ceph源码分析：Admin Socket机制

转自http://blog.chinaunix.net/uid-24774106-id-5059727.html

任何一个成熟的项目，必须要提供出接口，就像探针一样，可以让我们探查进程内部的运行情况，进程不能是一个黑盒子。对于ceph而言，Admin Socket 提供了该功能。Admin Socket 不仅可以查看当前的配置，进程运行的状态，而且，还可以修改配置，获取log等。

下面help可以看出，ceph给出了很多命令，来了解ceph内部的运行情况。

 root@test3:~# ceph daemon /var/run/ceph/ceph-osd.4.asok help
 { "config get": "config get : get the config value",
   "config set": "config set [ ...]: set a config variable",
   "config show": "dump current config settings",
   "dump_blacklist": "dump blacklisted clients and times",
   "dump_historic_ops": "show slowest recent ops",
   "dump_op_pq_state": "dump op priority queue state",
   "dump_ops_in_flight": "show the ops currently in flight",
   "dump_watchers": "show clients which have active watches, and on which objects",
   "get_command_descriptions": "list available commands",
   "getomap": "output entire object map",
   "git_version": "get git sha1",
   "help": "list available commands",
   "injectdataerr": "inject data error into omap",
   "injectmdataerr": "inject metadata error",
   "log dump": "dump recent log entries to log file",
   "log flush": "flush log entries to log file",
   "log reopen": "reopen log file",
   "perf dump": "dump perfcounters value",
   "perf schema": "dump perfcounters schema",
   "rmomapkey": "remove omap key",
   "setomapheader": "set omap header",
   "setomapval": "set omap key",
   "truncobj": "truncate object to length",
   "version": "get ceph version"}
 root@test3:~# ceph daemon /var/run/ceph/ceph-mon.*.asok help
 { "add_bootstrap_peer_hint": "add peer address as potential bootstrap peer for cluster bringup",
   "config get": "config get : get the config value",
   "config set": "config set [ ...]: set a config variable",
   "config show": "dump current config settings",
   "get_command_descriptions": "list available commands",
   "git_version": "get git sha1",
   "help": "list available commands",
   "log dump": "dump recent log entries to log file",
   "log flush": "flush log entries to log file",
   "log reopen": "reopen log file",
   "mon_status": "show current monitor status",
   "perf dump": "dump perfcounters value",
   "perf schema": "dump perfcounters schema",
   "quorum_status": "show current quorum status",
   "sync_force": "force sync of and clear monitor store",
   "version": "get ceph version"}
 root@test3:~# ceph daemon /var/run/ceph/ceph-mds.*.asok help
 { "config get": "config get : get the config value",
   "config set": "config set [ ...]: set a config variable",
   "config show": "dump current config settings",
   "get_command_descriptions": "list available commands",
   "git_version": "get git sha1",
   "help": "list available commands",
   "log dump": "dump recent log entries to log file",
   "log flush": "flush log entries to log file",
   "log reopen": "reopen log file",
   "objecter_requests": "show in-progress osd requests",
   "perf dump": "dump perfcounters value",
   "perf schema": "dump perfcounters schema",
   "version": "get ceph version"}

比如可以查看 ceph的各个模块的当前配置

 ceph daemon /var/run/ceph/ceph-mds.*.asok config show
 ceph daemon /var/run/ceph/ceph-mon.*.asok config show
 ceph daemon /var/run/ceph/ceph-osd.4.asok config show

这个是怎么实现的呢？ 这是依靠ceph的AdminSocket机制完成。
CephContext中会创建一个AdminSocket对象，该对象本质是一个线程。ceph-mon/ceph-osd/ceph-mds这些进程都会有创建一个AdminSocket的线程，负责响应用户的探查命令。

从上面的OSD MON MDS help的不同输出可以看出，他们支持的命令有共同的，也有各自独立的。
首先是共同支持的命令有：

• config show        显示所有的配置项
• config get           获取某个配置项
• config set           设置某个配置项
• log flush　　      将log 刷入日志文件
• log dump           将最近的若干笔log刷入到log文件
• log reopen         重新打开log文件
• perf dump         输出统计信息
• perf schema      输出统计信息的类型
• version             版本信息
• git_version   :    git 版本信息

