bpf: expose socket storage to BPF_PROG_TYPE_CGROUP_SOCK
Subsystems (6)
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
@@ -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
@@ -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
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
@@ -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
Re: [PATCH bpf-next 2/2] selftests/bpf: verify socket storage in cgroup/sock_{create,release}
Quoted 7 lines · Stanislav Fomichev
Augment
udp_limittest to set and verify socket storage value. That should be enough to exercise the changes from the previous patch.Cc Martin KaFai Lau <[email protected]>Signed-off-by Stanislav Fomichev <[email protected]>
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_BINDUsing existing '
bpf_sk_storage_get_proto' doesn't work because second argument isARG_PTR_TO_SOCKET. Even thoughBPF_PROG_TYPE_CGROUP_SOCKhooks operate on 'structbpf_sock', the verifier still considers it as aPTR_TO_CTX.
That's why I'm adding another 'bpf_sk_storage_get_cg_sock_proto' definition strictly forBPF_PROG_TYPE_CGROUP_SOCKwhich acceptsARG_PTR_TO_CTXwhich is really 'struct sock' for this program type.
Quoted 2 lines
Cc Martin KaFai Lau <[email protected]>Signed-off-by Stanislav Fomichev <[email protected]>
Daniel Borkmann
maintainerQuoted 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_BINDUsing existing '
bpf_sk_storage_get_proto' doesn't work because second argument isARG_PTR_TO_SOCKET. Even thoughBPF_PROG_TYPE_CGROUP_SOCKhooks operate on 'structbpf_sock', the verifier still considers it as aPTR_TO_CTX.
That's why I'm adding another 'bpf_sk_storage_get_cg_sock_proto' definition strictly forBPF_PROG_TYPE_CGROUP_SOCKwhich acceptsARG_PTR_TO_CTXwhich is really 'struct sock' for this program type.
Quoted 2 lines
Cc Martin KaFai Lau <[email protected]>Signed-off-by Stanislav Fomichev <[email protected]>
Makes sense, both applied, thanks!
[...]
Quoted 13 lines
net/core/filter.c +1
-0@@ -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
-0default :return bpf_base_func_proto (func_id); }
Quoted 1 line · Daniel Borkmann
On 7/29/20 2:31 AM, Stanislav Fomichev wrote:
[..]
Quoted 3 lines
sock_addr_func_proto()also lists theBPF_FUNC_sk_storage_delete. Should we add that one as well forsock_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!