/* * Copyright (c) 2013-2019 Huawei Technologies Co., Ltd. All rights reserved. * Copyright (c) 2020-2021 Huawei Device Co., Ltd. All rights reserved. * * Redistribution and use in source and binary forms, with or without modification, * are permitted provided that the following conditions are met: * * 1. Redistributions of source code must retain the above copyright notice, this list of * conditions and the following disclaimer. * * 2. Redistributions in binary form must reproduce the above copyright notice, this list * of conditions and the following disclaimer in the documentation and/or other materials * provided with the distribution. * * 3. Neither the name of the copyright holder nor the names of its contributors may be used * to endorse or promote products derived from this software without specific prior written * permission. * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "fatfs.h" #ifdef LOSCFG_FS_FAT #include "ff.h" #include "disk_pri.h" #include "diskio.h" #include "fs/fs.h" #include "fs/dirent_fs.h" #include "fs/mount.h" #include "vnode.h" #include "path_cache.h" #ifdef LOSCFG_FS_FAT_VIRTUAL_PARTITION #include "virpartff.h" #include "errcode_fat.h" #endif #include "los_tables.h" #include "user_copy.h" #include "los_vm_filemap.h" #include #include #include #include #include #include #include #include "los_hash.h" struct VnodeOps fatfs_vops; /* forward define */ struct file_operations_vfs fatfs_fops; #define BITMASK4 0x0F #define BITMASK5 0x1F #define BITMASK6 0x3F #define BITMASK7 0x7F #define FTIME_MIN_OFFSET 6 /* minute offset in word */ #define FTIME_HR_OFFSET 11 /* hour offset in word */ #define FTIME_MTH_OFFSET 5 /* month offset in word */ #define FTIME_YEAR_OFFSET 9 /* year offset in word */ #define FTIME_DATE_OFFSET 16 /* date offset in dword */ #define SEC_MULTIPLIER 2 #define YEAR_OFFSET 80 /* Year start from 1980 in FATFS, while start from 1900 in struct tm */ int fatfs_2_vfs(int result) { int status = ENOERR; #ifdef LOSCFG_FS_FAT_VIRTUAL_PARTITION if (result < 0 || result >= VIRERR_BASE) { return result; } #else if (result < 0) { return result; } #endif /* FatFs errno to Libc errno */ switch (result) { case FR_OK: break; case FR_NO_FILE: case FR_NO_PATH: status = ENOENT; break; case FR_NO_FILESYSTEM: status = ENOTSUP; break; case FR_INVALID_NAME: status = EINVAL; break; case FR_EXIST: case FR_INVALID_OBJECT: status = EEXIST; break; case FR_DISK_ERR: case FR_NOT_READY: case FR_INT_ERR: status = EIO; break; case FR_WRITE_PROTECTED: status = EROFS; break; case FR_MKFS_ABORTED: case FR_INVALID_PARAMETER: status = EINVAL; break; case FR_NO_SPACE_LEFT: status = ENOSPC; break; case FR_NO_DIRENTRY: status = ENFILE; break; case FR_NO_EMPTY_DIR: status = ENOTEMPTY; break; case FR_IS_DIR: status = EISDIR; break; case FR_NO_DIR: status = ENOTDIR; break; case FR_NO_EPERM: case FR_DENIED: status = EPERM; break; case FR_LOCKED: case FR_TIMEOUT: status = EBUSY; break; #ifdef LOSCFG_FS_FAT_VIRTUAL_PARTITION case FR_MODIFIED: status = -VIRERR_MODIFIED; break; case FR_CHAIN_ERR: status = -VIRERR_CHAIN_ERR; break; case FR_OCCUPIED: status = -VIRERR_OCCUPIED; break; case FR_NOTCLEAR: status = -VIRERR_NOTCLEAR; break; case FR_NOTFIT: status = -VIRERR_NOTFIT; break; case FR_INVAILD_FATFS: status = -VIRERR_INTER_ERR; break; #endif default: status = -FAT_ERROR; break; } return status; } static bool fatfs_is_last_cluster(FATFS *fs, DWORD cclust) { switch (fs->fs_type) { case FS_FAT12: return (cclust == FAT12_END_OF_CLUSTER); case FS_FAT16: return (cclust == FAT16_END_OF_CLUSTER); case FS_FAT32: default: return (cclust == FAT32_END_OF_CLUSTER); } } static int fatfs_sync(unsigned long mountflags, FATFS *fs) { #ifdef LOSCFG_FS_FAT_CACHE los_part *part = NULL; if (mountflags != MS_NOSYNC) { part = get_part((INT)fs->pdrv); if (part == NULL) { return -ENODEV; } (void)OsSdSync(part->disk_id); } #endif return 0; } int fatfs_hash_cmp(struct Vnode *vp, void *arg) { DIR_FILE *dfp_target = (DIR_FILE *)arg; DIR_FILE *dfp = (DIR_FILE *)vp->data; return dfp_target->f_dir.sect != dfp->f_dir.sect || dfp_target->f_dir.dptr != dfp->f_dir.dptr || dfp_target->fno.sclst != dfp->fno.sclst; } static DWORD fatfs_hash(QWORD sect, DWORD dptr, DWORD sclst) { DWORD hash = FNV1_32A_INIT; hash = LOS_HashFNV32aBuf(§, sizeof(QWORD), hash); hash = LOS_HashFNV32aBuf(&dptr, sizeof(DWORD), hash); hash = LOS_HashFNV32aBuf(&sclst, sizeof(DWORD), hash); return hash; } static mode_t fatfs_get_mode(BYTE attribute, mode_t fs_mode) { mode_t mask = 0; if (attribute & AM_RDO) { mask = S_IWUSR | S_IWGRP | S_IWOTH; } fs_mode &= ~mask; if (attribute & AM_DIR) { fs_mode |= S_IFDIR; } else if (attribute & AM_LNK) { fs_mode |= S_IFLNK; } else { fs_mode |= S_IFREG; } return fs_mode; } static enum VnodeType fatfstype_2_vnodetype(BYTE type) { switch (type) { case AM_ARC: return VNODE_TYPE_REG; case AM_DIR: return VNODE_TYPE_DIR; case AM_LNK: return VNODE_TYPE_LNK; default: return VNODE_TYPE_UNKNOWN; } } #define DIR_SIZE 32 static FRESULT init_cluster(DIR_FILE *pdfp, DIR *dp_new, FATFS *fs, int type, const char *target, DWORD *clust) { FRESULT result; BYTE *dir = NULL; QWORD sect; DWORD pclust; UINT n; /* Allocate a new cluster */ *clust = create_chain(&(dp_new->obj), 0); if (*clust == 0) { return FR_NO_SPACE_LEFT; } if (*clust == 1 || *clust == DISK_ERROR) { return FR_DISK_ERR; } result = sync_window(fs); /* Flush FAT */ if (result != FR_OK) { remove_chain(&(dp_new->obj), *clust, 0); return result; } /* Initialize the new cluster */ #ifndef LOSCFG_FS_FAT_VIRTUAL_PARTITION dir = fs->win; #else dir = PARENTFS(fs)->win; #endif sect = clst2sect(fs, *clust); mem_set(dir, 0, SS(fs)); if (type == AM_LNK && target) { /* Write target to symlink */ strcpy_s((char *)dir, SS(fs), target); } else { /* Write the dir cluster */ mem_set(dir, 0, SS(fs)); mem_set(dir + DIR_Name, ' ', 11); /* Create "." entry */ dir[DIR_Name] = '.'; dir[DIR_Attr] = AM_DIR; st_clust(fs, dir, *clust); mem_cpy(dir + DIR_SIZE, dir, DIR_SIZE); /* Create ".." entry */ dir[DIR_SIZE + 1] = '.'; /* Add extra "." */ pclust = pdfp->fno.sclst; if (fs->fs_type == FS_FAT32 && pclust == fs->dirbase) { pclust = 0; } st_clust(fs, dir + DIR_SIZE, pclust); } #ifndef LOSCFG_FS_FAT_VIRTUAL_PARTITION fs->winsect = sect++; fs->wflag = 1; #else PARENTFS(fs)->winsect = sect++; PARENTFS(fs)->wflag = 1; #endif result = sync_window(fs); if (result != FR_OK) { remove_chain(&(dp_new->obj), *clust, 0); return result; } /* Rest of directory cluster should set to be zero */ if (type == AM_DIR) { mem_set(dir, 0, SS(fs)); for (n = fs->csize - 1; n > 0; n--) { #ifndef LOSCFG_FS_FAT_VIRTUAL_PARTITION fs->winsect = sect++; fs->wflag = 1; #else PARENTFS(fs)->winsect = sect++; PARENTFS(fs)->wflag = 1; #endif result = sync_window(fs); if (result != FR_OK) { remove_chain(&(dp_new->obj), *clust, 0); return result; } } } return FR_OK; } static int fatfs_create_obj(struct Vnode *parent, const char *name, int mode, struct Vnode **vpp, BYTE type, const char *target) { struct Vnode *vp = NULL; FATFS *fs = (FATFS *)parent->originMount->data; DIR_FILE *dfp = (DIR_FILE *)parent->data; FILINFO *finfo = &(dfp->fno); DIR_FILE *dfp_new = NULL; FILINFO *finfo_new = NULL; DIR *dp_new = NULL; QWORD time; DWORD hash; DWORD clust = 0; FRESULT result; int ret; if ((type != AM_ARC) && (type != AM_DIR) && (type != AM_LNK)) { result = FR_INVALID_NAME; goto ERROR_EXIT; } dfp_new = (DIR_FILE *)zalloc(sizeof(DIR_FILE)); if (dfp_new == NULL) { result = FR_NOT_ENOUGH_CORE; goto ERROR_EXIT; } ret = lock_fs(fs); if (ret == FALSE) { /* lock failed */ result = FR_TIMEOUT; goto ERROR_FREE; } if (finfo->fattrib & AM_ARC || finfo->fattrib & AM_LNK) { result = FR_NO_DIR; goto ERROR_UNLOCK; } finfo_new = &(dfp_new->fno); LOS_ListInit(&finfo_new->fp_list); dp_new = &(dfp_new->f_dir); dp_new->obj.fs = fs; dp_new->obj.sclust = finfo->sclst; DEF_NAMBUF; INIT_NAMBUF(fs); result = create_name(dp_new, &name); if (result != FR_OK) { goto ERROR_UNLOCK; } result = dir_find(dp_new); if (result == FR_OK) { result = FR_EXIST; goto ERROR_UNLOCK; } if (type == AM_DIR || type == AM_LNK) { result = init_cluster(dfp, dp_new, fs, type, target, &clust); if (result != FR_OK) { goto ERROR_UNLOCK; } } result = dir_register(dp_new); if (result != FR_OK) { goto ERROR_REMOVE_CHAIN; } /* Set the directory entry attribute */ if (time_status == SYSTEM_TIME_ENABLE) { time = GET_FATTIME(); } else { time = 0; } st_dword(dp_new->dir + DIR_CrtTime, time); st_dword(dp_new->dir + DIR_ModTime, time); st_word(dp_new->dir + DIR_LstAccDate, time >> FTIME_DATE_OFFSET); dp_new->dir[DIR_Attr] = type; if (((DWORD)mode & S_IWUSR) == 0) { dp_new->dir[DIR_Attr] |= AM_RDO; } st_clust(fs, dp_new->dir, clust); if (type == AM_ARC) { st_dword(dp_new->dir + DIR_FileSize, 0); } else if (type == AM_LNK) { st_dword(dp_new->dir + DIR_FileSize, strlen(target)); } #ifdef LOSCFG_FS_FAT_VIRTUAL_PARTITION PARENTFS(fs)->wflag = 1; #else fs->wflag = 1; #endif result = sync_fs(fs); if (result != FR_OK) { goto ERROR_REMOVE_CHAIN; } result = dir_read(dp_new, 0); if (result != FR_OK) { goto ERROR_REMOVE_CHAIN; } dp_new->blk_ofs = dir_ofs(dp_new); get_fileinfo(dp_new, finfo_new); if (type == AM_ARC) { dp_new->obj.objsize = 0; } else if (type == AM_LNK) { dp_new->obj.objsize = strlen(target); } ret = VnodeAlloc(&fatfs_vops, &vp); if (ret != 0) { result = FR_NOT_ENOUGH_CORE; goto ERROR_REMOVE_CHAIN; } vp->parent = parent; vp->fop = &fatfs_fops; vp->data = dfp_new; vp->originMount = parent->originMount; vp->uid = fs->fs_uid; vp->gid = fs->fs_gid; vp->mode = fatfs_get_mode(finfo_new->fattrib, fs->fs_mode); vp->type = fatfstype_2_vnodetype(type); hash = fatfs_hash(dp_new->sect, dp_new->dptr, finfo_new->sclst); ret = VfsHashInsert(vp, hash); if (ret != 0) { result = FR_NOT_ENOUGH_CORE; goto ERROR_REMOVE_CHAIN; } *vpp = vp; unlock_fs(fs, FR_OK); FREE_NAMBUF(); return fatfs_sync(parent->originMount->mountFlags, fs); ERROR_REMOVE_CHAIN: remove_chain(&(dp_new->obj), clust, 0); ERROR_UNLOCK: unlock_fs(fs, result); FREE_NAMBUF(); ERROR_FREE: free(dfp_new); ERROR_EXIT: return -fatfs_2_vfs(result); } int fatfs_lookup(struct Vnode *parent, const char *path, int len, struct Vnode **vpp) { struct Vnode *vp = NULL; FATFS *fs = (FATFS *)(parent->originMount->data); DIR_FILE *dfp; DIR *dp = NULL; FILINFO *finfo = NULL; DWORD hash; FRESULT result; int ret; dfp = (DIR_FILE *)zalloc(sizeof(DIR_FILE)); if (dfp == NULL) { ret = ENOMEM; goto ERROR_EXIT; } ret = lock_fs(fs); if (ret == FALSE) { ret = EBUSY; goto ERROR_FREE; } finfo = &(dfp->fno); LOS_ListInit(&finfo->fp_list); dp = &(dfp->f_dir); dp->obj.fs = fs; dp->obj.sclust = ((DIR_FILE *)(parent->data))->fno.sclst; DEF_NAMBUF; INIT_NAMBUF(fs); result = create_name(dp, &path); if (result != FR_OK) { ret = fatfs_2_vfs(result); goto ERROR_UNLOCK; } result = dir_find(dp); if (result != FR_OK) { ret = fatfs_2_vfs(result); goto ERROR_UNLOCK; } if (dp->fn[NSFLAG] & NS_NONAME) { result = FR_INVALID_NAME; ret = fatfs_2_vfs(result); goto