Re: coreutils-5.92:du not AIX largefile safe --

[Jim Meyering]

Paul Townsend <address@hidden> wrote:
> When the `du' included with coreutils-5.92 is compiled on an AIX 5.1 box
> that has a 32 bit kernel and is compiled with large file support, it
> incorrectly computes the size of a file that exceeds 2**31-1 bytes.  If
> I've interpretted the code correctly, it's because the stat->st_blocks
> entry is a signed int that's multiplied by 512 before being promoted to
> an unsigned long long (uintmax_t).  The patch below forces the promotion
> before the multiplication.
>
> --  Paul Townsend (alpha a beta present at purdue dot edu)
>
> diff -u coreutils-5.92/src/du.c.orig coreutils-5.92/src/du.c
> --- coreutils-5.92/src/du.c.orig      2005-10-15 04:13:21.000000000 -0500
> +++ coreutils-5.92/src/du.c   2005-11-02 02:44:21.767391000 -0500
> @@ -537,7 +537,7 @@
>        duinfo_set (&dui,
>                 (apparent_size
>                  ? sb->st_size
> -                : ST_NBLOCKS (*sb) * ST_NBLOCKSIZE),
> +                : (uintmax_t) ST_NBLOCKS (*sb) * ST_NBLOCKSIZE),
>                 (time_type == time_mtime ? get_stat_mtime (sb)
>                  : time_type == time_atime ? get_stat_atime (sb)
>                  : get_stat_ctime (sb)));

Thanks for the report and patch.
That looks right to me, too.
Would you please test it with something like this:

printf %$(echo '2^31-1'|bc)s x > k
du -s k
printf x >> k
du -s k
printf x >> k
du -s k

I get this output:

2097152 k
2097152 k
2097156 k

Of course, your numbers may vary a little.
The important thing is that there be no errors.