git.m6w6.name Git - m6w6/libmemcached/blob - libmemcached/connect.cc

   1 /*  vim:expandtab:shiftwidth=2:tabstop=2:smarttab:
   2  *
   3  *  Libmemcached library
   4  *
   5  *  Copyright (C) 2011 Data Differential, http://datadifferential.com/
   6  *  Copyright (C) 2006-2010 Brian Aker All rights reserved.
   7  *
   8  *  Redistribution and use in source and binary forms, with or without
   9  *  modification, are permitted provided that the following conditions are
  10  *  met:
  11  *
  12  *      * Redistributions of source code must retain the above copyright
  13  *  notice, this list of conditions and the following disclaimer.
  14  *
  15  *      * Redistributions in binary form must reproduce the above
  16  *  copyright notice, this list of conditions and the following disclaimer
  17  *  in the documentation and/or other materials provided with the
  18  *  distribution.
  19  *
  20  *      * The names of its contributors may not be used to endorse or
  21  *  promote products derived from this software without specific prior
  22  *  written permission.
  23  *
  24  *  THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  25  *  "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  26  *  LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  27  *  A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  28  *  OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  29  *  SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  30  *  LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  31  *  DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  32  *  THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  33  *  (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  34  *  OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  35  *
  36  */
  37
  38
  39 #include <libmemcached/common.h>
  40 #include <cassert>
  41 #include <ctime>
  42 #include <sys/time.h>
  43
  44 static memcached_return_t connect_poll(memcached_server_st *ptr)
  45 {
  46   struct pollfd fds[1];
  47   fds[0].fd = ptr->fd;
  48   fds[0].events = POLLOUT;
  49
  50   size_t loop_max= 5;
  51
  52   if (ptr->root->poll_timeout == 0)
  53   {
  54     return memcached_set_error(*ptr, MEMCACHED_TIMEOUT, MEMCACHED_AT);
  55   }
  56
  57   while (--loop_max) // Should only loop on cases of ERESTART or EINTR
  58   {
  59     int error= poll(fds, 1, ptr->root->connect_timeout);
  60     switch (error)
  61     {
  62     case 1:
  63       {
  64         int err;
  65         socklen_t len= sizeof (err);
  66         (void)getsockopt(ptr->fd, SOL_SOCKET, SO_ERROR, &err, &len);
  67
  68         // We check the value to see what happened wth the socket.
  69         if (err == 0)
  70         {
  71           return MEMCACHED_SUCCESS;
  72         }
  73
  74         return memcached_set_errno(*ptr, err, MEMCACHED_AT);
  75       }
  76     case 0:
  77       {
  78         return memcached_set_error(*ptr, MEMCACHED_TIMEOUT, MEMCACHED_AT);
  79       }
  80
  81     default: // A real error occurred and we need to completely bail
  82       WATCHPOINT_ERRNO(get_socket_errno());
  83       switch (get_socket_errno())
  84       {
  85 #ifdef TARGET_OS_LINUX
  86       case ERESTART:
  87 #endif
  88       case EINTR:
  89         continue;
  90
  91       case EFAULT:
  92       case ENOMEM:
  93         return memcached_set_error(*ptr, MEMCACHED_MEMORY_ALLOCATION_FAILURE, MEMCACHED_AT);
  94
  95       case EINVAL:
  96         return memcached_set_error(*ptr, MEMCACHED_MEMORY_ALLOCATION_FAILURE, MEMCACHED_AT, memcached_literal_param("RLIMIT_NOFILE exceeded, or if OSX the timeout value was invalid"));
  97
  98       default: // This should not happen
  99         if (fds[0].revents & POLLERR)
 100         {
 101           int err;
 102           socklen_t len= sizeof (err);
 103           (void)getsockopt(ptr->fd, SOL_SOCKET, SO_ERROR, &err, &len);
 104           memcached_set_errno(*ptr, (err == 0) ? get_socket_errno() : err, MEMCACHED_AT);
 105         }
 106         else
 107         {
 108           memcached_set_errno(*ptr, get_socket_errno(), MEMCACHED_AT);
 109         }
 110
 111         WATCHPOINT_ASSERT(ptr->fd != INVALID_SOCKET);
 112         (void)closesocket(ptr->fd);
 113         ptr->fd= INVALID_SOCKET;
 114         ptr->state= MEMCACHED_SERVER_STATE_NEW;
 115
 116         return memcached_set_errno(*ptr, get_socket_errno(), MEMCACHED_AT);
 117       }
 118     }
 119   }
 120
 121   // This should only be possible from ERESTART or EINTR;
 122   return memcached_set_errno(*ptr, get_socket_errno(), MEMCACHED_AT);
 123 }
 124
 125 static memcached_return_t set_hostinfo(memcached_server_st *server)
