[U-Boot] [PATCH 1/1] fat: FAT sector offsets overflow on large disks and/or FAT partitions

[Aaron Williams]

Hi Anatolij,

On 05/12/2012 08:41 AM, Anatolij Gustschin wrote:
> Hello,
>
> On Wed, 02 May 2012 19:17:41 -0700
> Aaron Williams <Aaron.Williams at cavium.com> wrote:
>
>> This patch fixes several issues where sector offsets can overflow due to
>> being limited to 16-bits.  There are many cases which can cause an
>> overflow, including large FAT32 partitions and partitions that start at
>> a sufficiently large offset on the storage device.
> For large FAT32 partitions only changing of fatlength, rootdir_sect and
> data_begin is needed to avoid overflows. Changing of fat_sect shouldn't
> be needed.
>
> What do you mean exactly by "partitions starting at a sufficiently large
> offset on the storage device"? How do you create such partition?
> I've tested with a 210 GB FAT32 partition as the fourth primary partition
> on a 2 TB disk. This partition is the last partition on the disk, so its
> offset is sufficiently large. For this test only fatlength, rootdir_sect
> and data_begin was changed to __u32 and int and I do not see issues when
> listing or loading the files from this partition.
You are correct about fat_sect, I misread the code and changed that one
as well. I don't think it really matters since another changing it back
to a __u16 won't save any space in the data structure on most platforms.

It looks like the corruption problems I was seeing were due to the fact
that the file I was reading was after a number of very large files in a
64GB FAT32 partition and I ran into overflows. The partition location
shouldn't have any impact unless we're talking about 2+TB drives, in
which case I expect there will be numerous other things that will break.

>> Numerous issues were observed and fixed when a 64GB FAT32 filesystem was
>> accessed due to truncation.
>>
>> Signed-off-by: Aaron Williams <aaron.williams at caviumnetworks.com>
>> ---
>>  include/fat.h |   10 +++++-----
>>  1 files changed, 5 insertions(+), 5 deletions(-)
>>
>> diff --git a/include/fat.h b/include/fat.h
>> index 4c92442..7215628 100644
>> --- a/include/fat.h
>> +++ b/include/fat.h
>> @@ -178,12 +178,12 @@ typedef struct dir_slot {
>>  typedef struct {
>>         __u8    *fatbuf;        /* Current FAT buffer */
>>         int     fatsize;        /* Size of FAT in bits */
>> -       __u16   fatlength;      /* Length of FAT in sectors */
>> -       __u16   fat_sect;       /* Starting sector of the FAT */
>> -       __u16   rootdir_sect;   /* Start sector of root directory */
>> -       __u16   sect_size;      /* Size of sectors in bytes */
>> +       __u32   fat_sect;       /* Starting sector of the FAT */
>> +       __u32   rootdir_sect;   /* Start sector of root directory */
>> +       __u32   fatlength;      /* Length of FAT in sectors */
>>         __u16   clust_size;     /* Size of clusters in sectors */
>> -       short   data_begin;     /* The sector of the first cluster, can be negative */
>> +       __u16   sect_size;      /* Size of sectors in bytes */
>> +       int     data_begin;     /* The sector of the first cluster, can be negative */
>>         int     fatbufnum;      /* Used by get_fatent, init to -1 */
>>  } fsdata;
> The patch is probably corrupted by your mailer, it doesn't apply.
>
> Thanks,
> Anatolij
>
Unfortunately our outbound Exchange mail server tends to "fix" mail
formatting making patch submission rather difficult.
I'll see what I can do to try and bypass it.

-Aaron

-- 
Aaron Williams
Software Engineer
Cavium, Inc.
(408) 943-7198  (510) 789-8988 (cell)