lwt: fix return values of BPF ops

index 8b6b5e72b217..4a0797f0a154 100644
@@ -60,9 +60,8 @@ static int run_lwt_bpf(struct sk_buff *skb, struct bpf_lwt_prog *lwt,
 			ret = BPF_OK;
 		} else {
 			skb_reset_mac_header(skb);
-			ret = skb_do_redirect(skb);
-			if (ret == 0)
-				ret = BPF_REDIRECT;
+			skb_do_redirect(skb);
+			ret = BPF_REDIRECT;
 		}
 		break;
 

@@ -255,7 +254,7 @@ static int bpf_lwt_xmit_reroute(struct sk_buff *skb)
 
 	err = dst_output(dev_net(skb_dst(skb)->dev), skb->sk, skb);
 	if (unlikely(err))
-		return err;
+		return net_xmit_errno(err);
 
 	/* ip[6]_finish_output2 understand LWTUNNEL_XMIT_DONE */
 	return LWTUNNEL_XMIT_DONE;

index 6f15e6fa154e..53bd2d02a4f0 100644
@@ -16,9 +16,12 @@
 #define LWTUNNEL_STATE_INPUT_REDIRECT	BIT(1)
 #define LWTUNNEL_STATE_XMIT_REDIRECT	BIT(2)
 
+/* LWTUNNEL_XMIT_CONTINUE should be distinguishable from dst_output return
+ * values (NET_XMIT_xxx and NETDEV_TX_xxx in linux/netdevice.h) for safety.
+ */
 enum {
 	LWTUNNEL_XMIT_DONE,
-	LWTUNNEL_XMIT_CONTINUE,
+	LWTUNNEL_XMIT_CONTINUE = 0x100,
 };
 
 

index 6ba1a0fafbaa..a6e4c82615d7 100644
@@ -216,7 +216,7 @@ static int ip_finish_output2(struct net *net, struct sock *sk, struct sk_buff *s
 	if (lwtunnel_xmit_redirect(dst->lwtstate)) {
 		int res = lwtunnel_xmit(skb);
 
-		if (res < 0 || res == LWTUNNEL_XMIT_DONE)
+		if (res != LWTUNNEL_XMIT_CONTINUE)
 			return res;
 	}
 

index 1e8c90e97608..016b0a513259 100644
@@ -113,7 +113,7 @@ static int ip6_finish_output2(struct net *net, struct sock *sk, struct sk_buff *
 	if (lwtunnel_xmit_redirect(dst->lwtstate)) {
 		int res = lwtunnel_xmit(skb);
 
-		if (res < 0 || res == LWTUNNEL_XMIT_DONE)
+		if (res != LWTUNNEL_XMIT_CONTINUE)
 			return res;
 	}
 