 126 {
 127   WATCHPOINT_ASSERT(not server->address_info); // We cover the case where a programming mistake has been made.
 128   if (server->address_info)
 129   {
 130     freeaddrinfo(server->address_info);
 131     server->address_info= NULL;
 132     server->address_info_next= NULL;
 133   }
 134
 135   char str_port[NI_MAXSERV];
 136   int length= snprintf(str_port, NI_MAXSERV, "%u", (uint32_t)server->port);
 137   if (length >= NI_MAXSERV || length < 0)
 138     return MEMCACHED_FAILURE;
 139
 140   struct addrinfo hints;
 141   memset(&hints, 0, sizeof(struct addrinfo));
 142
 143 #if 0
 144   hints.ai_family= AF_INET;
 145 #endif
 146   if (server->type == MEMCACHED_CONNECTION_UDP)
 147   {
 148     hints.ai_protocol= IPPROTO_UDP;
 149     hints.ai_socktype= SOCK_DGRAM;
 150   }
 151   else
 152   {
 153     hints.ai_socktype= SOCK_STREAM;
 154     hints.ai_protocol= IPPROTO_TCP;
 155   }
 156
 157   int errcode;
 158   switch(errcode= getaddrinfo(server->hostname, str_port, &hints, &server->address_info))
 159   {
 160   case 0:
 161     break;
 162
 163   case EAI_AGAIN:
 164     return memcached_set_error(*server, MEMCACHED_TIMEOUT, MEMCACHED_AT, memcached_string_make_from_cstr(gai_strerror(errcode)));
 165
 166   case EAI_SYSTEM:
 167       return memcached_set_errno(*server, errno, MEMCACHED_AT, memcached_literal_param("getaddrinfo(EAI_SYSTEM)"));
 168
 169   case EAI_BADFLAGS:
 170     return memcached_set_error(*server, MEMCACHED_INVALID_ARGUMENTS, MEMCACHED_AT, memcached_literal_param("getaddrinfo(EAI_BADFLAGS)"));
 171
 172   case EAI_MEMORY:
 173     return memcached_set_error(*server, MEMCACHED_MEMORY_ALLOCATION_FAILURE, MEMCACHED_AT, memcached_literal_param("getaddrinfo(EAI_MEMORY)"));
 174
 175   default:
 176     {
 177       WATCHPOINT_STRING(server->hostname);
 178       WATCHPOINT_STRING(gai_strerror(errcode));
 179       return memcached_set_error(*server, MEMCACHED_HOST_LOOKUP_FAILURE, MEMCACHED_AT, memcached_string_make_from_cstr(gai_strerror(errcode)));
 180     }
 181   }
 182   server->address_info_next= server->address_info;
 183   server->state= MEMCACHED_SERVER_STATE_ADDRINFO;
 184
 185   return MEMCACHED_SUCCESS;
 186 }
 187
 188 static inline void set_socket_nonblocking(memcached_server_st *ptr)
 189 {
 190 #ifdef WIN32
 191   u_long arg = 1;
 192   if (ioctlsocket(ptr->fd, FIONBIO, &arg) == SOCKET_ERROR)
 193   {
 194     memcached_set_errno(*ptr, get_socket_errno(), NULL);
 195   }
 196 #else
 197   int flags;
 198
 199   do
 200   {
 201     flags= fcntl(ptr->fd, F_GETFL, 0);
 202   } while (flags == -1 && (errno == EINTR || errno == EAGAIN));
 203
 204   if (flags == -1)
 205   {
 206     memcached_set_errno(*ptr, errno, NULL);
 207   }
 208   else if ((flags & O_NONBLOCK) == 0)
 209   {
 210     int rval;
 211
 212     do
 213     {
 214       rval= fcntl(ptr->fd, F_SETFL, flags | O_NONBLOCK);
 215     } while (rval == -1 && (errno == EINTR || errno == EAGAIN));
 216
 217     unlikely (rval == -1)
 218     {
 219       memcached_set_errno(*ptr, errno, NULL);
 220     }
 221   }
