Add netstacklat tool

headers/vmlinux/vmlinux_net.h

@@ -67,6 +67,7 @@ struct sk_buff {
 	__u8 nf_trace: 1;
 	__u8 ip_summed: 2;
 	__u8 ooo_okay: 1;
+	__u8 tstamp_type: 2;
 	__u8 l4_hash: 1;
 	__u8 sw_hash: 1;
 	__u8 wifi_acked_valid: 1;
@@ -145,4 +146,8 @@ enum ip_conntrack_status {
 	IPS_CONFIRMED = (1 << IPS_CONFIRMED_BIT),
 };
 
+struct scm_timestamping_internal {
+        struct timespec64 ts[3];
+};
+
 #endif /* __VMLINUX_NET_H__ */

headers/vmlinux/vmlinux_types.h

@@ -11,4 +11,11 @@ typedef __u64 u64;
 
 typedef s64 ktime_t;
 
+typedef __s64 time64_t;
+
+struct timespec64 {
+        time64_t tv_sec;
+        long int tv_nsec;
+};
+
 #endif /* __VMLINUX_TYPES_H__ */

netstacklat/.gitignore

@@ -0,0 +1 @@
+netstacklat

netstacklat/Makefile

@@ -0,0 +1,13 @@
+# SPDX-License-Identifier: (GPL-2.0 OR BSD-2-Clause)
+
+USER_TARGETS   := netstacklat
+BPF_TARGETS    := netstacklat.bpf
+BPF_SKEL_OBJ   := netstacklat.bpf.o
+
+EXTRA_DEPS += netstacklat.h bits.bpf.h
+LDLIBS     += -lm
+
+LIB_DIR = ../lib
+
+include $(LIB_DIR)/common.mk
+

netstacklat/README.md

@@ -0,0 +1,58 @@
+# Netstacklat - Monitor latency within the network stack
+Netstacklat is a simple tool for monitoring latency within the Linux
+network stack for ingress traffic. The tool relies on the kernel time
+stamping received packets (`SOF_TIMESTAMPING_RX_SOFTWARE`),
+specifically setting `sk_buff->tstamp`. It then reports when packets
+arrive at various hooks relative to this timestamp, i.e. the time
+between the packet being timestamped by the kernel and reaching a
+specific hook.
+
+The tool is based on the following bpftrace script from Jesper
+Dangaard Brouer:
+```console
+sudo bpftrace -e '
+	kfunc:tcp_v4_do_rcv,
+	kfunc:tcp_data_queue,
+	kfunc:udp_queue_rcv_one_skb
+	{
+		$tai_offset=37000000000;
+		$now=nsecs(tai)-$tai_offset; @cnt[probe]=count(); @total[probe]=count();
+		$ts=args->skb->tstamp; $delta=$now-(uint64)$ts;
+		@hist_ns[probe]=hist($delta);
+		@stats[probe]=stats($delta);
+		//printf("now:%llu - ts:%llu = delta:%llu\n", $now, $ts, $delta);
+	}
+	interval:s:10 {time("\n%H:%M:%S\n");
+		print(@cnt); clear(@cnt);
+		print(@total);
+		print(@stats);
+		print(@hist_ns);
+	}'
+```
+
+The eBPF part of the tool (`netstacklat.bpf.c`) is designed to be
+compatible with
+[ebpf_exporter](https://github.com/cloudflare/ebpf_exporter), so that
+the data can easily be exported to Prometheus. The easiest way to use
+netstacklat together with ebpf-exporter is simply to point it to this
+directory, i.e.
+```console
+$ ebpf_exporter --config.dir=<path>/bpf-examples/netstacklat --config.names=netstacklat
+```
+
+Alternatively, you can copy over the files to ebpf-exporter's example
+repository.
+```console
+$ cp netstacklat.{bpf.c,h,yaml} -t <path>/ebpf_exporter/examples/
+# Fix up some header includes (e.g. "vmlinux_local.h" -> <vmlinux.h>
+$ make -C <path>/ebpf_exporter/examples build
+$ ebpf_exporter --config.dir=<path>/ebpf_exporter/examples --config.names
+```
+
+Note that when using together with ebpf-exporter, some of the
+functionality handled by netstacklat's userspace program will not be
+available. This includes setting the `TAI_OFFSET` constant in
+`netstacklat.bpf.c` to match your system's TAI offset (you can do this
+manually instead), and enabling RX timestamping by the kernel (see the
+`enable_sw_rx_tstamps()` function in `netstacklat.c` for an example of
+how to do this).

netstacklat/bits.bpf.h

@@ -0,0 +1,29 @@
+/* SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause) */
+/* From https://github.com/iovisor/bcc/blob/v0.25.0/libbpf-tools/bits.bpf.h*/
+
+#ifndef __BITS_BPF_H
+#define __BITS_BPF_H
+
+static __always_inline u64 log2(u32 v)
+{
+    u32 shift, r;
+
+    r = (v > 0xFFFF) << 4; v >>= r;
+    shift = (v > 0xFF) << 3; v >>= shift; r |= shift;
+    shift = (v > 0xF) << 2; v >>= shift; r |= shift;
+    shift = (v > 0x3) << 1; v >>= shift; r |= shift;
+    r |= (v >> 1);
+
+    return r;
+}
+
+static __always_inline u64 log2l(u64 v)
+{
+    u32 hi = v >> 32;
+    if (hi)
+        return log2(hi) + 32;
+    else
+        return log2(v);
+}
+
+#endif /* __BITS_BPF_H */