1#!/usr/bin/python 2# @lint-avoid-python-3-compatibility-imports 3# 4# mdflush Trace md flush events. 5# For Linux, uses BCC, eBPF. 6# 7# Todo: add more details of the flush (latency, I/O count). 8# 9# Copyright 2016 Netflix, Inc. 10# Licensed under the Apache License, Version 2.0 (the "License") 11# 12# 13-Feb-2015 Brendan Gregg Created this. 13 14from __future__ import print_function 15from bcc import BPF 16from time import strftime 17import ctypes as ct 18 19# load BPF program 20b = BPF(text=""" 21#include <uapi/linux/ptrace.h> 22#include <linux/sched.h> 23#include <linux/genhd.h> 24#include <linux/bio.h> 25 26struct data_t { 27 u64 pid; 28 char comm[TASK_COMM_LEN]; 29 char disk[DISK_NAME_LEN]; 30}; 31BPF_PERF_OUTPUT(events); 32 33int kprobe__md_flush_request(struct pt_regs *ctx, void *mddev, struct bio *bio) 34{ 35 struct data_t data = {}; 36 u32 pid = bpf_get_current_pid_tgid(); 37 data.pid = pid; 38 bpf_get_current_comm(&data.comm, sizeof(data.comm)); 39/* 40 * The following deals with a kernel version change (in mainline 4.14, although 41 * it may be backported to earlier kernels) with how the disk name is accessed. 42 * We handle both pre- and post-change versions here. Please avoid kernel 43 * version tests like this as much as possible: they inflate the code, test, 44 * and maintenance burden. 45 */ 46#ifdef bio_dev 47 struct gendisk *bi_disk = bio->bi_disk; 48#else 49 struct gendisk *bi_disk = bio->bi_bdev->bd_disk; 50#endif 51 bpf_probe_read(&data.disk, sizeof(data.disk), bi_disk->disk_name); 52 events.perf_submit(ctx, &data, sizeof(data)); 53 return 0; 54} 55""") 56 57# event data 58TASK_COMM_LEN = 16 # linux/sched.h 59DISK_NAME_LEN = 32 # linux/genhd.h 60class Data(ct.Structure): 61 _fields_ = [ 62 ("pid", ct.c_ulonglong), 63 ("comm", ct.c_char * TASK_COMM_LEN), 64 ("disk", ct.c_char * DISK_NAME_LEN) 65 ] 66 67# header 68print("Tracing md flush requests... Hit Ctrl-C to end.") 69print("%-8s %-6s %-16s %s" % ("TIME", "PID", "COMM", "DEVICE")) 70 71# process event 72def print_event(cpu, data, size): 73 event = ct.cast(data, ct.POINTER(Data)).contents 74 print("%-8s %-6d %-16s %s" % (strftime("%H:%M:%S"), event.pid, 75 event.comm.decode('utf-8', 'replace'), 76 event.disk.decode('utf-8', 'replace'))) 77 78# read events 79b["events"].open_perf_buffer(print_event) 80while 1: 81 b.perf_buffer_poll() 82