]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/commitdiff
selftests/bpf: add a test case to check verifier pointer arithmetic
authorYonghong Song <yhs@fb.com>
Wed, 3 May 2017 02:58:14 +0000 (19:58 -0700)
committerDavid S. Miller <davem@davemloft.net>
Wed, 3 May 2017 13:51:25 +0000 (09:51 -0400)
With clang/llvm 4.0+, the test case is able to generate
the following pattern:
....
440: (b7) r1 = 15
441: (05) goto pc+73
515: (79) r6 = *(u64 *)(r10 -152)
516: (bf) r7 = r10
517: (07) r7 += -112
518: (bf) r2 = r7
519: (0f) r2 += r1
520: (71) r1 = *(u8 *)(r8 +0)
521: (73) *(u8 *)(r2 +45) = r1
....

commit 332270fdc8b6 ("bpf: enhance verifier to understand stack
pointer arithmetic") improved verifier to handle such a pattern.
This patch adds a C test case to actually generate such a pattern.
A dummy tracepoint interface is used to load the program
into the kernel.

Signed-off-by: Yonghong Song <yhs@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
tools/testing/selftests/bpf/Makefile
tools/testing/selftests/bpf/test_progs.c
tools/testing/selftests/bpf/test_tcp_estats.c [new file with mode: 0644]

index d8d94b9bd76c7c4fb62c8ebf50e0033de97012da..f4341e160de6e4580deadc129f4c84d2172fe4d3 100644 (file)
@@ -13,7 +13,7 @@ LDLIBS += -lcap -lelf
 
 TEST_GEN_PROGS = test_verifier test_tag test_maps test_lru_map test_lpm_map test_progs
 
-TEST_GEN_FILES = test_pkt_access.o test_xdp.o test_l4lb.o
+TEST_GEN_FILES = test_pkt_access.o test_xdp.o test_l4lb.o test_tcp_estats.o
 
 TEST_PROGS := test_kmod.sh
 
index 4ed049a0b14bcc3054b8da0132adbc06e820f590..b59f5ed4ae40d36bf9755fa844e63597c2a6f2c1 100644 (file)
@@ -268,6 +268,21 @@ out:
        bpf_object__close(obj);
 }
 
+static void test_tcp_estats(void)
+{
+       const char *file = "./test_tcp_estats.o";
+       int err, prog_fd;
+       struct bpf_object *obj;
+       __u32 duration = 0;
+
+       err = bpf_prog_load(file, BPF_PROG_TYPE_TRACEPOINT, &obj, &prog_fd);
+       CHECK(err, "", "err %d errno %d\n", err, errno);
+       if (err)
+               return;
+
+       bpf_object__close(obj);
+}
+
 int main(void)
 {
        struct rlimit rinf = { RLIM_INFINITY, RLIM_INFINITY };
@@ -277,6 +292,7 @@ int main(void)
        test_pkt_access();
        test_xdp();
        test_l4lb();
+       test_tcp_estats();
 
