Adding upstream version 3.2.6.upstream/3.2.6upstream

Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>

1 files changed, 1302 insertions, 0 deletions

diff --git a/src/utils/kxdpgun/main.c b/src/utils/kxdpgun/main.c
new file mode 100644
index 0000000..2ec0c0c
--- /dev/null
+++ b/src/utils/kxdpgun/main.c
@@ -0,0 +1,1302 @@
+/*  Copyright (C) 2023 CZ.NIC, z.s.p.o. <knot-dns@labs.nic.cz>
+
+    This program is free software: you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation, either version 3 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License
+    along with this program.  If not, see <https://www.gnu.org/licenses/>.
+ */
+
+#include <assert.h>
+#include <errno.h>
+#include <getopt.h>
+#include <ifaddrs.h>
+#include <inttypes.h>
+#include <net/if.h>
+#include <poll.h>
+#include <pthread.h>
+#include <signal.h>
+#include <stdbool.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <time.h>
+#include <unistd.h>
+
+#include <arpa/inet.h>
+#include <netinet/in.h>
+#include <net/if.h>
+#include <sys/ioctl.h>
+#include <sys/socket.h>
+#include <sys/resource.h>
+
+#include "libknot/libknot.h"
+#include "libknot/xdp.h"
+#include "libknot/xdp/tcp_iobuf.h"
+#ifdef ENABLE_QUIC
+#include <gnutls/gnutls.h>
+#include "libknot/xdp/quic.h"
+#endif // ENABLE_QUIC
+#include "contrib/macros.h"
+#include "contrib/mempattern.h"
+#include "contrib/openbsd/strlcat.h"
+#include "contrib/openbsd/strlcpy.h"
+#include "contrib/os.h"
+#include "contrib/sockaddr.h"
+#include "contrib/toeplitz.h"
+#include "contrib/ucw/mempool.h"
+#include "utils/common/msg.h"
+#include "utils/common/params.h"
+#include "utils/kxdpgun/ip_route.h"
+#include "utils/kxdpgun/load_queries.h"
+
+#define PROGRAM_NAME "kxdpgun"
+#define SPACE        "  "
+
+enum {
+	KXDPGUN_WAIT,
+	KXDPGUN_START,
+	KXDPGUN_STOP,
+};
+
+volatile int xdp_trigger = KXDPGUN_WAIT;
+
+volatile unsigned stats_trigger = 0;
+
+unsigned global_cpu_aff_start = 0;
+unsigned global_cpu_aff_step = 1;
+
+#define REMOTE_PORT_DEFAULT       53
+#define REMOTE_PORT_DOQ_DEFAULT  853
+#define LOCAL_PORT_MIN          2000
+#define LOCAL_PORT_MAX         65535
+#define QUIC_THREAD_PORTS        100
+
+#define RCODE_MAX (0x0F + 1)
+
+typedef struct {
+	size_t collected;
+	uint64_t duration;
+	uint64_t qry_sent;
+	uint64_t synack_recv;
+	uint64_t ans_recv;
+	uint64_t finack_recv;
+	uint64_t rst_recv;
+	uint64_t size_recv;
+	uint64_t wire_recv;
+	uint64_t rcodes_recv[RCODE_MAX];
+	pthread_mutex_t mutex;
+} kxdpgun_stats_t;
+
+static kxdpgun_stats_t global_stats = { 0 };
+
+typedef enum {
+	KXDPGUN_IGNORE_NONE     = 0,
+	KXDPGUN_IGNORE_QUERY    = (1 << 0),
+	KXDPGUN_IGNORE_LASTBYTE = (1 << 1),
+	KXDPGUN_IGNORE_CLOSE    = (1 << 2),
+} xdp_gun_ignore_t;
+
+typedef struct {
+	union {
+		struct sockaddr_in local_ip4;
+		struct sockaddr_in6 local_ip;
+		struct sockaddr_storage local_ip_ss;
+	};
+	union {
+		struct sockaddr_in target_ip4;
+		struct sockaddr_in6 target_ip;
+		struct sockaddr_storage target_ip_ss;
+	};
+	char		dev[IFNAMSIZ];
+	uint64_t	qps, duration;
+	unsigned	at_once;
+	uint16_t	msgid;
+	uint16_t	edns_size;
+	uint16_t	vlan_tci;
+	uint8_t		local_mac[6], target_mac[6];
+	uint8_t		local_ip_range;
+	bool		ipv6;
+	bool		tcp;
+	bool		quic;
+	bool		quic_full_handshake;
+	const char	*sending_mode;
+	xdp_gun_ignore_t  ignore1;
+	knot_tcp_ignore_t ignore2;
+	uint16_t	target_port;
+	knot_xdp_filter_flag_t flags;
+	unsigned	n_threads, thread_id;
+	knot_eth_rss_conf_t *rss_conf;
+} xdp_gun_ctx_t;
+
+const static xdp_gun_ctx_t ctx_defaults = {
+	.dev[0] = '\0',
+	.edns_size = 1232,
+	.qps = 1000,
+	.duration = 5000000UL, // usecs
+	.at_once = 10,
+	.sending_mode = "",
+	.target_port = 0,
+	.flags = KNOT_XDP_FILTER_UDP | KNOT_XDP_FILTER_PASS,
+};
+
+static void sigterm_handler(int signo)
+{
+	assert(signo == SIGTERM || signo == SIGINT);
+	xdp_trigger = KXDPGUN_STOP;
+}
+
+static void sigusr_handler(int signo)
+{
+	assert(signo == SIGUSR1);
+	if (global_stats.collected == 0) {
+		stats_trigger++;
+	}
+}
+
+static void clear_stats(kxdpgun_stats_t *st)
+{
+	pthread_mutex_lock(&st->mutex);
+	st->duration    = 0;
+	st->qry_sent    = 0;
+	st->synack_recv = 0;
+	st->ans_recv    = 0;
+	st->finack_recv = 0;
+	st->rst_recv    = 0;
+	st->size_recv   = 0;
+	st->wire_recv   = 0;
+	st->collected   = 0;
+	memset(st->rcodes_recv, 0, sizeof(st->rcodes_recv));
+	pthread_mutex_unlock(&st->mutex);
+}
+
+static size_t collect_stats(kxdpgun_stats_t *into, const kxdpgun_stats_t *what)
+{
+	pthread_mutex_lock(&into->mutex);
+	into->duration = MAX(into->duration, what->duration);
+	into->qry_sent    += what->qry_sent;
+	into->synack_recv += what->synack_recv;
+	into->ans_recv    += what->ans_recv;
