selftests/bpf: Selftest for sys_bind post-hooks.
authorAndrey Ignatov <rdna@fb.com>
Fri, 30 Mar 2018 22:08:08 +0000 (15:08 -0700)
committerDaniel Borkmann <daniel@iogearbox.net>
Sat, 31 Mar 2018 00:16:40 +0000 (02:16 +0200)
Add selftest for attach types `BPF_CGROUP_INET4_POST_BIND` and
`BPF_CGROUP_INET6_POST_BIND`.

The main things tested are:
* prog load behaves as expected (valid/invalid accesses in prog);
* prog attach behaves as expected (load- vs attach-time attach types);
* `BPF_CGROUP_INET_SOCK_CREATE` can be attached in a backward compatible
  way;
* post-hooks return expected result and errno.

Example:
  # ./test_sock
  Test case: bind4 load with invalid access: src_ip6 .. [PASS]
  Test case: bind4 load with invalid access: mark .. [PASS]
  Test case: bind6 load with invalid access: src_ip4 .. [PASS]
  Test case: sock_create load with invalid access: src_port .. [PASS]
  Test case: sock_create load w/o expected_attach_type (compat mode) ..
  [PASS]
  Test case: sock_create load w/ expected_attach_type .. [PASS]
  Test case: attach type mismatch bind4 vs bind6 .. [PASS]
  Test case: attach type mismatch bind6 vs bind4 .. [PASS]
  Test case: attach type mismatch default vs bind4 .. [PASS]
  Test case: attach type mismatch bind6 vs sock_create .. [PASS]
  Test case: bind4 reject all .. [PASS]
  Test case: bind6 reject all .. [PASS]
  Test case: bind6 deny specific IP & port .. [PASS]
  Test case: bind4 allow specific IP & port .. [PASS]
  Test case: bind4 allow all .. [PASS]
  Test case: bind6 allow all .. [PASS]
  Summary: 16 PASSED, 0 FAILED

Signed-off-by: Andrey Ignatov <rdna@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
tools/include/uapi/linux/bpf.h
tools/testing/selftests/bpf/Makefile
tools/testing/selftests/bpf/test_sock.c [new file with mode: 0644]

index 71051d01e8dd5add858fb35ee896c3835ed98073..9d07465023a263ba6d1e357d43a97b8ef550941b 100644 (file)
@@ -152,6 +152,8 @@ enum bpf_attach_type {
        BPF_CGROUP_INET6_BIND,
        BPF_CGROUP_INET4_CONNECT,
        BPF_CGROUP_INET6_CONNECT,
        BPF_CGROUP_INET6_BIND,
        BPF_CGROUP_INET4_CONNECT,
        BPF_CGROUP_INET6_CONNECT,
+       BPF_CGROUP_INET4_POST_BIND,
+       BPF_CGROUP_INET6_POST_BIND,
        __MAX_BPF_ATTACH_TYPE
 };
 
        __MAX_BPF_ATTACH_TYPE
 };
 
@@ -947,6 +949,15 @@ struct bpf_sock {
        __u32 protocol;
        __u32 mark;
        __u32 priority;
        __u32 protocol;
        __u32 mark;
        __u32 priority;
+       __u32 src_ip4;          /* Allows 1,2,4-byte read.
+                                * Stored in network byte order.
+                                */
+       __u32 src_ip6[4];       /* Allows 1,2,4-byte read.
+                                * Stored in network byte order.
+                                */
+       __u32 src_port;         /* Allows 4-byte read.
+                                * Stored in host byte order
+                                */
 };
 
 #define XDP_PACKET_HEADROOM 256
 };
 
 #define XDP_PACKET_HEADROOM 256
index c64d4ebc77ff0eb4f4418aaeb6aa10f2abc37d11..0a315ddabbf46508e0bce464f52b0948fbf1cf4e 100644 (file)
@@ -23,7 +23,8 @@ urandom_read: urandom_read.c
 
 # Order correspond to 'make run_tests' order
 TEST_GEN_PROGS = test_verifier test_tag test_maps test_lru_map test_lpm_map test_progs \
 
 # Order correspond to 'make run_tests' order
 TEST_GEN_PROGS = test_verifier test_tag test_maps test_lru_map test_lpm_map test_progs \
-       test_align test_verifier_log test_dev_cgroup test_tcpbpf_user test_sock_addr
+       test_align test_verifier_log test_dev_cgroup test_tcpbpf_user \
+       test_sock test_sock_addr
 
