Merge pull request #1925 from gegles/gsro

Add UDP GSO/GRO support (Linux) and --gsro switch

Author: bmah888

configure

@@ -17399,6 +17399,78 @@ printf "%s\n" "#define HAVE_IPPROTO_MPTCP 1" >>confdefs.h
 
 fi
 
+# Check for UDP_SEGMENT sockopt
+{ printf "%s\n" "$as_me:${as_lineno-$LINENO}: checking UDP_SEGMENT socket option" >&5
+printf %s "checking UDP_SEGMENT socket option... " >&6; }
+if test ${iperf3_cv_header_udp_segment+y}
+then :
+  printf %s "(cached) " >&6
+else case e in #(
+  e) cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+#include <linux/udp.h>
+int
+main (void)
+{
+int foo = UDP_SEGMENT;
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_c_try_compile "$LINENO"
+then :
+  iperf3_cv_header_udp_segment=yes
+else case e in #(
+  e) iperf3_cv_header_udp_segment=no ;;
+esac
+fi
+rm -f core conftest.err conftest.$ac_objext conftest.beam conftest.$ac_ext ;;
+esac
+fi
+{ printf "%s\n" "$as_me:${as_lineno-$LINENO}: result: $iperf3_cv_header_udp_segment" >&5
+printf "%s\n" "$iperf3_cv_header_udp_segment" >&6; }
+if test "x$iperf3_cv_header_udp_segment" = "xyes"; then
+
+printf "%s\n" "#define HAVE_UDP_SEGMENT 1" >>confdefs.h
+
+fi
+
+# Check for UDP_GRO sockopt
+{ printf "%s\n" "$as_me:${as_lineno-$LINENO}: checking UDP_GRO socket option" >&5
+printf %s "checking UDP_GRO socket option... " >&6; }
+if test ${iperf3_cv_header_udp_gro+y}
+then :
+  printf %s "(cached) " >&6
+else case e in #(
+  e) cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+#include <linux/udp.h>
+int
+main (void)
+{
+int foo = UDP_GRO;
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_c_try_compile "$LINENO"
+then :
+  iperf3_cv_header_udp_gro=yes
+else case e in #(
+  e) iperf3_cv_header_udp_gro=no ;;
+esac
+fi
+rm -f core conftest.err conftest.$ac_objext conftest.beam conftest.$ac_ext ;;
+esac
+fi
+{ printf "%s\n" "$as_me:${as_lineno-$LINENO}: result: $iperf3_cv_header_udp_gro" >&5
+printf "%s\n" "$iperf3_cv_header_udp_gro" >&6; }
+if test "x$iperf3_cv_header_udp_gro" = "xyes"; then
+
+printf "%s\n" "#define HAVE_UDP_GRO 1" >>confdefs.h
+
+fi
+
 # Check if we need -lrt for clock_gettime
 { printf "%s\n" "$as_me:${as_lineno-$LINENO}: checking for library containing clock_gettime" >&5
 printf %s "checking for library containing clock_gettime... " >&6; }

configure.ac

@@ -375,6 +375,30 @@ if test "x$iperf3_cv_header_ipproto_mptcp" = "xyes"; then
     AC_DEFINE([HAVE_IPPROTO_MPTCP], [1], [Have MPTCP protocol.])
 fi
 
+# Check for UDP_SEGMENT sockopt
+AC_CACHE_CHECK([UDP_SEGMENT socket option],
+[iperf3_cv_header_udp_segment],
+AC_COMPILE_IFELSE(
+  [AC_LANG_PROGRAM([[#include <linux/udp.h>]],
+                   [[int foo = UDP_SEGMENT;]])],
+  iperf3_cv_header_udp_segment=yes,
+  iperf3_cv_header_udp_segment=no))
+if test "x$iperf3_cv_header_udp_segment" = "xyes"; then
+    AC_DEFINE([HAVE_UDP_SEGMENT], [1], [Have UDP_SEGMENT sockopt.])
+fi
+
+# Check for UDP_GRO sockopt
+AC_CACHE_CHECK([UDP_GRO socket option],
+[iperf3_cv_header_udp_gro],
+AC_COMPILE_IFELSE(
+  [AC_LANG_PROGRAM([[#include <linux/udp.h>]],
+                   [[int foo = UDP_GRO;]])],
+  iperf3_cv_header_udp_gro=yes,
+  iperf3_cv_header_udp_gro=no))
+if test "x$iperf3_cv_header_udp_gro" = "xyes"; then
+    AC_DEFINE([HAVE_UDP_GRO], [1], [Have UDP_GRO sockopt.])
+fi
+
 # Check if we need -lrt for clock_gettime
 AC_SEARCH_LIBS(clock_gettime, [rt posix4])
 # Check for clock_gettime support

