Linux探测工具BCC(网络)

承接上文，本节以ICMP和TCP为例介绍与网络相关的部分内容。

Linux探测工具BCC(网络)
- Icmp的探测
- TCP的探测

Icmp的探测

首先看下促使我学习bcc的这篇文章中的程序traceicmpsoftirq.py，使用该程序的本意是找出对ping响应的进程位于哪个CPU core上，然后使用perf扫描该core，找出造成网络延迟的原因。源码如下：

#!/usr/bin/python
bpf_text = """
#include <linux/ptrace.h>
#include <linux/sched.h>        /* For TASK_COMM_LEN */
#include <linux/icmp.h>
#include <linux/netdevice.h>
struct probe_icmp_data_t
{
        u64 timestamp_ns;
        u32 tgid;
        u32 pid;
        char comm[TASK_COMM_LEN];
        int v0;
};
BPF_PERF_OUTPUT(probe_icmp_events);
static inline unsigned char *my_skb_transport_header(const struct sk_buff *skb)
{
    return skb->head + skb->transport_header;
}
static inline struct icmphdr *my_icmp_hdr(const struct sk_buff *skb)
{
    return (struct icmphdr *)my_skb_transport_header(skb);
}
int probe_icmp(struct pt_regs *ctx, struct sk_buff *skb)
{
        u64 __pid_tgid = bpf_get_current_pid_tgid();
        u32 __tgid = __pid_tgid >> 32;
        u32 __pid = __pid_tgid; // implicit cast to u32 for bottom half
        struct probe_icmp_data_t __data = {0};
        __data.timestamp_ns = bpf_ktime_get_ns();
        __data.tgid = __tgid;
        __data.pid = __pid;
        bpf_get_current_comm(&__data.comm, sizeof(__data.comm));
        __be16 seq;
        bpf_probe_read_kernel(&seq, sizeof(seq), &my_icmp_hdr(skb)->un.echo.sequence);
        __data.v0 = (int)seq;
        probe_icmp_events.perf_submit(ctx, &__data, sizeof(__data));
        return 0;
}
"""
from bcc import BPF
import ctypes as ct
class Data_icmp(ct.Structure):
    _fields_ = [
        ("timestamp_ns", ct.c_ulonglong),
        ("tgid", ct.c_uint),
        ("pid", ct.c_uint),
        ("comm", ct.c_char * 16),       # TASK_COMM_LEN
        ('v0', ct.c_uint),
    ]
b = BPF(text=bpf_text)
def print_icmp_event(cpu, data, size):
    #event = b["probe_icmp_events"].event(data)
    event = ct.cast(data, ct.POINTER(Data_icmp)).contents
    print("%-7d %-7d %-15s %s" %
                      (event.tgid, event.pid,
                       event.comm.decode('utf-8', 'replace'),
                       event.v0))
b.attach_kprobe(event="icmp_echo", fn_name="probe_icmp")
b["probe_icmp_events"].open_perf_buffer(print_icmp_event)
while 1:
    try:
        b.kprobe_poll()
    except KeyboardInterrupt:
        exit()

上面程序对icmp_echo内核函数进行打点探测，当内核运行该函数时会执行自定义的函数probe_icmp，并获取当前的tgid，pid以及icmp报文的序列号。

