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 <assert.h>
  41 #include <sys/time.h>
  42 #include <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   int error;
  51   size_t loop_max= 5;
  52
  53   while (--loop_max) // Should only loop on cases of ERESTART or EINTR
  54   {
  55     error= poll(fds, 1, ptr->root->connect_timeout);
  56
  57     switch (error)
  58     {
  59     case 1:
  60       {
  61         int err;
  62         socklen_t len= sizeof (err);
  63         (void)getsockopt(ptr->fd, SOL_SOCKET, SO_ERROR, &err, &len);
  64
  65         // We check the value to see what happened wth the socket.
  66         if (err == 0)
  67         {
  68           return MEMCACHED_SUCCESS;
  69         }
  70         else
  71         {
  72           ptr->cached_errno= errno;
  73
  74           return MEMCACHED_ERRNO;
  75         }
  76       }
  77     case 0:
  78       return MEMCACHED_TIMEOUT;
  79     default: // A real error occurred and we need to completely bail
  80       WATCHPOINT_ERRNO(get_socket_errno());
  81       switch (get_socket_errno())
  82       {
  83 #ifdef TARGET_OS_LINUX
  84       case ERESTART:
  85 #endif
  86       case EINTR:
  87         continue;
  88       default:
  89         if (fds[0].revents & POLLERR)
  90         {
  91           int err;
  92           socklen_t len= sizeof (err);
  93           (void)getsockopt(ptr->fd, SOL_SOCKET, SO_ERROR, &err, &len);
  94           ptr->cached_errno= (err == 0) ? get_socket_errno() : err;
  95         }
  96         else
  97         {
  98           ptr->cached_errno= get_socket_errno();
  99         }
 100
 101         (void)closesocket(ptr->fd);
 102         ptr->fd= INVALID_SOCKET;
 103
 104         return MEMCACHED_ERRNO;
 105       }
 106     }
 107   }
 108
 109   // This should only be possible from ERESTART or EINTR;
 110   ptr->cached_errno= get_socket_errno();
 111
 112   return MEMCACHED_ERRNO;
 113 }
 114
 115 static memcached_return_t set_hostinfo(memcached_server_st *server)
 116 {
 117   char str_port[NI_MAXSERV];
 118
 119   assert(! server->address_info); // We cover the case where a programming mistake has been made.
 120   if (server->address_info)
 121   {
 122     freeaddrinfo(server->address_info);
 123     server->address_info= NULL;
 124     server->address_info_next= NULL;
 125   }
 126
 127   int length= snprintf(str_port, NI_MAXSERV, "%u", (uint32_t)server->port);
 128   if (length >= NI_MAXSERV || length < 0)
 129     return MEMCACHED_FAILURE;
 130
 131   struct addrinfo hints;
 132   memset(&hints, 0, sizeof(struct addrinfo));
 133
 134 #if 0
 135   hints.ai_family= AF_INET;
 136 #endif
 137   if (server->type == MEMCACHED_CONNECTION_UDP)
 138   {
 139     hints.ai_protocol= IPPROTO_UDP;
 140     hints.ai_socktype= SOCK_DGRAM;
 141   }
 142   else
 143   {
 144     hints.ai_socktype= SOCK_STREAM;
 145     hints.ai_protocol= IPPROTO_TCP;
 146   }
 147
 148   uint32_t counter= 5;
 149   while (--counter)
 150   {
 151     int e= getaddrinfo(server->hostname, str_port, &hints, &server->address_info);
 152
 153     if (e == 0)
 154     {
 155       break;
 156     }
