nuttx/fs/smartfs/smartfs_utils.c

2154 lines
65 KiB
C

/****************************************************************************
* fs/smartfs/smartfs_utils.c
*
* Copyright (C) 2013-2014 Ken Pettit. All rights reserved.
* Author: Ken Pettit <pettitkd@gmail.com>
*
* 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 NuttX 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 OWNER 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.
*
****************************************************************************/
/****************************************************************************
* Included Files
****************************************************************************/
#include <nuttx/config.h>
#include <sys/types.h>
#include <stdint.h>
#include <stdbool.h>
#include <stdlib.h>
#include <string.h>
#include <time.h>
#include <assert.h>
#include <errno.h>
#include <debug.h>
#include <nuttx/kmalloc.h>
#include <nuttx/fs/fs.h>
#include <nuttx/fs/ioctl.h>
#include "smartfs.h"
/****************************************************************************
* Pre-processor Definitions
****************************************************************************/
#define WORKBUFFER_SIZE 256
/****************************************************************************
* Private Data
****************************************************************************/
#if defined(CONFIG_SMARTFS_MULTI_ROOT_DIRS) || \
(defined(CONFIG_FS_PROCFS) && !defined(CONFIG_FS_PROCFS_EXCLUDE_SMARTFS))
static struct smartfs_mountpt_s *g_mounthead = NULL;
#endif
/****************************************************************************
* Public Functions
****************************************************************************/
/****************************************************************************
* Name: smartfs_semtake
****************************************************************************/
int smartfs_semtake(struct smartfs_mountpt_s *fs)
{
return nxsem_wait_uninterruptible(fs->fs_sem);
}
/****************************************************************************
* Name: smartfs_semgive
****************************************************************************/
void smartfs_semgive(struct smartfs_mountpt_s *fs)
{
nxsem_post(fs->fs_sem);
}
/****************************************************************************
* Name: smartfs_rdle16
*
* Description:
* Get a (possibly unaligned) 16-bit little endian value.
*
* Input Parameters:
* val - A pointer to the first byte of the little endian value.
*
* Returned Value:
* A uint16_t representing the whole 16-bit integer value
*
****************************************************************************/
uint16_t smartfs_rdle16(FAR const void *val)
{
return (uint16_t)((FAR const uint8_t *)val)[1] << 8 |
(uint16_t)((FAR const uint8_t *)val)[0];
}
/****************************************************************************
* Name: smartfs_wrle16
*
* Description:
* Put a (possibly unaligned) 16-bit little endian value.
*
* Input Parameters:
* dest - A pointer to the first byte to save the little endian value.
* val - The 16-bit value to be saved.
*
* Returned Value:
* None
*
****************************************************************************/
void smartfs_wrle16(FAR void *dest, uint16_t val)
{
((FAR uint8_t *) dest)[0] = val & 0xff; /* Little endian means LS byte first in byte stream */
((FAR uint8_t *) dest)[1] = val >> 8;
}
/****************************************************************************
* Name: smartfs_rdle32
*
* Description:
* Get a (possibly unaligned) 32-bit little endian value.
*
* Input Parameters:
* val - A pointer to the first byte of the little endian value.
*
* Returned Value:
* A uint32_t representing the whole 32-bit integer value
*
****************************************************************************/
uint32_t smartfs_rdle32(FAR const void *val)
{
/* Little endian means LS halfword first in byte stream */
return (uint32_t)smartfs_rdle16(&((FAR const uint8_t *)val)[2]) << 16 |
(uint32_t)smartfs_rdle16(val);
}
/****************************************************************************
* Name: smartfs_wrle32
*
* Description:
* Put a (possibly unaligned) 32-bit little endian value.
*
* Input Parameters:
* dest - A pointer to the first byte to save the little endian value.
* val - The 32-bit value to be saved.
*
* Returned Value:
* None
*
****************************************************************************/
void smartfs_wrle32(uint8_t *dest, uint32_t val)
{
/* Little endian means LS halfword first in byte stream */
smartfs_wrle16(dest, (uint16_t)(val & 0xffff));
smartfs_wrle16(dest + 2, (uint16_t)(val >> 16));
}
/****************************************************************************
* Name: smartfs_mount
*
* Description: This function is called only when the mountpoint is first
* established. It initializes the mountpoint structure and verifies
* that a valid SMART filesystem is provided by the block driver.
*
* The caller should hold the mountpoint semaphore
*
****************************************************************************/
int smartfs_mount(struct smartfs_mountpt_s *fs, bool writeable)
{
FAR struct inode *inode;
struct geometry geo;
int ret = OK;
#if defined(CONFIG_SMARTFS_MULTI_ROOT_DIRS)
struct smartfs_mountpt_s *nextfs;
#endif
/* Assume that the mount is not successful */
fs->fs_mounted = false;
/* Check if there is media available */
inode = fs->fs_blkdriver;
if (!inode || !inode->u.i_bops || !inode->u.i_bops->geometry ||
inode->u.i_bops->geometry(inode, &geo) != OK || !geo.geo_available)
{
ret = -ENODEV;
goto errout;
}
/* Make sure that that the media is write-able (if write access is needed) */
if (writeable && !geo.geo_writeenabled)
{
ret = -EACCES;
goto errout;
}
/* Get the SMART low-level format information to validate the device has
* been formatted and scan properly for logical to physical sector mapping.
*/
ret = FS_IOCTL(fs, BIOC_GETFORMAT, (unsigned long) &fs->fs_llformat);
if (ret != OK)
{
ferr("ERROR: Error getting device low level format: %d\n", ret);
goto errout;
}
/* Validate the low-level format is valid */
if (!(fs->fs_llformat.flags & SMART_FMT_ISFORMATTED))
{
ferr("ERROR: No low-level format found\n");
ret = -ENODEV;
goto errout;
}
/* Allocate a read/write buffer */
#ifdef CONFIG_SMARTFS_MULTI_ROOT_DIRS
/* Scan linked list of mounted file systems to find another FS with
* the same blockdriver. We will reuse the buffers.
*/
nextfs = g_mounthead;
while (nextfs != NULL)
{
/* Test if this FS uses the same block driver */
if (nextfs->fs_blkdriver == fs->fs_blkdriver)
{
/* Yep, it's the same block driver. Reuse the buffers.
* we can do this because we are protected by the same
* semaphore.
*/
fs->fs_rwbuffer = nextfs->fs_rwbuffer;
fs->fs_workbuffer = nextfs->fs_workbuffer;
break;
}
/* Advance to next FS */
nextfs = nextfs->fs_next;
}
/* If we didn't find a FS above, then allocate some buffers */
if (nextfs == NULL)
{
fs->fs_rwbuffer = (char *) kmm_malloc(fs->fs_llformat.availbytes);
fs->fs_workbuffer = (char *) kmm_malloc(WORKBUFFER_SIZE);
}
/* Now add ourselves to the linked list of SMART mounts */
fs->fs_next = g_mounthead;
g_mounthead = fs;
/* Set our root directory sector based on the directory entry
* reported by the block driver (based on which device is
* associated with this mount.
