bpf: expose bpf_{s,g}etsockopt helpers to bind{4,6} hooks
Subsystems (7)
cover bpf: expose bpf_{s,g}etsockopt helpers to bind{4,6} hooks
This might be useful for the listener sockets to pre-populate some options. Since those helpers require locked sockets, I'm changing bind hooks to lock/unlock the sockets. This should not cause any performance overhead because at this point there shouldn't be any socket lock contention and the locking/unlocking should be cheap.
Also, as part of the series, I convert test_sock_addr bpf assembly into C (and preserve the narrow load tests) to make it easier to extend with th bpf_setsockopt later on.
Stanislav Fomichev (3):
selftests/bpf: rewrite test_sock_addr bind bpf into C
bpf: allow bpf_{s,g}etsockopt from cgroup bind{4,6} hooks selftests/bpf: extend bind{4,6} programs with a call to bpf_setsockopt
include/linux/bpf-cgroup.h | 12 +- net/core/filter.c | 4 + net/ipv4/af_inet.c | 2 +- net/ipv6/af_inet6.c | 2 +- .../testing/selftests/bpf/progs/bind4_prog.c | 104 ++++++++++ .../testing/selftests/bpf/progs/bind6_prog.c | 121 +++++++++++ tools/testing/selftests/bpf/test_sock_addr.c | 196 ++---------------- 7 files changed, 249 insertions(+), 192 deletions(-) create mode 100644 tools/testing/selftests/bpf/progs/bind4_prog.c create mode 100644 tools/testing/selftests/bpf/progs/bind6_prog.c
Signature · 1 line
2.29.2.299.gdc1121823c-goog
PATCH 1/3 selftests/bpf: rewrite test_sock_addr bind bpf into C
I'm planning to extend it in the next patches. It's much easier to work with C than BPF assembly.
--- .../testing/selftests/bpf/progs/bind4_prog.c | 73 +++++++ .../testing/selftests/bpf/progs/bind6_prog.c | 90 ++++++++ tools/testing/selftests/bpf/test_sock_addr.c | 196 ++---------------- 3 files changed, 175 insertions(+), 184 deletions(-) create mode 100644 tools/testing/selftests/bpf/progs/bind4_prog.c create mode 100644 tools/testing/selftests/bpf/progs/bind6_prog.c
tools/testing/selftests/bpf/progs/bind4_prog.c new file +73 -0
@@ -0,0 +1,73 @@ +// SPDX-License-Identifier: GPL-2.0 + +#include <string.h> + +#include <linux/stddef.h> +#include <linux/bpf.h> +#include <linux/in.h> +#include <linux/in6.h> +#include <sys/socket.h> +#include <netinet/tcp.h> +#include <linux/if.h> +#include <errno.h> + +#include <bpf/bpf_helpers.h> +#include <bpf/bpf_endian.h> + +#define SERV4_IP 0xc0a801feU/* 192.168.1.254 */ +#define SERV4_PORT 4040 +#define SERV4_REWRITE_IP 0x7f000001U/* 127.0.0.1 */ +#define SERV4_REWRITE_PORT 4444 + +int _version SEC("version") = 1; + +SEC("cgroup/bind4") +int bind_v4_prog (struct bpf_sock_addr *ctx) +{ +struct bpf_sock *sk; +__u32 user_ip4; +__u16 user_port; + + sk = ctx->sk; +if (!sk) +return 0 ; + +if (sk->family !=AF_INET ) +return 0 ; + +if (ctx->type !=SOCK_STREAM && ctx->type !=SOCK_DGRAM ) +return 0 ; + +if (ctx->user_ip4 !=bpf_htonl (SERV4_IP ) || + ctx->user_port !=bpf_htons (SERV4_PORT )) +return 0 ; + +// u8 narrow loads: + user_ip4 =0 ; + user_ip4 |= ((volatile __u8 *)&ctx->user_ip4)[0 ] <<0 ; + user_ip4 |= ((volatile __u8 *)&ctx->user_ip4)[1 ] <<8 ; + user_ip4 |= ((volatile __u8 *)&ctx->user_ip4)[2 ] <<16 ; + user_ip4 |= ((volatile __u8 *)&ctx->user_ip4)[3 ] <<24 ; +if (ctx->user_ip4 != user_ip4) +return 0 ; + + user_port =0 ; + user_port |= ((volatile __u8 *)&ctx->user_port)[0 ] <<0 ; + user_port |= ((volatile __u8 *)&ctx->user_port)[1 ] <<8 ; +if (ctx->user_port != user_port) +return 0 ; + +// u16 narrow loads: + user_ip4 =0 ; + user_ip4 |= ((volatile __u16 *)&ctx->user_ip4)[0 ] <<0 ; + user_ip4 |= ((volatile __u16 *)&ctx->user_ip4)[1 ] <<16 ; +if (ctx->user_ip4 != user_ip4) +return 0 ; + + ctx->user_ip4 =bpf_htonl (SERV4_REWRITE_IP ); + ctx->user_port =bpf_htons (SERV4_REWRITE_PORT ); + +return 1 ; +} + +char _license[] SEC("license") = "GPL";
tools/testing/selftests/bpf/progs/bind6_prog.c new file +90 -0
90 lines not on this page · raw message
tools/testing/selftests/bpf/test_sock_addr.c +12 -184
214 lines not on this page · raw message
Signature · 1 line
2.29.2.299.gdc1121823c-goog
Re: [PATCH bpf-next 1/3] selftests/bpf: rewrite test_sock_addr bind bpf into C
Quoted 6 lines · Stanislav Fomichev
I'm planning to extend it in the next patches. It's much easier to work with C than BPF assembly.
