On Thu, Nov 17, 2022 at 1:52 PM Stanislav Fomichev <sdf@xxxxxxxxxx> wrote:
>
> On Thu, Nov 17, 2022 at 7:17 AM Rong Tao <rtoax@xxxxxxxxxxx> wrote:
> >
> > From: Rong Tao <rongtao@xxxxxxxx>
> >
> > commit 472caa69183f("netfilter: nat: un-export nf_nat_used_tuple")
> > introduce NF_NAT_MANIP_SRC/DST enum in include/net/netfilter/nf_nat.h,
> > and commit b06b45e82b59("selftests/bpf: add tests for bpf_ct_set_nat_info
> > kfunc") use NF_NAT_MANIP_SRC/DST in test_bpf_nf.c.
> >
> > In bpf kself-test config (tools/testing/selftests/bpf/config) nf_nat
> > is compiled as built-in, this issue occurs just if it is compiled as
> > module. We could use BPF CO-RE and ___suffix rule to avoid this.
> >
> > How to reproduce the error:
> >
> > $ make -C tools/testing/selftests/bpf/
> > ...
> > CLNG-BPF [test_maps] test_bpf_nf.bpf.o
> > error: use of undeclared identifier 'NF_NAT_MANIP_SRC'
> > bpf_ct_set_nat_info(ct, &saddr, sport, NF_NAT_MANIP_SRC);
> > ^
> > error: use of undeclared identifier 'NF_NAT_MANIP_DST'
> > bpf_ct_set_nat_info(ct, &daddr, dport, NF_NAT_MANIP_DST);
> > ^
> > 2 errors generated.
> >
> > Signed-off-by: Rong Tao <rongtao@xxxxxxxx>
> > ---
> > v2: use BPF CO-RE and ___suffix rule to avoid this error.
> > v1: https://lore.kernel.org/lkml/tencent_29D7ABD1744417031AA1B52C914B61158E07@xxxxxx/
> > ---
> > .../testing/selftests/bpf/progs/test_bpf_nf.c | 30 +++++++++++++++++--
> > 1 file changed, 27 insertions(+), 3 deletions(-)
> >
> > diff --git a/tools/testing/selftests/bpf/progs/test_bpf_nf.c b/tools/testing/selftests/bpf/progs/test_bpf_nf.c
> > index 227e85e85dda..1706984e1a6a 100644
> > --- a/tools/testing/selftests/bpf/progs/test_bpf_nf.c
> > +++ b/tools/testing/selftests/bpf/progs/test_bpf_nf.c
> > @@ -2,6 +2,7 @@
> > #include <vmlinux.h>
> > #include <bpf/bpf_helpers.h>
> > #include <bpf/bpf_endian.h>
> > +#include <bpf/bpf_core_read.h>
> >
> > #define EAFNOSUPPORT 97
> > #define EPROTO 71
> > @@ -11,6 +12,11 @@
> >
> > extern unsigned long CONFIG_HZ __kconfig;
> >
> > +enum nf_nat_manip_type___x {
> > + NF_NAT_MANIP_SRC___x,
> > + NF_NAT_MANIP_DST___x,
> > +};
> > +
> > int test_einval_bpf_tuple = 0;
> > int test_einval_reserved = 0;
> > int test_einval_netns_id = 0;
> > @@ -58,7 +64,7 @@ int bpf_ct_change_timeout(struct nf_conn *, u32) __ksym;
> > int bpf_ct_set_status(struct nf_conn *, u32) __ksym;
> > int bpf_ct_change_status(struct nf_conn *, u32) __ksym;
> > int bpf_ct_set_nat_info(struct nf_conn *, union nf_inet_addr *,
> > - int port, enum nf_nat_manip_type) __ksym;
> > + int port, int type) __ksym;
> >
> > static __always_inline void
> > nf_ct_test(struct nf_conn *(*lookup_fn)(void *, struct bpf_sock_tuple *, u32,
> > @@ -151,16 +157,34 @@ nf_ct_test(struct nf_conn *(*lookup_fn)(void *, struct bpf_sock_tuple *, u32,
> > union nf_inet_addr saddr = {};
> > union nf_inet_addr daddr = {};
> > struct nf_conn *ct_ins;
> > + int manip_src;
> > + int manip_dst;
> > + enum nf_nat_manip_type___x mapip_type_x;
> > +
> > + if (!bpf_core_type_exists(enum nf_nat_manip_type)) {
> > + bpf_printk("enum nf_nat_manip_type not exist.\n");
> > + return;
> > + }
> > +
> > + if (bpf_core_enum_value_exists(mapip_type_x, NF_NAT_MANIP_SRC___x))
> > + manip_src = bpf_core_enum_value(mapip_type_x, NF_NAT_MANIP_SRC___x);
> > + else
> > + return;
> > +
> > + if (bpf_core_enum_value_exists(mapip_type_x, NF_NAT_MANIP_DST___x))
> > + manip_dst = bpf_core_enum_value(mapip_type_x, NF_NAT_MANIP_DST___x);
> > + else
> > + return;
> >
> > bpf_ct_set_timeout(ct, 10000);
> > ct->mark = 77;
> >
> > /* snat */
> > saddr.ip = bpf_get_prandom_u32();
> > - bpf_ct_set_nat_info(ct, &saddr, sport, NF_NAT_MANIP_SRC);
> > + bpf_ct_set_nat_info(ct, &saddr, sport, manip_src);
>
> I'm not sure these co-re checks are helpful. Can we just hardcode 1/0
> here and below?
>
> bpf_ct_set_nat_info(ct, &saddr, sport, 0 /*NF_NAT_MANIP_SRC*/);
> bpf_ct_set_nat_info(ct, &daddr, dport, 1 /*NF_NAT_MANIP_DST*/);
>
> But I'm also overall not sure we need to make this test flexible; we
> have a lot of tests that depend on tools/testing/selftests/bpf/config;
> at some point I was trying to make the tests more tolerant to
> different environments, but it went nowhere..
Agreed. bpf_core_enum_value_exists() makes no sense here.
bpf_core_enum_value(enum nf_nat_manip_type___x, NF_NAT_MANIP_SRC___x)
would be ok, IMHO. It will compile but fail at runtime if the module
is not loaded.
>
>
> > /* dnat */
> > daddr.ip = bpf_get_prandom_u32();
> > - bpf_ct_set_nat_info(ct, &daddr, dport, NF_NAT_MANIP_DST);
> > + bpf_ct_set_nat_info(ct, &daddr, dport, manip_dst);
> >
> > ct_ins = bpf_ct_insert_entry(ct);
> > if (ct_ins) {
> > --
> > 2.31.1
> >