 157     else if (e == EAI_AGAIN)
 158     {
 159 #ifndef WIN32
 160       struct timespec dream, rem;
 161
 162       dream.tv_nsec= 1000;
 163       dream.tv_sec= 0;
 164
 165       nanosleep(&dream, &rem);
 166 #endif
 167       continue;
 168     }
 169     else
 170     {
 171       WATCHPOINT_STRING(server->hostname);
 172       WATCHPOINT_STRING(gai_strerror(e));
 173       return MEMCACHED_HOST_LOOKUP_FAILURE;
 174     }
 175   }
 176
 177   server->address_info_next= server->address_info;
 178
 179   return MEMCACHED_SUCCESS;
 180 }
 181
 182 static inline memcached_return_t set_socket_nonblocking(memcached_server_st *ptr)
 183 {
 184 #ifdef WIN32
 185   u_long arg = 1;
 186   if (ioctlsocket(ptr->fd, FIONBIO, &arg) == SOCKET_ERROR)
 187   {
 188     ptr->cached_errno= get_socket_errno();
 189     return MEMCACHED_CONNECTION_FAILURE;
 190   }
 191 #else
 192   int flags;
 193
 194   do
 195   {
 196     flags= fcntl(ptr->fd, F_GETFL, 0);
 197   }
 198   while (flags == -1 && (errno == EINTR || errno == EAGAIN));
 199
 200   unlikely (flags == -1)
 201   {
 202     ptr->cached_errno= errno;
 203     return MEMCACHED_CONNECTION_FAILURE;
 204   }
 205   else if ((flags & O_NONBLOCK) == 0)
 206   {
 207     int rval;
 208
 209     do
 210     {
 211       rval= fcntl(ptr->fd, F_SETFL, flags | O_NONBLOCK);
 212     }
 213     while (rval == -1 && (errno == EINTR || errno == EAGAIN));
 214
 215     unlikely (rval == -1)
 216     {
 217       ptr->cached_errno= errno;
 218       return MEMCACHED_CONNECTION_FAILURE;
 219     }
 220   }
 221 #endif
 222   return MEMCACHED_SUCCESS;
 223 }
 224
 225 static memcached_return_t set_socket_options(memcached_server_st *ptr)
 226 {
 227   WATCHPOINT_ASSERT(ptr->fd != -1);
 228
 229   if (ptr->type == MEMCACHED_CONNECTION_UDP)
 230     return MEMCACHED_SUCCESS;
 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     if (error)
 245       return MEMCACHED_FAILURE;
 246   }
 247 #endif
 248
 249 #ifdef HAVE_RCVTIMEO
 250   if (ptr->root->rcv_timeout)
 251   {
 252     int error;
 253     struct timeval waittime;
 254
 255     waittime.tv_sec= 0;
 256     waittime.tv_usec= ptr->root->rcv_timeout;
 257
 258     error= setsockopt(ptr->fd, SOL_SOCKET, SO_RCVTIMEO,
 259                       &waittime, (socklen_t)sizeof(struct timeval));
 260     WATCHPOINT_ASSERT(error == 0);
 261     if (error)
 262       return MEMCACHED_FAILURE;
 263   }
 264 #endif
 265
 266
 267 #if defined(__MACH__) && defined(__APPLE__) || defined(__FreeBSD__)
 268   {
 269     int set = 1;
 270     int error= setsockopt(ptr->fd, SOL_SOCKET, SO_NOSIGPIPE, (void *)&set, sizeof(int));
 271
 272     // This is not considered a fatal error
 273     if (error == -1)
 274     {
 275       WATCHPOINT_ERRNO(get_socket_errno());
 276       perror("setsockopt(SO_NOSIGPIPE)");
 277     }
 278   }
 279 #endif
 280
 281   if (ptr->root->flags.no_block)
 282   {
 283     int error;
 284     struct linger linger;
 285
 286     linger.l_onoff= 1;
 287     linger.l_linger= 0; /* By default on close() just drop the socket */
 288     error= setsockopt(ptr->fd, SOL_SOCKET, SO_LINGER,