        printf("Summary: %d PASSED, %d FAILED\n", pass_cnt, error_cnt);
        return 0;
diff --git a/tools/testing/selftests/bpf/test_tcp_estats.c b/tools/testing/selftests/bpf/test_tcp_estats.c
new file mode 100644 (file)
index 0000000..bee3bbe
--- /dev/null
@@ -0,0 +1,258 @@
+/* Copyright (c) 2017 Facebook
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of version 2 of the GNU General Public
+ * License as published by the Free Software Foundation.
+ */
+
+/* This program shows clang/llvm is able to generate code pattern
+ * like:
+ *   _tcp_send_active_reset:
+ *      0:       bf 16 00 00 00 00 00 00         r6 = r1
+ *    ......
+ *    335:       b7 01 00 00 0f 00 00 00         r1 = 15
+ *    336:       05 00 48 00 00 00 00 00         goto 72
+ *
+ *   LBB0_3:
+ *    337:       b7 01 00 00 01 00 00 00         r1 = 1
+ *    338:       63 1a d0 ff 00 00 00 00         *(u32 *)(r10 - 48) = r1
+ *    408:       b7 01 00 00 03 00 00 00         r1 = 3
+ *
+ *   LBB0_4:
+ *    409:       71 a2 fe ff 00 00 00 00         r2 = *(u8 *)(r10 - 2)
+ *    410:       bf a7 00 00 00 00 00 00         r7 = r10
+ *    411:       07 07 00 00 b8 ff ff ff         r7 += -72
+ *    412:       bf 73 00 00 00 00 00 00         r3 = r7
+ *    413:       0f 13 00 00 00 00 00 00         r3 += r1
+ *    414:       73 23 2d 00 00 00 00 00         *(u8 *)(r3 + 45) = r2
+ *
+ * From the above code snippet, the code generated by the compiler
+ * is reasonable. The "r1" is assigned to different values in basic
+ * blocks "_tcp_send_active_reset" and "LBB0_3", and used in "LBB0_4".
+ * The verifier should be able to handle such code patterns.
+ */
+#include <string.h>
+#include <linux/bpf.h>
+#include <linux/ipv6.h>
+#include <linux/version.h>
+#include <sys/socket.h>
+#include "bpf_helpers.h"
+
+#define _(P) ({typeof(P) val = 0; bpf_probe_read(&val, sizeof(val), &P); val;})
+#define TCP_ESTATS_MAGIC 0xBAADBEEF
+
+/* This test case needs "sock" and "pt_regs" data structure.
+ * Recursively, "sock" needs "sock_common" and "inet_sock".
+ * However, this is a unit test case only for
+ * verifier purpose without bpf program execution.
+ * We can safely mock much simpler data structures, basically
+ * only taking the necessary fields from kernel headers.
+ */
+typedef __u32 __bitwise __portpair;
+typedef __u64 __bitwise __addrpair;
+
+struct sock_common {
+       unsigned short          skc_family;
+       union {
+               __addrpair      skc_addrpair;
+               struct {
+                       __be32  skc_daddr;
+                       __be32  skc_rcv_saddr;
+               };
+       };
+       union {
+               __portpair      skc_portpair;
+               struct {
+                       __be16  skc_dport;
+                       __u16   skc_num;
+               };
+       };
+       struct in6_addr         skc_v6_daddr;
+       struct in6_addr         skc_v6_rcv_saddr;
+};
+
+struct sock {
+       struct sock_common      __sk_common;
+#define sk_family              __sk_common.skc_family
+#define sk_v6_daddr            __sk_common.skc_v6_daddr
+#define sk_v6_rcv_saddr                __sk_common.skc_v6_rcv_saddr
+};
+
+struct inet_sock {
+       struct sock             sk;
+#define inet_daddr             sk.__sk_common.skc_daddr
+#define inet_dport             sk.__sk_common.skc_dport
+       __be32                  inet_saddr;
+       __be16                  inet_sport;
+};
+
+struct pt_regs {
+       long di;
+};
+
+static inline struct inet_sock *inet_sk(const struct sock *sk)
+{
+       return (struct inet_sock *)sk;
+}
+
+/* Define various data structures for state recording.
+ * Some fields are not used due to test simplification.
+ */
+enum tcp_estats_addrtype {
+       TCP_ESTATS_ADDRTYPE_IPV4 = 1,
+       TCP_ESTATS_ADDRTYPE_IPV6 = 2
+};
+
+enum tcp_estats_event_type {
+       TCP_ESTATS_ESTABLISH,
+       TCP_ESTATS_PERIODIC,
+       TCP_ESTATS_TIMEOUT,
+       TCP_ESTATS_RETRANSMIT_TIMEOUT,
+       TCP_ESTATS_RETRANSMIT_OTHER,
+       TCP_ESTATS_SYN_RETRANSMIT,
+       TCP_ESTATS_SYNACK_RETRANSMIT,
+       TCP_ESTATS_TERM,
+       TCP_ESTATS_TX_RESET,
+       TCP_ESTATS_RX_RESET,
+       TCP_ESTATS_WRITE_TIMEOUT,
+       TCP_ESTATS_CONN_TIMEOUT,
+       TCP_ESTATS_ACK_LATENCY,
+       TCP_ESTATS_NEVENTS,
+};
+
+struct tcp_estats_event {
+       int pid;
+       int cpu;
+       unsigned long ts;
+       unsigned int magic;
+       enum tcp_estats_event_type event_type;
+};
+
+/* The below data structure is packed in order for
+ * llvm compiler to generate expected code.