*/
fs->fs_rootsector = SMARTFS_ROOT_DIR_SECTOR + fs->fs_llformat.rootdirnum;
#else /* CONFIG_SMARTFS_MULTI_ROOT_DIRS */
#if defined(CONFIG_FS_PROCFS) && !defined(CONFIG_FS_PROCFS_EXCLUDE_SMARTFS)
/* Now add ourselves to the linked list of SMART mounts */
fs->fs_next = g_mounthead;
g_mounthead = fs;
#endif
fs->fs_rwbuffer = (char *) kmm_malloc(fs->fs_llformat.availbytes);
fs->fs_workbuffer = (char *) kmm_malloc(WORKBUFFER_SIZE);
fs->fs_rootsector = SMARTFS_ROOT_DIR_SECTOR;
#endif /* CONFIG_SMARTFS_MULTI_ROOT_DIRS */
/* We did it! */
fs->fs_mounted = TRUE;
finfo("SMARTFS:\n");
finfo("\t Sector size: %d\n", fs->fs_llformat.sectorsize);
finfo("\t Bytes/sector %d\n", fs->fs_llformat.availbytes);
finfo("\t Num sectors: %d\n", fs->fs_llformat.nsectors);
finfo("\t Free sectors: %d\n", fs->fs_llformat.nfreesectors);
finfo("\t Max filename: %d\n", CONFIG_SMARTFS_MAXNAMLEN);
#ifdef CONFIG_SMARTFS_MULTI_ROOT_DIRS
finfo("\t RootDirEntries: %d\n", fs->fs_llformat.nrootdirentries);
#endif
finfo("\t RootDirSector: %d\n", fs->fs_rootsector);
errout:
return ret;
}
/****************************************************************************
* Name: smartfs_unmount
*
* Description: This function is called only when the mountpoint is being
* unbound. If we are serving multiple directories, then we have to
* remove ourselves from the mount linked list, and potentially free
* the shared buffers.
*
* The caller should hold the mountpoint semaphore
*
****************************************************************************/
int smartfs_unmount(struct smartfs_mountpt_s *fs)
{
int ret = OK;
struct inode *inode;
#if defined(CONFIG_SMARTFS_MULTI_ROOT_DIRS) || \
(defined(CONFIG_FS_PROCFS) && !defined(CONFIG_FS_PROCFS_EXCLUDE_SMARTFS))
struct smartfs_mountpt_s *nextfs;
struct smartfs_mountpt_s *prevfs;
int count = 0;
int found = FALSE;
#endif
#if defined(CONFIG_SMARTFS_MULTI_ROOT_DIRS) || \
(defined(CONFIG_FS_PROCFS) && !defined(CONFIG_FS_PROCFS_EXCLUDE_SMARTFS))
/* Start at the head of the mounts and search for our entry. Also
* count the number of entries that match our blkdriver.
*/
nextfs = g_mounthead;
prevfs = NULL;
while (nextfs != NULL)
{
/* Test if this FS's blkdriver matches ours (it could be us) */
if (nextfs->fs_blkdriver == fs->fs_blkdriver)
count++;
/* Test if this entry is our's */
if (nextfs == fs)
{
found = TRUE;
}
/* Keep track of the previous entry until our's is found */
if (!found)
{
/* Save this entry as the previous entry */
prevfs = nextfs;
}
/* Advance to the next entry */
nextfs = nextfs->fs_next;
}
/* Ensure we found our FS */
if (!found)
{
/* Our entry not found! Invalid unmount or bug somewhere */
return -EINVAL;
}
/* If the count is only one, then we need to delete the shared
* buffers because we are the last ones.
*/
if (count == 1)
{
/* Close the block driver */
if (fs->fs_blkdriver)
{
inode = fs->fs_blkdriver;
if (inode)
{
if (inode->u.i_bops && inode->u.i_bops->close)
{
inode->u.i_bops->close(inode);
}
}
}
/* Free the buffers */
kmm_free(fs->fs_rwbuffer);
kmm_free(fs->fs_workbuffer);
/* Set the buffer's to invalid value to catch program bugs */
fs->fs_rwbuffer = (char *) 0xdeadbeef;
fs->fs_workbuffer = (char *) 0xdeadbeef;
}
/* Now removed ourselves from the linked list */
if (fs == g_mounthead)
{
/* We were the first ones. Set a new head */
g_mounthead = fs->fs_next;
}
else
{
/* Remove from the middle of the list somewhere */
prevfs->fs_next = fs->fs_next;
}
#else
if (fs->fs_blkdriver)
{
inode = fs->fs_blkdriver;
if (inode)
{
if (inode->u.i_bops && inode->u.i_bops->close)
{
inode->u.i_bops->close(inode);
}
}
}
/* Release the mountpoint private data */
kmm_free(fs->fs_rwbuffer);
kmm_free(fs->fs_workbuffer);
#endif
return ret;
}
/****************************************************************************
* Name: smartfs_finddirentry
*
* Description: Finds an entry in the filesystem as specified by relpath.
* If found, the direntry will be populated with information
* for accessing the entry.
*
* If the final directory segment of relpath just before the
* last segment (the target file/dir) is valid, then the
* parentdirsector will indicate the logical sector number of
* the parent directory where a new entry should be created,
* and the filename pointer will point to the final segment
* (i.e. the "filename").
*
****************************************************************************/
int smartfs_finddirentry(struct smartfs_mountpt_s *fs,
struct smartfs_entry_s *direntry, const char *relpath,
uint16_t *parentdirsector, const char **filename)
{
int ret = -ENOENT;
const char *segment;
const char *ptr;
uint16_t seglen;
uint16_t depth = 0;
uint16_t dirstack[CONFIG_SMARTFS_DIRDEPTH];
uint16_t dirsector;
uint16_t entrysize;
uint16_t offset;
struct smartfs_chain_header_s *header;
struct smart_read_write_s readwrite;
struct smartfs_entry_header_s *entry;
/* Set the initial value of the output */
*parentdirsector = 0xffff;
*filename = NULL;
/* Initialize directory level zero as the root sector */
dirstack[0] = fs->fs_rootsector;
entrysize = sizeof(struct smartfs_entry_header_s) +
fs->fs_llformat.namesize;
/* Test if this is a request for the root directory */
if (*relpath == '\0')
{
direntry->firstsector = fs->fs_rootsector;
direntry->flags = SMARTFS_DIRENT_TYPE_DIR | 0777;
direntry->utc = 0;
direntry->dsector = 0;
direntry->doffset = 0;
direntry->dfirst = fs->fs_rootsector;
direntry->name = NULL;
direntry->datlen = 0;
*parentdirsector = 0; /* Our parent is the format sector I guess */
return OK;
}
/* Parse through each segment of relpath */
segment = relpath;
while (segment != NULL && *segment != '\0')
{
/* Find the end of this segment. It will be '/' or NULL. */
ptr = segment;
seglen = 0;
while (*ptr != '/' && *ptr != '\0')
{
seglen++;
ptr++;
}
/* Check to avoid buffer overflow */
if (seglen >= WORKBUFFER_SIZE)
{
ret = -ENAMETOOLONG;
goto errout;
}
strncpy(fs->fs_workbuffer, segment, seglen);
fs->fs_workbuffer[seglen] = '\0';
/* Search for "." and ".." as segment names */
if (strcmp(fs->fs_workbuffer, ".") == 0)
{
/* Just ignore this segment. Advance ptr if not on NULL */
if (*ptr == '/')
{
ptr++;
}
segment = ptr;
continue;
}
else if (strcmp(fs->fs_workbuffer, "..") == 0)
{
/* Up one level */
if (depth == 0)
{
/* We went up one level past our mount point! */
goto errout;
}
/* "Pop" to the previous directory level */
depth--;
if (*ptr == '/')
{
ptr++;
}
segment = ptr;
continue;
}
else
{
/* Search for the entry in the current directory */
dirsector = dirstack[depth];
/* Read the directory */
offset = 0xffff;
#if CONFIG_SMARTFS_ERASEDSTATE == 0xff
while (dirsector != 0xffff)
#else
while (dirsector != 0)
#endif
{
/* Read the next directory in the chain */
readwrite.logsector = dirsector;
readwrite.count = fs->fs_llformat.availbytes;
readwrite.buffer = (uint8_t *)fs->fs_rwbuffer;
readwrite.offset = 0;
ret = FS_IOCTL(fs, BIOC_READSECT, (unsigned long) &readwrite);
if (ret < 0)
{
goto errout;
}
/* Point to next sector in chain */
header = (struct smartfs_chain_header_s *) fs->fs_rwbuffer;
dirsector = SMARTFS_NEXTSECTOR(header);
/* Search for the entry */
offset = sizeof(struct smartfs_chain_header_s);
entry = (struct smartfs_entry_header_s *)
&fs->fs_rwbuffer[offset];
while (offset < readwrite.count)
{
/* Test if this entry is valid and active */
#ifdef CONFIG_SMARTFS_ALIGNED_ACCESS
if (((smartfs_rdle16(&entry->flags) &
SMARTFS_DIRENT_EMPTY) ==
(SMARTFS_ERASEDSTATE_16BIT & SMARTFS_DIRENT_EMPTY)) ||
((smartfs_rdle16(&entry->flags)
& SMARTFS_DIRENT_ACTIVE) !=
(SMARTFS_ERASEDSTATE_16BIT & SMARTFS_DIRENT_ACTIVE)))
#else
if (((entry->flags & SMARTFS_DIRENT_EMPTY) ==
(SMARTFS_ERASEDSTATE_16BIT & SMARTFS_DIRENT_EMPTY)) ||
((entry->flags & SMARTFS_DIRENT_ACTIVE) !=
(SMARTFS_ERASEDSTATE_16BIT & SMARTFS_DIRENT_ACTIVE)))
#endif
{
/* This entry isn't valid, skip it */
offset += entrysize;
entry = (struct smartfs_entry_header_s *)
&fs->fs_rwbuffer[offset];
continue;
}
/* Test if the name matches */
if (strncmp(entry->name, fs->fs_workbuffer,
fs->fs_llformat.namesize) == 0)
{
/* We found it! If this is the last segment entry,
* then report the entry. If it isn't the last
* entry, then validate it is a directory entry and
* open it and continue searching.
