On Tue, 2023-03-28 at 16:19 -0400, Paul Moore wrote: > On Tue, Mar 28, 2023 at 3:47 AM Roberto Sassu > <roberto.sassu@xxxxxxxxxxxxxxx> wrote: > > On Mon, 2023-03-27 at 17:02 -0400, Paul Moore wrote: > > > On Mon, Mar 27, 2023 at 3:30 AM Roberto Sassu > > > <roberto.sassu@xxxxxxxxxxxxxxx> wrote: > > > > On Fri, 2023-03-24 at 17:39 -0400, Paul Moore wrote: > > > > > On Fri, Mar 24, 2023 at 9:26 AM Roberto Sassu > > > > > <roberto.sassu@xxxxxxxxxxxxxxx> wrote: > > > > > > On Fri, 2023-03-24 at 11:18 +0100, Roberto Sassu wrote: > > > > > > > On Thu, 2023-03-23 at 20:09 -0400, Paul Moore wrote: > > > > > > > > On Tue, Mar 14, 2023 at 4:19 AM Roberto Sassu > > > > > > > > <roberto.sassu@xxxxxxxxxxxxxxx> wrote: > > > > > > > > > From: Roberto Sassu <roberto.sassu@xxxxxxxxxx> > > ... > > > > Okay, that's fair, but we could still pass the full xattrs array and a > > > reference to the current count which could be both read and updated by > > > the individual LSMs, right? > > > > Yes, we could do. > > > > > The issue is that the separate compaction stage is not something we > > > want to have to do if we can avoid it. Maybe we're stuck with it, but > > > I'm not yet convinced that we can't make some minor changes to the > > > LSMs to avoid the compaction step. > > > > I liked more the idea that LSMs do what they are most familiar with, > > get an offset in a security blob or, in this case, a starting slot in > > the new_xattrs array, and write there. > > > > v3 had the lsm_find_xattr_slot() helper, to get the starting slot, but > > somehow I find it less intuitive. > > > > Ok, if you prefer to avoid the compaction stage, I will rewrite this > > patch. > > My concern is having to look through the xattr array after each LSM > has been run and in at least one case having to then do a memcpy() to > keep the array packed. There are some cases where there is no way to > avoid all that extra work, but here I think we have the LSMs do the > Right Thing with respect to packing the xattr array without overly > burdening the individual LSMs. > > Does that make sense? It basically comes down to being smart about > our abstractions and both selectively, and carefully, breaking them > when there is a reasonable performance gain to be had. Yes, ok, it is a good approach. Thanks Roberto