 222 #endif
 223 }
 224
 225 static void set_socket_options(memcached_server_st *ptr)
 226 {
 227   WATCHPOINT_ASSERT(ptr->fd != -1);
 228
 229   if (ptr->type == MEMCACHED_CONNECTION_UDP)
 230     return;
 231
 232 #ifdef HAVE_SNDTIMEO
 233   if (ptr->root->snd_timeout)
 234   {
 235     int error;
 236     struct timeval waittime;
 237
 238     waittime.tv_sec= 0;
 239     waittime.tv_usec= ptr->root->snd_timeout;
 240
 241     error= setsockopt(ptr->fd, SOL_SOCKET, SO_SNDTIMEO,
 242                       &waittime, (socklen_t)sizeof(struct timeval));
 243     WATCHPOINT_ASSERT(error == 0);
 244   }
 245 #endif
 246
 247 #ifdef HAVE_RCVTIMEO
 248   if (ptr->root->rcv_timeout)
 249   {
 250     int error;
 251     struct timeval waittime;
 252
 253     waittime.tv_sec= 0;
 254     waittime.tv_usec= ptr->root->rcv_timeout;
 255
 256     error= setsockopt(ptr->fd, SOL_SOCKET, SO_RCVTIMEO,
 257                       &waittime, (socklen_t)sizeof(struct timeval));
 258     WATCHPOINT_ASSERT(error == 0);
 259   }
 260 #endif
 261
 262
 263 #if defined(__MACH__) && defined(__APPLE__) || defined(__FreeBSD__)
 264   {
 265     int set= 1;
 266     int error= setsockopt(ptr->fd, SOL_SOCKET, SO_NOSIGPIPE, (void *)&set, sizeof(int));
 267
 268     // This is not considered a fatal error
 269     if (error == -1)
 270     {
 271       WATCHPOINT_ERRNO(get_socket_errno());
 272       perror("setsockopt(SO_NOSIGPIPE)");
 273     }
 274   }
 275 #endif
 276
 277   if (ptr->root->flags.no_block)
 278   {
 279     int error;
 280     struct linger linger;
 281
 282     linger.l_onoff= 1;
 283     linger.l_linger= 0; /* By default on close() just drop the socket */
 284     error= setsockopt(ptr->fd, SOL_SOCKET, SO_LINGER,
 285                       &linger, (socklen_t)sizeof(struct linger));
 286     WATCHPOINT_ASSERT(error == 0);
 287   }
 288
 289   if (ptr->root->flags.tcp_nodelay)
 290   {
 291     int flag= 1;
 292     int error;
 293
 294     error= setsockopt(ptr->fd, IPPROTO_TCP, TCP_NODELAY,
 295                       &flag, (socklen_t)sizeof(int));
 296     WATCHPOINT_ASSERT(error == 0);
 297   }
 298
 299   if (ptr->root->flags.tcp_keepalive)
 300   {
 301     int flag= 1;
 302     int error;
 303
 304     error= setsockopt(ptr->fd, SOL_SOCKET, SO_KEEPALIVE,
 305                       &flag, (socklen_t)sizeof(int));
 306     WATCHPOINT_ASSERT(error == 0);
 307   }
 308
 309 #ifdef TCP_KEEPIDLE
 310   if (ptr->root->tcp_keepidle > 0)
 311   {
 312     int error;
 313
 314     error= setsockopt(ptr->fd, IPPROTO_TCP, TCP_KEEPIDLE,
 315                       &ptr->root->tcp_keepidle, (socklen_t)sizeof(int));
 316     WATCHPOINT_ASSERT(error == 0);
 317   }
 318 #endif
 319
 320   if (ptr->root->send_size > 0)
 321   {
 322     int error;
 323
 324     error= setsockopt(ptr->fd, SOL_SOCKET, SO_SNDBUF,
 325                       &ptr->root->send_size, (socklen_t)sizeof(int));
 326     WATCHPOINT_ASSERT(error == 0);
 327   }
 328
 329   if (ptr->root->recv_size > 0)
 330   {
 331     int error;
 332
 333     error= setsockopt(ptr->fd, SOL_SOCKET, SO_RCVBUF,