*/
if (*ptr == '\0')
{
/* We are at the last segment. Report the entry */
/* Fill in the entry */
#ifdef CONFIG_SMARTFS_ALIGNED_ACCESS
direntry->firstsector =
smartfs_rdle16(&entry->firstsector);
direntry->flags = smartfs_rdle16(&entry->flags);
direntry->utc = smartfs_rdle32(&entry->utc);
#else
direntry->firstsector = entry->firstsector;
direntry->flags = entry->flags;
direntry->utc = entry->utc;
#endif
direntry->dsector = readwrite.logsector;
direntry->doffset = offset;
direntry->dfirst = dirstack[depth];
if (direntry->name == NULL)
{
direntry->name = (FAR char *)
kmm_malloc(fs->fs_llformat.namesize + 1);
}
memset(direntry->name, 0,
fs->fs_llformat.namesize + 1);
strncpy(direntry->name, entry->name,
fs->fs_llformat.namesize);
direntry->datlen = 0;
/* Scan the file's sectors to calculate the length
* and perform a rudimentary check.
*/
#ifdef CONFIG_SMARTFS_ALIGNED_ACCESS
if ((smartfs_rdle16(&entry->flags) &
SMARTFS_DIRENT_TYPE) ==
SMARTFS_DIRENT_TYPE_FILE)
{
dirsector =
smartfs_rdle16(&entry->firstsector);
#else
if ((entry->flags & SMARTFS_DIRENT_TYPE) ==
SMARTFS_DIRENT_TYPE_FILE)
{
dirsector = entry->firstsector;
#endif
readwrite.count =
sizeof(struct smartfs_chain_header_s);
readwrite.buffer = (uint8_t *)fs->fs_rwbuffer;
readwrite.offset = 0;
while (dirsector != SMARTFS_ERASEDSTATE_16BIT)
{
/* Read the next sector of the file */
readwrite.logsector = dirsector;
ret = FS_IOCTL(fs, BIOC_READSECT,
(unsigned long) &readwrite);
if (ret < 0)
{
ferr("ERROR: Error in sector"
" chain at %d!\n", dirsector);
break;
}
/* Add used bytes to the total and point to next sector */
if (*((FAR uint16_t *)header->used) !=
SMARTFS_ERASEDSTATE_16BIT)
{
direntry->datlen +=
*((uint16_t *)header->used);
}
dirsector = SMARTFS_NEXTSECTOR(header);
}
}
*parentdirsector = dirstack[depth];
*filename = segment;
ret = OK;
goto errout;
}
else
{
/* Validate it's a directory */
#ifdef CONFIG_SMARTFS_ALIGNED_ACCESS
if ((smartfs_rdle16(&entry->flags) &
SMARTFS_DIRENT_TYPE) !=
SMARTFS_DIRENT_TYPE_DIR)
#else
if ((entry->flags & SMARTFS_DIRENT_TYPE) !=
SMARTFS_DIRENT_TYPE_DIR)
#endif
{
/* Not a directory! Report the error */
ret = -ENOTDIR;
goto errout;
}
/* "Push" the directory and continue searching */
if (depth >= CONFIG_SMARTFS_DIRDEPTH - 1)
{
/* Directory depth too big */
ret = -ENAMETOOLONG;
goto errout;
}
#ifdef CONFIG_SMARTFS_ALIGNED_ACCESS
dirstack[++depth] =
smartfs_rdle16(&entry->firstsector);
#else
dirstack[++depth] = entry->firstsector;
#endif
segment = ptr + 1;
break;
}
}
/* Not this entry. Skip to the next one */
offset += entrysize;
entry = (struct smartfs_entry_header_s *)
&fs->fs_rwbuffer[offset];
}
/* Test if a directory entry was found and break if it was */
if (offset < readwrite.count)
{
break;
}
}
/* If we found a dir entry, then continue searching */
if (offset < readwrite.count)
{
/* Update the segment pointer */
if (*ptr != '\0')
{
ptr++;
}
segment = ptr;
continue;
}
/* Entry not found! Report the error. Also, if this is the last
* segment, then report the parent directory sector.
*/
if (*ptr == '\0')
{
*parentdirsector = dirstack[depth];
*filename = segment;
}
else
{
*parentdirsector = 0xffff;
*filename = NULL;
}
ret = -ENOENT;
goto errout;
}
}
errout:
return ret;
}
/****************************************************************************
* Name: smartfs_createentry
*
* Description: Creates a new entry in the specified parent directory, using
* the specified type and name. If the given sectorno is
* 0xffff, then a new sector is allocated for the new entry,
* otherwise the supplied sectorno is used.
*
****************************************************************************/
int smartfs_createentry(FAR struct smartfs_mountpt_s *fs,
uint16_t parentdirsector, FAR const char *filename,
uint16_t type, mode_t mode,
FAR struct smartfs_entry_s *direntry,
uint16_t sectorno, FAR struct smartfs_ofile_s *sf)
{
struct smart_read_write_s readwrite;
int ret;
uint16_t psector;
uint16_t nextsector;
uint16_t offset;
uint16_t found;
uint16_t entrysize;
struct smartfs_entry_header_s *entry;
struct smartfs_chain_header_s *chainheader;
int update_chain = 0;
struct smart_read_write_s update_readwrite;
struct smartfs_chain_header_s update_header;
/* Start at the 1st sector in the parent directory */
psector = parentdirsector;
found = FALSE;
entrysize = sizeof(struct smartfs_entry_header_s) +
fs->fs_llformat.namesize;
/* Validate the name isn't too long */
if (strlen(filename) > fs->fs_llformat.namesize)
{
return -ENAMETOOLONG;
}
/* Read the parent directory sector and find a place to insert
* the new entry.
