bug#61386: [PATCH] cp,mv,install: Disable sparse copy on macOS

[Pádraig Brady]

On 13/02/2023 17:16, George Valkov wrote:
> > On 2023-02-13, at 4:05 PM, Pádraig Brady <P@draigBrady.com> wrote:
> > This amendment seems to make the treatment of umask and setuid consistent
> > between using fclonefileat() and not.
> >
> > diff --git a/src/copy.c b/src/copy.c
> > index 782fab98d..8370f55bd 100644
> > --- a/src/copy.c
> > +++ b/src/copy.c
> > @@ -1280,10 +1280,11 @@ copy_reg (char const *src_name, char const *dst_name,
> >                        return_val = false;
> >                      }
> >                  }
> > -              mode_already_preserved = (fc_flags & CLONE_ACL) != 0;
> > +              mode_already_preserved = (fc_flags & CLONE_ACL) != 0
> > +                                       && cloned_mode == src_mode;
> >                dest_desc = -1;
> >                omitted_permissions = dst_mode & ~cloned_mode;
> > -              extra_permissions = 0;
> > +              extra_permissions = dst_mode & ~ cached_umask ();
> >                goto set_dest_mode;
> >              }
> >            else if (! handle_clone_fail (dst_dirfd, dst_relname, src_name,
> >
> > With this, all tests pass (apart from thru-dangling.sh as mentioned above
> > which I'll skip in another patch).
> > Paul I'm happy to merge this amendment, mention the improvement in NEWS,
> > and push in your name if you're OK with it.
>
> To summarise:
> 1. We apply my original patch to disable sparse-copy on macOS. Otherwise since 
> fclonefileat is not used whenever we overwrite a file, corruption will still 
> occur. An updated copy of the patch is attached below:
> By doing this we fail the following test so you have to disable it on macOS:
> FAIL: tests/cp/sparse-perf.sh

Right. I'll adjust your patch to also get
tests/seek-data-capable to return false on Darwin,
which will then skip that test.

> 2. Next we add support for fclonefileat using cp: improve use of fclonefileat
> A few more tests fail.
>
> 3. We fix those tests using the patch from your last mail. Then we have to 
> disable the following test on macOS:
> FAIL: tests/cp/thru-dangling.sh
>
> Final results: copying sparse-files on macOS no longer produces corrupted 
> copies. Performance is greatly optimised by cloning instead of copying, where 
> possible. The following two tests now fail on macOS and will be disabled:
> FAIL: tests/cp/sparse-perf.sh

handled above

> FAIL: tests/cp/thru-dangling.sh

I'll fix this

> Looks good to me. Yes, please add me to the commit history. Also feel free to 
> add my web site if appropriate. Finally please let me know when the patches are 
> pushed, so that I can update OpenWRT to use the latest version of coreutils.
>
> Pádraig and Paul, thank you very much for your kind help! It was nice working 
> with you and learning new experience. Cheers, mates!

Summary sounds good.

Thanks for all the testing.

Pádraig