On Tue, Mar 19, 2024 at 12:54 AM Eduard Zingerman eddyz87@gmail.com wrote:
On Fri, 2024-03-15 at 15:29 +0100, Benjamin Tissoires wrote: [...]
@@ -5279,7 +5281,8 @@ static int map_kptr_match_type(struct bpf_verifier_env *env,
static bool in_sleepable(struct bpf_verifier_env *env) {
return env->prog->sleepable;
return env->prog->sleepable ||(env->cur_state && env->cur_state->in_sleepable);}
I was curious why 'env->cur_state &&' check was needed and found that removing it caused an error in the following fragment:
static int do_misc_fixups(struct bpf_verifier_env *env) { ... if (is_storage_get_function(insn->imm)) { if (!in_sleepable(env) || env->insn_aux_data[i + delta].storage_get_func_atomic) insn_buf[0] = BPF_MOV64_IMM(BPF_REG_5, (__force __s32)GFP_ATOMIC); else insn_buf[0] = BPF_MOV64_IMM(BPF_REG_5, (__force __s32)GFP_KERNEL); ... } ... }
When do_misc_fixups() is done env->cur_state is NULL. Current implementation would use GFP_ATOMIC allocation even for sleepable callbacks, where GFP_KERNEL is sufficient. Is this is something we want to address?
I honestly have no idea of the impact there.
AFAICT, if env->cur_state is not set, we don't even know if the callback will be sleepable or not, so if there is a small penalty, then it's the safest option, no?
Cheers, Benjamin
/* The non-sleepable programs and sleepable programs with explicit bpf_rcu_read_lock()