ceph/src/msg/async/AsyncMessenger.h

   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 #ifndef CEPH_ASYNCMESSENGER_H
  18 #define CEPH_ASYNCMESSENGER_H
  19
  20 #include <map>
  21
  22 #include "include/types.h"
  23 #include "include/xlist.h"
  24 #include "include/spinlock.h"
  25 #include "include/unordered_map.h"
  26 #include "include/unordered_set.h"
  27
  28 #include "common/ceph_mutex.h"
  29 #include "common/Cond.h"
  30 #include "common/Thread.h"
  31
  32 #include "msg/SimplePolicyMessenger.h"
  33 #include "msg/DispatchQueue.h"
  34 #include "AsyncConnection.h"
  35 #include "Event.h"
  36
  37 #include "include/ceph_assert.h"
  38
  39 class AsyncMessenger;
  40
  41 /**
  42  * If the Messenger binds to a specific address, the Processor runs
  43  * and listens for incoming connections.
  44  */
  45 class Processor {
  46   AsyncMessenger *msgr;
  47   ceph::NetHandler net;
  48   Worker *worker;
  49   std::vector<ServerSocket> listen_sockets;
  50   EventCallbackRef listen_handler;
  51
  52   class C_processor_accept;
  53
  54  public:
  55   Processor(AsyncMessenger *r, Worker *w, CephContext *c);
  56   ~Processor() { delete listen_handler; };
  57
  58   void stop();
  59   int bind(const entity_addrvec_t &bind_addrs,
  60            const std::set<int>& avoid_ports,
  61            entity_addrvec_t* bound_addrs);
  62   void start();
  63   void accept();
  64 };
  65
  66 /*
  67  * AsyncMessenger is represented for maintaining a set of asynchronous connections,
  68  * it may own a bind address and the accepted connections will be managed by
  69  * AsyncMessenger.
  70  *
  71  */
  72
  73 class AsyncMessenger : public SimplePolicyMessenger {
  74   // First we have the public Messenger interface implementation...
  75 public:
  76   /**
  77    * Initialize the AsyncMessenger!
  78    *
  79    * @param cct The CephContext to use
  80    * @param name The name to assign ourselves
  81    * _nonce A unique ID to use for this AsyncMessenger. It should not
  82    * be a value that will be repeated if the daemon restarts.
  83    */
  84   AsyncMessenger(CephContext *cct, entity_name_t name, const std::string &type,
  85                  std::string mname, uint64_t _nonce);
  86
  87   /**
  88    * Destroy the AsyncMessenger. Pretty simple since all the work is done
  89    * elsewhere.
  90    */
  91   ~AsyncMessenger() override;
  92
  93   /** @defgroup Accessors
  94    * @{
  95    */
  96   bool set_addr_unknowns(const entity_addrvec_t &addr) override;
  97   void set_addrs(const entity_addrvec_t &addrs) override;
  98
  99   int get_dispatch_queue_len() override {
 100     return dispatch_queue.get_queue_len();
 101   }
 102
 103   double get_dispatch_queue_max_age(utime_t now) override {
 104     return dispatch_queue.get_max_age(now);
 105   }
 106   /** @} Accessors */
 107
 108   /**
 109    * @defgroup Configuration functions
 110    * @{
 111    */
 112   void set_cluster_protocol(int p) override {
 113     ceph_assert(!started && !did_bind);
 114     cluster_protocol = p;
 115   }
 116
 117   int bind(const entity_addr_t& bind_addr) override;
 118   int rebind(const std::set<int>& avoid_ports) override;
 119   int bindv(const entity_addrvec_t& bind_addrs) override;
 120
 121   int client_bind(const entity_addr_t& bind_addr) override;
 122
 123   int client_reset() override;
 124
 125   bool should_use_msgr2() override;
 126
 127   /** @} Configuration functions */
 128
 129   /**
 130    * @defgroup Startup/Shutdown
 131    * @{
 132    */
 133   int start() override;
 134   void wait() override;
 135   int shutdown() override;
 136
 137   /** @} // Startup/Shutdown */
 138
 139   /**
 140    * @defgroup Messaging
 141    * @{
 142    */
 143   int send_to(Message *m, int type, const entity_addrvec_t& addrs) override;
 144
 145   /** @} // Messaging */
 146
 147   /**
 148    * @defgroup Connection Management
 149    * @{
 150    */
 151   ConnectionRef connect_to(int type,
 152                            const entity_addrvec_t& addrs,
 153                            bool anon, bool not_local_dest=false) override;
 154   ConnectionRef get_loopback_connection() override;
 155   void mark_down(const entity_addr_t& addr) override {
 156     mark_down_addrs(entity_addrvec_t(addr));
 157   }
 158   void mark_down_addrs(const entity_addrvec_t& addrs) override;
 159   void mark_down_all() override {
 160     shutdown_connections(true);
 161   }
 162   /** @} // Connection Management */
 163
 164   /**
 165    * @defgroup Inner classes
 166    * @{
 167    */
 168
 169   /**
 170    * @} // Inner classes
 171    */
 172
 173 protected:
 174   /**
 175    * @defgroup Messenger Interfaces
 176    * @{
 177    */
 178   /**
 179    * Start up the DispatchQueue thread once we have somebody to dispatch to.