 334                       &ptr->root->recv_size, (socklen_t)sizeof(int));
 335     WATCHPOINT_ASSERT(error == 0);
 336   }
 337
 338
 339   /* libmemcached will always use nonblocking IO to avoid write deadlocks */
 340   set_socket_nonblocking(ptr);
 341 }
 342
 343 static memcached_return_t unix_socket_connect(memcached_server_st *ptr)
 344 {
 345 #ifndef WIN32
 346   WATCHPOINT_ASSERT(ptr->fd == -1);
 347
 348   if ((ptr->fd= socket(AF_UNIX, SOCK_STREAM, 0)) < 0)
 349   {
 350     memcached_set_errno(*ptr, errno, NULL);
 351     return MEMCACHED_CONNECTION_FAILURE;
 352   }
 353
 354   struct sockaddr_un servAddr;
 355
 356   memset(&servAddr, 0, sizeof (struct sockaddr_un));
 357   servAddr.sun_family= AF_UNIX;
 358   strncpy(servAddr.sun_path, ptr->hostname, sizeof(servAddr.sun_path)); /* Copy filename */
 359
 360   do {
 361     if (connect(ptr->fd, (struct sockaddr *)&servAddr, sizeof(servAddr)) < 0)
 362     {
 363       switch (errno)
 364       {
 365       case EINPROGRESS:
 366       case EALREADY:
 367       case EINTR:
 368         continue;
 369
 370       case EISCONN: /* We were spinning waiting on connect */
 371         {
 372           WATCHPOINT_ASSERT(0); // Programmer error
 373           break;
 374         }
 375
 376       default:
 377         WATCHPOINT_ERRNO(errno);
 378         memcached_set_errno(*ptr, errno, MEMCACHED_AT);
 379         return MEMCACHED_CONNECTION_FAILURE;
 380       }
 381     }
 382   } while (0);
 383   ptr->state= MEMCACHED_SERVER_STATE_CONNECTED;
 384
 385   WATCHPOINT_ASSERT(ptr->fd != INVALID_SOCKET);
 386
 387   return MEMCACHED_SUCCESS;
 388 #else
 389   (void)ptr;
 390   return MEMCACHED_NOT_SUPPORTED;
 391 #endif
 392 }
 393
 394 static memcached_return_t network_connect(memcached_server_st *ptr)
 395 {
 396   bool timeout_error_occured= false;
 397
 398   WATCHPOINT_ASSERT(ptr->fd == INVALID_SOCKET);
 399   WATCHPOINT_ASSERT(ptr->cursor_active == 0);
 400
 401   if (not ptr->address_info)
 402   {
 403     WATCHPOINT_ASSERT(ptr->state == MEMCACHED_SERVER_STATE_NEW);
 404     memcached_return_t rc;
 405     uint32_t counter= 5;
 406     while (--counter)
 407     {
 408       if ((rc= set_hostinfo(ptr)) != MEMCACHED_TIMEOUT)
 409         break;
 410
 411 #ifndef WIN32
 412       struct timespec dream, rem;
 413
 414       dream.tv_nsec= 1000;
 415       dream.tv_sec= 0;
 416
 417       nanosleep(&dream, &rem);
 418 #endif
 419     }
 420
 421     if (memcached_failed(rc))
 422       return rc;
 423   }
 424
 425   /* Create the socket */
 426   while (ptr->address_info_next && ptr->fd == INVALID_SOCKET)
 427   {
 428     /* Memcache server does not support IPV6 in udp mode, so skip if not ipv4 */
 429     if (ptr->type == MEMCACHED_CONNECTION_UDP && ptr->address_info_next->ai_family != AF_INET)
 430     {
 431       ptr->address_info_next= ptr->address_info_next->ai_next;
 432       continue;
 433     }
 434
 435     if ((ptr->fd= socket(ptr->address_info_next->ai_family,
 436                          ptr->address_info_next->ai_socktype,
