On Thu, Jun 3, 2021 at 5:20 PM Jeff Layton <jlayton@xxxxxxxxxx> wrote:
>
> On Thu, 2021-06-03 at 17:19 +0200, Ilya Dryomov wrote:
> > On Thu, Jun 3, 2021 at 4:42 PM Jeff Layton <jlayton@xxxxxxxxxx> wrote:
> > >
> > > On Thu, 2021-06-03 at 16:33 +0200, Ilya Dryomov wrote:
> > > > On Thu, Jun 3, 2021 at 4:02 PM Jeff Layton <jlayton@xxxxxxxxxx> wrote:
> > > > >
> > > > > On Thu, 2021-06-03 at 15:57 +0200, Ilya Dryomov wrote:
> > > > > > On Thu, Jun 3, 2021 at 3:39 PM Jeff Layton <jlayton@xxxxxxxxxx> wrote:
> > > > > > >
> > > > > > > Currently ceph_update_snap_realm returns -EINVAL when it hits a decoding
> > > > > > > error, which is the wrong error code. -EINVAL implies that the user gave
> > > > > > > us a bogus argument to a syscall or something similar. -EIO is more
> > > > > > > descriptive when we hit a decoding error.
> > > > > > >
> > > > > > > Signed-off-by: Jeff Layton <jlayton@xxxxxxxxxx>
> > > > > > > ---
> > > > > > > fs/ceph/snap.c | 2 +-
> > > > > > > 1 file changed, 1 insertion(+), 1 deletion(-)
> > > > > > >
> > > > > > > diff --git a/fs/ceph/snap.c b/fs/ceph/snap.c
> > > > > > > index d07c1c6ac8fb..f8cac2abab3f 100644
> > > > > > > --- a/fs/ceph/snap.c
> > > > > > > +++ b/fs/ceph/snap.c
> > > > > > > @@ -807,7 +807,7 @@ int ceph_update_snap_trace(struct ceph_mds_client *mdsc,
> > > > > > > return 0;
> > > > > > >
> > > > > > > bad:
> > > > > > > - err = -EINVAL;
> > > > > > > + err = -EIO;
> > > > > > > fail:
> > > > > > > if (realm && !IS_ERR(realm))
> > > > > > > ceph_put_snap_realm(mdsc, realm);
> > > > > >
> > > > > > Hi Jeff,
> > > > > >
> > > > > > Is this error code propagated anywhere important?
> > > > > >
> > > > > > The vast majority of functions that have something to do with decoding
> > > > > > use EINVAL as a default (usually out-of-bounds) error. I agree that it
> > > > > > is totally ambiguous, but EIO doesn't seem to be any better to me. If
> > > > > > there is a desire to separate these errors, I think we need to pick
> > > > > > something much more distinctive.
> > > > > >
> > > > >
> > > > > When I see EINVAL, I automatically wonder what bogus argument I passed
> > > > > in somewhere, so I find that particularly deceptive here where the bug
> > > > > is either from the MDS or we had some sort of low-level socket handling
> > > > > problem.
> > > > >
> > > > > OTOH, you have a good point. The callers universally ignore the error
> > > > > code from this function. Perhaps we ought to just log a pr_warn message
> > > > > or something if the decoding fails here instead?
> > > >
> > > > There already is one:
> > > >
> > > > 793 bad:
> > > > 794 err = -EINVAL;
> > > > 795 fail:
> > > > 796 if (realm && !IS_ERR(realm))
> > > > 797 ceph_put_snap_realm(mdsc, realm);
> > > > 798 if (first_realm)
> > > > 799 ceph_put_snap_realm(mdsc, first_realm);
> > > > 800 pr_err("update_snap_trace error %d\n", err);
> > > > 801 return err;
> > > >
> > > > Or do you mean specifically the "bad" label?
> > > >
> > >
> > > Well, if we have a distinctive error code there, then we won't need a
> > > separate pr_err message or anything. I still think that -EINVAL is not
> > > descriptive of the issue though. I suppose if -EIO is too vague, we
> > > could use something like -EILSEQ ?
> >
> > In a sense it is an invalid argument because the buffer passed to the
> > decoding function is too short. This is what would lead to EINVAL here
> > and in many other decoding-related functions.
> >
> > EINVAL is the standard error code for "buffer/message too short" in
> > many other APIs. EILSEQ is certainly more distinctive, but I'm not
> > sure it is the "right" error code for this kind of error.
> >
>
> The issue is that almost everywhere else, decoding routines use -EIO for
> this. This function is a special snowflake. Why? I don't see any
> justification for it.
Ah, indeed I see that there is precedent for using EIO for this in
fs/ceph (although this particular EINVAL goes back to 2009). I just
wanted to keep things consistent with libceph and rbd where it is
mostly EINVAL.
Sorry for the noise.
Thanks,
Ilya
