RE: [PATCH v2 bpf-next] libbpf: Add bpf_obj_get_opts()

[Roberto Sassu]

> From: Stanislav Fomichev [mailto:sdf@xxxxxxxxxx]
> Sent: Thursday, July 21, 2022 12:38 AM
> On Wed, Jul 20, 2022 at 3:30 PM Roberto Sassu <roberto.sassu@xxxxxxxxxx>
> wrote:
> >
> > > From: Stanislav Fomichev [mailto:sdf@xxxxxxxxxx]
> > > Sent: Wednesday, July 20, 2022 5:57 PM
> > > On Wed, Jul 20, 2022 at 1:02 AM Roberto Sassu
> <roberto.sassu@xxxxxxxxxx>
> > > wrote:
> > > >
> > > > > From: Stanislav Fomichev [mailto:sdf@xxxxxxxxxx]
> > > > > Sent: Tuesday, July 19, 2022 10:40 PM
> > > > > On Tue, Jul 19, 2022 at 12:40 PM Joe Burton
> <jevburton.kernel@xxxxxxxxx>
> > > > > wrote:
> > > > > >
> > > > > > From: Joe Burton <jevburton@xxxxxxxxxx>
> > > > > >
> > > > > > Add an extensible variant of bpf_obj_get() capable of setting the
> > > > > > `file_flags` parameter.
> > > > > >
> > > > > > This parameter is needed to enable unprivileged access to BPF maps.
> > > > > > Without a method like this, users must manually make the syscall.
> > > > > >
> > > > > > Signed-off-by: Joe Burton <jevburton@xxxxxxxxxx>
> > > > >
> > > > > Reviewed-by: Stanislav Fomichev <sdf@xxxxxxxxxx>
> > > > >
> > > > > For context:
> > > > > We've found this out while we were trying to add support for unpriv
> > > > > processes to open pinned r-x maps.
> > > > > Maybe this deserves a test as well? Not sure.
> > > >
> > > > Hi Stanislav, Joe
> > > >
> > > > I noticed now this patch. I'm doing a broader work to add opts
> > > > to bpf_*_get_fd_by_id(). I also adjusted permissions of bpftool
> > > > depending on the operation type (e.g. show, dump: BPF_F_RDONLY).
> > > >
> > > > Will send it soon (I'm trying to solve an issue with the CI, where
> > > > libbfd is not available in the VM doing actual tests).
> > >
> > > Is something like this patch included in your series as well? Can you
> > > use this new interface or do you need something different?
> >
> > It is very similar. Except that I called it bpf_get_fd_opts, as it
> > is shared with the bpf_*_get_fd_by_id() functions. The member
> > name is just flags, plus an extra u32 for alignment.
> 
> We can bikeshed the naming, but we've been using existing conventions
> where opts fields match syscall fields, that seems like a sensible
> thing to do?

The only problem is that bpf_*_get_fd_by_id() functions would
set the open_flags member of bpf_attr.

Flags would be good for both, even if not exact. Believe me,
duplicating the opts would just create more confusion.

> > It needs to be shared, as there are functions in bpftool calling
> > both. Since the meaning of flags is the same, seems ok sharing.
> 
> So I guess there are no objections to the current patch? If it gets
> accepted, you should be able to drop some of your code and use this
> new bpf_obj_get_opts..

If you use a name good also for bpf_*_get_fd_by_id() and flags
as structure member name, that would be ok.