+	into->finack_recv += what->finack_recv;
+	into->rst_recv    += what->rst_recv;
+	into->size_recv   += what->size_recv;
+	into->wire_recv   += what->wire_recv;
+	for (int i = 0; i < RCODE_MAX; i++) {
+		into->rcodes_recv[i] += what->rcodes_recv[i];
+	}
+	size_t res = ++into->collected;
+	pthread_mutex_unlock(&into->mutex);
+	return res;
+}
+
+static void print_stats(kxdpgun_stats_t *st, bool tcp, bool quic, bool recv)
+{
+	pthread_mutex_lock(&st->mutex);
+
+#define ps(counter)  ((counter) * 1000 / (st->duration / 1000))
+#define pct(counter) ((counter) * 100 / st->qry_sent)
+
+	const char *name = tcp ? "SYNs:    " : quic ? "initials:" : "queries: ";
+	printf("total %s    %"PRIu64" (%"PRIu64" pps)\n", name,
+	       st->qry_sent, ps(st->qry_sent));
+	if (st->qry_sent > 0 && recv) {
+		if (tcp || quic) {
+		name = tcp ? "established:" : "handshakes: ";
+		printf("total %s %"PRIu64" (%"PRIu64" pps) (%"PRIu64"%%)\n", name,
+		       st->synack_recv, ps(st->synack_recv), pct(st->synack_recv));
+		}
+		printf("total replies:     %"PRIu64" (%"PRIu64" pps) (%"PRIu64"%%)\n",
+		       st->ans_recv, ps(st->ans_recv), pct(st->ans_recv));
+		if (tcp) {
+		printf("total closed:      %"PRIu64" (%"PRIu64" pps) (%"PRIu64"%%)\n",
+		       st->finack_recv, ps(st->finack_recv), pct(st->finack_recv));
+		}
+		if (st->rst_recv > 0) {
+		printf("total reset:       %"PRIu64" (%"PRIu64" pps) (%"PRIu64"%%)\n",
+		       st->rst_recv, ps(st->rst_recv), pct(st->rst_recv));
+		}
+		printf("average DNS reply size: %"PRIu64" B\n",
+		       st->ans_recv > 0 ? st->size_recv / st->ans_recv : 0);
+		printf("average Ethernet reply rate: %"PRIu64" bps (%.2f Mbps)\n",
+		       ps(st->wire_recv * 8), ps((float)st->wire_recv * 8 / (1000 * 1000)));
+
+		for (int i = 0; i < RCODE_MAX; i++) {
+			if (st->rcodes_recv[i] > 0) {
+				const knot_lookup_t *rcode = knot_lookup_by_id(knot_rcode_names, i);
+				const char *rcname = rcode == NULL ? "unknown" : rcode->name;
+				int space = MAX(9 - strlen(rcname), 0);
+				printf("responded %s: %.*s%"PRIu64"\n",
+				       rcname, space, "         ", st->rcodes_recv[i]);
+			}
+		}
+	}
+	printf("duration: %"PRIu64" s\n", (st->duration / (1000 * 1000)));
+
+	pthread_mutex_unlock(&st->mutex);
+}
+
+inline static void timer_start(struct timespec *timesp)
+{
+	clock_gettime(CLOCK_MONOTONIC, timesp);
+}
+
+inline static uint64_t timer_end(struct timespec *timesp)
+{
+	struct timespec end;
+	clock_gettime(CLOCK_MONOTONIC, &end);
+	uint64_t res = (end.tv_sec - timesp->tv_sec) * (uint64_t)1000000;
+	res += ((int64_t)end.tv_nsec - timesp->tv_nsec) / 1000;
+	return res;
+}
+
+static unsigned addr_bits(bool ipv6)
+{
+	return ipv6 ? 128 : 32;
+}
+
+static void shuffle_sockaddr4(struct sockaddr_in *dst, struct sockaddr_in *src, uint64_t increment)
+{
+	memcpy(&dst->sin_addr, &src->sin_addr, sizeof(dst->sin_addr));
+	if (increment > 0) {
+		dst->sin_addr.s_addr = htobe32(be32toh(src->sin_addr.s_addr) + increment);
+	}
+}
+
+static void shuffle_sockaddr6(struct sockaddr_in6 *dst, struct sockaddr_in6 *src, uint64_t increment)
+{
+	memcpy(&dst->sin6_addr, &src->sin6_addr, sizeof(dst->sin6_addr));
+	if (increment > 0) {
+		uint64_t *dst_addr = (uint64_t *)&dst->sin6_addr;
+		uint64_t *src_addr = (uint64_t *)&src->sin6_addr;
+		dst_addr[1] = htobe64(be64toh(src_addr[1]) + increment);
+	}
+}
+
+static void shuffle_sockaddr(struct sockaddr_in6 *dst, struct sockaddr_in6 *src,
+                             uint16_t port, uint64_t increment)
+{
+	dst->sin6_family = src->sin6_family;
+	dst->sin6_port = htobe16(port);
+	if (src->sin6_family == AF_INET6) {
+		shuffle_sockaddr6(dst, src, increment);
+	} else {
+		shuffle_sockaddr4((struct sockaddr_in *)dst, (struct sockaddr_in *)src, increment);
+	}
+}
+
+static void next_payload(struct pkt_payload **payload, int increment)
+{
+	if (*payload == NULL) {
+		*payload = global_payloads;
+	}
+	for (int i = 0; i < increment; i++) {
+		if ((*payload)->next == NULL) {
+			*payload = global_payloads;
+		} else {
+			*payload = (*payload)->next;
+		}
+	}
+}
+
+static void put_dns_payload(struct iovec *put_into, bool zero_copy, xdp_gun_ctx_t *ctx, struct pkt_payload **payl)
+{
+	if (zero_copy) {
+		put_into->iov_base = (*payl)->payload;
+	} else {
+		memcpy(put_into->iov_base, (*payl)->payload, (*payl)->len);
+	}
+	put_into->iov_len = (*payl)->len;
+	next_payload(payl, ctx->n_threads);
+}
+
+#ifdef ENABLE_QUIC
+static uint16_t get_rss_id(xdp_gun_ctx_t *ctx, uint16_t local_port)
+{
+	assert(ctx->rss_conf);
+
+	const uint8_t *key = (const uint8_t *)&(ctx->rss_conf->data[ctx->rss_conf->table_size]);
+	const size_t key_len = ctx->rss_conf->key_size;