+ */
+struct tcp_estats_conn_id {
+       unsigned int localaddressType;
+       struct {
+               unsigned char data[16];
+       } localaddress;
+       struct {
+               unsigned char data[16];
+       } remaddress;
+       unsigned short    localport;
+       unsigned short    remport;
+} __attribute__((__packed__));
+
+struct tcp_estats_basic_event {
+       struct tcp_estats_event event;
+       struct tcp_estats_conn_id conn_id;
+};
+
+struct bpf_map_def SEC("maps") ev_record_map = {
+       .type = BPF_MAP_TYPE_HASH,
+       .key_size = sizeof(__u32),
+       .value_size = sizeof(struct tcp_estats_basic_event),
+       .max_entries = 1024,
+};
+
+struct dummy_tracepoint_args {
+       unsigned long long pad;
+       struct sock *sock;
+};
+
+static __always_inline void tcp_estats_ev_init(struct tcp_estats_event *event,
+                                              enum tcp_estats_event_type type)
+{
+       event->magic = TCP_ESTATS_MAGIC;
+       event->ts = bpf_ktime_get_ns();
+       event->event_type = type;
+}
+
+static __always_inline void unaligned_u32_set(unsigned char *to, __u8 *from)
+{
+       to[0] = _(from[0]);
+       to[1] = _(from[1]);
+       to[2] = _(from[2]);
+       to[3] = _(from[3]);
+}
+
+static __always_inline void conn_id_ipv4_init(struct tcp_estats_conn_id *conn_id,
+                                             __be32 *saddr, __be32 *daddr)
+{
+       conn_id->localaddressType = TCP_ESTATS_ADDRTYPE_IPV4;
+
+       unaligned_u32_set(conn_id->localaddress.data, (__u8 *)saddr);
+       unaligned_u32_set(conn_id->remaddress.data, (__u8 *)daddr);
+}
+
+static __always_inline void conn_id_ipv6_init(struct tcp_estats_conn_id *conn_id,
+                                             __be32 *saddr, __be32 *daddr)
+{
+       conn_id->localaddressType = TCP_ESTATS_ADDRTYPE_IPV6;
+
+       unaligned_u32_set(conn_id->localaddress.data, (__u8 *)saddr);
+       unaligned_u32_set(conn_id->localaddress.data + sizeof(__u32),
+                         (__u8 *)(saddr + 1));
+       unaligned_u32_set(conn_id->localaddress.data + sizeof(__u32) * 2,
+                         (__u8 *)(saddr + 2));
+       unaligned_u32_set(conn_id->localaddress.data + sizeof(__u32) * 3,
+                         (__u8 *)(saddr + 3));
+
+       unaligned_u32_set(conn_id->remaddress.data,
+                         (__u8 *)(daddr));
+       unaligned_u32_set(conn_id->remaddress.data + sizeof(__u32),
+                         (__u8 *)(daddr + 1));
+       unaligned_u32_set(conn_id->remaddress.data + sizeof(__u32) * 2,
+                         (__u8 *)(daddr + 2));
+       unaligned_u32_set(conn_id->remaddress.data + sizeof(__u32) * 3,
+                         (__u8 *)(daddr + 3));
+}
+
+static __always_inline void tcp_estats_conn_id_init(struct tcp_estats_conn_id *conn_id,
+                                                   struct sock *sk)
+{
+       conn_id->localport = _(inet_sk(sk)->inet_sport);
+       conn_id->remport = _(inet_sk(sk)->inet_dport);
+
+       if (_(sk->sk_family) == AF_INET6)
+               conn_id_ipv6_init(conn_id,
+                                 sk->sk_v6_rcv_saddr.s6_addr32,
+                                 sk->sk_v6_daddr.s6_addr32);
+       else
+               conn_id_ipv4_init(conn_id,
+                                 &inet_sk(sk)->inet_saddr,
+                                 &inet_sk(sk)->inet_daddr);
+}
+
+static __always_inline void tcp_estats_init(struct sock *sk,
+                                           struct tcp_estats_event *event,
+                                           struct tcp_estats_conn_id *conn_id,
+                                           enum tcp_estats_event_type type)
+{
+       tcp_estats_ev_init(event, type);
+       tcp_estats_conn_id_init(conn_id, sk);
+}
+
+static __always_inline void send_basic_event(struct sock *sk,
+                                            enum tcp_estats_event_type type)
+{
+       struct tcp_estats_basic_event ev;
+       __u32 key = bpf_get_prandom_u32();
+
+       memset(&ev, 0, sizeof(ev));
+       tcp_estats_init(sk, &ev.event, &ev.conn_id, type);
+       bpf_map_update_elem(&ev_record_map, &key, &ev, BPF_ANY);
+}
+
+SEC("dummy_tracepoint")
+int _dummy_tracepoint(struct dummy_tracepoint_args *arg)
+{
+       if (!arg->sock)
+               return 0;
+
+       send_basic_event(arg->sock, TCP_ESTATS_TX_RESET);
+       return 0;
+}
+
+char _license[] SEC("license") = "GPL";
+__u32 _version SEC("version") = 1; /* ignored by tracepoints, required by libbpf.a */