*/
while (1)
{
/* Read the next sector */
readwrite.logsector = psector;
readwrite.count = fs->fs_llformat.availbytes;
readwrite.offset = 0;
readwrite.buffer = (uint8_t *) fs->fs_rwbuffer;
ret = FS_IOCTL(fs, BIOC_READSECT, (unsigned long) &readwrite);
if (ret < 0)
{
goto errout;
}
/* Get the next chained sector */
chainheader = (struct smartfs_chain_header_s *) fs->fs_rwbuffer;
nextsector = SMARTFS_NEXTSECTOR(chainheader);
/* Search for an empty entry in this sector */
offset = sizeof(struct smartfs_chain_header_s);
entry = (struct smartfs_entry_header_s *) &fs->fs_rwbuffer[offset];
while (offset + entrysize < readwrite.count)
{
/* Check if this entry is available */
#ifdef CONFIG_SMARTFS_ALIGNED_ACCESS
if ((smartfs_rdle16(&entry->flags) == SMARTFS_ERASEDSTATE_16BIT) ||
((smartfs_rdle16(&entry->flags) &
#else
if ((entry->flags == SMARTFS_ERASEDSTATE_16BIT) ||
((entry->flags &
#endif
(SMARTFS_DIRENT_EMPTY | SMARTFS_DIRENT_ACTIVE)) ==
(~SMARTFS_ERASEDSTATE_16BIT &
(SMARTFS_DIRENT_EMPTY | SMARTFS_DIRENT_ACTIVE))))
{
/* We found an empty entry. Use it. */
found = TRUE;
break;
}
/* Not available. Skip to next entry */
offset += entrysize;
entry = (struct smartfs_entry_header_s *) &fs->fs_rwbuffer[offset];
}
/* If we found an entry, stop the search */
if (found)
{
break;
}
/* If there are no more sectors, then we need to add one to make
* room for the new entry.
*/
if (nextsector == SMARTFS_ERASEDSTATE_16BIT)
{
/* Allocate a new sector and chain it to the last one */
ret = FS_IOCTL(fs, BIOC_ALLOCSECT, 0xffff);
if (ret < 0)
{
goto errout;
}
nextsector = (uint16_t) ret;
/* Chain the next sector into this sector. */
*((uint16_t *)update_header.nextsector) = nextsector;
update_readwrite.logsector = psector;
update_readwrite.offset = offsetof(struct smartfs_chain_header_s,
nextsector);
update_readwrite.count = sizeof(uint16_t);
update_readwrite.buffer = update_header.nextsector;
update_chain = 1;
}
/* Now update to the next sector */
psector = nextsector;
}
/* We found an insertion point. Create the entry at sector,offset */
#if CONFIG_SMARTFS_ERASEDSTATE == 0xff
#ifdef CONFIG_SMARTFS_ALIGNED_ACCESS
smartfs_wrle16(&entry->flags, (uint16_t) (SMARTFS_DIRENT_ACTIVE |
SMARTFS_DIRENT_DELETING | SMARTFS_DIRENT_RESERVED | type | (mode &
SMARTFS_DIRENT_MODE)));
#else
entry->flags = (uint16_t) (SMARTFS_DIRENT_ACTIVE |
SMARTFS_DIRENT_DELETING | SMARTFS_DIRENT_RESERVED | type | (mode &
SMARTFS_DIRENT_MODE));
#endif
#else /* CONFIG_SMARTFS_ERASEDSTATE == 0xff */
#ifdef CONFIG_SMARTFS_ALIGNED_ACCESS
smartfs_wrle16(&entry->flags, (uint16_t) (SMARTFS_DIRENT_EMPTY | type |
(mode & SMARTFS_DIRENT_MODE)));
#else
entry->flags = (uint16_t) (SMARTFS_DIRENT_EMPTY | type |
(mode & SMARTFS_DIRENT_MODE));
#endif
#endif /* CONFIG_SMARTFS_ERASEDSTATE == 0xff */
if (sectorno == 0xffff)
{
/* Allocate a new sector for the file / dir */
ret = FS_IOCTL(fs, BIOC_ALLOCSECT, 0xffff);
if (ret < 0)
{
goto errout;
}
nextsector = (uint16_t) ret;
/* Set the newly allocated sector's type (file or dir) */
#ifdef CONFIG_SMARTFS_USE_SECTOR_BUFFER
if (sf)
{
/* Using sector buffer and we have an open file context.
* Just update the sector buffer in the open file context.
*/
memset(sf->buffer, CONFIG_SMARTFS_ERASEDSTATE,
fs->fs_llformat.availbytes);
chainheader = (struct smartfs_chain_header_s *) sf->buffer;
chainheader->type = SMARTFS_SECTOR_TYPE_FILE;
sf->bflags = SMARTFS_BFLAG_DIRTY | SMARTFS_BFLAG_NEWALLOC;
}
else
#endif
{
if ((type & SMARTFS_DIRENT_TYPE) == SMARTFS_DIRENT_TYPE_DIR)
{
chainheader->type = SMARTFS_SECTOR_TYPE_DIR;
}
else
{
chainheader->type = SMARTFS_SECTOR_TYPE_FILE;
}
readwrite.count = 1;
readwrite.offset = offsetof(struct smartfs_chain_header_s, type);
readwrite.buffer = (uint8_t *) &chainheader->type;
readwrite.logsector = nextsector;
ret = FS_IOCTL(fs, BIOC_WRITESECT, (unsigned long) &readwrite);
if (ret < 0)
{
ferr("ERROR: Error %d setting new sector type for sector %d\n",
ret, nextsector);
goto errout;
}
}
}
else
{
/* Use the provided sector number */
nextsector = sectorno;
}
/* Create the directory entry to be written in the parent's sector */
#ifdef CONFIG_SMARTFS_ALIGNED_ACCESS
smartfs_wrle16(&entry->firstsector, nextsector);
smartfs_wrle16(&entry->utc, time(NULL));
#else
entry->firstsector = nextsector;
entry->utc = time(NULL);
#endif
memset(entry->name, 0, fs->fs_llformat.namesize);
strncpy(entry->name, filename, fs->fs_llformat.namesize);
/* Now write the new entry to the parent directory sector */
readwrite.logsector = psector;
readwrite.offset = offset;
readwrite.count = entrysize;
readwrite.buffer = (uint8_t *) &fs->fs_rwbuffer[offset];
ret = FS_IOCTL(fs, BIOC_WRITESECT, (unsigned long) &readwrite);
if (ret < 0)
{
goto errout;
}
if (update_chain)
{
/* Update chain header after the next sector was written */
ret = FS_IOCTL(fs, BIOC_WRITESECT, (unsigned long) &update_readwrite);
if (ret < 0)
{
ferr("ERROR: Error chaining sector %d\n",
update_readwrite.logsector);
goto errout;
}
}
/* Now fill in the entry */
direntry->firstsector = nextsector;
direntry->dsector = psector;
direntry->doffset = offset;
#ifdef CONFIG_SMARTFS_ALIGNED_ACCESS
direntry->flags = smartfs_rdle16(&entry->flags);
direntry->utc = smartfs_rdle32(&entry->utc);
#else
direntry->flags = entry->flags;
direntry->utc = entry->utc;
#endif
direntry->datlen = 0;
if (direntry->name == NULL)
{
direntry->name = (FAR char *)kmm_malloc(fs->fs_llformat.namesize + 1);
}
memset(direntry->name, 0, fs->fs_llformat.namesize + 1);
strncpy(direntry->name, filename, fs->fs_llformat.namesize);
ret = OK;
errout:
return ret;
}
/****************************************************************************
* Name: smartfs_deleteentry
*
* Description: Deletes an entry from the filesystem (file or dir) by
* freeing all the entry's sectors and then marking it inactive
* in it's parent's directory sector. For a directory, it
* does not validate the directory is empty, nor does it do
* a recursive delete.