支持的大部分common的命令，注册发生在 CephContext的构造函数：

   _admin_hook = new CephContextHook(this);
   _admin_socket->register_command("perfcounters_dump", "perfcounters_dump", _admin_hook, "");
   _admin_socket->register_command("1", "1", _admin_hook, "");
   _admin_socket->register_command("perf dump", "perf dump", _admin_hook, "dump perfcounters value");
   _admin_socket->register_command("perfcounters_schema", "perfcounters_schema", _admin_hook, "");
   _admin_socket->register_command("2", "2", _admin_hook, "");
   _admin_socket->register_command("perf schema", "perf schema", _admin_hook, "dump perfcounters schema");
   _admin_socket->register_command("config show", "config show", _admin_hook, "dump current config settings");
   _admin_socket->register_command("config set", "config set name=var,type=CephString name=val,type=CephString,n=N", _admin_hook, "config set [ ...]: set a config variable");
   _admin_socket->register_command("config get", "config get name=var,type=CephString", _admin_hook, "config get : get the config value");
   _admin_socket->register_command("log flush", "log flush", _admin_hook, "flush log entries to log file");
   _admin_socket->register_command("log dump", "log dump", _admin_hook, "dump recent log entries to log file");
   _admin_socket->register_command("log reopen", "log reopen", _admin_hook, "reopen log file")

首先定义了个CephContextHook，注册该Hook也作为一个参数传递进去，这个register_command比较简单，就是建立了command字符串和Hook的关联，这个关联有何作用，后面会分析到，暂时按下不表。

 int AdminSocket::register_command(std::string command, std::string cmddesc, AdminSocketHook *hook, std::string help)
 {
   int ret;
   m_lock.Lock();
   if (m_hooks.count(command)) {
     ldout(m_cct, 5) << "register_command " << command << " hook " << hook << " EEXIST" << dendl;
     ret = -EEXIST;
   } else {
     ldout(m_cct, 5) << "register_command " << command << " hook " << hook << dendl;
     m_hooks[command] = hook;
     m_descs[command] = cmddesc;
     m_help[command] = help;
     ret = 0;
   }
   m_lock.Unlock();
   return ret;
 }

后面从注册部分的代码我们可以知道下面三个命令是等效的。

 ceph daemon /var/run/ceph/ceph-mds.bfudz.asok perf dump
 ceph daemon /var/run/ceph/ceph-mds.bfudz.asok perfcounters_dump
 ceph daemon /var/run/ceph/ceph-mds.bfudz.asok 1

还有下面三个命令是等效的：

 ceph daemon /var/run/ceph/ceph-mds.bfudz.asok perf schema
 ceph daemon /var/run/ceph/ceph-mds.bfudz.asok perfcounters_schema
 ceph daemon /var/run/ceph/ceph-mds.bfudz.asok 2

 ceph-osd/ceph-mon/ceph-mds有一些自己的独特的注册函数，大家检索register_command可以检索的到这些独特的命令。


这个AdminSocket 对应的线程，在common_init_finish函数中负责创建。

 void common_init_finish(CephContext *cct)
 {
   ceph::crypto::init(cct);
   cct->start_service_thread();
   if (cct->_conf->lockdep) {
     g_lockdep = true;
     ldout(cct,0) << "lockdep is enabled" << dendl;
     lockdep_register_ceph_context(cct);
   }
 }

而在start_service_thread函数中会调用adminsocket的init函数：

 void CephContext::start_service_thread()
 {
   pthread_spin_lock(&_service_thread_lock);
   if (_service_thread) {
     pthread_spin_unlock(&_service_thread_lock);
     return;
   }
   _service_thread = new CephContextServiceThread(this);
   _service_thread->create();
   pthread_spin_unlock(&_service_thread_lock);
   // make logs flush on_exit()
   if (_conf->log_flush_on_exit)
     _log->set_flush_on_exit();
   // Trigger callbacks on any config observers that were waiting for
   // it to become safe to start threads.
   _conf->set_val("internal_safe_to_start_threads", "true");
   _conf->call_all_observers();
   // start admin socket
   if (_conf->admin_socket.length())
     _admin_socket->init(_conf->admin_socket);
 }