+	uint8_t data[2 * sizeof(struct in6_addr) + 2 * sizeof(uint16_t)];
+
+	size_t addr_len;
+	if (ctx->ipv6) {
+		addr_len = sizeof(struct in6_addr);
+		memcpy(data, &ctx->target_ip.sin6_addr, addr_len);
+		memcpy(data + addr_len, &ctx->local_ip.sin6_addr, addr_len);
+	} else {
+		addr_len = sizeof(struct in_addr);
+		memcpy(data, &ctx->target_ip4.sin_addr, addr_len);
+		memcpy(data + addr_len, &ctx->local_ip4.sin_addr, addr_len);
+	}
+
+	uint16_t src_port = htobe16(ctx->target_port);
+	memcpy(data + 2 * addr_len, &src_port, sizeof(src_port));
+	uint16_t dst_port = htobe16(local_port);
+	memcpy(data + 2 * addr_len + sizeof(uint16_t), &dst_port, sizeof(dst_port));
+
+	size_t data_len = 2 * addr_len + 2 * sizeof(uint16_t);
+	uint16_t hash = toeplitz_hash(key, key_len, data, data_len);
+
+	return ctx->rss_conf->data[hash & ctx->rss_conf->mask];
+}
+
+static uint16_t adjust_port(xdp_gun_ctx_t *ctx, uint16_t local_port)
+{
+	assert(UINT16_MAX == LOCAL_PORT_MAX);
+
+	if (local_port < LOCAL_PORT_MIN) {
+		local_port = LOCAL_PORT_MIN;
+	}
+
+	if (ctx->rss_conf == NULL) {
+		return local_port;
+	}
+
+	for (int i = 0; i < UINT16_MAX; i++) {
+		if (ctx->thread_id == get_rss_id(ctx, local_port)) {
+			break;
+		}
+		local_port++;
+		if (local_port < LOCAL_PORT_MIN) {
+			local_port = LOCAL_PORT_MIN;
+		}
+	}
+
+	return local_port;
+}
+#endif // ENABLE_QUIC
+
+static unsigned alloc_pkts(knot_xdp_msg_t *pkts, struct knot_xdp_socket *xsk,
+                           xdp_gun_ctx_t *ctx, uint64_t tick)
+{
+	uint64_t unique = (tick * ctx->n_threads + ctx->thread_id) * ctx->at_once;
+
+	knot_xdp_msg_flag_t flags = ctx->ipv6 ? KNOT_XDP_MSG_IPV6 : 0;
+	if (ctx->tcp) {
+		flags |= (KNOT_XDP_MSG_TCP | KNOT_XDP_MSG_SYN | KNOT_XDP_MSG_MSS);
+	} else if (ctx->quic) {
+		return ctx->at_once; // NOOP
+	}
+	if (ctx->vlan_tci != 0) {
+		flags |= KNOT_XDP_MSG_VLAN;
+	}
+
+	for (unsigned i = 0; i < ctx->at_once; i++) {
+		int ret = knot_xdp_send_alloc(xsk, flags, &pkts[i]);
+		if (ret != KNOT_EOK) {
+			return i;
+		}
+
+		uint16_t port_range = LOCAL_PORT_MAX - LOCAL_PORT_MIN + 1;
+		uint16_t local_port = LOCAL_PORT_MIN + unique % port_range;
+		uint64_t ip_incr = (unique / port_range) % (1 << (addr_bits(ctx->ipv6) - ctx->local_ip_range));
+		shuffle_sockaddr(&pkts[i].ip_from, &ctx->local_ip,  local_port, ip_incr);
+		shuffle_sockaddr(&pkts[i].ip_to,   &ctx->target_ip, ctx->target_port, 0);
+
+		memcpy(pkts[i].eth_from, ctx->local_mac, 6);
+		memcpy(pkts[i].eth_to, ctx->target_mac, 6);
+
+		pkts[i].vlan_tci = ctx->vlan_tci;
+
+		unique++;
+	}
+	return ctx->at_once;
+}
+
+inline static bool check_dns_payload(struct iovec *payl, xdp_gun_ctx_t *ctx,
+                                     kxdpgun_stats_t *st)
+{
+	if (payl->iov_len < KNOT_WIRE_HEADER_SIZE ||
+	    memcmp(payl->iov_base, &ctx->msgid, sizeof(ctx->msgid)) != 0) {
+		return false;
+	}
+	st->rcodes_recv[((uint8_t *)payl->iov_base)[3] & 0x0F]++;
+	st->size_recv += payl->iov_len;
+	st->ans_recv++;
+	return true;
+}
+
+void *xdp_gun_thread(void *_ctx)
+{
+	xdp_gun_ctx_t *ctx = _ctx;
+	struct knot_xdp_socket *xsk;
+	struct timespec timer;
+	knot_xdp_msg_t pkts[ctx->at_once];
+	uint64_t errors = 0, lost = 0, duration = 0;
+	kxdpgun_stats_t local_stats = { 0 };
+	unsigned stats_triggered = 0;
+	knot_tcp_table_t *tcp_table = NULL;
+#ifdef ENABLE_QUIC
+	knot_xquic_table_t *quic_table = NULL;
+	struct knot_quic_creds *quic_creds = NULL;
+	knot_xdp_msg_t quic_fake_req = { 0 };
+	list_t quic_sessions;
+	init_list(&quic_sessions);
+#endif // ENABLE_QUIC
+	const uint64_t extra_wait = ctx->quic ? 4000000 : 1000000;
+
+	if (ctx->tcp) {
+		tcp_table = knot_tcp_table_new(ctx->qps, NULL);
+		if (tcp_table == NULL) {
+			ERR2("failed to allocate TCP connection table");
+			return NULL;
+		}
+	}
+	if (ctx->quic) {
+#ifdef ENABLE_QUIC
+		quic_creds = knot_xquic_init_creds(false, NULL, NULL);
+		if (quic_creds == NULL) {
+			ERR2("failed to initialize QUIC context");
+			return NULL;
+		}
+		quic_table = knot_xquic_table_new(ctx->qps * 100, SIZE_MAX, SIZE_MAX, 1232, quic_creds);
+		if (quic_table == NULL) {
+			ERR2("failed to allocate QUIC connection table");
+			return NULL;
+		}
+		ctx->target_ip.sin6_port = htobe16(ctx->target_port);
+
+		memcpy(quic_fake_req.eth_from, ctx->target_mac,  sizeof(ctx->target_mac));
+		memcpy(quic_fake_req.eth_to,   ctx->local_mac,   sizeof(ctx->local_mac));
+		memcpy(&quic_fake_req.ip_from, &ctx->target_ip,  sizeof(quic_fake_req.ip_from));
+		memcpy(&quic_fake_req.ip_to,   &ctx->local_ip,   sizeof(quic_fake_req.ip_to));
+		quic_fake_req.flags = ctx->ipv6 ? KNOT_XDP_MSG_IPV6 : 0;
+#else
+		assert(0);
+#endif // ENABLE_QUIC
+	}
+
+	knot_xdp_load_bpf_t mode = (ctx->thread_id == 0 ?
