Re: [Qemu-devel] virtfs/9p duplicate inodes

[Greg Kurz]

Hi Christian,

Sorry for the late response. I'm quite busy on other topics these days...

On Sun, 21 Apr 2019 00:41:01 +0200
Christian Schoenebeck <address@hidden> wrote:

> On Samstag, 30. März 2019 21:01:28 CEST Christian Schoenebeck wrote:
> > On Samstag, 30. März 2019 17:47:51 CET Greg Kurz wrote:  
> > > Maybe have a look at this tentative to fix QID collisions:
> > > 
> > > https://lists.gnu.org/archive/html/qemu-devel/2018-02/msg02283.html  
> [snip]
> > Question: so far I just had a look at that patch set, but haven't tried it
> > yet. Am I correct that the inode numbers (of the same file) would actually
> > change on guest side with every reboot (i.e. depending on the precise
> > sequence individual files would be accessed by guest after each reboot)?  
> 

I have not checked.

> I intended to extend Antonios' patch set regarding 9p QID collisions with the 
> goal to make the ids constant beyond reboots by storing the qpp_table as fs 
> xattr.
> 

Hmm... why would you do that ? Even if some filesystems do have persistant
inode numbers, it isn't mandatory AFAIK.

> My plan was to load the qpp_table in v9fs_device_realize_common() and save 
> the 
> table only once in v9fs_device_unrealize_common(), instead of storing the 
> table on every new insertion. The problem though is that none of the 9p 
> unrealize functions is called on guest shutdowns.
> 

The unrealize function is called when the device is unplugged from
it's parent bus. It isn't related to guest shutdown.

> Is there any callback that is guaranteed to be called on guest shutdowns?
> 

There's no such thing.

> Best regards,
> Christian Schoenebeck

Cheers,

--
Greg