接下来可以分析下admin_socket的初始化函数init：

 bool AdminSocket::init(const std::string &path)
 {
   ldout(m_cct, 5) << "init " << path << dendl;
   /* Set up things for the new thread */
   std::string err;
   int pipe_rd = -1, pipe_wr = -1;
   err = create_shutdown_pipe(&pipe_rd, &pipe_wr);
   if (!err.empty()) {
     lderr(m_cct) << "AdminSocketConfigObs::init: error: " << err << dendl;
     return false;
   }
   int sock_fd;
   err = bind_and_listen(path, &sock_fd);
   if (!err.empty()) {
     lderr(m_cct) << "AdminSocketConfigObs::init: failed: " << err << dendl;
     close(pipe_rd);
     close(pipe_wr);
     return false;
   }
   /* Create new thread */
   m_sock_fd = sock_fd;
   m_shutdown_rd_fd = pipe_rd;
   m_shutdown_wr_fd = pipe_wr;
   m_path = path;
   m_version_hook = new VersionHook;
   register_command("0", "0", m_version_hook, "");
   register_command("version", "version", m_version_hook, "get ceph version");
   register_command("git_version", "git_version", m_version_hook, "get git sha1");
   m_help_hook = new HelpHook(this);
   register_command("help", "help", m_help_hook, "list available commands");
   m_getdescs_hook = new GetdescsHook(this);
   register_command("get_command_descriptions", "get_command_descriptions",
   m_getdescs_hook, "list available commands");
   create();
   add_cleanup_file(m_path.c_str());
   return true;
 }

首先是创建了管道，读取端的文件描述符记录在m_shutdown_rd_fd中，写入端的文件描述符记录在m_shutdown_wr_fd中。
从变量名字也可以看出，该文件描述符的作用是收取关闭信息。因为adminsocket一旦创建，必须能够通知到该线程及时退出。
退出的事情会写入管道的写入端，而线程会通过多路复用接口，监听读取端，一旦发现m_shutdown_rd_fd中读出内容，线程就知道，可以退出了。
按下不表。

AdminSocket最重要的是监听发过来的请求，它是用socket来实现的，初始化在bind_and_listen 函数：

 std::string AdminSocket::bind_and_listen(const std::string &sock_path, int *fd)
 {
   ldout(m_cct, 5) << "bind_and_listen " << sock_path << dendl;

   struct sockaddr_un address;
   if (sock_path.size() > sizeof(address.sun_path) - 1) {
     ostringstream oss;
     oss << "AdminSocket::bind_and_listen: "
     << "The UNIX domain socket path " << sock_path << " is too long! The "
     << "maximum length on this system is "
     << (sizeof(address.sun_path) - 1);
     return oss.str();
   }
   int sock_fd = socket(PF_UNIX, SOCK_STREAM, 0);
   if (sock_fd < 0) {
     int err = errno;
     ostringstream oss;
     oss << "AdminSocket::bind_and_listen: "
     << "failed to create socket: " << cpp_strerror(err);
     return oss.str();
   }
   int r = fcntl(sock_fd, F_SETFD, FD_CLOEXEC);
   if (r < 0) {
     r = errno;
     TEMP_FAILURE_RETRY(::close(sock_fd));
     ostringstream oss;
     oss << "AdminSocket::bind_and_listen: failed to fcntl on socket: " << cpp_strerror(r);
     return oss.str();
   }
   memset(&address, 0, sizeof(struct sockaddr_un));
   address.sun_family = AF_UNIX;
   snprintf(address.sun_path, sizeof(address.sun_path),
      "%s", sock_path.c_str());
   if (bind(sock_fd, (struct sockaddr*)&address,
      sizeof(struct sockaddr_un)) != 0) {
     int err = errno;
     if (err == EADDRINUSE) {
       AdminSocketClient client(sock_path);
       bool ok;
       client.ping(&ok);
       if (ok) {
     ldout(m_cct, 20) << "socket " << sock_path << " is in use" << dendl;
     err = EEXIST;
       } else {
     ldout(m_cct, 20) << "unlink stale file " << sock_path << dendl;
     TEMP_FAILURE_RETRY(unlink(sock_path.c_str()));
     if (bind(sock_fd, (struct sockaddr*)&address,
          sizeof(struct sockaddr_un)) == 0) {
      err = 0;
     } else {
      err = errno;
     }
       }
     }
     if (err != 0) {
       ostringstream oss;
       oss << "AdminSocket::bind_and_listen: "
      << "failed to bind the UNIX domain socket to '" << sock_path
      << "': " << cpp_strerror(err);
       close(sock_fd);
       return oss.str();
     }
   }
   if (listen(sock_fd, 5) != 0) {
     int err = errno;
     ostringstream oss;
     oss << "AdminSocket::bind_and_listen: "
      << "failed to listen to socket: " << cpp_strerror(err);
     close(sock_fd);
     TEMP_FAILURE_RETRY(unlink(sock_path.c_str()));
     return oss.str();
   }
   *fd = sock_fd;
   return "";
 }

