ceph/src/msg/async/AsyncMessenger.cc

   1 // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
   2 // vim: ts=8 sw=2 smarttab
   3 /*
   4  * Ceph - scalable distributed file system
   5  *
   6  * Copyright (C) 2014 UnitedStack <haomai@unitedstack.com>
   7  *
   8  * Author: Haomai Wang <haomaiwang@gmail.com>
   9  *
  10  * This is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU Lesser General Public
  12  * License version 2.1, as published by the Free Software
  13  * Foundation.  See file COPYING.
  14  *
  15  */
  16
  17 #include "acconfig.h"
  18
  19 #include <iostream>
  20 #include <fstream>
  21
  22 #include "AsyncMessenger.h"
  23
  24 #include "common/config.h"
  25 #include "common/Timer.h"
  26 #include "common/errno.h"
  27
  28 #include "messages/MOSDOp.h"
  29 #include "messages/MOSDOpReply.h"
  30 #include "common/EventTrace.h"
  31
  32 #define dout_subsys ceph_subsys_ms
  33 #undef dout_prefix
  34 #define dout_prefix _prefix(_dout, this)
  35 static ostream& _prefix(std::ostream *_dout, AsyncMessenger *m) {
  36   return *_dout << "-- " << m->get_myaddr() << " ";
  37 }
  38
  39 static ostream& _prefix(std::ostream *_dout, Processor *p) {
  40   return *_dout << " Processor -- ";
  41 }
  42
  43
  44 /*******************
  45  * Processor
  46  */
  47
  48 class Processor::C_processor_accept : public EventCallback {
  49   Processor *pro;
  50
  51  public:
  52   explicit C_processor_accept(Processor *p): pro(p) {}
  53   void do_request(int id) override {
  54     pro->accept();
  55   }
  56 };
  57
  58 Processor::Processor(AsyncMessenger *r, Worker *w, CephContext *c)
  59   : msgr(r), net(c), worker(w),
  60     listen_handler(new C_processor_accept(this)) {}
  61
  62 int Processor::bind(const entity_addr_t &bind_addr,
  63                     const set<int>& avoid_ports,
  64                     entity_addr_t* bound_addr)
  65 {
  66   const md_config_t *conf = msgr->cct->_conf;
  67   // bind to a socket
  68   ldout(msgr->cct, 10) << __func__ << dendl;
  69
  70   int family;
  71   switch (bind_addr.get_family()) {
  72     case AF_INET:
  73     case AF_INET6:
  74       family = bind_addr.get_family();
  75       break;
  76
  77     default:
  78       // bind_addr is empty
  79       family = conf->ms_bind_ipv6 ? AF_INET6 : AF_INET;
  80   }
  81
  82   SocketOptions opts;
  83   opts.nodelay = msgr->cct->_conf->ms_tcp_nodelay;
  84   opts.rcbuf_size = msgr->cct->_conf->ms_tcp_rcvbuf;
  85
  86   // use whatever user specified (if anything)
  87   entity_addr_t listen_addr = bind_addr;
  88   if (listen_addr.get_type() == entity_addr_t::TYPE_NONE) {
  89     listen_addr.set_type(entity_addr_t::TYPE_LEGACY);
  90   }
  91   listen_addr.set_family(family);
  92
  93   /* bind to port */
  94   int r = -1;
  95
  96   for (int i = 0; i < conf->ms_bind_retry_count; i++) {
  97     if (i > 0) {
  98       lderr(msgr->cct) << __func__ << " was unable to bind. Trying again in "
  99                        << conf->ms_bind_retry_delay << " seconds " << dendl;
 100       sleep(conf->ms_bind_retry_delay);
 101     }
 102
 103     if (listen_addr.get_port()) {
 104       worker->center.submit_to(worker->center.get_id(), [this, &listen_addr, &opts, &r]() {
 105         r = worker->listen(listen_addr, opts, &listen_socket);
 106       }, false);
 107       if (r < 0) {
 108         lderr(msgr->cct) << __func__ << " unable to bind to " << listen_addr
 109                          << ": " << cpp_strerror(r) << dendl;
 110         continue;
 111       }
 112     } else {
 113       // try a range of ports
 114       for (int port = msgr->cct->_conf->ms_bind_port_min; port <= msgr->cct->_conf->ms_bind_port_max; port++) {
 115         if (avoid_ports.count(port))
 116           continue;
 117
 118         listen_addr.set_port(port);
 119         worker->center.submit_to(worker->center.get_id(), [this, &listen_addr, &opts, &r]() {
 120           r = worker->listen(listen_addr, opts, &listen_socket);
 121         }, false);
 122         if (r == 0)
 123           break;
 124       }
 125       if (r < 0) {
 126         lderr(msgr->cct) << __func__ << " unable to bind to " << listen_addr
 127                          << " on any port in range " << msgr->cct->_conf->ms_bind_port_min
 128                          << "-" << msgr->cct->_conf->ms_bind_port_max << ": "
 129                          << cpp_strerror(r) << dendl;
 130         listen_addr.set_port(0); // Clear port before retry, otherwise we shall fail again.
