For now, ctx[-1] is used to store the nr_args in the trampoline. However,
1-byte is enough to store such information. Therefore, we use only the
last byte of ctx[-1] to store the nr_args, and reserve the rest for other
usages.

Signed-off-by: Menglong Dong <dongml2@chinatelecom.cn>
---
 kernel/bpf/verifier.c    | 35 +++++++++++++++++++----------------
 kernel/trace/bpf_trace.c |  4 ++--
 2 files changed, 21 insertions(+), 18 deletions(-)

diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index 774c9b0aafa3..bfff3f84fd91 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -23277,15 +23277,16 @@ static int do_misc_fixups(struct bpf_verifier_env *env)
 		    insn->imm == BPF_FUNC_get_func_arg) {
 			/* Load nr_args from ctx - 8 */
 			insn_buf[0] = BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_1, -8);
-			insn_buf[1] = BPF_JMP32_REG(BPF_JGE, BPF_REG_2, BPF_REG_0, 6);
-			insn_buf[2] = BPF_ALU64_IMM(BPF_LSH, BPF_REG_2, 3);
-			insn_buf[3] = BPF_ALU64_REG(BPF_ADD, BPF_REG_2, BPF_REG_1);
-			insn_buf[4] = BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_2, 0);
-			insn_buf[5] = BPF_STX_MEM(BPF_DW, BPF_REG_3, BPF_REG_0, 0);
-			insn_buf[6] = BPF_MOV64_IMM(BPF_REG_0, 0);
-			insn_buf[7] = BPF_JMP_A(1);
-			insn_buf[8] = BPF_MOV64_IMM(BPF_REG_0, -EINVAL);
-			cnt = 9;
+			insn_buf[1] = BPF_ALU64_IMM(BPF_AND, BPF_REG_0, 0xFF);
+			insn_buf[2] = BPF_JMP32_REG(BPF_JGE, BPF_REG_2, BPF_REG_0, 6);
+			insn_buf[3] = BPF_ALU64_IMM(BPF_LSH, BPF_REG_2, 3);
+			insn_buf[4] = BPF_ALU64_REG(BPF_ADD, BPF_REG_2, BPF_REG_1);
+			insn_buf[5] = BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_2, 0);
+			insn_buf[6] = BPF_STX_MEM(BPF_DW, BPF_REG_3, BPF_REG_0, 0);
+			insn_buf[7] = BPF_MOV64_IMM(BPF_REG_0, 0);
+			insn_buf[8] = BPF_JMP_A(1);
+			insn_buf[9] = BPF_MOV64_IMM(BPF_REG_0, -EINVAL);
+			cnt = 10;
 
 			new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt);
 			if (!new_prog)
@@ -23305,12 +23306,13 @@ static int do_misc_fixups(struct bpf_verifier_env *env)
 			    eatype == BPF_MODIFY_RETURN) {
 				/* Load nr_args from ctx - 8 */
 				insn_buf[0] = BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_1, -8);
-				insn_buf[1] = BPF_ALU64_IMM(BPF_LSH, BPF_REG_0, 3);
-				insn_buf[2] = BPF_ALU64_REG(BPF_ADD, BPF_REG_0, BPF_REG_1);
-				insn_buf[3] = BPF_LDX_MEM(BPF_DW, BPF_REG_3, BPF_REG_0, 0);
-				insn_buf[4] = BPF_STX_MEM(BPF_DW, BPF_REG_2, BPF_REG_3, 0);
-				insn_buf[5] = BPF_MOV64_IMM(BPF_REG_0, 0);
-				cnt = 6;
+				insn_buf[1] = BPF_ALU64_IMM(BPF_AND, BPF_REG_0, 0xFF);
+				insn_buf[2] = BPF_ALU64_IMM(BPF_LSH, BPF_REG_0, 3);
+				insn_buf[3] = BPF_ALU64_REG(BPF_ADD, BPF_REG_0, BPF_REG_1);
+				insn_buf[4] = BPF_LDX_MEM(BPF_DW, BPF_REG_3, BPF_REG_0, 0);
+				insn_buf[5] = BPF_STX_MEM(BPF_DW, BPF_REG_2, BPF_REG_3, 0);
+				insn_buf[6] = BPF_MOV64_IMM(BPF_REG_0, 0);
+				cnt = 7;
 			} else {
 				insn_buf[0] = BPF_MOV64_IMM(BPF_REG_0, -EOPNOTSUPP);
 				cnt = 1;
@@ -23331,8 +23333,9 @@ static int do_misc_fixups(struct bpf_verifier_env *env)
 		    insn->imm == BPF_FUNC_get_func_arg_cnt) {
 			/* Load nr_args from ctx - 8 */
 			insn_buf[0] = BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_1, -8);
+			insn_buf[1] = BPF_ALU64_IMM(BPF_AND, BPF_REG_0, 0xFF);
 
