[PATCH] ubifs: Fix a potential space leak problem while linking tmpfile

[Zhihao Cheng]

在 2020/7/7 20:09, Richard Weinberger 写道:
> ----- Ursprüngliche Mail -----
>> Perhaps I misunderstood what commit 32fe905c17f001 ("ubifs: Fix
>> O_TMPFILE corner case in ubifs_link()") wanted to fix.
>> I think orphan area is used to remind filesystem don't forget to delete
>> inodes (whose nlink is 0) in next unclean rebooting. Generally, the file
>> system is not corrupted caused by replaying orphan nodes.
>> Ralph reported a filesystem corruption in combination with overlayfs.
>> Can you tell me the details about that problem? Thanks.
> On my test bed I didn't see a fs corruption, what I saw was a failing orphan
> self test while playing with O_TMPFILE and linkat().
Do we have a reproducer, or can I get the fail testcase? Is it a xfstest 
case?
>
> When you create a tmpfile it has a link count of 0 and an orphan is
> installed. Such that the tmpfile is gone after a reboot but you can
> still use it prior to that.
> By using linkat() you can raise the link counter to 1 again.
> Thus, the orphan needs to be removed.
> This is pattern overlayfs uses a lot.
>
> Since UBIFS never supported raising the link counter from 0 to 1
> we have many corner cases and fixing all these turned out into a nightmare.
> ...as you can see from the amount broken patches from me :-(.
>
> Thanks,
> //richard
>
> .