syscalls/sendmmsg: add new test

Test basic functionality of sendmmsg and recvmmsg.

---

Known issues/questions:
- This test occasionally gets stuck when the two messages get sent by separate sendmmsg calls. The two sendmmsg calls complete successfully but recvmmsg blocks indefinitely. I am not yet sure what would cause this.
- I have not used tst_get_unused_port() because I'm unsure of how that function should be made available for both the new and old test API? I can just create a duplicate function but is there a cleaner way?

Changes since v1:
- rebased on recent autotools cleanup for libc func checks
- used test_variant to test direct syscalls and libc syscalls
- use tst_syscall instead of ltp_syscall
- check that both messages have been sent, retry sending messages if necessary
- use tst_brk(TBROK... if sendmmsg fails so recv thread is not kept waiting

 configure.ac                                  |   3 +
 include/lapi/socket.h                         |   8 +
 runtest/syscalls                              |   2 +
 testcases/kernel/syscalls/sendmmsg/.gitignore |   1 +
 testcases/kernel/syscalls/sendmmsg/Makefile   |   9 ++
 .../kernel/syscalls/sendmmsg/sendmmsg01.c     | 147 ++++++++++++++++++
 .../kernel/syscalls/sendmmsg/sendmmsg_var.h   |  60 +++++++
 7 files changed, 230 insertions(+)
 create mode 100644 testcases/kernel/syscalls/sendmmsg/.gitignore
 create mode 100644 testcases/kernel/syscalls/sendmmsg/Makefile
 create mode 100644 testcases/kernel/syscalls/sendmmsg/sendmmsg01.c
 create mode 100644 testcases/kernel/syscalls/sendmmsg/sendmmsg_var.h

index fad8f8396..182a0476f 100644
@@ -75,8 +75,10 @@ AC_CHECK_FUNCS([ \
     pwritev \
     pwritev2 \
     readlinkat \
+    recvmmsg \
     renameat \
     renameat2 \
+    sendmmsg \
     sigpending \
     splice \
     stime \

@@ -243,6 +245,7 @@ LTP_CHECK_TIME
 LTP_CHECK_TIMERFD
 test "x$with_tirpc" = xyes && LTP_CHECK_TIRPC
 LTP_CHECK_TPACKET_V3
+LTP_CHECK_MMSGHDR
 LTP_CHECK_UNAME_DOMAINNAME
 LTP_CHECK_XFS_QUOTACTL
 LTP_CHECK_X_TABLES

index 2605443e8..6d9e9fe30 100644
@@ -19,6 +19,7 @@
 #ifndef __LAPI_SOCKET_H__
 #define __LAPI_SOCKET_H__
 
+#include "config.h"
 #include <sys/socket.h>
 
 #ifndef MSG_ZEROCOPY

@@ -69,4 +70,11 @@
 # define SOL_ALG		279
 #endif
 
+#ifndef HAVE_STRUCT_MMSGHDR
+struct mmsghdr {
+	struct msghdr msg_hdr;
+	unsigned int msg_len;
+};
+#endif
+
 #endif /* __LAPI_SOCKET_H__ */

index 2b8ca719b..0e5d5ad3f 100644
@@ -1109,6 +1109,8 @@ sendfile09_64 sendfile09_64
 sendmsg01 sendmsg01
 sendmsg02 sendmsg02
 
+sendmmsg01 sendmmsg01
+
 sendto01 sendto01
 sendto02 sendto02
 

new file mode 100644
index 000000000..b703ececd
@@ -0,0 +1 @@
+sendmmsg01

new file mode 100644
index 000000000..47e063722
@@ -0,0 +1,9 @@
+# SPDX-License-Identifier: GPL-2.0-or-later
+
+top_srcdir		?= ../../../..
+
+sendmmsg01: CFLAGS+=-pthread
+
+include $(top_srcdir)/include/mk/testcases.mk
+
+include $(top_srcdir)/include/mk/generic_leaf_target.mk

2.21.0.593.g511ec345e18-goog

Hi Steve,

As I need tst_get_unused_port() as well, I've sent a patch to port it to new API and available for both new and legacy API via TST_GET_UNUSED_PORT() macro [1].

Kind regards,
Petr

[1] https://patchwork.ozlabs.org/patch/1089695/

Hi!

There is a race between the sender and receiver thread, if the sender thread finishes before the receiver thread binds the socket the packets are simply dropped since this is SOCK_DGRAM and nobody is listening at the other side of the socket yet.

You have to synchronize the threads with checkpoints so that the sender thread does not send anything until the receiver binds the socket.

Peter just merged patch that adds this functionality to new library.

You are also missing the SPDX licence identifier and other than this there are minor coding style violation (hint use checkpatch.pl).

Apart from these the code looks good.

Cyril Hrubis
[email protected]

Hi!
Ping. Are you going to work on this test or should I take over?

Cyril Hrubis
[email protected]

Hi sorry for the delay, yes I'm just getting back to this now.