selftests: net: fix/improve ip_defrag selftest

Commit ade446403bfb ("net: ipv4: do not handle duplicate fragments as
overlapping") changed IPv4 defragmentation so that duplicate fragments,
as well as _some_ fragments completely covered by previously delivered
fragments, do not lead to the whole frag queue being discarded. This
makes the existing ip_defrag selftest flaky.

This patch
* makes sure that negative IPv4 defrag tests generate truly overlapping
  fragments that trigger defrag queue drops;
* tests that duplicate IPv4 fragments do not trigger defrag queue drops;
* makes a couple of minor tweaks to the test aimed at increasing its code
  coverage and reduce flakiness.

Signed-off-by: Peter Oskolkov <posk@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

2 files changed, 95 insertions, 10 deletions

diff --git a/tools/testing/selftests/net/ip_defrag.c b/tools/testing/selftests/net/ip_defrag.c
index 61ae2782388e..5d56cc0838f6 100644
--- a/tools/testing/selftests/net/ip_defrag.c
+++ b/tools/testing/selftests/net/ip_defrag.c
@@ -203,6 +203,7 @@ static void send_udp_frags(int fd_raw, struct sockaddr *addr,
 {
 	struct ip *iphdr = (struct ip *)ip_frame;
 	struct ip6_hdr *ip6hdr = (struct ip6_hdr *)ip_frame;
+	const bool ipv4 = !ipv6;
 	int res;
 	int offset;
 	int frag_len;
@@ -239,19 +240,53 @@ static void send_udp_frags(int fd_raw, struct sockaddr *addr,
 		iphdr->ip_sum = 0;
 	}
 
+	/* Occasionally test in-order fragments. */
+	if (!cfg_overlap && (rand() % 100 < 15)) {
+		offset = 0;
+		while (offset < (UDP_HLEN + payload_len)) {
+			send_fragment(fd_raw, addr, alen, offset, ipv6);
+			offset += max_frag_len;
+		}
+		return;
+	}
+
+	/* Occasionally test IPv4 "runs" (see net/ipv4/ip_fragment.c) */
+	if (ipv4 && !cfg_overlap && (rand() % 100 < 20) &&
+			(payload_len > 9 * max_frag_len)) {
+		offset = 6 * max_frag_len;
+		while (offset < (UDP_HLEN + payload_len)) {
+			send_fragment(fd_raw, addr, alen, offset, ipv6);
+			offset += max_frag_len;
+		}
+		offset = 3 * max_frag_len;
+		while (offset < 6 * max_frag_len) {
+			send_fragment(fd_raw, addr, alen, offset, ipv6);
+			offset += max_frag_len;
+		}
+		offset = 0;
+		while (offset < 3 * max_frag_len) {
+			send_fragment(fd_raw, addr, alen, offset, ipv6);
+			offset += max_frag_len;
+		}
+		return;
+	}
+
 	/* Odd fragments. */
 	offset = max_frag_len;
 	while (offset < (UDP_HLEN + payload_len)) {
 		send_fragment(fd_raw, addr, alen, offset, ipv6);
+		/* IPv4 ignores duplicates, so randomly send a duplicate. */
+		if (ipv4 && (1 == rand() % 100))
+			send_fragment(fd_raw, addr, alen, offset, ipv6);
 		offset += 2 * max_frag_len;
 	}
 
 	if (cfg_overlap) {
 		/* Send an extra random fragment. */
-		offset = rand() % (UDP_HLEN + payload_len - 1);
-		/* sendto() returns EINVAL if offset + frag_len is too small. */
 		if (ipv6) {
 			struct ip6_frag *fraghdr = (struct ip6_frag *)(ip_frame + IP6_HLEN);
+			/* sendto() returns EINVAL if offset + frag_len is too small. */
+			offset = rand() % (UDP_HLEN + payload_len - 1);
 			frag_len = max_frag_len + rand() % 256;
 			/* In IPv6 if !!(frag_len % 8), the fragment is dropped. */
 			frag_len &= ~0x7;
@@ -259,13 +294,29 @@ static void send_udp_frags(int fd_raw, struct sockaddr *addr,
 			ip6hdr->ip6_plen = htons(frag_len);
 			frag_len += IP6_HLEN;
 		} else {
-			frag_len = IP4_HLEN + UDP_HLEN + rand() % 256;
+			/* In IPv4, duplicates and some fragments completely inside
+			 * previously sent fragments are dropped/ignored. So
+			 * random offset and frag_len can result in a dropped
+			 * fragment instead of a dropped queue/packet. So we
+			 * hard-code offset and frag_len.
+			 *
+			 * See ade446403bfb ("net: ipv4: do not handle duplicate
+			 * fragments as overlapping").
+			 */
+			if (max_frag_len * 4 < payload_len || max_frag_len < 16) {
+				/* not enough payload to play with random offset and frag_len. */
+				offset = 8;
+				frag_len = IP4_HLEN + UDP_HLEN + max_frag_len;
+			} else {
+				offset = rand() % (payload_len / 2);
+				frag_len = 2 * max_frag_len + 1 + rand() % 256;
+			}
 			iphdr->ip_off = htons(offset / 8 | IP4_MF);
 			iphdr->ip_len = htons(frag_len);
 		}
 		res = sendto(fd_raw, ip_frame, frag_len, 0, addr, alen);
 		if (res < 0)
-			error(1, errno, "sendto overlap");
+			error(1, errno, "sendto overlap: %d", frag_len);
 		if (res != frag_len)
 			error(1, 0, "sendto overlap: %d vs %d", (int)res, frag_len);
 		frag_counter++;
@@ -275,6 +326,9 @@ static void send_udp_frags(int fd_raw, struct sockaddr *addr,
 	offset = 0;
 	while (offset < (UDP_HLEN + payload_len)) {
 		send_fragment(fd_raw, addr, alen, offset, ipv6);
+		/* IPv4 ignores duplicates, so randomly send a duplicate. */
+		if (ipv4 && (1 == rand() % 100))
+			send_fragment(fd_raw, addr, alen, offset, ipv6);
 		offset += 2 * max_frag_len;
 	}
 }
@@ -282,7 +336,11 @@ static void send_udp_frags(int fd_raw, struct sockaddr *addr,
 static void run_test(struct sockaddr *addr, socklen_t alen, bool ipv6)
 {
 	int fd_tx_raw, fd_rx_udp;
-	struct timeval tv = { .tv_sec = 0, .tv_usec = 10 * 1000 };
+	/* Frag queue timeout is set to one second in the calling script;
+	 * socket timeout should be just a bit longer to avoid tests interfering
+	 * with each other.
+	 */
+	struct timeval tv = { .tv_sec = 1, .tv_usec = 10 };
 	int idx;
 	int min_frag_len = ipv6 ? 1280 : 8;
 
