test_xdp_frags: check for privs early

[dankamongmen]

As currently written, test_xdp_frags if run without necessary permissions will loudly print "this kernel does NOT support xdp frags", and then exit on the setrlimit(). The same thing happens if the argument list is invalid. Instead, verify all this before performing the frag support check, so as not to dismay users (like me) who try using this program to check for frag support.

I didn't touch the formatting in this file so as to produce a minimal diff for review, but it's kinda all over the place.

[tohojo]

Hmm, interesting; I didn't anticipate someone would actually use this for that purpose.

As for the second bit, can you be a bit more specific than "the formatting is all over the place"? 😅

[dankamongmen]

Hmm, interesting; I didn't anticipate someone would actually use this for that purpose.

as a pure user (i.e. didn't do any of the core development) of XDP and AF_XDP, there are a lot of questions unanswered by the documentation. in the hope of figuring some things out, i cast around for any helpful tool =].

and i'm going to at some point need to know for my application whether the hardware stack on which i'm running supports frags/multibufs (of course, as there is no AF_XDP support yet, the answer for me would be 'none of them').

please don't interpret this as complaining! just letting you know where i'm coming from.

i wrote this blog post last night you might find interesting:
https://nick-black.com/dankwiki/index.php/Extended_disquisitions_pertaining_to_eXpress_data_paths_(XDP)

As for the second bit, can you be a bit more specific than "the formatting is all over the place"? sweat_smile

using cat -T:

int main(int argc, char **argv)
{
^Istruct rlimit r = {RLIM_INFINITY, RLIM_INFINITY};
^Iint ifindex_bigmtu, ifindex_smallmtu, ret;
        char *envval;

        envval = secure_getenv("VERBOSE_TESTS");

^Isilence_libbpf_logging();
        if (envval && envval[0] == '1')
                verbose_libxdp_logging();
        else
                silence_libxdp_logging();

        kern_compat = check_frags_compat();

^Iif (argc != 3)
                usage(argv[0]);

^Iifindex_bigmtu = if_nametoindex(argv[1]);
^Iifindex_smallmtu = if_nametoindex(argv[2]);
^Iif (!ifindex_bigmtu || !ifindex_smallmtu) {
^I^Ifprintf(stderr, "Interface '%s' or '%s' not found.\n", argv[1], argv[2]);
                usage(argv[0]);
^I}

^Iif (setrlimit(RLIMIT_MEMLOCK, &r)) {
^I^Ifprintf(stderr, "ERROR: setrlimit(RLIMIT_MEMLOCK) \"%s\"\n",
^I^I^Istrerror(errno));
^I^Iexit(EXIT_FAILURE);
^I}

        ret = check_load_frags(kern_compat ? ifindex_bigmtu : 0, ifindex_smallmtu);
        ret = check_load_nofrags_success(ifindex_smallmtu) || ret;
^Iif (kern_compat) {
^I^Iret = check_load_nofrags_fail(ifindex_bigmtu) || ret;
^I^Iret = check_load_frags_multi(ifindex_bigmtu) || ret;
                ret = check_load_mix_big(ifindex_bigmtu) || ret;
^I}
^Iret = check_load_mix_small(ifindex_smallmtu) || ret;

^Ireturn ret;
}

the ^I are tabs. the spaces are spaces.

[tohojo]

nick black ***@***.***> writes:

> Hmm, interesting; I didn't anticipate someone would actually use this for that purpose. as a pure user (i.e. didn't do any of the core development) of XDP and `AF_XDP`, there are a lot of questions unanswered by the documentation. in the hope of figuring some things out, i cast around for any helpful tool =]. and i'm going to at some point need to know for my application whether the hardware stack on which i'm running supports frags/multibufs (of course, as there is no AF_XDP support yet, the answer for me would be 'none of them'). please don't interpret this as complaining! just letting you know where i'm coming from.

It's quite OK to complain - usability of these things is severely lacking in BPF/XDP land. I'm just intrigued that you actually found this use, I'm not trying to disparage it, certainly. But as you say, ideally we should have something more accessible. There was recently a series merged in the upstream kernel that adds explicit "feature flags" for XDP so userspace can query the kernel what a certain netdevice supports. This is still fairly new on the kernel side (I forget if it's 6.2 or 6.3 it landed in), but once it gets a bit more widespread, the plan is to add support for this in xdp-tools so you can do something like `xdp-loader features <dev>` and get a full list of what that device supports (or doesn't support). So there is some hope for a friendlier future, fortunately :)

i wrote this blog post last night you might find interesting: https://nick-black.com/dankwiki/index.php/Extended_disquisitions_pertaining_to_eXpress_data_paths_(XDP)

Ah yes, very interesting. You are...not wrong.. about many of your complaints. I opened #317 for that locking issue you mention; feel free to open another one with specifics about those valgrind issues.

> As for the second bit, can you be a bit more specific than "the formatting is all over the place"? sweat_smile using `cat -T`:

Ah, right. Yeah, those should really be tabs all of them; that's probably my fault for not configuring my editor properly when writing that code (Emacs has this weird issue with not always picking up the tabs-only indentation thing, and I don't always notice). If you want to fix it, doing so in a separate commit in the same PR should be fine...