On Monday, May 13th, 2024 at 5:34 AM, Brian Foster <[email protected]> wrote:
> On Sat, May 11, 2024 at 12:20:12AM +0000, Reed Riley wrote: > > > By removing the early-exit when REMAP_FILE_DEDUP is set, we should be > > able to support the fideduperange ioctl, albeit less efficiently than if > > we handled some of the extent locking and comparison logic inside > > bcachefs. Extent comparison logic already exists inside of > > `__generic_remap_file_range_prep`. > > > > Signed-off-by: Reed Riley [email protected] > > --- > > > Seems reasonable: > > Reviewed-by: Brian Foster [email protected] > > > Have you run any tests just to make sure there are no surprises? If not, > it looks like xfs_io has a 'dedupe' command that would make it easy to > run a quick test or two from the command line. fstests has a bunch of > tests in the dedupe group (which I presume this patch should now allow > to run on bcachefs) as well. I worked with Kent to run his CI tests against the patch (https://evilpiepirate.org/~testdashboard/ci?branch=bcachefs-fideduperange&commit=1945149c8d7549b924cd88f57f0cd938b3bb7125) and also used xfs_io to do some basic sanity checks. Specifically, I sanity checked that: 1. fideduperange doesn’t dedupe if file content doesn’t match, 2. fideduperange does dedupe stuff when they do (according to filefrag -v reporting shared extents), and 3. That neither of the above operations changed file checksums. I’d be happy to run more tests if anyone can suggest them? > > Brian > > > fs/bcachefs/fs-io.c | 3 --- > > 1 file changed, 3 deletions(-) > > > > diff --git a/fs/bcachefs/fs-io.c b/fs/bcachefs/fs-io.c > > index 20b40477425f..4f513f22a66a 100644 > > --- a/fs/bcachefs/fs-io.c > > +++ b/fs/bcachefs/fs-io.c > > @@ -857,9 +857,6 @@ loff_t bch2_remap_file_range(struct file *file_src, > > loff_t pos_src, > > if (remap_flags & ~(REMAP_FILE_DEDUP|REMAP_FILE_ADVISORY)) > > return -EINVAL; > > > > - if (remap_flags & REMAP_FILE_DEDUP) > > - return -EOPNOTSUPP; > > - > > if ((pos_src & (block_bytes(c) - 1)) || > > (pos_dst & (block_bytes(c) - 1))) > > return -EINVAL; > > -- > > 2.44.0
