Skip to contentSkip to conversation

bpf: add BPF_CGROUP_INET_SOCK_RELEASE hook

PATCH Superseded Stanislav Fomichev 4 participants 10 messages Jun 26, 2020 – Jun 28, 2020

Subsystems (9)

v2 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/4 bpf: add BPF_CGROUP_INET_SOCK_RELEASE hook

Sometimes it's handy to know when the socket gets freed. In particular, we'd like to try to use a smarter allocation of ports for bpf_bind and explore the possibility of limiting the number of SOCK_DGRAM sockets the process can have. Adding a release pair to existing BPF_CGROUP_INET_SOCK_CREATE can unlock both of the mentioned features.

The only questionable part here is the sock->sk check in the inet_release. Looking at the places where we do 'sock->sk = NULL', I don't understand how it can race with inet_release and why the check is there (it's been there since the initial git import). Otherwise, the change itself is pretty simple, we add a BPF hook to the inet_release and avoid calling it for kernel sockets.

v2:
* fix compile issue with CONFIG_CGROUP_BPF=n (kernel test robot)

---
 include/linux/bpf-cgroup.h | 4 ++++
 include/uapi/linux/bpf.h   | 1 +
 kernel/bpf/syscall.c       | 3 +++
 net/core/filter.c          | 1 +
 net/ipv4/af_inet.c         | 3 +++
 5 files changed, 12 insertions(+)
include/linux/bpf-cgroup.h +4 -0
index c66c545e161a..2c6f26670acc 100644
@@ -210,6 +210,9 @@ int bpf_percpu_cgroup_storage_update(struct bpf_map *map, void *key,
 #define BPF_CGROUP_RUN_PROG_INET_SOCK(sk)				       \
 	BPF_CGROUP_RUN_SK_PROG(sk, BPF_CGROUP_INET_SOCK_CREATE)
 
+#define BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk)			       \
+	BPF_CGROUP_RUN_SK_PROG(sk, BPF_CGROUP_INET_SOCK_RELEASE)
+
 #define BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk)				       \
 	BPF_CGROUP_RUN_SK_PROG(sk, BPF_CGROUP_INET4_POST_BIND)
 
@@ -401,6 +404,7 @@ static inline int bpf_percpu_cgroup_storage_update(struct bpf_map *map,
 #define BPF_CGROUP_RUN_PROG_INET_INGRESS(sk,skb) ({ 0; })
 #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_POST_BIND(sk) ({ 0; })
include/uapi/linux/bpf.h +1 -0
index c65b374a5090..d7aea1d0167a 100644
@@ -226,6 +226,7 @@ enum bpf_attach_type {
 	BPF_CGROUP_INET4_GETSOCKNAME,
 	BPF_CGROUP_INET6_GETSOCKNAME,
 	BPF_XDP_DEVMAP,
+	BPF_CGROUP_INET_SOCK_RELEASE,
 	__MAX_BPF_ATTACH_TYPE
 };
 
kernel/bpf/syscall.c +3 -0
index 4d530b1d5683..2a3d4b8f95c7 100644
@@ -1994,6 +1994,7 @@ bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
 	case BPF_PROG_TYPE_CGROUP_SOCK:
 		switch (expected_attach_type) {
 		case BPF_CGROUP_INET_SOCK_CREATE:
+		case BPF_CGROUP_INET_SOCK_RELEASE:
 		case BPF_CGROUP_INET4_POST_BIND:
 		case BPF_CGROUP_INET6_POST_BIND:
 			return 0;
@@ -2792,6 +2793,7 @@ attach_type_to_prog_type(enum bpf_attach_type attach_type)
 		return BPF_PROG_TYPE_CGROUP_SKB;
 		break;
 	case BPF_CGROUP_INET_SOCK_CREATE:
+	case BPF_CGROUP_INET_SOCK_RELEASE:
 	case BPF_CGROUP_INET4_POST_BIND:
 	case BPF_CGROUP_INET6_POST_BIND:
 		return BPF_PROG_TYPE_CGROUP_SOCK;
@@ -2942,6 +2944,7 @@ static int bpf_prog_query(const union bpf_attr *attr,
 	case BPF_CGROUP_INET_INGRESS:
 	case BPF_CGROUP_INET_EGRESS:
 	case BPF_CGROUP_INET_SOCK_CREATE:
+	case BPF_CGROUP_INET_SOCK_RELEASE:
 	case BPF_CGROUP_INET4_BIND:
 	case BPF_CGROUP_INET6_BIND:
 	case BPF_CGROUP_INET4_POST_BIND:
net/core/filter.c +1 -0
index 209482a4eaa2..7bcac182383c 100644
@@ -6855,6 +6855,7 @@ static bool __sock_filter_check_attach_type(int off,
 	case offsetof(struct bpf_sock, priority):
 		switch (attach_type) {
 		case BPF_CGROUP_INET_SOCK_CREATE:
+		case BPF_CGROUP_INET_SOCK_RELEASE:
 			goto full_access;
 		default:
 			return false;
net/ipv4/af_inet.c +3 -0
index 02aa5cb3a4fd..965a96ea1168 100644
@@ -411,6 +411,9 @@ int inet_release(struct socket *sock)
 	if (sk) {
 		long timeout;
 
