On 12/27/19 4:05 PM, Ming Lei wrote:
> Some filesystem, such as vfat, may send bio which crosses device boundary,
> and the worse thing is that the IO request starting within device boundaries
> can contain more than one segment past EOD.
>
> Commit dce30ca9e3b6 ("fs: fix guard_bio_eod to check for real EOD errors")
> tries to fix this issue by returning -EIO for this situation. However,
> this way lets fs user code lose chance to handle -EIO, then sync_inodes_sb()
> may hang for ever.
>
> Also the current truncating on last segment is dangerous by updating the
> last bvec, given bvec table becomes not immutable any more, and fs bio
> users may not retrieve the truncated pages via bio_for_each_segment_all() in
> its .end_io callback.
>
> Fixes this issue by supporting multi-segment truncating. And the
> approach is simpler:
>
> - just update bio size since block layer can make correct bvec with
> the updated bio size. Then bvec table becomes really immutable.
>
> - zero all truncated segments for read bio
Applied, thanks.
--
Jens Axboe
