Re: stat64 for over 2TB file returned invalid st_blocks

From: Takashi Sato
Date: Fri Dec 02 2005 - 08:17:41 EST


Hi,

> Hi all,
>
> I found a problem at stat64 on 32bit architecture.
>
> When I called stat64 for a file which is larger than 2TB, stat64
> returned an invalid number of blocks at st_blocks on 32bit
> architecture, although it returned a valid number of blocks on 64bit
> architecture(ia64).

For jfs, it's a bigger problem than just stat64. When writing the inode
to disk, jfs calculates the number of blocks from the 32-bit value:
dip->di_nblocks = cpu_to_le64(PBLK2LBLK(ip->i_sb, ip->i_blocks))

So it won't only report the wrong number of blocks, but it will actually
store the wrong number. :-(

I also found another problem on generic quota code. In
dquot_transfer(), the file usage is calculated from i_blocks via
inode_get_bytes(). If the file is over 2TB, the change of usage is
less than expected.

To solve this problem, I think inode.i_blocks should be 8 byte.

2. Change the type of architecture dependent stat64.st_blocks in
include/asm/asm-*/stat.h from unsigned long to unsigned long long.
I tried modifying only stat64 of 32bit architecture
(include/asm-i386/stat.h).

This changes the API, but the structure does suggest that the 4-byte pad
should be used for the high-order bytes of st_blocks, so that's not
really a problem. A correct fix would replace __pad4 with
st_blocks_high (or something like that) and ensure that the high-order
word was stored there. Your proposed fix would only be correct on
little-endian hardware, as Jörn pointed out.

Thank you for your advice. I'll research for glibc and consider
how to implement.
By the way I think, as Avi Kivity said, it's always little-endian on i386,
is it correct?

-- Takashi Sato

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/