 TEST_GEN_FILES = test_pkt_access.o test_xdp.o test_l4lb.o test_tcp_estats.o test_obj_id.o \
        test_pkt_md_access.o test_xdp_redirect.o test_xdp_meta.o sockmap_parse_prog.o     \
 
 TEST_GEN_FILES = test_pkt_access.o test_xdp.o test_l4lb.o test_tcp_estats.o test_obj_id.o \
        test_pkt_md_access.o test_xdp_redirect.o test_xdp_meta.o sockmap_parse_prog.o     \
@@ -52,6 +53,7 @@ $(TEST_GEN_PROGS): $(BPFOBJ)
 $(TEST_GEN_PROGS_EXTENDED): $(OUTPUT)/libbpf.a
 
 $(OUTPUT)/test_dev_cgroup: cgroup_helpers.c
 $(TEST_GEN_PROGS_EXTENDED): $(OUTPUT)/libbpf.a
 
 $(OUTPUT)/test_dev_cgroup: cgroup_helpers.c
+$(OUTPUT)/test_sock: cgroup_helpers.c
 $(OUTPUT)/test_sock_addr: cgroup_helpers.c
 
 .PHONY: force
 $(OUTPUT)/test_sock_addr: cgroup_helpers.c
 
 .PHONY: force
diff --git a/tools/testing/selftests/bpf/test_sock.c b/tools/testing/selftests/bpf/test_sock.c
new file mode 100644 (file)
index 0000000..73bb20c
--- /dev/null
@@ -0,0 +1,479 @@
+// SPDX-License-Identifier: GPL-2.0
+// Copyright (c) 2018 Facebook
+
+#include <stdio.h>
+#include <unistd.h>
+
+#include <arpa/inet.h>
+#include <sys/types.h>
+#include <sys/socket.h>
+
+#include <linux/filter.h>
+
+#include <bpf/bpf.h>
+
+#include "cgroup_helpers.h"
+
+#ifndef ARRAY_SIZE
+# define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
+#endif
+
+#define CG_PATH                "/foo"
+#define MAX_INSNS      512
+
+char bpf_log_buf[BPF_LOG_BUF_SIZE];
+
+struct sock_test {
+       const char *descr;
+       /* BPF prog properties */
+       struct bpf_insn insns[MAX_INSNS];
+       enum bpf_attach_type expected_attach_type;
+       enum bpf_attach_type attach_type;
+       /* Socket properties */
+       int domain;
+       int type;
+       /* Endpoint to bind() to */
+       const char *ip;
+       unsigned short port;
+       /* Expected test result */
+       enum {
+               LOAD_REJECT,
+               ATTACH_REJECT,
+               BIND_REJECT,
+               SUCCESS,
+       } result;
+};
+
+static struct sock_test tests[] = {
+       {
+               "bind4 load with invalid access: src_ip6",
+               .insns = {
+                       BPF_MOV64_REG(BPF_REG_6, BPF_REG_1),
+                       BPF_LDX_MEM(BPF_W, BPF_REG_7, BPF_REG_6,
+                                   offsetof(struct bpf_sock, src_ip6[0])),
+                       BPF_MOV64_IMM(BPF_REG_0, 1),
+                       BPF_EXIT_INSN(),
+               },
+               BPF_CGROUP_INET4_POST_BIND,
+               BPF_CGROUP_INET4_POST_BIND,
+               0,
+               0,
+               NULL,
+               0,
+               LOAD_REJECT,
+       },
+       {
+               "bind4 load with invalid access: mark",
+               .insns = {
+                       BPF_MOV64_REG(BPF_REG_6, BPF_REG_1),
+                       BPF_LDX_MEM(BPF_W, BPF_REG_7, BPF_REG_6,
+                                   offsetof(struct bpf_sock, mark)),
+                       BPF_MOV64_IMM(BPF_REG_0, 1),
+                       BPF_EXIT_INSN(),
+               },
+               BPF_CGROUP_INET4_POST_BIND,
+               BPF_CGROUP_INET4_POST_BIND,
+               0,
+               0,
+               NULL,
+               0,
+               LOAD_REJECT,
+       },
+       {
+               "bind6 load with invalid access: src_ip4",
+               .insns = {
+                       BPF_MOV64_REG(BPF_REG_6, BPF_REG_1),
+                       BPF_LDX_MEM(BPF_W, BPF_REG_7, BPF_REG_6,
+                                   offsetof(struct bpf_sock, src_ip4)),
+                       BPF_MOV64_IMM(BPF_REG_0, 1),
+                       BPF_EXIT_INSN(),
+               },
+               BPF_CGROUP_INET6_POST_BIND,
+               BPF_CGROUP_INET6_POST_BIND,
+               0,
+               0,
+               NULL,
+               0,
+               LOAD_REJECT,
+       },
+       {