+	                            KNOT_XDP_LOAD_BPF_ALWAYS : KNOT_XDP_LOAD_BPF_NEVER);
+	/*
+	 * This mutex prevents libbpf from logging:
+	 * 'libbpf: can't get link by id (5535): Resource temporarily unavailable'
+	*/
+	pthread_mutex_lock(&global_stats.mutex);
+	int ret = knot_xdp_init(&xsk, ctx->dev, ctx->thread_id, ctx->flags,
+	                        LOCAL_PORT_MIN, LOCAL_PORT_MIN, mode, NULL);
+	pthread_mutex_unlock(&global_stats.mutex);
+	if (ret != KNOT_EOK) {
+		ERR2("failed to initialize XDP socket#%u (%s)",
+		     ctx->thread_id, knot_strerror(ret));
+		knot_tcp_table_free(tcp_table);
+		return NULL;
+	}
+
+	if (ctx->thread_id == 0) {
+		INFO2("using interface %s, XDP threads %u, %s%s%s, %s mode",
+		      ctx->dev, ctx->n_threads,
+		      (ctx->tcp ? "TCP" : ctx->quic ? "QUIC" : "UDP"),
+		      (ctx->sending_mode[0] != '\0' ? " mode " : ""),
+		      (ctx->sending_mode[0] != '\0' ? ctx->sending_mode : ""),
+		      (knot_eth_xdp_mode(if_nametoindex(ctx->dev)) == KNOT_XDP_MODE_FULL ?
+		       "native" : "emulated"));
+	}
+
+	struct pollfd pfd = { knot_xdp_socket_fd(xsk), POLLIN, 0 };
+
+	while (xdp_trigger == KXDPGUN_WAIT) {
+		usleep(1000);
+	}
+
+	uint64_t tick = 0;
+	struct pkt_payload *payload_ptr = NULL;
+	next_payload(&payload_ptr, ctx->thread_id);
+
+#ifdef ENABLE_QUIC
+	knot_sweep_stats_t sweep_stats = { 0 };
+	uint16_t local_ports[QUIC_THREAD_PORTS];
+	uint16_t port = LOCAL_PORT_MIN;
+	for (int i = 0; i < QUIC_THREAD_PORTS; ++i) {
+		local_ports[i] = adjust_port(ctx, port);
+		port = local_ports[i] + 1;
+		assert(port >= LOCAL_PORT_MIN);
+	}
+	size_t local_ports_it = 0;
+#endif // ENABLE_QUIC
+
+	timer_start(&timer);
+
+	while (duration < ctx->duration + extra_wait) {
+
+		// sending part
+		if (duration < ctx->duration) {
+			while (1) {
+				knot_xdp_send_prepare(xsk);
+				unsigned alloced = alloc_pkts(pkts, xsk, ctx, tick);
+				if (alloced < ctx->at_once) {
+					lost += ctx->at_once - alloced;
+					if (alloced == 0) {
+						break;
+					}
+				}
+
+				if (ctx->tcp) {
+					for (int i = 0; i < alloced; i++) {
+						pkts[i].payload.iov_len = 0;
+					}
+				} else if (ctx->quic) {
+#ifdef ENABLE_QUIC
+					uint16_t local_port = local_ports[local_ports_it++ % QUIC_THREAD_PORTS];
+					for (unsigned i = 0; i < ctx->at_once; i++) {
+						knot_xquic_conn_t *newconn = NULL;
+						ctx->local_ip.sin6_port = htobe16(local_port);
+						ret = knot_xquic_client(quic_table, &ctx->target_ip, &ctx->local_ip, &newconn);
+						if (ret == KNOT_EOK) {
+							struct iovec tmp = { knot_xquic_stream_add_data(newconn, 0, NULL, payload_ptr->len), 0 };
+							put_dns_payload(&tmp, false, ctx, &payload_ptr);
+							if (EMPTY_LIST(quic_sessions)) {
+								newconn->streams_count = -1;
+							} else {
+								void *session = HEAD(quic_sessions);
+								rem_node(session);
+								(void)knot_xquic_session_load(newconn, session);
+							}
+							quic_fake_req.ip_to.sin6_port = htobe16(local_port);
+							ret = knot_xquic_send(quic_table, newconn, xsk, &quic_fake_req, KNOT_EOK, 1, (ctx->ignore1 & KXDPGUN_IGNORE_LASTBYTE));
+						}
+						if (ret == KNOT_EOK) {
+							local_stats.qry_sent++;
+						}
+					}
+					(void)knot_xdp_send_finish(xsk);
+#endif // ENABLE_QUIC
+					break;
+				} else {
+					for (int i = 0; i < alloced; i++) {
+						put_dns_payload(&pkts[i].payload, false,
+						                ctx, &payload_ptr);
+					}
+				}
+
+				uint32_t really_sent = 0;
+				if (knot_xdp_send(xsk, pkts, alloced, &really_sent) != KNOT_EOK) {
+					lost += alloced;
+				}
+				local_stats.qry_sent += really_sent;
+				(void)knot_xdp_send_finish(xsk);
+
+				break;
+			}
+		}
+
+		// receiving part
+		if (!(ctx->flags & KNOT_XDP_FILTER_DROP)) {
+			while (1) {
+				ret = poll(&pfd, 1, 0);
+				if (ret < 0) {
+					errors++;
+					break;
+				}
+				if (!pfd.revents) {
+					break;
+				}
+
+				uint32_t recvd = 0;
+				size_t wire = 0;
+				(void)knot_xdp_recv(xsk, pkts, ctx->at_once, &recvd, &wire);
+				if (recvd == 0) {
+					break;
+				}
+				if (ctx->tcp) {
+					knot_tcp_relay_t relays[recvd];
+					ret = knot_tcp_recv(relays, pkts, recvd, tcp_table, NULL, ctx->ignore2);
+					if (ret != KNOT_EOK) {
+						errors++;
+						break;
+					}
+
+					for (size_t i = 0; i < recvd; i++) {
+						knot_tcp_relay_t *rl = &relays[i];
+						struct iovec payl;
+						switch (rl->action) {
+						case XDP_TCP_ESTABLISH:
+							local_stats.synack_recv++;
+							if (ctx->ignore1 & KXDPGUN_IGNORE_QUERY) {
+								break;
+							}
+							put_dns_payload(&payl, true, ctx, &payload_ptr);
+							ret = knot_tcp_reply_data(rl, tcp_table,
+							                          (ctx->ignore1 & KXDPGUN_IGNORE_LASTBYTE),
+							                          payl.iov_base, payl.iov_len);
+							if (ret != KNOT_EOK) {
+								errors++;
+							}
+							break;
