samples: bpf: Refactor test_cgrp2_sock2 program with libbpf
authorDaniel T. Lee <danieltimlee@gmail.com>
Tue, 24 Nov 2020 09:03:05 +0000 (09:03 +0000)
committerAndrii Nakryiko <andrii@kernel.org>
Fri, 27 Nov 2020 03:33:35 +0000 (19:33 -0800)
This commit refactors the existing cgroup program with libbpf bpf
loader. The original test_cgrp2_sock2 has keeped the bpf program
attached to the cgroup hierarchy even after the exit of user program.
To implement the same functionality with libbpf, this commit uses the
BPF_LINK_PINNING to pin the link attachment even after it is closed.

Since this uses LINK instead of ATTACH, detach of bpf program from
cgroup with 'test_cgrp2_sock' is not used anymore.

The code to mount the bpf was added to the .sh file in case the bpff
was not mounted on /sys/fs/bpf. Additionally, to fix the problem that
shell script cannot find the binary object from the current path,
relative path './' has been added in front of binary.

Fixes: 554ae6e792ef3 ("samples/bpf: add userspace example for prohibiting sockets")
Signed-off-by: Daniel T. Lee <danieltimlee@gmail.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20201124090310.24374-3-danieltimlee@gmail.com
samples/bpf/Makefile
samples/bpf/test_cgrp2_sock2.c
samples/bpf/test_cgrp2_sock2.sh

index 7c61118..d31e082 100644 (file)
@@ -82,7 +82,7 @@ test_overhead-objs := bpf_load.o test_overhead_user.o
 test_cgrp2_array_pin-objs := test_cgrp2_array_pin.o
 test_cgrp2_attach-objs := test_cgrp2_attach.o
 test_cgrp2_sock-objs := test_cgrp2_sock.o
-test_cgrp2_sock2-objs := bpf_load.o test_cgrp2_sock2.o
+test_cgrp2_sock2-objs := test_cgrp2_sock2.o
 xdp1-objs := xdp1_user.o
 # reuse xdp1 source intentionally
 xdp2-objs := xdp1_user.o
index a9277b1..e7060aa 100644 (file)
@@ -20,9 +20,9 @@
 #include <net/if.h>
 #include <linux/bpf.h>
 #include <bpf/bpf.h>
+#include <bpf/libbpf.h>
 
 #include "bpf_insn.h"
-#include "bpf_load.h"
 
 static int usage(const char *argv0)
 {
@@ -32,37 +32,64 @@ static int usage(const char *argv0)
 
 int main(int argc, char **argv)
 {
-       int cg_fd, ret, filter_id = 0;
+       int cg_fd, err, ret = EXIT_FAILURE, filter_id = 0, prog_cnt = 0;
+       const char *link_pin_path = "/sys/fs/bpf/test_cgrp2_sock2";
+       struct bpf_link *link = NULL;
+       struct bpf_program *progs[2];
+       struct bpf_program *prog;
+       struct bpf_object *obj;
 
        if (argc < 3)
                return usage(argv[0]);
 
+       if (argc > 3)
+               filter_id = atoi(argv[3]);
+
        cg_fd = open(argv[1], O_DIRECTORY | O_RDONLY);
        if (cg_fd < 0) {
                printf("Failed to open cgroup path: '%s'\n", strerror(errno));
-               return EXIT_FAILURE;
+               return ret;
        }
 
-       if (load_bpf_file(argv[2]))
-               return EXIT_FAILURE;
-
-       printf("Output from kernel verifier:\n%s\n-------\n", bpf_log_buf);
+       obj = bpf_object__open_file(argv[2], NULL);
+       if (libbpf_get_error(obj)) {
+               printf("ERROR: opening BPF object file failed\n");
+               return ret;
+       }
 
-       if (argc > 3)
-               filter_id = atoi(argv[3]);
+       bpf_object__for_each_program(prog, obj) {
+               progs[prog_cnt] = prog;
+               prog_cnt++;
+       }
 
        if (filter_id >= prog_cnt) {
                printf("Invalid program id; program not found in file\n");
-               return EXIT_FAILURE;
+               goto cleanup;
+       }
+
+       /* load BPF program */
+       if (bpf_object__load(obj)) {
+               printf("ERROR: loading BPF object file failed\n");
+               goto cleanup;
        }
 
-       ret = bpf_prog_attach(prog_fd[filter_id], cg_fd,
-                             BPF_CGROUP_INET_SOCK_CREATE, 0);
-       if (ret < 0) {
-               printf("Failed to attach prog to cgroup: '%s'\n",
-                      strerror(errno));
-               return EXIT_FAILURE;
+       link = bpf_program__attach_cgroup(progs[filter_id], cg_fd);
+       if (libbpf_get_error(link)) {
+               printf("ERROR: bpf_program__attach failed\n");
+               link = NULL;
+               goto cleanup;
        }
 
-       return EXIT_SUCCESS;
+       err = bpf_link__pin(link, link_pin_path);
+       if (err < 0) {
+               printf("ERROR: bpf_link__pin failed: %d\n", err);
+               goto cleanup;
+       }
+
+       ret = EXIT_SUCCESS;
+
+cleanup:
+       bpf_link__destroy(link);
+       bpf_object__close(obj);
+       return ret;
 }
index 0f396a8..6a3dbe6 100755 (executable)
@@ -1,6 +1,9 @@
 #!/bin/bash
 # SPDX-License-Identifier: GPL-2.0
 
+BPFFS=/sys/fs/bpf
+LINK_PIN=$BPFFS/test_cgrp2_sock2
+
 function config_device {
        ip netns add at_ns0
        ip link add veth0 type veth peer name veth0b
@@ -21,16 +24,22 @@ function config_cgroup {
        echo $$ >> /tmp/cgroupv2/foo/cgroup.procs
 }
 
+function config_bpffs {
+       if mount | grep $BPFFS > /dev/null; then
+               echo "bpffs already mounted"
+       else
+               echo "bpffs not mounted. Mounting..."
+               mount -t bpf none $BPFFS
+       fi
+}
 
 function attach_bpf {
-       test_cgrp2_sock2 /tmp/cgroupv2/foo sock_flags_kern.o $1
+       ./test_cgrp2_sock2 /tmp/cgroupv2/foo sock_flags_kern.o $1
        [ $? -ne 0 ] && exit 1
 }
 
 function cleanup {
-       if [ -d /tmp/cgroupv2/foo ]; then
-               test_cgrp2_sock -d /tmp/cgroupv2/foo
-       fi
+       rm -rf $LINK_PIN
        ip link del veth0b
        ip netns delete at_ns0
        umount /tmp/cgroupv2
@@ -42,6 +51,7 @@ cleanup 2>/dev/null
 set -e
 config_device
 config_cgroup
+config_bpffs
 set +e
 
 #
@@ -62,6 +72,9 @@ if [ $? -eq 0 ]; then
        exit 1
 fi
 
+rm -rf $LINK_PIN
+sleep 1                 # Wait for link detach
+
 #
 # Test 2 - fail ping
 #