 289                       &linger, (socklen_t)sizeof(struct linger));
 290     WATCHPOINT_ASSERT(error == 0);
 291     if (error)
 292       return MEMCACHED_FAILURE;
 293   }
 294
 295   if (ptr->root->flags.tcp_nodelay)
 296   {
 297     int flag= 1;
 298     int error;
 299
 300     error= setsockopt(ptr->fd, IPPROTO_TCP, TCP_NODELAY,
 301                       &flag, (socklen_t)sizeof(int));
 302     WATCHPOINT_ASSERT(error == 0);
 303     if (error)
 304       return MEMCACHED_FAILURE;
 305   }
 306
 307   if (ptr->root->flags.tcp_keepalive)
 308   {
 309     int flag= 1;
 310     int error;
 311
 312     error= setsockopt(ptr->fd, SOL_SOCKET, SO_KEEPALIVE,
 313                       &flag, (socklen_t)sizeof(int));
 314     WATCHPOINT_ASSERT(error == 0);
 315     if (error)
 316       return MEMCACHED_FAILURE;
 317   }
 318
 319 #ifdef TCP_KEEPIDLE
 320   if (ptr->root->tcp_keepidle > 0)
 321   {
 322     int error;
 323
 324     error= setsockopt(ptr->fd, IPPROTO_TCP, TCP_KEEPIDLE,
 325                       &ptr->root->tcp_keepidle, (socklen_t)sizeof(int));
 326     WATCHPOINT_ASSERT(error == 0);
 327     if (error)
 328       return MEMCACHED_FAILURE;
 329   }
 330 #endif
 331
 332   if (ptr->root->send_size > 0)
 333   {
 334     int error;
 335
 336     error= setsockopt(ptr->fd, SOL_SOCKET, SO_SNDBUF,
 337                       &ptr->root->send_size, (socklen_t)sizeof(int));
 338     WATCHPOINT_ASSERT(error == 0);
 339     if (error)
 340       return MEMCACHED_FAILURE;
 341   }
 342
 343   if (ptr->root->recv_size > 0)
 344   {
 345     int error;
 346
 347     error= setsockopt(ptr->fd, SOL_SOCKET, SO_RCVBUF,
 348                       &ptr->root->recv_size, (socklen_t)sizeof(int));
 349     WATCHPOINT_ASSERT(error == 0);
 350     if (error)
 351       return MEMCACHED_FAILURE;
 352   }
 353
 354
 355   /* libmemcached will always use nonblocking IO to avoid write deadlocks */
 356   return set_socket_nonblocking(ptr);
 357 }
 358
 359 static memcached_return_t unix_socket_connect(memcached_server_st *ptr)
 360 {
 361 #ifndef WIN32
 362   WATCHPOINT_ASSERT(ptr->fd == -1);
 363
 364   if ((ptr->fd= socket(AF_UNIX, SOCK_STREAM, 0)) < 0)
 365   {
 366     ptr->cached_errno= errno;
 367     return MEMCACHED_CONNECTION_SOCKET_CREATE_FAILURE;
 368   }
 369
 370   struct sockaddr_un servAddr;
 371
 372   memset(&servAddr, 0, sizeof (struct sockaddr_un));
 373   servAddr.sun_family= AF_UNIX;
 374   strncpy(servAddr.sun_path, ptr->hostname, sizeof(servAddr.sun_path)); /* Copy filename */
 375
 376 test_connect:
 377   if (connect(ptr->fd,
 378               (struct sockaddr *)&servAddr,
 379               sizeof(servAddr)) < 0)
 380   {
 381     switch (errno)
 382     {
 383     case EINPROGRESS:
 384     case EALREADY:
 385     case EINTR:
 386       goto test_connect;
 387     case EISCONN: /* We were spinning waiting on connect */
 388       break;
 389     default:
 390       WATCHPOINT_ERRNO(errno);
 391       ptr->cached_errno= errno;
 392       return MEMCACHED_ERRNO;
 393     }
 394   }
 395
 396   WATCHPOINT_ASSERT(ptr->fd != -1);
 397
 398   return MEMCACHED_SUCCESS;
 399 #else
 400   (void)ptr;
 401   return MEMCACHED_NOT_SUPPORTED;
