sync-ebpf.py

#!/usr/bin/python3
#
# profile *sync* syscalls.
#
# Based on existing programs that have these copyrights:
# Copyright (c) 2015 Brendan Gregg.
# Licensed under the Apache License, Version 2.0 (the "License")
#
# 13-Aug-2015   Brendan Gregg   Created this.
# 19-Feb-2016   Allan McAleavy migrated to BPF_PERF_OUTPUT

from __future__ import print_function
from bcc import BPF
import os

syscalls = {
    1: "fdatasync",
    2: "fsync",
    3: "msync",
    4: "sync",
    5: "syncfs",
    6: "sync_file_range",
    7: "sync_file_range2",
}

code = """
struct data_t {
    u32 syscall;
    u64 ts;
    u32 pid;
};

BPF_HASH(start, u32);

BPF_PERF_OUTPUT(events);

int trace_func_entry(struct pt_regs *ctx)
{
    u64 pid_tgid = bpf_get_current_pid_tgid();
    u32 pid = pid_tgid;
    u32 tgid = pid_tgid >> 32;
    u64 sc;
    u64 ts = bpf_ktime_get_ns();

    start.update(&pid, &ts);

    return 0;
}
"""

for k, v in syscalls.items():
    code = code + \
"""
int trace_func_return_SYSCALL(struct pt_regs *ctx)
{
    struct data_t data = {};
    u64 *tsp;
    u64 pid_tgid = bpf_get_current_pid_tgid();
    u32 pid = pid_tgid;
    u32 tgid = pid_tgid >> 32;

    // calculate delta time
    tsp = start.lookup(&pid);
    if (tsp == 0) {
        return 0;   // missed start
    }
    start.delete(&pid);
    data.ts = (bpf_ktime_get_ns() - *tsp) / 1000;

    data.syscall = SYSCALL;
    data.pid = pid;

    events.perf_submit(ctx, &data, sizeof(data));

    return 0;
}

""".replace("SYSCALL", str(k))

# load BPF program
b = BPF(text=code)

for k, v in syscalls.items():
    b.attach_kprobe(event=b.get_syscall_fnname(v),
                    fn_name="trace_func_entry")
    b.attach_kretprobe(event=b.get_syscall_fnname(v),
                       fn_name="trace_func_return_%d" % k)

# header
print("%-18s %-18s %-12s %-12s %s" % ("TIME(s)", "TOTAL SO FAR(s)", "CALL", "PID", "EXE"))

total_so_far = 0.0

# process event
def print_event(cpu, data, size):
    global total_so_far
    event = b["events"].event(data)
    delta = float(event.ts) / 1000000
    total_so_far = total_so_far + delta
    link = ""
    try:
        link = os.readlink("/proc/%d/exe" % event.pid)
    except:
        pass
    print("%-18.9f %-18.9f %-12s %-12s %s" % (delta, total_so_far, syscalls[event.syscall], event.pid, link))

# loop with callback to print_event
b["events"].open_perf_buffer(print_event)
while 1:
    try:
        b.perf_buffer_poll()
    except KeyboardInterrupt:
        exit()