Signed-off-by Stanislav Fomichev <[email protected]>---
With nits below:
Quoted 25 lines
.../testing/selftests/bpf/progs/bind4_prog.c | 73 +++++++ .../testing/selftests/bpf/progs/bind6_prog.c | 90 ++++++++ tools/testing/selftests/bpf/test_sock_addr.c | 196 ++---------------- 3 files changed, 175 insertions(+), 184 deletions(-) create mode 100644 tools/testing/selftests/bpf/progs/bind4_prog.c create mode 100644 tools/testing/selftests/bpf/progs/bind6_prog.ctools/testing/selftests/bpf/progs/bind4_prog.c new file +12
-0@@ -0,0 +1,73 @@ +// SPDX-License-Identifier: GPL-2.0 + +#include <string.h> + +#include <linux/stddef.h> +#include <linux/bpf.h> +#include <linux/in.h> +#include <linux/in6.h> +#include <sys/socket.h> +#include <netinet/tcp.h> +#include <linux/if.h> +#include <errno.h>
tools/testing/selftests/bpf/progs/bind4_prog.c line 22
Quoted 10 lines
tools/testing/selftests/bpf/progs/bind4_prog.c new file +10
-0+ +#include <bpf/bpf_helpers.h> +#include <bpf/bpf_endian.h> + +#define SERV4_IP 0xc0a801feU/* 192.168.1.254 */ +#define SERV4_PORT 4040 +#define SERV4_REWRITE_IP 0x7f000001U/* 127.0.0.1 */ +#define SERV4_REWRITE_PORT 4444 + +int _version SEC("version") = 1;
not needed, let's not add it to a new test prog
Quoted 1 line
tools/testing/selftests/bpf/progs/bind4_prog.c new file +1
-0+
[...]
Quoted 24 lines
tools/testing/selftests/bpf/progs/bind6_prog.c new file +18
-0@@ -0,0 +1,90 @@ +// SPDX-License-Identifier: GPL-2.0 + +#include <string.h> + +#include <linux/stddef.h> +#include <linux/bpf.h> +#include <linux/in.h> +#include <linux/in6.h> +#include <sys/socket.h> +#include <netinet/tcp.h> +#include <linux/if.h> +#include <errno.h> + +#include <bpf/bpf_helpers.h> +#include <bpf/bpf_endian.h> + +#define SERV6_IP_0 0xfaceb00c/* face:b00c:1234:5678::abcd */ +#define SERV6_IP_1 0x12345678
tools/testing/selftests/bpf/progs/bind6_prog.c line 28
Quoted 10 lines
tools/testing/selftests/bpf/progs/bind6_prog.c new file +10
-0+#define SERV6_IP_2 0x00000000 +#define SERV6_IP_3 0x0000abcd +#define SERV6_PORT 6060 +#define SERV6_REWRITE_IP_0 0x00000000 +#define SERV6_REWRITE_IP_1 0x00000000 +#define SERV6_REWRITE_IP_2 0x00000000 +#define SERV6_REWRITE_IP_3 0x00000001 +#define SERV6_REWRITE_PORT 6666 + +int _version SEC("version") = 1;
same
Quoted 4 lines
tools/testing/selftests/bpf/progs/bind6_prog.c new file +4
-0+ +SEC("cgroup/bind6") +int bind_v6_prog (struct bpf_sock_addr *ctx) +{
[...]
Re: [PATCH bpf-next 1/3] selftests/bpf: rewrite test_sock_addr bind bpf into C
Quoted 11 lines · Andrii Nakryiko
Quoted 6 lines · Stanislav Fomichev
I'm planning to extend it in the next patches. It's much easier to work with C than BPF assembly.
Signed-off-by Stanislav Fomichev <[email protected]>---
With nits below:
Acked-by Andrii Nakryiko <[email protected]>
Thank you for the review! Will respin shortly with the nits addressed.
PATCH 2/3 bpf: allow bpf_{s,g}etsockopt from cgroup bind{4,6} hooks
In reply to Stanislav FomichevI have to now lock/unlock socket for the bind hook execution. That shouldn't cause any overhead because the socket is unbound and shouldn't receive any traffic.
--- include/linux/bpf-cgroup.h | 12 ++++++------ net/core/filter.c | 4 ++++ net/ipv4/af_inet.c | 2 +- net/ipv6/af_inet6.c | 2 +- 4 files changed, 12 insertions(+), 8 deletions(-)
include/linux/bpf-cgroup.h +6 -6
@@ -246,11 +246,11 @@ int bpf_percpu_cgroup_storage_update(struct bpf_map *map, void *key, __ret; \ }) -#define BPF_CGROUP_RUN_PROG_INET4_BIND (sk, uaddr) \ - BPF_CGROUP_RUN_SA_PROG(sk, uaddr, BPF_CGROUP_INET4_BIND) +#define BPF_CGROUP_RUN_PROG_INET4_BIND_LOCK (sk, uaddr) \ + BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, BPF_CGROUP_INET4_BIND, NULL) -#define BPF_CGROUP_RUN_PROG_INET6_BIND (sk, uaddr) \ - BPF_CGROUP_RUN_SA_PROG(sk, uaddr, BPF_CGROUP_INET6_BIND) +#define BPF_CGROUP_RUN_PROG_INET6_BIND_LOCK (sk, uaddr) \ + BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, BPF_CGROUP_INET6_BIND, NULL)#define BPF_CGROUP_PRE_CONNECT_ENABLED (sk) (cgroup_bpf_enabled && \ sk->sk_prot->pre_connect)
@@ -434,8 +434,8 @@ static inline int bpf_percpu_cgroup_storage_update(struct bpf_map *map,#define BPF_CGROUP_RUN_PROG_INET_EGRESS (sk,skb) ({ 0; })#define BPF_CGROUP_RUN_PROG_INET_SOCK (sk) ({ 0; })#define BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE (sk) ({ 0; }) -#define BPF_CGROUP_RUN_PROG_INET4_BIND (sk, uaddr) ({ 0; }) -#define BPF_CGROUP_RUN_PROG_INET6_BIND (sk, uaddr) ({ 0; }) +#define BPF_CGROUP_RUN_PROG_INET4_BIND_LOCK (sk, uaddr) ({ 0; }) +#define BPF_CGROUP_RUN_PROG_INET6_BIND_LOCK (sk, uaddr) ({ 0; })#define BPF_CGROUP_RUN_PROG_INET4_POST_BIND (sk) ({ 0; })#define BPF_CGROUP_RUN_PROG_INET6_POST_BIND (sk) ({ 0; })#define BPF_CGROUP_RUN_PROG_INET4_CONNECT (sk, uaddr) ({ 0; })
