[PATCH bpf-next v2] selftests/bpf: Fix spurious failures in accept due to EAGAIN

[PATCH bpf-next v2] selftests/bpf: Fix spurious failures in accept due to EAGAIN

[Jakub Sitnicki]

Andrii Nakryiko reports that sockmap_listen test suite is frequently
failing due to accept() calls erroring out with EAGAIN:

  ./test_progs:connect_accept_thread:733: accept: Resource temporarily unavailable
  connect_accept_thread:FAIL:733

This is because we are using a non-blocking listening TCP socket to
accept() connections without polling on the socket.

While at first switching to blocking mode seems like the right thing to do,
this could lead to test process blocking indefinitely in face of a network
issue, like loopback interface being down, as Andrii pointed out.

Hence, stick to non-blocking mode for TCP listening sockets but with
polling for incoming connection for a limited time before giving up.

Apply this approach to all socket I/O calls in the test suite that we
expect to block indefinitely, that is accept() for TCP and recv() for UDP.

Fixes: 44d28be2b8d4 ("selftests/bpf: Tests for sockmap/sockhash holding listening sockets")
Reported-by: Andrii Nakryiko <andrii.nakryiko@gmail.com>
Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
---

Notes:
    v2: Switch back to non-blocking mode, but with polling and timeout.
        Extend the fix to all I/O calls that we expect to block. (Andrii)

 .../selftests/bpf/prog_tests/sockmap_listen.c | 77 ++++++++++++++-----
 1 file changed, 58 insertions(+), 19 deletions(-)

diff --git a/tools/testing/selftests/bpf/prog_tests/sockmap_listen.c b/tools/testing/selftests/bpf/prog_tests/sockmap_listen.c
index 52aa468bdccd..d7d65a700799 100644
--- a/tools/testing/selftests/bpf/prog_tests/sockmap_listen.c
+++ b/tools/testing/selftests/bpf/prog_tests/sockmap_listen.c
@@ -16,6 +16,7 @@
 #include <pthread.h>
 #include <stdlib.h>
 #include <string.h>
+#include <sys/select.h>
 #include <unistd.h>
 
 #include <bpf/bpf.h>
@@ -25,6 +26,7 @@
 #include "test_progs.h"
 #include "test_sockmap_listen.skel.h"
 
+#define IO_TIMEOUT_SEC 30
 #define MAX_STRERR_LEN 256
 #define MAX_TEST_NAME 80
 
@@ -44,9 +46,10 @@
 
 /* Wrappers that fail the test on error and report it. */
 
-#define xaccept(fd, addr, len)                                                 \
+#define xaccept_nonblock(fd, addr, len)                                        \
 	({                                                                     \
-		int __ret = accept((fd), (addr), (len));                       \
+		int __ret =                                                    \
+			accept_timeout((fd), (addr), (len), IO_TIMEOUT_SEC);   \
 		if (__ret == -1)                                               \
 			FAIL_ERRNO("accept");                                  \
 		__ret;                                                         \
@@ -116,9 +119,10 @@
 		__ret;                                                         \
 	})
 
-#define xrecv(fd, buf, len, flags)                                             \
+#define xrecv_nonblock(fd, buf, len, flags)                                    \
 	({                                                                     \
-		ssize_t __ret = recv((fd), (buf), (len), (flags));             \
+		ssize_t __ret = recv_timeout((fd), (buf), (len), (flags),      \
+					     IO_TIMEOUT_SEC);                  \
 		if (__ret == -1)                                               \
 			FAIL_ERRNO("recv");                                    \
 		__ret;                                                         \
@@ -191,6 +195,40 @@
 		__ret;                                                         \
 	})
 
+static int poll_read(int fd, unsigned int timeout_sec)
+{
+	struct timeval timeout = { .tv_sec = timeout_sec };
+	fd_set rfds;
+	int r;
+
+	FD_ZERO(&rfds);
+	FD_SET(fd, &rfds);
+
+	r = select(fd + 1, &rfds, NULL, NULL, &timeout);
+	if (r == 0)
+		errno = ETIME;
+
+	return r == 1 ? 0 : -1;
+}
+
+static int accept_timeout(int fd, struct sockaddr *addr, socklen_t *len,
+			  unsigned int timeout_sec)
+{
+	if (poll_read(fd, timeout_sec))
+		return -1;
+
+	return accept(fd, addr, len);
+}
+
+static int recv_timeout(int fd, void *buf, size_t len, int flags,
+			unsigned int timeout_sec)
+{
+	if (poll_read(fd, timeout_sec))
+		return -1;
+
+	return recv(fd, buf, len, flags);
+}
+
 static void init_addr_loopback4(struct sockaddr_storage *ss, socklen_t *len)
 {
 	struct sockaddr_in *addr4 = memset(ss, 0, sizeof(*ss));
@@ -265,7 +303,7 @@ static int socket_loopback_reuseport(int family, int sotype, int progfd)
 	if (err)
 		goto close;
 
