, &kTime, &uTime);
646 log_err("fio: getrusage %d is not implemented\n", who);
650 time = ((uint64_t)uTime.dwHighDateTime << 32) + uTime.dwLowDateTime;
651 /* Divide by 10,000,000 to get the number of seconds and move the epoch from
653 time = (time_t)(((time)/10000000) - SECONDS_BETWEEN_1601_AND_1970);
654 r_usage->ru_utime.tv_sec = time;
655 /* getrusage() doesn't care about anything other than seconds, so set tv_usec to 0 */
656 r_usage->ru_utime.tv_usec = 0;
657 time = ((uint64_t)kTime.dwHighDateTime << 32) + kTime.dwLowDateTime;
658 /* Divide by 10,000,000 to get the number of seconds and move the epoch from
660 time = (time_t)(((time)/10000000) - SECONDS_BETWEEN_1601_AND_1970);
661 r_usage->ru_stime.tv_sec = time;
662 r_usage->ru_stime.tv_usec = 0;
666 int posix_madvise(void *addr, size_t len, int advice)
668 log_err("%s is not implemented\n", __func__);
672 /* Windows doesn't support advice for memory pages. Just ignore it. */
673 int msync(void *addr, size_t len, int flags)
679 int fdatasync(int fildes)
681 return fsync(fildes);
684 ssize_t pwrite(int fildes, const void *buf, size_t nbyte,
687 int64_t pos = _telli64(fildes);
688 ssize_t len = _write(fildes, buf, nbyte);
689 _lseeki64(fildes, pos, SEEK_SET);
693 ssize_t pread(int fildes, void *buf, size_t nbyte, off_t offset)
695 int64_t pos = _telli64(fildes);
696 ssize_t len = read(fildes, buf, nbyte);
697 _lseeki64(fildes, pos, SEEK_SET);
701 ssize_t readv(int fildes, const struct iovec *iov, int iovcnt)
703 log_err("%s is not implemented\n", __func__);
708 ssize_t writev(int fildes, const struct iovec *iov, int iovcnt)
710 log_err("%s is not implemented\n", __func__);
715 long long strtoll(const char *restrict str, char **restrict endptr,
718 return _strtoi64(str, endptr, base);
721 int poll(struct pollfd fds[], nfds_t nfds, int timeout)
724 struct timeval *to = NULL;
725 fd_set readfds, writefds, exceptfds;
731 to->tv_sec = timeout / 1000;
732 to->tv_usec = (timeout % 1000) * 1000;
739 for (i = 0; i < nfds; i++)
746 if (fds[i].events & POLLIN)
747 FD_SET(fds[i].fd, &readfds);
749 if (fds[i].events & POLLOUT)
750 FD_SET(fds[i].fd, &writefds);
752 FD_SET(fds[i].fd, &exceptfds);
755 rc = select(nfds, &readfds, &writefds, &exceptfds, to);
757 if (rc != SOCKET_ERROR) {
758 for (i = 0; i < nfds; i++)
764 if ((fds[i].events & POLLIN) && FD_ISSET(fds[i].fd, &readfds))
765 fds[i].revents |= POLLIN;
767 if ((fds[i].events & POLLOUT) && FD_ISSET(fds[i].fd, &writefds))
768 fds[i].revents |= POLLOUT;
770 if (FD_ISSET(fds[i].fd, &exceptfds))
771 fds[i].revents |= POLLHUP;
778 int nanosleep(const struct timespec *rqtp, struct timespec *rmtp)
782 DWORD ms_total = (rqtp->tv_sec * 1000) + (rqtp->tv_nsec / 1000000.0);
787 ms_remaining = ms_total;
789 /* Since Sleep() can sleep for less than the requested time, add a loop to
790 ensure we only return after the requested length of time has elapsed */
792 fio_gettime(&tv, NULL);
794 ms_remaining = ms_total - mtime_since_now(&tv);
795 } while (ms_remaining > 0 && ms_remaining < ms_total);
797 /* this implementation will never sleep for less than the requested time */