内容如下：

my_skb_transport_header：该函数通过偏移sk_buff指针获取传输层首部地址，用于后续获取icmp首部的序列号。此处的操作可以直接参考static bool icmp_echo(struct sk_buff *skb)的内核源码，其获取icmp首部的方式依次为：
```
static inline struct icmphdr *icmp_hdr(const struct sk_buff *skb)
{
	return (struct icmphdr *)skb_transport_header(skb);
}
static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
{
	return skb->head + skb->transport_header;
}
```
可以看到skb_transport_header的处理与本程序的方式是一样的，将该函数的实现直接移植过去即可。需要注意的是，不能直接调用内核函数skb_transport_header获取transport_header的地址。
bpf_get_current_pid_tgid()：获取当前的PID。需要注意的是该函数获取的是当前CPU上运行的进程ID，而不是某一个特定的进程ID。其内核源码如下：
```
BPF_CALL_1(bpf_get_current_ancestor_cgroup_id, int, ancestor_level)
{
	struct cgroup *cgrp = task_dfl_cgroup(current);
	struct cgroup *ancestor;
	ancestor = cgroup_ancestor(cgrp, ancestor_level);
	if (!ancestor)
		return 0;
	return cgroup_id(ancestor);
}
```
而current定义如下，用于获得当前执行进程的task_struct指针。更多参见这篇文章。
```
#define current get_current()
```
因此以本程序为例，如果对icmp_echo的打点采集中如果发生了上下文切换，可能bpf_get_current_pid_tgid获取到的可能是切换后的程序。本文也是借助这种机制，发现在切换到cadvisor导致了网络延时。
bpf_probe_read_kernel:读取内核结构体的成员，原文中使用的是bpf_probe_read，更多参见issue。