这个函数做的事情并不难理解，做的事情比较老套：
1 创建一个socket，该函数需要一个入参，指定socket路径名：
2  bind
3 listen
而传入的路径名是这个

 root@test3:/var/run/ceph# ceph daemon /var/run/ceph/ceph-mds.bfudz.asok config get admin_socket
 { "admin_socket": "\/var\/run\/ceph\/ceph-mds.bfudz.asok"}

socket建好之后，因为线程还没有创建，所以至今还没有accept接口 。然后我们再次回到init函数，该函数又注册了几个函数：

 m_version_hook = new VersionHook;
   register_command("0", "0", m_version_hook, "");
   register_command("version", "version", m_version_hook, "get ceph version");
   register_command("git_version", "git_version", m_version_hook, "get git sha1");
   m_help_hook = new HelpHook(this);
   register_command("help", "help", m_help_hook, "list available commands");
   m_getdescs_hook = new GetdescsHook(this);
   register_command("get_command_descriptions", "get_command_descriptions",
   m_getdescs_hook, "list available commands");

这几个函数的重要性并不大，基本是用来查版本信息的，如下所示，按下不表。

 root@test3:/var/run/ceph# ceph daemon /var/run/ceph/ceph-mds.bfudz.asok version
 {"version":"0.67.9-222-g014b35f"}
 root@test3:/var/run/ceph# ceph daemon /var/run/ceph/ceph-mds.bfudz.asok git_version
 {"git_version":"014b35fc1ee0a1ad1f699a3705f3481a88614d36"}
 root@test3:/var/run/ceph#

init函数最后调用了create函数。create函数是老朋友，前面分析Log的时候已经提到，对于Thread这个类，做的事情无非就是创建线程。 关键内容是，线程执行的函数是哪个？  
和Log一样，是entry函数。AdminSocket类也有entry函数，该函数是AdminSocket 对应线程指定的函数：

 void* AdminSocket::entry()
 {
   ldout(m_cct, 5) << "entry start" << dendl;
   while (true) {
     struct pollfd fds[2];
     memset(fds, 0, sizeof(fds));
     fds[0].fd = m_sock_fd;
     fds[0].events = POLLIN | POLLRDBAND;
     fds[1].fd = m_shutdown_rd_fd;
     fds[1].events = POLLIN | POLLRDBAND;
     int ret = poll(fds, 2, -1);
     if (ret < 0) {
       int err = errno;
       if (err == EINTR) {
 continue;
       }
       lderr(m_cct) << "AdminSocket: poll(2) error: '"
   << cpp_strerror(err) << dendl;
       return PFL_FAIL;
     }
     if (fds[0].revents & POLLIN) {
       // Send out some data
       do_accept();
     }
     if (fds[1].revents & POLLIN) {
       // Parent wants us to shut down
       return PFL_SUCCESS;
     }
   }
   ldout(m_cct, 5) << "entry exit" << dendl;
 }

