Re: [PATCH rdma-next] RDMA/uverbs: Propagate errors from rdma_lookup_get_uobject()

[Leon Romanovsky <leon@kernel.org>]

On Wed, Feb 19, 2025 at 01:53:35PM -0400, Jason Gunthorpe wrote:
> On Wed, Feb 19, 2025 at 05:56:47PM +0200, Leon Romanovsky wrote:
> > On Wed, Feb 19, 2025 at 10:46:16AM -0400, Jason Gunthorpe wrote:
> > > On Wed, Feb 19, 2025 at 03:52:05PM +0200, Leon Romanovsky wrote:
> > > > From: Maher Sanalla <msanalla@nvidia.com>
> > > > 
> > > > Currently, the IB uverbs API calls uobj_get_uobj_read(), which in turn
> > > > uses the rdma_lookup_get_uobject() helper to retrieve user objects.
> > > > In case of failure, uobj_get_uobj_read() returns NULL, overriding the
> > > > error code from rdma_lookup_get_uobject(). The IB uverbs API then
> > > > translates this NULL to -EINVAL, masking the actual error and
> > > > complicating debugging.
> > > 
> > > This may have been deliberate as this old stuff is not supposed to be
> > > returning weird error codes.
> > 
> > I assumed that this was the reason for such overwrite in the past, but
> > is this continue to be true in 2025?
> 
> Maybe, it is ABI that leaks out libiverbs
> 
> But also, maybe nobody cares. There is a small chance places are
> relying on detecting certain errnos.
> 
> > > What error code are you missing here?
> > 
> > Error returned from modify QP was masked by setting real error to be -EINVAL.
> 
> What errno was it though? What other errors are there that are now no
> longer supressed?

Mainly -EBUSY from FW command interface, so users can safely call again
to modify QP.

> 
> I think the commit message needs a deeper analysis to be convincing
> the ABI break is low risk

No problem, we will update commit message.

Thanks

> 
> Jason