 402 #endif
 403 }
 404
 405 static memcached_return_t network_connect(memcached_server_st *ptr)
 406 {
 407   bool timeout_error_occured= false;
 408
 409   WATCHPOINT_ASSERT(ptr->fd == INVALID_SOCKET);
 410   WATCHPOINT_ASSERT(ptr->cursor_active == 0);
 411
 412   if (! ptr->address_info)
 413   {
 414     memcached_return_t rc= set_hostinfo(ptr);
 415     if (rc != MEMCACHED_SUCCESS)
 416       return rc;
 417   }
 418
 419   /* Create the socket */
 420   while (ptr->address_info_next && ptr->fd == INVALID_SOCKET)
 421   {
 422     /* Memcache server does not support IPV6 in udp mode, so skip if not ipv4 */
 423     if (ptr->type == MEMCACHED_CONNECTION_UDP && ptr->address_info_next->ai_family != AF_INET)
 424     {
 425       ptr->address_info_next= ptr->address_info_next->ai_next;
 426       continue;
 427     }
 428
 429     if ((ptr->fd= socket(ptr->address_info_next->ai_family,
 430                          ptr->address_info_next->ai_socktype,
 431                          ptr->address_info_next->ai_protocol)) < 0)
 432     {
 433       ptr->cached_errno= get_socket_errno();
 434       WATCHPOINT_ERRNO(get_socket_errno());
 435       return MEMCACHED_CONNECTION_SOCKET_CREATE_FAILURE;
 436     }
 437
 438     (void)set_socket_options(ptr);
 439
 440     /* connect to server */
 441     if ((connect(ptr->fd, ptr->address_info_next->ai_addr, ptr->address_info_next->ai_addrlen) != SOCKET_ERROR))
 442     {
 443       break; // Success
 444     }
 445
 446     /* An error occurred */
 447     ptr->cached_errno= get_socket_errno();
 448     switch (ptr->cached_errno)
 449     {
 450     case EWOULDBLOCK:
 451     case EINPROGRESS: // nonblocking mode - first return
 452     case EALREADY: // nonblocking mode - subsequent returns
 453       {
 454         memcached_return_t rc;
 455         rc= connect_poll(ptr);
 456
 457         if (rc == MEMCACHED_TIMEOUT)
 458           timeout_error_occured= true;
 459
 460         if (rc == MEMCACHED_SUCCESS)
 461           break;
 462       }
 463
 464     case EISCONN: // we are connected :-)
 465       break;
 466
 467     case EINTR: // Special case, we retry ai_addr
 468       (void)closesocket(ptr->fd);
 469       ptr->fd= INVALID_SOCKET;
 470       continue;
 471
 472     default:
 473       (void)closesocket(ptr->fd);
 474       ptr->fd= INVALID_SOCKET;
 475       ptr->address_info_next= ptr->address_info_next->ai_next;
 476       break;
 477     }
 478   }
 479
 480   if (ptr->fd == INVALID_SOCKET)
 481   {
 482     WATCHPOINT_STRING("Never got a good file descriptor");
 483
 484     /* Failed to connect. schedule next retry */
 485     if (ptr->root->retry_timeout)
 486     {
 487       struct timeval next_time;
 488
 489       if (gettimeofday(&next_time, NULL) == 0)
 490         ptr->next_retry= next_time.tv_sec + ptr->root->retry_timeout;
 491     }
 492
 493     if (timeout_error_occured)
 494       return MEMCACHED_TIMEOUT;
 495
 496     return MEMCACHED_ERRNO; /* The last error should be from connect() */
 497   }
 498
 499   return MEMCACHED_SUCCESS; /* The last error should be from connect() */
 500 }
 501