new file mode 100644
index 000000000000..61333f2a03f9
@@ -0,0 +1,139 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+
+#ifndef __LWT_HELPERS_H
+#define __LWT_HELPERS_H
+
+#include <time.h>
+#include <net/if.h>
+#include <linux/if_tun.h>
+#include <linux/icmp.h>
+
+#include "test_progs.h"
+
+#define log_err(MSG, ...) \
+	fprintf(stderr, "(%s:%d: errno: %s) " MSG "\n", \
+		__FILE__, __LINE__, strerror(errno), ##__VA_ARGS__)
+
+#define RUN_TEST(name)                                                        \
+	({                                                                    \
+		if (test__start_subtest(#name))                               \
+			if (ASSERT_OK(netns_create(), "netns_create")) {      \
+				struct nstoken *token = open_netns(NETNS);    \
+				if (ASSERT_OK_PTR(token, "setns")) {          \
+					test_ ## name();                      \
+					close_netns(token);                   \
+				}                                             \
+				netns_delete();                               \
+			}                                                     \
+	})
+
+#define NETNS "ns_lwt"
+
+static inline int netns_create(void)
+{
+	return system("ip netns add " NETNS);
+}
+
+static inline int netns_delete(void)
+{
+	return system("ip netns del " NETNS ">/dev/null 2>&1");
+}
+
+static int open_tuntap(const char *dev_name, bool need_mac)
+{
+	int err = 0;
+	struct ifreq ifr;
+	int fd = open("/dev/net/tun", O_RDWR);
+
+	if (!ASSERT_GT(fd, 0, "open(/dev/net/tun)"))
+		return -1;
+
+	ifr.ifr_flags = IFF_NO_PI | (need_mac ? IFF_TAP : IFF_TUN);
+	memcpy(ifr.ifr_name, dev_name, IFNAMSIZ);
+
+	err = ioctl(fd, TUNSETIFF, &ifr);
+	if (!ASSERT_OK(err, "ioctl(TUNSETIFF)")) {
+		close(fd);
+		return -1;
+	}
+
+	err = fcntl(fd, F_SETFL, O_NONBLOCK);
+	if (!ASSERT_OK(err, "fcntl(O_NONBLOCK)")) {
+		close(fd);
+		return -1;
+	}
+
+	return fd;
+}
+
+#define ICMP_PAYLOAD_SIZE     100
+
+/* Match an ICMP packet with payload len ICMP_PAYLOAD_SIZE */
+static int __expect_icmp_ipv4(char *buf, ssize_t len)
+{
+	struct iphdr *ip = (struct iphdr *)buf;
+	struct icmphdr *icmp = (struct icmphdr *)(ip + 1);
+	ssize_t min_header_len = sizeof(*ip) + sizeof(*icmp);
+
+	if (len < min_header_len)
+		return -1;
+
+	if (ip->protocol != IPPROTO_ICMP)
+		return -1;
+
+	if (icmp->type != ICMP_ECHO)
+		return -1;
+
+	return len == ICMP_PAYLOAD_SIZE + min_header_len;
+}
+
+typedef int (*filter_t) (char *, ssize_t);
+
+/* wait_for_packet - wait for a packet that matches the filter
+ *
+ * @fd: tun fd/packet socket to read packet
+ * @filter: filter function, returning 1 if matches
+ * @timeout: timeout to wait for the packet
+ *
+ * Returns 1 if a matching packet is read, 0 if timeout expired, -1 on error.
+ */
+static int wait_for_packet(int fd, filter_t filter, struct timeval *timeout)
+{
+	char buf[4096];
+	int max_retry = 5; /* in case we read some spurious packets */
+	fd_set fds;
+
+	FD_ZERO(&fds);
+	while (max_retry--) {
+		/* Linux modifies timeout arg... So make a copy */
+		struct timeval copied_timeout = *timeout;
+		ssize_t ret = -1;
+
+		FD_SET(fd, &fds);
+
+		ret = select(1 + fd, &fds, NULL, NULL, &copied_timeout);
+		if (ret <= 0) {
+			if (errno == EINTR)
+				continue;
+			else if (errno == EAGAIN || ret == 0)
+				return 0;
+
+			log_err("select failed");
+			return -1;
+		}
+
+		ret = read(fd, buf, sizeof(buf));
+
+		if (ret <= 0) {
+			log_err("read(dev): %ld", ret);
+			return -1;
+		}
+
+		if (filter && filter(buf, ret) > 0)
+			return 1;
+	}
+
+	return 0;
+}
+
+#endif /* __LWT_HELPERS_H */