net/core/filter.c +4 -0
@@ -6995,6 +6995,8 @@ sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)return &bpf_sk_storage_delete_proto;case BPF_FUNC_setsockopt:switch (prog->expected_attach_type) { +case BPF_CGROUP_INET4_BIND : +case BPF_CGROUP_INET6_BIND :case BPF_CGROUP_INET4_CONNECT :case BPF_CGROUP_INET6_CONNECT :return &bpf_sock_addr_setsockopt_proto;
@@ -7003,6 +7005,8 @@ sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) }case BPF_FUNC_getsockopt:switch (prog->expected_attach_type) { +case BPF_CGROUP_INET4_BIND : +case BPF_CGROUP_INET6_BIND :case BPF_CGROUP_INET4_CONNECT :case BPF_CGROUP_INET6_CONNECT :return &bpf_sock_addr_getsockopt_proto;
net/ipv4/af_inet.c +1 -1
@@ -450,7 +450,7 @@ int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)/* BPF prog is run before any checks are done so that if the prog * changes context in a wrong way it will be caught. */ - err =BPF_CGROUP_RUN_PROG_INET4_BIND (sk, uaddr); + err =BPF_CGROUP_RUN_PROG_INET4_BIND_LOCK (sk, uaddr);if (err)return err;
net/ipv6/af_inet6.c +1 -1
@@ -451,7 +451,7 @@ int inet6_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)/* BPF prog is run before any checks are done so that if the prog * changes context in a wrong way it will be caught. */ - err =BPF_CGROUP_RUN_PROG_INET6_BIND (sk, uaddr); + err =BPF_CGROUP_RUN_PROG_INET6_BIND_LOCK (sk, uaddr);if (err)return err;
Signature · 1 line
2.29.2.299.gdc1121823c-goog
Re: [PATCH bpf-next 2/3] bpf: allow bpf_{s,g}etsockopt from cgroup bind{4,6} hooks
Quoted 71 lines · Stanislav Fomichev
I have to now lock/unlock socket for the bind hook execution. That shouldn't cause any overhead because the socket is unbound and shouldn't receive any traffic.
Signed-off-by Stanislav Fomichev <[email protected]>--- include/linux/bpf-cgroup.h | 12 ++++++------ net/core/filter.c | 4 ++++ net/ipv4/af_inet.c | 2 +- net/ipv6/af_inet6.c | 2 +- 4 files changed, 12 insertions(+), 8 deletions(-)include/linux/bpf-cgroup.h +6
-6@@ -246,11 +246,11 @@ int bpf_percpu_cgroup_storage_update(struct bpf_map *map, void *key, __ret; \ }) -#define BPF_CGROUP_RUN_PROG_INET4_BIND (sk, uaddr) \ - BPF_CGROUP_RUN_SA_PROG(sk, uaddr, BPF_CGROUP_INET4_BIND) +#define BPF_CGROUP_RUN_PROG_INET4_BIND_LOCK (sk, uaddr) \ + BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, BPF_CGROUP_INET4_BIND, NULL) -#define BPF_CGROUP_RUN_PROG_INET6_BIND (sk, uaddr) \ - BPF_CGROUP_RUN_SA_PROG(sk, uaddr, BPF_CGROUP_INET6_BIND) +#define BPF_CGROUP_RUN_PROG_INET6_BIND_LOCK (sk, uaddr) \ + BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, BPF_CGROUP_INET6_BIND, NULL)#define BPF_CGROUP_PRE_CONNECT_ENABLED (sk) (cgroup_bpf_enabled && \ sk->sk_prot->pre_connect)@@ -434,8 +434,8 @@ static inline int bpf_percpu_cgroup_storage_update(struct bpf_map *map,#define BPF_CGROUP_RUN_PROG_INET_EGRESS (sk,skb) ({ 0; })#define BPF_CGROUP_RUN_PROG_INET_SOCK (sk) ({ 0; })#define BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE (sk) ({ 0; }) -#define BPF_CGROUP_RUN_PROG_INET4_BIND (sk, uaddr) ({ 0; }) -#define BPF_CGROUP_RUN_PROG_INET6_BIND (sk, uaddr) ({ 0; }) +#define BPF_CGROUP_RUN_PROG_INET4_BIND_LOCK (sk, uaddr) ({ 0; }) +#define BPF_CGROUP_RUN_PROG_INET6_BIND_LOCK (sk, uaddr) ({ 0; })#define BPF_CGROUP_RUN_PROG_INET4_POST_BIND (sk) ({ 0; })#define BPF_CGROUP_RUN_PROG_INET6_POST_BIND (sk) ({ 0; })#define BPF_CGROUP_RUN_PROG_INET4_CONNECT (sk, uaddr) ({ 0; })net/core/filter.c +4
-0@@ -6995,6 +6995,8 @@ sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)return &bpf_sk_storage_delete_proto;case BPF_FUNC_setsockopt:switch (prog->expected_attach_type) { +case BPF_CGROUP_INET4_BIND : +case BPF_CGROUP_INET6_BIND :case BPF_CGROUP_INET4_CONNECT :case BPF_CGROUP_INET6_CONNECT :return &bpf_sock_addr_setsockopt_proto;@@ -7003,6 +7005,8 @@ sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) }case BPF_FUNC_getsockopt:switch (prog->expected_attach_type) { +case BPF_CGROUP_INET4_BIND : +case BPF_CGROUP_INET6_BIND :case BPF_CGROUP_INET4_CONNECT :case BPF_CGROUP_INET6_CONNECT :return &bpf_sock_addr_getsockopt_proto;net/ipv4/af_inet.c +0
-0
net/ipv4/af_inet.c int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) · line 453
Quoted 5 lines
net/ipv4/af_inet.c +1
-1@@ -450,7 +450,7 @@ int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)/* BPF prog is run before any checks are done so that if the prog * changes context in a wrong way it will be caught. */ - err =BPF_CGROUP_RUN_PROG_INET4_BIND (sk, uaddr); + err =BPF_CGROUP_RUN_PROG_INET4_BIND_LOCK (sk, uaddr);
I think it is ok, but I need to go through the locking paths more. Andrey,
please take a look as well.