 437                          ptr->address_info_next->ai_protocol)) < 0)
 438     {
 439       return memcached_set_errno(*ptr, get_socket_errno(), NULL);
 440     }
 441
 442     set_socket_options(ptr);
 443
 444     /* connect to server */
 445     if ((connect(ptr->fd, ptr->address_info_next->ai_addr, ptr->address_info_next->ai_addrlen) != SOCKET_ERROR))
 446     {
 447       ptr->state= MEMCACHED_SERVER_STATE_CONNECTED;
 448       break; // Success
 449     }
 450
 451     /* An error occurred */
 452     switch (get_socket_errno())
 453     {
 454     case ETIMEDOUT:
 455       timeout_error_occured= true;
 456       break;
 457
 458     case EWOULDBLOCK:
 459     case EINPROGRESS: // nonblocking mode - first return
 460     case EALREADY: // nonblocking mode - subsequent returns
 461       {
 462         ptr->state= MEMCACHED_SERVER_STATE_IN_PROGRESS;
 463         memcached_return_t rc= connect_poll(ptr);
 464
 465         if (memcached_success(rc))
 466         {
 467           ptr->state= MEMCACHED_SERVER_STATE_CONNECTED;
 468           return MEMCACHED_SUCCESS;
 469         }
 470
 471         // A timeout here is treated as an error, we will not retry
 472         if (rc == MEMCACHED_TIMEOUT)
 473         {
 474           timeout_error_occured= true;
 475         }
 476       }
 477       break;
 478
 479     case EISCONN: // we are connected :-)
 480       WATCHPOINT_ASSERT(0); // This is a programmer's error
 481       break;
 482
 483     case EINTR: // Special case, we retry ai_addr
 484       WATCHPOINT_ASSERT(ptr->fd != INVALID_SOCKET);
 485       (void)closesocket(ptr->fd);
 486       ptr->fd= INVALID_SOCKET;
 487       continue;
 488
 489     default:
 490       break;
 491     }
 492
 493     WATCHPOINT_ASSERT(ptr->fd != INVALID_SOCKET);
 494     (void)closesocket(ptr->fd);
 495     ptr->fd= INVALID_SOCKET;
 496     ptr->address_info_next= ptr->address_info_next->ai_next;
 497   }
 498
 499   WATCHPOINT_ASSERT(ptr->fd == INVALID_SOCKET);
 500
 501   if (timeout_error_occured)
 502   {
 503     if (ptr->fd != INVALID_SOCKET)
 504     {
 505       (void)closesocket(ptr->fd);
 506       ptr->fd= INVALID_SOCKET;
 507     }
 508   }
 509
 510   WATCHPOINT_STRING("Never got a good file descriptor");
 511   /* Failed to connect. schedule next retry */
 512   if (ptr->root->retry_timeout)
 513   {
 514     struct timeval next_time;
 515
 516     if (gettimeofday(&next_time, NULL) == 0)
 517       ptr->next_retry= next_time.tv_sec + ptr->root->retry_timeout;
 518   }
 519
 520   if (timeout_error_occured)
 521     return memcached_set_error(*ptr, MEMCACHED_TIMEOUT, MEMCACHED_AT);
 522
 523   return memcached_set_error(*ptr, MEMCACHED_CONNECTION_FAILURE, MEMCACHED_AT); /* The last error should be from connect() */
 524 }
 525
 526 void set_last_disconnected_host(memcached_server_write_instance_st self)
 527 {
 528   // const_cast
 529   memcached_st *root= (memcached_st *)self->root;
 530
 531 #if 0
 532   WATCHPOINT_STRING(self->hostname);
 533   WATCHPOINT_NUMBER(self->port);
 534   WATCHPOINT_ERRNO(self->cached_errno);