-	if (sotype == SOCK_DGRAM)
+	if (sotype & SOCK_DGRAM)
 		return s;
 
 	err = xlisten(s, SOMAXCONN);
@@ -589,7 +627,7 @@ static void test_accept_after_delete(int family, int sotype, int mapfd)
 	socklen_t len;
 	u64 value;
 
-	s = socket_loopback(family, sotype);
+	s = socket_loopback(family, sotype | SOCK_NONBLOCK);
 	if (s == -1)
 		return;
 
@@ -617,7 +655,7 @@ static void test_accept_after_delete(int family, int sotype, int mapfd)
 	if (err)
 		goto close_cli;
 
-	p = xaccept(s, NULL, NULL);
+	p = xaccept_nonblock(s, NULL, NULL);
 	if (p == -1)
 		goto close_cli;
 
@@ -643,7 +681,7 @@ static void test_accept_before_delete(int family, int sotype, int mapfd)
 	socklen_t len;
 	u64 value;
 
-	s = socket_loopback(family, sotype);
+	s = socket_loopback(family, sotype | SOCK_NONBLOCK);
 	if (s == -1)
 		return;
 
@@ -666,7 +704,7 @@ static void test_accept_before_delete(int family, int sotype, int mapfd)
 	if (err)
 		goto close_cli;
 
-	p = xaccept(s, NULL, NULL);
+	p = xaccept_nonblock(s, NULL, NULL);
 	if (p == -1)
 		goto close_cli;
 
@@ -730,7 +768,7 @@ static void *connect_accept_thread(void *arg)
 			break;
 		}
 
-		p = xaccept(s, NULL, NULL);
+		p = xaccept_nonblock(s, NULL, NULL);
 		if (p < 0) {
 			xclose(c);
 			break;
@@ -912,7 +950,7 @@ static void redir_to_connected(int family, int sotype, int sock_mapfd,
 	if (err)
 		goto close_cli0;
 
-	p0 = xaccept(s, NULL, NULL);
+	p0 = xaccept_nonblock(s, NULL, NULL);
 	if (p0 < 0)
 		goto close_cli0;
 
@@ -923,7 +961,7 @@ static void redir_to_connected(int family, int sotype, int sock_mapfd,
 	if (err)
 		goto close_cli1;
 
-	p1 = xaccept(s, NULL, NULL);
+	p1 = xaccept_nonblock(s, NULL, NULL);
 	if (p1 < 0)
 		goto close_cli1;
 
@@ -1044,7 +1082,7 @@ static void redir_to_listening(int family, int sotype, int sock_mapfd,
 	if (err)
 		goto close_cli;
 
-	p = xaccept(s, NULL, NULL);
+	p = xaccept_nonblock(s, NULL, NULL);
 	if (p < 0)
 		goto close_cli;
 
@@ -1139,7 +1177,8 @@ static void test_reuseport_select_listening(int family, int sotype,
 
 	zero_verdict_count(verd_map);
 
-	s = socket_loopback_reuseport(family, sotype, reuseport_prog);
+	s = socket_loopback_reuseport(family, sotype | SOCK_NONBLOCK,
+				      reuseport_prog);
 	if (s < 0)
 		return;
 
@@ -1164,7 +1203,7 @@ static void test_reuseport_select_listening(int family, int sotype,
 	if (sotype == SOCK_STREAM) {
 		int p;
 
-		p = xaccept(s, NULL, NULL);
+		p = xaccept_nonblock(s, NULL, NULL);
 		if (p < 0)
 			goto close_cli;
 		xclose(p);
@@ -1176,7 +1215,7 @@ static void test_reuseport_select_listening(int family, int sotype,
 		if (n == -1)
 			goto close_cli;
 
-		n = xrecv(s, &b, sizeof(b), 0);
+		n = xrecv_nonblock(s, &b, sizeof(b), 0);
 		if (n == -1)
 			goto close_cli;
 	}
@@ -1232,7 +1271,7 @@ static void test_reuseport_select_connected(int family, int sotype,
 		goto close_cli0;
 
 	if (sotype == SOCK_STREAM) {
-		p0 = xaccept(s, NULL, NULL);
+		p0 = xaccept_nonblock(s, NULL, NULL);
 		if (p0 < 0)
 			goto close_cli0;
 	} else {
@@ -1276,7 +1315,7 @@ static void test_reuseport_select_connected(int family, int sotype,
 		if (n == -1)
 			goto close_cli1;
 
-		n = recv(c1, &b, sizeof(b), 0);
+		n = recv_timeout(c1, &b, sizeof(b), 0, IO_TIMEOUT_SEC);
 		err = n == -1;
 	}
 	if (!err || errno != ECONNREFUSED)
@@ -1350,7 +1389,7 @@ static void test_reuseport_mixed_groups(int family, int sotype, int sock_map,
 		if (n == -1)
 			goto close_cli;
 
