summaryrefslogtreecommitdiffstats
path: root/src/test/test-bpf.c
blob: cd8d68f215b2074b702784da880e231dab5c9a8a (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
/* SPDX-License-Identifier: LGPL-2.1+ */

#include <linux/libbpf.h>
#include <string.h>
#include <sys/mman.h>
#include <unistd.h>

#include "bpf-firewall.h"
#include "bpf-program.h"
#include "load-fragment.h"
#include "manager.h"
#include "rm-rf.h"
#include "service.h"
#include "test-helper.h"
#include "tests.h"
#include "unit.h"

/* We use the same limit here that PID 1 bumps RLIMIT_MEMLOCK to if it can */
#define CAN_MEMLOCK_SIZE (64U*1024U*1024U)

static bool can_memlock(void) {
        void *p;
        bool b;

        /* Let's see if we can mlock() a larger blob of memory. BPF programs are charged against
         * RLIMIT_MEMLOCK, hence let's first make sure we can lock memory at all, and skip the test if we
         * cannot. Why not check RLIMIT_MEMLOCK explicitly? Because in container environments the
         * RLIMIT_MEMLOCK value we see might not match the RLIMIT_MEMLOCK value actually in effect. */

        p = mmap(NULL, CAN_MEMLOCK_SIZE, PROT_READ|PROT_WRITE, MAP_ANONYMOUS|MAP_SHARED, -1, 0);
        if (p == MAP_FAILED)
                return false;

        b = mlock(p, CAN_MEMLOCK_SIZE) >= 0;
        if (b)
                assert_se(munlock(p, CAN_MEMLOCK_SIZE) >= 0);

        assert_se(munmap(p, CAN_MEMLOCK_SIZE) >= 0);
        return b;
}

int main(int argc, char *argv[]) {
        struct bpf_insn exit_insn[] = {
                BPF_MOV64_IMM(BPF_REG_0, 1),
                BPF_EXIT_INSN()
        };

        _cleanup_(rm_rf_physical_and_freep) char *runtime_dir = NULL;
        CGroupContext *cc = NULL;
        _cleanup_(bpf_program_unrefp) BPFProgram *p = NULL;
        _cleanup_(manager_freep) Manager *m = NULL;
        Unit *u;
        char log_buf[65535];
        struct rlimit rl;
        int r;

        test_setup_logging(LOG_DEBUG);

        if (is_run_on_travis_ci())
                return log_tests_skipped("test-bpf fails on Travis CI: https://github.com/systemd/systemd/issues/9666");

        assert_se(getrlimit(RLIMIT_MEMLOCK, &rl) >= 0);
        rl.rlim_cur = rl.rlim_max = MAX3(rl.rlim_cur, rl.rlim_max, CAN_MEMLOCK_SIZE);
        (void) setrlimit(RLIMIT_MEMLOCK, &rl);

        if (!can_memlock())
                return log_tests_skipped("Can't use mlock(), skipping.");

        r = enter_cgroup_subroot();
        if (r == -ENOMEDIUM)
                return log_tests_skipped("cgroupfs not available");

        assert_se(set_unit_path(get_testdata_dir()) >= 0);
        assert_se(runtime_dir = setup_fake_runtime_dir());

        r = bpf_program_new(BPF_PROG_TYPE_CGROUP_SKB, &p);
        assert(r == 0);

        r = bpf_program_add_instructions(p, exit_insn, ELEMENTSOF(exit_insn));
        assert(r == 0);

        if (getuid() != 0)
                return log_tests_skipped("not running as root");

        r = bpf_firewall_supported();
        if (r == BPF_FIREWALL_UNSUPPORTED)
                return log_tests_skipped("BPF firewalling not supported");
        assert_se(r > 0);

        if (r == BPF_FIREWALL_SUPPORTED_WITH_MULTI)
                log_notice("BPF firewalling with BPF_F_ALLOW_MULTI supported. Yay!");
        else
                log_notice("BPF firewalling (though without BPF_F_ALLOW_MULTI) supported. Good.");

        r = bpf_program_load_kernel(p, log_buf, ELEMENTSOF(log_buf));
        assert(r >= 0);

        p = bpf_program_unref(p);