+						case XDP_TCP_CLOSE:
+							local_stats.finack_recv++;
+							break;
+						case XDP_TCP_RESET:
+							local_stats.rst_recv++;
+							break;
+						default:
+							break;
+						}
+						for (size_t j = 0; j < rl->inbufs_count; j++) {
+							if (check_dns_payload(&rl->inbufs[j], ctx, &local_stats)) {
+								if (!(ctx->ignore1 & KXDPGUN_IGNORE_CLOSE)) {
+									rl->answer = XDP_TCP_CLOSE;
+								}
+							}
+						}
+					}
+
+					ret = knot_tcp_send(xsk, relays, recvd, ctx->at_once);
+					if (ret != KNOT_EOK) {
+						errors++;
+					}
+					(void)knot_xdp_send_finish(xsk);
+
+					knot_tcp_cleanup(tcp_table, relays, recvd);
+				} else if (ctx->quic) {
+#ifdef ENABLE_QUIC
+					knot_xquic_conn_t *relays[recvd];
+					for (size_t i = 0; i < recvd; i++) {
+						ret = knot_xquic_handle(quic_table, &pkts[i], 5000000000L, &relays[i]);
+						if (ret == KNOT_ECONN) {
+							local_stats.rst_recv++;
+							continue;
+						} else if (ret != 0) {
+							errors++;
+							break;
+						}
+
+						knot_xquic_conn_t *rl = relays[i];
+						if (rl == NULL) {
+							continue;
+						}
+
+						bool sess_ticket = (gnutls_session_get_flags(rl->tls_session) & GNUTLS_SFLAGS_SESSION_TICKET);
+
+						if (sess_ticket && !rl->session_taken && !ctx->quic_full_handshake) {
+							rl->session_taken = true;
+							void *session = knot_xquic_session_save(rl);
+							if (session != NULL) {
+								add_tail(&quic_sessions, session);
+							}
+						}
+
+						if (rl->handshake_done && rl->streams_count == -1) {
+							rl->streams_count = 1;
+
+							local_stats.synack_recv++;
+							if ((ctx->ignore1 & KXDPGUN_IGNORE_QUERY)) {
+								knot_xquic_table_rem(relays[i], quic_table);
+								knot_xquic_cleanup(&relays[i], 1);
+								relays[i] = NULL;
+								continue;
+							}
+						}
+						if (!rl->handshake_done && rl->streams_count == -1) {
+							continue;
+						}
+
+						knot_xquic_stream_t *stream0 = knot_xquic_conn_get_stream(rl, 0, false);
+						assert(stream0 != NULL);
+
+						if ((ctx->ignore2 & XDP_TCP_IGNORE_ESTABLISH)) {
+							knot_xquic_table_rem(relays[i], quic_table);
+							knot_xquic_cleanup(&relays[i], 1);
+							relays[i] = NULL;
+							local_stats.synack_recv++;
+							continue;
+						}
+
+						stream0 = knot_xquic_conn_get_stream(rl, 0, false);
+						if (stream0 != NULL && stream0->inbuf_fin != NULL) {
+							check_dns_payload(stream0->inbuf_fin, ctx, &local_stats);
+							free(stream0->inbuf_fin);
+							stream0->inbuf_fin = NULL;
+
+							if ((ctx->ignore2 & XDP_TCP_IGNORE_DATA_ACK)) {
+								knot_xquic_table_rem(relays[i], quic_table);
+								knot_xquic_cleanup(&relays[i], 1);
+								relays[i] = NULL;
+								continue;
+							}
+						}
+						ret = knot_xquic_send(quic_table, rl, xsk, &pkts[i], KNOT_EOK, 4, (ctx->ignore1 & KXDPGUN_IGNORE_LASTBYTE));
+						if (ret != KNOT_EOK) {
+							errors++;
+						}
+					}
+					knot_xquic_cleanup(relays, recvd);
+					(void)knot_xdp_send_finish(xsk);
+#endif // ENABLE_QUIC
+				} else {
+					for (int i = 0; i < recvd; i++) {
+						(void)check_dns_payload(&pkts[i].payload, ctx,
+						                        &local_stats);
+					}
+				}
+				local_stats.wire_recv += wire;
+				knot_xdp_recv_finish(xsk, pkts, recvd);
+				pfd.revents = 0;
+			}
+		}
+
+#ifdef ENABLE_QUIC
+		if (ctx->quic) {
+			(void)knot_xquic_table_sweep(quic_table, &sweep_stats);
+		}
+#endif // ENABLE_QUIC
+
+		// speed and signal part
+		uint64_t dura_exp = (local_stats.qry_sent * 1000000) / ctx->qps;
+		duration = timer_end(&timer);
+		if (xdp_trigger == KXDPGUN_STOP && ctx->duration > duration) {
+			ctx->duration = duration;
+		}
+		if (stats_trigger > stats_triggered) {
+			assert(stats_trigger == stats_triggered + 1);
+			stats_triggered++;
+
+			local_stats.duration = duration;
+			size_t collected = collect_stats(&global_stats, &local_stats);
+			assert(collected <= ctx->n_threads);
+			if (collected == ctx->n_threads) {
+				print_stats(&global_stats, ctx->tcp, ctx->quic,
+				            !(ctx->flags & KNOT_XDP_FILTER_DROP));
+				clear_stats(&global_stats);
+			}
+		}
+		if (dura_exp > duration) {
+			usleep(dura_exp - duration);
+		}
+		if (duration > ctx->duration) {
+			usleep(1000);
+		}
+		tick++;
+	}
+
+	knot_xdp_deinit(xsk);
+
+	knot_tcp_table_free(tcp_table);
+#ifdef ENABLE_QUIC
+	knot_xquic_table_free(quic_table);
+	struct knot_quic_session *n, *nxt;
+	WALK_LIST_DELSAFE(n, nxt, quic_sessions) {
+		knot_xquic_session_load(NULL, n);
+	}
+	knot_xquic_free_creds(quic_creds);
+#endif // ENABLE_QUIC
+
+	char recv_str[40] = "", lost_str[40] = "", err_str[40] = "";
+	if (!(ctx->flags & KNOT_XDP_FILTER_DROP)) {
+		(void)snprintf(recv_str, sizeof(recv_str), ", received %"PRIu64, local_stats.ans_recv);
+	}
+	if (lost > 0) {