 131         continue;
 132       }
 133       ldout(msgr->cct, 10) << __func__ << " bound on random port " << listen_addr << dendl;
 134     }
 135     if (r == 0)
 136       break;
 137   }
 138   // It seems that binding completely failed, return with that exit status
 139   if (r < 0) {
 140     lderr(msgr->cct) << __func__ << " was unable to bind after " << conf->ms_bind_retry_count
 141                      << " attempts: " << cpp_strerror(r) << dendl;
 142     return r;
 143   }
 144
 145   ldout(msgr->cct, 10) << __func__ << " bound to " << listen_addr << dendl;
 146   *bound_addr = listen_addr;
 147   return 0;
 148 }
 149
 150 void Processor::start()
 151 {
 152   ldout(msgr->cct, 1) << __func__ << dendl;
 153
 154   // start thread
 155   if (listen_socket) {
 156     worker->center.submit_to(worker->center.get_id(), [this]() {
 157       worker->center.create_file_event(listen_socket.fd(), EVENT_READABLE, listen_handler); }, false);
 158   }
 159 }
 160
 161 void Processor::accept()
 162 {
 163   ldout(msgr->cct, 10) << __func__ << " listen_fd=" << listen_socket.fd() << dendl;
 164   SocketOptions opts;
 165   opts.nodelay = msgr->cct->_conf->ms_tcp_nodelay;
 166   opts.rcbuf_size = msgr->cct->_conf->ms_tcp_rcvbuf;
 167   opts.priority = msgr->get_socket_priority();
 168   while (true) {
 169     entity_addr_t addr;
 170     ConnectedSocket cli_socket;
 171     Worker *w = worker;
 172     if (!msgr->get_stack()->support_local_listen_table())
 173       w = msgr->get_stack()->get_worker();
 174     int r = listen_socket.accept(&cli_socket, opts, &addr, w);
 175     if (r == 0) {
 176       ldout(msgr->cct, 10) << __func__ << " accepted incoming on sd " << cli_socket.fd() << dendl;
 177
 178       msgr->add_accept(w, std::move(cli_socket), addr);
 179       continue;
 180     } else {
 181       if (r == -EINTR) {
 182         continue;
 183       } else if (r == -EAGAIN) {
 184         break;
 185       } else if (r == -EMFILE || r == -ENFILE) {
 186         lderr(msgr->cct) << __func__ << " open file descriptions limit reached sd = " << listen_socket.fd()
 187                          << " errno " << r << " " << cpp_strerror(r) << dendl;
 188         break;
 189       } else if (r == -ECONNABORTED) {
 190         ldout(msgr->cct, 0) << __func__ << " it was closed because of rst arrived sd = " << listen_socket.fd()
 191                             << " errno " << r << " " << cpp_strerror(r) << dendl;
 192         continue;
 193       } else {
 194         lderr(msgr->cct) << __func__ << " no incoming connection?"
