Skip to contentSkip to conversation

bpf: expose socket storage to BPF_PROG_TYPE_CGROUP_SOCK

PATCH In review Stanislav Fomichev 3 participants 6 messages Jul 29, 2020

Subsystems (6)
lore
Share & tools
lore
#0 lore raw

PATCH 1/2 bpf: expose socket storage to BPF_PROG_TYPE_CGROUP_SOCK

This lets us use socket storage from the following hooks:
* BPF_CGROUP_INET_SOCK_CREATE
* BPF_CGROUP_INET_SOCK_RELEASE
* BPF_CGROUP_INET4_POST_BIND
* BPF_CGROUP_INET6_POST_BIND

Using existing 'bpf_sk_storage_get_proto' doesn't work because second argument is ARG_PTR_TO_SOCKET. Even though BPF_PROG_TYPE_CGROUP_SOCK hooks operate on 'struct bpf_sock', the verifier still considers it as a PTR_TO_CTX.
That's why I'm adding another 'bpf_sk_storage_get_cg_sock_proto' definition strictly for BPF_PROG_TYPE_CGROUP_SOCK which accepts ARG_PTR_TO_CTX which is really 'struct sock' for this program type.

---
 net/core/bpf_sk_storage.c | 10 ++++++++++
 net/core/filter.c         |  3 +++
 2 files changed, 13 insertions(+)
net/core/bpf_sk_storage.c +10 -0
index eafcd15e7dfd..d3377c90a291 100644
@@ -944,6 +944,16 @@ const struct bpf_func_proto bpf_sk_storage_get_proto = {
 	.arg4_type	= ARG_ANYTHING,
 };
 
+const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto = {
+	.func		= bpf_sk_storage_get,
+	.gpl_only	= false,
+	.ret_type	= RET_PTR_TO_MAP_VALUE_OR_NULL,
+	.arg1_type	= ARG_CONST_MAP_PTR,
+	.arg2_type	= ARG_PTR_TO_CTX, /* context is 'struct sock' */
+	.arg3_type	= ARG_PTR_TO_MAP_VALUE_OR_NULL,
+	.arg4_type	= ARG_ANYTHING,
+};
+
 const struct bpf_func_proto bpf_sk_storage_delete_proto = {
 	.func		= bpf_sk_storage_delete,
 	.gpl_only	= false,
net/core/filter.c +3 -0
index 29e3455122f7..7124f0fe6974 100644
@@ -6187,6 +6187,7 @@ bool bpf_helper_changes_pkt_data(void *func)
 }
 
 const struct bpf_func_proto bpf_event_output_data_proto __weak;
+const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
 
 static const struct bpf_func_proto *
 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
@@ -6219,6 +6220,8 @@ sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
 	case BPF_FUNC_get_cgroup_classid:
 		return &bpf_get_cgroup_classid_curr_proto;
 #endif
+	case BPF_FUNC_sk_storage_get:
+		return &bpf_sk_storage_get_cg_sock_proto;
 	default:
 		return bpf_base_func_proto(func_id);
 	}
Signature · 1 line
2.28.0.163.g6104cc2f0b6-goog
#1 lore raw

PATCH 2/2 selftests/bpf: verify socket storage in cgroup/sock_{create,release}

Augment udp_limit test to set and verify socket storage value. That should be enough to exercise the changes from the previous patch.

---
 tools/testing/selftests/bpf/progs/udp_limit.c | 19 +++++++++++++++++++
 1 file changed, 19 insertions(+)
tools/testing/selftests/bpf/progs/udp_limit.c +19 -0
index 8429b22525a7..165e3c2dd9a3 100644
@@ -6,14 +6,28 @@
 
 int invocations = 0, in_use = 0;
 
+struct {
+	__uint(type, BPF_MAP_TYPE_SK_STORAGE);
+	__uint(map_flags, BPF_F_NO_PREALLOC);
+	__type(key, int);
+	__type(value, int);
+} sk_map SEC(".maps");
+
 SEC("cgroup/sock_create")
 int sock(struct bpf_sock *ctx)
 {
+	int *sk_storage;
 	__u32 key;
 
 	if (ctx->type != SOCK_DGRAM)
 		return 1;
 
