容器之namespace

1. 介绍

简单玩了下Linux kernel为容器技术提供的基础设施之一namespace(另一个是cgroups)，包括uts/user/pid/mnt/ipc/net六个(3.13.0的内核). 这东西主要用来做资源的隔离，我感觉本质上是全局资源的映射，映射之间独立了自然隔离了。主要涉及到的东西是:

• clone
• setns
• unshare
• /proc/pid/ns, /proc/pid/uid_map, /proc/pid/gid_map等

2. 测试流程及代码

下面是一些简单的例子，主要测试uts/pid/user/mnt四个namespace的效果，测试代码主要用到三个进程，一个是clone系统调用执行/bin/bash后的进程，也是生成新的子namespace的初始进程，然后是打开/proc/pid/ns下的namespace链接文件，用setns将第二个可执行文件的进程加入/bin/bash的进程的namespace(容器)，并让其fork出一个子进程，测试pid namespace的差异。值得注意的几个点:

• 不同版本的内核setns和unshare对namespace的支持不一样，较老的内核可能只支持ipc/net/uts三个namespace
• 某个进程创建后其pid namespace就固定了，使用setns和unshare改变后，其本身的pid namespace不会改变，只有fork出的子进程的pid namespace改变
• setns将进程加入的新namespace需是此进程的后代namespace
• 用setns添加mnt namespace应该放在其他namespace之后，否则可能出现无法打开/proc/pid/ns/…的错误

// 代码1: 开一些新的namespace(启动新容器)
#define _GNU_SOURCE
#include <sys/wait.h>
#include <sched.h>
#include <string.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>

#define errExit(msg)  do { perror(msg); exit(EXIT_FAILURE); \
} while (0)

/* Start function for cloned child */
static int childFunc(void *arg)
{
  const char *binary = "/bin/bash";
  char *const argv[] = {
    "/bin/bash",
    NULL
  };
  char *const envp[] = { NULL };

  /* wrappers for execve */
  // has const char * as argument list
  // execl 
  // execle  => has envp
  // execlp  => need search PATH 

  // has char *const arr[] as argument list 
  // execv 
  // execvpe => need search PATH and has envp
  // execvp  => need search PATH 

  //int ret = execve(binary, argv, envp);
  int ret = execv(binary, argv);
  if (ret < 0) {
    errExit("execve error");
  }
  return ret;
}

#define STACK_SIZE (1024 * 1024)    /* Stack size for cloned child */

int main(int argc, char *argv[])
{
  char *stack; 
  char *stackTop;                 
  pid_t pid;
  stack = malloc(STACK_SIZE);
  if (stack == NULL)
    errExit("malloc");
  stackTop = stack + STACK_SIZE;  /* Assume stack grows downward */

  //pid = clone(childFunc, stackTop, CLONE_NEWUTS | CLONE_NEWNS | CLONE_NEWPID | CLONE_NEWUSER | SIGCHLD, NULL);
  pid = clone(childFunc, stackTop, CLONE_NEWUTS | CLONE_NEWNS | CLONE_NEWPID | CLONE_NEWUSER | CLONE_NEWIPC | SIGCHLD, NULL);
//pid = clone(childFunc, stackTop, CLONE_NEWUTS | //CLONE_NEWNS | CLONE_NEWPID | CLONE_NEWUSER | CLONE_NEWIPC //| CLONE_NEWNET | SIGCHLD, NULL);
  if (pid == -1)
    errExit("clone");
  printf("clone() returned %ld\n", (long) pid);

  if (waitpid(pid, NULL, 0) == -1)  
    errExit("waitpid");
  printf("child has terminated\n");

  exit(EXIT_SUCCESS);
}

// 代码2: 使用setns加入新进程
#define _GNU_SOURCE  // ?
#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <errno.h>
#include <sys/utsname.h>
#include <unistd.h>
#include <sys/types.h>
#include <sched.h>
#include <fcntl.h>
#include <wait.h>

// mainly setns and unshare system calls

/* int setns(int fd, int nstype); */

// 不同版本内核/proc/pid/ns下namespace文件情况
/*
   CLONE_NEWCGROUP (since Linux 4.6)
   fd must refer to a cgroup namespace.

   CLONE_NEWIPC (since Linux 3.0)
   fd must refer to an IPC namespace.

   CLONE_NEWNET (since Linux 3.0)
   fd must refer to a network namespace.

   CLONE_NEWNS (since Linux 3.8)
   fd must refer to a mount namespace.

   CLONE_NEWPID (since Linux 3.8)
   fd must refer to a descendant PID namespace.

   CLONE_NEWUSER (since Linux 3.8)
   fd must refer to a user namespace.