*
****************************************************************************/
int smartfs_deleteentry(struct smartfs_mountpt_s *fs,
struct smartfs_entry_s *entry)
{
int ret;
uint16_t nextsector;
uint16_t sector;
uint16_t count;
uint16_t entrysize;
uint16_t offset;
struct smartfs_entry_header_s *direntry;
struct smartfs_chain_header_s *header;
struct smart_read_write_s readwrite;
/* Okay, delete the file. Loop through each sector and release them
*
* TODO: We really should walk the list backward to avoid lost
* sectors in the event we lose power. However this requires
* allocating a buffer to build the sector list since we don't
* store a doubly-linked list of sectors on the device. We
* could test if the sector data buffer is big enough and
* just use that, and only allocate a new buffer if the
* sector buffer isn't big enough. Do do this, however, we
* need to change the code below as it is using the a few
* bytes of the buffer to read in header info.
*/
nextsector = entry->firstsector;
header = (struct smartfs_chain_header_s *) fs->fs_rwbuffer;
readwrite.offset = 0;
readwrite.count = sizeof(struct smartfs_chain_header_s);
readwrite.buffer = (uint8_t *) fs->fs_rwbuffer;
while (nextsector != SMARTFS_ERASEDSTATE_16BIT)
{
/* Read the next sector into our buffer */
sector = nextsector;
readwrite.logsector = sector;
ret = FS_IOCTL(fs, BIOC_READSECT, (unsigned long) &readwrite);
if (ret < 0)
{
ferr("ERROR: Error reading sector %d\n", nextsector);
break;
}
/* Release this sector */
nextsector = SMARTFS_NEXTSECTOR(header);
ret = FS_IOCTL(fs, BIOC_FREESECT, sector);
}
/* Remove the entry from the directory tree */
readwrite.logsector = entry->dsector;
readwrite.offset = 0;
readwrite.count = fs->fs_llformat.availbytes;
readwrite.buffer = (uint8_t *) fs->fs_rwbuffer;
ret = FS_IOCTL(fs, BIOC_READSECT, (unsigned long) &readwrite);
if (ret < 0)
{
ferr("ERROR: Error reading directory info at sector %d\n",
entry->dsector);
goto errout;
}
/* Mark this entry as inactive */
direntry = (struct smartfs_entry_header_s *)
&fs->fs_rwbuffer[entry->doffset];
#if CONFIG_SMARTFS_ERASEDSTATE == 0xff
#ifdef CONFIG_SMARTFS_ALIGNED_ACCESS
smartfs_wrle16(&direntry->flags,
smartfs_rdle16(&direntry->flags) & ~SMARTFS_DIRENT_ACTIVE);
#else
direntry->flags &= ~SMARTFS_DIRENT_ACTIVE;
#endif
#else /* CONFIG_SMARTFS_ERASEDSTATE == 0xff */
#ifdef CONFIG_SMARTFS_ALIGNED_ACCESS
smartfs_wrle16(&direntry->flags,
smartfs_rdle16(&direntry->flags) | SMARTFS_DIRENT_ACTIVE);
#else
direntry->flags |= SMARTFS_DIRENT_ACTIVE;
#endif
#endif /* CONFIG_SMARTFS_ERASEDSTATE == 0xff */
/* Write the updated flags back to the sector */
readwrite.offset = entry->doffset;
readwrite.count = sizeof(uint16_t);
readwrite.buffer = (uint8_t *) &direntry->flags;
ret = FS_IOCTL(fs, BIOC_WRITESECT, (unsigned long) &readwrite);
if (ret < 0)
{
ferr("ERROR: Error marking entry inactive at sector %d\n",
entry->dsector);
goto errout;
}
/* Test if any entries in this sector are being used */
if ((entry->dsector != fs->fs_rootsector) &&
(entry->dsector != entry->dfirst))
{
/* Scan the sector and count used entries */
count = 0;
offset = sizeof(struct smartfs_chain_header_s);
entrysize = sizeof(struct smartfs_entry_header_s) +
fs->fs_llformat.namesize;
while (offset + entrysize < fs->fs_llformat.availbytes)
{
/* Test the next entry */
direntry = (struct smartfs_entry_header_s *)
&fs->fs_rwbuffer[offset];
#ifdef CONFIG_SMARTFS_ALIGNED_ACCESS
if (((smartfs_rdle16(&direntry->flags) & SMARTFS_DIRENT_EMPTY) !=
(SMARTFS_ERASEDSTATE_16BIT & SMARTFS_DIRENT_EMPTY)) &&
((smartfs_rdle16(&direntry->flags) & SMARTFS_DIRENT_ACTIVE) ==
(SMARTFS_ERASEDSTATE_16BIT & SMARTFS_DIRENT_ACTIVE)))
#else
if (((direntry->flags & SMARTFS_DIRENT_EMPTY) !=
(SMARTFS_ERASEDSTATE_16BIT & SMARTFS_DIRENT_EMPTY)) &&
((direntry->flags & SMARTFS_DIRENT_ACTIVE) ==
(SMARTFS_ERASEDSTATE_16BIT & SMARTFS_DIRENT_ACTIVE)))
#endif
{
/* Count this entry */
count++;
}
/* Advance to next entry */
offset += entrysize;
}
/* Test if the count it zero. If it is, then we will release the sector */
if (count == 0)
{
/* Okay, to release the sector, we must find the sector that we
* are chained to and remove ourselves from the chain. First
* save our nextsector value so we can "unchain" ourselves.
*/
nextsector = SMARTFS_NEXTSECTOR(header);
/* Now loop through the dir sectors to find ourselves in the chain */
sector = entry->dfirst;
readwrite.offset = 0;
readwrite.count = sizeof(struct smartfs_chain_header_s);
readwrite.buffer = (uint8_t *) fs->fs_rwbuffer;
while (sector != SMARTFS_ERASEDSTATE_16BIT)
{
/* Read the header for the next sector */
readwrite.logsector = sector;
ret = FS_IOCTL(fs, BIOC_READSECT, (unsigned long) &readwrite);
if (ret < 0)
{
ferr("ERROR: Error reading sector %d\n", nextsector);
break;
}
/* Test if this sector "points" to us */
if (SMARTFS_NEXTSECTOR(header) == entry->dsector)
{
/* We found ourselves in the chain. Update the chain. */
SMARTFS_NEXTSECTOR(header) = nextsector;
readwrite.offset = offsetof(struct smartfs_chain_header_s,
nextsector);
readwrite.count = sizeof(uint16_t);
readwrite.buffer = header->nextsector;
ret = FS_IOCTL(fs, BIOC_WRITESECT,
(unsigned long)&readwrite);
if (ret < 0)
{
ferr("ERROR: Error unchaining sector (%d)\n",
nextsector);
goto errout;
}
/* Now release our sector */
ret = FS_IOCTL(fs, BIOC_FREESECT,
(unsigned long)entry->dsector);
if (ret < 0)
{
ferr("ERROR: Error freeing sector %d\n",
entry->dsector);
goto errout;
}
/* Break out of the loop, we are done! */
break;
}
/* Chain to the next sector */
sector = SMARTFS_NEXTSECTOR(header);
}
}
}
ret = OK;
errout:
return ret;
}
/****************************************************************************
* Name: smartfs_countdirentries
*
* Description: Counts the number of items in the specified directory entry.
