bpf: Add ene-to-end test for bpf_sk_storage_* helpers
authorMartin KaFai Lau <kafai@fb.com>
Fri, 26 Apr 2019 23:39:54 +0000 (16:39 -0700)
committerAlexei Starovoitov <ast@kernel.org>
Sat, 27 Apr 2019 16:07:05 +0000 (09:07 -0700)
This patch rides on an existing BPF_PROG_TYPE_CGROUP_SKB test
(test_sock_fields.c) to do a TCP end-to-end test on the new
bpf_sk_storage_* helpers.

Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
tools/testing/selftests/bpf/bpf_helpers.h
tools/testing/selftests/bpf/progs/test_sock_fields_kern.c
tools/testing/selftests/bpf/test_sock_fields.c

index 59e221586cf7ccfed4024bbdfe39a7d66bdae7bb..6e80b66d7fb1d7a715782bdbffe74fbea005d0a6 100644 (file)
@@ -211,6 +211,11 @@ static int (*bpf_strtol)(const char *buf, unsigned long long buf_len,
 static int (*bpf_strtoul)(const char *buf, unsigned long long buf_len,
                          unsigned long long flags, unsigned long *res) =
        (void *) BPF_FUNC_strtoul;
+static void *(*bpf_sk_storage_get)(void *map, struct bpf_sock *sk,
+                                  void *value, __u64 flags) =
+       (void *) BPF_FUNC_sk_storage_get;
+static int (*bpf_sk_storage_delete)(void *map, struct bpf_sock *sk) =
+       (void *)BPF_FUNC_sk_storage_delete;
 
 /* llvm builtin functions that eBPF C program may use to
  * emit BPF_LD_ABS and BPF_LD_IND instructions
index 37328f1485384b756c3120f428a3348206a9eb47..1c39e4ccb7f1bf208b7debd62c633f1f719794c8 100644 (file)
@@ -55,6 +55,31 @@ struct bpf_map_def SEC("maps") linum_map = {
        .max_entries = __NR_BPF_LINUM_ARRAY_IDX,
 };
 
+struct bpf_spinlock_cnt {
+       struct bpf_spin_lock lock;
+       __u32 cnt;
+};
+
+struct bpf_map_def SEC("maps") sk_pkt_out_cnt = {
+       .type = BPF_MAP_TYPE_SK_STORAGE,
+       .key_size = sizeof(int),
+       .value_size = sizeof(struct bpf_spinlock_cnt),
+       .max_entries = 0,
+       .map_flags = BPF_F_NO_PREALLOC,
+};
+
+BPF_ANNOTATE_KV_PAIR(sk_pkt_out_cnt, int, struct bpf_spinlock_cnt);
+
+struct bpf_map_def SEC("maps") sk_pkt_out_cnt10 = {
+       .type = BPF_MAP_TYPE_SK_STORAGE,
+       .key_size = sizeof(int),
+       .value_size = sizeof(struct bpf_spinlock_cnt),
+       .max_entries = 0,
+       .map_flags = BPF_F_NO_PREALLOC,
+};
+
+BPF_ANNOTATE_KV_PAIR(sk_pkt_out_cnt10, int, struct bpf_spinlock_cnt);
+
 static bool is_loopback6(__u32 *a6)
 {
        return !a6[0] && !a6[1] && !a6[2] && a6[3] == bpf_htonl(1);
@@ -120,7 +145,9 @@ static void tpcpy(struct bpf_tcp_sock *dst,
 SEC("cgroup_skb/egress")
 int egress_read_sock_fields(struct __sk_buff *skb)
 {
+       struct bpf_spinlock_cnt cli_cnt_init = { .lock = 0, .cnt = 0xeB9F };
        __u32 srv_idx = ADDR_SRV_IDX, cli_idx = ADDR_CLI_IDX, result_idx;
+       struct bpf_spinlock_cnt *pkt_out_cnt, *pkt_out_cnt10;
        struct sockaddr_in6 *srv_sa6, *cli_sa6;
        struct bpf_tcp_sock *tp, *tp_ret;
        struct bpf_sock *sk, *sk_ret;
@@ -161,6 +188,32 @@ int egress_read_sock_fields(struct __sk_buff *skb)
        skcpy(sk_ret, sk);
        tpcpy(tp_ret, tp);
 