+               "sock_create load with invalid access: src_port",
+               .insns = {
+                       BPF_MOV64_REG(BPF_REG_6, BPF_REG_1),
+                       BPF_LDX_MEM(BPF_W, BPF_REG_7, BPF_REG_6,
+                                   offsetof(struct bpf_sock, src_port)),
+                       BPF_MOV64_IMM(BPF_REG_0, 1),
+                       BPF_EXIT_INSN(),
+               },
+               BPF_CGROUP_INET_SOCK_CREATE,
+               BPF_CGROUP_INET_SOCK_CREATE,
+               0,
+               0,
+               NULL,
+               0,
+               LOAD_REJECT,
+       },
+       {
+               "sock_create load w/o expected_attach_type (compat mode)",
+               .insns = {
+                       BPF_MOV64_IMM(BPF_REG_0, 1),
+                       BPF_EXIT_INSN(),
+               },
+               0,
+               BPF_CGROUP_INET_SOCK_CREATE,
+               AF_INET,
+               SOCK_STREAM,
+               "127.0.0.1",
+               8097,
+               SUCCESS,
+       },
+       {
+               "sock_create load w/ expected_attach_type",
+               .insns = {
+                       BPF_MOV64_IMM(BPF_REG_0, 1),
+                       BPF_EXIT_INSN(),
+               },
+               BPF_CGROUP_INET_SOCK_CREATE,
+               BPF_CGROUP_INET_SOCK_CREATE,
+               AF_INET,
+               SOCK_STREAM,
+               "127.0.0.1",
+               8097,
+               SUCCESS,
+       },
+       {
+               "attach type mismatch bind4 vs bind6",
+               .insns = {
+                       BPF_MOV64_IMM(BPF_REG_0, 1),
+                       BPF_EXIT_INSN(),
+               },
+               BPF_CGROUP_INET4_POST_BIND,
+               BPF_CGROUP_INET6_POST_BIND,
+               0,
+               0,
+               NULL,
+               0,
+               ATTACH_REJECT,
+       },
+       {
+               "attach type mismatch bind6 vs bind4",
+               .insns = {
+                       BPF_MOV64_IMM(BPF_REG_0, 1),
+                       BPF_EXIT_INSN(),
+               },
+               BPF_CGROUP_INET6_POST_BIND,
+               BPF_CGROUP_INET4_POST_BIND,
+               0,
+               0,
+               NULL,
+               0,
+               ATTACH_REJECT,
+       },
+       {
+               "attach type mismatch default vs bind4",
+               .insns = {
+                       BPF_MOV64_IMM(BPF_REG_0, 1),
+                       BPF_EXIT_INSN(),
+               },
+               0,
+               BPF_CGROUP_INET4_POST_BIND,
+               0,
+               0,
+               NULL,
+               0,
+               ATTACH_REJECT,
+       },
+       {
+               "attach type mismatch bind6 vs sock_create",
+               .insns = {
+                       BPF_MOV64_IMM(BPF_REG_0, 1),
+                       BPF_EXIT_INSN(),
+               },
+               BPF_CGROUP_INET6_POST_BIND,
+               BPF_CGROUP_INET_SOCK_CREATE,
+               0,
+               0,
+               NULL,
+               0,
+               ATTACH_REJECT,
+       },
+       {
+               "bind4 reject all",
+               .insns = {
+                       BPF_MOV64_IMM(BPF_REG_0, 0),
+                       BPF_EXIT_INSN(),
+               },
+               BPF_CGROUP_INET4_POST_BIND,
+               BPF_CGROUP_INET4_POST_BIND,
+               AF_INET,
+               SOCK_STREAM,
+               "0.0.0.0",
+               0,
+               BIND_REJECT,
+       },
+       {
+               "bind6 reject all",
+               .insns = {
+                       BPF_MOV64_IMM(BPF_REG_0, 0),
+                       BPF_EXIT_INSN(),
+               },
+               BPF_CGROUP_INET6_POST_BIND,
+               BPF_CGROUP_INET6_POST_BIND,
+               AF_INET6,
+               SOCK_STREAM,
+               "::",
+               0,
+               BIND_REJECT,