   CLONE_NEWUTS (since Linux 3.0)
   fd must refer to a UTS namespace.
   */

/* // 特殊的pid namespace 
   CLONE_NEWPID behaves somewhat differently from the other nstype
values: reassociating the calling thread with a PID namespace changes
only the PID namespace that child processes of the caller will be
created in; it does not change the PID namespace of the caller
itself.  Reassociating with a PID namespace is allowed only if the
PID namespace specified by fd is a descendant (child, grandchild,
etc.)  of the PID namespace of the caller.  For further details on
PID namespaces, see pid_namespaces(7).
*/


/*
int unshare(int flags);
CLONE_FILES | CLONE_FS | CLONE_NEWCGROUP | CLONE_NEWIPC | CLONE_NEWNET 
| CLONE_NEWNS | CLONE_NEWPID | CLONE_NEWUSER | CLONE_NEWUTS | CLONE_SYSVSEM
*/



#define MAX_PROCPATH_LEN 1024

#define errorExit(msg) \
  do { fprintf(stderr, "%s in file %s in line %d\n", msg, __FILE__, __LINE__);    exit(EXIT_FAILURE); } while (0)

void printInfo();
int openAndSetns(const char *path);

int main(int argc, char *argv[])
{
  if (argc < 2) {
    fprintf(stdout, "usage : execname pid(find namespaces of this process)\n");
    return 0;
  }
  printInfo();

  fprintf(stdout, "---- setns for uts ----\n");
  char uts[MAX_PROCPATH_LEN];
  snprintf(uts, MAX_PROCPATH_LEN, "/proc/%s/ns/uts", argv[1]);
  openAndSetns(uts);
  printInfo();

  fprintf(stdout, "---- setns for user ----\n");
  char user[MAX_PROCPATH_LEN];
  snprintf(user, MAX_PROCPATH_LEN, "/proc/%s/ns/user", argv[1]);
  openAndSetns(user);
  printInfo();

  // 注意pid namespace的不同行为，只有后续创建的子进程进入setns设置
  // 的新的pid namespace，本进程不会改变
  fprintf(stdout, "---- setns for pid ----\n");
  char pidpath[MAX_PROCPATH_LEN];
  snprintf(pidpath, MAX_PROCPATH_LEN, "/proc/%s/ns/pid", argv[1]);
  openAndSetns(pidpath);
  printInfo();


  fprintf(stdout, "---- setns for ipc ----\n");
  char ipc[MAX_PROCPATH_LEN];
  snprintf(ipc, MAX_PROCPATH_LEN, "/proc/%s/ns/ipc", argv[1]);
  openAndSetns(ipc);
  printInfo();

  fprintf(stdout, "---- setns for net ----\n");
  char net[MAX_PROCPATH_LEN];
  snprintf(net, MAX_PROCPATH_LEN, "/proc/%s/ns/net", argv[1]);
  openAndSetns(net);
  printInfo();

  // 注意mnt namespace需要放在其他后面，避免mnt namespace改变后
  // 找不到/proc/pid/ns下的文件
  fprintf(stdout, "---- setns for mount ----\n");
  char mount[MAX_PROCPATH_LEN];
  snprintf(mount, MAX_PROCPATH_LEN, "/proc/%s/ns/mnt", argv[1]);
  openAndSetns(mount);
  printInfo();

  // 测试子进程的pid namespace
  int ret = fork();
  if (-1 == ret) {
    errorExit("failed to fork");
  } else if (ret == 0) {
    fprintf(stdout, "********\n");
    fprintf(stdout, "in child process\n");
    printInfo();
    fprintf(stdout, "********\n");
    for (;;) {
      sleep(5);
    }
  } else {
    fprintf(stdout, "child pid : %d\n", ret);
  }
  for (;;) {
    sleep(5);
  }
  waitpid(ret, NULL, 0);
  return 0;
}

void printInfo()
{
  pid_t pid;
  struct utsname uts;
  uid_t uid;
  gid_t gid;
  // pid namespace 
  pid = getpid();
  // user namespace 
  uid = getuid();
  gid = getgid();
  // uts namespace 
  uname(&uts);
  fprintf(stdout, "pid : %d\n", pid);
  fprintf(stdout, "uid : %d\n", uid);
  fprintf(stdout, "gid : %d\n", gid);
  fprintf(stdout, "hostname : %s\n", uts.nodename);
}

int openAndSetns(const char *path)
{
  int ret = open(path, O_RDONLY, 0);
  if (-1 == ret) {
    fprintf(stderr, "%s\n", strerror(errno));
    errorExit("failed to open fd");
  }
  if (-1 == (ret = setns(ret, 0))) {
    fprintf(stderr, "%s\n", strerror(errno));
    errorExit("failed to setns");
  }
  return ret;
}

3. 测试效果

• user的效果 : 通过/proc/pid/uid_map和/proc/pid/gid_map设置container外用户id和容器内用户id的映射关系(把这放前面是因为后面hostname和mount需要权限…)
  
  
  

• uts的效果 : 改变container中的hostname不会影响container外面的hostname
  
  

• pid和mnt的效果 : container中进程id被重新映射，在container中重新挂载/proc filesystem不会影响容器外的/proc
  
  

• setns的测试

  • 依次为init进程，container init进程(6个namespace的flag都指定了)，新加入container的进程以及其fork出的子进程的namespace情况，可以看到container init进程与init进程的namespace完全不同了，新加入container的进程除了pid与init相同外，其他namespace与container init进程相同，而新加入container的进程fork出的子进程的namespace则与container init进程完全相同
    

  • 新加入container init进程pid namespace的子进程
    
    

  • 程序2输出