 195                          << " errno " << r << " " << cpp_strerror(r) << dendl;
 196         break;
 197       }
 198     }
 199   }
 200 }
 201
 202 void Processor::stop()
 203 {
 204   ldout(msgr->cct,10) << __func__ << dendl;
 205
 206   if (listen_socket) {
 207     worker->center.submit_to(worker->center.get_id(), [this]() {
 208       worker->center.delete_file_event(listen_socket.fd(), EVENT_READABLE);
 209       listen_socket.abort_accept();
 210     }, false);
 211   }
 212 }
 213
 214
 215 struct StackSingleton {
 216   CephContext *cct;
 217   std::shared_ptr<NetworkStack> stack;
 218
 219   StackSingleton(CephContext *c): cct(c) {}
 220   void ready(std::string &type) {
 221     if (!stack)
 222       stack = NetworkStack::create(cct, type);
 223   }
 224   ~StackSingleton() {
 225     stack->stop();
 226   }
 227 };
 228
 229
 230 class C_handle_reap : public EventCallback {
 231   AsyncMessenger *msgr;
 232
 233   public:
 234   explicit C_handle_reap(AsyncMessenger *m): msgr(m) {}
 235   void do_request(int id) override {
 236     // judge whether is a time event
 237     msgr->reap_dead();
 238   }
 239 };
 240
 241 /*******************
 242  * AsyncMessenger
 243  */
 244
 245 AsyncMessenger::AsyncMessenger(CephContext *cct, entity_name_t name,
 246                                const std::string &type, string mname, uint64_t _nonce)
 247   : SimplePolicyMessenger(cct, name,mname, _nonce),
 248     dispatch_queue(cct, this, mname),
 249     lock("AsyncMessenger::lock"),
 250     nonce(_nonce), need_addr(true), did_bind(false),
 251     global_seq(0), deleted_lock("AsyncMessenger::deleted_lock"),
 252     cluster_protocol(0), stopped(true)
 253 {
 254   std::string transport_type = "posix";
 255   if (type.find("rdma") != std::string::npos)
 256     transport_type = "rdma";
 257   else if (type.find("dpdk") != std::string::npos)
 258     transport_type = "dpdk";
 259
 260   ceph_spin_init(&global_seq_lock);
 261   StackSingleton *single;
 262   cct->lookup_or_create_singleton_object<StackSingleton>(single, "AsyncMessenger::NetworkStack::"+transport_type);
 263   single->ready(transport_type);
 264   stack = single->stack.get();
 265   stack->start();
 266   local_worker = stack->get_worker();
 267   local_connection = new AsyncConnection(cct, this, &dispatch_queue, local_worker);
 268   init_local_connection();
 269   reap_handler = new C_handle_reap(this);
 270   unsigned processor_num = 1;
 271   if (stack->support_local_listen_table())
 272     processor_num = stack->get_num_worker();
 273   for (unsigned i = 0; i < processor_num; ++i)
 274     processors.push_back(new Processor(this, stack->get_worker(i), cct));
 275 }
 276
 277 /**
 278  * Destroy the AsyncMessenger. Pretty simple since all the work is done
 279  * elsewhere.
 280  */
 281 AsyncMessenger::~AsyncMessenger()
 282 {
 283   delete reap_handler;
 284   assert(!did_bind); // either we didn't bind or we shut down the Processor
 285   local_connection->mark_down();
 286   for (auto &&p : processors)
 287     delete p;
 288 }
 289
 290 void AsyncMessenger::ready()
 291 {
 292   ldout(cct,10) << __func__ << " " << get_myaddr() << dendl;
 293
 294   stack->ready();
 295   if (pending_bind) {
 296     int err = bind(pending_bind_addr);
 297     if (err) {
 298       lderr(cct) << __func__ << " postponed bind failed" << dendl;
 299       ceph_abort();
 300     }
 301   }
 302
 303   Mutex::Locker l(lock);
 304   for (auto &&p : processors)
 305     p->start();
 306   dispatch_queue.start();
 307 }
 308
 309 int AsyncMessenger::shutdown()
 310 {
 311   ldout(cct,10) << __func__ << " " << get_myaddr() << dendl;
 312
 313   // done!  clean up.
