zephyr/subsys/fs/fat_fs.c
Carlo Kirchmeier 1ea0267a31 fs: enable fstab devicetree integration for fatfs
In order to use fstab with fatfs in addition to
littlefs a new driver fstab,fatfs was introduced
containing the required logic to detect and if
needed automount fatfs devices specified in fstab.
Additionally as the partition phandle is not needed
by fatfs and currently is specific to littlefs it
was moved from the common dts binding into the
littlefs binding.

Signed-off-by: Carlo Kirchmeier <carlo.kirchmeier@zuehlke.com>
Co-authored-by: Carlo Kirchmeier <carlo.kirchmeier@zuehlke.com>
Co-authored-by: Dominik Ermel <dominik.ermel@nordicsemi.no>
2025-03-15 13:07:40 +01:00

639 lines
15 KiB
C

/*
* Copyright (c) 2016 Intel Corporation.
* Copyright 2024 NXP
*
* SPDX-License-Identifier: Apache-2.0
*/
#include <stdio.h>
#include <string.h>
#include <zephyr/kernel.h>
#include <zephyr/types.h>
#include <errno.h>
#include <zephyr/init.h>
#include <zephyr/fs/fs.h>
#include <zephyr/fs/fs_sys.h>
#include <zephyr/sys/__assert.h>
#include <ff.h>
#include <diskio.h>
#include <zfs_diskio.h> /* Zephyr specific FatFS API */
#include <zephyr/logging/log.h>
LOG_MODULE_DECLARE(fs, CONFIG_FS_LOG_LEVEL);
#define FATFS_MAX_FILE_NAME 12 /* Uses 8.3 SFN */
/* Memory pool for FatFs directory objects */
K_MEM_SLAB_DEFINE(fatfs_dirp_pool, sizeof(DIR),
CONFIG_FS_FATFS_NUM_DIRS, 4);
/* Memory pool for FatFs file objects */
K_MEM_SLAB_DEFINE(fatfs_filep_pool, sizeof(FIL),
CONFIG_FS_FATFS_NUM_FILES, 4);
static int translate_error(int error)
{
switch (error) {
case FR_OK:
return 0;
case FR_NO_FILE:
case FR_NO_PATH:
case FR_INVALID_NAME:
return -ENOENT;
case FR_DENIED:
return -EACCES;
case FR_EXIST:
return -EEXIST;
case FR_INVALID_OBJECT:
return -EBADF;
case FR_WRITE_PROTECTED:
return -EROFS;
case FR_INVALID_DRIVE:
case FR_NOT_ENABLED:
case FR_NO_FILESYSTEM:
return -ENODEV;
case FR_NOT_ENOUGH_CORE:
return -ENOMEM;
case FR_TOO_MANY_OPEN_FILES:
return -EMFILE;
case FR_INVALID_PARAMETER:
return -EINVAL;
case FR_LOCKED:
case FR_TIMEOUT:
case FR_MKFS_ABORTED:
case FR_DISK_ERR:
case FR_INT_ERR:
case FR_NOT_READY:
return -EIO;
}
return -EIO;
}
static int translate_disk_error(int error)
{
switch (error) {
case RES_OK:
return 0;
case RES_WRPRT:
return -EPERM;
case RES_PARERR:
return -EINVAL;
case RES_NOTRDY:
case RES_ERROR:
return -EIO;
}
return -EIO;
}
/* Converts a zephyr path like /SD:/foo into a path digestible by FATFS by stripping the
* leading slash, i.e. SD:/foo.
*/
static const char *translate_path(const char *path)
{
/* this is guaranteed by the fs subsystem */
__ASSERT_NO_MSG(path[0] == '/');
return &path[1];
}
static uint8_t translate_flags(fs_mode_t flags)
{
uint8_t fat_mode = 0;
fat_mode |= (flags & FS_O_READ) ? FA_READ : 0;
fat_mode |= (flags & FS_O_WRITE) ? FA_WRITE : 0;
fat_mode |= (flags & FS_O_CREATE) ? FA_OPEN_ALWAYS : 0;
/* NOTE: FA_APPEND is not translated because FAT FS does not
* support append semantics of the Zephyr, where file position
* is forwarded to the end before each write, the fatfs_write
* will be tasked with setting a file position to the end,
* if FA_APPEND flag is present.
