History

bpf: Add SO_KEEPALIVE and related options to bpf_setsockopt

This patch adds support of SO_KEEPALIVE flag and TCP related options to bpf_setsockopt() routine. This is helpful if we want to enable or tune TCP keepalive for applications which don't do it in the userspace code. v3: - update kernel-doc in uapi (Nikita Vetoshkin <nekto0n@yandex-team.ru>) v4: - update kernel-doc in tools too (Alexei Starovoitov) - add test to selftests (Alexei Starovoitov)

Dmitry Yakunin committed 6 years ago

Commit f9bcf961 parent aad4a0a4 files · +72 −5Browse files

Unified · Split

4 files
  1. include/uapi/linux/bpf.h +5 −2
  2. net/core/filter.c +35 −1
  3. tools/include/uapi/linux/bpf.h +5 −2
  4. tools/testing/selftests/bpf/progs/connect4_prog.c +27 −0
include/uapi/linux/bpf.h+5−2
@@ -1621,10 +1621,13 @@ union bpf_attr { · 1620 unmodified lines
1621 *1621 *
1622 * * **SOL_SOCKET**, which supports the following *optname*\ s:1622 * * **SOL_SOCKET**, which supports the following *optname*\ s:
1623 * **SO_RCVBUF**, **SO_SNDBUF**, **SO_MAX_PACING_RATE**,1623 * **SO_RCVBUF**, **SO_SNDBUF**, **SO_MAX_PACING_RATE**,
1624 * **SO_PRIORITY**, **SO_RCVLOWAT**, **SO_MARK**.1624 * **SO_PRIORITY**, **SO_RCVLOWAT**, **SO_MARK**,
1625 * **SO_BINDTODEVICE**, **SO_KEEPALIVE**.
1625 * * **IPPROTO_TCP**, which supports the following *optname*\ s:1626 * * **IPPROTO_TCP**, which supports the following *optname*\ s:
1626 * **TCP_CONGESTION**, **TCP_BPF_IW**,1627 * **TCP_CONGESTION**, **TCP_BPF_IW**,
1627 * **TCP_BPF_SNDCWND_CLAMP**.1628 * **TCP_BPF_SNDCWND_CLAMP**, **TCP_SAVE_SYN**,
1629 * **TCP_KEEPIDLE**, **TCP_KEEPINTVL**, **TCP_KEEPCNT**,
1630 * **TCP_SYNCNT**, **TCP_USER_TIMEOUT**.
1628 * * **IPPROTO_IP**, which supports *optname* **IP_TOS**.1631 * * **IPPROTO_IP**, which supports *optname* **IP_TOS**.
1629 * * **IPPROTO_IPV6**, which supports *optname* **IPV6_TCLASS**.1632 * * **IPPROTO_IPV6**, which supports *optname* **IPV6_TCLASS**.
1630 * Return1633 * Return
net/core/filter.c+35−1
@@ -4289,10 +4289,10 @@ static int _bpf_setsockopt(struct sock *sk, int level, int optname, · 4288 unmodified lines
4289 char *optval, int optlen, u32 flags)4289 char *optval, int optlen, u32 flags)
4290{4290{
4291 char devname[IFNAMSIZ];4291 char devname[IFNAMSIZ];
4292 int val, valbool;
4292 struct net *net;4293 struct net *net;
4293 int ifindex;4294 int ifindex;
4294 int ret = 0;4295 int ret = 0;
4295 int val;
4296 4296
4297 if (!sk_fullsock(sk))4297 if (!sk_fullsock(sk))
4298 return -EINVAL;4298 return -EINVAL;
@@ -4303,6 +4303,7 @@ static int _bpf_setsockopt(struct sock *sk, int level, int optname, · 4 unmodified lines
4303 if (optlen != sizeof(int) && optname != SO_BINDTODEVICE)4303 if (optlen != sizeof(int) && optname != SO_BINDTODEVICE)
4304 return -EINVAL;4304 return -EINVAL;
4305 val = *((int *)optval);4305 val = *((int *)optval);
4306 valbool = val ? 1 : 0;
4306 4307
4307 /* Only some socketops are supported */4308 /* Only some socketops are supported */
4308 switch (optname) {4309 switch (optname) {
@@ -4361,6 +4362,11 @@ static int _bpf_setsockopt(struct sock *sk, int level, int optname, · 52 unmodified lines
4361 }4362 }
4362 ret = sock_bindtoindex(sk, ifindex, false);4363 ret = sock_bindtoindex(sk, ifindex, false);
4363 break;4364 break;
4365 case SO_KEEPALIVE:
4366 if (sk->sk_prot->keepalive)
4367 sk->sk_prot->keepalive(sk, valbool);
4368 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
4369 break;
4364 default:4370 default:
4365 ret = -EINVAL;4371 ret = -EINVAL;
4366 }4372 }
@@ -4421,6 +4427,7 @@ static int _bpf_setsockopt(struct sock *sk, int level, int optname, · 54 unmodified lines
4421 ret = tcp_set_congestion_control(sk, name, false,4427 ret = tcp_set_congestion_control(sk, name, false,
4422 reinit, true);4428 reinit, true);
4423 } else {4429 } else {
4430 struct inet_connection_sock *icsk = inet_csk(sk);
