Skip to contentSkip to conversation

bpf: Enable bpf_skc_to_* sock casting helper to networking prog type

RFC PATCH Superseded Martin KaFai Lau 1 participant 3 messages Sep 15, 2020

Subsystems (5)

v2 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

v2 cover bpf: Enable bpf_skc_to_* sock casting helper to networking prog type

This set allows networking prog type to directly read fields from the in-kernel socket type, e.g. "struct tcp_sock".

Patch 2 has the details on the use case.

It is currently under RFC since it needs proper tests and it builds on top of Lorenz's patches:
https://lore.kernel.org/bpf/[email protected]/

v2:
- Add ARG_PTR_TO_SOCK_COMMON_OR_NULL (Lorenz)

Martin KaFai Lau (2):
bpf: Move the PTR_TO_BTF_ID check to check_reg_type()
bpf: Enable bpf_skc_to_* sock casting helper to networking prog type

 include/linux/bpf.h   |  1 +
 kernel/bpf/verifier.c | 97 +++++++++++++++++++++++++++++--------------
 net/core/filter.c     | 73 ++++++++++++++++++++++----------
 3 files changed, 117 insertions(+), 54 deletions(-)
Signature · 1 line
2.24.1
#1 lore raw

RFC PATCH v2 1/2 bpf: Move the PTR_TO_BTF_ID check to check_reg_type()

check_reg_type() checks whether a reg can be used as an arg of a func_proto. For PTR_TO_BTF_ID, the check is actually not completely done until the reg->btf_id is pointing to a kernel struct that is acceptable by the func_proto.

Thus, this patch moves the btf_id check into check_reg_type(). The compatible_reg_types[] usage is localized in check_reg_type() now which I found it easier to reason in the next patch.

The "if (!btf_id) verbose(...); " is removed for now since it won't happen. It will be added back in the next patch with new error log specific to mis-configured compatible_reg_types[].

---
 kernel/bpf/verifier.c | 65 +++++++++++++++++++++++--------------------
 1 file changed, 35 insertions(+), 30 deletions(-)
kernel/bpf/verifier.c +35 -30
index 8fdef5656e7f..3a5932bd7c22 100644
@@ -3981,20 +3981,29 @@ static const struct bpf_reg_types *compatible_reg_types[] = {
 	[__BPF_ARG_TYPE_MAX]		= NULL,
 };
 
-static int check_reg_type(struct bpf_verifier_env *env, u32 regno,
-			  const struct bpf_reg_types *compatible)
+static int check_reg_type(struct bpf_verifier_env *env, u32 arg,
+			  enum bpf_arg_type arg_type,
+			  const struct bpf_func_proto *fn)
 {
+	u32 regno = BPF_REG_1 + arg;
 	struct bpf_reg_state *regs = cur_regs(env), *reg = &regs[regno];
 	enum bpf_reg_type expected, type = reg->type;
+	const struct bpf_reg_types *compatible;
 	int i, j;
 
+	compatible = compatible_reg_types[arg_type];
+	if (!compatible) {
+		verbose(env, "verifier internal error: unsupported arg type %d\n", arg_type);
+		return -EFAULT;
+	}
+
 	for (i = 0; i < ARRAY_SIZE(compatible->types); i++) {
 		expected = compatible->types[i];
 		if (expected == NOT_INIT)
 			break;
 
 		if (type == expected)
-			return 0;
+			goto found;
 	}
 
 	verbose(env, "R%d type=%s expected=", regno, reg_type_str[type]);
@@ -4002,6 +4011,27 @@ static int check_reg_type(struct bpf_verifier_env *env, u32 regno,
 		verbose(env, "%s, ", reg_type_str[compatible->types[j]]);
 	verbose(env, "%s\n", reg_type_str[compatible->types[j]]);
 	return -EACCES;
+
+found:
+	if (type == PTR_TO_BTF_ID) {
+		u32 *expected_btf_id = fn->arg_btf_id[arg];
+
+		if (!btf_struct_ids_match(&env->log, reg->off, reg->btf_id,
+					  *expected_btf_id)) {
+			verbose(env, "R%d is a pointer to in-kernel struct %s but %s is expected instead\n",
+				regno, kernel_type_name(reg->btf_id),
+				kernel_type_name(*expected_btf_id));
+			return -EACCES;
+		}
+
+		if (!tnum_is_const(reg->var_off) || reg->var_off.value) {
+			verbose(env, "R%d is a pointer to in-kernel struct with non-zero offset\n",
+				regno);
+			return -EACCES;
+		}
+	}
+
+	return 0;
 }
 