        /* The simple tests suceeded. Now let's try full unit-based use-case. */

        assert_se(manager_new(UNIT_FILE_USER, MANAGER_TEST_RUN_BASIC, &m) >= 0);
        assert_se(manager_startup(m, NULL, NULL) >= 0);

        assert_se(u = unit_new(m, sizeof(Service)));
        assert_se(unit_add_name(u, "foo.service") == 0);
        assert_se(cc = unit_get_cgroup_context(u));
        u->perpetual = true;

        cc->ip_accounting = true;

        assert_se(config_parse_ip_address_access(u->id, "filename", 1, "Service", 1, "IPAddressAllow", 0, "10.0.1.0/24", &cc->ip_address_allow, NULL) == 0);
        assert_se(config_parse_ip_address_access(u->id, "filename", 1, "Service", 1, "IPAddressAllow", 0, "127.0.0.2", &cc->ip_address_allow, NULL) == 0);
        assert_se(config_parse_ip_address_access(u->id, "filename", 1, "Service", 1, "IPAddressDeny", 0, "127.0.0.3", &cc->ip_address_deny, NULL) == 0);
        assert_se(config_parse_ip_address_access(u->id, "filename", 1, "Service", 1, "IPAddressDeny", 0, "10.0.3.2/24", &cc->ip_address_deny, NULL) == 0);
        assert_se(config_parse_ip_address_access(u->id, "filename", 1, "Service", 1, "IPAddressDeny", 0, "127.0.0.1/25", &cc->ip_address_deny, NULL) == 0);
        assert_se(config_parse_ip_address_access(u->id, "filename", 1, "Service", 1, "IPAddressDeny", 0, "127.0.0.4", &cc->ip_address_deny, NULL) == 0);

        assert(cc->ip_address_allow);
        assert(cc->ip_address_allow->items_next);
        assert(!cc->ip_address_allow->items_next->items_next);

        /* The deny list is defined redundantly, let's ensure it got properly reduced */
        assert(cc->ip_address_deny);
        assert(cc->ip_address_deny->items_next);
        assert(!cc->ip_address_deny->items_next->items_next);

        assert_se(config_parse_exec(u->id, "filename", 1, "Service", 1, "ExecStart", SERVICE_EXEC_START, "/bin/ping -c 1 127.0.0.2 -W 5", SERVICE(u)->exec_command, u) == 0);
        assert_se(config_parse_exec(u->id, "filename", 1, "Service", 1, "ExecStart", SERVICE_EXEC_START, "/bin/ping -c 1 127.0.0.3 -W 5", SERVICE(u)->exec_command, u) == 0);

        assert_se(SERVICE(u)->exec_command[SERVICE_EXEC_START]);
        assert_se(SERVICE(u)->exec_command[SERVICE_EXEC_START]->command_next);
        assert_se(!SERVICE(u)->exec_command[SERVICE_EXEC_START]->command_next->command_next);

        SERVICE(u)->type = SERVICE_ONESHOT;
        u->load_state = UNIT_LOADED;

        unit_dump(u, stdout, NULL);

        r = bpf_firewall_compile(u);
        if (IN_SET(r, -ENOTTY, -ENOSYS, -EPERM))
                return log_tests_skipped("Kernel doesn't support the necessary bpf bits (masked out via seccomp?)");
        assert_se(r >= 0);

        assert(u->ip_bpf_ingress);
        assert(u->ip_bpf_egress);

        r = bpf_program_load_kernel(u->ip_bpf_ingress, log_buf, ELEMENTSOF(log_buf));

        log_notice("log:");
        log_notice("-------");
        log_notice("%s", log_buf);
        log_notice("-------");

        assert(r >= 0);

        r = bpf_program_load_kernel(u->ip_bpf_egress, log_buf, ELEMENTSOF(log_buf));

        log_notice("log:");
        log_notice("-------");
        log_notice("%s", log_buf);
        log_notice("-------");

        assert(r >= 0);

        assert_se(unit_start(u) >= 0);

        while (!IN_SET(SERVICE(u)->state, SERVICE_DEAD, SERVICE_FAILED))
                assert_se(sd_event_run(m->event, UINT64_MAX) >= 0);

        assert_se(SERVICE(u)->exec_command[SERVICE_EXEC_START]->exec_status.code == CLD_EXITED &&
                  SERVICE(u)->exec_command[SERVICE_EXEC_START]->exec_status.status == EXIT_SUCCESS);

        assert_se(SERVICE(u)->exec_command[SERVICE_EXEC_START]->command_next->exec_status.code != CLD_EXITED ||
                  SERVICE(u)->exec_command[SERVICE_EXEC_START]->command_next->exec_status.status != EXIT_SUCCESS);

        return 0;
}