Linux跟踪技术之Ebpf

#!/usr/bin/python
# @lint-avoid-python-3-compatibility-imports
#
# execsnoop Trace new processes via exec() syscalls.
#           For Linux, uses BCC, eBPF. Embedded C.
#
# USAGE: execsnoop [-h] [-T] [-t] [-x] [-q] [-n NAME] [-l LINE]
#                  [--max-args MAX_ARGS]
#
# This currently will print up to a maximum of 19 arguments, plus the process
# name, so 20 fields in total (MAXARG).
#
# This won't catch all new processes: an application may fork() but not exec().
#
# Copyright 2016 Netflix, Inc.
# Licensed under the Apache License, Version 2.0 (the "License")
#
# 07-Feb-2016   Brendan Gregg   Created this.

from __future__ import print_function
from bcc import BPF
from bcc.containers import filter_by_containers
from bcc.utils import ArgString, printb
import bcc.utils as utils
import argparse
import re
import time
import pwd
from collections import defaultdict
from time import strftime


def parse_uid(user):
try:
result = int(user)
except ValueError:
try:
user_info = pwd.getpwnam(user)
except KeyError:
raise argparse.ArgumentTypeError(
"{0!r} is not valid UID or user entry".format(user))
else:
return user_info.pw_uid
else:
# Maybe validate if UID < 0 ?
return result


# arguments
examples = """examples:
./execsnoop           # trace all exec() syscalls
./execsnoop -x        # include failed exec()s
./execsnoop -T        # include time (HH:MM:SS)
./execsnoop -U        # include UID
./execsnoop -u 1000   # only trace UID 1000
./execsnoop -u user   # get user UID and trace only them
./execsnoop -t        # include timestamps
./execsnoop -q        # add "quotemarks" around arguments
./execsnoop -n main   # only print command lines containing "main"
./execsnoop -l tpkg   # only print command where arguments contains "tpkg"
./execsnoop --cgroupmap mappath  # only trace cgroups in this BPF map
./execsnoop --mntnsmap mappath   # only trace mount namespaces in the map
"""
parser = argparse.ArgumentParser(
description="Trace exec() syscalls",
formatter_class=argparse.RawDescriptionHelpFormatter,
epilog=examples)
parser.add_argument("-T", "--time", action="store_true",
help="include time column on output (HH:MM:SS)")
parser.add_argument("-t", "--timestamp", action="store_true",
help="include timestamp on output")
parser.add_argument("-x", "--fails", action="store_true",
help="include failed exec()s")
parser.add_argument("--cgroupmap",
help="trace cgroups in this BPF map only")
parser.add_argument("--mntnsmap",
help="trace mount namespaces in this BPF map only")
parser.add_argument("-u", "--uid", type=parse_uid, metavar='USER',
help="trace this UID only")
parser.add_argument("-q", "--quote", action="store_true",
help="Add quotemarks (\") around arguments."
)
parser.add_argument("-n", "--name",
type=ArgString,
help="only print commands matching this name (regex), any arg")
parser.add_argument("-l", "--line",
type=ArgString,
help="only print commands where arg contains this line (regex)")
parser.add_argument("-U", "--print-uid", action="store_true",
help="print UID column")
parser.add_argument("--max-args", default="20",
help="maximum number of arguments parsed and displayed, defaults to 20")
parser.add_argument("--ebpf", action="store_true",
help=argparse.SUPPRESS)
args = parser.parse_args()

# define BPF program
bpf_text = """
#include <uapi/linux/ptrace.h>
#include <linux/sched.h>
#include <linux/fs.h>

#define ARGSIZE  128

enum event_type {
EVENT_ARG,
EVENT_RET,
};

struct data_t {
u32 pid;  // PID as in the userspace term (i.e. task->tgid in kernel)
u32 ppid; // Parent PID as in the userspace term (i.e task->real_parent->tgid in kernel)
u32 uid;
char comm[TASK_COMM_LEN];
enum event_type type;
char argv[ARGSIZE];
int retval;
};