 static int check_func_arg(struct bpf_verifier_env *env, u32 arg,
@@ -4011,7 +4041,6 @@ static int check_func_arg(struct bpf_verifier_env *env, u32 arg,
 	u32 regno = BPF_REG_1 + arg;
 	struct bpf_reg_state *regs = cur_regs(env), *reg = &regs[regno];
 	enum bpf_arg_type arg_type = fn->arg_type[arg];
-	const struct bpf_reg_types *compatible;
 	enum bpf_reg_type type = reg->type;
 	int err = 0;
 
@@ -4051,35 +4080,11 @@ static int check_func_arg(struct bpf_verifier_env *env, u32 arg,
 		 */
 		goto skip_type_check;
 
-	compatible = compatible_reg_types[arg_type];
-	if (!compatible) {
-		verbose(env, "verifier internal error: unsupported arg type %d\n", arg_type);
-		return -EFAULT;
-	}
-
-	err = check_reg_type(env, regno, compatible);
+	err = check_reg_type(env, arg, arg_type, fn);
 	if (err)
 		return err;
 
-	if (type == PTR_TO_BTF_ID) {
-		const u32 *btf_id = fn->arg_btf_id[arg];
-
-		if (!btf_id) {
-			verbose(env, "verifier internal error: missing BTF ID\n");
-			return -EFAULT;
-		}
-
-		if (!btf_struct_ids_match(&env->log, reg->off, reg->btf_id, *btf_id)) {
-			verbose(env, "R%d has incompatible type %s\n", regno,
-				kernel_type_name(reg->btf_id));
-			return -EACCES;
-		}
-		if (!tnum_is_const(reg->var_off) || reg->var_off.value) {
-			verbose(env, "R%d is a pointer to in-kernel struct with non-zero offset\n",
-				regno);
-			return -EACCES;
-		}
-	} else if (type == PTR_TO_CTX) {
+	if (type == PTR_TO_CTX) {
 		err = check_ctx_reg(env, reg, regno);
 		if (err < 0)
 			return err;
Signature · 1 line
2.24.1
#2 lore raw

RFC PATCH v2 2/2 bpf: Enable bpf_skc_to_* sock casting helper to networking prog type

In reply to Martin KaFai Lau

There is a constant need to add more fields into the bpf_tcp_sock for the bpf programs running at tc, sock_ops...etc.

A current workaround could be to use bpf_probe_read_kernel(). However, other than making another helper call for reading each field and missing CO-RE, it is also not as intuitive to use as directly reading "tp->lsndtime" for example. While already having perfmon cap to do bpf_probe_read_kernel(), it will be much easier if the bpf prog can directly read from the tcp_sock.

This patch tries to do that by using the existing casting-helpers bpf_skc_to_*() whose func_proto returns a btf_id. For example, the func_proto of bpf_skc_to_tcp_sock returns the btf_id of the kernel "struct tcp_sock".

This patch modifies the existing bpf_skc_to_* func_proto to take ARG_PTR_TO_SOCK_COMMON_OR_NULL instead of taking "ARG_PTR_TO_BTF_ID + &btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON]". That will allow tc, sock_ops,...etc to call these casting helpers because they already hold the PTR_TO_SOCK_COMMON (or its equivalent). For example:

	sk = sock_ops->sk;
	if (!sk)
		return;
	tp = bpf_skc_to_tcp_sock(sk);
	if (!tp)
		return;
	/* Read tp as a PTR_TO_BTF_ID */
	lsndtime = tp->lsndtime;

