/* ArduPilot filesystem interface for systems using the FATFS filesystem */ #include "AP_Filesystem.h" #include #include #if HAVE_FILESYSTEM_SUPPORT && CONFIG_HAL_BOARD == HAL_BOARD_CHIBIOS #include #if 0 #define debug(fmt, args ...) do {printf("%s:%d: " fmt "\n", __FUNCTION__, __LINE__, ## args); } while(0) #else #define debug(fmt, args ...) #endif extern const AP_HAL::HAL& hal; static bool remount_needed; #define FATFS_R (S_IRUSR | S_IRGRP | S_IROTH) /*< FatFs Read perms */ #define FATFS_W (S_IWUSR | S_IWGRP | S_IWOTH) /*< FatFs Write perms */ #define FATFS_X (S_IXUSR | S_IXGRP | S_IXOTH) /*< FatFs Execute perms */ // use a semaphore to ensure that only one filesystem operation is // happening at a time. A recursive semaphore is used to cope with the // mkdir() inside sdcard_retry() static HAL_Semaphore_Recursive sem; typedef struct { FIL *fh; char *name; } FAT_FILE; #define MAX_FILES 16 static FAT_FILE *file_table[MAX_FILES]; static bool isatty(int fileno) { if (fileno >= 0 && fileno <= 2) { return true; } return false; } /* allocate a file descriptor */ static int new_file_descriptor(const char *pathname) { int i; FAT_FILE *stream; FIL *fh; for (i=0; iname = fname; file_table[i] = stream; stream->fh = fh; return i; } } errno = ENFILE; return -1; } static FAT_FILE *fileno_to_stream(int fileno) { FAT_FILE *stream; if (fileno < 0 || fileno >= MAX_FILES) { errno = EBADF; return nullptr; } stream = file_table[fileno]; if (stream == NULL) { errno = EBADF; return nullptr; } return stream; } static int free_file_descriptor(int fileno) { FAT_FILE *stream; FIL *fh; if (isatty( fileno )) { errno = EBADF; return -1; } // checks if fileno out of bounds stream = fileno_to_stream(fileno); if (stream == NULL) { return -1; } fh = stream->fh; if (fh != NULL) { free(fh); } free(stream->name); stream->name = NULL; file_table[fileno] = NULL; free(stream); return fileno; } static FIL *fileno_to_fatfs(int fileno) { FAT_FILE *stream; FIL *fh; if (isatty( fileno )) { errno = EBADF; return nullptr; } // checks if fileno out of bounds stream = fileno_to_stream(fileno); if ( stream == NULL ) { return nullptr; } fh = stream->fh; if (fh == NULL) { errno = EBADF; return nullptr; } return fh; } static int fatfs_to_errno( FRESULT Result ) { switch ( Result ) { case FR_OK: /* FatFS (0) Succeeded */ return 0; /* POSIX OK */ case FR_DISK_ERR: /* FatFS (1) A hard error occurred in the low level disk I/O layer */ return EIO; /* POSIX Input/output error (POSIX.1) */ case FR_INT_ERR: /* FatFS (2) Assertion failed */ return EPERM; /* POSIX Operation not permitted (POSIX.1) */ case FR_NOT_READY: /* FatFS (3) The physical drive cannot work */ return EBUSY; /* POSIX Device or resource busy (POSIX.1) */ case FR_NO_FILE: /* FatFS (4) Could not find the file */ return ENOENT; /* POSIX No such file or directory (POSIX.1) */ case FR_NO_PATH: /* FatFS (5) Could not find the path */ return ENOENT; /* POSIX No such file or directory (POSIX.1) */ case FR_INVALID_NAME: /* FatFS (6) The path name format is invalid */ return EINVAL; /* POSIX Invalid argument (POSIX.1) */ case FR_DENIED: /* FatFS (7) Access denied due to prohibited access or directory full */ return EACCES; /* POSIX Permission denied (POSIX.1) */ case FR_EXIST: /* file exists */ return EEXIST; /* file exists */ case FR_INVALID_OBJECT: /* FatFS (9) The