*/
return fat_mode;
}
static int fatfs_open(struct fs_file_t *zfp, const char *file_name,
fs_mode_t mode)
{
FRESULT res;
uint8_t fs_mode;
void *ptr;
if (k_mem_slab_alloc(&fatfs_filep_pool, &ptr, K_NO_WAIT) == 0) {
(void)memset(ptr, 0, sizeof(FIL));
zfp->filep = ptr;
} else {
return -ENOMEM;
}
fs_mode = translate_flags(mode);
res = f_open(zfp->filep, translate_path(file_name), fs_mode);
if (res != FR_OK) {
k_mem_slab_free(&fatfs_filep_pool, ptr);
zfp->filep = NULL;
}
return translate_error(res);
}
static int fatfs_close(struct fs_file_t *zfp)
{
FRESULT res;
res = f_close(zfp->filep);
/* Free file ptr memory */
k_mem_slab_free(&fatfs_filep_pool, zfp->filep);
zfp->filep = NULL;
return translate_error(res);
}
static int fatfs_unlink(struct fs_mount_t *mountp, const char *path)
{
int res = -ENOTSUP;
#if !defined(CONFIG_FS_FATFS_READ_ONLY)
res = f_unlink(translate_path(path));
res = translate_error(res);
#endif
return res;
}
static int fatfs_rename(struct fs_mount_t *mountp, const char *from,
const char *to)
{
int res = -ENOTSUP;
#if !defined(CONFIG_FS_FATFS_READ_ONLY)
FILINFO fno;
/* Check if 'to' path exists; remove it if it does */
res = f_stat(translate_path(to), &fno);
if (res == FR_OK) {
res = f_unlink(translate_path(to));
if (res != FR_OK) {
return translate_error(res);
}
}
res = f_rename(translate_path(from), translate_path(to));
res = translate_error(res);
#endif
return res;
}
static ssize_t fatfs_read(struct fs_file_t *zfp, void *ptr, size_t size)
{
FRESULT res;
unsigned int br;
res = f_read(zfp->filep, ptr, size, &br);
if (res != FR_OK) {
return translate_error(res);
}
return br;
}
static ssize_t fatfs_write(struct fs_file_t *zfp, const void *ptr, size_t size)
{
int res = -ENOTSUP;
#if !defined(CONFIG_FS_FATFS_READ_ONLY)
unsigned int bw;
off_t pos = f_size((FIL *)zfp->filep);
res = FR_OK;
/* FA_APPEND flag means that file has been opened for append.
* The FAT FS write does not support the POSIX append semantics,
* to always write at the end of file, so set file position
* at the end before each write if FA_APPEND is set.
*/
if (zfp->flags & FS_O_APPEND) {
res = f_lseek(zfp->filep, pos);
}
if (res == FR_OK) {
res = f_write(zfp->filep, ptr, size, &bw);
}
if (res != FR_OK) {
res = translate_error(res);
} else {
res = bw;
}
#endif
return res;
}
static int fatfs_seek(struct fs_file_t *zfp, off_t offset, int whence)
{
FRESULT res = FR_OK;
off_t pos;
switch (whence) {
case FS_SEEK_SET:
pos = offset;
break;
case FS_SEEK_CUR:
pos = f_tell((FIL *)zfp->filep) + offset;
break;
case FS_SEEK_END:
pos = f_size((FIL *)zfp->filep) + offset;
break;
default:
return -EINVAL;
}
if ((pos < 0) || (pos > f_size((FIL *)zfp->filep))) {
return -EINVAL;
}
res = f_lseek(zfp->filep, pos);
return translate_error(res);
}
static off_t fatfs_tell(struct fs_file_t *zfp)
{
return f_tell((FIL *)zfp->filep);
}
static int fatfs_truncate(struct fs_file_t *zfp, off_t length)
{
int res = -ENOTSUP;
#if !defined(CONFIG_FS_FATFS_READ_ONLY)
off_t cur_length = f_size((FIL *)zfp->filep);
/* f_lseek expands file if new position is larger than file size */
res = f_lseek(zfp->filep, length);
if (res != FR_OK) {
return translate_error(res);
}
if (length < cur_length) {
res = f_truncate(zfp->filep);
} else {
/*
* Get actual length after expansion. This could be
* less if there was not enough space in the volume
* to expand to the requested length
*/
length = f_tell((FIL *)zfp->filep);
res = f_lseek(zfp->filep, cur_length);
if (res != FR_OK) {
return translate_error(res);
}
/*
* The FS module does caching and optimization of
* writes. Here we write 1 byte at a time to avoid
* using additional code and memory for doing any
* optimization.