+		if (!sk->sk_kern_sock)
+			BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk);
+
 		/* Applications forget to leave groups before exiting */
 		ip_mc_drop_socket(sk);
 
Signature · 1 line
2.27.0.212.ge8ba1cc988-goog
#1 lore raw

PATCH v2 2/4 libbpf: add support for BPF_CGROUP_INET_SOCK_RELEASE

Add auto-detection for the cgroup/sock_release programs.

---
 tools/include/uapi/linux/bpf.h | 1 +
 tools/lib/bpf/libbpf.c         | 2 ++
 2 files changed, 3 insertions(+)
tools/include/uapi/linux/bpf.h +1 -0
index c65b374a5090..d7aea1d0167a 100644
@@ -226,6 +226,7 @@ enum bpf_attach_type {
 	BPF_CGROUP_INET4_GETSOCKNAME,
 	BPF_CGROUP_INET6_GETSOCKNAME,
 	BPF_XDP_DEVMAP,
+	BPF_CGROUP_INET_SOCK_RELEASE,
 	__MAX_BPF_ATTACH_TYPE
 };
 
tools/lib/bpf/libbpf.c +2 -0
index 7f01be2b88b8..acbab6d0672d 100644
@@ -6670,6 +6670,8 @@ static const struct bpf_sec_def section_defs[] = {
 	BPF_APROG_SEC("cgroup_skb/egress",	BPF_PROG_TYPE_CGROUP_SKB,
 						BPF_CGROUP_INET_EGRESS),
 	BPF_APROG_COMPAT("cgroup/skb",		BPF_PROG_TYPE_CGROUP_SKB),
+	BPF_EAPROG_SEC("cgroup/sock_release",	BPF_PROG_TYPE_CGROUP_SOCK,
+						BPF_CGROUP_INET_SOCK_RELEASE),
 	BPF_APROG_SEC("cgroup/sock",		BPF_PROG_TYPE_CGROUP_SOCK,
 						BPF_CGROUP_INET_SOCK_CREATE),
 	BPF_EAPROG_SEC("cgroup/post_bind4",	BPF_PROG_TYPE_CGROUP_SOCK,
Signature · 1 line
2.27.0.212.ge8ba1cc988-goog
#2 lore raw

Re: [PATCH bpf-next v2 2/4] libbpf: add support for BPF_CGROUP_INET_SOCK_RELEASE

Quoted 5 lines · Stanislav Fomichev

Add auto-detection for the cgroup/sock_release programs.

---

Left a bunch of comments on v1 of this series, oops.

But where is the cover letter? Did I miss it?

Quoted 4 lines
 tools/include/uapi/linux/bpf.h | 1 +
 tools/lib/bpf/libbpf.c         | 2 ++
 2 files changed, 3 insertions(+)

[...]

#3 lore raw

Re: [PATCH bpf-next v2 2/4] libbpf: add support for BPF_CGROUP_INET_SOCK_RELEASE

Quoted 9 lines · Andrii Nakryiko
Quoted 5 lines · Stanislav Fomichev

Add auto-detection for the cgroup/sock_release programs.

---

Left a bunch of comments on v1 of this series, oops.

No worries, there is not much changed since then, thanks for the review!

Quoted 1 line

But where is the cover letter? Did I miss it?

I didn't feel like it deserves a cover letter. The main patch is small, most of the changes are in libbpf/selftests and I mostly split them to make it easier for you to push to github (is it still a requirement or I shouldn't worry about it anymore?).

#4 lore raw

Re: [PATCH bpf-next v2 2/4] libbpf: add support for BPF_CGROUP_INET_SOCK_RELEASE

Quoted 11 lines · Stanislav Fomichev
Quoted 8 lines · Andrii Nakryiko
Quoted 5 lines · Stanislav Fomichev

Add auto-detection for the cgroup/sock_release programs.