 314   for (auto &&p : processors)
 315     p->stop();
 316   mark_down_all();
 317   // break ref cycles on the loopback connection
 318   local_connection->set_priv(NULL);
 319   did_bind = false;
 320   lock.Lock();
 321   stop_cond.Signal();
 322   stopped = true;
 323   lock.Unlock();
 324   stack->drain();
 325   return 0;
 326 }
 327
 328
 329 int AsyncMessenger::bind(const entity_addr_t &bind_addr)
 330 {
 331   lock.Lock();
 332
 333   if (!pending_bind && started) {
 334     ldout(cct,10) << __func__ << " already started" << dendl;
 335     lock.Unlock();
 336     return -1;
 337   }
 338
 339   ldout(cct,10) << __func__ << " bind " << bind_addr << dendl;
 340
 341   if (!stack->is_ready()) {
 342     ldout(cct, 10) << __func__ << " Network Stack is not ready for bind yet - postponed" << dendl;
 343     pending_bind_addr = bind_addr;
 344     pending_bind = true;
 345     lock.Unlock();
 346     return 0;
 347   }
 348
 349   lock.Unlock();
 350
 351   // bind to a socket
 352   set<int> avoid_ports;
 353   entity_addr_t bound_addr;
 354   unsigned i = 0;
 355   for (auto &&p : processors) {
 356     int r = p->bind(bind_addr, avoid_ports, &bound_addr);
 357     if (r) {
 358       // Note: this is related to local tcp listen table problem.
 359       // Posix(default kernel implementation) backend shares listen table
 360       // in the kernel, so all threads can use the same listen table naturally
 361       // and only one thread need to bind. But other backends(like dpdk) uses local
 362       // listen table, we need to bind/listen tcp port for each worker. So if the
 363       // first worker failed to bind, it could be think the normal error then handle
 364       // it, like port is used case. But if the first worker successfully to bind
 365       // but the second worker failed, it's not expected and we need to assert
 366       // here
 367       assert(i == 0);
 368       return r;
 369     }
 370     ++i;
 371   }
 372   _finish_bind(bind_addr, bound_addr);
 373   return 0;
 374 }
 375
 376 int AsyncMessenger::rebind(const set<int>& avoid_ports)
 377 {
 378   ldout(cct,1) << __func__ << " rebind avoid " << avoid_ports << dendl;
 379   assert(did_bind);
 380
 381   for (auto &&p : processors)
 382     p->stop();
 383   mark_down_all();
 384
 385   // adjust the nonce; we want our entity_addr_t to be truly unique.
 386   nonce += 1000000;
 387   ldout(cct, 10) << __func__ << " new nonce " << nonce
 388                  << " and inst " << get_myinst() << dendl;
 389
 390   entity_addr_t bound_addr;
 391   entity_addr_t bind_addr = get_myaddr();
 392   bind_addr.set_port(0);
 393   set<int> new_avoid(avoid_ports);
 394   new_avoid.insert(bind_addr.get_port());
 395   ldout(cct, 10) << __func__ << " will try " << bind_addr
 396                  << " and avoid ports " << new_avoid << dendl;
 397   unsigned i = 0;
 398   for (auto &&p : processors) {
 399     int r = p->bind(bind_addr, avoid_ports, &bound_addr);
 400     if (r) {
 401       assert(i == 0);
 402       return r;
 403     }
 404     ++i;
 405   }
 406   _finish_bind(bind_addr, bound_addr);
 407   for (auto &&p : processors) {
 408     p->start();
 409   }
 410   return 0;
 411 }
 412
 413 int AsyncMessenger::client_bind(const entity_addr_t &bind_addr)
 414 {
 415   if (!cct->_conf->ms_bind_before_connect)
 416     return 0;
 417   Mutex::Locker l(lock);
 418   if (did_bind) {
 419     assert(my_inst.addr == bind_addr);
 420     return 0;
 421   }
 422   if (started) {
 423     ldout(cct, 10) << __func__ << " already started" << dendl;
 424     return -1;
 425   }
 426   ldout(cct, 10) << __func__ << " " << bind_addr << dendl;
 427
 428   set_myaddr(bind_addr);
 429   return 0;
 430 }
 431
 432 void AsyncMessenger::_finish_bind(const entity_addr_t& bind_addr,
 433                                   const entity_addr_t& listen_addr)
 434 {
 435   set_myaddr(bind_addr);
 436   if (bind_addr != entity_addr_t())
 437     learned_addr(bind_addr);
 438
 439   if (get_myaddr().get_port() == 0) {
 440     set_myaddr(listen_addr);
 441   }
 442   entity_addr_t addr = get_myaddr();
 443   addr.set_nonce(nonce);
 444   set_myaddr(addr);
 445
 446   init_local_connection();
 447
 448   ldout(cct,1) << __func__ << " bind my_inst.addr is " << get_myaddr() << dendl;
 449   did_bind = true;
 450 }
 451
 452 int AsyncMessenger::start()
 453 {
 454   lock.Lock();
 455   ldout(cct,1) << __func__ << " start" << dendl;
 456
 457   // register at least one entity, first!