To ensure the current bpf prog passing a PTR_TO_BTF_ID to bpf_skc_to_*() still works as is, the verifier is modified such that ARG_PTR_TO_SOCK_COMMON_OR_NULL can accept a reg with reg->type == PTR_TO_BTF_ID and reg->btf_id is btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON]

To do that, an idea is borrowed from one of the Lorenz's patch:
https://lore.kernel.org/bpf/[email protected]/ . It adds PTR_TO_BTF_ID as one of the acceptable reg->type for ARG_PTR_TO_SOCK_COMMON_OR_NULL and also specifies what btf_id it can take. By doing this, the bpf_skc_to_* will work as before and can still take PTR_TO_BTF_ID as the arg. e.g. The bpf tcp iter will work as is.

ARG_PTR_TO_SOCK_COMMON_OR_NULL is newly added. The existing helpers have to change to take this new arg type in order to work with the PTR_TO_BTF_ID reg. OR_NULL is needed because PTR_TO_BTF_ID could be NULL. For example,

	req_sk = bpf_skc_to_tcp_request_sock(sk);
	/* req_sk->sk is a PTR_TO_BTF_ID and it could be NULL */
	bpf_sk_cgroup_id(req_sk->sk);

[ Note:
To handle PTR_TO_BTF_ID that could be NULL,
one considered option is to introduce PTR_TO_BTF_ID_NOFAULT which will be used in the return value of the bpf_skc_to_*() helpers. It is because the helper has verified the pointer can access the whole struct described by the btf_id and NULL check must have been done by the program. However, the above req_sk->sk example will still have a PTR_TO_BTF_ID which may be NULL. ]

bpf_sk_release() is changed in this patch to take ARG_PTR_TO_SOCK_COMMON_OR_NULL such that the following will work:

 	sk = bpf_skc_lookup_tcp(skb, tuple, tuplen, BPF_F_CURRENT_NETNS, 0);
	if (!sk)
		return;
	tp = bpf_skc_to_tcp_sock(sk);
	if (!tp) {
		bpf_sk_release(sk);
		return;
	}
	lsndtime = tp->lsndtime;
	/* Pass tp to bpf_sk_release() will also work */
	bpf_sk_release(tp);

Other helpers will be changed to take ARG_PTR_TO_SOCK_COMMONN_OR_NULL later.

The BPF_FUNC_skc_to_* func_id is added to is_ptr_cast_function(). It ensures the returning reg (BPF_REF_0) which is a PTR_TO_BTF_ID_OR_NULL also carries the ref_obj_id. That will keep the ref-tracking works properly.

The bpf_skc_to_* helpers are made available to most of the bpf prog types in filter.c. They are limited by perfmon cap.

---
 include/linux/bpf.h   |  1 +
 kernel/bpf/verifier.c | 32 +++++++++++++++++--
 net/core/filter.c     | 73 ++++++++++++++++++++++++++++++-------------
 3 files changed, 82 insertions(+), 24 deletions(-)
include/linux/bpf.h +1 -0
index 7cf7e90f55bb..2e9baeaa6dae 100644
@@ -284,6 +284,7 @@ enum bpf_arg_type {
 	ARG_ANYTHING,		/* any (initialized) argument is ok */
 	ARG_PTR_TO_SPIN_LOCK,	/* pointer to bpf_spin_lock */
 	ARG_PTR_TO_SOCK_COMMON,	/* pointer to sock_common */
+	ARG_PTR_TO_SOCK_COMMON_OR_NULL,	/* pointer to sock_common or NULL */
 	ARG_PTR_TO_INT,		/* pointer to int */
 	ARG_PTR_TO_LONG,	/* pointer to long */
 	ARG_PTR_TO_SOCKET,	/* pointer to bpf_sock (fullsock) */
kernel/bpf/verifier.c +30 -2
index 3a5932bd7c22..a88841ed17a7 100644
@@ -432,7 +432,8 @@ static bool reg_type_may_be_refcounted_or_null(enum bpf_reg_type type)
 
 static bool arg_type_may_be_refcounted(enum bpf_arg_type type)
 {
-	return type == ARG_PTR_TO_SOCK_COMMON;
+	return type == ARG_PTR_TO_SOCK_COMMON ||
+		type == ARG_PTR_TO_SOCK_COMMON_OR_NULL;
 }
 