+       },
+       {
+               "bind6 deny specific IP & port",
+               .insns = {
+                       BPF_MOV64_REG(BPF_REG_6, BPF_REG_1),
+
+                       /* if (ip == expected && port == expected) */
+                       BPF_LDX_MEM(BPF_W, BPF_REG_7, BPF_REG_6,
+                                   offsetof(struct bpf_sock, src_ip6[3])),
+                       BPF_JMP_IMM(BPF_JNE, BPF_REG_7, 0x01000000, 4),
+                       BPF_LDX_MEM(BPF_W, BPF_REG_7, BPF_REG_6,
+                                   offsetof(struct bpf_sock, src_port)),
+                       BPF_JMP_IMM(BPF_JNE, BPF_REG_7, 0x2001, 2),
+
+                       /* return DENY; */
+                       BPF_MOV64_IMM(BPF_REG_0, 0),
+                       BPF_JMP_A(1),
+
+                       /* else return ALLOW; */
+                       BPF_MOV64_IMM(BPF_REG_0, 1),
+                       BPF_EXIT_INSN(),
+               },
+               BPF_CGROUP_INET6_POST_BIND,
+               BPF_CGROUP_INET6_POST_BIND,
+               AF_INET6,
+               SOCK_STREAM,
+               "::1",
+               8193,
+               BIND_REJECT,
+       },
+       {
+               "bind4 allow specific IP & port",
+               .insns = {
+                       BPF_MOV64_REG(BPF_REG_6, BPF_REG_1),
+
+                       /* if (ip == expected && port == expected) */
+                       BPF_LDX_MEM(BPF_W, BPF_REG_7, BPF_REG_6,
+                                   offsetof(struct bpf_sock, src_ip4)),
+                       BPF_JMP_IMM(BPF_JNE, BPF_REG_7, 0x0100007F, 4),
+                       BPF_LDX_MEM(BPF_W, BPF_REG_7, BPF_REG_6,
+                                   offsetof(struct bpf_sock, src_port)),
+                       BPF_JMP_IMM(BPF_JNE, BPF_REG_7, 0x1002, 2),
+
+                       /* return ALLOW; */
+                       BPF_MOV64_IMM(BPF_REG_0, 1),
+                       BPF_JMP_A(1),
+
+                       /* else return DENY; */
+                       BPF_MOV64_IMM(BPF_REG_0, 0),
+                       BPF_EXIT_INSN(),
+               },
+               BPF_CGROUP_INET4_POST_BIND,
+               BPF_CGROUP_INET4_POST_BIND,
+               AF_INET,
+               SOCK_STREAM,
+               "127.0.0.1",
+               4098,
+               SUCCESS,
+       },
+       {
+               "bind4 allow all",
+               .insns = {
+                       BPF_MOV64_IMM(BPF_REG_0, 1),
+                       BPF_EXIT_INSN(),
+               },
+               BPF_CGROUP_INET4_POST_BIND,
+               BPF_CGROUP_INET4_POST_BIND,
+               AF_INET,
+               SOCK_STREAM,
+               "0.0.0.0",
+               0,
+               SUCCESS,
+       },
+       {
+               "bind6 allow all",
+               .insns = {
+                       BPF_MOV64_IMM(BPF_REG_0, 1),
+                       BPF_EXIT_INSN(),
+               },
+               BPF_CGROUP_INET6_POST_BIND,
+               BPF_CGROUP_INET6_POST_BIND,
+               AF_INET6,
+               SOCK_STREAM,
+               "::",
+               0,
+               SUCCESS,
+       },
+};
+
+static size_t probe_prog_length(const struct bpf_insn *fp)
+{
+       size_t len;
+
+       for (len = MAX_INSNS - 1; len > 0; --len)
+               if (fp[len].code != 0 || fp[len].imm != 0)
+                       break;
+       return len + 1;
+}
+
+static int load_sock_prog(const struct bpf_insn *prog,
+                         enum bpf_attach_type attach_type)
+{
+       struct bpf_load_program_attr attr;
+
+       memset(&attr, 0, sizeof(struct bpf_load_program_attr));
+       attr.prog_type = BPF_PROG_TYPE_CGROUP_SOCK;
+       attr.expected_attach_type = attach_type;
+       attr.insns = prog;
+       attr.insns_cnt = probe_prog_length(attr.insns);
+       attr.license = "GPL";
+