 458   assert(my_inst.name.type() >= 0);
 459
 460   assert(!started);
 461   started = true;
 462   stopped = false;
 463
 464   if (!did_bind) {
 465     my_inst.addr.nonce = nonce;
 466     _init_local_connection();
 467   }
 468
 469   lock.Unlock();
 470   return 0;
 471 }
 472
 473 void AsyncMessenger::wait()
 474 {
 475   lock.Lock();
 476   if (!started) {
 477     lock.Unlock();
 478     return;
 479   }
 480   if (!stopped)
 481     stop_cond.Wait(lock);
 482
 483   lock.Unlock();
 484
 485   dispatch_queue.shutdown();
 486   if (dispatch_queue.is_started()) {
 487     ldout(cct, 10) << __func__ << ": waiting for dispatch queue" << dendl;
 488     dispatch_queue.wait();
 489     dispatch_queue.discard_local();
 490     ldout(cct, 10) << __func__ << ": dispatch queue is stopped" << dendl;
 491   }
 492
 493   // close all connections
 494   shutdown_connections(false);
 495   stack->drain();
 496
 497   ldout(cct, 10) << __func__ << ": done." << dendl;
 498   ldout(cct, 1) << __func__ << " complete." << dendl;
 499   started = false;
 500 }
 501
 502 void AsyncMessenger::add_accept(Worker *w, ConnectedSocket cli_socket, entity_addr_t &addr)
 503 {
 504   lock.Lock();
 505   AsyncConnectionRef conn = new AsyncConnection(cct, this, &dispatch_queue, w);
 506   conn->accept(std::move(cli_socket), addr);
 507   accepting_conns.insert(conn);
 508   lock.Unlock();
 509 }
 510
 511 AsyncConnectionRef AsyncMessenger::create_connect(const entity_addr_t& addr, int type)
 512 {
 513   assert(lock.is_locked());
 514   assert(addr != my_inst.addr);
 515
 516   ldout(cct, 10) << __func__ << " " << addr
 517       << ", creating connection and registering" << dendl;
 518
 519   // create connection
 520   Worker *w = stack->get_worker();
 521   AsyncConnectionRef conn = new AsyncConnection(cct, this, &dispatch_queue, w);
 522   conn->connect(addr, type);
 523   assert(!conns.count(addr));
 524   conns[addr] = conn;
 525   w->get_perf_counter()->inc(l_msgr_active_connections);
 526
 527   return conn;
 528 }
 529
 530 ConnectionRef AsyncMessenger::get_connection(const entity_inst_t& dest)
 531 {
 532   Mutex::Locker l(lock);
 533   if (my_inst.addr == dest.addr) {
 534     // local
 535     return local_connection;
 536   }
 537
 538   AsyncConnectionRef conn = _lookup_conn(dest.addr);
 539   if (conn) {
 540     ldout(cct, 10) << __func__ << " " << dest << " existing " << conn << dendl;
 541   } else {
 542     conn = create_connect(dest.addr, dest.name.type());
 543     ldout(cct, 10) << __func__ << " " << dest << " new " << conn << dendl;
 544   }
 545
 546   return conn;
 547 }
 548