+		(void)snprintf(lost_str, sizeof(lost_str), ", lost %"PRIu64, lost);
+	}
+	if (errors > 0) {
+		(void)snprintf(err_str, sizeof(err_str), ", errors %"PRIu64, errors);
+	}
+	INFO2("thread#%02u: sent %"PRIu64"%s%s%s",
+	      ctx->thread_id, local_stats.qry_sent, recv_str, lost_str, err_str);
+	local_stats.duration = ctx->duration;
+	collect_stats(&global_stats, &local_stats);
+
+	return NULL;
+}
+
+static int dev2mac(const char *dev, uint8_t *mac)
+{
+	struct ifreq ifr;
+	memset(&ifr, 0, sizeof(ifr));
+	int fd = socket(AF_INET, SOCK_DGRAM, 0);
+	if (fd < 0) {
+		return -errno;
+	}
+	strlcpy(ifr.ifr_name, dev, IFNAMSIZ);
+
+	int ret = ioctl(fd, SIOCGIFHWADDR, &ifr);
+	if (ret >= 0) {
+		memcpy(mac, ifr.ifr_hwaddr.sa_data, 6);
+	} else {
+		ret = -errno;
+	}
+	close(fd);
+	return ret;
+}
+
+static bool mac_empty(const uint8_t *mac)
+{
+	static const uint8_t unset_mac[6] = { 0 };
+	return (memcmp(mac, unset_mac, sizeof(unset_mac)) == 0);
+}
+
+static int mac_sscan(const char *src, uint8_t *dst)
+{
+	int tmp[6];
+	if (6 != sscanf(src, "%2x:%2x:%2x:%2x:%2x:%2x",
+	                &tmp[0], &tmp[1], &tmp[2], &tmp[3], &tmp[4], &tmp[5])) {
+		return KNOT_EINVAL;
+	}
+
+	for (int i = 0; i < 6; i++) {
+		dst[i] = (uint8_t)tmp[i];
+	}
+
+	return KNOT_EOK;
+}
+
+static bool configure_target(char *target_str, char *local_ip, xdp_gun_ctx_t *ctx)
+{
+	int val;
+	char *at = strrchr(target_str, '@');
+	if (at != NULL && (val = atoi(at + 1)) > 0 && val <= 0xffff) {
+		ctx->target_port = val;
+		*at = '\0';
+	}
+
+	ctx->ipv6 = false;
+	if (inet_pton(AF_INET, target_str, &ctx->target_ip4.sin_addr) <= 0) {
+		ctx->ipv6 = true;
+		ctx->target_ip.sin6_family = AF_INET6;
+		if (inet_pton(AF_INET6, target_str, &ctx->target_ip.sin6_addr) <= 0) {
+			ERR2("invalid target IP");
+			return false;
+		}
+	} else {
+		ctx->target_ip.sin6_family = AF_INET;
+	}
+
+	struct sockaddr_storage via = { 0 };
+	if (local_ip == NULL || ctx->dev[0] == '\0' || mac_empty(ctx->target_mac)) {
+		char auto_dev[IFNAMSIZ];
+		int ret = ip_route_get(&ctx->target_ip_ss,
+		                       &via,
+		                       &ctx->local_ip_ss,
+		                       (ctx->dev[0] == '\0') ? ctx->dev : auto_dev);
+		if (ret < 0) {
+			ERR2("can't find route to '%s' (%s)", target_str, strerror(-ret));
+			return false;
+		}
+	}
+
+	ctx->local_ip_range = addr_bits(ctx->ipv6); // by default use one IP
+	if (local_ip != NULL) {
+		at = strrchr(local_ip, '/');
+		if (at != NULL && (val = atoi(at + 1)) > 0 && val <= ctx->local_ip_range) {
+			ctx->local_ip_range = val;
+			*at = '\0';
+		}
+		if (ctx->ipv6) {
+			if (ctx->local_ip_range < 64 ||
+			    inet_pton(AF_INET6, local_ip, &ctx->local_ip.sin6_addr) <= 0) {
+				ERR2("invalid local IPv6 or unsupported prefix length");
+				return false;
+			}
+		} else {
+			if (inet_pton(AF_INET, local_ip, &ctx->local_ip4.sin_addr) <= 0) {
+				ERR2("invalid local IPv4");
+				return false;
+			}
+		}
+	}
+
+	if (mac_empty(ctx->target_mac)) {
+		const struct sockaddr_storage *neigh = (via.ss_family == AF_UNSPEC) ?
+		                                       &ctx->target_ip_ss : &via;
+		int ret = ip_neigh_get(neigh, true, ctx->target_mac);
+		if (ret < 0) {
+			char neigh_str[256] = { 0 };
+			sockaddr_tostr(neigh_str, sizeof(neigh_str), (struct sockaddr_storage *)neigh);
+			ERR2("failed to get remote MAC of target/gateway '%s' (%s)",
+			     neigh_str, strerror(-ret));
+			return false;
+		}
+	}
+
+	if (mac_empty(ctx->local_mac)) {
+		int ret = dev2mac(ctx->dev, ctx->local_mac);
+		if (ret < 0) {
+			ERR2("failed to get MAC of device '%s' (%s)", ctx->dev, strerror(-ret));
+			return false;
+		}
+	}
+
+	int ret = knot_eth_queues(ctx->dev);
+	if (ret >= 0) {
+		ctx->n_threads = ret;
+	} else {
+		ERR2("unable to get number of queues for '%s' (%s)", ctx->dev,
+		     knot_strerror(ret));
+		return false;
+	}
+
+	if (ctx->n_threads > 1 && ctx->quic) {
+		ret = knot_eth_rss(ctx->dev, &ctx->rss_conf);
+		if (ret != 0) {
+			WARN2("unable to read NIC RSS configuration for '%s' (%s)",
+			      ctx->dev, knot_strerror(ret));
+		}
+	}
+
+	return true;
+}
+
+static void print_help(void)
+{
+	printf("Usage: %s [parameters] -i <queries_file> <dest_ip>\n"
+	       "\n"
+	       "Parameters:\n"
+	       " -t, --duration <sec>     "SPACE"Duration of traffic generation.\n"
+	       "                          "SPACE" (default is %"PRIu64" seconds)\n"
+	       " -T, --tcp[=debug_mode]   "SPACE"Send queries over TCP.\n"
+	       " -U, --quic[=debug_mode]  "SPACE"Send queries over QUIC.\n"
+	       " -Q, --qps <qps>          "SPACE"Number of queries-per-second (approximately) to be sent.\n"
+	       "                          "SPACE" (default is %"PRIu64" qps)\n"
+	       " -b, --batch <size>       "SPACE"Send queries in a batch of defined size.\n"
+	       "                          "SPACE" (default is %d for UDP, %d for TCP)\n"