319 lines not on this page · raw message

58 lines not on this page · raw message

new file mode 100644
index 000000000000..e5fa5451bd2c
@@ -0,0 +1,256 @@
+// SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause
+
+/*
+ * Test suite of lwt BPF programs that reroutes packets
+ *   The file tests focus not only if these programs work as expected normally,
+ *   but also if they can handle abnormal situations gracefully. This test
+ *   suite currently only covers lwt_xmit hook. lwt_in tests have not been
+ *   implemented.
+ *
+ * WARNING
+ * -------
+ *  This test suite can crash the kernel, thus should be run in a VM.
+ *
+ * Setup:
+ * ---------
+ *  all tests are performed in a single netns. A lwt encap route is setup for
+ *  each subtest:
+ *
+ *    ip route add 10.0.0.0/24 encap bpf xmit <obj> sec "<section_N>" dev link_err
+ *
+ *  Here <obj> is statically defined to test_lwt_reroute.bpf.o, and it contains
+ *  a single test program entry. This program sets packet mark by last byte of
+ *  the IPv4 daddr. For example, a packet going to 1.2.3.4 will receive a skb
+ *  mark 4. A packet will only be marked once, and IP x.x.x.0 will be skipped
+ *  to avoid route loop. We didn't use generated BPF skeleton since the
+ *  attachment for lwt programs are not supported by libbpf yet.
+ *
+ *  The test program will bring up a tun device, and sets up the following
+ *  routes:
+ *
+ *    ip rule add pref 100 from all fwmark <tun_index> lookup 100
+ *    ip route add table 100 default dev tun0
+ *
+ *  For normal testing, a ping command is running in the test netns:
+ *
+ *    ping 10.0.0.<tun_index> -c 1 -w 1 -s 100
+ *
+ *  For abnormal testing, fq is used as the qdisc of the tun device. Then a UDP
+ *  socket will try to overflow the fq queue and trigger qdisc drop error.
+ *
+ * Scenarios:
+ * --------------------------------
+ *  1. Reroute to a running tun device
+ *  2. Reroute to a device where qdisc drop
+ *
+ *  For case 1, ping packets should be received by the tun device.
+ *
+ *  For case 2, force UDP packets to overflow fq limit. As long as kernel
+ *  is not crashed, it is considered successful.
+ */
+#include "lwt_helpers.h"
+#include "network_helpers.h"
+#include <linux/net_tstamp.h>
+
+#define BPF_OBJECT            "test_lwt_reroute.bpf.o"
+#define LOCAL_SRC             "10.0.0.1"
+#define TEST_CIDR             "10.0.0.0/24"
+#define XMIT_HOOK             "xmit"
+#define XMIT_SECTION          "lwt_xmit"
+#define NSEC_PER_SEC          1000000000ULL
+
+/* send a ping to be rerouted to the target device */
+static void ping_once(const char *ip)
+{
+	/* We won't get a reply. Don't fail here */
+	SYS_NOFAIL("ping %s -c1 -w1 -s %d >/dev/null 2>&1",
+		   ip, ICMP_PAYLOAD_SIZE);
+}
+
+/* Send snd_target UDP packets to overflow the fq queue and trigger qdisc drop
+ * error. This is done via TX tstamp to force buffering delayed packets.
+ */
+static int overflow_fq(int snd_target, const char *target_ip)
+{
+	struct sockaddr_in addr = {
+		.sin_family = AF_INET,
+		.sin_port = htons(1234),
+	};
+
+	char data_buf[8]; /* only #pkts matter, so use a random small buffer */
+	char control_buf[CMSG_SPACE(sizeof(uint64_t))];
+	struct iovec iov = {
+		.iov_base = data_buf,
+		.iov_len = sizeof(data_buf),
+	};
+	int err = -1;
+	int s = -1;
+	struct sock_txtime txtime_on = {
+		.clockid = CLOCK_MONOTONIC,
+		.flags = 0,
+	};
+	struct msghdr msg = {
+		.msg_name = &addr,
+		.msg_namelen = sizeof(addr),
+		.msg_control = control_buf,
+		.msg_controllen = sizeof(control_buf),
+		.msg_iovlen = 1,
+		.msg_iov = &iov,
+	};
+	struct cmsghdr *cmsg = CMSG_FIRSTHDR(&msg);
+
+	memset(data_buf, 0, sizeof(data_buf));
+
+	s = socket(AF_INET, SOCK_DGRAM, 0);
+	if (!ASSERT_GE(s, 0, "socket"))
+		goto out;
+
+	err = setsockopt(s, SOL_SOCKET, SO_TXTIME, &txtime_on, sizeof(txtime_on));
+	if (!ASSERT_OK(err, "setsockopt(SO_TXTIME)"))
+		goto out;
+
+	err = inet_pton(AF_INET, target_ip, &addr.sin_addr);
+	if (!ASSERT_EQ(err, 1, "inet_pton"))
+		goto out;
+
+	while (snd_target > 0) {
+		struct timespec now;
+
+		memset(control_buf, 0, sizeof(control_buf));
+		cmsg->cmsg_type = SCM_TXTIME;
+		cmsg->cmsg_level = SOL_SOCKET;