 549 ConnectionRef AsyncMessenger::get_loopback_connection()
 550 {
 551   return local_connection;
 552 }
 553
 554 int AsyncMessenger::_send_message(Message *m, const entity_inst_t& dest)
 555 {
 556   FUNCTRACE();
 557   assert(m);
 558
 559   if (m->get_type() == CEPH_MSG_OSD_OP)
 560     OID_EVENT_TRACE(((MOSDOp *)m)->get_oid().name.c_str(), "SEND_MSG_OSD_OP");
 561   else if (m->get_type() == CEPH_MSG_OSD_OPREPLY)
 562     OID_EVENT_TRACE(((MOSDOpReply *)m)->get_oid().name.c_str(), "SEND_MSG_OSD_OP_REPLY");
 563
 564   ldout(cct, 1) << __func__ << "--> " << dest.name << " "
 565       << dest.addr << " -- " << *m << " -- ?+"
 566       << m->get_data().length() << " " << m << dendl;
 567
 568   if (dest.addr == entity_addr_t()) {
 569     ldout(cct,0) << __func__ <<  " message " << *m
 570         << " with empty dest " << dest.addr << dendl;
 571     m->put();
 572     return -EINVAL;
 573   }
 574
 575   AsyncConnectionRef conn = _lookup_conn(dest.addr);
 576   submit_message(m, conn, dest.addr, dest.name.type());
 577   return 0;
 578 }
 579
 580 void AsyncMessenger::submit_message(Message *m, AsyncConnectionRef con,
 581                                     const entity_addr_t& dest_addr, int dest_type)
 582 {
 583   if (cct->_conf->ms_dump_on_send) {
 584     m->encode(-1, MSG_CRC_ALL);
 585     ldout(cct, 0) << __func__ << "submit_message " << *m << "\n";
 586     m->get_payload().hexdump(*_dout);
 587     if (m->get_data().length() > 0) {
 588       *_dout << " data:\n";
 589       m->get_data().hexdump(*_dout);
 590     }
 591     *_dout << dendl;
 592     m->clear_payload();
 593   }
 594
 595   // existing connection?
 596   if (con) {
 597     con->send_message(m);
 598     return ;
 599   }
 600
 601   // local?
 602   if (my_inst.addr == dest_addr) {
 603     // local
 604     local_connection->send_message(m);
 605     return ;
 606   }
 607
 608   // remote, no existing connection.
 609   const Policy& policy = get_policy(dest_type);
 610   if (policy.server) {
 611     ldout(cct, 20) << __func__ << " " << *m << " remote, " << dest_addr
 612         << ", lossy server for target type "
 613         << ceph_entity_type_name(dest_type) << ", no session, dropping." << dendl;
 614     m->put();
 615   } else {
 616     ldout(cct,20) << __func__ << " " << *m << " remote, " << dest_addr << ", new connection." << dendl;
 617     con = create_connect(dest_addr, dest_type);
 618     con->send_message(m);
 619   }
 620 }
 621
 622 /**
 623  * If my_inst.addr doesn't have an IP set, this function
 624  * will fill it in from the passed addr. Otherwise it does nothing and returns.