+       if (result_idx == EGRESS_SRV_IDX) {
+               /* The userspace has created it for srv sk */
+               pkt_out_cnt = bpf_sk_storage_get(&sk_pkt_out_cnt, sk, 0, 0);
+               pkt_out_cnt10 = bpf_sk_storage_get(&sk_pkt_out_cnt10, sk,
+                                                  0, 0);
+       } else {
+               pkt_out_cnt = bpf_sk_storage_get(&sk_pkt_out_cnt, sk,
+                                                &cli_cnt_init,
+                                                BPF_SK_STORAGE_GET_F_CREATE);
+               pkt_out_cnt10 = bpf_sk_storage_get(&sk_pkt_out_cnt10,
+                                                  sk, &cli_cnt_init,
+                                                  BPF_SK_STORAGE_GET_F_CREATE);
+       }
+
+       if (!pkt_out_cnt || !pkt_out_cnt10)
+               RETURN;
+
+       /* Even both cnt and cnt10 have lock defined in their BTF,
+        * intentionally one cnt takes lock while one does not
+        * as a test for the spinlock support in BPF_MAP_TYPE_SK_STORAGE.
+        */
+       pkt_out_cnt->cnt += 1;
+       bpf_spin_lock(&pkt_out_cnt10->lock);
+       pkt_out_cnt10->cnt += 10;
+       bpf_spin_unlock(&pkt_out_cnt10->lock);
+
        RETURN;
 }
 
index dcae7f664dce0827f5ddda4bc5f693d191def9b0..e089477fa0a3ec98cdf29450f8d133fe7b492933 100644 (file)
@@ -35,6 +35,11 @@ enum bpf_linum_array_idx {
        __NR_BPF_LINUM_ARRAY_IDX,
 };
 
+struct bpf_spinlock_cnt {
+       struct bpf_spin_lock lock;
+       __u32 cnt;
+};
+
 #define CHECK(condition, tag, format...) ({                            \
        int __ret = !!(condition);                                      \
        if (__ret) {                                                    \
@@ -50,6 +55,8 @@ enum bpf_linum_array_idx {
 #define DATA_LEN sizeof(DATA)
 
 static struct sockaddr_in6 srv_sa6, cli_sa6;
+static int sk_pkt_out_cnt10_fd;
+static int sk_pkt_out_cnt_fd;
 static int linum_map_fd;
 static int addr_map_fd;
 static int tp_map_fd;
@@ -220,28 +227,90 @@ static void check_result(void)
              "Unexpected listen_tp", "Check listen_tp output. ingress_linum:%u",
              ingress_linum);
 
-       CHECK(srv_tp.data_segs_out != 1 ||
+       CHECK(srv_tp.data_segs_out != 2 ||
              srv_tp.data_segs_in ||
              srv_tp.snd_cwnd != 10 ||
              srv_tp.total_retrans ||
-             srv_tp.bytes_acked != DATA_LEN,
+             srv_tp.bytes_acked != 2 * DATA_LEN,
              "Unexpected srv_tp", "Check srv_tp output. egress_linum:%u",
              egress_linum);
 
        CHECK(cli_tp.data_segs_out ||
-             cli_tp.data_segs_in != 1 ||
+             cli_tp.data_segs_in != 2 ||
              cli_tp.snd_cwnd != 10 ||
              cli_tp.total_retrans ||
-             cli_tp.bytes_received != DATA_LEN,
+             cli_tp.bytes_received != 2 * DATA_LEN,
              "Unexpected cli_tp", "Check cli_tp output. egress_linum:%u",
              egress_linum);
 }
 
+static void check_sk_pkt_out_cnt(int accept_fd, int cli_fd)
+{
+       struct bpf_spinlock_cnt pkt_out_cnt = {}, pkt_out_cnt10 = {};
+       int err;
+
+       pkt_out_cnt.cnt = ~0;
+       pkt_out_cnt10.cnt = ~0;
+       err = bpf_map_lookup_elem(sk_pkt_out_cnt_fd, &accept_fd, &pkt_out_cnt);
+       if (!err)
+               err = bpf_map_lookup_elem(sk_pkt_out_cnt10_fd, &accept_fd,
+                                         &pkt_out_cnt10);
+
+       /* The bpf prog only counts for fullsock and
+        * passive conneciton did not become fullsock until 3WHS
+        * had been finished.
+        * The bpf prog only counted two data packet out but we
+        * specially init accept_fd's pkt_out_cnt by 2 in
+        * init_sk_storage().  Hence, 4 here.
+        */
+       CHECK(err || pkt_out_cnt.cnt != 4 || pkt_out_cnt10.cnt != 40,
+             "bpf_map_lookup_elem(sk_pkt_out_cnt, &accept_fd)",
+             "err:%d errno:%d pkt_out_cnt:%u pkt_out_cnt10:%u",
+             err, errno, pkt_out_cnt.cnt, pkt_out_cnt10.cnt);
+
+       pkt_out_cnt.cnt = ~0;
+       pkt_out_cnt10.cnt = ~0;
+       err = bpf_map_lookup_elem(sk_pkt_out_cnt_fd, &cli_fd, &pkt_out_cnt);
+       if (!err)
+               err = bpf_map_lookup_elem(sk_pkt_out_cnt10_fd, &cli_fd,
+                                         &pkt_out_cnt10);
+       /* Active connection is fullsock from the beginning.
+        * 1 SYN and 1 ACK during 3WHS
+        * 2 Acks on data packet.
+        *
+        * The bpf_prog initialized it to 0xeB9F.
+        */
+       CHECK(err || pkt_out_cnt.cnt != 0xeB9F + 4 ||
+             pkt_out_cnt10.cnt != 0xeB9F + 40,
+             "bpf_map_lookup_elem(sk_pkt_out_cnt, &cli_fd)",
+             "err:%d errno:%d pkt_out_cnt:%u pkt_out_cnt10:%u",
+             err, errno, pkt_out_cnt.cnt, pkt_out_cnt10.cnt);
+}
+
+static void init_sk_storage(int sk_fd, __u32 pkt_out_cnt)
+{
+       struct bpf_spinlock_cnt scnt = {};
+       int err;
+
+       scnt.cnt = pkt_out_cnt;
+       err = bpf_map_update_elem(sk_pkt_out_cnt_fd, &sk_fd, &scnt,
+                                 BPF_NOEXIST);
+       CHECK(err, "bpf_map_update_elem(sk_pkt_out_cnt_fd)",
+             "err:%d errno:%d", err, errno);
+
+       scnt.cnt *= 10;
+       err = bpf_map_update_elem(sk_pkt_out_cnt10_fd, &sk_fd, &scnt,
+                                 BPF_NOEXIST);
+       CHECK(err, "bpf_map_update_elem(sk_pkt_out_cnt10_fd)",
+             "err:%d errno:%d", err, errno);
+}
+
 static void test(void)
 {
        int listen_fd, cli_fd, accept_fd, epfd, err;
        struct epoll_event ev;
        socklen_t addrlen;
+       int i;
 