Quoted 19 lines
net/ipv4/af_inet.c +0
-0if (err)return err;net/ipv6/af_inet6.c +1
-2@@ -451,7 +451,7 @@ int inet6_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)/* BPF prog is run before any checks are done so that if the prog * changes context in a wrong way it will be caught. */ - err =BPF_CGROUP_RUN_PROG_INET6_BIND (sk, uaddr); + err =BPF_CGROUP_RUN_PROG_INET6_BIND_LOCK (sk, uaddr);if (err)return err; --2.29.2.299.gdc1121823c-goog
Re: [PATCH bpf-next 2/3] bpf: allow bpf_{s,g}etsockopt from cgroup bind{4,6} hooks
Alexei Starovoitov <[email protected]> [Tue, 2020-11-17 20:05 -0800]:
Quoted 73 lines · Stanislav Fomichev
Quoted 72 lines
I have to now lock/unlock socket for the bind hook execution. That shouldn't cause any overhead because the socket is unbound and shouldn't receive any traffic.
Signed-off-by Stanislav Fomichev <[email protected]>--- include/linux/bpf-cgroup.h | 12 ++++++------ net/core/filter.c | 4 ++++ net/ipv4/af_inet.c | 2 +- net/ipv6/af_inet6.c | 2 +- 4 files changed, 12 insertions(+), 8 deletions(-)include/linux/bpf-cgroup.h +6
-6@@ -246,11 +246,11 @@ int bpf_percpu_cgroup_storage_update(struct bpf_map *map, void *key, __ret; \ }) -#define BPF_CGROUP_RUN_PROG_INET4_BIND (sk, uaddr) \ - BPF_CGROUP_RUN_SA_PROG(sk, uaddr, BPF_CGROUP_INET4_BIND) +#define BPF_CGROUP_RUN_PROG_INET4_BIND_LOCK (sk, uaddr) \ + BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, BPF_CGROUP_INET4_BIND, NULL) -#define BPF_CGROUP_RUN_PROG_INET6_BIND (sk, uaddr) \ - BPF_CGROUP_RUN_SA_PROG(sk, uaddr, BPF_CGROUP_INET6_BIND) +#define BPF_CGROUP_RUN_PROG_INET6_BIND_LOCK (sk, uaddr) \ + BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, BPF_CGROUP_INET6_BIND, NULL)#define BPF_CGROUP_PRE_CONNECT_ENABLED (sk) (cgroup_bpf_enabled && \ sk->sk_prot->pre_connect)@@ -434,8 +434,8 @@ static inline int bpf_percpu_cgroup_storage_update(struct bpf_map *map,#define BPF_CGROUP_RUN_PROG_INET_EGRESS (sk,skb) ({ 0; })#define BPF_CGROUP_RUN_PROG_INET_SOCK (sk) ({ 0; })#define BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE (sk) ({ 0; }) -#define BPF_CGROUP_RUN_PROG_INET4_BIND (sk, uaddr) ({ 0; }) -#define BPF_CGROUP_RUN_PROG_INET6_BIND (sk, uaddr) ({ 0; }) +#define BPF_CGROUP_RUN_PROG_INET4_BIND_LOCK (sk, uaddr) ({ 0; }) +#define BPF_CGROUP_RUN_PROG_INET6_BIND_LOCK (sk, uaddr) ({ 0; })#define BPF_CGROUP_RUN_PROG_INET4_POST_BIND (sk) ({ 0; })#define BPF_CGROUP_RUN_PROG_INET6_POST_BIND (sk) ({ 0; })#define BPF_CGROUP_RUN_PROG_INET4_CONNECT (sk, uaddr) ({ 0; })net/core/filter.c +4
-0@@ -6995,6 +6995,8 @@ sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)return &bpf_sk_storage_delete_proto;case BPF_FUNC_setsockopt:switch (prog->expected_attach_type) { +case BPF_CGROUP_INET4_BIND : +case BPF_CGROUP_INET6_BIND :case BPF_CGROUP_INET4_CONNECT :case BPF_CGROUP_INET6_CONNECT :return &bpf_sock_addr_setsockopt_proto;@@ -7003,6 +7005,8 @@ sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) }case BPF_FUNC_getsockopt:switch (prog->expected_attach_type) { +case BPF_CGROUP_INET4_BIND : +case BPF_CGROUP_INET6_BIND :case BPF_CGROUP_INET4_CONNECT :case BPF_CGROUP_INET6_CONNECT :return &bpf_sock_addr_getsockopt_proto;net/ipv4/af_inet.c +0
-0@@ -450,7 +450,7 @@ int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)/* BPF prog is run before any checks are done so that if the prog
Quoted 8 lines
net/ipv4/af_inet.c int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) · line 453
Quoted 4 lines
net/ipv4/af_inet.c +1
-1* changes context in a wrong way it will be caught. */ - err =BPF_CGROUP_RUN_PROG_INET4_BIND (sk, uaddr); + err =BPF_CGROUP_RUN_PROG_INET4_BIND_LOCK (sk, uaddr);I think it is ok, but I need to go through the locking paths more. Andrey,
please take a look as well.
Sorry for delay, I was offline for the last two weeks.
From the correctness perspective it looks fine to me.
From the performance perspective I can think of one relevant scenario. Quite common use-case in applications is to use bind(2) not before listen(2) but before connect(2) for client sockets so that connection can be set up from specific source IP and, optionally, port.