BPF_PERF_OUTPUT(events);

static int __submit_arg(struct pt_regs *ctx, void *ptr, struct data_t *data)
{
bpf_probe_read_user(data->argv, sizeof(data->argv), ptr);
events.perf_submit(ctx, data, sizeof(struct data_t));
return 1;
}

static int submit_arg(struct pt_regs *ctx, void *ptr, struct data_t *data)
{
const char *argp = NULL;
bpf_probe_read_user(&argp, sizeof(argp), ptr);
if (argp) {
return __submit_arg(ctx, (void *)(argp), data);
}
return 0;
}

int syscall__execve(struct pt_regs *ctx,
const char __user *filename,
const char __user *const __user *__argv,
const char __user *const __user *__envp)
{

u32 uid = bpf_get_current_uid_gid() & 0xffffffff;

UID_FILTER

if (container_should_be_filtered()) {
return 0;
}

// create data here and pass to submit_arg to save stack space (#555)
struct data_t data = {};
struct task_struct *task;

data.pid = bpf_get_current_pid_tgid() >> 32;

task = (struct task_struct *)bpf_get_current_task();
// Some kernels, like Ubuntu 4.13.0-generic, return 0
// as the real_parent->tgid.
// We use the get_ppid function as a fallback in those cases. (#1883)
data.ppid = task->real_parent->tgid;

bpf_get_current_comm(&data.comm, sizeof(data.comm));
data.type = EVENT_ARG;

__submit_arg(ctx, (void *)filename, &data);

// skip first arg, as we submitted filename
#pragma unroll
for (int i = 1; i < MAXARG; i++) {
if (submit_arg(ctx, (void *)&__argv[i], &data) == 0)
goto out;
}

// handle truncated argument list
char ellipsis[] = "...";
__submit_arg(ctx, (void *)ellipsis, &data);
out:
return 0;
}

int do_ret_sys_execve(struct pt_regs *ctx)
{
if (container_should_be_filtered()) {
return 0;
}

struct data_t data = {};
struct task_struct *task;

u32 uid = bpf_get_current_uid_gid() & 0xffffffff;
UID_FILTER

data.pid = bpf_get_current_pid_tgid() >> 32;
data.uid = uid;

task = (struct task_struct *)bpf_get_current_task();
// Some kernels, like Ubuntu 4.13.0-generic, return 0
// as the real_parent->tgid.
// We use the get_ppid function as a fallback in those cases. (#1883)
data.ppid = task->real_parent->tgid;

bpf_get_current_comm(&data.comm, sizeof(data.comm));
data.type = EVENT_RET;
data.retval = PT_REGS_RC(ctx);
events.perf_submit(ctx, &data, sizeof(data));

return 0;
}
"""

bpf_text = bpf_text.replace("MAXARG", args.max_args)

if args.uid:
bpf_text = bpf_text.replace('UID_FILTER',
'if (uid != %s) { return 0; }' % args.uid)
else:
bpf_text = bpf_text.replace('UID_FILTER', '')
bpf_text = filter_by_containers(args) + bpf_text
if args.ebpf:
print(bpf_text)
exit()

# initialize BPF
b = BPF(text=bpf_text)
execve_fnname = b.get_syscall_fnname("execve")
b.attach_kprobe(event=execve_fnname, fn_name="syscall__execve")
b.attach_kretprobe(event=execve_fnname, fn_name="do_ret_sys_execve")

# header
if args.time:
print("%-9s" % ("TIME"), end="")
if args.timestamp:
print("%-8s" % ("TIME(s)"), end="")
if args.print_uid:
print("%-6s" % ("UID"), end="")
print("%-16s %-7s %-7s %3s %s" % ("PCOMM", "PID", "PPID", "RET", "ARGS"))

class EventType(object):
EVENT_ARG = 0
EVENT_RET = 1

start_ts = time.time()
argv = defaultdict(list)