 625  */
 626 void AsyncMessenger::set_addr_unknowns(const entity_addr_t &addr)
 627 {
 628   Mutex::Locker l(lock);
 629   if (my_inst.addr.is_blank_ip()) {
 630     int port = my_inst.addr.get_port();
 631     my_inst.addr.u = addr.u;
 632     my_inst.addr.set_port(port);
 633     _init_local_connection();
 634   }
 635 }
 636
 637 void AsyncMessenger::set_addr(const entity_addr_t &addr)
 638 {
 639   Mutex::Locker l(lock);
 640   entity_addr_t t = addr;
 641   t.set_nonce(nonce);
 642   set_myaddr(t);
 643   _init_local_connection();
 644 }
 645
 646 void AsyncMessenger::shutdown_connections(bool queue_reset)
 647 {
 648   ldout(cct,1) << __func__ << " " << dendl;
 649   lock.Lock();
 650   for (set<AsyncConnectionRef>::iterator q = accepting_conns.begin();
 651        q != accepting_conns.end(); ++q) {
 652     AsyncConnectionRef p = *q;
 653     ldout(cct, 5) << __func__ << " accepting_conn " << p.get() << dendl;
 654     p->stop(queue_reset);
 655   }
 656   accepting_conns.clear();
 657
 658   while (!conns.empty()) {
 659     ceph::unordered_map<entity_addr_t, AsyncConnectionRef>::iterator it = conns.begin();
 660     AsyncConnectionRef p = it->second;
 661     ldout(cct, 5) << __func__ << " mark down " << it->first << " " << p << dendl;
 662     conns.erase(it);
 663     p->get_perf_counter()->dec(l_msgr_active_connections);
 664     p->stop(queue_reset);
 665   }
 666
 667   {
 668     Mutex::Locker l(deleted_lock);
 669     while (!deleted_conns.empty()) {
 670       set<AsyncConnectionRef>::iterator it = deleted_conns.begin();
 671       AsyncConnectionRef p = *it;
 672       ldout(cct, 5) << __func__ << " delete " << p << dendl;
 673       deleted_conns.erase(it);
 674     }
 675   }
 676   lock.Unlock();
 677 }
 678
 679 void AsyncMessenger::mark_down(const entity_addr_t& addr)
 680 {
 681   lock.Lock();
 682   AsyncConnectionRef p = _lookup_conn(addr);
 683   if (p) {
 684     ldout(cct, 1) << __func__ << " " << addr << " -- " << p << dendl;
 685     p->stop(true);
 686   } else {
 687     ldout(cct, 1) << __func__ << " " << addr << " -- connection dne" << dendl;
 688   }
 689   lock.Unlock();
 690 }
 691
 692 int AsyncMessenger::get_proto_version(int peer_type, bool connect) const
 693 {
 694   int my_type = my_inst.name.type();
 695
 696   // set reply protocol version
 697   if (peer_type == my_type) {
 698     // internal
 699     return cluster_protocol;
 700   } else {
 701     // public
 702     switch (connect ? peer_type : my_type) {
 703       case CEPH_ENTITY_TYPE_OSD: return CEPH_OSDC_PROTOCOL;
 704       case CEPH_ENTITY_TYPE_MDS: return CEPH_MDSC_PROTOCOL;
 705       case CEPH_ENTITY_TYPE_MON: return CEPH_MONC_PROTOCOL;
 706     }
 707   }
 708   return 0;
 709 }
 710
 711 void AsyncMessenger::learned_addr(const entity_addr_t &peer_addr_for_me)
 712 {
 713   // be careful here: multiple threads may block here, and readers of
 714   // my_inst.addr do NOT hold any lock.
 715
 716   // this always goes from true -> false under the protection of the
 717   // mutex.  if it is already false, we need not retake the mutex at
 718   // all.
 719   if (!need_addr)
 720     return ;
 721   lock.Lock();
 722   if (need_addr) {
 723     need_addr = false;
 724     entity_addr_t t = peer_addr_for_me;
 725     t.set_port(my_inst.addr.get_port());
 726     t.set_nonce(my_inst.addr.get_nonce());
 727     my_inst.addr = t;
 728     ldout(cct, 1) << __func__ << " learned my addr " << my_inst.addr << dendl;
 729     _init_local_connection();
 730   }
 731   lock.Unlock();
 732 }
 733
 734 int AsyncMessenger::reap_dead()
 735 {
 736   ldout(cct, 1) << __func__ << " start" << dendl;
 737   int num = 0;
 738
 739   Mutex::Locker l1(lock);
 740   Mutex::Locker l2(deleted_lock);
 741
 742   while (!deleted_conns.empty()) {
 743     auto it = deleted_conns.begin();
 744     AsyncConnectionRef p = *it;
 745     ldout(cct, 5) << __func__ << " delete " << p << dendl;
 746     auto conns_it = conns.find(p->peer_addr);
 747     if (conns_it != conns.end() && conns_it->second == p)
 748       conns.erase(conns_it);
 749     accepting_conns.erase(p);
 750     deleted_conns.erase(it);
 751     ++num;
 752   }
 753
 754   return num;
 755 }