Roberto

> > Roberto
> >
> > > > Roberto
> > > >
> > > > > > ---
> > > > > >  tools/lib/bpf/bpf.c      | 10 ++++++++++
> > > > > >  tools/lib/bpf/bpf.h      |  9 +++++++++
> > > > > >  tools/lib/bpf/libbpf.map |  1 +
> > > > > >  3 files changed, 20 insertions(+)
> > > > > >
> > > > > > diff --git a/tools/lib/bpf/bpf.c b/tools/lib/bpf/bpf.c
> > > > > > index 5eb0df90eb2b..5acb0e8bd13c 100644
> > > > > > --- a/tools/lib/bpf/bpf.c
> > > > > > +++ b/tools/lib/bpf/bpf.c
> > > > > > @@ -578,12 +578,22 @@ int bpf_obj_pin(int fd, const char
> *pathname)
> > > > > >  }
> > > > > >
> > > > > >  int bpf_obj_get(const char *pathname)
> > > > > > +{
> > > > > > +       LIBBPF_OPTS(bpf_obj_get_opts, opts);
> > > > > > +       return bpf_obj_get_opts(pathname, &opts);
> > > > > > +}
> > > > > > +
> > > > > > +int bpf_obj_get_opts(const char *pathname, const struct
> > > bpf_obj_get_opts
> > > > > *opts)
> > > > > >  {
> > > > > >         union bpf_attr attr;
> > > > > >         int fd;
> > > > > >
> > > > > > +       if (!OPTS_VALID(opts, bpf_obj_get_opts))
> > > > > > +               return libbpf_err(-EINVAL);
> > > > > > +
> > > > > >         memset(&attr, 0, sizeof(attr));
> > > > > >         attr.pathname = ptr_to_u64((void *)pathname);
> > > > > > +       attr.file_flags = OPTS_GET(opts, file_flags, 0);
> > > > > >
> > > > > >         fd = sys_bpf_fd(BPF_OBJ_GET, &attr, sizeof(attr));
> > > > > >         return libbpf_err_errno(fd);
> > > > > > diff --git a/tools/lib/bpf/bpf.h b/tools/lib/bpf/bpf.h
> > > > > > index 88a7cc4bd76f..f31b493b5f9a 100644
> > > > > > --- a/tools/lib/bpf/bpf.h
> > > > > > +++ b/tools/lib/bpf/bpf.h
> > > > > > @@ -270,8 +270,17 @@ LIBBPF_API int bpf_map_update_batch(int fd,
> > > const
> > > > > void *keys, const void *values
> > > > > >                                     __u32 *count,
> > > > > >                                     const struct bpf_map_batch_opts *opts);
> > > > > >
> > > > > > +struct bpf_obj_get_opts {
> > > > > > +       size_t sz; /* size of this struct for forward/backward compatibility
> */
> > > > > > +
> > > > > > +       __u32 file_flags;
> > > > > > +};
> > > > > > +#define bpf_obj_get_opts__last_field file_flags
> > > > > > +
> > > > > >  LIBBPF_API int bpf_obj_pin(int fd, const char *pathname);
> > > > > >  LIBBPF_API int bpf_obj_get(const char *pathname);
> > > > > > +LIBBPF_API int bpf_obj_get_opts(const char *pathname,
> > > > > > +                               const struct bpf_obj_get_opts *opts);
> > > > > >
> > > > > >  struct bpf_prog_attach_opts {
> > > > > >         size_t sz; /* size of this struct for forward/backward compatibility
> */
> > > > > > diff --git a/tools/lib/bpf/libbpf.map b/tools/lib/bpf/libbpf.map
> > > > > > index 0625adb9e888..119e6e1ea7f1 100644
> > > > > > --- a/tools/lib/bpf/libbpf.map
> > > > > > +++ b/tools/lib/bpf/libbpf.map
> > > > > > @@ -355,6 +355,7 @@ LIBBPF_0.8.0 {
> > > > > >
> > > > > >  LIBBPF_1.0.0 {
> > > > > >         global:
> > > > > > +               bpf_obj_get_opts;
> > > > > >                 bpf_prog_query_opts;
> > > > > >                 bpf_program__attach_ksyscall;
> > > > > >                 btf__add_enum64;
> > > > > > --
> > > > > > 2.37.0.170.g444d1eabd0-goog
> > > > > >