*/
unsigned int bw;
uint8_t c = 0U;
for (int i = cur_length; i < length; i++) {
res = f_write(zfp->filep, &c, 1, &bw);
if (res != FR_OK) {
break;
}
}
}
res = translate_error(res);
#endif
return res;
}
static int fatfs_sync(struct fs_file_t *zfp)
{
int res = -ENOTSUP;
#if !defined(CONFIG_FS_FATFS_READ_ONLY)
res = f_sync(zfp->filep);
res = translate_error(res);
#endif
return res;
}
static int fatfs_mkdir(struct fs_mount_t *mountp, const char *path)
{
int res = -ENOTSUP;
#if !defined(CONFIG_FS_FATFS_READ_ONLY)
res = f_mkdir(translate_path(path));
res = translate_error(res);
#endif
return res;
}
static int fatfs_opendir(struct fs_dir_t *zdp, const char *path)
{
FRESULT res;
void *ptr;
if (k_mem_slab_alloc(&fatfs_dirp_pool, &ptr, K_NO_WAIT) == 0) {
(void)memset(ptr, 0, sizeof(DIR));
zdp->dirp = ptr;
} else {
return -ENOMEM;
}
res = f_opendir(zdp->dirp, translate_path(path));
if (res != FR_OK) {
k_mem_slab_free(&fatfs_dirp_pool, ptr);
zdp->dirp = NULL;
}
return translate_error(res);
}
static int fatfs_readdir(struct fs_dir_t *zdp, struct fs_dirent *entry)
{
FRESULT res;
FILINFO fno;
res = f_readdir(zdp->dirp, &fno);
if (res == FR_OK) {
strcpy(entry->name, fno.fname);
if (entry->name[0] != 0) {
entry->type = ((fno.fattrib & AM_DIR) ?
FS_DIR_ENTRY_DIR : FS_DIR_ENTRY_FILE);
entry->size = fno.fsize;
}
}
return translate_error(res);
}
static int fatfs_closedir(struct fs_dir_t *zdp)
{
FRESULT res;
res = f_closedir(zdp->dirp);
/* Free file ptr memory */
k_mem_slab_free(&fatfs_dirp_pool, zdp->dirp);
return translate_error(res);
}
static int fatfs_stat(struct fs_mount_t *mountp,
const char *path, struct fs_dirent *entry)
{
FRESULT res;
FILINFO fno;
res = f_stat(translate_path(path), &fno);
if (res == FR_OK) {
entry->type = ((fno.fattrib & AM_DIR) ?
FS_DIR_ENTRY_DIR : FS_DIR_ENTRY_FILE);
strcpy(entry->name, fno.fname);
entry->size = fno.fsize;
}
return translate_error(res);
}
static int fatfs_statvfs(struct fs_mount_t *mountp,
const char *path, struct fs_statvfs *stat)
{
int res = -ENOTSUP;
#if !defined(CONFIG_FS_FATFS_READ_ONLY)
FATFS *fs;
DWORD f_bfree = 0;
res = f_getfree(translate_path(mountp->mnt_point), &f_bfree, &fs);
if (res != FR_OK) {
return -EIO;
}
stat->f_bfree = f_bfree;
/*
* If FF_MIN_SS and FF_MAX_SS differ, variable sector size support is
* enabled and the file system object structure contains the actual sector
* size, otherwise it is configured to a fixed value give by FF_MIN_SS.
