[U-Boot,V2] fs: fat: Fix mkcksum() function parameters

The mkcksum() function now takes one parameter, the pointer to
11-byte wide character array, which it then operates on.

Currently, the function is wrongly passed (dir_entry)->name, which
is only 8-byte wide character array. Though by further inspecting
the dir_entry structure, it can be noticed that the name[8] entry
is immediatelly followed by ext[3] entry. Thus, name[8] and ext[3]
in the dir_entry structure actually work as this 11-byte wide array
since they're placed right next to each other by current compiler
behavior.

Depending on this is obviously wrong, thus fix this by correctly
passing both (dir_entry)->name and (dir_entry)->ext to the mkcksum()
function and adjust the function appropriately.

Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Tom Rini <trini@ti.com>
---
 fs/fat/fat.c       |   20 ++++++++++++--------
 fs/fat/fat_write.c |    2 +-
 2 files changed, 13 insertions(+), 9 deletions(-)

V2: Fix fat write by adding braces

On Tue, Oct 09, 2012 at 07:20:22AM -0000, Marek Vasut wrote:

> The mkcksum() function now takes one parameter, the pointer to
> 11-byte wide character array, which it then operates on.
> 
> Currently, the function is wrongly passed (dir_entry)->name, which
> is only 8-byte wide character array. Though by further inspecting
> the dir_entry structure, it can be noticed that the name[8] entry
> is immediatelly followed by ext[3] entry. Thus, name[8] and ext[3]
> in the dir_entry structure actually work as this 11-byte wide array
> since they're placed right next to each other by current compiler
> behavior.
> 
> Depending on this is obviously wrong, thus fix this by correctly
> passing both (dir_entry)->name and (dir_entry)->ext to the mkcksum()
> function and adjust the function appropriately.
> 
> Signed-off-by: Marek Vasut <marex@denx.de>
> Cc: Tom Rini <trini@ti.com>

Applied to u-boot/master, thanks!

Hi Marek,

This patch is broken. It breaks detecting duplicate filenames. The 
problem is that you are using sizeof(name) and sizeof(ext). This does 
not work since this just reports the size of the pointer in gcc 4.7, 
which according to our compiler guy is the correct behavior. Instead of 
using sizeof, just use 8 and 3 respectively. Otherwise it is doing a 
checksum of 4 and 4 characters.