-			new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, 1);
+			new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, 2);
 			if (!new_prog)
 				return -ENOMEM;
 
diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c
index 6e076485bf70..6b58f9a4dc92 100644
--- a/kernel/trace/bpf_trace.c
+++ b/kernel/trace/bpf_trace.c
@@ -1194,7 +1194,7 @@ const struct bpf_func_proto bpf_get_branch_snapshot_proto = {
 BPF_CALL_3(get_func_arg, void *, ctx, u32, n, u64 *, value)
 {
 	/* This helper call is inlined by verifier. */
-	u64 nr_args = ((u64 *)ctx)[-1];
+	u64 nr_args = ((u64 *)ctx)[-1] & 0xFF;
 
 	if ((u64) n >= nr_args)
 		return -EINVAL;
@@ -1214,7 +1214,7 @@ static const struct bpf_func_proto bpf_get_func_arg_proto = {
 BPF_CALL_2(get_func_ret, void *, ctx, u64 *, value)
 {
 	/* This helper call is inlined by verifier. */
-	u64 nr_args = ((u64 *)ctx)[-1];
+	u64 nr_args = ((u64 *)ctx)[-1] & 0xFF;
 
 	*value = ((u64 *)ctx)[nr_args];
 	return 0;
-- 
2.52.0

> diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c
> index 6e076485bf70..6b58f9a4dc92 100644
> --- a/kernel/trace/bpf_trace.c
> +++ b/kernel/trace/bpf_trace.c

[ ... ]

> @@ -1194,7 +1194,7 @@ const struct bpf_func_proto bpf_get_branch_snapshot_proto = {
>  BPF_CALL_3(get_func_arg, void *, ctx, u32, n, u64 *, value)
>  {
>  	/* This helper call is inlined by verifier. */
> -	u64 nr_args = ((u64 *)ctx)[-1];
> +	u64 nr_args = ((u64 *)ctx)[-1] & 0xFF;

[ ... ]

> @@ -1214,7 +1214,7 @@ static const struct bpf_func_proto bpf_get_func_arg_proto = {
>  BPF_CALL_2(get_func_ret, void *, ctx, u64 *, value)
>  {
>  	/* This helper call is inlined by verifier. */
> -	u64 nr_args = ((u64 *)ctx)[-1];
> +	u64 nr_args = ((u64 *)ctx)[-1] & 0xFF;
>
>  	*value = ((u64 *)ctx)[nr_args];
>  	return 0;

Both get_func_arg() and get_func_ret() are updated with the 0xFF mask, but
should get_func_arg_cnt() also receive the same treatment?

Current code in get_func_arg_cnt():

    BPF_CALL_1(get_func_arg_cnt, void *, ctx)
    {
        /* This helper call is inlined by verifier. */
        return ((u64 *)ctx)[-1];
    }

The comment says "inlined by verifier", but the verifier only inlines this
helper for BPF_PROG_TYPE_TRACING programs. For BPF_PROG_TYPE_LSM programs
with BPF_LSM_MAC attachment type, bpf_prog_has_trampoline() returns true,
allowing access to this helper via tracing_prog_func_proto(). However,
do_misc_fixups() only inlines for BPF_PROG_TYPE_TRACING, so LSM programs
would call this non-inlined fallback function.

If subsequent patches store metadata in the upper bits of ctx[-1], could
LSM programs calling bpf_get_func_arg_cnt() receive incorrect values?

Later commits in this series (93d4641b9513 "bpf,x86: add fsession support
for x86_64") do use the upper bits for BPF_TRAMP_M_IS_RETURN and
BPF_TRAMP_M_COOKIE flags. This regression is not fixed in the series.


---
AI reviewed your patch. Please fix the bug or email reply why it's not a bug.
See: https://github.com/kernel-patches/vmtest/blob/master/ci/claude/README.md

CI run summary: https://github.com/kernel-patches/bpf/actions/runs/20773233136