---

Quoted 8 lines
Quoted 1 line

Left a bunch of comments on v1 of this series, oops.

No worries, there is not much changed since then, thanks for the review!

Quoted 1 line

But where is the cover letter? Did I miss it?

I didn't feel like it deserves a cover letter. The main patch is small, most of the changes are in libbpf/selftests and I mostly split them to make it easier for you to push to github (is it still a requirement or I shouldn't worry about it anymore?).

It's not a requirement, sync script is pretty smart about all that, so no need to do it specifically for the sync purposes. Still a good idea to have selftests separately, IMO.

As for the cover letter, unless it's some trivial independent fixes, cover letter ties everything together and is a good place to track "what changed between versions". So I'd recommend adding it anyways.

#5 lore raw

Re: [PATCH bpf-next v2 2/4] libbpf: add support for BPF_CGROUP_INET_SOCK_RELEASE

Quoted 4 lines · Andrii Nakryiko

As for the cover letter, unless it's some trivial independent fixes, cover letter ties everything together and is a good place to track "what changed between versions". So I'd recommend adding it anyways.

+1
For independent fixes it's ok to skip cover, but it's often nice to have one. New release hook like this one certainly deserves a cover.

#6 lore raw

PATCH v2 3/4 bpftool: support BPF_CGROUP_INET_SOCK_RELEASE

In reply to Stanislav Fomichev

Support attaching to sock_release from the bpftool.

---
 tools/bpf/bpftool/main.h | 1 +
 1 file changed, 1 insertion(+)
tools/bpf/bpftool/main.h +1 -0
index 5cdf0bc049bd..0a281d3cceb8 100644
@@ -92,6 +92,7 @@ static const char * const attach_type_name[__MAX_BPF_ATTACH_TYPE] = {
 	[BPF_CGROUP_INET_INGRESS] = "ingress",
 	[BPF_CGROUP_INET_EGRESS] = "egress",
 	[BPF_CGROUP_INET_SOCK_CREATE] = "sock_create",
+	[BPF_CGROUP_INET_SOCK_RELEASE] = "sock_release",
 	[BPF_CGROUP_SOCK_OPS] = "sock_ops",
 	[BPF_CGROUP_DEVICE] = "device",
 	[BPF_CGROUP_INET4_BIND] = "bind4",
Signature · 1 line
2.27.0.212.ge8ba1cc988-goog
maintainer
#7 lore raw

Re: [PATCH bpf-next v2 3/4] bpftool: support BPF_CGROUP_INET_SOCK_RELEASE

Quoted 12 lines · Stanislav Fomichev

Support attaching to sock_release from the bpftool.

---
  tools/bpf/bpftool/main.h | 1 +
  1 file changed, 1 insertion(+)
tools/bpf/bpftool/main.h +0 -0
index 5cdf0bc049bd..0a281d3cceb8 100644

tools/bpf/bpftool/main.h static const char * const attach_type_name[__MAX_BPF_ATTACH_TYPE] = { · line 98

Quoted 7 lines
tools/bpf/bpftool/main.h +1 -0
@@ -92,6 +92,7 @@ static const char * const attach_type_name[__MAX_BPF_ATTACH_TYPE] = {
  	[BPF_CGROUP_INET_INGRESS] = "ingress",
  	[BPF_CGROUP_INET_EGRESS] = "egress",
  	[BPF_CGROUP_INET_SOCK_CREATE] = "sock_create",
+	[BPF_CGROUP_INET_SOCK_RELEASE] = "sock_release",
  	[BPF_CGROUP_SOCK_OPS] = "sock_ops",
  	[BPF_CGROUP_DEVICE] = "device",
  	[BPF_CGROUP_INET4_BIND] = "bind4",

This one is not on latest bpf-next, needs rebase due to 16d37ee3d2b1 ("tools, bpftool: Define attach_type_name array only once").

Thanks,
Daniel

#8 lore raw

Re: [PATCH bpf-next v2 3/4] bpftool: support BPF_CGROUP_INET_SOCK_RELEASE

Quoted 9 lines · Daniel Borkmann
Quoted 7 lines · Stanislav Fomichev

Support attaching to sock_release from the bpftool.