Binding to both IP and port case is not interesting since it's already slow due to get_port().
But some applications do care about connection setup performance and at the same time need to set source IP only (no port). In this case they use IP_BIND_ADDRESS_NO_PORT socket option, what makes bind(2) fast (we've discussed it with Stanislav earlier in [0]).
I can imagine some pathological case when an application sets up tons of connections with bind(2) before connect(2) for sockets with IP_BIND_ADDRESS_NO_PORT enabled (that by itself requires setsockopt(2) though, i.e. socket lock/unlock) and that another lock/unlock to run bind hook may add some overhead. Though I do not know how critical that overhead may be and whether it's worth to benchmark or not (maybe too much paranoia).
[0] https://lore.kernel.org/bpf/20200505182010.GB55644@rdna-mbp/
Quoted 19 lines
Quoted 19 lines
net/ipv4/af_inet.c +0
-0if (err)return err;net/ipv6/af_inet6.c +1
-2@@ -451,7 +451,7 @@ int inet6_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)/* BPF prog is run before any checks are done so that if the prog * changes context in a wrong way it will be caught. */ - err =BPF_CGROUP_RUN_PROG_INET6_BIND (sk, uaddr); + err =BPF_CGROUP_RUN_PROG_INET6_BIND_LOCK (sk, uaddr);if (err)return err; --2.29.2.299.gdc1121823c-goog
Signature · 1 line
Andrey Ignatov
Re: [PATCH bpf-next 2/3] bpf: allow bpf_{s,g}etsockopt from cgroup bind{4,6} hooks
Quoted 4 lines · Andrey Ignatov
Alexei Starovoitov <[email protected]> [Tue, 2020-11-17 20:05 -0800]:
Quoted 1 line
On Tue, Nov 17, 2020 at 4:17 PM Stanislav Fomichev <[email protected]>
wrote:
[..]
Quoted 6 lines
Quoted 4 lines
I think it is ok, but I need to go through the locking paths more. Andrey,
please take a look as well.Sorry for delay, I was offline for the last two weeks.
No worries, I was OOO myself last week, thanks for the feedback!
Quoted 23 lines
From the correctness perspective it looks fine to me.
From the performance perspective I can think of one relevant scenario.
Quite common use-case in applications is to use bind(2) not before listen(2) but before connect(2) for client sockets so that connection can be set up from specific source IP and, optionally, port.Binding to both IP and port case is not interesting since it's already slow due to
get_port().But some applications do care about connection setup performance and at the same time need to set source IP only (no port). In this case they use
IP_BIND_ADDRESS_NO_PORTsocket option, what makes bind(2) fast (we've discussed it with Stanislav earlier in [0]).I can imagine some pathological case when an application sets up tons of connections with bind(2) before connect(2) for sockets with
IP_BIND_ADDRESS_NO_PORTenabled (that by itself requires setsockopt(2) though, i.e. socket lock/unlock) and that another lock/unlock to run bind hook may add some overhead. Though I do not know how critical that overhead may be and whether it's worth to benchmark or not (maybe too much paranoia).
Quoted 1 line
[0] https://lore.kernel.org/bpf/20200505182010.GB55644@rdna-mbp/
Even in case of IP_BIND_ADDRESS_NO_PORT, inet[6]_bind() does lock_sock down the line, so it's not like we are switching a lockless path to the one with the lock, right?
And in this case, similar to listen, the socket is still uncontended and owned by the userspace. So that extra lock/unlock should be cheap enough to be ignored (spin_lock_bh on the warm cache line).
Am I missing something?
Re: [PATCH bpf-next 2/3] bpf: allow bpf_{s,g}etsockopt from cgroup bind{4,6} hooks
[email protected] <[email protected]> [Mon, 2020-11-30 08:38 -0800]:
Quoted 37 lines · Andrey Ignatov
Quoted 4 lines
Alexei Starovoitov <[email protected]> [Tue, 2020-11-17 20:05 -0800]:
Quoted 1 line
On Tue, Nov 17, 2020 at 4:17 PM Stanislav Fomichev <[email protected]>
wrote:
[..]
Quoted 6 lines
Quoted 4 lines
I think it is ok, but I need to go through the locking paths more. Andrey,
please take a look as well.Sorry for delay, I was offline for the last two weeks.
No worries, I was OOO myself last week, thanks for the feedback!
Quoted 22 lines
From the correctness perspective it looks fine to me.
From the performance perspective I can think of one relevant scenario.
Quite common use-case in applications is to use bind(2) not before listen(2) but before connect(2) for client sockets so that connection can be set up from specific source IP and, optionally, port.Binding to both IP and port case is not interesting since it's already slow due to
get_port().But some applications do care about connection setup performance and at the same time need to set source IP only (no port). In this case they use
IP_BIND_ADDRESS_NO_PORTsocket option, what makes bind(2) fast (we've discussed it with Stanislav earlier in [0]).I can imagine some pathological case when an application sets up tons of connections with bind(2) before connect(2) for sockets with
IP_BIND_ADDRESS_NO_PORTenabled (that by itself requires setsockopt(2) though, i.e. socket lock/unlock) and that another lock/unlock to run bind hook may add some overhead. Though I do not know how critical that overhead may be and whether it's worth to benchmark or not (maybe too much paranoia).
Quoted 4 lines
Quoted 1 line
[0] https://lore.kernel.org/bpf/20200505182010.GB55644@rdna-mbp/
Even in case of
IP_BIND_ADDRESS_NO_PORT, inet[6]_bind() doeslock_sockdown the line, so it's not like we are switching a lockless path to the one with the lock, right?
Right, I understand that it's going from one lock/unlock to two (not from zero to one), that's what I meant by "another". My point was about this one more lock.
Quoted 5 lines
And in this case, similar to listen, the socket is still uncontended and owned by the userspace. So that extra lock/unlock should be cheap enough to be ignored (
spin_lock_bhon the warm cache line).Am I missing something?