其余部分与检测可观测性相同。

TCP的探测

下面看一下TCP的探测，用于跟踪内核代码tcp_v4_connect或tcp_v6_connect，代码源自官方库tools/tcpconnect

#!/usr/bin/python
from __future__ import print_function
from bcc import BPF
from bcc.containers import filter_by_containers
from bcc.utils import printb
import argparse
from socket import inet_ntop, ntohs, AF_INET, AF_INET6
from struct import pack
from time import sleep
# arguments
examples = """examples:
    ./tcpconnect           # trace all TCP connect()s
    ./tcpconnect -t        # include timestamps
    ./tcpconnect -p 181    # only trace PID 181
    ./tcpconnect -P 80     # only trace port 80
    ./tcpconnect -P 80,81  # only trace port 80 and 81
    ./tcpconnect -U        # include UID
    ./tcpconnect -u 1000   # only trace UID 1000
    ./tcpconnect -c        # count connects per src ip and dest ip/port
    ./tcpconnect --cgroupmap mappath  # only trace cgroups in this BPF map
    ./tcpconnect --mntnsmap mappath   # only trace mount namespaces in the map
"""
parser = argparse.ArgumentParser(
    description="Trace TCP connects",
    formatter_class=argparse.RawDescriptionHelpFormatter,
    epilog=examples)
parser.add_argument("-t", "--timestamp", action="store_true",
    help="include timestamp on output")
parser.add_argument("-p", "--pid",
    help="trace this PID only")
parser.add_argument("-P", "--port",
    help="comma-separated list of destination ports to trace.")
parser.add_argument("-U", "--print-uid", action="store_true",
    help="include UID on output")
parser.add_argument("-u", "--uid",
    help="trace this UID only")
parser.add_argument("-c", "--count", action="store_true",
    help="count connects per src ip and dest ip/port")
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("--ebpf", action="store_true",
    help=argparse.SUPPRESS)
args = parser.parse_args() #解析入参
debug = 0
# define BPF program
bpf_text = """
#include <uapi/linux/ptrace.h>
#include <net/sock.h>
#include <bcc/proto.h>
BPF_HASH(currsock, u32, struct sock *); #创建保存socket指针的哈希
// separate data structs for ipv4 and ipv6
struct ipv4_data_t {
    u64 ts_us;
    u32 pid;
    u32 uid;
    u32 saddr;
    u32 daddr;
    u64 ip;
    u16 dport;
    char task[TASK_COMM_LEN];
};
BPF_PERF_OUTPUT(ipv4_events); //创建ipv4的输出
struct ipv6_data_t {
    u64 ts_us;
    u32 pid;
    u32 uid;
    unsigned __int128 saddr;
    unsigned __int128 daddr;
    u64 ip;
    u16 dport;
    char task[TASK_COMM_LEN];
};
BPF_PERF_OUTPUT(ipv6_events); //创建ipv6的输出
// separate flow keys per address family
struct ipv4_flow_key_t { //用于根据地址统计执行tcp_v4_connect的次数，即指定了"-c"或"--count"选项
    u32 saddr;
    u32 daddr;
    u16 dport;
};
BPF_HASH(ipv4_count, struct ipv4_flow_key_t); //统计执行tcp_v4_connect的次数
struct ipv6_flow_key_t { //用于根据地址统计执行tcp_v6_connect的次数，即指定了"-c"或"--count"选项
    unsigned __int128 saddr;
    unsigned __int128 daddr;
    u16 dport;
};
BPF_HASH(ipv6_count, struct ipv6_flow_key_t); //统计执行tcp_v6_connect的次数
int trace_connect_entry(struct pt_regs *ctx, struct sock *sk) //在进入tcp_v4_connect时调用
{
    if (container_should_be_filtered()) {
        return 0;
    }
    u64 pid_tgid = bpf_get_current_pid_tgid(); //获取64位的pid_tgid
    u32 pid = pid_tgid >> 32; //tgid位于高32位，右移32位获取
    u32 tid = pid_tgid;       //tid线程唯一
    FILTER_PID //bpf程序对python来说就是一段字符串，此处可以看作是一个标记符，后续使用python的string.replace进行替换。此处表示过滤特定的PID
    u32 uid = bpf_get_current_uid_gid();
    FILTER_UID //过滤特定的UID
    // stash the sock ptr for lookup on return
    currsock.update(&tid, &sk); //使用tid作为key，保存sk指针指向的地址
    return 0;
};
static int trace_connect_return(struct pt_regs *ctx, short ipver) //在从tcp_v4_connect返回时调用
{
    int ret = PT_REGS_RC(ctx); //获取tcp_v4_connect函数的返回值
    u64 pid_tgid = bpf_get_current_pid_tgid();