* This routine assumes you have validated the entry you are
* passing is in fact a directory sector, though it checks
* just in case you were stupid :-)
*
****************************************************************************/
int smartfs_countdirentries(struct smartfs_mountpt_s *fs,
struct smartfs_entry_s *entry)
{
int ret;
uint16_t nextsector;
uint16_t offset;
uint16_t entrysize;
int count;
struct smartfs_entry_header_s *direntry;
struct smartfs_chain_header_s *header;
struct smart_read_write_s readwrite;
/* Walk through the directory's sectors and count entries */
count = 0;
nextsector = entry->firstsector;
while (nextsector != SMARTFS_ERASEDSTATE_16BIT)
{
/* Read the next sector into our buffer */
readwrite.logsector = nextsector;
readwrite.offset = 0;
readwrite.count = fs->fs_llformat.availbytes;
readwrite.buffer = (uint8_t *) fs->fs_rwbuffer;
ret = FS_IOCTL(fs, BIOC_READSECT, (unsigned long) &readwrite);
if (ret < 0)
{
ferr("ERROR: Error reading sector %d\n", nextsector);
break;
}
/* Validate this is a directory type sector */
header = (struct smartfs_chain_header_s *) fs->fs_rwbuffer;
if (header->type != SMARTFS_SECTOR_TYPE_DIR)
{
ferr("ERROR: Sector %d is not a DIR sector!\n", nextsector);
goto errout;
}
/* Loop for all entries in this sector and count them */
offset = sizeof(struct smartfs_chain_header_s);
entrysize = sizeof(struct smartfs_entry_header_s) +
fs->fs_llformat.namesize;
direntry = (struct smartfs_entry_header_s *) &fs->fs_rwbuffer[offset];
while (offset + entrysize < readwrite.count)
{
#ifdef CONFIG_SMARTFS_ALIGNED_ACCESS
if (((smartfs_rdle16(&direntry->flags) & SMARTFS_DIRENT_EMPTY) !=
(SMARTFS_ERASEDSTATE_16BIT & SMARTFS_DIRENT_EMPTY)) &&
((smartfs_rdle16(&direntry->flags) & SMARTFS_DIRENT_ACTIVE) ==
(SMARTFS_ERASEDSTATE_16BIT & SMARTFS_DIRENT_ACTIVE)))
#else
if (((direntry->flags & SMARTFS_DIRENT_EMPTY) !=
(SMARTFS_ERASEDSTATE_16BIT & SMARTFS_DIRENT_EMPTY)) &&
((direntry->flags & SMARTFS_DIRENT_ACTIVE) ==
(SMARTFS_ERASEDSTATE_16BIT & SMARTFS_DIRENT_ACTIVE)))
#endif
{
/* Count this entry */
count++;
}
offset += entrysize;
direntry = (struct smartfs_entry_header_s *)
&fs->fs_rwbuffer[offset];
}
/* Get the next sector from the header */
nextsector = SMARTFS_NEXTSECTOR(header);
}
ret = count;
errout:
return ret;
}
/****************************************************************************
* Name: smartfs_sync_internal
*
* Description:
* Synchronize the file state on disk to match internal, in-memory state.
*
****************************************************************************/
int smartfs_sync_internal(FAR struct smartfs_mountpt_s *fs,
FAR struct smartfs_ofile_s *sf)
{
FAR struct smartfs_chain_header_s *header;
struct smart_read_write_s readwrite;
int ret = OK;
#ifdef CONFIG_SMARTFS_USE_SECTOR_BUFFER
if (sf->bflags & SMARTFS_BFLAG_DIRTY)
{
/* Update the header with the number of bytes written */
header = (struct smartfs_chain_header_s *)sf->buffer;
if (*((uint16_t *)header->used) == SMARTFS_ERASEDSTATE_16BIT)
{
*((uint16_t *)header->used) = sf->byteswritten;
}
else
{
*((uint16_t *)header->used) += sf->byteswritten;
}
/* Write the entire sector to FLASH */
readwrite.logsector = sf->currsector;
readwrite.offset = 0;
readwrite.count = fs->fs_llformat.availbytes;
readwrite.buffer = sf->buffer;
ret = FS_IOCTL(fs, BIOC_WRITESECT, (unsigned long) &readwrite);
if (ret < 0)
{
ferr("ERROR: Error %d writing used bytes for sector %d\n",
ret, sf->currsector);
goto errout;
}
sf->byteswritten = 0;
sf->bflags = 0;
}
#else /* CONFIG_SMARTFS_USE_SECTOR_BUFFER */
/* Test if we have written bytes to the current sector that
* need to be recorded in the chain header's used bytes field.
*/
if (sf->byteswritten > 0)
{
finfo("Syncing sector %d\n", sf->currsector);
/* Read the existing sector used bytes value */
readwrite.logsector = sf->currsector;
readwrite.offset = 0;
readwrite.buffer = (uint8_t *) fs->fs_rwbuffer;
readwrite.count = sizeof(struct smartfs_chain_header_s);
ret = FS_IOCTL(fs, BIOC_READSECT, (unsigned long) &readwrite);
if (ret < 0)
{
ferr("ERROR: Error %d reading sector %d data\n",
ret, sf->currsector);
goto errout;
}
/* Add new byteswritten to existing value */
header = (struct smartfs_chain_header_s *) fs->fs_rwbuffer;
if (*((uint16_t *) header->used) == SMARTFS_ERASEDSTATE_16BIT)
{
*((uint16_t *) header->used) = sf->byteswritten;
}
else
{
*((uint16_t *) header->used) += sf->byteswritten;
}
readwrite.offset = offsetof(struct smartfs_chain_header_s, used);
readwrite.count = sizeof(uint16_t);
readwrite.buffer = (uint8_t *) &fs->fs_rwbuffer[readwrite.offset];
ret = FS_IOCTL(fs, BIOC_WRITESECT, (unsigned long) &readwrite);
if (ret < 0)
{
ferr("ERROR: Error %d writing used bytes for sector %d\n",
ret, sf->currsector);
goto errout;
}
sf->byteswritten = 0;
}
#endif /* CONFIG_SMARTFS_USE_SECTOR_BUFFER */
errout:
return ret;
}
/****************************************************************************
* Name: smartfs_seek_internal
*
* Description:
* Performs the logic of the seek function. This is an internal function
* because it does not provide semaphore protection and therefore must be
* called from one of the other public interface routines (open, seek,
* etc.).
*
****************************************************************************/
off_t smartfs_seek_internal(FAR struct smartfs_mountpt_s *fs,
FAR struct smartfs_ofile_s *sf,
off_t offset, int whence)
{
FAR struct smartfs_chain_header_s *header;
struct smart_read_write_s readwrite;
off_t newpos;
off_t sectorstartpos;
int ret;
/* Test if this is a seek to get the current file pos */
if ((whence == SEEK_CUR) && (offset == 0))
{
return sf->filepos;
}
/* Test if we need to sync the file */
if (sf->byteswritten > 0)
{
/* Perform a sync */
smartfs_sync_internal(fs, sf);
}
/* Calculate the file position to seek to based on current position */
switch (whence)
{
case SEEK_SET:
default:
newpos = offset;
break;
case SEEK_CUR:
newpos = sf->filepos + offset;
break;
case SEEK_END:
newpos = sf->entry.datlen + offset;
break;
}
/* Ensure newpos is in range */
if (newpos < 0)
{
newpos = 0;
}
if (newpos > sf->entry.datlen)
{
newpos = sf->entry.datlen;
}
/* Now perform the seek. Test if we are seeking within the current
* sector and can skip the search to save time.