src/iperf.h

@@ -191,6 +191,12 @@ struct iperf_settings
     int       cntl_ka_keepidle;     /* Control TCP connection Keepalive idle time (TCP_KEEPIDLE) */
     int       cntl_ka_interval;     /* Control TCP connection Keepalive interval between retries (TCP_KEEPINTV) */
     int       cntl_ka_count;        /* Control TCP connection Keepalive number of retries (TCP_KEEPCNT) */
+    /* GSO/GRO fields always present to allow client-server negotiation regardless of local support */
+    int       gso;
+    int       gso_dg_size;
+    int       gso_bf_size;
+    int       gro;
+    int       gro_bf_size;
 };
 
 struct iperf_test;
@@ -486,4 +492,7 @@ extern int gerror; /* error value from getaddrinfo(3), for use in internal error
 /* In Reverse mode, maximum number of packets to wait for "accept" response - to handle out of order packets */
 #define MAX_REVERSE_OUT_OF_ORDER_PACKETS 2
 
+#define GSO_BF_MAX_SIZE MAX_UDP_BLOCKSIZE
+#define GRO_BF_MAX_SIZE MAX_UDP_BLOCKSIZE
+
 #endif /* !__IPERF_H */

src/iperf3.1

@@ -514,6 +514,17 @@ or high-bitrate UDP tests.  Both client and server need to be running
 at least version 3.1 for this option to work.  It may become the
 default behavior at some point in the future.
 .TP
+.BR --gsro
+Enable UDP Generic Segmentation Offload (GSO) on the sender and
+Generic Receive Offload (GRO) on the receiver, where supported by
+the operating system and network hardware (currently Linux only).
+GSO allows the network stack to aggregate multiple UDP datagrams
+into larger packets, improving throughput and reducing CPU overhead.
+This is a client-only option; the client communicates the GSO/GRO
+settings to the server automatically.
+This feature is disabled by default and must be explicitly enabled
+with this flag.
+.TP
 .BR --repeating-payload
 Use repeating pattern in payload, instead of random bytes.
 The same payload is used in iperf2 (ASCII '0..9' repeating).

src/iperf_api.c

@@ -1189,6 +1189,7 @@ iperf_parse_arguments(struct iperf_test *test, int argc, char **argv)
 #if defined(HAVE_IPPROTO_MPTCP)
         {"mptcp", no_argument, NULL, 'm'},
 #endif
+        {"gsro", no_argument, NULL, OPT_GSRO},
         {"debug", optional_argument, NULL, 'd'},
         {"help", no_argument, NULL, 'h'},
         {NULL, 0, NULL, 0}
@@ -1212,6 +1213,7 @@ iperf_parse_arguments(struct iperf_test *test, int argc, char **argv)
 
     blksize = 0;
     server_flag = client_flag = rate_flag = duration_flag = rcv_timeout_flag = snd_timeout_flag =0;
+    int gsro_flag = 0;
 #if defined(HAVE_SSL)
     char *client_username = NULL, *client_rsa_public_key = NULL, *server_rsa_private_key = NULL;
     FILE *ptr_file;
@@ -1790,6 +1792,13 @@ iperf_parse_arguments(struct iperf_test *test, int argc, char **argv)
 		test->mptcp = 1;
 		break;
 #endif
+            case OPT_GSRO:
+		/* Enable GSO/GRO which is disabled by default */
+		/* Flag is available regardless of local support to allow client to request server to use it */
+		gsro_flag = 1;
+		test->settings->gso = 1;
+		test->settings->gro = 1;
+                break;
 	    case 'h':
 		usage_long(stdout);
 		exit(0);
@@ -1809,6 +1818,21 @@ iperf_parse_arguments(struct iperf_test *test, int argc, char **argv)
         i_errno = IECLIENTONLY;
         return -1;
     }
+    if (test->role == 's' && gsro_flag) {
+        i_errno = IECLIENTONLY;
+        return -1;
+    }
+
+    /* Show platform support warnings only after confirming we're in client mode */
+    if (gsro_flag) {
+#if !defined(HAVE_UDP_SEGMENT) && !defined(HAVE_UDP_GRO)
+        warning("--gsro requested but UDP GSO/GRO not supported on this client; will only be enabled on server if supported");
+#elif !defined(HAVE_UDP_SEGMENT)
+        warning("--gsro requested but UDP GSO not supported on this client; will be enabled on server if supported");
+#elif !defined(HAVE_UDP_GRO)
+        warning("--gsro requested but UDP GRO not supported on this client; will be enabled on server if supported");
+#endif
+    }
 
 #if defined(HAVE_SSL)
 