这个线程函数比较简单，它监听socket fd和管道的读取端。
1 管道的读取端负责管理何时退出
2 socket fd 负责监听用户发过来的指令。

处理用户发过来的命令，是do_accept函数干的事情：

 bool AdminSocket::do_accept()
 {
   struct sockaddr_un address;
   socklen_t address_length = sizeof(address);
   ldout(m_cct, 30) << "AdminSocket: calling accept" << dendl;
   int connection_fd = accept(m_sock_fd, (struct sockaddr*) &address,
              &address_length);
   ldout(m_cct, 30) << "AdminSocket: finished accept" << dendl;
   if (connection_fd < 0) {
     int err = errno;
     lderr(m_cct) << "AdminSocket: do_accept error: '"
              << cpp_strerror(err) << dendl;
     return false;
   }

   char cmd[1024];
   int pos = 0;
   string c;
   while (1) {
     int ret = safe_read(connection_fd, &cmd[pos], 1);
     if (ret <= 0) {
       lderr(m_cct) << "AdminSocket: error reading request code: "
          << cpp_strerror(ret) << dendl;
       close(connection_fd);
       return false;
     }
     //ldout(m_cct, 0) << "AdminSocket read byte " << (int)cmd[pos] << " pos " << pos << dendl;
     if (cmd[0] == '\0') {
       // old protocol: __be32
       if (pos == 3 && cmd[0] == '\0') {
     switch (cmd[3]) {
     case 0:
      c = "0";
      break;
     case 1:
      c = "perfcounters_dump";
      break;
     case 2:
      c = "perfcounters_schema";
      break;
     default:
      c = "foo";
      break;
     }
     break;
       }
     } else {
       // new protocol: null or \n terminated string
       if (cmd[pos] == '\n' || cmd[pos] == '\0') {
     cmd[pos] = '\0';
     c = cmd;
     break;
       }
     }
     pos++;
   }

   bool rval = false;

   map<string, cmd_vartype> cmdmap;
   string format;
   vector<string> cmdvec;
   stringstream errss;
   cmdvec.push_back(cmd);
   if (!cmdmap_from_json(cmdvec, &cmdmap, errss)) {
     ldout(m_cct, 0) << "AdminSocket: " << errss << dendl;
     return false;
   }
   cmd_getval(m_cct, cmdmap, "format", format);
   if (format != "json" && format != "json-pretty" &&
       format != "xml" && format != "xml-pretty")
     format = "json-pretty";
   cmd_getval(m_cct, cmdmap, "prefix", c);

   string firstword;
   if (c.find(" ") == string::npos)
     firstword = c;
   else
     firstword = c.substr(0, c.find(" "));

   m_lock.Lock();
   map<string,AdminSocketHook*>::iterator p;
   string match = c;
   while (match.size()) {
     p = m_hooks.find(match);
     if (p != m_hooks.end())
       break;

     // drop right-most word
     size_t pos = match.rfind(' ');
     if (pos == std::string::npos) {
       match.clear(); // we fail
       break;
     } else {
       match.resize(pos);
     }
   }

   bufferlist out;
   if (p == m_hooks.end()) {
     lderr(m_cct) << "AdminSocket: request '" << c << "' not defined" << dendl;
   } else {
     string args;
     if (match != c)
       args = c.substr(match.length() + 1);
     bool success = p->second->call(match, cmdmap, format, out);
     if (!success) {
       ldout(m_cct, 0) << "AdminSocket: request '" << match << "' args '" << args
          << "' to " << p->second << " failed" << dendl;
       out.append("failed");
     } else {
       ldout(m_cct, 5) << "AdminSocket: request '" << match << "' '" << args
          << "' to " << p->second
          << " returned " << out.length() << " bytes" << dendl;
     }
     uint32_t len = htonl(out.length());
     int ret = safe_write(connection_fd, &len, sizeof(len));
     if (ret < 0) {
       lderr(m_cct) << "AdminSocket: error writing response length "
          << cpp_strerror(ret) << dendl;
     } else {
       if (out.write_fd(connection_fd) >= 0)
     rval = true;
     }
   }
   m_lock.Unlock();

