trafgen: take advantage of PACKET_QDISC_BYPASS sock option

Since Linux 3.14, the kernel supports a socket option PACKET_QDISC_BYPASS,
which trafgen enables by default. That allow us to bypass the kernels
normal qdisc (traffic control) layer.

An option -q, --qdisc-path is added to allow enabling the qdisc path
explicity, useful for testing purposes.

This will be avail in kernels >= 3.14 via
 commit d346a3fae3 (packet: introduce PACKET_QDISC_BYPASS socket option).

Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>

5 files changed, 40 insertions, 2 deletions

diff --git a/built_in.h b/built_in.h
index 6f6625e..20d7317 100644
--- a/built_in.h
+++ b/built_in.h
@@ -370,6 +370,10 @@ static inline u64 cpu_to_le64(u64 val)
 # define TP_STATUS_TS_RAW_HARDWARE	(1 << 31)
 #endif
 
+#ifndef PACKET_QDISC_BYPASS
+# define PACKET_QDISC_BYPASS 20
+#endif
+
 #ifndef POLLRDNORM
 # define POLLRDNORM			0x0040
 #endif
diff --git a/sock.c b/sock.c
index a8b4f82..e5f6eac 100644
--- a/sock.c
+++ b/sock.c
@@ -42,6 +42,24 @@ int pf_tx_socket(void)
 	return sock;
 }
 
+/* Avail in kernel >= 3.14
+ * in commit d346a3fae3 (packet: introduce PACKET_QDISC_BYPASS socket option)
+ */
+void set_sock_qdisc_bypass(int fd, int verbose)
+{
+	int ret, val = 1;
+
+	ret = setsockopt(fd, SOL_PACKET, PACKET_QDISC_BYPASS, &val, sizeof(val));
+	if (ret < 0) {
+		if (errno == ENOPROTOOPT) {
+			if (verbose)
+				printf("No kernel support for PACKET_QDISC_BYPASS"
+				       " (kernel < 3.14?)\n");
+		} else
+			perror("Cannot set PACKET_QDISC_BYPASS");
+	} else
+		if (verbose) printf("Enabled kernel qdisc bypass\n");
+}
 
 void set_sock_prio(int fd, int prio)
 {
diff --git a/sock.h b/sock.h
index 75be501..0e680ef 100644
--- a/sock.h
+++ b/sock.h
@@ -7,6 +7,7 @@ extern int pf_tx_socket(void);
 extern void set_nonblocking(int fd);
 extern int set_nonblocking_sloppy(int fd);
 extern int set_reuseaddr(int fd);
+extern void set_sock_qdisc_bypass(int fd, int verbose);
 extern void set_sock_prio(int fd, int prio);
 extern void set_tcp_nodelay(int fd);
 extern void set_socket_keepalive(int fd);
diff --git a/trafgen.8 b/trafgen.8
index 6a49fd0..72c2575 100644
--- a/trafgen.8
+++ b/trafgen.8
@@ -161,6 +161,12 @@ threads.
 .SS -Q, --notouch-irq
 Do not reassign the NIC's IRQ CPU affinity settings.
 .PP
+.SS -q, --qdisc-path
+Since Linux 3.14, the kernel supports a socket option PACKET_QDISC_BYPASS,
+which trafgen enables by default.  This options disables the qdisc bypass,
+and uses the normal send path through the kernel's qdisc (traffic control)
+layer, which can be usefully for testing the qdisc path.
+.PP
 .SS -V, --verbose
 Let trafgen be more talkative and let it print the parsed configuration and
 some ring buffer statistics.
diff --git a/trafgen.c b/trafgen.c
index e69fcb5..6f19677 100644
--- a/trafgen.c
+++ b/trafgen.c
@@ -55,7 +55,7 @@
 #include "csum.h"
 
 struct ctx {
-	bool rand, rfraw, jumbo_support, verbose, smoke_test, enforce;
+	bool rand, rfraw, jumbo_support, verbose, smoke_test, enforce, qdisc_path;
 	unsigned long kpull, num, reserve_size;
 	unsigned int cpus;
 	uid_t uid; gid_t gid;
@@ -80,7 +80,7 @@ size_t plen = 0;
 struct packet_dyn *packet_dyn = NULL;
 size_t dlen = 0;
 
-static const char *short_options = "d:c:n:t:vJhS:rk:i:o:VRs:P:eE:pu:g:CHQ";
+static const char *short_options = "d:c:n:t:vJhS:rk:i:o:VRs:P:eE:pu:g:CHQq";
 static const struct option long_options[] = {
 	{"dev",			required_argument,	NULL, 'd'},
 	{"out",			required_argument,	NULL, 'o'},
@@ -97,6 +97,7 @@ static const struct option long_options[] = {
 	{"group",		required_argument,	NULL, 'g'},
 	{"prio-high",		no_argument,		NULL, 'H'},
 	{"notouch-irq",		no_argument,		NULL, 'Q'},
+	{"qdisc-path",		no_argument,		NULL, 'q'},
 	{"jumbo-support",	no_argument,		NULL, 'J'},
 	{"no-cpu-stats",	no_argument,		NULL, 'C'},
 	{"cpp",			no_argument,		NULL, 'p'},
@@ -193,6 +194,7 @@ static void __noreturn help(void)
 	     "  -g|--group <groupid>           Drop privileges and change to groupid\n"
 	     "  -H|--prio-high                 Make this high priority process\n"
 	     "  -Q|--notouch-irq               Do not touch IRQ CPU affinity of NIC\n"
+	     "  -q|--qdisc-path                Enabled qdisc kernel path (default off since 3.14)\n"
 	     "  -V|--verbose                   Be more verbose\n"
 	     "  -C|--no-cpu-stats              Do not print CPU time statistics on exit\n"
 	     "  -v|--version                   Show version and exit\n"
@@ -856,6 +858,9 @@ static void main_loop(struct ctx *ctx, char *confname, bool slow,
 
 	sock = pf_tx_socket();
 
+	if (ctx->qdisc_path == false)
+		set_sock_qdisc_bypass(sock, ctx->verbose);
+
 	if (slow)
 		xmit_slowpath_or_die(ctx, cpu, orig_num);
 	else
@@ -897,6 +902,7 @@ int main(int argc, char **argv)
 	ctx.cpus = get_number_cpus_online();
 	ctx.uid = getuid();
 	ctx.gid = getgid();
+	ctx.qdisc_path = false;
 
 	while ((c = getopt_long(argc, argv, short_options, long_options,
 				&opt_index)) != EOF) {
@@ -934,6 +940,9 @@ int main(int argc, char **argv)
 		case 'Q':
 			set_irq_aff = false;
 			break;
+		case 'q':
+			ctx.qdisc_path = true;
+			break;
 		case 'r':
 			ctx.rand = true;
 			break;