@@ -308,12 +366,32 @@ static void run_test(struct sockaddr *addr, socklen_t alen, bool ipv6)
 			payload_len += (rand() % 4096)) {
 		if (cfg_verbose)
 			printf("payload_len: %d\n", payload_len);
-		max_frag_len = min_frag_len;
-		do {
+
+		if (cfg_overlap) {
+			/* With overlaps, one send/receive pair below takes
+			 * at least one second (== timeout) to run, so there
+			 * is not enough test time to run a nested loop:
+			 * the full overlap test takes 20-30 seconds.
+			 */
+			max_frag_len = min_frag_len +
+				rand() % (1500 - FRAG_HLEN - min_frag_len);
 			send_udp_frags(fd_tx_raw, addr, alen, ipv6);
 			recv_validate_udp(fd_rx_udp);
-			max_frag_len += 8 * (rand() % 8);
-		} while (max_frag_len < (1500 - FRAG_HLEN) && max_frag_len <= payload_len);
+		} else {
+			/* Without overlaps, each packet reassembly (== one
+			 * send/receive pair below) takes very little time to
+			 * run, so we can easily afford more thourough testing
+			 * with a nested loop: the full non-overlap test takes
+			 * less than one second).
+			 */
+			max_frag_len = min_frag_len;
+			do {
+				send_udp_frags(fd_tx_raw, addr, alen, ipv6);
+				recv_validate_udp(fd_rx_udp);
+				max_frag_len += 8 * (rand() % 8);
+			} while (max_frag_len < (1500 - FRAG_HLEN) &&
+				 max_frag_len <= payload_len);
+		}
 	}
 
 	/* Cleanup. */
diff --git a/tools/testing/selftests/net/ip_defrag.sh b/tools/testing/selftests/net/ip_defrag.sh
index f34672796044..7dd79a9efb17 100755
--- a/tools/testing/selftests/net/ip_defrag.sh
+++ b/tools/testing/selftests/net/ip_defrag.sh
@@ -11,10 +11,17 @@ readonly NETNS="ns-$(mktemp -u XXXXXX)"
 setup() {
 	ip netns add "${NETNS}"
 	ip -netns "${NETNS}" link set lo up
+
 	ip netns exec "${NETNS}" sysctl -w net.ipv4.ipfrag_high_thresh=9000000 >/dev/null 2>&1
 	ip netns exec "${NETNS}" sysctl -w net.ipv4.ipfrag_low_thresh=7000000 >/dev/null 2>&1
+	ip netns exec "${NETNS}" sysctl -w net.ipv4.ipfrag_time=1 >/dev/null 2>&1
+
 	ip netns exec "${NETNS}" sysctl -w net.ipv6.ip6frag_high_thresh=9000000 >/dev/null 2>&1
 	ip netns exec "${NETNS}" sysctl -w net.ipv6.ip6frag_low_thresh=7000000 >/dev/null 2>&1
+	ip netns exec "${NETNS}" sysctl -w net.ipv6.ip6frag_time=1 >/dev/null 2>&1
+
+	# DST cache can get full with a lot of frags, with GC not keeping up with the test.
+	ip netns exec "${NETNS}" sysctl -w net.ipv6.route.max_size=65536 >/dev/null 2>&1
 }
 
 cleanup() {
@@ -27,7 +34,6 @@ setup
 echo "ipv4 defrag"
 ip netns exec "${NETNS}" ./ip_defrag -4
 
-
 echo "ipv4 defrag with overlaps"
 ip netns exec "${NETNS}" ./ip_defrag -4o
 
@@ -37,3 +43,4 @@ ip netns exec "${NETNS}" ./ip_defrag -6
 echo "ipv6 defrag with overlaps"
 ip netns exec "${NETNS}" ./ip_defrag -6o
 
+echo "all tests done"