+	       " -r, --drop               "SPACE"Drop incoming responses (disables response statistics).\n"
+	       " -p, --port <port>        "SPACE"Remote destination port.\n"
+	       "                          "SPACE" (default is %d for UDP/TCP, %u for QUIC)\n"
+	       " -F, --affinity <spec>    "SPACE"CPU affinity in the format [<cpu_start>][s<cpu_step>].\n"
+	       "                          "SPACE" (default is %s)\n"
+	       " -i, --infile <file>      "SPACE"Path to a file with query templates.\n"
+	       " -I, --interface <ifname> "SPACE"Override auto-detected interface for outgoing communication.\n"
+	       " -l, --local <ip[/prefix]>"SPACE"Override auto-detected source IP address or subnet.\n"
+	       " -L, --local-mac <MAC>    "SPACE"Override auto-detected local MAC address.\n"
+	       " -R, --remote-mac <MAC>   "SPACE"Override auto-detected remote MAC address.\n"
+	       " -v, --vlan <id>          "SPACE"Add VLAN 802.1Q header with the given id.\n"
+	       " -h, --help               "SPACE"Print the program help.\n"
+	       " -V, --version            "SPACE"Print the program version.\n"
+	       "\n"
+	       "Arguments:\n"
+	       " <dest_ip>                "SPACE"IPv4 or IPv6 address of the remote destination.\n",
+	       PROGRAM_NAME, ctx_defaults.duration / 1000000, ctx_defaults.qps,
+	       ctx_defaults.at_once, 1, REMOTE_PORT_DEFAULT, REMOTE_PORT_DOQ_DEFAULT, "0s1");
+}
+
+static bool sending_mode(const char *arg, xdp_gun_ctx_t *ctx)
+{
+	if (arg == NULL) {
+		ctx->sending_mode = "";
+		return true;
+	} else if (strlen(arg) != 1) {
+		goto mode_invalid;
+	}
+	ctx->sending_mode = arg;
+
+	switch (ctx->sending_mode[0]) {
+	case '0':
+		if (!ctx->quic) {
+			goto mode_unavailable;
+		}
+		ctx->quic_full_handshake = true;
+		break;
+	case '1':
+		ctx->ignore1 = KXDPGUN_IGNORE_QUERY;
+		ctx->ignore2 = XDP_TCP_IGNORE_ESTABLISH | XDP_TCP_IGNORE_FIN;
+		break;
+	case '2':
+		ctx->ignore1 = KXDPGUN_IGNORE_QUERY;
+		break;
+	case '3':
+		ctx->ignore1 = KXDPGUN_IGNORE_QUERY;
+		ctx->ignore2 = XDP_TCP_IGNORE_FIN;
+		break;
+	case '5':
+		ctx->ignore1 = KXDPGUN_IGNORE_LASTBYTE;
+		ctx->ignore2 = XDP_TCP_IGNORE_FIN;
+		break;
+	case '7':
+		ctx->ignore1 = KXDPGUN_IGNORE_CLOSE;
+		ctx->ignore2 = XDP_TCP_IGNORE_DATA_ACK | XDP_TCP_IGNORE_FIN;
+		break;
+	case '8':
+		if (!ctx->tcp) {
+			goto mode_unavailable;
+		}
+		ctx->ignore1 = KXDPGUN_IGNORE_CLOSE;
+		ctx->ignore2 = XDP_TCP_IGNORE_FIN;
+		break;
+	case '9':
+		if (!ctx->tcp) {
+			goto mode_unavailable;
+		}
+		ctx->ignore2 = XDP_TCP_IGNORE_FIN;
+		break;
+	default:
+		goto mode_invalid;
+	}
+
+	return true;
+mode_unavailable:
+	ERR2("mode '%s' not available", optarg);
+	return false;
+mode_invalid:
+	ERR2("invalid mode '%s'", optarg);
+	return false;
+}
+
+static bool get_opts(int argc, char *argv[], xdp_gun_ctx_t *ctx)
+{
+	struct option opts[] = {
+		{ "help",       no_argument,       NULL, 'h' },
+		{ "version",    no_argument,       NULL, 'V' },
+		{ "duration",   required_argument, NULL, 't' },
+		{ "qps",        required_argument, NULL, 'Q' },
+		{ "batch",      required_argument, NULL, 'b' },
+		{ "drop",       no_argument,       NULL, 'r' },
+		{ "port",       required_argument, NULL, 'p' },
+		{ "tcp",        optional_argument, NULL, 'T' },
+		{ "quic",       optional_argument, NULL, 'U' },
+		{ "affinity",   required_argument, NULL, 'F' },
+		{ "interface",  required_argument, NULL, 'I' },
+		{ "local",      required_argument, NULL, 'l' },
+		{ "infile",     required_argument, NULL, 'i' },
+		{ "local-mac",  required_argument, NULL, 'L' },
+		{ "remote-mac", required_argument, NULL, 'R' },
+		{ "vlan",       required_argument, NULL, 'v' },
+		{ NULL }
+	};
+
+	int opt = 0, arg;
+	bool default_at_once = true;
+	double argf;
+	char *argcp, *local_ip = NULL;
+	while ((opt = getopt_long(argc, argv, "hVt:Q:b:rp:T::U::F:I:l:i:L:R:v:", opts, NULL)) != -1) {
+		switch (opt) {
+		case 'h':
+			print_help();
+			exit(EXIT_SUCCESS);
+		case 'V':
+			print_version(PROGRAM_NAME);
+			exit(EXIT_SUCCESS);
+		case 't':
+			assert(optarg);
+			argf = atof(optarg);
+			if (argf > 0) {
+				ctx->duration = argf * 1000000.0;
+				assert(ctx->duration >= 1000);
+			} else {
+				ERR2("invalid duration '%s'", optarg);
+				return false;
+			}
+			break;
+		case 'Q':
+			assert(optarg);
+			arg = atoi(optarg);
+			if (arg > 0) {
+				ctx->qps = arg;
+			} else {
+				ERR2("invalid QPS '%s'", optarg);
+				return false;
+			}
+			break;
+		case 'b':
+			assert(optarg);
+			arg = atoi(optarg);
+			if (arg > 0) {
+				default_at_once = false;
+				ctx->at_once = arg;
+			} else {
+				ERR2("invalid batch size '%s'", optarg);
+				return false;
+			}
+			break;
+		case 'r':
+			ctx->flags &= ~KNOT_XDP_FILTER_PASS;
+			ctx->flags |= KNOT_XDP_FILTER_DROP;
+			break;
+		case 'p':