+	sk_storage = bpf_sk_storage_get(&sk_map, ctx, 0,
+					BPF_SK_STORAGE_GET_F_CREATE);
+	if (!sk_storage)
+		return 0;
+	*sk_storage = 0xdeadbeef;
+
 	__sync_fetch_and_add(&invocations, 1);
 
 	if (in_use > 0) {
@@ -31,11 +45,16 @@ int sock(struct bpf_sock *ctx)
 SEC("cgroup/sock_release")
 int sock_release(struct bpf_sock *ctx)
 {
+	int *sk_storage;
 	__u32 key;
 
 	if (ctx->type != SOCK_DGRAM)
 		return 1;
 
+	sk_storage = bpf_sk_storage_get(&sk_map, ctx, 0, 0);
+	if (!sk_storage || *sk_storage != 0xdeadbeef)
+		return 0;
+
 	__sync_fetch_and_add(&invocations, 1);
 	__sync_fetch_and_add(&in_use, -1);
 	return 1;
Signature · 1 line
2.28.0.163.g6104cc2f0b6-goog
reviewer Acked-by on v1 1/2
#3 lore raw
In reply to Stanislav Fomichev
Quoted 15 lines · Stanislav Fomichev

This lets us use socket storage from the following hooks:
* BPF_CGROUP_INET_SOCK_CREATE
* BPF_CGROUP_INET_SOCK_RELEASE
* BPF_CGROUP_INET4_POST_BIND
* BPF_CGROUP_INET6_POST_BIND

Using existing 'bpf_sk_storage_get_proto' doesn't work because second argument is ARG_PTR_TO_SOCKET. Even though BPF_PROG_TYPE_CGROUP_SOCK hooks operate on 'struct bpf_sock', the verifier still considers it as a PTR_TO_CTX.
That's why I'm adding another 'bpf_sk_storage_get_cg_sock_proto' definition strictly for BPF_PROG_TYPE_CGROUP_SOCK which accepts ARG_PTR_TO_CTX which is really 'struct sock' for this program type.

Quoted 2 lines
maintainer
#4 lore raw
In reply to Stanislav Fomichev
Quoted 14 lines · Stanislav Fomichev

This lets us use socket storage from the following hooks:
* BPF_CGROUP_INET_SOCK_CREATE
* BPF_CGROUP_INET_SOCK_RELEASE
* BPF_CGROUP_INET4_POST_BIND
* BPF_CGROUP_INET6_POST_BIND

Using existing 'bpf_sk_storage_get_proto' doesn't work because second argument is ARG_PTR_TO_SOCKET. Even though BPF_PROG_TYPE_CGROUP_SOCK hooks operate on 'struct bpf_sock', the verifier still considers it as a PTR_TO_CTX.
That's why I'm adding another 'bpf_sk_storage_get_cg_sock_proto' definition strictly for BPF_PROG_TYPE_CGROUP_SOCK which accepts ARG_PTR_TO_CTX which is really 'struct sock' for this program type.

Quoted 2 lines

Makes sense, both applied, thanks!

[...]

Quoted 13 lines
net/core/filter.c +1 -0
index 29e3455122f7..7124f0fe6974 100644
@@ -6187,6 +6187,7 @@ bool bpf_helper_changes_pkt_data(void *func)
  }
 
  const struct bpf_func_proto bpf_event_output_data_proto __weak;
+const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
 
  static const struct bpf_func_proto *
  sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)

net/core/filter.c sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) · line 6224

Quoted 5 lines
net/core/filter.c +2 -0
@@ -6219,6 +6220,8 @@ sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
  	case BPF_FUNC_get_cgroup_classid:
  		return &bpf_get_cgroup_classid_curr_proto;
  #endif
+	case BPF_FUNC_sk_storage_get:
+		return &bpf_sk_storage_get_cg_sock_proto;

Been wondering whether we need these for connect/sendmsg/etc hooks that operate on sock_addr, but for those we have them already covered in sock_addr_func_proto() therefore all good.

sock_addr_func_proto() also lists the BPF_FUNC_sk_storage_delete. Should we add that one as well for sock_filter_func_proto()? Presumably create/release doesn't make sense, but any use case for bind hook?

Quoted 4 lines
net/core/filter.c +0 -0
  	default:
  		return bpf_base_func_proto(func_id);
  	}
#5 lore raw
Quoted 1 line · Daniel Borkmann

On 7/29/20 2:31 AM, Stanislav Fomichev wrote:

[..]

Quoted 3 lines

sock_addr_func_proto() also lists the BPF_FUNC_sk_storage_delete. Should we add that one as well for sock_filter_func_proto()? Presumably create/release doesn't make sense, but any use case for bind hook?

Right, I didn't think delete makes sense for create/release, but maybe it does for post_bind :-/
Let me put it on my list, I'll follow up!