From mboxrd@z Thu Jan  1 00:00:00 1970
From: =?utf-8?Q?Beno=C3=AEt_Th=C3=A9baudeau?= <benoit.thebaudeau@advansee.com>
Date: Mon, 22 Oct 2012 01:25:30 +0200 (CEST)
Subject: [U-Boot] fat: Handle moved FAT32 root directory
In-Reply-To: <10395409.7297877.1350859974867.JavaMail.root@advansee.com>
Message-ID: <848164565.7298149.1350861930354.JavaMail.root@advansee.com>
List-Id: <u-boot.lists.denx.de>
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit
To: u-boot@lists.denx.de

Dear Romain Izard,

On Thu, 09 Jun 2011 21:36:08 -0000, Romain Izard wrote:
> The default location of the root directory in a FAT32 partition is the
> same as its location in a FAT12/16 partition. But a difference is that
> in FAT32, it is possible for the root directory to move.
> 
> This fix ensures that the parsing uses the root directory location
> stored in the PBR, instead of always using the default location.
> 
> Signed-off-by: Romain Izard <romain.izard.pro@gmail.com>
> 
> ---
> fs/fat/fat.c |    2 ++
>  1 files changed, 2 insertions(+), 0 deletions(-)
> 
> diff --git a/fs/fat/fat.c b/fs/fat/fat.c
> index c450bf6..22d34d3 100644
> --- a/fs/fat/fat.c
> +++ b/fs/fat/fat.c
> @@ -798,6 +798,8 @@ do_fat_read (const char *filename, void *buffer, unsigned long maxsize,
>  	if (mydata->fatsize == 32) {
>  		mydata->data_begin = mydata->rootdir_sect -
>  					(mydata->clust_size * 2);
> +		cursect = mydata->data_begin +
> +				(root_cluster * mydata->clust_size);
>  	} else {
>  		rootdir_size = ((bs.dir_entries[1]  * (int)256 +
>  				 bs.dir_entries[0]) *
> 

Good find!

However, I don't think that hacking cursect is an appropriate fix.
mydata->rootdir_sect should be fixed instead in the first place because:
 - it's cleaner,
 - mydata->rootdir_sect is used in a debug trace after that,
 - mydata->rootdir_sect is used by get_cluster(), and .. entries must give a 0
   start cluster for the root directory, even on FAT32.

By the way, while reviewing the FAT code, I noticed that it must have the
following bugs:
 - .. will be an issue in some way, especially if they reach the root folder on
   FAT12 or FAT16,
 - VFAT long file names will be broken if their entries cross the
   PREFETCH_BLOCKS barrier on FAT12/16,
 - do_fat_write() is also broken by the bug that you want to fix.

Also, do_fat_write() duplicates a lot of init code from do_fat_read_at(). A
common function could be created from that to init a mydata (fsdata) structure
from read_bootsectandvi().

So your patch could be changed to the following series:
1. Factorize do_fat_read_at() and do_fat_write() init (no change of behavior).
2. Fix this init by fixing mydata->rootdir_sect.

The other issues that I mentioned are unrelated to your patch and should be
verified and addressed separately.

Best regards,
Beno?t