On Tue, 2025-01-14 at 13:28 -0700, Daniel Xu wrote: > This commit allows progs to elide a null check on statically known > map > lookup keys. In other words, if the verifier can statically prove > that > the lookup will be in-bounds, allow the prog to drop the null check. > > This is useful for two reasons: > > 1. Large numbers of nullness checks (especially when they cannot > fail) > unnecessarily pushes prog towards BPF_COMPLEXITY_LIMIT_JMP_SEQ. > 2. It forms a tighter contract between programmer and verifier. > > For (1), bpftrace is starting to make heavier use of percpu scratch > maps. As a result, for user scripts with large number of unrolled > loops, > we are starting to hit jump complexity verification errors. These > percpu lookups cannot fail anyways, as we only use static key values. > Eliding nullness probably results in less work for verifier as well. > > For (2), percpu scratch maps are often used as a larger stack, as the > currrent stack is limited to 512 bytes. In these situations, it is > desirable for the programmer to express: "this lookup should never > fail, > and if it does, it means I messed up the code". By omitting the null > check, the programmer can "ask" the verifier to double check the > logic. > > Tests also have to be updated in sync with these changes, as the > verifier is more efficient with this change. Notable, iters.c tests > had > to be changed to use a map type that still requires null checks, as > it's > exercising verifier tracking logic w.r.t iterators. > > Signed-off-by: Daniel Xu <dxu@xxxxxxxxx> > --- > kernel/bpf/verifier.c | 92 > ++++++++++++++++++- > tools/testing/selftests/bpf/progs/iters.c | 14 +-- > .../selftests/bpf/progs/map_kptr_fail.c | 2 +- > .../selftests/bpf/progs/verifier_map_in_map.c | 2 +- > .../testing/selftests/bpf/verifier/map_kptr.c | 2 +- > 5 files changed, 99 insertions(+), 13 deletions(-) [...] > @@ -9158,6 +9216,7 @@ static int check_func_arg(struct > bpf_verifier_env *env, u32 arg, > enum bpf_arg_type arg_type = fn->arg_type[arg]; > enum bpf_reg_type type = reg->type; > u32 *arg_btf_id = NULL; > + u32 key_size; > int err = 0; > > if (arg_type == ARG_DONTCARE) > @@ -9291,8 +9350,13 @@ static int check_func_arg(struct > bpf_verifier_env *env, u32 arg, > verbose(env, "invalid map_ptr to access map- > >key\n"); > return -EACCES; > } > - err = check_helper_mem_access(env, regno, meta- > >map_ptr->key_size, > - BPF_READ, false, > NULL); > + key_size = meta->map_ptr->key_size; > + err = check_helper_mem_access(env, regno, key_size, > BPF_READ, false, NULL); > + if (err) > + return err; > + meta->const_map_key = get_constant_map_key(env, reg, > key_size); > + if (meta->const_map_key < 0 && meta->const_map_key > != -EOPNOTSUPP) > + return meta->const_map_key; Mark Hartmayer reported a problem that after this commit the verifier started refusing to load libvirt's virCgroupV2DevicesLoadProg(), which contains the following snippet: 53: (b7) r1 = -1 ; R1_w=-1 54: (7b) *(u64 *)(r10 -8) = r1 ; R1_w=-1 R10=fp0 fp-8_w=-1 55: (bf) r2 = r10 ; R2_w=fp0 R10=fp0 56: (07) r2 += -8 ; R2_w=fp-8 57: (18) r1 = 0x9553c800 ; R1_w=map_ptr(ks=8,vs=4) 59: (85) call bpf_map_lookup_elem#1 IIUC here the actual constant value is -1, which this code confuses with an error.
