1#!/usr/bin/env python 2# @lint-avoid-python-3-compatibility-imports 3# 4# btrfsdist Summarize btrfs operation latency. 5# For Linux, uses BCC, eBPF. 6# 7# USAGE: btrfsdist [-h] [-T] [-m] [-p PID] [interval] [count] 8# 9# Copyright 2016 Netflix, Inc. 10# Licensed under the Apache License, Version 2.0 (the "License") 11# 12# 15-Feb-2016 Brendan Gregg Created this. 13 14from __future__ import print_function 15from bcc import BPF 16from time import sleep, strftime 17import argparse 18 19# symbols 20kallsyms = "/proc/kallsyms" 21 22# arguments 23examples = """examples: 24 ./btrfsdist # show operation latency as a histogram 25 ./btrfsdist -p 181 # trace PID 181 only 26 ./btrfsdist 1 10 # print 1 second summaries, 10 times 27 ./btrfsdist -m 5 # 5s summaries, milliseconds 28""" 29parser = argparse.ArgumentParser( 30 description="Summarize btrfs operation latency", 31 formatter_class=argparse.RawDescriptionHelpFormatter, 32 epilog=examples) 33parser.add_argument("-T", "--notimestamp", action="store_true", 34 help="don't include timestamp on interval output") 35parser.add_argument("-m", "--milliseconds", action="store_true", 36 help="output in milliseconds") 37parser.add_argument("-p", "--pid", 38 help="trace this PID only") 39parser.add_argument("interval", nargs="?", 40 help="output interval, in seconds") 41parser.add_argument("count", nargs="?", default=99999999, 42 help="number of outputs") 43parser.add_argument("--ebpf", action="store_true", 44 help=argparse.SUPPRESS) 45args = parser.parse_args() 46pid = args.pid 47countdown = int(args.count) 48if args.milliseconds: 49 factor = 1000000 50 label = "msecs" 51else: 52 factor = 1000 53 label = "usecs" 54if args.interval and int(args.interval) == 0: 55 print("ERROR: interval 0. Exiting.") 56 exit() 57debug = 0 58 59# define BPF program 60bpf_text = """ 61#include <uapi/linux/ptrace.h> 62#include <linux/fs.h> 63#include <linux/sched.h> 64 65#define OP_NAME_LEN 8 66typedef struct dist_key { 67 char op[OP_NAME_LEN]; 68 u64 slot; 69} dist_key_t; 70BPF_HASH(start, u32); 71BPF_HISTOGRAM(dist, dist_key_t); 72 73// time operation 74int trace_entry(struct pt_regs *ctx) 75{ 76 u32 pid = bpf_get_current_pid_tgid(); 77 if (FILTER_PID) 78 return 0; 79 u64 ts = bpf_ktime_get_ns(); 80 start.update(&pid, &ts); 81 return 0; 82} 83 84// The current btrfs (Linux 4.5) uses generic_file_read_iter() instead of it's 85// own read function. So we need to trace that and then filter on btrfs, which 86// I do by checking file->f_op. 87int trace_read_entry(struct pt_regs *ctx, struct kiocb *iocb) 88{ 89 u32 pid = bpf_get_current_pid_tgid(); 90 if (FILTER_PID) 91 return 0; 92 93 // btrfs filter on file->f_op == btrfs_file_operations 94 struct file *fp = iocb->ki_filp; 95 if ((u64)fp->f_op != BTRFS_FILE_OPERATIONS) 96 return 0; 97 98 u64 ts = bpf_ktime_get_ns(); 99 start.update(&pid, &ts); 100 return 0; 101} 102 103// The current btrfs (Linux 4.5) uses generic_file_open(), instead of it's own 104// function. Same as with reads. Trace the generic path and filter: 105int trace_open_entry(struct pt_regs *ctx, struct inode *inode, 106 struct file *file) 107{ 108 u32 pid; 109 pid = bpf_get_current_pid_tgid(); 110 if (FILTER_PID) 111 return 0; 112 113 // btrfs filter