On Fri, Jun 2, 2023 at 12:01 AM <menglong8.dong@xxxxxxxxx> wrote:
>
> From: Menglong Dong <imagedong@xxxxxxxxxxx>
Please trim your cc when you respin. It's unnecessary huge.
> For now, the BPF program of type BPF_PROG_TYPE_TRACING can only be used
> on the kernel functions whose arguments count less than 6. This is not
> friendly at all, as too many functions have arguments count more than 6.
>
> Therefore, let's enhance it by increasing the function arguments count
> allowed in arch_prepare_bpf_trampoline(), for now, only x86_64.
>
> For the case that we don't need to call origin function, which means
> without BPF_TRAMP_F_CALL_ORIG, we need only copy the function arguments
> that stored in the frame of the caller to current frame. The arguments
> of arg6-argN are stored in "$rbp + 0x18", we need copy them to
> "$rbp - regs_off + (6 * 8)".
>
> For the case with BPF_TRAMP_F_CALL_ORIG, we need prepare the arguments
> in stack before call origin function, which means we need alloc extra
> "8 * (arg_count - 6)" memory in the top of the stack. Note, there should
> not be any data be pushed to the stack before call the origin function.
> Then, we have to store rbx with 'mov' instead of 'push'.
>
> It works well for the FENTRY and FEXIT, I'm not sure if there are other
> complicated cases.
>
> Reviewed-by: Jiang Biao <benbjiang@xxxxxxxxxxx>
> Signed-off-by: Menglong Dong <imagedong@xxxxxxxxxxx>
> ---
> v2:
> - instead EMIT4 with EMIT3_off32 for "lea" to prevent overflow
> - make MAX_BPF_FUNC_ARGS as the maximum argument count
> ---
> arch/x86/net/bpf_jit_comp.c | 96 +++++++++++++++++++++++++++++++------
> 1 file changed, 81 insertions(+), 15 deletions(-)
>
> diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c
> index 1056bbf55b17..0e247bb7d6f6 100644
> --- a/arch/x86/net/bpf_jit_comp.c
> +++ b/arch/x86/net/bpf_jit_comp.c
> @@ -1868,7 +1868,7 @@ static void save_regs(const struct btf_func_model *m, u8 **prog, int nr_regs,
> * mov QWORD PTR [rbp-0x10],rdi
> * mov QWORD PTR [rbp-0x8],rsi
> */
> - for (i = 0, j = 0; i < min(nr_regs, 6); i++) {
> + for (i = 0, j = 0; i < min(nr_regs, MAX_BPF_FUNC_ARGS); i++) {
> /* The arg_size is at most 16 bytes, enforced by the verifier. */
> arg_size = m->arg_size[j];
> if (arg_size > 8) {
> @@ -1876,10 +1876,22 @@ static void save_regs(const struct btf_func_model *m, u8 **prog, int nr_regs,
> next_same_struct = !next_same_struct;
> }
>
> - emit_stx(prog, bytes_to_bpf_size(arg_size),
> - BPF_REG_FP,
> - i == 5 ? X86_REG_R9 : BPF_REG_1 + i,
> - -(stack_size - i * 8));
> + if (i <= 5) {
> + /* store function arguments in regs */
The comment is confusing.
It's not storing arguments in regs.
It copies them from regs into stack.
> + emit_stx(prog, bytes_to_bpf_size(arg_size),
> + BPF_REG_FP,
> + i == 5 ? X86_REG_R9 : BPF_REG_1 + i,
> + -(stack_size - i * 8));
> + } else {
> + /* store function arguments in stack */
> + emit_ldx(prog, bytes_to_bpf_size(arg_size),
> + BPF_REG_0, BPF_REG_FP,
> + (i - 6) * 8 + 0x18);
> + emit_stx(prog, bytes_to_bpf_size(arg_size),
and we will have garbage values in upper bytes.
Probably should fix both here and in regular copy from reg.
> + BPF_REG_FP,
> + BPF_REG_0,
> + -(stack_size - i * 8));
> + }
>
> j = next_same_struct ? j : j + 1;
> }
> @@ -1913,6 +1925,41 @@ static void restore_regs(const struct btf_func_model *m, u8 **prog, int nr_regs,
> }
> }
>
> +static void prepare_origin_stack(const struct btf_func_model *m, u8 **prog,
> + int nr_regs, int stack_size)
> +{
> + int i, j, arg_size;
> + bool next_same_struct = false;
> +
> + if (nr_regs <= 6)
> + return;
> +
> + /* Prepare the function arguments in stack before call origin
> + * function. These arguments must be stored in the top of the
> + * stack.
