zephyr/subsys/fs/nffs_fs.c
Peter A. Bigot 312f05eaef subsys/fs: add implementation helper module
File system API functions that operate on paths are passed both the
absolute path including the mount point prefix, and the mount point
within which the path lies.

Unfortunately it's not entirely trivial to convert an arbitrary path
within the file system space to an absolute path within its mount point,
because the path may be to the mount point itself and not end with a
directory separator.  The effect is that a file system implementation
like nffs may be given an empty path when "/" is required.

Add an implementation module that does this transformation and use it to
transform paths within each filesystem wrapper.

Signed-off-by: Peter A. Bigot <pab@pabigot.com>
2019-07-22 12:46:50 +02:00

614 lines
12 KiB
C

/*
* Copyright (c) 2017 Codecoup
*
* SPDX-License-Identifier: Apache-2.0
*/
#include <stdio.h>
#include <string.h>
#include <zephyr/types.h>
#include <errno.h>
#include <init.h>
#include <drivers/flash.h>
#include <fs/fs.h>
#include <sys/crc.h>
#include <sys/__assert.h>
#include <sys/printk.h>
#include <nffs/os.h>
#include <nffs/nffs.h>
#include "fs_impl.h"
#define NFFS_MAX_FILE_NAME 256
/*
* NFFS code keeps fs state in RAM but access to these structures is not
* thread-safe - we need global lock for each fs operation to guarantee two
* threads won't modify NFFS at the same time.
*/
static struct k_mutex nffs_lock;
/*
* TODO: Get rid of global flash_dev which limits
* system to have multiple instances of NFFS.
*/
static struct device *flash_dev;
/* nffs flash area descriptors */
static struct nffs_area_desc descs[CONFIG_NFFS_FILESYSTEM_MAX_AREAS + 1];
K_MEM_SLAB_DEFINE(nffs_file_pool, sizeof(struct nffs_file),
CONFIG_FS_NFFS_NUM_FILES, 4);
K_MEM_SLAB_DEFINE(nffs_dir_pool, sizeof(struct nffs_dir),
CONFIG_FS_NFFS_NUM_DIRS, 4);
K_MEM_SLAB_DEFINE(nffs_inode_entry_pool, sizeof(struct nffs_inode_entry),
CONFIG_FS_NFFS_NUM_INODES, 4);
K_MEM_SLAB_DEFINE(nffs_block_entry_pool, sizeof(struct nffs_hash_entry),
CONFIG_FS_NFFS_NUM_BLOCKS, 4);
K_MEM_SLAB_DEFINE(nffs_cache_inode_pool, sizeof(struct nffs_cache_inode),
CONFIG_FS_NFFS_NUM_CACHE_INODES, 4);
K_MEM_SLAB_DEFINE(nffs_cache_block_pool, sizeof(struct nffs_cache_block),
CONFIG_FS_NFFS_NUM_CACHE_BLOCKS, 4);
static int translate_error(int error)
{
switch (error) {
case FS_EOK:
return 0;
case FS_ECORRUPT:
case FS_EHW:
return -EIO;
case FS_EOFFSET:
case FS_EINVAL:
return -EINVAL;
case FS_ENOMEM:
return -ENOMEM;
case FS_ENOENT:
return -ENOENT;
case FS_EEMPTY:
return -ENODEV;
case FS_EFULL:
return -ENOSPC;
case FS_EUNEXP:
case FS_EOS:
return -EIO;
case FS_EEXIST:
return -EEXIST;
case FS_EACCESS:
return -EACCES;
case FS_EUNINIT:
return -EIO;
}
return -EIO;
}
int nffs_os_mempool_init(void)
{
/*
* Just reinitialize slabs here - this is what original implementation
* does. We assume all references to previously allocated blocks, if
* any, are invalidated in NFFS code already.