 502 void set_last_disconnected_host(memcached_server_write_instance_st ptr)
 503 {
 504   // const_cast
 505   memcached_st *root= (memcached_st *)ptr->root;
 506
 507 #if 0
 508   WATCHPOINT_STRING(ptr->hostname);
 509   WATCHPOINT_NUMBER(ptr->port);
 510   WATCHPOINT_ERRNO(ptr->cached_errno);
 511 #endif
 512   if (root->last_disconnected_server)
 513     memcached_server_free(root->last_disconnected_server);
 514   root->last_disconnected_server= memcached_server_clone(NULL, ptr);
 515 }
 516
 517 memcached_return_t memcached_connect(memcached_server_write_instance_st ptr)
 518 {
 519   memcached_return_t rc= MEMCACHED_NO_SERVERS;
 520
 521   if (ptr->fd != INVALID_SOCKET)
 522     return MEMCACHED_SUCCESS;
 523
 524   LIBMEMCACHED_MEMCACHED_CONNECT_START();
 525
 526   /* both retry_timeout and server_failure_limit must be set in order to delay retrying a server on error. */
 527   WATCHPOINT_ASSERT(ptr->root);
 528   if (ptr->root->retry_timeout && ptr->next_retry)
 529   {
 530     struct timeval curr_time;
 531
 532     gettimeofday(&curr_time, NULL);
 533
 534     // We should optimize this to remove the allocation if the server was
 535     // the last server to die
 536     if (ptr->next_retry > curr_time.tv_sec)
 537     {
 538       set_last_disconnected_host(ptr);
 539
 540       return MEMCACHED_SERVER_MARKED_DEAD;
 541     }
 542   }
 543
 544   // If we are over the counter failure, we just fail. Reject host only
 545   // works if you have a set number of failures.
 546   if (ptr->root->server_failure_limit && ptr->server_failure_counter >= ptr->root->server_failure_limit)
 547   {
 548     set_last_disconnected_host(ptr);
 549
 550     // @todo fix this by fixing behavior to no longer make use of
 551     // memcached_st
 552     if (_is_auto_eject_host(ptr->root))
 553     {
 554       run_distribution((memcached_st *)ptr->root);
 555     }
 556
 557     return MEMCACHED_SERVER_MARKED_DEAD;
 558   }
 559
 560   /* We need to clean up the multi startup piece */
 561   switch (ptr->type)
 562   {
 563   case MEMCACHED_CONNECTION_UNKNOWN:
 564     WATCHPOINT_ASSERT(0);
 565     rc= MEMCACHED_NOT_SUPPORTED;
 566     break;
 567   case MEMCACHED_CONNECTION_UDP:
 568   case MEMCACHED_CONNECTION_TCP:
 569     rc= network_connect(ptr);
 570 #ifdef LIBMEMCACHED_WITH_SASL_SUPPORT
 571     if (ptr->fd != INVALID_SOCKET && ptr->root->sasl.callbacks)
 572     {
 573       rc= memcached_sasl_authenticate_connection(ptr);
 574       if (rc != MEMCACHED_SUCCESS)
 575       {
 576         (void)closesocket(ptr->fd);
 577         ptr->fd= INVALID_SOCKET;
 578       }
 579     }
 580 #endif
 581     break;
 582   case MEMCACHED_CONNECTION_UNIX_SOCKET:
 583     rc= unix_socket_connect(ptr);
 584     break;
 585   case MEMCACHED_CONNECTION_MAX:
 586   default:
 587     WATCHPOINT_ASSERT(0);
 588   }
 589
 590   if (rc == MEMCACHED_SUCCESS)
 591   {
 592     ptr->server_failure_counter= 0;
 593     ptr->next_retry= 0;
 594   }
 595   else
 596   {
 597     ptr->server_failure_counter++;
 598
 599     set_last_disconnected_host(ptr);
 600   }
 601
 602   LIBMEMCACHED_MEMCACHED_CONNECT_END();
 603
 604   return rc;
 605 }