+			assert(optarg);
+			arg = atoi(optarg);
+			if (arg > 0 && arg <= 0xffff) {
+				ctx->target_port = arg;
+			} else {
+				ERR2("invalid port '%s'", optarg);
+				return false;
+			}
+			break;
+		case 'T':
+			ctx->tcp = true;
+			ctx->quic = false;
+			ctx->flags &= ~(KNOT_XDP_FILTER_UDP | KNOT_XDP_FILTER_QUIC);
+			ctx->flags |= KNOT_XDP_FILTER_TCP;
+			if (default_at_once) {
+				ctx->at_once = 1;
+			}
+			if (!sending_mode(optarg, ctx)) {
+				return false;
+			}
+			break;
+		case 'U':
+#ifdef ENABLE_QUIC
+			ctx->quic = true;
+			ctx->tcp = false;
+			ctx->flags &= ~(KNOT_XDP_FILTER_UDP | KNOT_XDP_FILTER_TCP);
+			ctx->flags |= KNOT_XDP_FILTER_QUIC;
+			if (ctx->target_port == 0) {
+				ctx->target_port = REMOTE_PORT_DOQ_DEFAULT;
+			}
+			if (default_at_once) {
+				ctx->at_once = 1;
+			}
+			if (!sending_mode(optarg, ctx)) {
+				return false;
+			}
+#else
+			ERR2("QUIC not available");
+			return false;
+#endif // ENABLE_QUIC
+			break;
+		case 'F':
+			assert(optarg);
+			if ((arg = atoi(optarg)) > 0) {
+				global_cpu_aff_start = arg;
+			}
+			argcp = strchr(optarg, 's');
+			if (argcp != NULL && (arg = atoi(argcp + 1)) > 0) {
+				global_cpu_aff_step = arg;
+			}
+			break;
+		case 'I':
+			strlcpy(ctx->dev, optarg, IFNAMSIZ);
+			break;
+		case 'l':
+			local_ip = optarg;
+			break;
+		case 'i':
+			if (!load_queries(optarg, ctx->edns_size, ctx->msgid)) {
+				return false;
+			}
+			break;
+		case 'L':
+			if (mac_sscan(optarg, ctx->local_mac) != KNOT_EOK) {
+				ERR2("invalid local MAC address '%s'", optarg);
+				return false;
+			}
+			break;
+		case 'R':
+			if (mac_sscan(optarg, ctx->target_mac) != KNOT_EOK) {
+				ERR2("invalid remote MAC address '%s'", optarg);
+				return false;
+			}
+			break;
+		case 'v':
+			assert(optarg);
+			arg = atoi(optarg);
+			if (arg > 0 && arg < 4095) {
+				uint16_t id = arg;
+				ctx->vlan_tci = htobe16(id);
+			} else {
+				ERR2("invalid VLAN id '%s'", optarg);
+				return false;
+			}
+			break;
+		default:
+			print_help();
+			return false;
+		}
+	}
+	if (global_payloads == NULL || argc - optind != 1) {
+		print_help();
+		return false;
+	}
+
+	if (ctx->target_port == 0) {
+		ctx->target_port = REMOTE_PORT_DEFAULT;
+	}
+
+	if (!configure_target(argv[optind], local_ip, ctx)) {
+		return false;
+	}
+
+	if (ctx->qps < ctx->n_threads) {
+		WARN2("QPS increased to the number of threads/queues: %u", ctx->n_threads);
+		ctx->qps = ctx->n_threads;
+	}
+	ctx->qps /= ctx->n_threads;
+
+	return true;
+}
+
+int main(int argc, char *argv[])
+{
+	xdp_gun_ctx_t ctx = ctx_defaults, *thread_ctxs = NULL;
+	ctx.msgid = time(NULL) % UINT16_MAX;
+	pthread_t *threads = NULL;
+
+	if (!get_opts(argc, argv, &ctx)) {
+		free_global_payloads();
+		return EXIT_FAILURE;
+	}
+
+	thread_ctxs = calloc(ctx.n_threads, sizeof(*thread_ctxs));
+	threads = calloc(ctx.n_threads, sizeof(*threads));
+	if (thread_ctxs == NULL || threads == NULL) {
+		ERR2("out of memory");
+		free(thread_ctxs);
+		free(threads);
+		free_global_payloads();
+		return EXIT_FAILURE;
+	}
+	for (int i = 0; i < ctx.n_threads; i++) {
+		thread_ctxs[i] = ctx;
+		thread_ctxs[i].thread_id = i;
+	}
+
+	if (!linux_at_least(5, 11)) {
+		struct rlimit min_limit = { RLIM_INFINITY, RLIM_INFINITY }, cur_limit = { 0 };
+		if (getrlimit(RLIMIT_MEMLOCK, &cur_limit) != 0 ||
+		    cur_limit.rlim_cur != min_limit.rlim_cur ||
+		    cur_limit.rlim_max != min_limit.rlim_max) {
+			int ret = setrlimit(RLIMIT_MEMLOCK, &min_limit);
+			if (ret != 0) {
+				WARN2("unable to increase RLIMIT_MEMLOCK: %s",
+				      strerror(errno));
+			}
+		}
+	}
+
+	pthread_mutex_init(&global_stats.mutex, NULL);
+
+	struct sigaction stop_action = { .sa_handler = sigterm_handler };
+	struct sigaction stats_action = { .sa_handler = sigusr_handler };
+	sigaction(SIGINT,  &stop_action, NULL);
+	sigaction(SIGTERM, &stop_action, NULL);
+	sigaction(SIGUSR1, &stats_action, NULL);
+
+	for (size_t i = 0; i < ctx.n_threads; i++) {
+		unsigned affinity = global_cpu_aff_start + i * global_cpu_aff_step;
+		cpu_set_t set;
+		CPU_ZERO(&set);
+		CPU_SET(affinity, &set);
+		(void)pthread_create(&threads[i], NULL, xdp_gun_thread, &thread_ctxs[i]);
+		int ret = pthread_setaffinity_np(threads[i], sizeof(cpu_set_t), &set);
+		if (ret != 0) {
+			WARN2("failed to set affinity of thread#%zu to CPU#%u", i, affinity);
+		}
+		usleep(20000);
+	}
+	usleep(1000000);
+
+	xdp_trigger = KXDPGUN_START;
+	usleep(1000000);
+
+	for (size_t i = 0; i < ctx.n_threads; i++) {
+		pthread_join(threads[i], NULL);
+	}
+	if (global_stats.duration > 0 && global_stats.qry_sent > 0) {
+		print_stats(&global_stats, ctx.tcp, ctx.quic, !(ctx.flags & KNOT_XDP_FILTER_DROP));
+	}
+	pthread_mutex_destroy(&global_stats.mutex);
+
+	free(ctx.rss_conf);
+	free(thread_ctxs);
+	free(threads);
+	free_global_payloads();
+
+	return EXIT_SUCCESS;
+}