*/
k_mem_slab_init(&nffs_file_pool, _k_mem_slab_buf_nffs_file_pool,
sizeof(struct nffs_file),
CONFIG_FS_NFFS_NUM_FILES);
k_mem_slab_init(&nffs_dir_pool, _k_mem_slab_buf_nffs_dir_pool,
sizeof(struct nffs_dir),
CONFIG_FS_NFFS_NUM_DIRS);
k_mem_slab_init(&nffs_inode_entry_pool,
_k_mem_slab_buf_nffs_inode_entry_pool,
sizeof(struct nffs_inode_entry),
CONFIG_FS_NFFS_NUM_INODES);
k_mem_slab_init(&nffs_block_entry_pool,
_k_mem_slab_buf_nffs_block_entry_pool,
sizeof(struct nffs_hash_entry),
CONFIG_FS_NFFS_NUM_BLOCKS);
k_mem_slab_init(&nffs_cache_inode_pool,
_k_mem_slab_buf_nffs_cache_inode_pool,
sizeof(struct nffs_cache_inode),
CONFIG_FS_NFFS_NUM_CACHE_INODES);
k_mem_slab_init(&nffs_cache_block_pool,
_k_mem_slab_buf_nffs_cache_block_pool,
sizeof(struct nffs_cache_block),
CONFIG_FS_NFFS_NUM_CACHE_BLOCKS);
return 0;
}
void *nffs_os_mempool_get(nffs_os_mempool_t *pool)
{
int rc;
void *ptr;
rc = k_mem_slab_alloc(pool, &ptr, K_NO_WAIT);
if (rc) {
ptr = NULL;
}
return ptr;
}
int nffs_os_mempool_free(nffs_os_mempool_t *pool, void *block)
{
k_mem_slab_free(pool, &block);
return 0;
}
int nffs_os_flash_read(uint8_t id, uint32_t address, void *dst,
uint32_t num_bytes)
{
int rc;
rc = flash_read(flash_dev, address, dst, num_bytes);
return rc;
}
int nffs_os_flash_write(uint8_t id, uint32_t address, const void *src,
uint32_t num_bytes)
{
int rc;
rc = flash_write_protection_set(flash_dev, false);
if (rc) {
return rc;
}
rc = flash_write(flash_dev, address, src, num_bytes);
/* Ignore errors here - this does not affect write operation */
(void) flash_write_protection_set(flash_dev, true);
return rc;
}
int nffs_os_flash_erase(uint8_t id, uint32_t address, uint32_t num_bytes)
{
int rc;
rc = flash_write_protection_set(flash_dev, false);
if (rc) {
return rc;
}
rc = flash_erase(flash_dev, address, num_bytes);
/* Ignore errors here - this does not affect erase operation */
(void) flash_write_protection_set(flash_dev, true);
return rc;
}
int nffs_os_flash_info(uint8_t id, uint32_t sector, uint32_t *address,
uint32_t *size)
{
struct flash_pages_info pi;
int rc;
rc = flash_get_page_info_by_idx(flash_dev, sector, &pi);
__ASSERT(rc == 0, "Failed to obtain flash page data");
*address = pi.start_offset;
*size = pi.size;
return 0;
}
uint16_t nffs_os_crc16_ccitt(uint16_t initial_crc, const void *buf, int len,
int final)
{
return crc16(buf, len, 0x1021, initial_crc, final);
}
static int inode_to_dirent(struct nffs_inode_entry *inode,
struct fs_dirent *entry)
{
u8_t name_len;
uint32_t size;
int rc;
rc = nffs_inode_read_filename(inode, sizeof(entry->name), entry->name,
&name_len);
if (rc) {
return rc;
}
if (nffs_hash_id_is_dir(inode->nie_hash_entry.nhe_id)) {
entry->type = FS_DIR_ENTRY_DIR;
entry->size = 0;
} else {
entry->type = FS_DIR_ENTRY_FILE;
rc = nffs_inode_data_len(inode, &size);
