On 5/24/22 06:37, Linus Torvalds wrote:
> On Sun, May 22, 2022 at 10:01 PM Damien Le Moal
> <damien.lemoal@xxxxxxxxxxxxxxxxxx> wrote:
>>
>> Note: I made a mistake during this PR preparation and inadvertantly deleted the
>> for-5.19 branch used for this PR. I recreated it and prepared the PR using this
>> newly pushed for-5.19 branch. All patches have been in linux-next for a while.
>> I hope this does not trigger any problem on your end.
>
> Grr.
>
> That seems to be the cause of repeated commits, which in turn then
> caused conflicts because you had further changes.
>
> IOW, I already had gotten
>
> zonefs: Fix management of open zones
> zonefs: Clear inode information flags on inode creation
>
> from the block tree (your commits), and your branch now had different
> copies of those commits.
>
> And you don't actually list those commits, which makes me think that
> you then did some manual editing of the pull request.
Yes I did. I really messed up. My apologies about that.
>
> The duplicate commits with identical contents then caused commit
> 87c9ce3ffec9 ("zonefs: Add active seq file accounting") to show as a
> conflict, because the different branches did *some* things the same,
> but then that commit added other changes.
>
> And honestly, I think that commit is buggy.
>
> In particular, notice how the locking changes in that commit means
> that zonefs_init_file_inode() now always returns 0, even if
> zonefs_zone_mgmt() failed with an error.
>
> The error cause it to skip "zonefs_account_active()", but then it
> returns success anyway.
>
> Was that really intentional?
Absolutely not. That is definitely a bug. Will send a fix for that.
Thank you for catching this.
>
> I've merged this - with that apparent bug and all - because maybe it
> *was* intentional. But please double-check and confirm that you really
> intended zonefs_init_file_inode() to always return success, unlike the
> old behavior.
>
> Linus
--
Damien Le Moal
Western Digital Research