@@ -1914,6 +1938,18 @@ iperf_parse_arguments(struct iperf_test *test, int argc, char **argv)
 	i_errno = IEUDPBLOCKSIZE;
 	return -1;
     }
+
+    if (test->protocol->id == Pudp && test->settings->gso) {
+        test->settings->gso_dg_size = blksize;
+        /* use the multiple of datagram size for the best efficiency. */
+        if (test->settings->gso_dg_size > 0) {
+            test->settings->gso_bf_size = (test->settings->gso_bf_size / test->settings->gso_dg_size) * test->settings->gso_dg_size;
+        } else {
+            /* If gso_dg_size is 0 (unlimited bandwidth), use default UDP datagram size */
+            test->settings->gso_dg_size = DEFAULT_UDP_BLKSIZE;
+        }
+    }
+
     test->settings->blksize = blksize;
 
     if (!rate_flag)
@@ -2457,6 +2493,16 @@ send_parameters(struct iperf_test *test)
 	    cJSON_AddNumberToObject(j, "pacing_timer", test->settings->pacing_timer);
 	if (test->settings->burst)
 	    cJSON_AddNumberToObject(j, "burst", test->settings->burst);
+
+	/* Send UDP GSO/GRO settings from client to server */
+	/* Always send these fields to allow server to use GSO/GRO even if client doesn't support it */
+	if (test->protocol->id == Pudp) {
+	    cJSON_AddNumberToObject(j, "gso", test->settings->gso);
+	    cJSON_AddNumberToObject(j, "gso_dg_size", test->settings->gso_dg_size);
+	    cJSON_AddNumberToObject(j, "gso_bf_size", test->settings->gso_bf_size);
+	    cJSON_AddNumberToObject(j, "gro", test->settings->gro);
+	    cJSON_AddNumberToObject(j, "gro_bf_size", test->settings->gro_bf_size);
+	}
 	if (test->settings->tos)
 	    cJSON_AddNumberToObject(j, "TOS", test->settings->tos);
 	if (test->settings->flowlabel)
@@ -2580,6 +2626,31 @@ get_parameters(struct iperf_test *test)
 	    test->settings->socket_bufsize = j_p->valueint;
 	if ((j_p = iperf_cJSON_GetObjectItemType(j, "len", cJSON_Number)) != NULL)
 	    test->settings->blksize = j_p->valueint;
+
+	/* Accept UDP GSO/GRO settings provided by the client */
+	/* Always accept these fields to allow server to use GSO/GRO based on its own support */
+	if ((j_p = iperf_cJSON_GetObjectItemType(j, "gso", cJSON_Number)) != NULL)
+	    test->settings->gso = j_p->valueint;
+	if ((j_p = iperf_cJSON_GetObjectItemType(j, "gso_dg_size", cJSON_Number)) != NULL)
+	    test->settings->gso_dg_size = j_p->valueint;
+	if ((j_p = iperf_cJSON_GetObjectItemType(j, "gso_bf_size", cJSON_Number)) != NULL)
+	    test->settings->gso_bf_size = j_p->valueint;
+
+	/* Backward-compatibility: If client didn't send GSO params, derive from blksize. */
+	if (test->protocol->id == Pudp && test->settings->gso == 1 && test->settings->gso_dg_size == 0) {
+	    test->settings->gso_dg_size = test->settings->blksize;
+	    if (test->settings->gso_dg_size > 0) {
+	        test->settings->gso_bf_size = (test->settings->gso_bf_size / test->settings->gso_dg_size) * test->settings->gso_dg_size;
+	    } else {
+	        test->settings->gso_dg_size = DEFAULT_UDP_BLKSIZE;
+	    }
+	}
+
+	if ((j_p = iperf_cJSON_GetObjectItemType(j, "gro", cJSON_Number)) != NULL)
+	    test->settings->gro = j_p->valueint;
+	if ((j_p = iperf_cJSON_GetObjectItemType(j, "gro_bf_size", cJSON_Number)) != NULL)
+	    test->settings->gro_bf_size = j_p->valueint;
+
 	if ((j_p = iperf_cJSON_GetObjectItemType(j, "bandwidth", cJSON_Number)) != NULL)
 	    test->settings->rate = j_p->valueint;
 	if ((j_p = iperf_cJSON_GetObjectItemType(j, "fqrate", cJSON_Number)) != NULL)
@@ -3239,6 +3310,12 @@ iperf_defaults(struct iperf_test *testp)
     testp->settings->fqrate = 0;
     testp->settings->pacing_timer = DEFAULT_PACING_TIMER;
     testp->settings->burst = 0;
+    /* Always initialize GSO/GRO fields to allow client-server negotiation */
+    testp->settings->gso = 0;  /* Disable GSO by default, enabled via --gsro */
+    testp->settings->gso_dg_size = 0;
+    testp->settings->gso_bf_size = GSO_BF_MAX_SIZE;
+    testp->settings->gro = 0;  /* Disable GRO by default, enabled via --gsro */
+    testp->settings->gro_bf_size = GRO_BF_MAX_SIZE;
     testp->settings->mss = 0;
     testp->settings->bytes = 0;
     testp->settings->blocks = 0;
@@ -3552,6 +3629,10 @@ iperf_reset_test(struct iperf_test *test)
     test->settings->burst = 0;
     test->settings->mss = 0;
     test->settings->tos = 0;
+    /* Always initialize GSO/GRO fields */
+    test->settings->gso_dg_size = 0;
+    test->settings->gso_bf_size = GSO_BF_MAX_SIZE;
+    test->settings->gro_bf_size = GRO_BF_MAX_SIZE;
     test->settings->dont_fragment = 0;
     test->zerocopy = 0;
     test->settings->skip_rx_copy = 0;
@@ -4716,6 +4797,7 @@ iperf_new_stream(struct iperf_test *test, int s, int sender)
 {
     struct iperf_stream *sp;
     int ret = 0;
+    int size;
 