if (rc) {
return rc;
}
entry->size = size;
}
return rc;
}
static int nffs_open(struct fs_file_t *zfp, const char *file_name)
{
int rc;
file_name = fs_impl_strip_prefix(file_name, zfp->mp);
k_mutex_lock(&nffs_lock, K_FOREVER);
zfp->filep = NULL;
if (!nffs_misc_ready()) {
k_mutex_unlock(&nffs_lock);
return -ENODEV;
}
rc = nffs_file_open((struct nffs_file **)&zfp->filep, file_name,
FS_ACCESS_READ | FS_ACCESS_WRITE);
k_mutex_unlock(&nffs_lock);
return translate_error(rc);
}
static int nffs_close(struct fs_file_t *zfp)
{
int rc;
k_mutex_lock(&nffs_lock, K_FOREVER);
rc = nffs_file_close(zfp->filep);
if (!rc) {
zfp->filep = NULL;
}
k_mutex_unlock(&nffs_lock);
return translate_error(rc);
}
static int nffs_unlink(struct fs_mount_t *mountp, const char *path)
{
int rc;
path = fs_impl_strip_prefix(path, mountp);
k_mutex_lock(&nffs_lock, K_FOREVER);
rc = nffs_path_unlink(path);
k_mutex_unlock(&nffs_lock);
return translate_error(rc);
}
static ssize_t nffs_read(struct fs_file_t *zfp, void *ptr, size_t size)
{
uint32_t br;
int rc;
k_mutex_lock(&nffs_lock, K_FOREVER);
rc = nffs_file_read(zfp->filep, size, ptr, &br);
k_mutex_unlock(&nffs_lock);
if (rc) {
return translate_error(rc);
}
return br;
}
static ssize_t nffs_write(struct fs_file_t *zfp, const void *ptr, size_t size)
{
int rc;
k_mutex_lock(&nffs_lock, K_FOREVER);
rc = nffs_write_to_file(zfp->filep, ptr, size);
k_mutex_unlock(&nffs_lock);
if (rc) {
return translate_error(rc);
}
/* We need to assume all bytes were written */
return size;
}
static int nffs_seek(struct fs_file_t *zfp, off_t offset, int whence)
{
uint32_t len;
u32_t pos;
int rc;
k_mutex_lock(&nffs_lock, K_FOREVER);
switch (whence) {
case FS_SEEK_SET:
pos = offset;
break;
case FS_SEEK_CUR:
pos = ((struct nffs_file *)zfp->filep)->nf_offset + offset;
break;
case FS_SEEK_END:
rc = nffs_inode_data_len(((struct nffs_file *)zfp->filep)->nf_inode_entry, &len);
if (rc) {
k_mutex_unlock(&nffs_lock);
return -EINVAL;
}
pos = len + offset;
break;
default:
k_mutex_unlock(&nffs_lock);
return -EINVAL;
}
rc = nffs_file_seek(zfp->filep, pos);
k_mutex_unlock(&nffs_lock);
return translate_error(rc);
}
static off_t nffs_tell(struct fs_file_t *zfp)
{
u32_t offset;
k_mutex_lock(&nffs_lock, K_FOREVER);
if (!zfp->filep) {
k_mutex_unlock(&nffs_lock);
return -EIO;
}
offset = ((struct nffs_file *)zfp->filep)->nf_offset;
k_mutex_unlock(&nffs_lock);
return offset;
}
static int nffs_truncate(struct fs_file_t *zfp, off_t length)
{
/*
* FIXME:
* There is no API in NFFS to truncate opened file. For now we return
* ENOTSUP, but this should be revisited if truncation is implemented
* in NFFS at some point.
*/
return -ENOTSUP;
}
static int nffs_sync(struct fs_file_t *zfp)
{
/*
* Files are written to flash immediately so we do not need to support
* sync call - just return success.
