如何以编程方式从其父级获取子进程的堆栈跟踪？

2023年2月27日 400次阅读

假设我在我的程序中分叉子进程.在某些时候,我用kill(child,SIGSTOP)暂停子进程,并想检查堆栈的内容.有没有办法以编程方式从其父进程获取子进程的堆栈跟踪？

我知道ptrace是跟踪子进程并检查其内存/寄存器的标准方法.我也知道backtrace为调用线程提供了这个功能.是否有合并这些功能的功能或库？或者我需要使用ptrace手动移动堆栈吗？

最佳答案 回答我自己的问题 – 这是可行的.你需要
libunwind和ptrace. libunwind为ptrace提供了一个包装器,允许您展开远程目标.这是示例代码,运行NPB基准(cg,A类)：

#include <stdio.h>
#include <stdlib.h>
#include <time.h>
#include <unistd.h>
#include <wait.h>
#include <sys/ptrace.h>
#include <libunwind.h>
#include <libunwind-x86_64.h>
#include <libunwind-ptrace.h>
#include <signal.h>

#define panic(X) fprintf(stderr, #X "\n");

static unw_addr_space_t as;
static struct UPT_info *ui;

void do_backtrace(pid_t child) {

    ui = _UPT_create(child);
    if (!ui) {
        panic("_UPT_create failed");
    }

    ptrace(PTRACE_ATTACH, child, 0, 0);
    struct timespec t = { .tv_sec = 0, t.tv_nsec = 1000000 };
    nanosleep(&t, NULL);

    unw_cursor_t c;
    int rc = unw_init_remote(&c, as, ui);
    if (rc != 0) {
        if (rc == UNW_EINVAL) {
            panic("unw_init_remote: UNW_EINVAL");
        } else if (rc == UNW_EUNSPEC) {
            panic("unw_init_remote: UNW_EUNSPEC");
        } else if (rc == UNW_EBADREG) {
            panic("unw_init_remote: UNW_EBADREG");
        } else {
            panic("unw_init_remote: UNKNOWN");
        }
    }

    do {
        unw_word_t  offset, pc;
        char        fname[64];

        unw_get_reg(&c, UNW_REG_IP, &pc);
        fname[0] = '\0';
        (void) unw_get_proc_name(&c, fname, sizeof(fname), &offset);

        printf("\n%p : (%s+0x%x) [%p]\n", (void *)pc,
                                          fname,
                                          (int) offset,
                                          (void *) pc);
    } while (unw_step(&c) > 0);


    ptrace(PTRACE_DETACH, child, 0, 0);

    _UPT_destroy(ui);
}


int main(int argc __attribute__((unused)), char **argv, char **envp) {

    as = unw_create_addr_space(&_UPT_accessors, 0);
    if (!as) {
        panic("unw_create_addr_space failed");
    }

    pid_t child;
    child = fork();

    if (!child) {

        execve("/home/#######/#######/my_utilities/child_bt/cg.A.x",
                argv, envp);

        return 0;

    } else {

        struct timespec t = { .tv_sec = 1, .tv_nsec = 0 };
        nanosleep(&t, NULL);

        do_backtrace(child);

        int status;
        waitpid(child, &status, 0);

    }


    return 0;
}

并输出：

#######-######-desktop:~/popcorn/my_utilities/child_bt$./child_bt 

 NAS Parallel Benchmarks (NPB3.3-SER) - CG Benchmark

 Size:       14000
 Iterations:    15

 Initialization time =           0.422 seconds

   iteration           ||r||                 zeta
        1       0.26065081214763E-12    19.9997581277040
        2       0.25753187736717E-14    17.1140495745506
        3       0.25934878907518E-14    17.1296668946143
        4       0.25626292684826E-14    17.1302113581193
        5       0.25110613524700E-14    17.1302338856353
        6       0.25581937582088E-14    17.1302349879482
        7       0.25456477041068E-14    17.1302350498916
        8       0.24494068328538E-14    17.1302350537510

0x400c85 : (conj_grad_+0x135) [0x400c85]

0x401ec8 : (MAIN__+0x739) [0x401ec8]

0x402b39 : (main+0x1d) [0x402b39]

0x7f8ee80c2ec5 : (__libc_start_main+0xf5) [0x7f8ee80c2ec5]

0x400a89 : (_start+0x29) [0x400a89]
        9       0.24885235903729E-14    17.1302350540101
       10       0.24771507610856E-14    17.1302350540284
       11       0.24928441017003E-14    17.1302350540298
       12       0.24443706061229E-14    17.1302350540299
       13       0.24709361922612E-14    17.1302350540299
       14       0.24381630450112E-14    17.1302350540299
       15       0.24296673223448E-14    17.1302350540299
 Benchmark completed 
 VERIFICATION SUCCESSFUL 
 Zeta is     0.1713023505403E+02
 Error is    0.5122640033228E-13


 CG Benchmark Completed.
 Class           =                        A
 Size            =                    14000
 Iterations      =                       15
 Time in seconds =                     1.01
 Mop/s total     =                  1483.11
 Operation type  =           floating point
 Verification    =               SUCCESSFUL
 Version         =                    3.3.1
 Compile date    =              16 Jul 2015

 Compile options:
    F77          = gfortran 
    FLINK        = $(F77)
    F_LIB        = (none)
    F_INC        = (none)
    FFLAGS       = -O
    FLINKFLAGS   = -O
    RAND         = randi8


 Please send all errors/feedbacks to:

 NPB Development Team
 npb@nas.nasa.gov

我将do_backtrace函数基于libunwind发行版的tests文件夹中的test-ptrace.c文件以及this blog中的代码.

点赞