+		cmsg->cmsg_len = CMSG_LEN(sizeof(uint64_t));
+
+		err = clock_gettime(CLOCK_MONOTONIC, &now);
+		if (!ASSERT_OK(err, "clock_gettime(CLOCK_MONOTONIC)")) {
+			err = -1;
+			goto out;
+		}
+
+		*(uint64_t *)CMSG_DATA(cmsg) = (now.tv_nsec + 1) * NSEC_PER_SEC +
+					       now.tv_nsec;
+
+		/* we will intentionally send more than fq limit, so ignore
+		 * the error here.
+		 */
+		sendmsg(s, &msg, MSG_NOSIGNAL);
+		snd_target--;
+	}
+
+	/* no kernel crash so far is considered success */
+	err = 0;
+
+out:
+	if (s >= 0)
+		close(s);
+
+	return err;
+}
+
+static int setup(const char *tun_dev)
+{
+	int target_index = -1;
+	int tap_fd = -1;
+
+	tap_fd = open_tuntap(tun_dev, false);
+	if (!ASSERT_GE(tap_fd, 0, "open_tun"))
+		return -1;
+
+	target_index = if_nametoindex(tun_dev);
+	if (!ASSERT_GE(target_index, 0, "if_nametoindex"))
+		return -1;
+
+	SYS(fail, "ip link add link_err type dummy");
+	SYS(fail, "ip link set lo up");
+	SYS(fail, "ip addr add dev lo " LOCAL_SRC "/32");
+	SYS(fail, "ip link set link_err up");
+	SYS(fail, "ip link set %s up", tun_dev);
+
+	SYS(fail, "ip route add %s dev link_err encap bpf xmit obj %s sec lwt_xmit",
+	    TEST_CIDR, BPF_OBJECT);
+
+	SYS(fail, "ip rule add pref 100 from all fwmark %d lookup 100",
+	    target_index);
+	SYS(fail, "ip route add t 100 default dev %s", tun_dev);
+
+	return tap_fd;
+
+fail:
+	if (tap_fd >= 0)
+		close(tap_fd);
+	return -1;
+}
+
+static void test_lwt_reroute_normal_xmit(void)
+{
+	const char *tun_dev = "tun0";
+	int tun_fd = -1;
+	int ifindex = -1;
+	char ip[256];
+	struct timeval timeo = {
+		.tv_sec = 0,
+		.tv_usec = 250000,
+	};
+
+	tun_fd = setup(tun_dev);
+	if (!ASSERT_GE(tun_fd, 0, "setup_reroute"))
+		return;
+
+	ifindex = if_nametoindex(tun_dev);
+	snprintf(ip, 256, "10.0.0.%d", ifindex);
+
+	/* ping packets should be received by the tun device */
+	ping_once(ip);
+
+	if (!ASSERT_EQ(wait_for_packet(tun_fd, __expect_icmp_ipv4, &timeo), 1,
+		       "wait_for_packet"))
+		log_err("%s xmit", __func__);
+}
+
+/*
+ * Test the failure case when the skb is dropped at the qdisc. This is a
+ * regression prevention at the xmit hook only.
+ */
+static void test_lwt_reroute_qdisc_dropped(void)
+{
+	const char *tun_dev = "tun0";
+	int tun_fd = -1;
+	int ifindex = -1;
+	char ip[256];
+
+	tun_fd = setup(tun_dev);
+	if (!ASSERT_GE(tun_fd, 0, "setup_reroute"))
+		goto fail;
+
+	SYS(fail, "tc qdisc replace dev %s root fq limit 5 flow_limit 5", tun_dev);
+
+	ifindex = if_nametoindex(tun_dev);
+	snprintf(ip, 256, "10.0.0.%d", ifindex);
+	ASSERT_EQ(overflow_fq(10, ip), 0, "overflow_fq");
+
+fail:
+	if (tun_fd >= 0)
+		close(tun_fd);
+}
+
+static void *test_lwt_reroute_run(void *arg)
+{
+	netns_delete();
+	RUN_TEST(lwt_reroute_normal_xmit);
+	RUN_TEST(lwt_reroute_qdisc_dropped);
+	return NULL;
+}
+
+void test_lwt_reroute(void)
+{
+	pthread_t test_thread;
+	int err;
+
+	/* Run the tests in their own thread to isolate the namespace changes
+	 * so they do not affect the environment of other tests.
+	 * (specifically needed because of unshare(CLONE_NEWNS) in open_netns())
+	 */
+	err = pthread_create(&test_thread, NULL, &test_lwt_reroute_run, NULL);
+	if (ASSERT_OK(err, "pthread_create"))
+		ASSERT_OK(pthread_join(test_thread, NULL), "pthread_join");
+}

new file mode 100644
index 000000000000..1dc64351929c
@@ -0,0 +1,36 @@
+// SPDX-License-Identifier: GPL-2.0
+#include <inttypes.h>
+#include <linux/bpf.h>
+#include <bpf/bpf_endian.h>
+#include <bpf/bpf_helpers.h>
+#include <linux/if_ether.h>
+#include <linux/ip.h>
+
+/* This function extracts the last byte of the daddr, and uses it
+ * as output dev index.
+ */
+SEC("lwt_xmit")
+int test_lwt_reroute(struct __sk_buff *skb)
+{
+	struct iphdr *iph = NULL;
+	void *start = (void *)(long)skb->data;
+	void *end = (void *)(long)skb->data_end;
+
+	/* set mark at most once */
+	if (skb->mark != 0)
+		return BPF_OK;
+
+	if (start + sizeof(*iph) > end)
+		return BPF_DROP;
+
+	iph = (struct iphdr *)start;
+	skb->mark = bpf_ntohl(iph->daddr) & 0xff;
+
+	/* do not reroute x.x.x.0 packets */
+	if (skb->mark == 0)
+		return BPF_OK;
+
+	return BPF_LWT_REROUTE;
+}
+
+char _license[] SEC("license") = "GPL";