file/directory object is invalid */ return EINVAL; /* POSIX Invalid argument (POSIX.1) */ case FR_WRITE_PROTECTED: /* FatFS (10) The physical drive is write protected */ return EROFS; /* POSIX Read-only filesystem (POSIX.1) */ case FR_INVALID_DRIVE: /* FatFS (11) The logical drive number is invalid */ return ENXIO; /* POSIX No such device or address (POSIX.1) */ case FR_NOT_ENABLED: /* FatFS (12) The volume has no work area */ return ENOSPC; /* POSIX No space left on device (POSIX.1) */ case FR_NO_FILESYSTEM: /* FatFS (13) There is no valid FAT volume */ return ENXIO; /* POSIX No such device or address (POSIX.1) */ case FR_MKFS_ABORTED: /* FatFS (14) The f_mkfs() aborted due to any parameter error */ return EINVAL; /* POSIX Invalid argument (POSIX.1) */ case FR_TIMEOUT: /* FatFS (15) Could not get a grant to access the volume within defined period */ return EBUSY; /* POSIX Device or resource busy (POSIX.1) */ case FR_LOCKED: /* FatFS (16) The operation is rejected according to the file sharing policy */ return EBUSY; /* POSIX Device or resource busy (POSIX.1) */ case FR_NOT_ENOUGH_CORE: /* FatFS (17) LFN working buffer could not be allocated */ return ENOMEM; /* POSIX Not enough space (POSIX.1) */ case FR_TOO_MANY_OPEN_FILES:/* FatFS (18) Number of open files > _FS_SHARE */ return EMFILE; /* POSIX Too many open files (POSIX.1) */ case FR_INVALID_PARAMETER:/* FatFS (19) Given parameter is invalid */ return EINVAL; /* POSIX Invalid argument (POSIX.1) */ } return EBADMSG; /* POSIX Bad message (POSIX.1) */ } // check for a remount and return -1 if it fails #define CHECK_REMOUNT() do { if (remount_needed && !remount_file_system()) { errno = EIO; return -1; }} while (0) #define CHECK_REMOUNT_NULL() do { if (remount_needed && !remount_file_system()) { errno = EIO; return NULL; }} while (0) // we allow for IO retries if either not armed or not in main thread #define RETRY_ALLOWED() (!hal.scheduler->in_main_thread() || !hal.util->get_soft_armed()) /* try to remount the file system on disk error */ static bool remount_file_system(void) { EXPECT_DELAY_MS(3000); if (!remount_needed) { sdcard_stop(); } if (!sdcard_retry()) { remount_needed = true; EXPECT_DELAY_MS(0); return false; } remount_needed = false; for (uint16_t i=0; ifh; FSIZE_t offset = fh->fptr; uint8_t flags = fh->flag & (FA_READ | FA_WRITE); memset(fh, 0, sizeof(*fh)); if (flags & FA_WRITE) { // the file may not have been created yet on the sdcard flags |= FA_OPEN_ALWAYS; } FRESULT res = f_open(fh, f->name, flags); debug("reopen %s flags=0x%x ofs=%u -> %d\n", f->name, unsigned(flags), unsigned(offset), int(res)); if (res == FR_OK) { f_lseek(fh, offset); } } EXPECT_DELAY_MS(0); return true; } int AP_Filesystem::open(const char *pathname, int flags) { int fileno; int fatfs_modes; FAT_FILE *stream; FIL *fh; int res; WITH_SEMAPHORE(sem); CHECK_REMOUNT(); errno = 0; debug("Open %s 0x%x", pathname, flags); if ((flags & O_ACCMODE) == O_RDWR) { fatfs_modes = FA_READ | FA_WRITE; } else if ((flags & O_ACCMODE) == O_RDONLY) { fatfs_modes = FA_READ; } else { fatfs_modes = FA_WRITE; } if (flags & O_CREAT) { if (flags & O_TRUNC) { fatfs_modes |= FA_CREATE_ALWAYS; } else { fatfs_modes |= FA_OPEN_ALWAYS; } } fileno = new_file_descriptor(pathname); // checks if fileno out of bounds stream = fileno_to_stream(fileno); if (stream == NULL) { free_file_descriptor(fileno); return -1; } // fileno_to_fatfs checks for fileno out of bounds fh = fileno_to_fatfs(fileno); if (fh == NULL) { free_file_descriptor(fileno); errno = EBADF; return -1; } res = f_open(fh, pathname, (BYTE) (fatfs_modes & 0xff)); if (res == FR_DISK_ERR && RETRY_ALLOWED()) { // one retry on disk error hal.scheduler->delay(100); if (remount_file_system()) { res = f_open(fh, pathname, (BYTE) (fatfs_modes & 0xff)); } } if (res != FR_OK) { errno = fatfs_to_errno((FRESULT)res); free_file_descriptor(fileno); return -1; } if (flags & O_APPEND) { /// Seek to end of the file res = f_lseek(fh, f_size(fh)); if (res != FR_OK) { errno = fatfs_to_errno((FRESULT)res); f_close(fh); free_file_descriptor(fileno); return -1; } } debug("Open %s -> %d", pathname, fileno); return fileno; } int AP_Filesystem::close(int fileno) { FAT_FILE *stream; FIL *fh; int res; WITH_SEMAPHORE(sem); errno = 0; // checks if fileno out of bounds stream = fileno_to_stream(fileno); if (stream == NULL) { return -1; } // fileno_to_fatfs checks for fileno out of bounds fh = fileno_to_fatfs(fileno); if (fh == NULL) { return -1; } res = f_close(fh); free_file_descriptor(fileno); if (res != FR_OK) { errno = fatfs_to_errno((FRESULT)res); return -1; } return 0; } ssize_t AP_Filesystem::read(int fd, void *buf, size_t count) { UINT size; UINT bytes = count; int res; FIL *fh; WITH_SEMAPHORE(sem); CHECK_REMOUNT(); if (count > 0) { *(char *) buf = 0; } errno = 0; // fileno_to_fatfs checks for fd out of bounds fh = fileno_to_fatfs(fd); if ( fh == NULL ) { errno = EBADF; return -1; } res = f_read(fh, (void *) buf, bytes, &size); if (res != FR_OK) { errno = fatfs_to_errno((FRESULT)res); return -1; } return (ssize_t)size; } ssize_t AP_Filesystem::write(int fd, const void *buf, size_t count) { UINT size; UINT bytes = count; FRESULT res; FIL *fh; errno = 0; WITH_SEMAPHORE(sem); CHECK_REMOUNT(); // fileno_to_fatfs checks for fd out of bounds fh = fileno_to_fatfs(fd); if ( fh == NULL ) { errno = EBADF; return -1; } res = f_write(fh, buf, bytes, &size); if (res == FR_DISK_ERR && RETRY_ALLOWED()) { // one retry on disk error hal.scheduler->delay(100); if (remount_file_system()) { res = f_write(fh, buf, bytes, &size); } } if (res != FR_OK) { errno = fatfs_to_errno(res); return -1; } return (ssize_t)size; } int AP_Filesystem::fsync(int fileno) { FAT_FILE *stream; FIL *fh; int res; WITH_SEMAPHORE(sem); errno = 0; // checks if fileno out of bounds stream = fileno_to_stream(fileno); if (stream == NULL) { return -1; } // fileno_to_fatfs checks for fileno out of bounds fh = fileno_to_fatfs(fileno); if (fh == NULL) { return -1; } res = f_sync(fh); if (res != FR_OK) { errno = fatfs_to_errno((FRESULT)res); return -1; } return 0; } off_t AP_Filesystem::lseek(int fileno, off_t position, int whence) { FRESULT res; FIL *fh; errno = 0; WITH_SEMAPHORE(sem); // fileno_to_fatfs checks for fd out of bounds fh = fileno_to_fatfs(fileno); if (fh == NULL) { errno = EMFILE; return -1; } if (isatty(fileno)) { return -1; } if (whence == SEEK_END) { position += f_size(fh); } else if (whence==SEEK_CUR) { position += fh->fptr; } res = f_lseek(fh, position); if (res) { errno = fatfs_to_errno(res); return -1; } return fh->fptr; } /* mktime replacement from Samba */ static time_t replace_mktime(const struct tm *t) { time_t