On Apr 23, 2008 18:48 -0500, Eric Sandeen wrote:
> Andreas Dilger wrote:
> > On Apr 23, 2008 15:39 -0400, Josef Bacik wrote:
> >> + /*
> >> + * we want the comparisons to be unsigned, in case somebody passes -1,
> >> + * meaning they want they want the entire file, but the result has to be
> >> + * signed so we can handle the case where we get more blocks than the
> >> + * size of the file
> >> + */
> >> + length = (long)min((unsigned long)fiemap_s->fm_length,
> >> + (unsigned long)i_size_read(inode));
> >
> > This might be written as:
> >
> > length = (long)min_t(unsigned long,fiemap_s->fm_len,i_size_read(inode));
> >
> > Also, what about files that have blocks mapped after i_size?
>
> That'll be tough for ext3, though I guess for a generic interface it
> could happen, so I guess it needs to be handled.
Right, because some filesystems may preallocate blocks beyond i_size to
avoid fragmentation.
> Maybe check i_blocks
> against i_size, see if i_blocks indicates blocks past EOF? Hm, I guess
> that's not going to work in general; you could be completely sparse up
> to an EOF at 100G and have 100M of blocks past that...
...and there are also indirect blocks, and EA blocks that are not counted
toward i_size. The issue is that getblock() doesn't have any way of
reporting that it is beyond EOF. If it was an ext2/ext3-specific mechanism
then it could check in the i_block[] array and in the end of the
{t,d,}indirect blocks to know conclusively whether there are any blocks
beyond EOF.
That said, I don't think the generic interface can know everything about
each filesystem. My suggestion was that blocks beyond i_size continue
to be mapped until a hole (block == 0) is returned. It isn't perfect,
but would likely cover 99.9% of the cases where some small number of blocks
(<= 64kB or whatever) were allocated beyond EOF to avoid fragmentation.
Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