> + */
> + for (i = 0, j = 0; i < min(nr_regs, MAX_BPF_FUNC_ARGS); i++) {
> + /* The arg_size is at most 16 bytes, enforced by the verifier. */
> + arg_size = m->arg_size[j];
> + if (arg_size > 8) {
> + arg_size = 8;
> + next_same_struct = !next_same_struct;
> + }
> +
> + if (i > 5) {
> + emit_ldx(prog, bytes_to_bpf_size(arg_size),
> + BPF_REG_0, BPF_REG_FP,
> + (i - 6) * 8 + 0x18);
> + emit_stx(prog, bytes_to_bpf_size(arg_size),
> + BPF_REG_FP,
> + BPF_REG_0,
> + -(stack_size - (i - 6) * 8));
> + }
> +
> + j = next_same_struct ? j : j + 1;
> + }
> +}
> +
> static int invoke_bpf_prog(const struct btf_func_model *m, u8 **pprog,
> struct bpf_tramp_link *l, int stack_size,
> int run_ctx_off, bool save_ret)
> @@ -1938,7 +1985,7 @@ static int invoke_bpf_prog(const struct btf_func_model *m, u8 **pprog,
> /* arg1: mov rdi, progs[i] */
> emit_mov_imm64(&prog, BPF_REG_1, (long) p >> 32, (u32) (long) p);
> /* arg2: lea rsi, [rbp - ctx_cookie_off] */
> - EMIT4(0x48, 0x8D, 0x75, -run_ctx_off);
> + EMIT3_off32(0x48, 0x8D, 0xB5, -run_ctx_off);
>
> if (emit_rsb_call(&prog, bpf_trampoline_enter(p), prog))
> return -EINVAL;
> @@ -1954,7 +2001,7 @@ static int invoke_bpf_prog(const struct btf_func_model *m, u8 **pprog,
> emit_nops(&prog, 2);
>
> /* arg1: lea rdi, [rbp - stack_size] */
> - EMIT4(0x48, 0x8D, 0x7D, -stack_size);
> + EMIT3_off32(0x48, 0x8D, 0xBD, -stack_size);
> /* arg2: progs[i]->insnsi for interpreter */
> if (!p->jited)
> emit_mov_imm64(&prog, BPF_REG_2,
> @@ -1984,7 +2031,7 @@ static int invoke_bpf_prog(const struct btf_func_model *m, u8 **pprog,
> /* arg2: mov rsi, rbx <- start time in nsec */
> emit_mov_reg(&prog, true, BPF_REG_2, BPF_REG_6);
> /* arg3: lea rdx, [rbp - run_ctx_off] */
> - EMIT4(0x48, 0x8D, 0x55, -run_ctx_off);
> + EMIT3_off32(0x48, 0x8D, 0x95, -run_ctx_off);
> if (emit_rsb_call(&prog, bpf_trampoline_exit(p), prog))
> return -EINVAL;
>
> @@ -2136,7 +2183,7 @@ int arch_prepare_bpf_trampoline(struct bpf_tramp_image *im, void *image, void *i
> void *func_addr)
> {
> int i, ret, nr_regs = m->nr_args, stack_size = 0;
> - int regs_off, nregs_off, ip_off, run_ctx_off;
> + int regs_off, nregs_off, ip_off, run_ctx_off, arg_stack_off, rbx_off;
> struct bpf_tramp_links *fentry = &tlinks[BPF_TRAMP_FENTRY];
> struct bpf_tramp_links *fexit = &tlinks[BPF_TRAMP_FEXIT];
> struct bpf_tramp_links *fmod_ret = &tlinks[BPF_TRAMP_MODIFY_RETURN];
> @@ -2150,8 +2197,10 @@ int arch_prepare_bpf_trampoline(struct bpf_tramp_image *im, void *image, void *i
> if (m->arg_flags[i] & BTF_FMODEL_STRUCT_ARG)
> nr_regs += (m->arg_size[i] + 7) / 8 - 1;
>
> - /* x86-64 supports up to 6 arguments. 7+ can be added in the future */
> - if (nr_regs > 6)
> + /* x86-64 supports up to MAX_BPF_FUNC_ARGS arguments. 1-6
> + * are passed through regs, the remains are through stack.