epoch = 0; int n; int mon [] = { 31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31 }, y, m, i; const unsigned MINUTE = 60; const unsigned HOUR = 60*MINUTE; const unsigned DAY = 24*HOUR; const unsigned YEAR = 365*DAY; if (t->tm_year < 70) { return (time_t)-1; } n = t->tm_year + 1900 - 1; epoch = (t->tm_year - 70) * YEAR + ((n / 4 - n / 100 + n / 400) - (1969 / 4 - 1969 / 100 + 1969 / 400)) * DAY; y = t->tm_year + 1900; m = 0; for (i = 0; i < t->tm_mon; i++) { epoch += mon [m] * DAY; if (m == 1 && y % 4 == 0 && (y % 100 != 0 || y % 400 == 0)) { epoch += DAY; } if (++m > 11) { m = 0; y++; } } epoch += (t->tm_mday - 1) * DAY; epoch += t->tm_hour * HOUR + t->tm_min * MINUTE + t->tm_sec; return epoch; } static time_t fat_time_to_unix(uint16_t date, uint16_t time) { struct tm tp; time_t unix; memset(&tp, 0, sizeof(struct tm)); tp.tm_sec = (time << 1) & 0x3e; // 2 second resolution tp.tm_min = ((time >> 5) & 0x3f); tp.tm_hour = ((time >> 11) & 0x1f); tp.tm_mday = (date & 0x1f); tp.tm_mon = ((date >> 5) & 0x0f) - 1; tp.tm_year = ((date >> 9) & 0x7f) + 80; unix = replace_mktime( &tp ); return unix; } int AP_Filesystem::stat(const char *name, struct stat *buf) { FILINFO info; int res; time_t epoch; uint16_t mode; WITH_SEMAPHORE(sem); CHECK_REMOUNT(); errno = 0; // f_stat does not handle / or . as root directory if (strcmp(name,"/") == 0 || strcmp(name,".") == 0) { buf->st_atime = 0; buf->st_mtime = 0; buf->st_ctime = 0; buf->st_uid= 0; buf->st_gid= 0; buf->st_size = 0; buf->st_mode = S_IFDIR; return 0; } res = f_stat(name, &info); if (res == FR_DISK_ERR && RETRY_ALLOWED()) { // one retry on disk error if (remount_file_system()) { res = f_stat(name, &info); } } if (res != FR_OK) { errno = fatfs_to_errno((FRESULT)res); return -1; } buf->st_size = info.fsize; epoch = fat_time_to_unix(info.fdate, info.ftime); buf->st_atime = epoch; // Access time buf->st_mtime = epoch; // Modification time buf->st_ctime = epoch; // Creation time // We only handle read only case mode = (FATFS_R | FATFS_X); if ( !(info.fattrib & AM_RDO)) { mode |= (FATFS_W); // enable write if NOT read only } if (info.fattrib & AM_SYS) { buf->st_uid= 0; buf->st_gid= 0; } { buf->st_uid=1000; buf->st_gid=1000; } if (info.fattrib & AM_DIR) { mode |= S_IFDIR; } else { mode |= S_IFREG; } buf->st_mode = mode; return 0; } int AP_Filesystem::unlink(const char *pathname) { WITH_SEMAPHORE(sem); errno = 0; int res = f_unlink(pathname); if (res != FR_OK) { errno = fatfs_to_errno((FRESULT)res); return -1; } return 0; } int AP_Filesystem::mkdir(const char *pathname) { WITH_SEMAPHORE(sem); errno = 0; int res = f_mkdir(pathname); if (res != FR_OK) { errno = fatfs_to_errno((FRESULT)res); return -1; } return 0; } /* wrapper structure to associate a dirent with a DIR */ struct DIR_Wrapper { DIR d; // must be first structure element struct dirent de; }; DIR *AP_Filesystem::opendir(const char *pathdir) { WITH_SEMAPHORE(sem); CHECK_REMOUNT_NULL(); debug("Opendir %s", pathdir); struct DIR_Wrapper *ret = new DIR_Wrapper; if (!ret) { return nullptr; } int res = f_opendir(&ret->d, pathdir); if (res == FR_DISK_ERR && RETRY_ALLOWED()) { // one retry on disk error if (remount_file_system()) { res = f_opendir(&ret->d, pathdir); } } if (res != FR_OK) { errno = fatfs_to_errno((FRESULT)res); delete ret; return nullptr; } debug("Opendir %s -> %p", pathdir, ret); return &ret->d; } struct dirent *AP_Filesystem::readdir(DIR *dirp) { WITH_SEMAPHORE(sem); struct DIR_Wrapper *d = (struct DIR_Wrapper *)dirp; if (!d) { errno = EINVAL; return nullptr; } FILINFO fno; int len; int res; d->de.d_name[0] = 0; res = f_readdir ( dirp, &fno ); if (res != FR_OK || fno.fname[0] == 0) { errno = fatfs_to_errno((FRESULT)res); return nullptr; } len = strlen(fno.fname); strncpy(d->de.d_name,fno.fname,len); d->de.d_name[len] = 0; return &d->de; } int AP_Filesystem::closedir(DIR *dirp) { WITH_SEMAPHORE(sem); struct DIR_Wrapper *d = (struct DIR_Wrapper *)dirp; if (!d) { errno = EINVAL; return -1; } int res = f_closedir (dirp); delete d; if (res != FR_OK) { errno = fatfs_to_errno((FRESULT)res); return -1; } debug("closedir"); return 0; } // return free disk space in bytes int64_t AP_Filesystem::disk_free(const char *path) { WITH_SEMAPHORE(sem); FATFS *fs; DWORD fre_clust, fre_sect; CHECK_REMOUNT(); /* Get volume information and free clusters of drive 1 */ FRESULT res = f_getfree("/", &fre_clust, &fs); if (res) { return res; } /* Get total sectors and free sectors */ fre_sect = fre_clust * fs->csize; return (int64_t)(fre_sect)*512; } // return total disk space in bytes int64_t AP_Filesystem::disk_space(const char *path) { WITH_SEMAPHORE(sem); CHECK_REMOUNT(); FATFS *fs; DWORD fre_clust, tot_sect; /* Get volume information and free clusters of drive 1 */ FRESULT res = f_getfree("/", &fre_clust, &fs); if (res) { return -1; } /* Get total sectors and free sectors */ tot_sect = (fs->n_fatent - 2) * fs->csize; return (int64_t)(tot_sect)*512; } /* convert unix time_t to FATFS timestamp */ static void unix_time_to_fat(time_t epoch, uint16_t &date, uint16_t &time) { struct tm *t = gmtime((time_t *)&epoch); /* Pack date and time into a uint32_t variable */ date = ((uint16_t)(t->tm_year - 80) << 9) | (((uint16_t)t->tm_mon+1) << 5) | (((uint16_t)t->tm_mday)); time = ((uint16_t)t->tm_hour << 11) | ((uint16_t)t->tm_min << 5) | ((uint16_t)t->tm_sec >> 1); } /* set mtime on a file */ bool AP_Filesystem::set_mtime(const char *filename, const time_t mtime_sec) { FILINFO fno; uint16_t fdate, ftime; unix_time_to_fat(mtime_sec, fdate, ftime); fno.fdate = fdate; fno.ftime = ftime; return f_utime(filename, (FILINFO *)&fno) == FR_OK; } /* convert POSIX errno to text with user message. */ char *strerror(int errnum) { #define SWITCH_ERROR(errno) case errno: return const_cast(#errno); break switch (errnum) { SWITCH_ERROR(EPERM); SWITCH_ERROR(ENOENT); SWITCH_ERROR(ESRCH); SWITCH_ERROR(EINTR); SWITCH_ERROR(EIO); SWITCH_ERROR(ENXIO); SWITCH_ERROR(E2BIG); SWITCH_ERROR(ENOEXEC); SWITCH_ERROR(EBADF); SWITCH_ERROR(ECHILD); SWITCH_ERROR(EAGAIN); SWITCH_ERROR(ENOMEM); SWITCH_ERROR(EACCES); SWITCH_ERROR(EFAULT); #ifdef ENOTBLK SWITCH_ERROR(ENOTBLK); #endif // ENOTBLK SWITCH_ERROR(EBUSY); SWITCH_ERROR(EEXIST); SWITCH_ERROR(EXDEV); SWITCH_ERROR(ENODEV); SWITCH_ERROR(ENOTDIR); SWITCH_ERROR(EISDIR); SWITCH_ERROR(EINVAL); SWITCH_ERROR(ENFILE); SWITCH_ERROR(EMFILE); SWITCH_ERROR(ENOTTY); SWITCH_ERROR(ETXTBSY); SWITCH_ERROR(EFBIG); SWITCH_ERROR(ENOSPC); SWITCH_ERROR(ESPIPE); SWITCH_ERROR(EROFS); SWITCH_ERROR(EMLINK); SWITCH_ERROR(EPIPE); SWITCH_ERROR(EDOM); SWITCH_ERROR(ERANGE); SWITCH_ERROR(EBADMSG); } #undef SWITCH_ERROR return NULL; } #endif // CONFIG_HAL_BOARD