on file->f_op == btrfs_file_operations 114 if ((u64)file->f_op != BTRFS_FILE_OPERATIONS) 115 return 0; 116 117 u64 ts = bpf_ktime_get_ns(); 118 start.update(&pid, &ts); 119 return 0; 120} 121 122static int trace_return(struct pt_regs *ctx, const char *op) 123{ 124 u64 *tsp; 125 u32 pid = bpf_get_current_pid_tgid(); 126 127 // fetch timestamp and calculate delta 128 tsp = start.lookup(&pid); 129 if (tsp == 0) { 130 return 0; // missed start or filtered 131 } 132 u64 delta = (bpf_ktime_get_ns() - *tsp) / FACTOR; 133 134 // store as histogram 135 dist_key_t key = {.slot = bpf_log2l(delta)}; 136 __builtin_memcpy(&key.op, op, sizeof(key.op)); 137 dist.increment(key); 138 139 start.delete(&pid); 140 return 0; 141} 142 143int trace_read_return(struct pt_regs *ctx) 144{ 145 char *op = "read"; 146 return trace_return(ctx, op); 147} 148 149int trace_write_return(struct pt_regs *ctx) 150{ 151 char *op = "write"; 152 return trace_return(ctx, op); 153} 154 155int trace_open_return(struct pt_regs *ctx) 156{ 157 char *op = "open"; 158 return trace_return(ctx, op); 159} 160 161int trace_fsync_return(struct pt_regs *ctx) 162{ 163 char *op = "fsync"; 164 return trace_return(ctx, op); 165} 166""" 167 168# code replacements 169with open(kallsyms) as syms: 170 ops = '' 171 for line in syms: 172 a = line.rstrip().split() 173 (addr, name) = (a[0], a[2]) 174 name = name.split("\t")[0] 175 if name == "btrfs_file_operations": 176 ops = "0x" + addr 177 break 178 if ops == '': 179 print("ERROR: no btrfs_file_operations in /proc/kallsyms. Exiting.") 180 print("HINT: the kernel should be built with CONFIG_KALLSYMS_ALL.") 181 exit() 182 bpf_text = bpf_text.replace('BTRFS_FILE_OPERATIONS', ops) 183bpf_text = bpf_text.replace('FACTOR', str(factor)) 184if args.pid: 185 bpf_text = bpf_text.replace('FILTER_PID', 'pid != %s' % pid) 186else: 187 bpf_text = bpf_text.replace('FILTER_PID', '0') 188if debug or args.ebpf: 189 print(bpf_text) 190 if args.ebpf: 191 exit() 192 193# load BPF program 194b = BPF(text=bpf_text) 195 196# Common file functions. See earlier comment about generic_file_read_iter(). 197b.attach_kprobe(event="generic_file_read_iter", fn_name="trace_read_entry") 198b.attach_kprobe(event="btrfs_file_write_iter", fn_name="trace_entry") 199b.attach_kprobe(event="generic_file_open", fn_name="trace_open_entry") 200b.attach_kprobe(event="btrfs_sync_file", fn_name="trace_entry") 201b.attach_kretprobe(event="generic_file_read_iter", fn_name="trace_read_return") 202b.attach_kretprobe(event="btrfs_file_write_iter", fn_name="trace_write_return") 203b.attach_kretprobe(event="generic_file_open", fn_name="trace_open_return") 204b.attach_kretprobe(event="btrfs_sync_file", fn_name="trace_fsync_return") 205 206print("Tracing btrfs operation latency... Hit Ctrl-C to end.") 207 208# output 209exiting = 0 210dist = b.get_table("dist") 211while (1): 212 try: 213 if args.interval: 214 sleep(int(args.interval)) 215 else: 216 sleep(99999999) 217 except KeyboardInterrupt: 218 exiting = 1 219 220 print() 221 if args.interval and (not args.notimestamp): 222 print(strftime("%H:%M:%S:")) 223 224 dist.print_log2_hist(label, "operation") 225 dist.clear() 226 227 countdown -= 1 228 if exiting or countdown == 0: 229 exit() 230