# This is best-effort PPID matching. Short-lived processes may exit
# before we get a chance to read the PPID.
# This is a fallback for when fetching the PPID from task->real_parent->tgip
# returns 0, which happens in some kernel versions.
def get_ppid(pid):
try:
with open("/proc/%d/status" % pid) as status:
for line in status:
if line.startswith("PPid:"):
return int(line.split()[1])
except IOError:
pass
return 0

# process event
def print_event(cpu, data, size):
event = b["events"].event(data)
skip = False

if event.type == EventType.EVENT_ARG:
argv[event.pid].append(event.argv)
elif event.type == EventType.EVENT_RET:
if event.retval != 0 and not args.fails:
skip = True
if args.name and not re.search(bytes(args.name), event.comm):
skip = True
if args.line and not re.search(bytes(args.line),
b' '.join(argv[event.pid])):
skip = True
if args.quote:
argv[event.pid] = [
b"\"" + arg.replace(b"\"", b"\\\"") + b"\""
for arg in argv[event.pid]
]

if not skip:
if args.time:
printb(b"%-9s" % strftime("%H:%M:%S").encode('ascii'), nl="")
if args.timestamp:
printb(b"%-8.3f" % (time.time() - start_ts), nl="")
if args.print_uid:
printb(b"%-6d" % event.uid, nl="")
ppid = event.ppid if event.ppid > 0 else get_ppid(event.pid)
ppid = b"%d" % ppid if ppid > 0 else b"?"
argv_text = b' '.join(argv[event.pid]).replace(b'\n', b'\\n')
printb(b"%-16s %-7d %-7s %3d %s" % (event.comm, event.pid,
ppid, event.retval, argv_text))
try:
del(argv[event.pid])
except Exception:
pass


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

#!/usr/bin/python
#
# mallocstacks  Trace malloc() calls in a process and print the full
#               stack trace for all callsites.
#               For Linux, uses BCC, eBPF. Embedded C.
#
# This script is a basic example of the new Linux 4.6+ BPF_STACK_TRACE
# table API.
#
# Copyright 2016 GitHub, Inc.
# Licensed under the Apache License, Version 2.0 (the "License")

from __future__ import print_function
from bcc import BPF
from bcc.utils import printb
from time import sleep
import sys

if len(sys.argv) < 2:
print("USAGE: mallocstacks PID [NUM_STACKS=1024]")
exit()
pid = int(sys.argv[1])
if len(sys.argv) == 3:
try:
assert int(sys.argv[2]) > 0, ""
except (ValueError, AssertionError) as e:
print("USAGE: mallocstacks PID [NUM_STACKS=1024]")
print("NUM_STACKS must be a non-zero, positive integer")
exit()
stacks = sys.argv[2]
else:
stacks = "1024"

# load BPF program
b = BPF(text="""
#include <uapi/linux/ptrace.h>

BPF_HASH(calls, int);
BPF_STACK_TRACE(stack_traces, """ + stacks + """);

int alloc_enter(struct pt_regs *ctx, size_t size) {
int key = stack_traces.get_stackid(ctx, BPF_F_USER_STACK);
if (key < 0)
return 0;

// could also use `calls.increment(key, size);`
u64 zero = 0, *val;
val = calls.lookup_or_try_init(&key, &zero);
if (val) {
(*val) += size;
}
return 0;
};
""")

b.attach_uprobe(name="c", sym="malloc", fn_name="alloc_enter", pid=pid)
print("Attaching to malloc in pid %d, Ctrl+C to quit." % pid)

# sleep until Ctrl-C
try:
sleep(99999999)
except KeyboardInterrupt:
pass

calls = b.get_table("calls")
stack_traces = b.get_table("stack_traces")

for k, v in reversed(sorted(calls.items(), key=lambda c: c[1].value)):
print("%d bytes allocated at:" % v.value)
if k.value > 0 :
for addr in stack_traces.walk(k.value):
printb(b"\t%s" % b.sym(addr, pid, show_offset=True))

// SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
/* Copyright (c) 2020 Facebook */
#include <stdio.h>
#include <unistd.h>
#include <sys/resource.h>
#include <bpf/libbpf.h>
#include "minimal.skel.h"

static int libbpf_print_fn(enum libbpf_print_level level, const char *format, va_list args)
{
return vfprintf(stderr, format, args);
}

int main(int argc, char **argv)
{
struct minimal_bpf *skel;
int err;

libbpf_set_strict_mode(LIBBPF_STRICT_ALL);
/* Set up libbpf errors and debug info callback */
libbpf_set_print(libbpf_print_fn);

/* Open BPF application */
skel = minimal_bpf__open();
if (!skel) {
fprintf(stderr, "Failed to open BPF skeleton\n");
return 1;
}

/* ensure BPF program only handles write() syscalls from our process */
skel->bss->my_pid = getpid();

/* Load & verify BPF programs */
err = minimal_bpf__load(skel);
if (err) {
fprintf(stderr, "Failed to load and verify BPF skeleton\n");
goto cleanup;
}

/* Attach tracepoint handler */
err = minimal_bpf__attach(skel);
if (err) {
fprintf(stderr, "Failed to attach BPF skeleton\n");
goto cleanup;
}

printf("Successfully started! Please run `sudo cat /sys/kernel/debug/tracing/trace_pipe` "
"to see output of the BPF programs.\n");

for (;;) {
/* trigger our BPF program */
fprintf(stderr, ".");
sleep(1);
}

cleanup:
minimal_bpf__destroy(skel);
return -err;
}

// SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause
/* Copyright (c) 2020 Facebook */
#include <linux/bpf.h>
#include <bpf/bpf_helpers.h>

char LICENSE[] SEC("license") = "Dual BSD/GPL";

int my_pid = 0;

SEC("tp/syscalls/sys_enter_write")
int handle_tp(void *ctx)
{
int pid = bpf_get_current_pid_tgid() >> 32;

if (pid != my_pid)
return 0;

bpf_printk("BPF triggered from PID %d.\n", pid);

return 0;
}

# bpftrace
USAGE:
bpftrace [options] filename
bpftrace [options] -e 'program'

OPTIONS:
-B MODE        output buffering mode ('line', 'full', or 'none')
-d             debug info dry run
-dd            verbose debug info dry run
-e 'program'   execute this program
-h             show this help message
-I DIR         add the specified DIR to the search path for include files.
--include FILE adds an implicit #include which is read before the source file is preprocessed.
-l [search]    list probes
-p PID         enable USDT probes on PID
-c 'CMD'       run CMD and enable USDT probes on resulting process
-q             keep messages quiet
-v             verbose messages
-k             emit a warning when a bpf helper returns an error (except read functions)
-kk            check all bpf helper functions
--version      bpftrace version

ENVIRONMENT:
BPFTRACE_STRLEN             [default: 64] bytes on BPF stack per str()
BPFTRACE_NO_CPP_DEMANGLE    [default: 0] disable C++ symbol demangling
BPFTRACE_MAP_KEYS_MAX       [default: 4096] max keys in a map
BPFTRACE_MAX_PROBES         [default: 512] max number of probes bpftrace can attach to
BPFTRACE_MAX_BPF_PROGS      [default: 512] max number of generated BPF programs
BPFTRACE_CACHE_USER_SYMBOLS [default: auto] enable user symbol cache
BPFTRACE_VMLINUX            [default: none] vmlinux path used for kernel symbol resolution
BPFTRACE_BTF                [default: none] BTF file

EXAMPLES:
bpftrace -l '*sleep*'
list probes containing "sleep"
bpftrace -e 'kprobe:do_nanosleep { printf("PID %d sleeping...\n", pid); }'
trace processes calling sleep
bpftrace -e 'tracepoint:raw_syscalls:sys_enter { @[comm] = count(); }'
count syscalls by process name

probe[,probe]
/predicate/ {
action
}