 180    */
 181   void ready() override;
 182   /** @} // Messenger Interfaces */
 183
 184 private:
 185
 186   /**
 187    * @defgroup Utility functions
 188    * @{
 189    */
 190
 191   /**
 192    * Create a connection associated with the given entity (of the given type).
 193    * Initiate the connection. (This function returning does not guarantee
 194    * connection success.)
 195    *
 196    * @param addrs The address(es) of the entity to connect to.
 197    * @param type The peer type of the entity at the address.
 198    *
 199    * @return a pointer to the newly-created connection. Caller does not own a
 200    * reference; take one if you need it.
 201    */
 202   AsyncConnectionRef create_connect(const entity_addrvec_t& addrs, int type,
 203                                     bool anon);
 204
 205
 206   void _finish_bind(const entity_addrvec_t& bind_addrs,
 207                     const entity_addrvec_t& listen_addrs);
 208
 209   entity_addrvec_t _filter_addrs(const entity_addrvec_t& addrs);
 210
 211  private:
 212   NetworkStack *stack;
 213   std::vector<Processor*> processors;
 214   friend class Processor;
 215   DispatchQueue dispatch_queue;
 216
 217   // the worker run messenger's cron jobs
 218   Worker *local_worker;
 219
 220   std::string ms_type;
 221
 222   /// overall lock used for AsyncMessenger data structures
 223   ceph::mutex lock = ceph::make_mutex("AsyncMessenger::lock");
 224   // AsyncMessenger stuff
 225   /// approximately unique ID set by the Constructor for use in entity_addr_t
 226   uint64_t nonce;
 227
 228   /// true, specifying we haven't learned our addr; set false when we find it.
 229   // maybe this should be protected by the lock?
 230   bool need_addr = true;
 231
 232   /**
 233    * set to bind addresses if bind was called before NetworkStack was ready to
 234    * bind
 235    */
 236   entity_addrvec_t pending_bind_addrs;
 237
 238   /**
 239    * false; set to true if a pending bind exists
 240    */
 241   bool pending_bind = false;
 242
 243   /**
 244    *  The following aren't lock-protected since you shouldn't be able to race
 245    *  the only writers.
 246    */
 247
 248   /**
 249    *  false; set to true if the AsyncMessenger bound to a specific address;
 250    *  and set false again by Accepter::stop().
 251    */
 252   bool did_bind = false;
 253   /// counter for the global seq our connection protocol uses
 254   __u32 global_seq = 0;
 255   /// lock to protect the global_seq
 256   ceph::spinlock global_seq_lock;
 257
 258   /**
 259    * hash map of addresses to Asyncconnection
 260    *
 261    * NOTE: a Asyncconnection* with state CLOSED may still be in the map but is considered
 262    * invalid and can be replaced by anyone holding the msgr lock
 263    */
 264   ceph::unordered_map<entity_addrvec_t, AsyncConnectionRef> conns;
 265
 266   /**
 267    * list of connection are in the process of accepting
 268    *
 269    * These are not yet in the conns map.
 270    */
 271   std::set<AsyncConnectionRef> accepting_conns;
 272
 273   /// anonymous outgoing connections
 274   std::set<AsyncConnectionRef> anon_conns;
 275
 276   /**
 277    * list of connection are closed which need to be clean up
 278    *
 279    * Because AsyncMessenger and AsyncConnection follow a lock rule that
 280    * we can lock AsyncMesenger::lock firstly then lock AsyncConnection::lock
 281    * but can't reversed. This rule is aimed to avoid dead lock.
 282    * So if AsyncConnection want to unregister itself from AsyncMessenger,
 283    * we pick up this idea that just queue itself to this set and do lazy
 284    * deleted for AsyncConnection. "_lookup_conn" must ensure not return a
 285    * AsyncConnection in this set.