ERROR_UNLOCK; } get_fileinfo(dp, finfo); dp->obj.objsize = 0; hash = fatfs_hash(dp->sect, dp->dptr, finfo->sclst); ret = VfsHashGet(parent->originMount, hash, &vp, fatfs_hash_cmp, dfp); if (ret != 0) { ret = VnodeAlloc(&fatfs_vops, &vp); if (ret != 0) { ret = ENOMEM; result = FR_NOT_ENOUGH_CORE; goto ERROR_UNLOCK; } vp->parent = parent; vp->fop = &fatfs_fops; vp->data = dfp; vp->originMount = parent->originMount; vp->uid = fs->fs_uid; vp->gid = fs->fs_gid; vp->mode = fatfs_get_mode(finfo->fattrib, fs->fs_mode); if (finfo->fattrib & AM_DIR) { vp->type = VNODE_TYPE_DIR; } else { vp->type = VNODE_TYPE_REG; } ret = VfsHashInsert(vp, hash); if (ret != 0) { result = FR_INVALID_PARAMETER; goto ERROR_UNLOCK; } } else { vp->parent = parent; free(dfp); /* hash hit dfp is no needed */ } unlock_fs(fs, FR_OK); FREE_NAMBUF(); *vpp = vp; return 0; ERROR_UNLOCK: unlock_fs(fs, result); FREE_NAMBUF(); ERROR_FREE: free(dfp); ERROR_EXIT: return -ret; } int fatfs_create(struct Vnode *parent, const char *name, int mode, struct Vnode **vpp) { return fatfs_create_obj(parent, name, mode, vpp, AM_ARC, NULL); } int fatfs_open(struct file *filep) { struct Vnode *vp = filep->f_vnode; FATFS *fs = (FATFS *)vp->originMount->data; DIR_FILE *dfp = (DIR_FILE *)vp->data; DIR *dp = &(dfp->f_dir); FILINFO *finfo = &(dfp->fno); FIL *fp; int ret; fp = (FIL *)zalloc(sizeof(FIL)); if (fp == NULL) { ret = ENOMEM; goto ERROR_EXIT; } ret = lock_fs(fs); if (ret == FALSE) { ret = EBUSY; goto ERROR_EXIT; } fp->dir_sect = dp->sect; fp->dir_ptr = dp->dir; fp->obj.sclust = finfo->sclst; fp->obj.objsize = finfo->fsize; #if FF_USE_FASTSEEK fp->cltbl = 0; /* Disable fast seek mode */ #endif fp->obj.fs = fs; fp->obj.id = fs->id; fp->flag = FA_READ | FA_WRITE; fp->err = 0; fp->sect = 0; fp->fptr = 0; fp->buf = (BYTE*) ff_memalloc(SS(fs)); if (fp->buf == NULL) { ret = ENOMEM; goto ERROR_FREE; } LOS_ListAdd(&finfo->fp_list, &fp->fp_entry); unlock_fs(fs, FR_OK); filep->f_priv = fp; return fatfs_sync(vp->originMount->mountFlags, fs); ERROR_FREE: unlock_fs(fs, FR_OK); free(fp); ERROR_EXIT: return -ret; } int fatfs_close(struct file *filep) { FIL *fp = (FIL *)filep->f_priv; FATFS *fs = fp->obj.fs; FRESULT result; int ret; ret = lock_fs(fs); if (ret == FALSE) { return -EBUSY; } #if !FF_FS_READONLY result = f_sync(fp); /* Flush cached data */ if (result != FR_OK) { goto EXIT; } ret = fatfs_sync(filep->f_vnode->originMount->mountFlags, fs); if (ret != 0) { unlock_fs(fs, FR_OK); return ret; } #endif LOS_ListDelete(&fp->fp_entry); ff_memfree(fp->buf); free(fp); filep->f_priv = NULL; EXIT: unlock_fs(fs, result); return -fatfs_2_vfs(result); } int fatfs_read(struct file *filep, char *buff, size_t count) { FIL *fp = (FIL *)filep->f_priv; FATFS *fs = fp->obj.fs; struct Vnode *vp = filep->f_vnode; FILINFO *finfo = &((DIR_FILE *)(vp->data))->fno; size_t rcount; FRESULT result; int ret; ret = lock_fs(fs); if (ret == FALSE) { return -EBUSY; } fp->obj.objsize = finfo->fsize; fp->obj.sclust = finfo->sclst; result = f_read(fp, buff, count, &rcount); if (result != FR_OK) { goto EXIT; } filep->f_pos = fp->fptr; EXIT: unlock_fs(fs, result); return rcount; } static FRESULT update_dir(DIR *dp, FILINFO *finfo) { FATFS *fs = dp->obj.fs; DWORD tm; BYTE *dbuff = NULL; FRESULT result; result = move_window(fs, dp->sect); if (result != FR_OK) { return result; } dbuff = fs->win + dp->dptr % SS(fs); dbuff[DIR_Attr] = finfo->fattrib; st_clust(fs, dbuff, finfo->sclst); /* Update start cluster */ st_dword(dbuff + DIR_FileSize, (DWORD)finfo->fsize); /* Update file size */ if (time_status == SYSTEM_TIME_ENABLE) { tm = GET_FATTIME(); } else { tm = 0; } st_dword(dbuff + DIR_ModTime, tm); /* Update mtime */ st_word(dbuff + DIR_LstAccDate, tm >> FTIME_DATE_OFFSET); /* Update access date */ #ifndef LOSCFG_FS_FAT_VIRTUAL_PARTITION fs->wflag = 1; #else PARENTFS(fs)->wflag = 1; #endif return sync_fs(fs); } off64_t fatfs_lseek64(struct file *filep, off64_t offset, int whence) { FIL *fp = (FIL *)filep->f_priv; FATFS *fs = fp->obj.fs; struct Vnode *vp = filep->f_vnode; DIR_FILE *dfp = (DIR_FILE *)vp->data; FILINFO *finfo = &(dfp->fno); FSIZE_t fpos; FRESULT result; int ret; switch (whence) { case SEEK_CUR: offset = filep->f_pos + offset; if (offset < 0) { return -EINVAL; } fpos = offset; break; case SEEK_SET: if (offset < 0) { return -EINVAL; } fpos = offset; break; case SEEK_END: offset = (off_t)((long long)finfo->fsize + offset); if (offset < 0) { return -EINVAL; } fpos = offset; break; default: return -EINVAL; } if (offset >= FAT32_MAXSIZE) { return -EINVAL; } ret = lock_fs(fs); if (ret == FALSE) { return -EBUSY; } fp->obj.sclust = finfo->sclst; fp->obj.objsize = finfo->fsize; result = f_lseek(fp, fpos); finfo->fsize = fp->obj.objsize; finfo->sclst = fp->obj.sclust; if (result != FR_OK) { goto ERROR_EXIT; } result = f_sync(fp); if (result != FR_OK) { goto ERROR_EXIT; } filep->f_pos = fpos; unlock_fs(fs, FR_OK); return fpos; ERROR_EXIT: unlock_fs(fs, result); return -fatfs_2_vfs(result); } off_t fatfs_lseek(struct file *filep, off_t offset, int whence) { return (off_t)fatfs_lseek64(filep, offset, whence); } static int update_filbuff(FILINFO *finfo, FIL *wfp, const char *data) { LOS_DL_LIST *list = &finfo->fp_list; FATFS *fs = wfp->obj.fs; FIL *entry = NULL; int ret = 0; LOS_DL_LIST_FOR_EACH_ENTRY(entry, list, FIL, fp_entry) { if (entry == wfp) { continue; } if (entry->sect != 0) { if (disk_read(fs->pdrv, entry->buf, entry->sect, 1) != RES_OK) { ret = -1; } } } return ret; } int fatfs_write(struct file *filep, const char *buff, size_t count) { FIL *fp = (FIL *)filep->f_priv; FATFS *fs = fp->obj.fs; struct Vnode *vp = filep->f_vnode; FILINFO *finfo = &(((DIR_FILE *)vp->data)->fno); size_t wcount; FRESULT result; int ret; ret = lock_fs(fs); if (ret == FALSE) { return -EBUSY; } fp->obj.objsize = finfo->fsize; fp->obj.sclust = finfo->sclst; result = f_write(fp, buff, count, &wcount); if (result != FR_OK) { goto ERROR_EXIT; } finfo->fsize = fp->obj.objsize; finfo->sclst = fp->obj.sclust; result = f_sync(fp); if (result != FR_OK) { goto ERROR_EXIT; } update_filbuff(finfo, fp, buff); filep->f_pos = fp->fptr; unlock_fs(fs, FR_OK); return wcount; ERROR_EXIT: unlock_fs(fs, result); return -fatfs_2_vfs(result); } int fatfs_fsync(struct file *filep) { FIL *fp = filep->f_priv; FATFS *fs = fp->obj.fs; FRESULT result; int ret; ret = lock_fs(fs); if (ret == FALSE) { return -EBUSY; } result = f_sync(fp); unlock_fs(fs, result); return -fatfs_2_vfs(result); } int fatfs_fallocate64(struct file *filep, int mode, off64_t offset, off64_t len) { FIL *fp = (FIL *)filep->f_priv; FATFS *fs = fp->obj.fs; struct Vnode *vp = filep->f_vnode; FILINFO *finfo = &((DIR_FILE *)(vp->data))->fno; FRESULT result; int ret; if (offset < 0 || len <= 0) { return -EINVAL; } if (len >= FAT32_MAXSIZE || offset >= FAT32_MAXSIZE || len + offset >= FAT32_MAXSIZE) { return -EINVAL; } if (mode != FALLOC_FL_KEEP_SIZE) { return -EINVAL; } ret = lock_fs(fs); if (ret == FALSE) { return -EBUSY; } result = f_expand(fp, (FSIZE_t)offset, (FSIZE_t)len, 1); if (result == FR_OK && finfo->sclst == 0) { finfo->sclst = fp->obj.sclust; } result = f_sync(fp); unlock_fs(fs, FR_OK); return -fatfs_2_vfs(result); } static FRESULT realloc_cluster(FILINFO *finfo, FFOBJID *obj, FSIZE_t size) { FATFS *fs = obj->fs; off64_t remain; DWORD cclust; DWORD pclust; QWORD csize; FRESULT result; if (size == 0) { /* Remove cluster chain */ if (finfo->sclst != 0) { result = remove_chain(obj, finfo->sclst, 0); if (result != FR_OK) { return result; } finfo->sclst = 0; } return FR_OK; } remain = size; csize = SS(fs) * fs->csize; if (finfo->sclst == 0) { /* Allocate one cluster if file doesn't have any cluster */ cclust = create_chain(obj, 0); if (cclust == 0) { return FR_NO_SPACE_LEFT; } if (cclust == 1 || cclust == DISK_ERROR) { return FR_DISK_ERR; } finfo->sclst = cclust; } cclust = finfo->sclst; while (remain > csize) { /* Follow or strentch the cluster chain */ pclust = cclust; cclust = create_chain(obj, pclust); if (cclust == 0) { return FR_NO_SPACE_LEFT; } if (cclust == 1 || cclust == DISK_ERROR) { return FR_DISK_ERR; } remain -= csize; } pclust = cclust; cclust = get_fat(obj, pclust); if ((cclust == BAD_CLUSTER) || (cclust == DISK_ERROR)) { return FR_DISK_ERR; } if (!fatfs_is_last_cluster(obj->fs, cclust)) { /* Remove extra cluster if existing */ result = remove_chain(obj, cclust, pclust); if (result != FR_OK) { return result; } } return FR_OK; } int fatfs_fallocate(struct file *filep, int mode, off_t offset, off_t len) { return fatfs_fallocate64(filep, mode, offset, len); } int fatfs_truncate64(struct Vnode *vp, off64_t len) { FATFS *fs = (FATFS *)vp->originMount->data; DIR_FILE *dfp = (DIR_FILE *)vp->data; DIR *dp = &(dfp->f_dir); FILINFO *finfo = &(dfp->fno); FFOBJID object; FRESULT result = FR_OK; int ret; if (len < 0 || len >= FAT32_MAXSIZE) { return -EINVAL; } ret = lock_fs(fs); if (ret == FALSE) { result = FR_TIMEOUT; goto ERROR_OUT; } if (len == finfo->fsize) { unlock_fs(fs, FR_OK); return 0; } object.fs = fs; result = realloc_cluster(finfo, &object, (FSIZE_t)len); if (result != FR_OK) { goto ERROR_UNLOCK; } finfo->fsize = (FSIZE_t)len; result = update_dir(dp, finfo); if (result != FR_OK) { goto ERROR_UNLOCK; } unlock_fs(fs, FR_OK); return fatfs_sync(vp->originMount->mountFlags, fs); ERROR_UNLOCK: unlock_fs(fs, result); ERROR_OUT: return -fatfs_2_vfs(result); } int fatfs_truncate(struct Vnode *vp, off_t len) { return fatfs_truncate64(vp, len); } static int fat_bind_check(struct Vnode *blk_driver, los_part **partition) { los_part *part = NULL; if (blk_driver == NULL || blk_driver->data == NULL) { return ENODEV; } struct drv_data *dd = blk_driver->data; if (dd->ops == NULL) { return ENODEV; } const struct block_operations *bops = dd->ops; if (bops->open == NULL) { return EINVAL; } if (bops->open(blk_driver) < 0) { return EBUSY; } part = los_part_find(blk_driver); if (part == NULL) { return ENODEV; } if (part->part_name != NULL) { bops->close(blk_driver); return EBUSY; } #ifndef FF_MULTI_PARTITION if (part->part_no_mbr > 1) { bops->close(blk_driver); return EPERM; } #endif *partition = part; return 0; } int fatfs_mount(struct Mount *mnt, struct Vnode *blk_device, const void *data) { struct Vnode *vp = NULL; FATFS *fs = NULL; DIR_FILE *dfp = NULL; los_part *part = NULL; QWORD start_sector; BYTE fmt; DWORD hash; FRESULT result; int ret; ret = fat_bind_check(blk_device, &part); if (ret != 0) { goto ERROR_EXIT; } ret = SetDiskPartName(part, "vfat"); if (ret != 0) { ret = EIO; goto ERROR_EXIT; } fs = (FATFS *)zalloc(sizeof(FATFS)); if (fs == NULL) { ret = ENOMEM; goto ERROR_PARTNAME; } #ifdef LOSCFG_FS_FAT_VIRTUAL_PARTITION fs->vir_flag = FS_PARENT; fs->parent_fs = fs; fs->vir_amount = DISK_ERROR; fs->vir_avail = FS_VIRDISABLE; #endif ret = ff_cre_syncobj(0, &fs->sobj); if (ret == 0) { /* create sync object failed */ ret = EINVAL; goto ERROR_WITH_FS; } ret = lock_fs(fs); if (ret == FALSE) { ret = EBUSY; goto ERROR_WITH_MUX; } fs->fs_type = 0; fs->pdrv = part->part_id; #if