*/
#if FF_MAX_SS != FF_MIN_SS
stat->f_bsize = fs->ssize;
#else
stat->f_bsize = FF_MIN_SS;
#endif
stat->f_frsize = fs->csize * stat->f_bsize;
stat->f_blocks = (fs->n_fatent - 2);
res = translate_error(res);
#endif
return res;
}
static int fatfs_mount(struct fs_mount_t *mountp)
{
FRESULT res;
res = f_mount((FATFS *)mountp->fs_data, translate_path(mountp->mnt_point), 1);
#if defined(CONFIG_FS_FATFS_MOUNT_MKFS)
if (res == FR_NO_FILESYSTEM &&
(mountp->flags & FS_MOUNT_FLAG_READ_ONLY) != 0) {
return -EROFS;
}
/* If no file system found then create one */
if (res == FR_NO_FILESYSTEM &&
(mountp->flags & FS_MOUNT_FLAG_NO_FORMAT) == 0) {
uint8_t work[FF_MAX_SS];
MKFS_PARM mkfs_opt = {
.fmt = FM_ANY | FM_SFD, /* Any suitable FAT */
.n_fat = 1, /* One FAT fs table */
.align = 0, /* Get sector size via diskio query */
.n_root = CONFIG_FS_FATFS_MAX_ROOT_ENTRIES,
.au_size = 0 /* Auto calculate cluster size */
};
res = f_mkfs(translate_path(mountp->mnt_point), &mkfs_opt, work, sizeof(work));
if (res == FR_OK) {
res = f_mount((FATFS *)mountp->fs_data,
translate_path(mountp->mnt_point), 1);
}
}
#endif /* CONFIG_FS_FATFS_MOUNT_MKFS */
if (res == FR_OK) {
mountp->flags |= FS_MOUNT_FLAG_USE_DISK_ACCESS;
}
return translate_error(res);
}
static int fatfs_unmount(struct fs_mount_t *mountp)
{
FRESULT res;
DRESULT disk_res;
uint8_t param = DISK_IOCTL_POWER_OFF;
res = f_mount(NULL, translate_path(mountp->mnt_point), 0);
if (res != FR_OK) {
LOG_ERR("Unmount failed (%d)", res);
return translate_error(res);
}
/* Make direct disk IOCTL call to deinit disk */
disk_res = disk_ioctl(((FATFS *)mountp->fs_data)->pdrv, CTRL_POWER, &param);
if (disk_res != RES_OK) {
LOG_ERR("Could not power off disk (%d)", disk_res);
return translate_disk_error(disk_res);
}
return 0;
}
#if defined(CONFIG_FILE_SYSTEM_MKFS) && defined(CONFIG_FS_FATFS_MKFS)
static MKFS_PARM def_cfg = {
.fmt = FM_ANY | FM_SFD, /* Any suitable FAT */
.n_fat = 1, /* One FAT fs table */
.align = 0, /* Get sector size via diskio query */
.n_root = CONFIG_FS_FATFS_MAX_ROOT_ENTRIES,
.au_size = 0 /* Auto calculate cluster size */
};
static int fatfs_mkfs(uintptr_t dev_id, void *cfg, int flags)
{
FRESULT res;
uint8_t work[FF_MAX_SS];
MKFS_PARM *mkfs_opt = &def_cfg;
if (cfg != NULL) {
mkfs_opt = (MKFS_PARM *)cfg;
}
res = f_mkfs((char *)dev_id, mkfs_opt, work, sizeof(work));
return translate_error(res);
}
#endif /* CONFIG_FILE_SYSTEM_MKFS && FS_FATFS_MKFS */
/* File system interface */
static const struct fs_file_system_t fatfs_fs = {
.open = fatfs_open,
.close = fatfs_close,
.read = fatfs_read,
.write = fatfs_write,
.lseek = fatfs_seek,
.tell = fatfs_tell,
.truncate = fatfs_truncate,
.sync = fatfs_sync,