806 DIR *opendir(const char *dirname)
808 struct dirent_ctx *dc = NULL;
810 /* See if we can open it. If not, we'll return an error here */
811 HANDLE file = CreateFileA(dirname, 0, FILE_SHARE_READ | FILE_SHARE_WRITE, NULL, OPEN_EXISTING, FILE_FLAG_BACKUP_SEMANTICS, NULL);
812 if (file != INVALID_HANDLE_VALUE) {
814 dc = (struct dirent_ctx*)malloc(sizeof(struct dirent_ctx));
815 StringCchCopyA(dc->dirname, MAX_PATH, dirname);
816 dc->find_handle = INVALID_HANDLE_VALUE;
818 DWORD error = GetLastError();
819 if (error == ERROR_FILE_NOT_FOUND)
822 else if (error == ERROR_PATH_NOT_FOUND)
824 else if (error == ERROR_TOO_MANY_OPEN_FILES)
826 else if (error == ERROR_ACCESS_DENIED)
835 int closedir(DIR *dirp)
837 if (dirp != NULL && dirp->find_handle != INVALID_HANDLE_VALUE)
838 FindClose(dirp->find_handle);
844 struct dirent *readdir(DIR *dirp)
846 static struct dirent de;
847 WIN32_FIND_DATA find_data;
852 if (dirp->find_handle == INVALID_HANDLE_VALUE) {
853 char search_pattern[MAX_PATH];
854 StringCchPrintfA(search_pattern, MAX_PATH, "%s\\*", dirp->dirname);
855 dirp->find_handle = FindFirstFileA(search_pattern, &find_data);
856 if (dirp->find_handle == INVALID_HANDLE_VALUE)
859 if (!FindNextFile(dirp->find_handle, &find_data))
863 StringCchCopyA(de.d_name, MAX_PATH, find_data.cFileName);
871 log_err("%s is not implemented\n", __func__);
876 in_addr_t inet_network(const char *cp)
879 in_addr_t nbo = inet_addr(cp);
880 hbo = ((nbo & 0xFF) << 24) + ((nbo & 0xFF00) << 8) + ((nbo & 0xFF0000) >> 8) + ((nbo & 0xFF000000) >> 24);
884 const char* inet_ntop(int af, const void *restrict src,
885 char *restrict dst, socklen_t size)
887 INT status = SOCKET_ERROR;
891 if (af != AF_INET && af != AF_INET6) {
892 errno = EAFNOSUPPORT;
896 WSAStartup(MAKEWORD(2,2), &wsd);
899 struct sockaddr_in si;
901 memset(&si, 0, sizeof(si));
903 memcpy(&si.sin_addr, src, sizeof(si.sin_addr));
904 status = WSAAddressToString((struct sockaddr*)&si, sizeof(si), NULL, dst, &len);
905 } else if (af == AF_INET6) {
906 struct sockaddr_in6 si6;
908 memset(&si6, 0, sizeof(si6));
909 si6.sin6_family = af;
910 memcpy(&si6.sin6_addr, src, sizeof(si6.sin6_addr));
911 status = WSAAddressToString((struct sockaddr*)&si6, sizeof(si6), NULL, dst, &len);
914 if (status != SOCKET_ERROR)
924 int inet_pton(int af, const char *restrict src, void *restrict dst)
926 INT status = SOCKET_ERROR;
930 if (af != AF_INET && af != AF_INET6) {
931 errno = EAFNOSUPPORT;
935 WSAStartup(MAKEWORD(2,2), &wsd);
938 struct sockaddr_in si;
939 INT len = sizeof(si);
940 memset(&si, 0, sizeof(si));
942 status = WSAStringToAddressA((char*)src, af, NULL, (struct sockaddr*)&si, &len);
943 if (status != SOCKET_ERROR)
944 memcpy(dst, &si.sin_addr, sizeof(si.sin_addr));
945 } else if (af == AF_INET6) {
946 struct sockaddr_in6 si6;
947 INT len = sizeof(si6);
948 memset(&si6, 0, sizeof(si6));
949 si6.sin6_family = af;
950 status = WSAStringToAddressA((char*)src, af, NULL, (struct sockaddr*)&si6, &len);
951 if (status != SOCKET_ERROR)
952 memcpy(dst, &si6.sin6_addr, sizeof(si6.sin6_addr));
955 if (status == SOCKET_ERROR) {