通过strace抓取之前这个loader的系统调用：

bpf(BPF_OBJ_GET, {pathname="/sys/fs/bpf/prog_bpftest_tracepoint_raw_syscalls_sys_enter", bpf_fd=0, file_flags=0}, 120) = 3
openat(AT_FDCWD, "/sys/kernel/tracing/events/raw_syscalls/sys_enter/id", O_RDONLY|O_CLOEXEC) = 4
read(4, "21\n", 4096)                   = 3
close(4)                                = 0
perf_event_open({type=PERF_TYPE_TRACEPOINT, size=0, config=21, ...}, -1, 0, -1, PERF_FLAG_FD_CLOEXEC) = 4
ioctl(4, PERF_EVENT_IOC_SET_BPF, 3)     = 0
ioctl(4, PERF_EVENT_IOC_ENABLE, 0)      = 0
nanosleep({tv_sec=1, tv_nsec=0}, 0x7ff104b788) = 0
bpf(BPF_OBJ_GET, {pathname="/sys/fs/bpf/map_bpftest_execve_map", bpf_fd=0, file_flags=0}, 120) = 5
nanosleep({tv_sec=0, tv_nsec=40000000}, NULL) = 0
bpf(BPF_MAP_LOOKUP_ELEM, {map_fd=5, key=0x7ff104b5f4, value=0x7ff104b5e8}, 120) = 0

通过上面的系统调用，我们就能理清楚，loader程序到底做了哪些工作。

接着我找到了Linux内核中的一个bpf_load.c，参考了一下在普通的Linux系统中，loader是怎么处理的，所以我对该程序进行了修改，增加了以下代码：

struct androidBPF {
    char *prog_path;
    char *map_path;
    char *tp_category;
    char *tp_name;
};
static int load_prog(char *prog_path)
{
    int pfd;
    pfd = bpf_obj_get(prog_path);
    if (pfd < 0) {
        printf("bpf_prog_load() err=%d\n%s", errno, prog_path);
        return -1;
    }
    return pfd;
}
int attach_tracepoint(char *tp_category, char *tp_name)
{
    char buf[256];
    int efd, err, id;
    struct perf_event_attr attr = {};
    attr.type = PERF_TYPE_TRACEPOINT;
    attr.sample_type = PERF_SAMPLE_RAW;
    attr.sample_period = 1;
    attr.wakeup_events = 1;
    strcpy(buf, DEBUGFS);
    strcat(buf, "events/");
    strcat(buf, tp_category);
    strcat(buf, "/");
    strcat(buf, tp_name);
    strcat(buf, "/id");
    efd = open(buf, O_RDONLY, 0);
    if (efd < 0) {
        printf("failed to open %s\n", buf);
        return -1;
    }
    err = read(efd, buf, sizeof(buf));
    if (err < 0 || err >= sizeof(buf)) {
        printf("read from failed '%s'\n", strerror(errno));
        return -1;
    }
    close(efd);
    buf[err] = 0;
    id = atoi(buf);
    attr.config = id;
    efd = perf_event_open(&attr, -1/*pid*/, 0/*cpu*/, -1/*group_fd*/, 0);
    if (efd < 0) {
        printf("event %d fd %d err %s\n", id, efd, strerror(errno));
        return -1;
    }
    return efd;
}
static int load_map(char *map_path)
{
    int mfd;
    mfd = bpf_obj_get(map_path);
    if (mfd < 0) {
        printf("bpf_map_load() err=%d\n%s", errno, map_path);
        return -1;
    }
    return mfd;
}
int get_map_by_int_key(int *key, void *value)
{
    int mfd, ret;
    mfd = map_fd[prog_cnt-1];
    ret = bpf_lookup_elem(mfd, key, value);
    return ret;
}
int load_bpf_from_fs(struct androidBPF *abpf)
{
    int fd, efd, mfd;
    fd = load_prog(abpf->prog_path);
    if (fd <= 0) {
        printf("[debug] load prog error.\n");
        return fd;
    }
    prog_fd[prog_cnt] = fd;
    efd = attach_tracepoint(abpf->tp_category, abpf->tp_name);
    if (efd <= 0) {
        printf("[debug] attach_tracepoint error.\n");
        return efd;
    }
    event_fd[prog_cnt] = efd;
    ioctl(efd, PERF_EVENT_IOC_ENABLE, 0);
    ioctl(efd, PERF_EVENT_IOC_SET_BPF, fd);
    printf("[debug] load bpf prog success.\n");
    mfd = load_map(abpf->map_path);
    if (mfd <= 0) {
        printf("[debug] load_map error.\n");
        return mfd;
    }
    map_fd[prog_cnt++] = mfd;
    return 0;
}
void read_trace_pipe(int times)
{
    int trace_fd;
    trace_fd = open(DEBUGFS "trace_pipe", O_RDONLY, 0);
    if (trace_fd < 0)
        return;
    // times = 0, loop 0xffffffff
    do {
        static char buf[4096];
        ssize_t sz;
        sz = read(trace_fd, buf, sizeof(buf) - 1);
        if (sz > 0) {
            buf[sz] = 0;
            puts(buf);
        }
    } while (--times);
}

