[Linux-cachefs] Kernel 2.6.30-rc2

[Edward Konetzko]

David Howells wrote:
> David Howells <dhowells at redhat.com> wrote:
>
>   
>> I'm slightly perplexed that I have one retrieval that persists in failing to
>> retrieve anything.  It is, however, something I've reduced to a read of a
>> single file, so I should be able to debug that.
>>     
>
> Okay, it's not a bug...  at least not in my code.  The problem is that my test
> isn't just a tar, it's a cat and a tar, and when doing the cat, nfs_readpage()
> is asked to read one page beyond the end of the file (which is odd), and oddly
> enough, FS-Cache never has any data for that extra page.
>
> David
>
> --
> Linux-cachefs mailing list
> Linux-cachefs at redhat.com
> https://www.redhat.com/mailman/listinfo/linux-cachefs
>   
David

I was going to say I have found away to reproduce.  I boot the 
2.6.29-rc3 kernel and start my copy of the kernel source code tree, I do 
this a few times to know cache is populated and working.  I then boot 
into the 2.6.30-rc2 kernel and start my copy again.  Everything works as 
expected until I run "find linux-2.6-nfs-fscache -exec touch {} \;" and 
as expected all files are invalidated and the cache starts to repopulate 
this continues on forever. I was going to suggest files getting created 
in the cache wrong.    If I then reboot back into 2.6.29-rc3, restart 
copy, everything is copied over to populate the cache and after that it 
works as expected.  I think this is the same behavior other people are 
seeing.


So I guess the question is now where does the bug lie?  Also what 
information would be helpful for me to provide to you.  Looking at my 
graphs every file is just invalidated by the cache since nothing ever 
matches, I believe this is the behavior that you saw.

Thanks
Edward