> + */
> + if (nr_regs > MAX_BPF_FUNC_ARGS)
> return -ENOTSUPP;
>
> /* Generated trampoline stack layout:
> @@ -2170,7 +2219,14 @@ int arch_prepare_bpf_trampoline(struct bpf_tramp_image *im, void *image, void *i
> *
> * RBP - ip_off [ traced function ] BPF_TRAMP_F_IP_ARG flag
> *
> + * RBP - rbx_off [ rbx value ] always
> + *
That is the case already and we just didn't document it, right?
> * RBP - run_ctx_off [ bpf_tramp_run_ctx ]
> + *
> + * [ stack_argN ] BPF_TRAMP_F_CALL_ORIG
> + * [ ... ]
> + * [ stack_arg2 ]
> + * RBP - arg_stack_off [ stack_arg1 ]
> */
>
> /* room for return value of orig_call or fentry prog */
> @@ -2190,9 +2246,17 @@ int arch_prepare_bpf_trampoline(struct bpf_tramp_image *im, void *image, void *i
>
> ip_off = stack_size;
>
> + stack_size += 8;
> + rbx_off = stack_size;
> +
> stack_size += (sizeof(struct bpf_tramp_run_ctx) + 7) & ~0x7;
> run_ctx_off = stack_size;
>
> + if (nr_regs > 6 && (flags & BPF_TRAMP_F_CALL_ORIG))
> + stack_size += (nr_regs - 6) * 8;
> +
> + arg_stack_off = stack_size;
> +
> if (flags & BPF_TRAMP_F_SKIP_FRAME) {
> /* skip patched call instruction and point orig_call to actual
> * body of the kernel function.
> @@ -2212,8 +2276,9 @@ int arch_prepare_bpf_trampoline(struct bpf_tramp_image *im, void *image, void *i
> x86_call_depth_emit_accounting(&prog, NULL);
> EMIT1(0x55); /* push rbp */
> EMIT3(0x48, 0x89, 0xE5); /* mov rbp, rsp */
> - EMIT4(0x48, 0x83, 0xEC, stack_size); /* sub rsp, stack_size */
> - EMIT1(0x53); /* push rbx */
> + EMIT3_off32(0x48, 0x81, 0xEC, stack_size); /* sub rsp, stack_size */
> + /* mov QWORD PTR [rbp - rbx_off], rbx */
> + emit_stx(&prog, BPF_DW, BPF_REG_FP, BPF_REG_6, -rbx_off);
>
> /* Store number of argument registers of the traced function:
> * mov rax, nr_regs
> @@ -2262,6 +2327,7 @@ int arch_prepare_bpf_trampoline(struct bpf_tramp_image *im, void *image, void *i
>
> if (flags & BPF_TRAMP_F_CALL_ORIG) {
> restore_regs(m, &prog, nr_regs, regs_off);
> + prepare_origin_stack(m, &prog, nr_regs, arg_stack_off);
>
> if (flags & BPF_TRAMP_F_ORIG_STACK) {
> emit_ldx(&prog, BPF_DW, BPF_REG_0, BPF_REG_FP, 8);
> @@ -2321,14 +2387,14 @@ int arch_prepare_bpf_trampoline(struct bpf_tramp_image *im, void *image, void *i
> if (save_ret)
> emit_ldx(&prog, BPF_DW, BPF_REG_0, BPF_REG_FP, -8);
>
> - EMIT1(0x5B); /* pop rbx */
> + emit_ldx(&prog, BPF_DW, BPF_REG_6, BPF_REG_FP, -rbx_off);
It can stay as 'pop', no?
> EMIT1(0xC9); /* leave */
> if (flags & BPF_TRAMP_F_SKIP_FRAME)
> /* skip our return address and return to parent */
> EMIT4(0x48, 0x83, 0xC4, 8); /* add rsp, 8 */
> emit_return(&prog, prog);
> /* Make sure the trampoline generation logic doesn't overflow */
> - if (WARN_ON_ONCE(prog > (u8 *)image_end - BPF_INSN_SAFETY)) {
> + if (prog > (u8 *)image_end - BPF_INSN_SAFETY) {
> ret = -EFAULT;
> goto cleanup;
> }
> --
> 2.40.1
>