*/
sectorstartpos = sf->filepos - (sf->curroffset - sizeof(struct
smartfs_chain_header_s));
if (newpos >= sectorstartpos && newpos < sectorstartpos +
fs->fs_llformat.availbytes - sizeof(struct smartfs_chain_header_s))
{
/* Seeking within the current sector. Just update the offset */
sf->curroffset = sizeof(struct smartfs_chain_header_s) +
newpos - sectorstartpos;
sf->filepos = newpos;
return newpos;
}
/* Nope, we have to search for the sector and offset. If the new pos is
* greater than the current pos, then we can start from the beginning of
* the current sector, otherwise we have to start from the beginning of
* the file.
*/
if (newpos > sf->filepos)
{
sf->filepos = sectorstartpos;
}
else
{
sf->currsector = sf->entry.firstsector;
sf->filepos = 0;
}
header = (struct smartfs_chain_header_s *) fs->fs_rwbuffer;
while ((sf->currsector != SMARTFS_ERASEDSTATE_16BIT) &&
(sf->filepos + fs->fs_llformat.availbytes -
sizeof(struct smartfs_chain_header_s) < newpos))
{
/* Read the sector's header */
readwrite.logsector = sf->currsector;
readwrite.offset = 0;
readwrite.count = sizeof(struct smartfs_chain_header_s);
readwrite.buffer = (uint8_t *) fs->fs_rwbuffer;
ret = FS_IOCTL(fs, BIOC_READSECT, (unsigned long) &readwrite);
if (ret < 0)
{
ferr("ERROR: Error %d reading sector %d header\n",
ret, sf->currsector);
goto errout;
}
/* Point to next sector and update filepos */
sf->currsector = SMARTFS_NEXTSECTOR(header);
sf->filepos += SMARTFS_USED(header);
}
#ifdef CONFIG_SMARTFS_USE_SECTOR_BUFFER
/* When using sector buffering, we must read in the last buffer to our
* sf->buffer in case any changes are made.
*/
if (sf->currsector != SMARTFS_ERASEDSTATE_16BIT)
{
readwrite.logsector = sf->currsector;
readwrite.offset = 0;
readwrite.count = fs->fs_llformat.availbytes;
readwrite.buffer = (uint8_t *) sf->buffer;
ret = FS_IOCTL(fs, BIOC_READSECT, (unsigned long) &readwrite);
if (ret < 0)
{
ferr("ERROR: Error %d reading sector %d header\n",
ret, sf->currsector);
goto errout;
}
}
#endif
/* Now calculate the offset */
sf->curroffset = sizeof(struct smartfs_chain_header_s) + newpos -
sf->filepos;
sf->filepos = newpos;
return newpos;
errout:
return ret;
}
/****************************************************************************
* Name: smartfs_shrinkfile
*
* Description:
* Shrink the size of an existing file to the specified length
*
****************************************************************************/
int smartfs_shrinkfile(FAR struct smartfs_mountpt_s *fs,
FAR struct smartfs_ofile_s *sf, off_t length)
{
FAR struct smartfs_chain_header_s *header;
FAR struct smartfs_entry_s *entry;
FAR uint8_t *dest;
struct smart_read_write_s readwrite;
uint16_t nextsector;
uint16_t sector;
off_t remaining;
off_t destsize;
off_t available;
off_t offset;
int ret;
/* Walk through the directory's sectors and count entries */
entry = &sf->entry;
nextsector = entry->firstsector;
header = (struct smartfs_chain_header_s *)fs->fs_rwbuffer;
remaining = length;
available = fs->fs_llformat.availbytes -
sizeof(struct smartfs_chain_header_s);
while (nextsector != SMARTFS_ERASEDSTATE_16BIT)
{
/* Read the next sector into our buffer */
readwrite.logsector = nextsector;
readwrite.offset = 0;
readwrite.count = fs->fs_llformat.availbytes;
readwrite.buffer = (FAR uint8_t *)fs->fs_rwbuffer;
ret = FS_IOCTL(fs, BIOC_READSECT, (unsigned long) &readwrite);
if (ret < 0)
{
ferr("ERROR: Error reading sector %d header\n", nextsector);
return ret;
}
/* Get the next chained sector */
sector = SMARTFS_NEXTSECTOR(header);
#ifdef CONFIG_SMARTFS_USE_SECTOR_BUFFER
/* When we have a sector buffer in use, simply skip the first sector.
* It will be handled below.
*/
if (nextsector == entry->firstsector)
{
if (remaining > available)
{
remaining -= available;
}
else
{
remaining = 0;
}
}
else
#endif
/* Are we retaining the sector it its entirety? */
if (remaining >= available)
{
/* Yes... skip to the next sector */
remaining -= available;
}
/* Are we removing the sector it its entirety? */
else if (remaining <= 0 && nextsector != entry->firstsector)
{
/* Yes.. just release the sector */
ret = FS_IOCTL(fs, BIOC_FREESECT, (unsigned long)nextsector);
if (ret < 0)
{
ferr("ERROR: Error freeing sector %d\n", nextsector);
return ret;
}
}
else
{
/* No.. Fill our buffer with erased data, retaining any still-
* valid bytes at the beginning of the buffer.
*
* Because of the preceding tests we know that
* 0 <= remaining < available. A special case is remaining == 0
* and nextsector == firstsector. In that case, we need to
* overwrite the sector data with the erased state value. The
* underlying SMART block driver will detect this and release the
* old sector and create a new one with the new (blank) data.
*
* Otherwise, we need to preserve the header and overwrite some of
* the data.
*/
if (remaining == 0)
{
dest = (FAR uint8_t *)fs->fs_rwbuffer;
destsize = fs->fs_llformat.availbytes;
}
else
{
offset = sizeof(struct smartfs_chain_header_s) + remaining;
dest = (FAR uint8_t *)&fs->fs_rwbuffer[offset];
destsize = fs->fs_llformat.availbytes - offset;
*((uint16_t *)header->used) = remaining;
*((uint16_t *)header->nextsector) = SMARTFS_ERASEDSTATE_16BIT;
remaining = 0;
}
memset(dest, CONFIG_SMARTFS_ERASEDSTATE, destsize);
header->type = SMARTFS_SECTOR_TYPE_FILE;
/* Now write the new sector data */
readwrite.count = fs->fs_llformat.availbytes;
ret = FS_IOCTL(fs, BIOC_WRITESECT, (unsigned long)&readwrite);
if (ret < 0)
{
ferr("ERROR: Error blanking 1st sector (%d) of file\n",
nextsector);
return ret;
}
}
/* Now move on to the next sector */
nextsector = sector;
}
#ifdef CONFIG_SMARTFS_USE_SECTOR_BUFFER
/* Now deal with the first sector in the event we are using a sector buffer
* like we would be if CRC is enabled.
*
* Using sector buffer and we have an open file context. Just update the
* sector buffer in the open file context.
*/
if (length < fs->fs_llformat.availbytes)
{
/* Read the entire sector */
readwrite.logsector = entry->firstsector;
readwrite.offset = 0;
readwrite.count = fs->fs_llformat.availbytes;
readwrite.buffer = (uint8_t *)sf->buffer;
ret = FS_IOCTL(fs, BIOC_READSECT, (unsigned long)&readwrite);
if (ret < 0)
{
return ret;
}
/* Retain any valid data at the beginning of the sector, including the
* header. Special case length == 0
*/
if (length == 0)
{
dest = (FAR uint8_t *)sf->buffer;
destsize = fs->fs_llformat.availbytes;
}
else
{
offset = sizeof(struct smartfs_chain_header_s) + length;
dest = (FAR uint8_t *)&sf->buffer[offset];
destsize = fs->fs_llformat.availbytes - offset;
header = (struct smartfs_chain_header_s *)sf->buffer;
*((uint16_t *)header->used) = length;
*((uint16_t *)header->nextsector) = SMARTFS_ERASEDSTATE_16BIT;
}
memset(dest, CONFIG_SMARTFS_ERASEDSTATE, destsize);
header = (struct smartfs_chain_header_s *)sf->buffer;
header->type = SMARTFS_SECTOR_TYPE_FILE;
sf->bflags = SMARTFS_BFLAG_DIRTY;
}
#endif
entry->datlen = length;
return OK;
}
/****************************************************************************
* Name: smartfs_extendfile
*
* Description:
* Zero-extend the length of a regular file to 'length'.