   TEMP_FAILURE_RETRY(close(connection_fd));
   return rval;
 }

这个函数有点长，但是并不复杂。简单说，如果有个client尝试 connect ，该线程就poll就会感知到，然后进入do_accept函数。
do_accept首先执行accept，和client 搭上线，然后开始通信。 

safe_read负责 读取客户发过来的指令。前面已经提到过，AdminSocket支持的命令是有限的，初始化之前都已注册过了。
如果client 发来的指令时注册过的指令，就见招拆招，返回相应的结果给客户端。

每一个命令的字符串，都是和一个AdminSocketHook 的类型关联的，但是一个AdminSocketHook可以对应多个command

 std::map<std::string,AdminSocketHook*> m_hooks

比如说 config show  / config get   / perf dump对应的多事 CephContextHook，前面已经提到过。

见招拆招的函数，就记录在对应的Hook上：

 class CephContextHook : public AdminSocketHook {
   CephContext *m_cct;
 public:
   CephContextHook(CephContext *cct) : m_cct(cct) {}
   bool call(std::string command, cmdmap_t& cmdmap, std::string format,
    bufferlist& out) {
     m_cct->do_command(command, cmdmap, format, &out);
     return true;
   }
 };

do_accept函数中黄色的一行，具体是实现，就是对应Hook的call函数，对于CephContextHook，就是这个类的call方法。

下面我们看下CephContextHook的call方法：即它的do_command函数：

 void CephContext::do_command(std::string command, cmdmap_t& cmdmap,
              std::string format, bufferlist *out)
 {
   Formatter *f = new_formatter(format);
   if (!f)
     f = new_formatter("json-pretty");
   stringstream ss;
   for (cmdmap_t::iterator it = cmdmap.begin(); it != cmdmap.end(); ++it) {
     if (it->first != "prefix") {
       ss << it->first << ":" << cmd_vartype_stringify(it->second) << " ";
     }
   }
   lgeneric_dout(this, 1) << "do_command '" << command << "' '"
              << ss.str() << dendl;
   if (command == "perfcounters_dump" || command == "1" ||
       command == "perf dump") {
     _perf_counters_collection->dump_formatted(f, false);
   }
   else if (command == "perfcounters_schema" || command == "2" ||
     command == "perf schema") {
     _perf_counters_collection->dump_formatted(f, true);
   }
   else {
     f->open_object_section(command.c_str());
     if (command == "config show") {
       _conf->show_config(f);
     }
     else if (command == "config set") {
       std::string var;
       std::vector<std::string> val;

       if (!(cmd_getval(this, cmdmap, "var", var)) ||
           !(cmd_getval(this, cmdmap, "val", val))) {
         f->dump_string("error", "syntax error: 'config set '");
       } else {
     // val may be multiple words
     string valstr = str_join(val, " ");
         int r = _conf->set_val(var.c_str(), valstr.c_str());
         if (r < 0) {
           f->dump_stream("error") << "error setting '" << var << "' to '" << valstr << "': " << cpp_strerror(r);
         } else {
           ostringstream ss;
           _conf->apply_changes(&ss);
           f->dump_string("success", ss.str());
         }
       }
     } else if (command == "config get") {
       std::string var;
       if (!cmd_getval(this, cmdmap, "var", var)) {
     f->dump_string("error", "syntax error: 'config get '");
       } else {
     char buf[4096];
     memset(buf, 0, sizeof(buf));
     char *tmp = buf;
     int r = _conf->get_val(var.c_str(), &tmp, sizeof(buf));
     if (r < 0) {
      f->dump_stream("error") << "error getting '" << var << "': " << cpp_strerror(r);
     } else {
      f->dump_string(var.c_str(), buf);
     }
       }
     } else if (command == "log flush") {
       _log->flush();
     }
     else if (command == "log dump") {
       _log->dump_recent();
     }
     else if (command == "log reopen") {
       _log->reopen_log_file();
     }
     else {
       assert(0 == "registered under wrong command?");
     }
     f->close_section();
   }
   f->flush(*out);
   delete f;
   lgeneric_dout(this, 1) << "do_command '" << command << "' '" << ss.str()
          << "result is " << out->length() << " bytes" << dendl;
 };

  毫不意外，前面注册的每一个方法，在do_command函数中都做了处理。