On Thu, Aug 10, 2023 at 11:43:26PM -0700, Yonghong Song wrote:
>
>
> On 8/10/23 3:04 PM, David Vernet wrote:
> > Currently, if a struct_ops map is loaded with BPF_F_LINK, it must also
> > define the .validate() and .update() callbacks in its corresponding
> > struct bpf_struct_ops in the kernel. Enabling struct_ops link is useful
> > in its own right to ensure that the map is unloaded if an application
> > crashes. For example, with sched_ext, we want to automatically unload
> > the host-wide scheduler if the application crashes. We would likely
> > never support updating elements of a sched_ext struct_ops map, so we'd
> > have to implement these callbacks showing that they _can't_ support
> > element updates just to benefit from the basic lifetime management of
> > struct_ops links.
> >
> > Let's enable struct_ops maps to work with BPF_F_LINK even if they
> > haven't defined these callbacks, by assuming that a struct_ops map
> > element cannot be updated by default.
>
> Maybe you want to add one map_flag to indicate validate/update callbacks
> are optional for a struct_ops link? In this case, some struct_ops maps
> can still require validate() and update(), but others can skip them?
Are you proposing that a map flag be added that a user space caller can
specify to say that they're OK with a struct_ops implementation not
supporting .validate() and .update(), but still want to use a link to
manage registration and unregistration? Assuming I'm understanding your
suggestion correctly, I don't think it's what we want. Updating a
struct_ops map value is arguably orthogonal to the bpf link handling
registration and unregistration, so it seems confusing to require a user
to specify that it's the behavior they want as there's no reason they
shouldn't want it. If they mistakenly thought that update element is
supposed for that struct_ops variant, they'll just get an -EOPNOTSUPP
error at runtime, which seems reasonable. If a struct_ops implementation
should have implemented .validate() and/or .update() and neglects to,
that would just be a bug in the struct_ops implementation.
Apologies if I've misunderstood your proposal, and please feel free to
clarify if I have.
Thanks,
David
>
> >
> > Signed-off-by: David Vernet <void@xxxxxxxxxxxxx>
> > ---
> > kernel/bpf/bpf_struct_ops.c | 17 +++++++++++------
> > 1 file changed, 11 insertions(+), 6 deletions(-)
> >
> > diff --git a/kernel/bpf/bpf_struct_ops.c b/kernel/bpf/bpf_struct_ops.c
> > index eaff04eefb31..3d2fb85186a9 100644
> > --- a/kernel/bpf/bpf_struct_ops.c
> > +++ b/kernel/bpf/bpf_struct_ops.c
> > @@ -509,9 +509,12 @@ static long bpf_struct_ops_map_update_elem(struct bpf_map *map, void *key,
> > }
> > if (st_map->map.map_flags & BPF_F_LINK) {
> > - err = st_ops->validate(kdata);
> > - if (err)
> > - goto reset_unlock;
> > + err = 0;
> > + if (st_ops->validate) {
> > + err = st_ops->validate(kdata);
> > + if (err)
> > + goto reset_unlock;
> > + }
> > set_memory_rox((long)st_map->image, 1);
> > /* Let bpf_link handle registration & unregistration.
> > *
> > @@ -663,9 +666,6 @@ static struct bpf_map *bpf_struct_ops_map_alloc(union bpf_attr *attr)
> > if (attr->value_size != vt->size)
> > return ERR_PTR(-EINVAL);
> > - if (attr->map_flags & BPF_F_LINK && (!st_ops->validate || !st_ops->update))
> > - return ERR_PTR(-EOPNOTSUPP);
> > -
> > t = st_ops->type;
> > st_map_size = sizeof(*st_map) +
> > @@ -838,6 +838,11 @@ static int bpf_struct_ops_map_link_update(struct bpf_link *link, struct bpf_map
> > goto err_out;
> > }
> > + if (!st_map->st_ops->update) {
> > + err = -EOPNOTSUPP;
> > + goto err_out;
> > + }
> > +
> > err = st_map->st_ops->update(st_map->kvalue.data, old_st_map->kvalue.data);
> > if (err)
> > goto err_out;