---
  tools/bpf/bpftool/main.h | 1 +
  1 file changed, 1 insertion(+)
Quoted 15 lines

tools/bpf/bpftool/main.h static const char * const attach_type_name[__MAX_BPF_ATTACH_TYPE] = { · line 98

Quoted 12 lines
tools/bpf/bpftool/main.h +1 -0
index 5cdf0bc049bd..0a281d3cceb8 100644
@@ -92,6 +92,7 @@ static const char * const attach_type_name[__MAX_BPF_ATTACH_TYPE] = {
      [BPF_CGROUP_INET_INGRESS] = "ingress",
      [BPF_CGROUP_INET_EGRESS] = "egress",
      [BPF_CGROUP_INET_SOCK_CREATE] = "sock_create",
+     [BPF_CGROUP_INET_SOCK_RELEASE] = "sock_release",
      [BPF_CGROUP_SOCK_OPS] = "sock_ops",
      [BPF_CGROUP_DEVICE] = "device",
      [BPF_CGROUP_INET4_BIND] = "bind4",

This one is not on latest bpf-next, needs rebase due to 16d37ee3d2b1 ("tools, bpftool: Define attach_type_name array only once").

Sure, will follow up with a v3 to address Andrii's suggestions + will rebase on top of the latest bpf-next!

#9 lore raw

PATCH v2 4/4 selftests/bpf: test BPF_CGROUP_INET_SOCK_RELEASE

In reply to Stanislav Fomichev

Simple test that enforces a single SOCK_DGRAM socker per cgroup.

---
 .../selftests/bpf/prog_tests/udp_limit.c      | 71 +++++++++++++++++++
 tools/testing/selftests/bpf/progs/udp_limit.c | 42 +++++++++++
 2 files changed, 113 insertions(+)
 create mode 100644 tools/testing/selftests/bpf/prog_tests/udp_limit.c
 create mode 100644 tools/testing/selftests/bpf/progs/udp_limit.c
tools/testing/selftests/bpf/prog_tests/udp_limit.c new file +71 -0
new file mode 100644
index 000000000000..fe359a927d92
@@ -0,0 +1,71 @@
+// SPDX-License-Identifier: GPL-2.0
+#include <test_progs.h>
+#include "udp_limit.skel.h"
+
+#include <sys/types.h>
+#include <sys/socket.h>
+
+void test_udp_limit(void)
+{
+	struct udp_limit *skel;
+	int cgroup_fd;
+	int fd1, fd2;
+	int err;
+
+	cgroup_fd = test__join_cgroup("/udp_limit");
+	if (CHECK_FAIL(cgroup_fd < 0))
+		return;
+
+	skel = udp_limit__open_and_load();
+	if (CHECK_FAIL(!skel))
+		goto close_cgroup_fd;
+
+	err = bpf_prog_attach(bpf_program__fd(skel->progs.sock),
+			      cgroup_fd, BPF_CGROUP_INET_SOCK_CREATE, 0);
+	if (CHECK_FAIL(err))
+		goto close_skeleton;
+
+	err = bpf_prog_attach(bpf_program__fd(skel->progs.sock_release),
+			      cgroup_fd, BPF_CGROUP_INET_SOCK_RELEASE, 0);
+	if (CHECK_FAIL(err))
+		goto close_skeleton;
+
+	/* BPF program enforces a single UDP socket per cgroup,
+	 * verify that.
+	 */
+	fd1 = socket(AF_INET, SOCK_DGRAM, 0);
+	if (CHECK_FAIL(fd1 < 0))
+		goto close_skeleton;
+
+	fd2 = socket(AF_INET, SOCK_DGRAM, 0);
+	if (CHECK_FAIL(fd2 != -1))
+		goto close_fd1;
+
+	/* We can reopen again after close. */
+	close(fd1);
+
+	fd1 = socket(AF_INET, SOCK_DGRAM, 0);
+	if (CHECK_FAIL(fd1 < 0))
+		goto close_skeleton;
+
+	/* Make sure the program was invoked the expected
+	 * number of times:
+	 * - open fd1           - BPF_CGROUP_INET_SOCK_CREATE
+	 * - attempt to openfd2 - BPF_CGROUP_INET_SOCK_CREATE
+	 * - close fd1          - BPF_CGROUP_INET_SOCK_RELEASE
+	 * - open fd1 again     - BPF_CGROUP_INET_SOCK_CREATE
+	 */
+	if (CHECK_FAIL(skel->bss->invocations != 4))
+		goto close_fd1;
+
+	/* We should still have a single socket in use */
+	if (CHECK_FAIL(skel->bss->in_use != 1))
+		goto close_fd1;
+
+close_fd1:
+	close(fd1);
+close_skeleton:
+	udp_limit__destroy(skel);
+close_cgroup_fd:
+	close(cgroup_fd);
+}
tools/testing/selftests/bpf/progs/udp_limit.c new file +42 -0

42 lines not on this page · raw message

Signature · 1 line
2.27.0.212.ge8ba1cc988-goog