        addrlen = sizeof(struct sockaddr_in6);
        ev.events = EPOLLIN;
@@ -308,24 +377,30 @@ static void test(void)
              accept_fd, errno);
        close(listen_fd);
 
-       /* Send some data from accept_fd to cli_fd */
-       err = send(accept_fd, DATA, DATA_LEN, 0);
-       CHECK(err != DATA_LEN, "send(accept_fd)", "err:%d errno:%d",
-             err, errno);
-
-       /* Have some timeout in recv(cli_fd). Just in case. */
        ev.data.fd = cli_fd;
        err = epoll_ctl(epfd, EPOLL_CTL_ADD, cli_fd, &ev);
        CHECK(err, "epoll_ctl(EPOLL_CTL_ADD, cli_fd)", "err:%d errno:%d",
              err, errno);
 
-       err = epoll_wait(epfd, &ev, 1, 1000);
-       CHECK(err != 1 || ev.data.fd != cli_fd,
-             "epoll_wait(cli_fd)", "err:%d errno:%d ev.data.fd:%d cli_fd:%d",
-             err, errno, ev.data.fd, cli_fd);
+       init_sk_storage(accept_fd, 2);
 
-       err = recv(cli_fd, NULL, 0, MSG_TRUNC);
-       CHECK(err, "recv(cli_fd)", "err:%d errno:%d", err, errno);
+       for (i = 0; i < 2; i++) {
+               /* Send some data from accept_fd to cli_fd */
+               err = send(accept_fd, DATA, DATA_LEN, 0);
+               CHECK(err != DATA_LEN, "send(accept_fd)", "err:%d errno:%d",
+                     err, errno);
+
+               /* Have some timeout in recv(cli_fd). Just in case. */
+               err = epoll_wait(epfd, &ev, 1, 1000);
+               CHECK(err != 1 || ev.data.fd != cli_fd,
+                     "epoll_wait(cli_fd)", "err:%d errno:%d ev.data.fd:%d cli_fd:%d",
+                     err, errno, ev.data.fd, cli_fd);
+
+               err = recv(cli_fd, NULL, 0, MSG_TRUNC);
+               CHECK(err, "recv(cli_fd)", "err:%d errno:%d", err, errno);
+       }
+
+       check_sk_pkt_out_cnt(accept_fd, cli_fd);
 
        close(epfd);
        close(accept_fd);
@@ -395,6 +470,14 @@ int main(int argc, char **argv)
        CHECK(!map, "cannot find linum_map", "(null)");
        linum_map_fd = bpf_map__fd(map);
 
+       map = bpf_object__find_map_by_name(obj, "sk_pkt_out_cnt");
+       CHECK(!map, "cannot find sk_pkt_out_cnt", "(null)");
+       sk_pkt_out_cnt_fd = bpf_map__fd(map);
+
+       map = bpf_object__find_map_by_name(obj, "sk_pkt_out_cnt10");
+       CHECK(!map, "cannot find sk_pkt_out_cnt10", "(null)");
+       sk_pkt_out_cnt10_fd = bpf_map__fd(map);
+
        test();
 
        bpf_object__close(obj);