4424 struct tcp_sock *tp = tcp_sk(sk);4431 struct tcp_sock *tp = tcp_sk(sk);
4425 4432
4426 if (optlen != sizeof(int))4433 if (optlen != sizeof(int))
@@ -4449,6 +4456,33 @@ static int _bpf_setsockopt(struct sock *sk, int level, int optname, · 22 unmodified lines
4449 else4456 else
4450 tp->save_syn = val;4457 tp->save_syn = val;
4451 break;4458 break;
4459 case TCP_KEEPIDLE:
4460 ret = tcp_sock_set_keepidle_locked(sk, val);
4461 break;
4462 case TCP_KEEPINTVL:
4463 if (val < 1 || val > MAX_TCP_KEEPINTVL)
4464 ret = -EINVAL;
4465 else
4466 tp->keepalive_intvl = val * HZ;
4467 break;
4468 case TCP_KEEPCNT:
4469 if (val < 1 || val > MAX_TCP_KEEPCNT)
4470 ret = -EINVAL;
4471 else
4472 tp->keepalive_probes = val;
4473 break;
4474 case TCP_SYNCNT:
4475 if (val < 1 || val > MAX_TCP_SYNCNT)
4476 ret = -EINVAL;
4477 else
4478 icsk->icsk_syn_retries = val;
4479 break;
4480 case TCP_USER_TIMEOUT:
4481 if (val < 0)
4482 ret = -EINVAL;
4483 else
4484 icsk->icsk_user_timeout = val;
4485 break;
4452 default:4486 default:
4453 ret = -EINVAL;4487 ret = -EINVAL;
4454 }4488 }
tools/include/uapi/linux/bpf.h+5−2
@@ -1621,10 +1621,13 @@ union bpf_attr { · 1620 unmodified lines
1621 *1621 *
1622 * * **SOL_SOCKET**, which supports the following *optname*\ s:1622 * * **SOL_SOCKET**, which supports the following *optname*\ s:
1623 * **SO_RCVBUF**, **SO_SNDBUF**, **SO_MAX_PACING_RATE**,1623 * **SO_RCVBUF**, **SO_SNDBUF**, **SO_MAX_PACING_RATE**,
1624 * **SO_PRIORITY**, **SO_RCVLOWAT**, **SO_MARK**.1624 * **SO_PRIORITY**, **SO_RCVLOWAT**, **SO_MARK**,
1625 * **SO_BINDTODEVICE**, **SO_KEEPALIVE**.
1625 * * **IPPROTO_TCP**, which supports the following *optname*\ s:1626 * * **IPPROTO_TCP**, which supports the following *optname*\ s:
1626 * **TCP_CONGESTION**, **TCP_BPF_IW**,1627 * **TCP_CONGESTION**, **TCP_BPF_IW**,
1627 * **TCP_BPF_SNDCWND_CLAMP**.1628 * **TCP_BPF_SNDCWND_CLAMP**, **TCP_SAVE_SYN**,
1629 * **TCP_KEEPIDLE**, **TCP_KEEPINTVL**, **TCP_KEEPCNT**,
1630 * **TCP_SYNCNT**, **TCP_USER_TIMEOUT**.
1628 * * **IPPROTO_IP**, which supports *optname* **IP_TOS**.1631 * * **IPPROTO_IP**, which supports *optname* **IP_TOS**.
1629 * * **IPPROTO_IPV6**, which supports *optname* **IPV6_TCLASS**.1632 * * **IPPROTO_IPV6**, which supports *optname* **IPV6_TCLASS**.
1630 * Return1633 * Return
tools/testing/selftests/bpf/progs/connect4_prog.c+27−0
@@ -104,6 +104,30 @@ static __inline int bind_to_device(struct bpf_sock_addr *ctx) · 103 unmodified lines
104 return 0;104 return 0;
105}105}
106 106
107static __inline int set_keepalive(struct bpf_sock_addr *ctx)
108{
109 int zero = 0, one = 1;
110
111 if (bpf_setsockopt(ctx, SOL_SOCKET, SO_KEEPALIVE, &one, sizeof(one)))
112 return 1;
113 if (ctx->type == SOCK_STREAM) {
114 if (bpf_setsockopt(ctx, SOL_TCP, TCP_KEEPIDLE, &one, sizeof(one)))
115 return 1;
116 if (bpf_setsockopt(ctx, SOL_TCP, TCP_KEEPINTVL, &one, sizeof(one)))
117 return 1;
118 if (bpf_setsockopt(ctx, SOL_TCP, TCP_KEEPCNT, &one, sizeof(one)))
119 return 1;
120 if (bpf_setsockopt(ctx, SOL_TCP, TCP_SYNCNT, &one, sizeof(one)))
121 return 1;
122 if (bpf_setsockopt(ctx, SOL_TCP, TCP_USER_TIMEOUT, &one, sizeof(one)))
123 return 1;
124 }
125 if (bpf_setsockopt(ctx, SOL_SOCKET, SO_KEEPALIVE, &zero, sizeof(zero)))
126 return 1;
127
128 return 0;
129}
130
107SEC("cgroup/connect4")131SEC("cgroup/connect4")
108int connect_v4_prog(struct bpf_sock_addr *ctx)132int connect_v4_prog(struct bpf_sock_addr *ctx)
109{133{
@@ -121,6 +145,9 @@ int connect_v4_prog(struct bpf_sock_addr *ctx) · 11 unmodified lines
121 if (bind_to_device(ctx))145 if (bind_to_device(ctx))
122 return 0;146 return 0;
123 147
148 if (set_keepalive(ctx))
149 return 0;
150
124 if (ctx->type != SOCK_STREAM && ctx->type != SOCK_DGRAM)151 if (ctx->type != SOCK_STREAM && ctx->type != SOCK_DGRAM)
125 return 0;152 return 0;
126 else if (ctx->type == SOCK_STREAM)153 else if (ctx->type == SOCK_STREAM)