 535 #endif
 536   memcached_server_free(root->last_disconnected_server);
 537   root->last_disconnected_server= memcached_server_clone(NULL, self);
 538 }
 539
 540 memcached_return_t memcached_connect(memcached_server_write_instance_st ptr)
 541 {
 542   memcached_return_t rc= MEMCACHED_NO_SERVERS;
 543
 544   if (ptr->fd != INVALID_SOCKET)
 545   {
 546     return MEMCACHED_SUCCESS;
 547   }
 548
 549   LIBMEMCACHED_MEMCACHED_CONNECT_START();
 550
 551   /* both retry_timeout and server_failure_limit must be set in order to delay retrying a server on error. */
 552   WATCHPOINT_ASSERT(ptr->root);
 553   if (ptr->root->retry_timeout && ptr->next_retry)
 554   {
 555     struct timeval curr_time;
 556
 557     gettimeofday(&curr_time, NULL);
 558
 559     // We should optimize this to remove the allocation if the server was
 560     // the last server to die
 561     if (ptr->next_retry > curr_time.tv_sec)
 562     {
 563       set_last_disconnected_host(ptr);
 564
 565       return memcached_set_error(*ptr, MEMCACHED_SERVER_MARKED_DEAD, MEMCACHED_AT);
 566     }
 567   }
 568
 569   // If we are over the counter failure, we just fail. Reject host only
 570   // works if you have a set number of failures.
 571   if (ptr->root->server_failure_limit && ptr->server_failure_counter >= ptr->root->server_failure_limit)
 572   {
 573     set_last_disconnected_host(ptr);
 574
 575     // @todo fix this by fixing behavior to no longer make use of
 576     // memcached_st
 577     if (_is_auto_eject_host(ptr->root))
 578     {
 579       run_distribution((memcached_st *)ptr->root);
 580     }
 581
 582     return memcached_set_error(*ptr, MEMCACHED_SERVER_MARKED_DEAD, MEMCACHED_AT);
 583   }
 584
 585   /* We need to clean up the multi startup piece */
 586   switch (ptr->type)
 587   {
 588   case MEMCACHED_CONNECTION_UNKNOWN:
 589     WATCHPOINT_ASSERT(0);
 590     rc= MEMCACHED_NOT_SUPPORTED;
 591     break;
 592
 593   case MEMCACHED_CONNECTION_UDP:
 594   case MEMCACHED_CONNECTION_TCP:
 595     rc= network_connect(ptr);
 596 #ifdef LIBMEMCACHED_WITH_SASL_SUPPORT
 597     if (ptr->fd != INVALID_SOCKET and ptr->root->sasl.callbacks)
 598     {
 599       rc= memcached_sasl_authenticate_connection(ptr);
 600       if (memcached_failed(rc) and ptr->fd != INVALID_SOCKET)
 601       {
 602         WATCHPOINT_ASSERT(ptr->fd != INVALID_SOCKET);
 603         (void)closesocket(ptr->fd);
 604         ptr->fd= INVALID_SOCKET;
 605       }
 606     }
 607 #endif
 608     break;
 609
 610   case MEMCACHED_CONNECTION_UNIX_SOCKET:
 611     rc= unix_socket_connect(ptr);
 612     break;
 613
 614   case MEMCACHED_CONNECTION_MAX:
 615   default:
 616     WATCHPOINT_ASSERT(0);
 617   }
 618
 619   if (memcached_success(rc))
 620   {
 621     ptr->server_failure_counter= 0;
 622     ptr->next_retry= 0;
 623   }
 624   else
 625   {
 626     memcached_set_error(*ptr, rc, MEMCACHED_AT);
 627     ptr->server_failure_counter++;
 628     set_last_disconnected_host(ptr);
 629   }
 630
 631   LIBMEMCACHED_MEMCACHED_CONNECT_END();
 632
 633   return rc;
 634 }