 286    */
 287   ceph::mutex deleted_lock = ceph::make_mutex("AsyncMessenger::deleted_lock");
 288   std::set<AsyncConnectionRef> deleted_conns;
 289
 290   EventCallbackRef reap_handler;
 291
 292   /// internal cluster protocol version, if any, for talking to entities of the same type.
 293   int cluster_protocol = 0;
 294
 295   ceph::condition_variable  stop_cond;
 296   bool stopped = true;
 297
 298   /* You must hold this->lock for the duration of use! */
 299   const auto& _lookup_conn(const entity_addrvec_t& k) {
 300     static const AsyncConnectionRef nullref;
 301     ceph_assert(ceph_mutex_is_locked(lock));
 302     auto p = conns.find(k);
 303     if (p == conns.end()) {
 304       return nullref;
 305     }
 306
 307     // lazy delete, see "deleted_conns"
 308     // don't worry omit, Connection::send_message can handle this case.
 309     if (p->second->is_unregistered()) {
 310       std::lock_guard l{deleted_lock};
 311       if (deleted_conns.erase(p->second)) {
 312         p->second->get_perf_counter()->dec(l_msgr_active_connections);
 313         conns.erase(p);
 314         return nullref;
 315       }
 316     }
 317
 318     return p->second;
 319   }
 320
 321   void _init_local_connection() {
 322     ceph_assert(ceph_mutex_is_locked(lock));
 323     local_connection->peer_addrs = *my_addrs;
 324     local_connection->peer_type = my_name.type();
 325     local_connection->set_features(CEPH_FEATURES_ALL);
 326     ms_deliver_handle_fast_connect(local_connection.get());
 327   }
 328
 329   void shutdown_connections(bool queue_reset);
 330
 331 public:
 332
 333   /// con used for sending messages to ourselves
 334   AsyncConnectionRef local_connection;
 335
 336   /**
 337    * @defgroup AsyncMessenger internals
 338    * @{
 339    */
 340   /**
 341    * This wraps _lookup_conn.
 342    */
 343   AsyncConnectionRef lookup_conn(const entity_addrvec_t& k) {
 344     std::lock_guard l{lock};
 345     return _lookup_conn(k); /* make new ref! */
 346   }
 347
 348   int accept_conn(const AsyncConnectionRef& conn);
 349   bool learned_addr(const entity_addr_t &peer_addr_for_me);
 350   void add_accept(Worker *w, ConnectedSocket cli_socket,
 351                   const entity_addr_t &listen_addr,
 352                   const entity_addr_t &peer_addr);
 353   NetworkStack *get_stack() {
 354     return stack;
 355   }
 356
 357   uint64_t get_nonce() const {
 358     return nonce;
 359   }
 360
 361   /**
 362    * Increment the global sequence for this AsyncMessenger and return it.
 363    * This is for the connect protocol, although it doesn't hurt if somebody
 364    * else calls it.
 365    *
 366    * @return a global sequence ID that nobody else has seen.
 367    */
 368   __u32 get_global_seq(__u32 old=0) {
 369     std::lock_guard<ceph::spinlock> lg(global_seq_lock);
 370
 371     if (old > global_seq)
 372       global_seq = old;
 373     __u32 ret = ++global_seq;
 374
 375     return ret;
 376   }
 377   /**
 378    * Get the protocol version we support for the given peer type: either
 379    * a peer protocol (if it matches our own), the protocol version for the
 380    * peer (if we're connecting), or our protocol version (if we're accepting).
 381    */
 382   int get_proto_version(int peer_type, bool connect) const;
 383
 384   /**
 385    * Fill in the address and peer type for the local connection, which
 386    * is used for delivering messages back to ourself.
 387    */
 388   void init_local_connection() {
 389     std::lock_guard l{lock};
 390     local_connection->is_loopback = true;
 391     _init_local_connection();
 392   }
 393
 394   /**
 395    * Unregister connection from `conns`
 396    *
 397    * See "deleted_conns"
 398    */
 399   void unregister_conn(const AsyncConnectionRef& conn) {
 400     std::lock_guard l{deleted_lock};
 401     deleted_conns.emplace(std::move(conn));
 402     conn->unregister();
 403
 404     if (deleted_conns.size() >= cct->_conf->ms_async_reap_threshold) {
 405       local_worker->center.dispatch_event_external(reap_handler);
 406     }
 407   }
 408
 409   /**
 410    * Reap dead connection from `deleted_conns`
 411    *
 412    * @return the number of dead connections
 413    *
 414    * See "deleted_conns"
 415    */
 416   void reap_dead();
 417
 418   /**
 419    * @} // AsyncMessenger Internals
 420    */
 421 } ;
 422
 423 #endif /* CEPH_ASYNCMESSENGER_H */