*
****************************************************************************/
int smartfs_extendfile(FAR struct smartfs_mountpt_s *fs,
FAR struct smartfs_ofile_s *sf, off_t length)
{
struct smart_read_write_s readwrite;
FAR struct smartfs_chain_header_s *header;
#ifndef CONFIG_SMARTFS_USE_SECTOR_BUFFER
FAR uint8_t *buffer;
#endif
off_t remaining;
off_t savepos;
off_t oldsize;
int ret;
/* We are zero-extending the file. This is essentially the same as a
* write except that (1) we write zeros and (2) we don't update the file
* position.
*/
#ifndef CONFIG_SMARTFS_USE_SECTOR_BUFFER
/* In order to perform the writes we will have to have a sector buffer. If
* SmartFS is not configured with a sector buffer then we will, then we
* will, unfortunately, need to allocate one.
*/
buffer = (FAR uint8_t *)kmm_malloc(SMARTFS_TRUNCBUFFER_SIZE);
if (buffer == NULL)
{
return -ENOMEM;
}
#endif
/* Loop until either (1) the file has been fully extended with zeroed data
* or (2) an error occurs. We assume we start with the current sector in
* cache (ff_currentsector).
*/
oldsize = sf->entry.datlen;
remaining = length - oldsize;
DEBUGASSERT(length > oldsize);
/* Seek to the end of the file for the append/write operation, remembering
* the current file position. It will be restored before returning; the
* truncate operation must not alter the file position.
*/
savepos = sf->filepos;
smartfs_seek_internal(fs, sf, 0, SEEK_END);
while (remaining > 0)
{
/* We will fill up the current sector. Write data to the current
* sector first.
*/
#ifdef CONFIG_SMARTFS_USE_SECTOR_BUFFER
readwrite.count = fs->fs_llformat.availbytes - sf->curroffset;
if (readwrite.count > remaining)
{
readwrite.count = remaining;
}
memset(&sf->buffer[sf->curroffset], 0, readwrite.count);
sf->bflags |= SMARTFS_BFLAG_DIRTY;
#else /* CONFIG_SMARTFS_USE_SECTOR_BUFFER */
readwrite.offset = sf->curroffset;
readwrite.logsector = sf->currsector;
readwrite.buffer = buffer;
/* Select max size that available in the current sector */
readwrite.count = fs->fs_llformat.availbytes - sf->curroffset;
if (readwrite.count > remaining)
{
/* Limit the write to the size for our smaller working buffer */
readwrite.count = SMARTFS_TRUNCBUFFER_SIZE;
}
if (readwrite.count > remaining)
{
/* Further limit the write to the remaining bytes to write */
readwrite.count = remaining;
}
/* Perform the write */
if (readwrite.count > 0)
{
ret = FS_IOCTL(fs, BIOC_WRITESECT, (unsigned long) &readwrite);
if (ret < 0)
{
ferr("ERROR: Error %d writing sector %d data\n",
ret, sf->currsector);
goto errout_with_buffer;
}
}
#endif /* CONFIG_SMARTFS_USE_SECTOR_BUFFER */
/* Update our control variables */
sf->entry.datlen += readwrite.count;
sf->byteswritten += readwrite.count;
sf->curroffset += readwrite.count;
remaining -= readwrite.count;
/* Test if we wrote a full sector of data */
#ifdef CONFIG_SMARTFS_USE_SECTOR_BUFFER
if (sf->curroffset == fs->fs_llformat.availbytes && remaining)
{
/* First get a new chained sector */
ret = FS_IOCTL(fs, BIOC_ALLOCSECT, 0xffff);
if (ret < 0)
{
ferr("ERROR: Error %d allocating new sector\n", ret);
goto errout_with_buffer;
}
/* Copy the new sector to the old one and chain it */
header = (struct smartfs_chain_header_s *) sf->buffer;
*((uint16_t *)header->nextsector) = (uint16_t)ret;
/* Now sync the file to write this sector out */
ret = smartfs_sync_internal(fs, sf);
if (ret != OK)
{
goto errout_with_buffer;
}
/* Record the new sector in our tracking variables and reset the
* offset to "zero".
*/
if (sf->currsector == SMARTFS_NEXTSECTOR(header))
{
/* Error allocating logical sector! */
ferr("ERROR: Duplicate logical sector %d\n", sf->currsector);
}
sf->bflags = SMARTFS_BFLAG_DIRTY;
sf->currsector = SMARTFS_NEXTSECTOR(header);
sf->curroffset = sizeof(struct smartfs_chain_header_s);
memset(sf->buffer, CONFIG_SMARTFS_ERASEDSTATE,
fs->fs_llformat.availbytes);
header->type = SMARTFS_DIRENT_TYPE_FILE;
}
#else /* CONFIG_SMARTFS_USE_SECTOR_BUFFER */
if (sf->curroffset == fs->fs_llformat.availbytes)
{
/* Sync the file to write this sector out */
ret = smartfs_sync_internal(fs, sf);
if (ret != OK)
{
goto errout_with_buffer;
}
/* Allocate a new sector if needed */
if (remaining > 0)
{
/* Allocate a new sector */
ret = FS_IOCTL(fs, BIOC_ALLOCSECT, 0xffff);
if (ret < 0)
{
ferr("ERROR: Error %d allocating new sector\n", ret);
goto errout_with_buffer;
}
/* Copy the new sector to the old one and chain it */
header = (struct smartfs_chain_header_s *)fs->fs_rwbuffer;
*((FAR uint16_t *)header->nextsector) = (uint16_t)ret;
readwrite.offset = offsetof(struct smartfs_chain_header_s,
nextsector);
readwrite.buffer = (FAR uint8_t *)header->nextsector;
readwrite.count = sizeof(uint16_t);
ret = FS_IOCTL(fs, BIOC_WRITESECT, (unsigned long) &readwrite);
if (ret < 0)
{
ferr("ERROR: Error %d writing next sector\n", ret);
goto errout_with_buffer;
}
/* Record the new sector in our tracking variables and
* reset the offset to "zero".
*/
if (sf->currsector == SMARTFS_NEXTSECTOR(header))
{
/* Error allocating logical sector! */
ferr("ERROR: Duplicate logical sector %d\n",
sf->currsector);
}
sf->currsector = SMARTFS_NEXTSECTOR(header);
sf->curroffset = sizeof(struct smartfs_chain_header_s);
}
}
#endif /* CONFIG_SMARTFS_USE_SECTOR_BUFFER */
}
/* The file was successfully extended with zeros */
ret = OK;
errout_with_buffer:
#ifndef CONFIG_SMARTFS_USE_SECTOR_BUFFER
/* Release the allocated buffer */
kmm_free(buffer);
#endif
/* Restore the original file position */
smartfs_seek_internal(fs, sf, savepos, SEEK_SET);
return ret;
}
/****************************************************************************
* Name: smartfs_get_first_mount
*
* Description: Returns a pointer to the first mounted smartfs volume.
*
****************************************************************************/
#if defined(CONFIG_FS_PROCFS) && !defined(CONFIG_FS_PROCFS_EXCLUDE_SMARTFS)
FAR struct smartfs_mountpt_s *smartfs_get_first_mount(void)
{
return g_mounthead;
}
#endif