接着，我就能使用C代码来写loader了：

#include "bpf_load.h"
#include <stdio.h>
#include <string.h>
int main(int argc, char **argv) {
  struct androidBPF abpf = {0, };
    if (argc < 3)
        return 0;
  abpf.prog_path = argv[1];
  abpf.map_path = argv[2];
  abpf.tp_category = "raw_syscalls";
  abpf.tp_name = "sys_enter";
  if (load_bpf_from_fs(&abpf) != 0) { // 用于加载 ELF 格式的 BPF 程序
    printf("The kernel didn't load the BPF program\n");
    return -1;
  }
  int key, ret;
  key = 1;
  struct event_execv value;
  for (int i = 0; i < 10; i ++) {
    memset(&value, 0, sizeof(value));
    ret = get_map_by_int_key(&key, &value);
    printf("[debug] ret = %d, pid = %d, gid = %d, comm = %s\n", ret, value.pid, value.gid, value.cmd);
  }
  // for (int i=0; i < 88; i++)
  // {
  //   printf("debug value[%d] = 0x%x\n", i, *((char *)&value+i));
  // }
  read_trace_pipe(1);
  return 0;
}

在本地的arm64机器上就能编译了：

$ ls -alF
total 916
drwxr-xr-x 1 hehe hehe    320 Oct 31 11:54 ./
drwxr-xr-x 1 hehe hehe    416 Oct 30 22:34 ../
-rw-rw-r-- 1 hehe hehe   4025 Oct 30 22:37 bpf_helpers.h
-rw-rw-r-- 1 hehe hehe  10940 Oct 31 11:35 bpf_load.c
-rw-rw-r-- 1 hehe hehe   1112 Oct 31 11:35 bpf_load.h
-rw-rw-r-- 1 hehe hehe   3432 Oct 30 22:49 libbpf.c
-rw-rw-r-- 1 hehe hehe   5294 Oct 30 22:50 libbpf.h
-rw-r--r-- 1 hehe hehe 117176 Oct 30 22:54 libelf.so
-rwxrwxr-x 1 hehe hehe 773016 Oct 31 11:54 loader*
-rw-rw-r-- 1 hehe hehe    868 Oct 31 11:54 loader.c
$ clang loader.c bpf_load.c libbpf.c -lelf -lz -o loader -static

1. BPF_OBJ_GET prog_bpftest_tracepoint_raw_syscalls_sys_enter，获取prog对象
2. 读取SEC定义的section的id，从/sys/kernel/tracing/events/raw_syscalls/sys_enter/id获取
3. perf_event_open打开相应时间，因为是tracepoint，所以type要设置为PERF_TYPE_TRACEPOINT，config等于上面获取id
4. 打开事件后，获取了一个文件描述符，对该文件描述符进行ioctl操作，操作的命令有两个，PERF_EVENT_IOC_SET_BPF和PERF_EVENT_IOC_ENABLE，PERF_EVENT_IOC_SET_BPF设置为prog对象的文件描述符

1. BPF_OBJ_GET /sys/fs/bpf/map_bpftest_execve_map，获取map对象。
2. BPF_MAP_LOOKUP_ELEM {map_fd=5, key=0x7ff104b5f4, value=0x7ff104b5e8}，从map_fd中搜索key对应的value，储存在value的指针中返回。

1. https://elixir.bootlin.com/linux/v4.14.2/source/samples/bpf/bpf_load.c