Re: [PATCH bpf-next 0/5] libbpf: error reporting changes for v1.0

["Toke Høiland-Jørgensen" <toke@redhat.com>]

Andrii Nakryiko <andrii.nakryiko@gmail.com> writes:

> On Mon, May 24, 2021 at 2:34 PM Toke Høiland-Jørgensen <toke@redhat.com> wrote:
>>
>> Andrii Nakryiko <andrii.nakryiko@gmail.com> writes:
>>
>> > On Mon, May 24, 2021 at 1:53 PM Toke Høiland-Jørgensen <toke@redhat.com> wrote:
>> >>
>> >> Andrii Nakryiko <andrii.nakryiko@gmail.com> writes:
>> >>
>> >> > On Sun, May 23, 2021 at 11:36 PM John Fastabend
>> >> > <john.fastabend@gmail.com> wrote:
>> >> >>
>> >> >> Andrii Nakryiko wrote:
>> >> >> > Implement error reporting changes discussed in "Libbpf: the road to v1.0"
>> >> >> > ([0]) document.
>> >> >> >
>> >> >> > Libbpf gets a new API, libbpf_set_strict_mode() which accepts a set of flags
>> >> >> > that turn on a set of libbpf 1.0 changes, that might be potentially breaking.
>> >> >> > It's possible to opt-in into all current and future 1.0 features by specifying
>> >> >> > LIBBPF_STRICT_ALL flag.
>> >> >> >
>> >> >> > When some of the 1.0 "features" are requested, libbpf APIs might behave
>> >> >> > differently. In this patch set a first set of changes are implemented, all
>> >> >> > related to the way libbpf returns errors. See individual patches for details.
>> >> >> >
>> >> >> > Patch #1 adds a no-op libbpf_set_strict_mode() functionality to enable
>> >> >> > updating selftests.
>> >> >> >
>> >> >> > Patch #2 gets rid of all the bad code patterns that will break in libbpf 1.0
>> >> >> > (exact -1 comparison for low-level APIs, direct IS_ERR() macro usage to check
>> >> >> > pointer-returning APIs for error, etc). These changes make selftest work in
>> >> >> > both legacy and 1.0 libbpf modes. Selftests also opt-in into 100% libbpf 1.0
>> >> >> > mode to automatically gain all the subsequent changes, which will come in
>> >> >> > follow up patches.
>> >> >> >
>> >> >> > Patch #3 streamlines error reporting for low-level APIs wrapping bpf() syscall.
>> >> >> >
>> >> >> > Patch #4 streamlines errors for all the rest APIs.
>> >> >> >
>> >> >> > Patch #5 ensures that BPF skeletons propagate errors properly as well, as
>> >> >> > currently on error some APIs will return NULL with no way of checking exact
>> >> >> > error code.
>> >> >> >
>> >> >> >   [0] https://docs.google.com/document/d/1UyjTZuPFWiPFyKk1tV5an11_iaRuec6U-ZESZ54nNTY
>> >> >> >
>> >> >> > Andrii Nakryiko (5):
>> >> >> >   libbpf: add libbpf_set_strict_mode() API to turn on libbpf 1.0
>> >> >> >     behaviors
>> >> >> >   selftests/bpf: turn on libbpf 1.0 mode and fix all IS_ERR checks
>> >> >> >   libbpf: streamline error reporting for low-level APIs
>> >> >> >   libbpf: streamline error reporting for high-level APIs
>> >> >> >   bpftool: set errno on skeleton failures and propagate errors
>> >> >> >
>> >> >>
>> >> >> LGTM for the series,
>> >> >>
>> >> >> Acked-by: John Fastabend <john.fastabend@gmail.com>
>> >> >
>> >> > Thanks, John!
>> >> >
>> >> > Toke, Stanislav, you cared about these aspects of libbpf 1.0 (by
>> >> > commenting on the doc itself), do you mind also taking a brief look
>> >> > and letting me know if this works for your use cases? Thanks!
>> >>
>> >> Changes LGTM:
>> >>
>> >> Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
>> >>
>> >> As a side note, the series seems to have been chopped up into individual
>> >> emails with no threading; was a bit weird that I had to go hunting for
>> >> the individual patches in my mailbox...
>> >>
>> >
>> > That's my bad, I messed up and sent them individually and probably
>> > that's why they weren't threaded properly.
>>
>> Right, OK, I'll stop looking for bugs on my end, then :)
>>
>> BTW, one more thing that just came to mind: since that gdoc is not
>> likely to be around forever, would it be useful to make the reference in
>> the commit message(s) point to something more stable? IDK what that
>> shoul be, really. Maybe just pasting (an abbreviated outline of?) the
>> text in the document into the cover letter / merge commit could work?
>
> I was hoping Google won't deprecate Google Docs any time soon and I
> had no intention to remove that document. But I was also thinking to
> start wiki page at github.com/libbpf/libbpf with migration
> instructions, so once that is up and running I can link that from
> libbpf_set_strict_mode() doc comment.

Right, that sounds reasonable :)

> But I'd like to avoid blocking on that.

Understandable; but just pasting an outline into the commit message (and
keeping the link) could work in the meantime?

-Toke