As I mentioned it may come up only in "pathological case" what is probably fine to ignore, i.e. I'd rather agree with "cheap enough to be ignored" and benchmark would likely confirm it, I just couldn't say that for sure w/o numbers so brought this point.
Given that we both agree that it should be fine to ignore this +1 lock, IMO it should be good to go unless someone else has objections.
Signature · 1 line
Andrey Ignatov
Re: [PATCH bpf-next 2/3] bpf: allow bpf_{s,g}etsockopt from cgroup bind{4,6} hooks
Quoted 58 lines · Andrey Ignatov
[email protected] <[email protected]> [Mon, 2020-11-30 08:38 -0800]:
Quoted 2 lines · Andrey Ignatov
Quoted 1 line
Alexei Starovoitov <[email protected]> [Tue, 2020-11-17
20:05
Quoted 15 lines
Quoted 3 lines
-0800]:
Quoted 1 line
On Tue, Nov 17, 2020 at 4:17 PM Stanislav Fomichev <[email protected]>
wrote:
[..]
Quoted 6 lines
Quoted 4 lines
I think it is ok, but I need to go through the locking paths more. Andrey,
please take a look as well.Sorry for delay, I was offline for the last two weeks.
No worries, I was OOO myself last week, thanks for the feedback!
Quoted 3 lines
From the correctness perspective it looks fine to me.
From the performance perspective I can think of one relevant
scenario.
Quoted 8 lines
Quoted 8 lines
Quite common use-case in applications is to use bind(2) not before listen(2) but before connect(2) for client sockets so that connection can be set up from specific source IP and, optionally, port.
Binding to both IP and port case is not interesting since it's already slow due to
get_port().But some applications do care about connection setup performance and
at
Quoted 5 lines
Quoted 5 lines
the same time need to set source IP only (no port). In this case they use
IP_BIND_ADDRESS_NO_PORTsocket option, what makes bind(2) fast (we've discussed it with Stanislav earlier in [0]).I can imagine some pathological case when an application sets up tons
of
Quoted 4 lines
Quoted 4 lines
connections with bind(2) before connect(2) for sockets with
IP_BIND_ADDRESS_NO_PORTenabled (that by itself requires setsockopt(2) though, i.e. socket lock/unlock) and that another lock/unlock to run bind hook may add some overhead. Though I do not know how criticalthat
Quoted 7 lines
Quoted 4 lines
overhead may be and whether it's worth to benchmark or not (maybe too much paranoia).
[0] https://lore.kernel.org/bpf/20200505182010.GB55644@rdna-mbp/
Even in case of
IP_BIND_ADDRESS_NO_PORT, inet[6]_bind() doeslock_sockdown the line, so it's not like we are switching a lockless path to the one with the lock, right?Right, I understand that it's going from one lock/unlock to two (not from zero to one), that's what I meant by "another". My point was about this one more lock.
Quoted 5 lines
And in this case, similar to listen, the socket is still uncontended and owned by the userspace. So that extra lock/unlock should be cheap enough to be ignored (
spin_lock_bhon the warm cache line).Am I missing something?
Quoted 7 lines
As I mentioned it may come up only in "pathological case" what is probably fine to ignore, i.e. I'd rather agree with "cheap enough to be ignored" and benchmark would likely confirm it, I just couldn't say that for sure w/o numbers so brought this point.
Given that we both agree that it should be fine to ignore this +1 lock, IMO it should be good to go unless someone else has objections.
Thanks, agreed. Do you mind giving it an acked-by so it gets some attention in the patchwork? ;-)
Re: [PATCH bpf-next 2/3] bpf: allow bpf_{s,g}etsockopt from cgroup bind{4,6} hooks
[email protected] <[email protected]> [Tue, 2020-12-01 10:43 -0800]:
Quoted 64 lines · Andrey Ignatov
Quoted 62 lines
[email protected] <[email protected]> [Mon, 2020-11-30 08:38 -0800]:
Quoted 2 lines · Andrey Ignatov
Quoted 1 line
Alexei Starovoitov <[email protected]> [Tue, 2020-11-17
20:05
Quoted 15 lines
Quoted 3 lines
-0800]:
Quoted 1 line
On Tue, Nov 17, 2020 at 4:17 PM Stanislav Fomichev <[email protected]>
wrote:
[..]
Quoted 6 lines
Quoted 4 lines
I think it is ok, but I need to go through the locking paths more. Andrey,
please take a look as well.Sorry for delay, I was offline for the last two weeks.
No worries, I was OOO myself last week, thanks for the feedback!
Quoted 3 lines
From the correctness perspective it looks fine to me.
From the performance perspective I can think of one relevant
scenario.
Quoted 8 lines
Quoted 8 lines
Quite common use-case in applications is to use bind(2) not before listen(2) but before connect(2) for client sockets so that connection can be set up from specific source IP and, optionally, port.
Binding to both IP and port case is not interesting since it's already slow due to
get_port().But some applications do care about connection setup performance and
at
Quoted 5 lines
Quoted 5 lines
the same time need to set source IP only (no port). In this case they use
IP_BIND_ADDRESS_NO_PORTsocket option, what makes bind(2) fast (we've discussed it with Stanislav earlier in [0]).I can imagine some pathological case when an application sets up
tons of
Quoted 4 lines
Quoted 4 lines
connections with bind(2) before connect(2) for sockets with
IP_BIND_ADDRESS_NO_PORTenabled (that by itself requires setsockopt(2) though, i.e. socket lock/unlock) and that another lock/unlock to run bind hook may add some overhead. Though I do not know how criticalthat
Quoted 7 lines
Quoted 4 lines
overhead may be and whether it's worth to benchmark or not (maybe too much paranoia).
[0] https://lore.kernel.org/bpf/20200505182010.GB55644@rdna-mbp/
Even in case of
IP_BIND_ADDRESS_NO_PORT, inet[6]_bind() doeslock_sockdown the line, so it's not like we are switching a lockless path to the one with the lock, right?Right, I understand that it's going from one lock/unlock to two (not from zero to one), that's what I meant by "another". My point was about this one more lock.
