On Thu, Mar 18, 2021 at 09:34:14PM +0100, David Sterba wrote:
> On Thu, Mar 18, 2021 at 02:22:20AM +0000, Sidong Yang wrote:
> > On Wed, Mar 17, 2021 at 07:36:47PM +0100, David Sterba wrote:
> > > On Tue, Mar 16, 2021 at 01:27:46PM +0000, Sidong Yang wrote:
> > > > When user assign qgroup with qgroup id that is too big to exceeds
> > > > range and invade level value, and it works without any error. but
> > > > this action would be make undefined error. this code make sure that
> > > > qgroup id doesn't exceed range(0 ~ 2^48-1).
> > > 
> > > Should the level be also validate? The function parse_qgroupid does not
> > > do full validation, so eg 0//0 would be parsed as a path and not as a
> > > typo, level larger than 64K will be silently clamped.
> > 
> > I agree. 0//0 would be parsed as path but it failed in
> > btrfs_util_is_subvolume() and goes to err. I understand that upper 16
> > bits of qgroupid is for level. so, The valid llevel range is [0~2^16-1].
> > But I can't get it that level larger than 64K will be clampled. 
> 
> The way the level gets stored into the final qgroup id is level << 48.
> For example invalid values 70000/281474976710779 would be stored
> as subvol id 123 and level 4465, where the u64 is 0x117100000000007b
> 
> > one more question about that, I see that the ioctl calls just store the
> > qgroupid without any opeartion with level. is the level meaningless in
> > kernel?
> 
> The quota groups are hierarchical and the level denotes the level, where
> 0/subvolid is the lowest always attached to a subvolume and the higher
> levels are artificial and may contain any qgroups and do the whole
> accounting on the subtree. The original design doc .pdf can be found in
> https://git.kernel.org/pub/scm/linux/kernel/git/arne/qgroups-doc.git/tree/

Thanks for detailed description.
I'll write the new patch for checking level also.

Thanks,
Sidong

Reply via email to