FF_MAX_SS != FF_MIN_SS /* Get sector size (multiple sector size cfg only) */ if (disk_ioctl(fs->pdrv, GET_SECTOR_SIZE, &(fs->ssize)) != RES_OK) { ret = EIO; goto ERROR_WITH_LOCK; } if (fs->ssize > FF_MAX_SS || fs->ssize < FF_MIN_SS || (fs->ssize & (fs->ssize - 1))) { ret = EIO; goto ERROR_WITH_LOCK; } #endif fs->win = (BYTE *)ff_memalloc(SS(fs)); if (fs->win == NULL) { ret = ENOMEM; goto ERROR_WITH_LOCK; } result = find_fat_partition(fs, part, &fmt, &start_sector); if (result != FR_OK) { ret = fatfs_2_vfs(result); goto ERROR_WITH_FSWIN; } result = init_fatobj(fs, fmt, start_sector); if (result != FR_OK) { ret = fatfs_2_vfs(result); goto ERROR_WITH_FSWIN; } fs->fs_uid = mnt->vnodeBeCovered->uid; fs->fs_gid = mnt->vnodeBeCovered->gid; fs->fs_dmask = GetUmask(); fs->fs_fmask = GetUmask(); fs->fs_mode = mnt->vnodeBeCovered->mode & (S_IRWXU | S_IRWXG | S_IRWXO); dfp = (DIR_FILE *)zalloc(sizeof(DIR_FILE)); if (dfp == NULL) { ret = ENOMEM; goto ERROR_WITH_FSWIN; } dfp->f_dir.obj.fs = fs; dfp->f_dir.obj.sclust = 0; /* set start clust 0, root */ dfp->f_dir.obj.attr = AM_DIR; dfp->f_dir.obj.objsize = 0; /* dir size is 0 */ dfp->fno.fsize = 0; dfp->fno.fdate = 0; dfp->fno.ftime = 0; dfp->fno.fattrib = AM_DIR; dfp->fno.sclst = 0; dfp->fno.fname[0] = '/'; /* Mark as root dir */ dfp->fno.fname[1] = '\0'; LOS_ListInit(&(dfp->fno.fp_list)); ret = VnodeAlloc(&fatfs_vops, &vp); if (ret != 0) { ret = ENOMEM; goto ERROR_WITH_FSWIN; } mnt->data = fs; mnt->vnodeCovered = vp; vp->parent = mnt->vnodeBeCovered; vp->fop = &fatfs_fops; vp->data = dfp; vp->originMount = mnt; vp->uid = fs->fs_uid; vp->gid = fs->fs_gid; vp->mode = mnt->vnodeBeCovered->mode; vp->type = VNODE_TYPE_DIR; hash = fatfs_hash(0, 0, 0); ret = VfsHashInsert(vp, hash); if (ret != 0) { ret = -ret; goto ERROR_WITH_LOCK; } unlock_fs(fs, FR_OK); return 0; ERROR_WITH_FSWIN: ff_memfree(fs->win); ERROR_WITH_LOCK: unlock_fs(fs, FR_OK); ERROR_WITH_MUX: ff_del_syncobj(&fs->sobj); ERROR_WITH_FS: free(fs); ERROR_PARTNAME: if (part->part_name) { free(part->part_name); part->part_name = NULL; } ERROR_EXIT: return -ret; } int fatfs_umount(struct Mount *mnt, struct Vnode **blkdriver) { struct Vnode *device; FATFS *fs = (FATFS *)mnt->data; los_part *part; int ret; ret = lock_fs(fs); if (ret == FALSE) { return -EBUSY; } part = get_part(fs->pdrv); if (part == NULL) { unlock_fs(fs, FR_OK); return -ENODEV; } device = part->dev; if (device == NULL) { unlock_fs(fs, FR_OK); return -ENODEV; } #ifdef LOSCFG_FS_FAT_CACHE ret = OsSdSync(part->disk_id); if (ret != 0) { unlock_fs(fs, FR_DISK_ERR); return -EIO; } #endif if (part->part_name != NULL) { free(part->part_name); part->part_name = NULL; } struct drv_data *dd = device->data; if (dd->ops == NULL) { unlock_fs(fs, FR_OK); return ENODEV; } const struct block_operations *bops = dd->ops; if (bops != NULL && bops->close != NULL) { bops->close(*blkdriver); } if (fs->win != NULL) { ff_memfree(fs->win); } unlock_fs(fs, FR_OK); ret = ff_del_syncobj(&fs->sobj); if (ret == FALSE) { return -EINVAL; } free(fs); *blkdriver = device; return 0; } int fatfs_statfs(struct Mount *mnt, struct statfs *info) { FATFS *fs = (FATFS *)mnt->data; DWORD nclst = 0; FRESULT result = FR_OK; int ret; info->f_type = MSDOS_SUPER_MAGIC; #if FF_MAX_SS != FF_MIN_SS info->f_bsize = fs->ssize * fs->csize; #else info->f_bsize = FF_MIN_SS * fs->csize; #endif info->f_blocks = fs->n_fatent; ret = lock_fs(fs); if (ret == FALSE) { return -EBUSY; } /* free cluster is unavailable, update it */ if (fs->free_clst == DISK_ERROR) { result = fat_count_free_entries(&nclst, fs); } info->f_bfree = fs->free_clst; info->f_bavail = fs->free_clst; unlock_fs(fs, result); #if FF_USE_LFN /* Maximum length of filenames */ info->f_namelen = FF_MAX_LFN; #else /* Maximum length of filenames: 8 is the basename length, 1 is the dot, 3 is the extension length */ info->f_namelen = (8 + 1 + 3); #endif info->f_fsid.__val[0] = MSDOS_SUPER_MAGIC; info->f_fsid.__val[1] = 1; info->f_frsize = SS(fs) * fs->csize; info->f_files = 0; info->f_ffree = 0; info->f_flags = mnt->mountFlags; return -fatfs_2_vfs(result); } static inline int GET_SECONDS(WORD ftime) { return (ftime & BITMASK5) * SEC_MULTIPLIER; } static inline int GET_MINUTES(WORD ftime) { return (ftime >> FTIME_MIN_OFFSET) & BITMASK6; } static inline int GET_HOURS(WORD ftime) { return (ftime >> FTIME_HR_OFFSET) & BITMASK5; } static inline int GET_DAY(WORD fdate) { return fdate & BITMASK5; } static inline int GET_MONTH(WORD fdate) { return (fdate >> FTIME_MTH_OFFSET) & BITMASK4; } static inline int GET_YEAR(WORD fdate) { return (fdate >> FTIME_YEAR_OFFSET) & BITMASK7; } static time_t fattime_transfer(WORD fdate, WORD ftime) { struct tm time = { 0 }; time.tm_sec = GET_SECONDS(ftime); time.tm_min = GET_MINUTES(ftime); time.tm_hour = GET_HOURS(ftime); time.tm_mday = GET_DAY(fdate); time.tm_mon = GET_MONTH(fdate); time.tm_year = GET_YEAR(fdate) + YEAR_OFFSET; /* Year start from 1980 in FATFS */ time_t ret = mktime(&time); return ret; } DWORD fattime_format(time_t time) { struct tm st; DWORD ftime; localtime_r(&time, &st); ftime = (DWORD)st.tm_mday; ftime |= ((DWORD)st.tm_mon) << FTIME_MTH_OFFSET; ftime |= ((DWORD)((st.tm_year > YEAR_OFFSET) ? (st.tm_year - YEAR_OFFSET) : 0)) << FTIME_YEAR_OFFSET; ftime <<= FTIME_DATE_OFFSET; ftime = (DWORD)st.tm_sec / SEC_MULTIPLIER; ftime |= ((DWORD)st.tm_min) << FTIME_MIN_OFFSET; ftime |= ((DWORD)st.tm_hour) << FTIME_HR_OFFSET; return ftime; } int fatfs_stat(struct Vnode *vp, struct