 static bool arg_type_may_be_null(enum bpf_arg_type type)
@@ -440,6 +441,7 @@ static bool arg_type_may_be_null(enum bpf_arg_type type)
 	return type == ARG_PTR_TO_MAP_VALUE_OR_NULL ||
 	       type == ARG_PTR_TO_MEM_OR_NULL ||
 	       type == ARG_PTR_TO_CTX_OR_NULL ||
+	       type == ARG_PTR_TO_SOCK_COMMON_OR_NULL ||
 	       type == ARG_PTR_TO_SOCKET_OR_NULL ||
 	       type == ARG_PTR_TO_ALLOC_MEM_OR_NULL;
 }
@@ -486,7 +488,12 @@ static bool is_acquire_function(enum bpf_func_id func_id,
 static bool is_ptr_cast_function(enum bpf_func_id func_id)
 {
 	return func_id == BPF_FUNC_tcp_sock ||
-		func_id == BPF_FUNC_sk_fullsock;
+		func_id == BPF_FUNC_sk_fullsock ||
+		func_id == BPF_FUNC_skc_to_tcp_sock ||
+		func_id == BPF_FUNC_skc_to_tcp6_sock ||
+		func_id == BPF_FUNC_skc_to_udp6_sock ||
+		func_id == BPF_FUNC_skc_to_tcp_timewait_sock ||
+		func_id == BPF_FUNC_skc_to_tcp_request_sock;
 }
 
 /* string representation of 'enum bpf_reg_type' */
@@ -3906,6 +3913,7 @@ static int resolve_map_arg_type(struct bpf_verifier_env *env,
 
 struct bpf_reg_types {
 	const enum bpf_reg_type types[10];
+	u32 *btf_id;
 };
 
 static const struct bpf_reg_types map_key_value_types = {
@@ -3926,6 +3934,17 @@ static const struct bpf_reg_types sock_types = {
 	},
 };
 
+static const struct bpf_reg_types sock_or_null_types = {
+	.types = {
+		PTR_TO_SOCK_COMMON,
+		PTR_TO_SOCKET,
+		PTR_TO_TCP_SOCK,
+		PTR_TO_XDP_SOCK,
+		PTR_TO_BTF_ID,
+	},
+	.btf_id = &btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON],
+};
+
 static const struct bpf_reg_types mem_types = {
 	.types = {
 		PTR_TO_STACK,
@@ -3967,6 +3986,7 @@ static const struct bpf_reg_types *compatible_reg_types[] = {
 	[ARG_PTR_TO_CTX]		= &context_types,
 	[ARG_PTR_TO_CTX_OR_NULL]	= &context_types,
 	[ARG_PTR_TO_SOCK_COMMON]	= &sock_types,
+	[ARG_PTR_TO_SOCK_COMMON_OR_NULL]= &sock_or_null_types,
 	[ARG_PTR_TO_SOCKET]		= &fullsock_types,
 	[ARG_PTR_TO_SOCKET_OR_NULL]	= &fullsock_types,
 	[ARG_PTR_TO_BTF_ID]		= &btf_ptr_types,
@@ -4016,6 +4036,14 @@ static int check_reg_type(struct bpf_verifier_env *env, u32 arg,
 	if (type == PTR_TO_BTF_ID) {
 		u32 *expected_btf_id = fn->arg_btf_id[arg];
 
+		if (!expected_btf_id) {
+			if (!compatible->btf_id) {
+				 verbose(env, "verifier internal error: missing arg compatible BTF ID\n");
+				 return -EFAULT;
+			}
+			expected_btf_id = compatible->btf_id;
+		}
+
 		if (!btf_struct_ids_match(&env->log, reg->off, reg->btf_id,
 					  *expected_btf_id)) {
 			verbose(env, "R%d is a pointer to in-kernel struct %s but %s is expected instead\n",
net/core/filter.c +51 -22

180 lines not on this page · raw message

Signature · 1 line
2.24.1