+       return bpf_load_program_xattr(&attr, bpf_log_buf, BPF_LOG_BUF_SIZE);
+}
+
+static int attach_sock_prog(int cgfd, int progfd,
+                           enum bpf_attach_type attach_type)
+{
+       return bpf_prog_attach(progfd, cgfd, attach_type, BPF_F_ALLOW_OVERRIDE);
+}
+
+static int bind_sock(int domain, int type, const char *ip, unsigned short port)
+{
+       struct sockaddr_storage addr;
+       struct sockaddr_in6 *addr6;
+       struct sockaddr_in *addr4;
+       int sockfd = -1;
+       socklen_t len;
+       int err = 0;
+
+       sockfd = socket(domain, type, 0);
+       if (sockfd < 0)
+               goto err;
+
+       memset(&addr, 0, sizeof(addr));
+
+       if (domain == AF_INET) {
+               len = sizeof(struct sockaddr_in);
+               addr4 = (struct sockaddr_in *)&addr;
+               addr4->sin_family = domain;
+               addr4->sin_port = htons(port);
+               if (inet_pton(domain, ip, (void *)&addr4->sin_addr) != 1)
+                       goto err;
+       } else if (domain == AF_INET6) {
+               len = sizeof(struct sockaddr_in6);
+               addr6 = (struct sockaddr_in6 *)&addr;
+               addr6->sin6_family = domain;
+               addr6->sin6_port = htons(port);
+               if (inet_pton(domain, ip, (void *)&addr6->sin6_addr) != 1)
+                       goto err;
+       } else {
+               goto err;
+       }
+
+       if (bind(sockfd, (const struct sockaddr *)&addr, len) == -1)
+               goto err;
+
+       goto out;
+err:
+       err = -1;
+out:
+       close(sockfd);
+       return err;
+}
+
+static int run_test_case(int cgfd, const struct sock_test *test)
+{
+       int progfd = -1;
+       int err = 0;
+
+       printf("Test case: %s .. ", test->descr);
+       progfd = load_sock_prog(test->insns, test->expected_attach_type);
+       if (progfd < 0) {
+               if (test->result == LOAD_REJECT)
+                       goto out;
+               else
+                       goto err;
+       }
+
+       if (attach_sock_prog(cgfd, progfd, test->attach_type) == -1) {
+               if (test->result == ATTACH_REJECT)
+                       goto out;
+               else
+                       goto err;
+       }
+
+       if (bind_sock(test->domain, test->type, test->ip, test->port) == -1) {
+               /* sys_bind() may fail for different reasons, errno has to be
+                * checked to confirm that BPF program rejected it.
+                */
+               if (test->result == BIND_REJECT && errno == EPERM)
+                       goto out;
+               else
+                       goto err;
+       }
+
+
+       if (test->result != SUCCESS)
+               goto err;
+
+       goto out;
+err:
+       err = -1;
+out:
+       /* Detaching w/o checking return code: best effort attempt. */
+       if (progfd != -1)
+               bpf_prog_detach(cgfd, test->attach_type);
+       close(progfd);
+       printf("[%s]\n", err ? "FAIL" : "PASS");
+       return err;
+}
+
+static int run_tests(int cgfd)
+{
+       int passes = 0;
+       int fails = 0;
+       int i;
+
+       for (i = 0; i < ARRAY_SIZE(tests); ++i) {
+               if (run_test_case(cgfd, &tests[i]))
+                       ++fails;
+               else
+                       ++passes;
+       }
+       printf("Summary: %d PASSED, %d FAILED\n", passes, fails);
+       return fails ? -1 : 0;
+}
+
+int main(int argc, char **argv)
+{
+       int cgfd = -1;
+       int err = 0;
+
+       if (setup_cgroup_environment())
+               goto err;
+
+       cgfd = create_and_get_cgroup(CG_PATH);
+       if (!cgfd)
+               goto err;
+
+       if (join_cgroup(CG_PATH))
+               goto err;
+
+       if (run_tests(cgfd))
+               goto err;
+
+       goto out;
+err:
+       err = -1;
+out:
+       close(cgfd);
+       cleanup_cgroup_environment();
+       return err;
+}