     char template[1024];
     if (test->tmp_template) {
@@ -4774,13 +4856,20 @@ iperf_new_stream(struct iperf_test *test, int s, int sender)
         free(sp);
         return NULL;
     }
-    if (ftruncate(sp->buffer_fd, test->settings->blksize) < 0) {
+    size = test->settings->blksize;
+    if (test->protocol->id == Pudp && test->settings->gso && (size < test->settings->gso_bf_size))
+        size = test->settings->gso_bf_size;
+    if (test->protocol->id == Pudp && test->settings->gro && (size < test->settings->gro_bf_size))
+        size = test->settings->gro_bf_size;
+    if (sp->test->debug)
+        printf("Buffer %d bytes\n", size);
+    if (ftruncate(sp->buffer_fd, size) < 0) {
         i_errno = IECREATESTREAM;
         free(sp->result);
         free(sp);
         return NULL;
     }
-    sp->buffer = (char *) mmap(NULL, test->settings->blksize, PROT_READ|PROT_WRITE, MAP_SHARED, sp->buffer_fd, 0);
+    sp->buffer = (char *) mmap(NULL, size, PROT_READ|PROT_WRITE, MAP_SHARED, sp->buffer_fd, 0);
     if (sp->buffer == MAP_FAILED) {
         i_errno = IECREATESTREAM;
         free(sp->result);

src/iperf_api.h

@@ -106,6 +106,7 @@ typedef atomic_uint_fast64_t atomic_iperf_size_t;
 #define OPT_SKIP_RX_COPY 32
 #define OPT_JSON_STREAM_FULL_OUTPUT 33
 #define OPT_SERVER_MAX_DURATION 34
+#define OPT_GSRO 35
 
 /* states */
 #define TEST_START 1

src/iperf_client_api.c

@@ -516,6 +516,17 @@ iperf_connect(struct iperf_test *test)
 		printf("Setting UDP block size to %d\n", test->settings->blksize);
 	    }
 	}
+	/* Initialize GSO parameters when --gsro is used */
+	if (test->settings->gso) {
+	    test->settings->gso_dg_size = test->settings->blksize;
+	    /* use the multiple of datagram size for the best efficiency. */
+	    if (test->settings->gso_dg_size > 0) {
+		test->settings->gso_bf_size = (test->settings->gso_bf_size / test->settings->gso_dg_size) * test->settings->gso_dg_size;
+	    } else {
+		/* If gso_dg_size is 0 (unlimited bandwidth), use default UDP datagram size */
+		test->settings->gso_dg_size = DEFAULT_UDP_BLKSIZE;
+	    }
+	}
 
 	/*
 	 * Regardless of whether explicitly or implicitly set, if the

src/iperf_config.h.in

@@ -132,6 +132,12 @@
 /* Have TCP_USER_TIMEOUT sockopt. */
 #undef HAVE_TCP_USER_TIMEOUT
 
+/* Have UDP_GRO sockopt. */
+#undef HAVE_UDP_GRO
+
+/* Have UDP_SEGMENT sockopt. */
+#undef HAVE_UDP_SEGMENT
+
 /* Define to 1 if you have the <unistd.h> header file. */
 #undef HAVE_UNISTD_H
 

src/iperf_locale.c

@@ -219,6 +219,7 @@ const char usage_longstr[] = "Usage: iperf3 [-s|-c host] [options]\n"
                            "  --extra-data str          data string to include in client and server JSON\n"
                            "  --get-server-output       get results from server\n"
                            "  --udp-counters-64bit      use 64-bit counters in UDP test packets\n"
+                           "  --gsro                    enable UDP GSO/GRO on both client and server (client-only option)\n"
                            "  --repeating-payload       use repeating pattern in payload, instead of\n"
                            "                            randomized payload (like in iperf2)\n"
 #if defined(HAVE_DONT_FRAGMENT)