    u32 pid = pid_tgid >> 32;
    u32 tid = pid_tgid;
    struct sock **skpp;
    skpp = currsock.lookup(&tid); //判断当前线程在进入tcp_v4_connect时是否打点采集，即是否执行了上面的trace_connect_entry
    if (skpp == 0) {
        return 0;   // missed entry
    }
    if (ret != 0) { //如果tcp_v4_connect的返回值非0，表示无法发送SYNC报文
        // failed to send SYNC packet, may not have populated
        // socket __sk_common.{skc_rcv_saddr, ...}
        currsock.delete(&tid); //本次采集失败，删除哈希
        return 0;
    }
    // pull in details
    struct sock *skp = *skpp;
    u16 dport = skp->__sk_common.skc_dport;
    FILTER_PORT //过滤特定的端口
    if (ipver == 4) {
        IPV4_CODE //根据入参替换为IPV4的处理
    } else /* 6 */ {
        IPV6_CODE //根据入参替换为位IPV6的处理
    }
    currsock.delete(&tid);
    return 0;
}
int trace_connect_v4_return(struct pt_regs *ctx)
{
    return trace_connect_return(ctx, 4);
}
int trace_connect_v6_return(struct pt_regs *ctx)
{
    return trace_connect_return(ctx, 6);
}
"""
struct_init = { 'ipv4':
        { 'count' : #统计执行tcp_v4_connect的次数
               """
               struct ipv4_flow_key_t flow_key = {};
               flow_key.saddr = skp->__sk_common.skc_rcv_saddr;
               flow_key.daddr = skp->__sk_common.skc_daddr;
               flow_key.dport = ntohs(dport);
               ipv4_count.increment(flow_key);""",
          'trace' : #默认执行tcp_v4_connect的跟踪，记录地址，端口等信息
               """
               struct ipv4_data_t data4 = {.pid = pid, .ip = ipver};
               data4.uid = bpf_get_current_uid_gid();
               data4.ts_us = bpf_ktime_get_ns() / 1000;
               data4.saddr = skp->__sk_common.skc_rcv_saddr;
               data4.daddr = skp->__sk_common.skc_daddr;
               data4.dport = ntohs(dport);
               bpf_get_current_comm(&data4.task, sizeof(data4.task));
               ipv4_events.perf_submit(ctx, &data4, sizeof(data4));"""
               },
        'ipv6':
        { 'count' :#统计执行tcp_v6_connect的次数
               """
               struct ipv6_flow_key_t flow_key = {};
               bpf_probe_read_kernel(&flow_key.saddr, sizeof(flow_key.saddr),
                   skp->__sk_common.skc_v6_rcv_saddr.in6_u.u6_addr32);
               bpf_probe_read_kernel(&flow_key.daddr, sizeof(flow_key.daddr),
                   skp->__sk_common.skc_v6_daddr.in6_u.u6_addr32);
               flow_key.dport = ntohs(dport);
               ipv6_count.increment(flow_key);""",
          'trace' : #默认执行tcp_v6_connect的跟踪，记录地址，端口等信息
               """
               struct ipv6_data_t data6 = {.pid = pid, .ip = ipver};
               data6.uid = bpf_get_current_uid_gid();
               data6.ts_us = bpf_ktime_get_ns() / 1000;
               bpf_probe_read_kernel(&data6.saddr, sizeof(data6.saddr),
                   skp->__sk_common.skc_v6_rcv_saddr.in6_u.u6_addr32);
               bpf_probe_read_kernel(&data6.daddr, sizeof(data6.daddr),
                   skp->__sk_common.skc_v6_daddr.in6_u.u6_addr32);
               data6.dport = ntohs(dport);
               bpf_get_current_comm(&data6.task, sizeof(data6.task));
               ipv6_events.perf_submit(ctx, &data6, sizeof(data6));"""
               }
        }
# code substitutions
if args.count: #如果入参指定了"-c"或"-count",则执行count
    bpf_text = bpf_text.replace("IPV4_CODE", struct_init['ipv4']['count'])
    bpf_text = bpf_text.replace("IPV6_CODE", struct_init['ipv6']['count'])
else: #如果入参没有指定"-c"或"-count",则执行trace
    bpf_text = bpf_text.replace("IPV4_CODE", struct_init['ipv4']['trace'])
    bpf_text = bpf_text.replace("IPV6_CODE", struct_init['ipv6']['trace'])
if args.pid: #如果入参指定了"-p"或"--pid",则对PID进行过滤
    bpf_text = bpf_text.replace('FILTER_PID',
        'if (pid != %s) { return 0; }' % args.pid)