stat* sp) { FATFS *fs = (FATFS *)vp->originMount->data; DIR_FILE *dfp = (DIR_FILE *)vp->data; FILINFO *finfo = &(dfp->fno); time_t time; int ret; ret = lock_fs(fs); if (ret == FALSE) { return EBUSY; } sp->st_dev = fs->pdrv; sp->st_mode = vp->mode; sp->st_nlink = 1; sp->st_uid = fs->fs_uid; sp->st_gid = fs->fs_gid; sp->st_size = finfo->fsize; sp->st_blksize = fs->csize * SS(fs); sp->st_blocks = finfo->fsize ? ((finfo->fsize - 1) / SS(fs) / fs->csize + 1) : 0; time = fattime_transfer(finfo->fdate, finfo->ftime); sp->st_mtime = time; unlock_fs(fs, FR_OK); return 0; } void fatfs_chtime(DIR *dp, struct IATTR *attr) { BYTE *dir = dp->dir; DWORD ftime; if (attr->attr_chg_valid & CHG_ATIME) { ftime = fattime_format(attr->attr_chg_atime); st_word(dir + DIR_LstAccDate, (ftime >> FTIME_DATE_OFFSET)); } if (attr->attr_chg_valid & CHG_CTIME) { ftime = fattime_format(attr->attr_chg_ctime); st_dword(dir + DIR_CrtTime, ftime); } if (attr->attr_chg_valid & CHG_MTIME) { ftime = fattime_format(attr->attr_chg_mtime); st_dword(dir + DIR_ModTime, ftime); } } int fatfs_chattr(struct Vnode *vp, struct IATTR *attr) { FATFS *fs = (FATFS *)vp->originMount->data; DIR_FILE *dfp = (DIR_FILE *)vp->data; DIR *dp = &(dfp->f_dir); FILINFO *finfo = &(dfp->fno); BYTE *dir = dp->dir; DWORD ftime; FRESULT result; int ret; if (finfo->fname[0] == '/') { /* Is root dir of fatfs ? */ return 0; } ret = lock_fs(fs); if (ret == FALSE) { result = FR_TIMEOUT; goto ERROR_OUT; } result = move_window(fs, dp->sect); if (result != FR_OK) { goto ERROR_UNLOCK; } if (attr->attr_chg_valid & CHG_MODE) { /* FAT only support readonly flag */ if ((attr->attr_chg_mode & S_IWUSR) == 0 && (finfo->fattrib & AM_RDO) == 0) { dir[DIR_Attr] |= AM_RDO; finfo->fattrib |= AM_RDO; fs->wflag = 1; } else if ((attr->attr_chg_mode & S_IWUSR) != 0 && (finfo->fattrib & AM_RDO) != 0) { dir[DIR_Attr] &= ~AM_RDO; finfo->fattrib &= ~AM_RDO; fs->wflag = 1; } vp->mode = fatfs_get_mode(finfo->fattrib, fs->fs_mode); } if (attr->attr_chg_valid & (CHG_ATIME | CHG_CTIME | CHG_MTIME)) { fatfs_chtime(dp, attr); ftime = ld_dword(dp->dir + DIR_ModTime); finfo->fdate = (WORD)(ftime >> FTIME_DATE_OFFSET); finfo->ftime = (WORD)ftime; } result = sync_window(fs); if (result != FR_OK) { goto ERROR_UNLOCK; } unlock_fs(fs, FR_OK); return fatfs_sync(vp->originMount->mountFlags, fs); ERROR_UNLOCK: unlock_fs(fs, result); ERROR_OUT: return -fatfs_2_vfs(result); } int fatfs_opendir(struct Vnode *vp, struct fs_dirent_s *idir) { FATFS *fs = vp->originMount->data; DIR_FILE *dfp = (DIR_FILE *)vp->data; FILINFO *finfo = &(dfp->fno); DIR *dp; DWORD clst; FRESULT result; int ret; dp = (DIR*)zalloc(sizeof(DIR)); if (dp == NULL) { return -ENOMEM; } ret = lock_fs(fs); if (ret == FALSE) { return -EBUSY; } clst = finfo->sclst; dp->obj.fs = fs; dp->obj.sclust = clst; result = dir_sdi(dp, 0); if (result != FR_OK) { free(dp); unlock_fs(fs, result); return -fatfs_2_vfs(result); } unlock_fs(fs, result); idir->u.fs_dir = dp; return 0; } int fatfs_readdir(struct Vnode *vp, struct fs_dirent_s *idir) { FATFS *fs = vp->originMount->data; FILINFO fno; DIR* dp = (DIR*)idir->u.fs_dir; struct dirent *dirp = NULL; FRESULT result; int ret, i; ret = lock_fs(fs); if (ret == FALSE) { /* Lock fs failed */ return -EBUSY; } DEF_NAMBUF; INIT_NAMBUF(fs); for (i = 0; i < idir->read_cnt; i++) { /* using dir_read_massive to promote performance */ result = dir_read_massive(dp, 0); if (result == FR_NO_FILE) { break; } else if (result != FR_OK) { goto ERROR_UNLOCK; } get_fileinfo(dp, &fno); /* 0x00 for end of directory and 0xFF for directory is empty */ if (fno.fname[0] == 0x00 || fno.fname[0] == (TCHAR)0xFF) { break; } dirp = &(idir->fd_dir[i]); if (fno.fattrib & AM_DIR) { /* is dir */ dirp->d_type = DT_DIR; } else { dirp->d_type = DT_REG; } if (strncpy_s(dirp->d_name, sizeof(dirp->d_name), fno.fname, strlen(fno.fname)) != EOK) { result = FR_NOT_ENOUGH_CORE; goto ERROR_UNLOCK; } result = dir_next(dp, 0); if (result != FR_OK && result != FR_NO_FILE) { goto ERROR_UNLOCK; } idir->fd_position++; idir->fd_dir[i].d_off = idir->fd_position; idir->fd_dir[i].d_reclen = (uint16_t)sizeof(struct dirent); } unlock_fs(fs, FR_OK); FREE_NAMBUF(); return i; ERROR_UNLOCK: unlock_fs(fs, result); FREE_NAMBUF(); return -fatfs_2_vfs(result); } int fatfs_rewinddir(struct Vnode *vp, struct fs_dirent_s *dir) { DIR *dp = (DIR *)dir->u.fs_dir; FATFS *fs = dp->obj.fs; FRESULT result; int ret; ret = lock_fs(fs); if (ret == FALSE) { return -EBUSY; } result = dir_sdi(dp, 0); unlock_fs(fs, result); return -fatfs_2_vfs(result); } int fatfs_closedir(struct Vnode *vp, struct fs_dirent_s *dir) { DIR *dp = (DIR *)dir->u.fs_dir; free(dp); dir->u.fs_dir = NULL; return 0; } static FRESULT rename_check(DIR *dp_new, FILINFO *finfo_new, DIR *dp_old, FILINFO *finfo_old) { DIR dir_sub; FRESULT result; if (finfo_new->fattrib & AM_ARC) { /* new path is file */ if (finfo_old->fattrib & AM_DIR) { /* but old path is dir */ return FR_NO_DIR; } } else if (finfo_new->fattrib & AM_DIR) { /* new path is dir */ if (finfo_old->fattrib & AM_ARC) { /* old path is file */ return FR_IS_DIR; } dir_sub.obj.fs = dp_old->obj.fs; dir_sub.obj.sclust = finfo_new->sclst; result = dir_sdi(&dir_sub, 0); if (result != FR_OK) { return result; } result = dir_read(&dir_sub, 0); if (result == FR_OK) { /* new path isn't empty file */ return FR_NO_EMPTY_DIR; } } else { /* System file or volume label */ return FR_DENIED; } return FR_OK; } int fatfs_rename(struct Vnode *old_vnode, struct Vnode *new_parent, const