*/
return 0;
}
static int nffs_mkdir(struct fs_mount_t *mountp, const char *path)
{
int rc;
path = fs_impl_strip_prefix(path, mountp);
k_mutex_lock(&nffs_lock, K_FOREVER);
if (!nffs_misc_ready()) {
k_mutex_unlock(&nffs_lock);
return -ENODEV;
}
rc = nffs_path_new_dir(path, NULL);
k_mutex_unlock(&nffs_lock);
return translate_error(rc);
}
static int nffs_opendir(struct fs_dir_t *zdp, const char *path)
{
int rc;
path = fs_impl_strip_prefix(path, zdp->mp);
k_mutex_lock(&nffs_lock, K_FOREVER);
zdp->dirp = NULL;
if (!nffs_misc_ready()) {
k_mutex_unlock(&nffs_lock);
return -ENODEV;
}
rc = nffs_dir_open(path, (struct nffs_dir **)&zdp->dirp);
k_mutex_unlock(&nffs_lock);
return translate_error(rc);
}
static int nffs_readdir(struct fs_dir_t *zdp, struct fs_dirent *entry)
{
struct nffs_dirent *dirent;
int rc;
k_mutex_lock(&nffs_lock, K_FOREVER);
rc = nffs_dir_read(zdp->dirp, &dirent);
switch (rc) {
case 0:
rc = inode_to_dirent(dirent->nde_inode_entry, entry);
break;
case FS_ENOENT:
entry->name[0] = 0;
rc = 0;
break;
default:
break;
}
k_mutex_unlock(&nffs_lock);
return translate_error(rc);
}
static int nffs_closedir(struct fs_dir_t *zdp)
{
int rc;
k_mutex_lock(&nffs_lock, K_FOREVER);
rc = nffs_dir_close(zdp->dirp);
if (!rc) {
zdp->dirp = NULL;
}
k_mutex_unlock(&nffs_lock);
return translate_error(rc);
}
static int nffs_stat(struct fs_mount_t *mountp,
const char *path, struct fs_dirent *entry)
{
struct nffs_path_parser parser;
struct nffs_inode_entry *parent;
struct nffs_inode_entry *inode;
int rc;
path = fs_impl_strip_prefix(path, mountp);
k_mutex_lock(&nffs_lock, K_FOREVER);
nffs_path_parser_new(&parser, path);
rc = nffs_path_find(&parser, &inode, &parent);
if (rc == 0) {
rc = inode_to_dirent(inode, entry);
}
k_mutex_unlock(&nffs_lock);
return translate_error(rc);
}
static int nffs_statvfs(struct fs_mount_t *mountp,
const char *path, struct fs_statvfs *stat)
{
/*
* FIXME:
* There is not API to retrieve such data in NFFS.
*/
return -ENOTSUP;
}
static int nffs_rename(struct fs_mount_t *mountp, const char *from,
const char *to)
{
int rc;
from = fs_impl_strip_prefix(from, mountp);
to = fs_impl_strip_prefix(to, mountp);
k_mutex_lock(&nffs_lock, K_FOREVER);
if (!nffs_misc_ready()) {
k_mutex_unlock(&nffs_lock);
return -ENODEV;
}
rc = nffs_path_rename(from, to);
k_mutex_unlock(&nffs_lock);
return translate_error(rc);
}
static int nffs_mount(struct fs_mount_t *mountp)
{
struct nffs_flash_desc *flash_desc =
(struct nffs_flash_desc *)mountp->fs_data;
int cnt;
int rc;
/* Set flash device */
flash_dev = (struct device *)mountp->storage_dev;
/* Set flash descriptor fields */
flash_desc->id = 0;
flash_desc->sector_count = flash_get_page_count(flash_dev);
flash_desc->area_offset = DT_FLASH_AREA_STORAGE_OFFSET;
flash_desc->area_size = DT_FLASH_AREA_STORAGE_SIZE;
rc = nffs_misc_reset();
if (rc) {
return -EIO;
}
cnt = CONFIG_NFFS_FILESYSTEM_MAX_AREAS;
rc = nffs_misc_desc_from_flash_area(flash_desc, &cnt, descs);
if (rc) {
return -EIO;
}
rc = nffs_restore_full(descs);
switch (rc) {
case 0:
break;
case FS_ECORRUPT:
rc = nffs_format_full(descs);
if (rc) {
return -EIO;
}
break;
default:
return -EIO;
}
return 0;
}
/* File system interface */
static struct fs_file_system_t nffs_fs = {
.open = nffs_open,
.close = nffs_close,
.read = nffs_read,
.write = nffs_write,
.lseek = nffs_seek,
.tell = nffs_tell,
.truncate = nffs_truncate,
.sync = nffs_sync,
.opendir = nffs_opendir,
.readdir = nffs_readdir,
.closedir = nffs_closedir,
.mount = nffs_mount,
.unlink = nffs_unlink,
.rename = nffs_rename,
.mkdir = nffs_mkdir,
.stat = nffs_stat,
.statvfs = nffs_statvfs,
};
static int nffs_init(struct device *dev)
{
ARG_UNUSED(dev);
k_mutex_init(&nffs_lock);
return fs_register(FS_NFFS, &nffs_fs);
}
SYS_INIT(nffs_init, APPLICATION, CONFIG_KERNEL_INIT_PRIORITY_DEFAULT);