if args.port:#如果入参指定了"-P"或"--port",则对端口进行过滤
    dports = [int(dport) for dport in args.port.split(',')]
    dports_if = ' && '.join(['dport != %d' % ntohs(dport) for dport in dports])
    bpf_text = bpf_text.replace('FILTER_PORT',
        'if (%s) { currsock.delete(&pid); return 0; }' % dports_if)
if args.uid:#如果入参指定了"-u"或"--uid",则对UID进行过滤
    bpf_text = bpf_text.replace('FILTER_UID',
        'if (uid != %s) { return 0; }' % args.uid)
bpf_text = filter_by_containers(args) + bpf_text
#下面的处理在没有指定特定的过滤时去除标记符
bpf_text = bpf_text.replace('FILTER_PID', '')
bpf_text = bpf_text.replace('FILTER_PORT', '')
bpf_text = bpf_text.replace('FILTER_UID', '')
if debug or args.ebpf:
    print(bpf_text)
    if args.ebpf:
        exit()
# process event
def print_ipv4_event(cpu, data, size): #TCP4跟踪的打印函数
    event = b["ipv4_events"].event(data)
    global start_ts
    if args.timestamp:
        if start_ts == 0:
            start_ts = event.ts_us
        printb(b"%-9.3f" % ((float(event.ts_us) - start_ts) / 1000000), nl="")
    if args.print_uid:
        printb(b"%-6d" % event.uid, nl="")
    printb(b"%-6d %-12.12s %-2d %-16s %-16s %-4d" % (event.pid,
        event.task, event.ip,
        inet_ntop(AF_INET, pack("I", event.saddr)).encode(), #转换为主机序地址
        inet_ntop(AF_INET, pack("I", event.daddr)).encode(), event.dport)) #转换为主机序地址和端口
def print_ipv6_event(cpu, data, size): #TCP6跟踪的打印函数
    event = b["ipv6_events"].event(data)
    global start_ts
    if args.timestamp:
        if start_ts == 0:
            start_ts = event.ts_us
        printb(b"%-9.3f" % ((float(event.ts_us) - start_ts) / 1000000), nl="")
    if args.print_uid:
        printb(b"%-6d" % event.uid, nl="")
    printb(b"%-6d %-12.12s %-2d %-16s %-16s %-4d" % (event.pid,
        event.task, event.ip,
        inet_ntop(AF_INET6, event.saddr).encode(), inet_ntop(AF_INET6, event.daddr).encode(),
        event.dport))
def depict_cnt(counts_tab, l3prot='ipv4'): #
    for k, v in sorted(counts_tab.items(), key=lambda counts: counts[1].value, reverse=True):
        depict_key = ""
        if l3prot == 'ipv4':
            depict_key = "%-25s %-25s %-20s" %  ((inet_ntop(AF_INET, pack('I', k.saddr))),
                                              inet_ntop(AF_INET, pack('I', k.daddr)), k.dport)
        else:
            depict_key = "%-25s %-25s %-20s" % ((inet_ntop(AF_INET6, k.saddr)),
                                              inet_ntop(AF_INET6, k.daddr), k.dport)
        print ("%s %-10d" % (depict_key, v.value))
# initialize BPF
b = BPF(text=bpf_text)
b.attach_kprobe(event="tcp_v4_connect", fn_name="trace_connect_entry")
b.attach_kprobe(event="tcp_v6_connect", fn_name="trace_connect_entry")
b.attach_kretprobe(event="tcp_v4_connect", fn_name="trace_connect_v4_return")
b.attach_kretprobe(event="tcp_v6_connect", fn_name="trace_connect_v6_return")
print("Tracing connect ... Hit Ctrl-C to end")
if args.count:
    try:
        while 1:
            sleep(99999999)
    except KeyboardInterrupt:
        pass
    # header
    print("\n%-25s %-25s %-20s %-10s" % (
        "LADDR", "RADDR", "RPORT", "CONNECTS"))
    depict_cnt(b["ipv4_count"])
    depict_cnt(b["ipv6_count"], l3prot='ipv6')
# read events
else:
    # header
    if args.timestamp:
        print("%-9s" % ("TIME(s)"), end="")
    if args.print_uid:
        print("%-6s" % ("UID"), end="")
    print("%-6s %-12s %-2s %-16s %-16s %-4s" % ("PID", "COMM", "IP", "SADDR",
        "DADDR", "DPORT"))
    start_ts = 0
    # read events
    b["ipv4_events"].open_perf_buffer(print_ipv4_event)
    b["ipv6_events"].open_perf_buffer(print_ipv6_event)
    while 1:
        try:
            b.perf_buffer_poll()
        except KeyboardInterrupt:
            exit()