Quoted 5 lines
And in this case, similar to listen, the socket is still uncontended and owned by the userspace. So that extra lock/unlock should be cheap enough to be ignored (
spin_lock_bhon the warm cache line).Am I missing something?
As I mentioned it may come up only in "pathological case" what is probably fine to ignore, i.e. I'd rather agree with "cheap enough to be ignored" and benchmark would likely confirm it, I just couldn't say that for sure w/o numbers so brought this point.
Quoted 4 lines
Quoted 2 lines
Given that we both agree that it should be fine to ignore this +1 lock, IMO it should be good to go unless someone else has objections.
Thanks, agreed. Do you mind giving it an acked-by so it gets some attention in the patchwork? ;-)
Sure. Acked this one.
Signature · 1 line
Andrey Ignatov
Re: [PATCH bpf-next 2/3] bpf: allow bpf_{s,g}etsockopt from cgroup bind{4,6} hooks
In reply to Stanislav FomichevStanislav Fomichev <[email protected]> [Tue, 2020-11-17 16:18 -0800]:
Quoted 5 lines
I have to now lock/unlock socket for the bind hook execution. That shouldn't cause any overhead because the socket is unbound and shouldn't receive any traffic.
Signed-off-by Stanislav Fomichev <[email protected]>
Quoted 89 lines
--- include/linux/bpf-cgroup.h | 12 ++++++------ net/core/filter.c | 4 ++++ net/ipv4/af_inet.c | 2 +- net/ipv6/af_inet6.c | 2 +- 4 files changed, 12 insertions(+), 8 deletions(-)include/linux/bpf-cgroup.h +6
-6@@ -246,11 +246,11 @@ int bpf_percpu_cgroup_storage_update(struct bpf_map *map, void *key, __ret; \ }) -#define BPF_CGROUP_RUN_PROG_INET4_BIND (sk, uaddr) \ - BPF_CGROUP_RUN_SA_PROG(sk, uaddr, BPF_CGROUP_INET4_BIND) +#define BPF_CGROUP_RUN_PROG_INET4_BIND_LOCK (sk, uaddr) \ + BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, BPF_CGROUP_INET4_BIND, NULL) -#define BPF_CGROUP_RUN_PROG_INET6_BIND (sk, uaddr) \ - BPF_CGROUP_RUN_SA_PROG(sk, uaddr, BPF_CGROUP_INET6_BIND) +#define BPF_CGROUP_RUN_PROG_INET6_BIND_LOCK (sk, uaddr) \ + BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, BPF_CGROUP_INET6_BIND, NULL)#define BPF_CGROUP_PRE_CONNECT_ENABLED (sk) (cgroup_bpf_enabled && \ sk->sk_prot->pre_connect)@@ -434,8 +434,8 @@ static inline int bpf_percpu_cgroup_storage_update(struct bpf_map *map,#define BPF_CGROUP_RUN_PROG_INET_EGRESS (sk,skb) ({ 0; })#define BPF_CGROUP_RUN_PROG_INET_SOCK (sk) ({ 0; })#define BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE (sk) ({ 0; }) -#define BPF_CGROUP_RUN_PROG_INET4_BIND (sk, uaddr) ({ 0; }) -#define BPF_CGROUP_RUN_PROG_INET6_BIND (sk, uaddr) ({ 0; }) +#define BPF_CGROUP_RUN_PROG_INET4_BIND_LOCK (sk, uaddr) ({ 0; }) +#define BPF_CGROUP_RUN_PROG_INET6_BIND_LOCK (sk, uaddr) ({ 0; })#define BPF_CGROUP_RUN_PROG_INET4_POST_BIND (sk) ({ 0; })#define BPF_CGROUP_RUN_PROG_INET6_POST_BIND (sk) ({ 0; })#define BPF_CGROUP_RUN_PROG_INET4_CONNECT (sk, uaddr) ({ 0; })net/core/filter.c +4
-0@@ -6995,6 +6995,8 @@ sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)return &bpf_sk_storage_delete_proto;case BPF_FUNC_setsockopt:switch (prog->expected_attach_type) { +case BPF_CGROUP_INET4_BIND : +case BPF_CGROUP_INET6_BIND :case BPF_CGROUP_INET4_CONNECT :case BPF_CGROUP_INET6_CONNECT :return &bpf_sock_addr_setsockopt_proto;@@ -7003,6 +7005,8 @@ sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) }case BPF_FUNC_getsockopt:switch (prog->expected_attach_type) { +case BPF_CGROUP_INET4_BIND : +case BPF_CGROUP_INET6_BIND :case BPF_CGROUP_INET4_CONNECT :case BPF_CGROUP_INET6_CONNECT :return &bpf_sock_addr_getsockopt_proto;net/ipv4/af_inet.c +1
-1@@ -450,7 +450,7 @@ int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)/* BPF prog is run before any checks are done so that if the prog * changes context in a wrong way it will be caught. */ - err =BPF_CGROUP_RUN_PROG_INET4_BIND (sk, uaddr); + err =BPF_CGROUP_RUN_PROG_INET4_BIND_LOCK (sk, uaddr);if (err)return err;net/ipv6/af_inet6.c +1
-1@@ -451,7 +451,7 @@ int inet6_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)/* BPF prog is run before any checks are done so that if the prog * changes context in a wrong way it will be caught. */ - err =BPF_CGROUP_RUN_PROG_INET6_BIND (sk, uaddr); + err =BPF_CGROUP_RUN_PROG_INET6_BIND_LOCK (sk, uaddr);if (err)return err;Signature · 1 line
2.29.2.299.gdc1121823c-goog
Signature · 1 line
Andrey Ignatov
PATCH 3/3 selftests/bpf: extend bind{4,6} programs with a call to bpf_setsockopt
In reply to Stanislav FomichevTo make sure it doesn't trigger sock_owned_by_me splat.