char *oldname, const char *newname) { FATFS *fs = (FATFS *)(old_vnode->originMount->data); DIR_FILE *dfp_old = (DIR_FILE *)old_vnode->data; DIR *dp_old = &(dfp_old->f_dir); FILINFO *finfo_old = &(dfp_old->fno); DIR_FILE *dfp_new = NULL; DIR* dp_new = NULL; FILINFO* finfo_new = NULL; DWORD clust; FRESULT result; int ret; ret = lock_fs(fs); if (ret == FALSE) { /* Lock fs failed */ return -EBUSY; } dfp_new = (DIR_FILE *)zalloc(sizeof(DIR_FILE)); if (dfp_new == NULL) { result = FR_NOT_ENOUGH_CORE; goto ERROR_UNLOCK; } dp_new = &(dfp_new->f_dir); finfo_new = &(dfp_new->fno); dp_new->obj.sclust = ((DIR_FILE *)(new_parent->data))->fno.sclst; dp_new->obj.fs = fs; /* Find new path */ DEF_NAMBUF; INIT_NAMBUF(fs); result = create_name(dp_new, &newname); if (result != FR_OK) { goto ERROR_FREE; } result = dir_find(dp_new); if (result == FR_OK) { /* new path name exist */ get_fileinfo(dp_new, finfo_new); result = rename_check(dp_new, finfo_new, dp_old, finfo_old); if (result != FR_OK) { goto ERROR_FREE; } result = dir_remove(dp_old); if (result != FR_OK) { goto ERROR_FREE; } clust = finfo_new->sclst; if (clust != 0) { /* remove the new path cluster chain if exists */ result = remove_chain(&(dp_new->obj), clust, 0); if (result != FR_OK) { goto ERROR_FREE; } } } else { /* new path name not exist */ result = dir_remove(dp_old); if (result != FR_OK) { goto ERROR_FREE; } result = dir_register(dp_new); if (result != FR_OK) { goto ERROR_FREE; } } /* update new dir entry with old info */ result = update_dir(dp_new, finfo_old); if (result != FR_OK) { goto ERROR_FREE; } result = dir_read(dp_new, 0); if (result != FR_OK) { goto ERROR_FREE; } dp_new->blk_ofs = dir_ofs(dp_new); get_fileinfo(dp_new, finfo_new); dfp_new->fno.fp_list.pstNext = dfp_old->fno.fp_list.pstNext; dfp_new->fno.fp_list.pstPrev = dfp_old->fno.fp_list.pstPrev; ret = memcpy_s(dfp_old, sizeof(DIR_FILE), dfp_new, sizeof(DIR_FILE)); if (ret != 0) { result = FR_NOT_ENOUGH_CORE; goto ERROR_FREE; } free(dfp_new); unlock_fs(fs, FR_OK); FREE_NAMBUF(); return fatfs_sync(old_vnode->originMount->mountFlags, fs); ERROR_FREE: free(dfp_new); ERROR_UNLOCK: unlock_fs(fs, result); FREE_NAMBUF(); return -fatfs_2_vfs(result); } static int fatfs_erase(los_part *part, int option) { int opt = option; if ((UINT)opt & FMT_ERASE) { opt = (UINT)opt & (~FMT_ERASE); if (EraseDiskByID(part->disk_id, part->sector_start, part->sector_count) != LOS_OK) { PRINTK("Disk erase error.\n"); return -1; } } if (opt != FM_FAT && opt != FM_FAT32) { opt = FM_ANY; } return opt; } static int fatfs_set_part_info(los_part *part) { los_disk *disk = NULL; char *buf = NULL; int ret; /* If there is no MBR before, the partition info needs to be changed after mkfs */ if (part->type != EMMC && part->part_no_mbr == 0) { disk = get_disk(part->disk_id); if (disk == NULL) { return -EIO; } buf = (char *)zalloc(disk->sector_size); if (buf == NULL) { return -ENOMEM; } (void)memset_s(buf, disk->sector_size, 0, disk->sector_size); ret = los_disk_read(part->disk_id, buf, 0, 1, TRUE); /* TRUE when not reading large data */ if (ret < 0) { free(buf); return -EIO; } part->sector_start = LD_DWORD_DISK(&buf[PAR_OFFSET + PAR_START_OFFSET]); part->sector_count = LD_DWORD_DISK(&buf[PAR_OFFSET + PAR_COUNT_OFFSET]); part->part_no_mbr = 1; part->filesystem_type = buf[PAR_OFFSET + PAR_TYPE_OFFSET]; free(buf); } return 0; } int fatfs_mkfs (struct Vnode *device, int sectors, int option) { BYTE *work_buff = NULL; los_part *part = NULL; FRESULT result; int ret; part = los_part_find(device); if (part == NULL || device->data == NULL) { return -ENODEV; } if (sectors < 0 || sectors > FAT32_MAX_CLUSTER_SIZE || ((DWORD)sectors & ((DWORD)sectors - 1))) { return -EINVAL; } if (option != FMT_FAT && option != FMT_FAT32 && option != FMT_ANY && option != FMT_ERASE) { return -EINVAL; } if (part->part_name != NULL) { /* The part is mounted */ return -EBUSY; } option = fatfs_erase(part, option); work_buff = (BYTE *)zalloc(FF_MAX_SS); if (work_buff == NULL) { return -ENOMEM; } result = _mkfs(part, sectors, option, work_buff, FF_MAX_SS); free(work_buff); if (result != FR_OK) { return -fatfs_2_vfs(result); } #ifdef LOSCFG_FS_FAT_CACHE ret = OsSdSync(part->disk_id); if (ret != 0) { return -EIO; } #endif ret = fatfs_set_part_info(part); if (ret != 0) { return -EIO; } return 0; } int fatfs_mkdir(struct Vnode *parent, const char *name, mode_t mode, struct Vnode **vpp) { return fatfs_create_obj(parent, name, mode, vpp, AM_DIR, NULL); } int fatfs_rmdir(struct Vnode *parent, struct Vnode *vp, const char *name) { FATFS *fs = (FATFS *)vp->originMount->data; DIR_FILE *dfp = (DIR_FILE *)vp->data; FILINFO *finfo = &(dfp->fno); DIR *dp = &(dfp->f_dir); DIR dir_sub; FRESULT result = FR_OK; int ret; if (finfo->fattrib & AM_ARC) { result = FR_NO_DIR; goto ERROR_OUT; } DEF_NAMBUF; INIT_NAMBUF(fs); ret = lock_fs(fs); if (ret == FALSE) { result = FR_TIMEOUT; goto ERROR_OUT; } dir_sub.obj.fs = fs; dir_sub.obj.sclust = finfo->sclst; result = dir_sdi(&dir_sub, 0); if (result != FR_OK) { goto ERROR_UNLOCK; } result = dir_read(&dir_sub, 0); if (result == FR_OK) { result = FR_NO_EMPTY_DIR; goto ERROR_UNLOCK; } result = dir_remove(dp); /* remove directory entry */ if (result != FR_OK) { goto ERROR_UNLOCK; } /* Directory entry contains at least one cluster */ result = remove_chain(&(dp->obj), finfo->sclst, 0); if (result != FR_OK) { goto ERROR_UNLOCK; } unlock_fs(fs, FR_OK); FREE_NAMBUF(); return fatfs_sync(vp->originMount->mountFlags, fs); ERROR_UNLOCK: unlock_fs(fs, result); FREE_NAMBUF(); ERROR_OUT: return -fatfs_2_vfs(result); } int fatfs_reclaim(struct Vnode *vp) { free(vp->data); vp->data = NULL; return 0; } int fatfs_unlink(struct Vnode *parent, struct Vnode *vp, const char *name) { FATFS *fs = (FATFS *)vp->originMount->data; DIR_FILE *dfp = (DIR_FILE *)vp->data; FILINFO *finfo = &(dfp->fno); DIR *dp = &(dfp->f_dir); FRESULT result = FR_OK; int ret; if (finfo->fattrib & AM_DIR) { result = FR_IS_DIR; goto ERROR_OUT; } ret = lock_fs(fs); if (ret == FALSE) { result = FR_TIMEOUT; goto ERROR_OUT; } result = dir_remove(dp); /* remove directory entry */ if (result != FR_OK) { goto ERROR_UNLOCK; } if (finfo->sclst != 0) { /* if cluster chain exists */ result = remove_chain(&(dp->obj), finfo->sclst, 0); if (result != FR_OK) { goto ERROR_UNLOCK; } } result = sync_fs(fs); if (result != FR_OK) { goto ERROR_UNLOCK; } unlock_fs(fs, FR_OK); return fatfs_sync(vp->originMount->mountFlags, fs); ERROR_UNLOCK: unlock_fs(fs, result); ERROR_OUT: return -fatfs_2_vfs(result); } int fatfs_ioctl(struct file *filep, int req, unsigned long arg) { return -ENOSYS; } #define CHECK_FILE_NUM 3 static inline DWORD combine_time(FILINFO *finfo) { return (finfo->fdate << FTIME_DATE_OFFSET) | finfo->ftime; } static UINT get_oldest_time(DIR_FILE df[], DWORD *oldest_time, UINT len) { int i; DWORD old_time = combine_time(&(df[0].fno)); DWORD time; UINT index = 0; for (i = 1; i < len; i++) { time = combine_time(&(df[i].fno)); if (time < old_time) { old_time = time; index = i; } } *oldest_time = old_time; return index; } int fatfs_fscheck(struct Vnode* vp, struct fs_dirent_s *dir) { FATFS *fs = (FATFS *)vp->originMount->data; DIR_FILE df[CHECK_FILE_NUM] = {0}; DIR *dp = NULL; FILINFO *finfo = &(((DIR_FILE *)(vp->data))->fno); FILINFO fno; DWORD old_time = -1; DWORD time; UINT count; UINT index = 0; los_part *part = NULL; FRESULT result; int ret; if (fs->fs_type != FS_FAT32) { return -EINVAL; } if ((finfo->fattrib & AM_DIR) == 0) { return -ENOTDIR; } ret = fatfs_opendir(vp, dir); if (ret < 0) { return ret; } ret = lock_fs(fs); if (ret == FALSE) { result = FR_TIMEOUT; goto ERROR_WITH_DIR; } dp = (DIR *)dir->u.fs_dir; dp->obj.id = fs->id; for (count = 0; count < CHECK_FILE_NUM; count++) { if ((result = f_readdir(dp, &fno)) != FR_OK) { goto ERROR_UNLOCK; } else { if (fno.fname[0] == 0 || fno.fname[0] == (TCHAR)0xFF) { break; } (void)memcpy_s(&df[count].f_dir, sizeof(DIR), dp, sizeof(DIR)); (void)memcpy_s(&df[count].fno, sizeof(FILINFO), &fno, sizeof(FILINFO)); time = combine_time(&(df[count].fno)); if (time < old_time) { old_time = time; index = count; } } } while ((result = f_readdir(dp, &fno)) == FR_OK) { if (fno.fname[0] == 0 || fno.fname[0] == (TCHAR)0xFF) { break; } time = combine_time(&fno); if (time < old_time) { (void)memcpy_s(&df[index].f_dir, sizeof(DIR), dp, sizeof(DIR)); (void)memcpy_s(&df[index].fno, sizeof(FILINFO), &fno, sizeof(FILINFO)); index = get_oldest_time(df, &old_time, CHECK_FILE_NUM); } } if (result != FR_OK) { goto ERROR_UNLOCK; } for (index = 0; index < count; index++) { result = f_fcheckfat(&df[index]); if (result != FR_OK) { goto ERROR_UNLOCK; } } unlock_fs(fs, FR_OK); ret = fatfs_closedir(vp, dir); if (ret < 0) { return ret; } #ifdef LOSCFG_FS_FAT_CACHE part = get_part((INT)fs->pdrv); if (part != NULL) { (void)OsSdSync(part->disk_id); } #endif return 0; ERROR_UNLOCK: unlock_fs(fs, result); ERROR_WITH_DIR: fatfs_closedir(vp, dir); return -fatfs_2_vfs(result); } int fatfs_symlink(struct Vnode *parentVnode, struct Vnode **newVnode, const char *path, const char *target) { return fatfs_create_obj(parentVnode, path, 0, newVnode, AM_LNK, target); } ssize_t fatfs_readlink(struct Vnode *vnode, char *buffer, size_t bufLen) { int ret; FRESULT res = FR_OK; DWORD clust; QWORD sect; DIR_FILE *dfp = (DIR_FILE *)(vnode->data); DIR *dp = &(dfp->f_dir); FATFS *fs = dp->obj.fs; FILINFO *finfo = &(dfp->fno); size_t targetLen = finfo->fsize; size_t cnt; ret = lock_fs(fs); if (ret == FALSE) { return -EBUSY; } clust = finfo->sclst; sect = clst2sect(fs, clust); /* Get current sector */ if (sect == 0) { res = FR_DISK_ERR; goto ERROUT; } if (move_window(fs, sect) != FR_OK) { res = FR_DISK_ERR; goto ERROUT; } cnt = (bufLen - 1) < targetLen ? (bufLen - 1) : targetLen; ret = LOS_CopyFromKernel(buffer, bufLen, fs->win, cnt); if (ret != EOK) { res = FR_INVALID_PARAMETER; goto ERROUT; } buffer[cnt] = '\0'; unlock_fs(fs, FR_OK); return cnt; ERROUT: unlock_fs(fs, res); return -fatfs_2_vfs(res); } struct VnodeOps fatfs_vops = { /* file ops */ .Getattr = fatfs_stat, .Chattr = fatfs_chattr, .Lookup = fatfs_lookup, .Rename = fatfs_rename, .Create = fatfs_create, .Unlink = fatfs_unlink, .Reclaim = fatfs_reclaim, .Truncate = fatfs_truncate, .Truncate64 = fatfs_truncate64, /* dir ops */ .Opendir = fatfs_opendir, .Readdir = fatfs_readdir, .Rewinddir = fatfs_rewinddir, .Closedir = fatfs_closedir, .Mkdir = fatfs_mkdir, .Rmdir = fatfs_rmdir, .Fscheck = fatfs_fscheck, .Symlink = fatfs_symlink, .Readlink = fatfs_readlink, }; struct MountOps fatfs_mops = { .Mount = fatfs_mount, .Unmount = fatfs_umount, .Statfs = fatfs_statfs, }; struct file_operations_vfs fatfs_fops = { .open = fatfs_open, .read = fatfs_read, .write = fatfs_write, .seek = fatfs_lseek, .close = fatfs_close, .mmap = OsVfsFileMmap, .fallocate = fatfs_fallocate, .fallocate64 = fatfs_fallocate64, .fsync = fatfs_fsync, .ioctl = fatfs_ioctl, }; FSMAP_ENTRY(fat_fsmap, "vfat", fatfs_mops, FALSE, TRUE); #endif /* LOSCFG_FS_FAT */