.opendir = fatfs_opendir,
.readdir = fatfs_readdir,
.closedir = fatfs_closedir,
.mount = fatfs_mount,
.unmount = fatfs_unmount,
.unlink = fatfs_unlink,
.rename = fatfs_rename,
.mkdir = fatfs_mkdir,
.stat = fatfs_stat,
.statvfs = fatfs_statvfs,
#if defined(CONFIG_FILE_SYSTEM_MKFS) && defined(CONFIG_FS_FATFS_MKFS)
.mkfs = fatfs_mkfs,
#endif
};
#define DT_DRV_COMPAT zephyr_fstab_fatfs
#define DEFINE_FS(inst) \
BUILD_ASSERT(DT_INST_PROP(inst, disk_access), "FATFS needs disk-access"); \
BUILD_ASSERT(!DT_INST_PROP(inst, read_only), \
"READ_ONLY not supported for individual instances see FS_FATFS_READ_ONLY"); \
BUILD_ASSERT(!DT_INST_PROP(inst, no_format), \
"NO_FORMAT not supported for individual instanzes FS_FATFS_MKFS"); \
static FATFS fs_data_##inst; \
struct fs_mount_t FS_FSTAB_ENTRY(DT_DRV_INST(inst)) = { \
.type = FS_FATFS, \
.mnt_point = DT_INST_PROP(inst, mount_point), \
.fs_data = &fs_data_##inst, \
.storage_dev = NULL, \
.flags = FSTAB_ENTRY_DT_MOUNT_FLAGS(DT_DRV_INST(inst)), \
};
DT_INST_FOREACH_STATUS_OKAY(DEFINE_FS);
#ifdef CONFIG_FS_FATFS_FSTAB_AUTOMOUNT
#define REFERENCE_MOUNT(inst) (&FS_FSTAB_ENTRY(DT_DRV_INST(inst))),
static void automount_if_enabled(struct fs_mount_t *mountp)
{
int ret = 0;
if ((mountp->flags & FS_MOUNT_FLAG_AUTOMOUNT) != 0) {
ret = fs_mount(mountp);
if (ret < 0) {
LOG_ERR("Error mounting filesystem: at %s: %d", mountp->mnt_point, ret);
} else {
LOG_DBG("FATFS Filesystem \"%s\" initialized", mountp->mnt_point);
}
}
}
#endif /* CONFIG_FS_FATFS_FSTAB_AUTOMOUNT */
#if CONFIG_FS_FATFS_CUSTOM_MOUNT_POINT_COUNT
const char *VolumeStr[CONFIG_FS_FATFS_CUSTOM_MOUNT_POINT_COUNT];
#endif /* CONFIG_FS_FATFS_CUSTOM_MOUNT_POINT_COUNT */
static int fatfs_init(void)
{
#if CONFIG_FS_FATFS_CUSTOM_MOUNT_POINT_COUNT
static char mount_points[] = CONFIG_FS_FATFS_CUSTOM_MOUNT_POINTS;
int mount_point_count = 0;
VolumeStr[0] = mount_points;
for (int i = 0; i < ARRAY_SIZE(mount_points) - 1; i++) {
if (mount_points[i] == ',') {
mount_points[i] = 0;
mount_point_count++;
if (mount_point_count >= ARRAY_SIZE(VolumeStr)) {
LOG_ERR("Mount point count not sufficient for defined mount "
"points.");
return -1;
}
VolumeStr[mount_point_count] = &mount_points[i + 1];
}
}
#endif /* CONFIG_FS_FATFS_CUSTOM_MOUNT_POINT_COUNT */
int rc = fs_register(FS_FATFS, &fatfs_fs);
#ifdef CONFIG_FS_FATFS_FSTAB_AUTOMOUNT
if (rc == 0) {
struct fs_mount_t *partitions[] = {DT_INST_FOREACH_STATUS_OKAY(REFERENCE_MOUNT)};
for (size_t i = 0; i < ARRAY_SIZE(partitions); i++) {
struct fs_mount_t *mpi = partitions[i];
automount_if_enabled(mpi);
}
}
#endif /* CONFIG_FS_FATFS_FSTAB_AUTOMOUNT */
return rc;
}
SYS_INIT(fatfs_init, POST_KERNEL, CONFIG_FILE_SYSTEM_INIT_PRIORITY);