-		n = recv(c, &b, sizeof(b), 0);
+		n = recv_timeout(c, &b, sizeof(b), 0, IO_TIMEOUT_SEC);
 		err = n == -1;
 	}
 	if (!err || errno != ECONNREFUSED) {
-- 
2.24.1

Re: [PATCH bpf-next v2] selftests/bpf: Fix spurious failures in accept due to EAGAIN

[Andrii Nakryiko]

On Fri, Mar 13, 2020 at 9:10 AM Jakub Sitnicki <jakub@cloudflare.com> wrote:
>
> Andrii Nakryiko reports that sockmap_listen test suite is frequently
> failing due to accept() calls erroring out with EAGAIN:
>
>   ./test_progs:connect_accept_thread:733: accept: Resource temporarily unavailable
>   connect_accept_thread:FAIL:733
>
> This is because we are using a non-blocking listening TCP socket to
> accept() connections without polling on the socket.
>
> While at first switching to blocking mode seems like the right thing to do,
> this could lead to test process blocking indefinitely in face of a network
> issue, like loopback interface being down, as Andrii pointed out.
>
> Hence, stick to non-blocking mode for TCP listening sockets but with
> polling for incoming connection for a limited time before giving up.
>
> Apply this approach to all socket I/O calls in the test suite that we
> expect to block indefinitely, that is accept() for TCP and recv() for UDP.
>
> Fixes: 44d28be2b8d4 ("selftests/bpf: Tests for sockmap/sockhash holding listening sockets")
> Reported-by: Andrii Nakryiko <andrii.nakryiko@gmail.com>
> Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
> ---

This looks good. Unfortunately can't repro the issue locally anymore.
But once this gets into bpf-next and we update libbpf in Github, I'll
enable sockmap_listen tests again and see if it's still flaky. Thanks
for following up!

Stanislav, would you get a chance to do something similar for tcp_rtt
as well? Seems like all the tests dealing with sockets might use this
approach?

Acked-by: Andrii Nakryiko <andriin@fb.com>

>
> Notes:
>     v2: Switch back to non-blocking mode, but with polling and timeout.
>         Extend the fix to all I/O calls that we expect to block. (Andrii)
>
>  .../selftests/bpf/prog_tests/sockmap_listen.c | 77 ++++++++++++++-----
>  1 file changed, 58 insertions(+), 19 deletions(-)
>
> diff --git a/tools/testing/selftests/bpf/prog_tests/sockmap_listen.c b/tools/testing/selftests/bpf/prog_tests/sockmap_listen.c
> index 52aa468bdccd..d7d65a700799 100644
> --- a/tools/testing/selftests/bpf/prog_tests/sockmap_listen.c
> +++ b/tools/testing/selftests/bpf/prog_tests/sockmap_listen.c
> @@ -16,6 +16,7 @@

[...]

Re: [PATCH bpf-next v2] selftests/bpf: Fix spurious failures in accept due to EAGAIN

[Daniel Borkmann]

On 3/13/20 8:01 PM, Andrii Nakryiko wrote:
> On Fri, Mar 13, 2020 at 9:10 AM Jakub Sitnicki <jakub@cloudflare.com> wrote:
>> Andrii Nakryiko reports that sockmap_listen test suite is frequently
>> failing due to accept() calls erroring out with EAGAIN:
>>
>>    ./test_progs:connect_accept_thread:733: accept: Resource temporarily unavailable
>>    connect_accept_thread:FAIL:733
>>
>> This is because we are using a non-blocking listening TCP socket to
>> accept() connections without polling on the socket.
>>
>> While at first switching to blocking mode seems like the right thing to do,
>> this could lead to test process blocking indefinitely in face of a network
>> issue, like loopback interface being down, as Andrii pointed out.
>>
>> Hence, stick to non-blocking mode for TCP listening sockets but with
>> polling for incoming connection for a limited time before giving up.
>>
>> Apply this approach to all socket I/O calls in the test suite that we
>> expect to block indefinitely, that is accept() for TCP and recv() for UDP.
>>
>> Fixes: 44d28be2b8d4 ("selftests/bpf: Tests for sockmap/sockhash holding listening sockets")
>> Reported-by: Andrii Nakryiko <andrii.nakryiko@gmail.com>
>> Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
> 
> This looks good. Unfortunately can't repro the issue locally anymore.
> But once this gets into bpf-next and we update libbpf in Github, I'll
> enable sockmap_listen tests again and see if it's still flaky. Thanks
> for following up!

Ok, applied, thanks!

Re: [PATCH bpf-next v2] selftests/bpf: Fix spurious failures in accept due to EAGAIN

[Stanislav Fomichev]

On 03/13, Andrii Nakryiko wrote:
> On Fri, Mar 13, 2020 at 9:10 AM Jakub Sitnicki <jakub@cloudflare.com> wrote:
> Stanislav, would you get a chance to do something similar for tcp_rtt
> as well? Seems like all the tests dealing with sockets might use this
> approach?
Ack, will take a look ty!