-Aaron

On 10/09/2012 10:20 AM, Marek Vasut wrote:
> The mkcksum() function now takes one parameter, the pointer to
> 11-byte wide character array, which it then operates on.
>
> Currently, the function is wrongly passed (dir_entry)->name, which
> is only 8-byte wide character array. Though by further inspecting
> the dir_entry structure, it can be noticed that the name[8] entry
> is immediatelly followed by ext[3] entry. Thus, name[8] and ext[3]
> in the dir_entry structure actually work as this 11-byte wide array
> since they're placed right next to each other by current compiler
> behavior.
>
> Depending on this is obviously wrong, thus fix this by correctly
> passing both (dir_entry)->name and (dir_entry)->ext to the mkcksum()
> function and adjust the function appropriately.
>
> Signed-off-by: Marek Vasut <marex@denx.de>
> Cc: Tom Rini <trini@ti.com>
> ---
>   fs/fat/fat.c       |   20 ++++++++++++--------
>   fs/fat/fat_write.c |    2 +-
>   2 files changed, 13 insertions(+), 9 deletions(-)
>
> V2: Fix fat write by adding braces
>
> diff --git a/fs/fat/fat.c b/fs/fat/fat.c
> index 41ae15e..8870292 100644
> --- a/fs/fat/fat.c
> +++ b/fs/fat/fat.c
> @@ -567,15 +567,16 @@ get_vfatname(fsdata *mydata, int curclust, __u8 *cluster,
>   }
>   
>   /* Calculate short name checksum */
> -static __u8 mkcksum(const char *str)
> +static __u8 mkcksum(const char name[8], const char ext[3])
>   {
>   	int i;
>   
>   	__u8 ret = 0;
>   
> -	for (i = 0; i < 11; i++) {
> -		ret = (((ret & 1) << 7) | ((ret & 0xfe) >> 1)) + str[i];
> -	}
> +	for (i = 0; i < sizeof(name); i++)
> +		ret = (((ret & 1) << 7) | ((ret & 0xfe) >> 1)) + name[i];
> +	for (i = 0; i < sizeof(ext); i++)
> +		ret = (((ret & 1) << 7) | ((ret & 0xfe) >> 1)) + ext[i];
>   
>   	return ret;
>   }
> @@ -678,7 +679,8 @@ static dir_entry *get_dentfromdir(fsdata *mydata, int startsect,
>   				return NULL;
>   			}
>   #ifdef CONFIG_SUPPORT_VFAT
> -			if (dols && mkcksum(dentptr->name) == prevcksum) {
> +			__u8 csum = mkcksum(dentptr->name, dentptr->ext);
> +			if (dols && csum == prevcksum) {
>   				prevcksum = 0xffff;
>   				dentptr++;
>   				continue;
> @@ -946,13 +948,16 @@ do_fat_read_at(const char *filename, unsigned long pos, void *buffer,
>   
>   		for (i = 0; i < DIRENTSPERBLOCK; i++) {
>   			char s_name[14], l_name[VFAT_MAXLEN_BYTES];
> +			__u8 csum;
>   
>   			l_name[0] = '\0';
>   			if (dentptr->name[0] == DELETED_FLAG) {
>   				dentptr++;
>   				continue;
>   			}
> -			if ((dentptr->attr & ATTR_VOLUME)) {
> +
> +			csum = mkcksum(dentptr->name, dentptr->ext);
> +			if (dentptr->attr & ATTR_VOLUME) {
>   #ifdef CONFIG_SUPPORT_VFAT
>   				if ((dentptr->attr & ATTR_VFAT) == ATTR_VFAT &&
>   				    (dentptr->name[0] & LAST_LONG_ENTRY_MASK)) {
> @@ -1015,8 +1020,7 @@ do_fat_read_at(const char *filename, unsigned long pos, void *buffer,
>   				goto exit;
>   			}
>   #ifdef CONFIG_SUPPORT_VFAT
> -			else if (dols == LS_ROOT &&
> -				 mkcksum(dentptr->name) == prevcksum) {
> +			else if (dols == LS_ROOT && csum == prevcksum) {
>   				prevcksum = 0xffff;
>   				dentptr++;
>   				continue;
> diff --git a/fs/fat/fat_write.c b/fs/fat/fat_write.c
> index 5829adf..4a1bda0 100644
> --- a/fs/fat/fat_write.c
> +++ b/fs/fat/fat_write.c
> @@ -335,7 +335,7 @@ fill_dir_slot(fsdata *mydata, dir_entry **dentptr, const char *l_name)
>   
>   	/* Get short file name and checksum value */
>   	strncpy(s_name, (*dentptr)->name, 16);
> -	checksum = mkcksum(s_name);
> +	checksum = mkcksum((*dentptr)->name, (*dentptr)->ext);
>   
>   	do {
>   		memset(slotptr, 0x00, sizeof(dir_slot));

Dear Aaron Williams,

> Hi Marek,
> 
> This patch is broken. It breaks detecting duplicate filenames. The
> problem is that you are using sizeof(name) and sizeof(ext). This does
> not work since this just reports the size of the pointer in gcc 4.7,
> which according to our compiler guy is the correct behavior. Instead of
> using sizeof, just use 8 and 3 respectively. Otherwise it is doing a
> checksum of 4 and 4 characters.

You're right. Can you post a patch for that please? This was already applied. 
Also, do you see any symptoms?

Best regards,
Marek Vasut