On 11/1/24 1:13 PM, Alexei Starovoitov wrote:
On Thu, Oct 31, 2024 at 8:10 PM Yonghong Song <yonghong.song@xxxxxxxxx> wrote:For struct_ops progs, whether a particular prog will use private stack or not (prog->aux->use_priv_stack) will be set before actual insn-level verification for that prog. One particular implementation is to piggyback on struct_ops->check_member(). The next patch will have an example for this. The struct_ops->check_member() will set prog->aux->use_priv_stack to be true which enables private stack usage with ignoring BPF_PRIV_STACK_MIN_SIZE limit. If use_priv_stack is true for a particular struct_ops prog, bpf trampoline will need to do recursion checks (one level at this point) to avoid stack overwrite. A field (recursion_skipped()) is added to bpf_prog_aux structure such that if bpf_prog->aux->recursion_skipped is set by the struct_ops subsystem, the function will be called to terminate the prog run, collect related info, etc. Acked-by: Tejun Heo <tj@xxxxxxxxxx> Signed-off-by: Yonghong Song <yonghong.song@xxxxxxxxx> --- include/linux/bpf.h | 1 + include/linux/bpf_verifier.h | 1 + kernel/bpf/trampoline.c | 4 ++++ kernel/bpf/verifier.c | 36 ++++++++++++++++++++++++++++++++---- 4 files changed, 38 insertions(+), 4 deletions(-) diff --git a/include/linux/bpf.h b/include/linux/bpf.h index 8a3ea7440a4a..7a34108c6974 100644 --- a/include/linux/bpf.h +++ b/include/linux/bpf.h @@ -1528,6 +1528,7 @@ struct bpf_prog_aux { u64 prog_array_member_cnt; /* counts how many times as member of prog_array */ struct mutex ext_mutex; /* mutex for is_extended and prog_array_member_cnt */ struct bpf_arena *arena; + void (*recursion_skipped)(struct bpf_prog *prog); /* callback if recursion is skipped */The name doesn't fit. The recursion wasn't skipped. It's the execution of the program that was skipped. 'recursion_detected' or 'recursion_disallowed' would be a better name.
I will use recursion_detected().
/* BTF_KIND_FUNC_PROTO for valid attach_btf_id */ const struct btf_type *attach_func_proto; /* function name for valid attach_btf_id */ diff --git a/include/linux/bpf_verifier.h b/include/linux/bpf_verifier.h index bc28ce7996ac..ff0fba935f89 100644 --- a/include/linux/bpf_verifier.h +++ b/include/linux/bpf_verifier.h @@ -889,6 +889,7 @@ static inline bool bpf_prog_check_recur(const struct bpf_prog *prog) case BPF_PROG_TYPE_TRACING: return prog->expected_attach_type != BPF_TRACE_ITER; case BPF_PROG_TYPE_STRUCT_OPS: + return prog->aux->use_priv_stack; case BPF_PROG_TYPE_LSM: return false; default: diff --git a/kernel/bpf/trampoline.c b/kernel/bpf/trampoline.c index 9f36c049f4c2..a84e60efbf89 100644 --- a/kernel/bpf/trampoline.c +++ b/kernel/bpf/trampoline.c @@ -899,6 +899,8 @@ static u64 notrace __bpf_prog_enter_recur(struct bpf_prog *prog, struct bpf_tram if (unlikely(this_cpu_inc_return(*(prog->active)) != 1)) { bpf_prog_inc_misses_counter(prog); + if (prog->aux->recursion_skipped) + prog->aux->recursion_skipped(prog); return 0; } return bpf_prog_start_time(); @@ -975,6 +977,8 @@ u64 notrace __bpf_prog_enter_sleepable_recur(struct bpf_prog *prog, if (unlikely(this_cpu_inc_return(*(prog->active)) != 1)) { bpf_prog_inc_misses_counter(prog); + if (prog->aux->recursion_skipped) + prog->aux->recursion_skipped(prog); return 0; } return bpf_prog_start_time(); diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c index 30e74db6a85f..865191c5d21b 100644 --- a/kernel/bpf/verifier.c +++ b/kernel/bpf/verifier.c @@ -6023,17 +6023,31 @@ static int check_ptr_alignment(struct bpf_verifier_env *env, static int bpf_enable_priv_stack(struct bpf_verifier_env *env) { + bool force_priv_stack = env->prog->aux->use_priv_stack; struct bpf_subprog_info *si; + int ret; + + if (!bpf_jit_supports_private_stack()) { + if (force_priv_stack) { + verbose(env, "Private stack not supported by jit\n"); + return -EACCES; + }This logic would fit better in the patch 2. Less code churn and the whole approach is easier to understand. I don't like this inband signaling. Now I see why you had that weird <0 check in patch 2 :( This is ugly. May be it should be a separate bool request_priv_stack:1 that struct_ops callback will set and it will clean up this logic.
I can add this logic to function check_struct_ops_btf_id(), which is struct_ops preparation for verification. This will ensure in bpf_enable_priv_stack(), if !bpf_jit_supports_private_stack(), it is guaranteed to return NO_PRIV_STACK.
- if (!bpf_jit_supports_private_stack()) return NO_PRIV_STACK; + } + ret = PRIV_STACK_ADAPTIVE; switch (env->prog->type) { case BPF_PROG_TYPE_KPROBE: case BPF_PROG_TYPE_TRACEPOINT: case BPF_PROG_TYPE_PERF_EVENT: case BPF_PROG_TYPE_RAW_TRACEPOINT: break; + case BPF_PROG_TYPE_STRUCT_OPS: + if (!force_priv_stack) + return NO_PRIV_STACK; + ret = PRIV_STACK_ALWAYS; + break; case BPF_PROG_TYPE_TRACING: if (env->prog->expected_attach_type != BPF_TRACE_ITER) break; @@ -6044,11 +6058,18 @@ static int bpf_enable_priv_stack(struct bpf_verifier_env *env) si = env->subprog_info; for (int i = 0; i < env->subprog_cnt; i++) { - if (si[i].has_tail_call) + if (si[i].has_tail_call) { + if (ret == PRIV_STACK_ALWAYS) { + verbose(env, + "Private stack not supported due to tail call presence\n"); + return -EACCES; + } + return NO_PRIV_STACK; + } } - return PRIV_STACK_ADAPTIVE; + return ret; }
[...]