上面C程序采集了内核数据skp->sk_common.skc_dport，skp->sk_common.skc_rcv_saddr和skp->__sk_common.skc_daddr。与第一个例子类似，这类数据可以直接参考tcp_v4_connect内核源码的实现，源码中通过struct inet_sock *inet = inet_sk(sk);来获取源目的地址和端口，inet_sock的结构体定义如下，可以明显看到inet_daddr，inet_rcv_saddr和inet_dport与上述代码获取的内容相同，进而可以了解到获取这些成员的方式。

struct inet_sock {
	/* sk and pinet6 has to be the first two members of inet_sock */
	struct sock		sk;
#if IS_ENABLED(CONFIG_IPV6)
	struct ipv6_pinfo	*pinet6;
#endif
	/* Socket demultiplex comparisons on incoming packets. */
#define inet_daddr		sk.__sk_common.skc_daddr
#define inet_rcv_saddr		sk.__sk_common.skc_rcv_saddr
#define inet_dport		sk.__sk_common.skc_dport
#define inet_num		sk.__sk_common.skc_num
...

此外在inet_sock结构体的注释中给出详细的说明，非常明了：

 * @inet_daddr - Foreign IPv4 addr
 * @inet_rcv_saddr - Bound local IPv4 addr
 * @inet_dport - Destination port
 * @inet_num - Local port

因此可以直接参考tcp_v4_connect的源码修改ipv4中获取地址和端口的实现，效果是一样的：

struct_init = { 'ipv4':
        { 'count' :
               """
               struct ipv4_flow_key_t flow_key = {};
               struct inet_sock *inet  = inet_sk(skp);
               flow_key.saddr = inet->inet_rcv_saddr;
               flow_key.daddr = inet->inet_daddr;
               u16 dport = inet->inet_dport;
               flow_key.dport = ntohs(dport);
               ipv4_count.increment(flow_key);""",
          'trace' :
               """
               struct ipv4_data_t data4 = {.pid = pid, .ip = ipver};
               data4.uid = bpf_get_current_uid_gid();
               data4.ts_us = bpf_ktime_get_ns() / 1000;
               struct inet_sock *inet  = inet_sk(skp);
               data4.saddr = inet->inet_rcv_saddr;
               data4.daddr = inet->inet_daddr;
               u16 dport = inet->inet_dport;
               data4.dport = ntohs(dport);
               bpf_get_current_comm(&data4.task, sizeof(data4.task));
               ipv4_events.perf_submit(ctx, &data4, sizeof(data4));"""
               },
        'ipv6':
        { 'count' :
               """
               struct ipv6_flow_key_t flow_key = {};
               bpf_probe_read_kernel(&flow_key.saddr, sizeof(flow_key.saddr),
                   skp->__sk_common.skc_v6_rcv_saddr.in6_u.u6_addr32);
               bpf_probe_read_kernel(&flow_key.daddr, sizeof(flow_key.daddr),
                   skp->__sk_common.skc_v6_daddr.in6_u.u6_addr32);
               flow_key.dport = ntohs(dport);
               ipv6_count.increment(flow_key);""",
          'trace' :
               """
               struct ipv6_data_t data6 = {.pid = pid, .ip = ipver};
               data6.uid = bpf_get_current_uid_gid();
               data6.ts_us = bpf_ktime_get_ns() / 1000;
               bpf_probe_read_kernel(&data6.saddr, sizeof(data6.saddr),
                   skp->__sk_common.skc_v6_rcv_saddr.in6_u.u6_addr32);
               bpf_probe_read_kernel(&data6.daddr, sizeof(data6.daddr),
                   skp->__sk_common.skc_v6_daddr.in6_u.u6_addr32);
               data6.dport = ntohs(dport);
               bpf_get_current_comm(&data6.task, sizeof(data6.task));
               ipv6_events.perf_submit(ctx, &data6, sizeof(data6));"""
               }
        }

此外注意到读取TCP4的数据时没有用到bpf_probe_read_kernel，但读取TCP6的数据时用到了bpf_probe_read_kernel，这是因为TCP4的地址是一个u32类型的数据，直接赋值即可；而TCP6的地址结构如下，无法通过直接赋值获取，因此需要调用bpf_probe_read_kernel拷贝内存。

struct in6_addr {
	union {
		__u8		u6_addr8[16];
#if __UAPI_DEF_IN6_ADDR_ALT
		__be16		u6_addr16[8];
		__be32		u6_addr32[4];
#endif
	} in6_u;
#define s6_addr			in6_u.u6_addr8
#if __UAPI_DEF_IN6_ADDR_ALT
#define s6_addr16		in6_u.u6_addr16
#define s6_addr32		in6_u.u6_addr32
#endif
};

整体看，上面代码使用了python处理了一些C程序的替换和拼接，大部分跟可观测性并没有什么不同，当然，最主要的还是需要了解内核处理流程，选择正确的内核函数进行打点。

上述仅给出的网络并无法修改报文内容以及对报文进行重定向等操作。ebpf提供了XDP和tc两种管理网络的方式，更多可以参见下一篇博客。