--- .../testing/selftests/bpf/progs/bind4_prog.c | 31 +++++++++++++++++++ .../testing/selftests/bpf/progs/bind6_prog.c | 31 +++++++++++++++++++ 2 files changed, 62 insertions(+)
tools/testing/selftests/bpf/progs/bind4_prog.c +31 -0
@@ -19,8 +19,35 @@#define SERV4_REWRITE_IP 0x7f000001U/* 127.0.0.1 */ #define SERV4_REWRITE_PORT 4444 +#ifndef IFNAMSIZ +#define IFNAMSIZ 16 +#endif + int _version SEC("version") = 1; +static __inlineint bind_to_device (struct bpf_sock_addr *ctx) +{ +char veth1[IFNAMSIZ ] ="test_sock_addr1" ; +char veth2[IFNAMSIZ ] ="test_sock_addr2" ; +char missing[IFNAMSIZ ] ="nonexistent_dev" ; +char del_bind[IFNAMSIZ ] ="" ; + +if (bpf_setsockopt (ctx,SOL_SOCKET ,SO_BINDTODEVICE , + &veth1,sizeof (veth1))) +return 1 ; +if (bpf_setsockopt (ctx,SOL_SOCKET ,SO_BINDTODEVICE , + &veth2,sizeof (veth2))) +return 1 ; +if (bpf_setsockopt (ctx,SOL_SOCKET ,SO_BINDTODEVICE , + &missing,sizeof (missing)) != -ENODEV ) +return 1 ; +if (bpf_setsockopt (ctx,SOL_SOCKET ,SO_BINDTODEVICE , + &del_bind,sizeof (del_bind))) +return 1 ; + +return 0 ; +} + SEC("cgroup/bind4")int bind_v4_prog (struct bpf_sock_addr *ctx) {
@@ -64,6 +91,10 @@ int bind_v4_prog(struct bpf_sock_addr *ctx)if (ctx->user_ip4 != user_ip4)return 0 ; +/* Bind to device and unbind it. */ +if (bind_to_device (ctx)) +return 0 ; + ctx->user_ip4 =bpf_htonl (SERV4_REWRITE_IP ); ctx->user_port =bpf_htons (SERV4_REWRITE_PORT );
tools/testing/selftests/bpf/progs/bind6_prog.c +31 -0
@@ -25,8 +25,35 @@#define SERV6_REWRITE_IP_3 0x00000001#define SERV6_REWRITE_PORT 6666 +#ifndef IFNAMSIZ +#define IFNAMSIZ 16 +#endif + int _version SEC("version") = 1; +static __inlineint bind_to_device (struct bpf_sock_addr *ctx) +{ +char veth1[IFNAMSIZ ] ="test_sock_addr1" ; +char veth2[IFNAMSIZ ] ="test_sock_addr2" ; +char missing[IFNAMSIZ ] ="nonexistent_dev" ; +char del_bind[IFNAMSIZ ] ="" ; + +if (bpf_setsockopt (ctx,SOL_SOCKET ,SO_BINDTODEVICE , + &veth1,sizeof (veth1))) +return 1 ; +if (bpf_setsockopt (ctx,SOL_SOCKET ,SO_BINDTODEVICE , + &veth2,sizeof (veth2))) +return 1 ; +if (bpf_setsockopt (ctx,SOL_SOCKET ,SO_BINDTODEVICE , + &missing,sizeof (missing)) != -ENODEV ) +return 1 ; +if (bpf_setsockopt (ctx,SOL_SOCKET ,SO_BINDTODEVICE , + &del_bind,sizeof (del_bind))) +return 1 ; + +return 0 ; +} + SEC("cgroup/bind6")int bind_v6_prog (struct bpf_sock_addr *ctx) {
@@ -78,6 +105,10 @@ int bind_v6_prog(struct bpf_sock_addr *ctx)return 0 ; } +/* Bind to device and unbind it. */ +if (bind_to_device (ctx)) +return 0 ; + ctx->user_ip6[0 ] =bpf_htonl (SERV6_REWRITE_IP_0 ); ctx->user_ip6[1 ] =bpf_htonl (SERV6_REWRITE_IP_1 ); ctx->user_ip6[2 ] =bpf_htonl (SERV6_REWRITE_IP_2 );
Signature · 1 line
2.29.2.299.gdc1121823c-goog
Re: [PATCH bpf-next 3/3] selftests/bpf: extend bind{4,6} programs with a call to bpf_setsockopt
Quoted 5 lines · Stanislav Fomichev
To make sure it doesn't trigger
sock_owned_by_mesplat.Signed-off-by Stanislav Fomichev <[email protected]>---
Quoted 9 lines
.../testing/selftests/bpf/progs/bind4_prog.c | 31 +++++++++++++++++++ .../testing/selftests/bpf/progs/bind6_prog.c | 31 +++++++++++++++++++ 2 files changed, 62 insertions(+)tools/testing/selftests/bpf/progs/bind4_prog.c +0
-0
tools/testing/selftests/bpf/progs/bind4_prog.c line 26
Quoted 8 lines
tools/testing/selftests/bpf/progs/bind4_prog.c +4
-0@@ -19,8 +19,35 @@#define SERV4_REWRITE_IP 0x7f000001U/* 127.0.0.1 */ #define SERV4_REWRITE_PORT 4444 +#ifndef IFNAMSIZ +#define IFNAMSIZ 16 +#endif + int _version SEC("version") = 1;
nit: would be nice to drop this anachronism
Quoted 1 line
[...]
tools/testing/selftests/bpf/progs/bind6_prog.c line 32
Quoted 13 lines
tools/testing/selftests/bpf/progs/bind6_prog.c +4
-0@@ -25,8 +25,35 @@#define SERV6_REWRITE_IP_3 0x00000001#define SERV6_REWRITE_PORT 6666 +#ifndef IFNAMSIZ +#define IFNAMSIZ 16 +#endif + int _version SEC("version") = 1;
nit: same
Quoted 1 line
[...]