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   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   int errcode;
 149   switch(errcode= getaddrinfo(server->hostname, str_port, &hints, &server->address_info))
 150   {
 151   case 0:
 152     break;
 153
 154   case EAI_AGAIN:
 155     return memcached_set_error_string(*server, MEMCACHED_TIMEOUT, gai_strerror(errcode), strlen(gai_strerror(errcode)));
 156
 157   case EAI_SYSTEM:
 158     {
 159       static memcached_string_t mesg= { memcached_literal_param("getaddrinfo") };
 160       return memcached_set_errno(*server, errno, &mesg);
 161     }
 162   case EAI_BADFLAGS:
 163     return memcached_set_error_string(*server, MEMCACHED_INVALID_ARGUMENTS,  memcached_literal_param("getaddrinfo(EAI_BADFLAGS)"));
 164
 165   case EAI_MEMORY:
 166     return memcached_set_error_string(*server, MEMCACHED_ERRNO,  memcached_literal_param("getaddrinfo(EAI_MEMORY)"));
 167
 168   default:
 169     {
 170       WATCHPOINT_STRING(server->hostname);
 171       WATCHPOINT_STRING(gai_strerror(e));
 172       return memcached_set_error_string(*server, MEMCACHED_HOST_LOOKUP_FAILURE, gai_strerror(errcode), strlen(gai_strerror(errcode)));
 173     }
 174   }
 175   server->address_info_next= server->address_info;
 176
 177   return MEMCACHED_SUCCESS;
 178 }
 179
 180 static inline memcached_return_t set_socket_nonblocking(memcached_server_st *ptr)
 181 {
 182 #ifdef WIN32
 183   u_long arg = 1;
 184   if (ioctlsocket(ptr->fd, FIONBIO, &arg) == SOCKET_ERROR)
 185   {
 186     return memcached_set_errno(*ptr, get_socket_errno(), NULL);
 187   }
 188 #else
 189   int flags;
 190
 191   do
 192   {
 193     flags= fcntl(ptr->fd, F_GETFL, 0);
 194   }
 195   while (flags == -1 && (errno == EINTR || errno == EAGAIN));
 196
 197   unlikely (flags == -1)
 198   {
 199     return memcached_set_errno(*ptr, errno, NULL);
 200   }
 201   else if ((flags & O_NONBLOCK) == 0)
 202   {
 203     int rval;
 204
 205     do
 206     {
 207       rval= fcntl(ptr->fd, F_SETFL, flags | O_NONBLOCK);
 208     }
 209     while (rval == -1 && (errno == EINTR || errno == EAGAIN));
 210
 211     unlikely (rval == -1)
 212     {
 213       return memcached_set_errno(*ptr, errno, NULL);
 214     }
 215   }
 216 #endif
 217   return MEMCACHED_SUCCESS;
 218 }
 219
 220 static memcached_return_t set_socket_options(memcached_server_st *ptr)
 221 {
 222   WATCHPOINT_ASSERT(ptr->fd != -1);
 223
 224   if (ptr->type == MEMCACHED_CONNECTION_UDP)
 225     return MEMCACHED_SUCCESS;
 226
 227 #ifdef HAVE_SNDTIMEO
 228   if (ptr->root->snd_timeout)
 229   {
 230     int error;
 231     struct timeval waittime;
 232
 233     waittime.tv_sec= 0;
 234     waittime.tv_usec= ptr->root->snd_timeout;
 235
 236     error= setsockopt(ptr->fd, SOL_SOCKET, SO_SNDTIMEO,
 237                       &waittime, (socklen_t)sizeof(struct timeval));
 238     WATCHPOINT_ASSERT(error == 0);
 239     if (error)
 240       return MEMCACHED_FAILURE;
 241   }
 242 #endif
 243
 244 #ifdef HAVE_RCVTIMEO
 245   if (ptr->root->rcv_timeout)
 246   {
 247     int error;
 248     struct timeval waittime;
 249
 250     waittime.tv_sec= 0;
 251     waittime.tv_usec= ptr->root->rcv_timeout;
 252
 253     error= setsockopt(ptr->fd, SOL_SOCKET, SO_RCVTIMEO,
 254                       &waittime, (socklen_t)sizeof(struct timeval));
 255     WATCHPOINT_ASSERT(error == 0);
 256     if (error)
 257       return MEMCACHED_FAILURE;
 258   }
 259 #endif
 260
 261
 262 #if defined(__MACH__) && defined(__APPLE__) || defined(__FreeBSD__)
 263   {
 264     int set = 1;
 265     int error= setsockopt(ptr->fd, SOL_SOCKET, SO_NOSIGPIPE, (void *)&set, sizeof(int));
 266
 267     // This is not considered a fatal error
 268     if (error == -1)
 269     {
 270       WATCHPOINT_ERRNO(get_socket_errno());
 271       perror("setsockopt(SO_NOSIGPIPE)");
 272     }
 273   }
 274 #endif
 275
 276   if (ptr->root->flags.no_block)
 277   {
 278     int error;
 279     struct linger linger;
 280
 281     linger.l_onoff= 1;
 282     linger.l_linger= 0; /* By default on close() just drop the socket */
 283     error= setsockopt(ptr->fd, SOL_SOCKET, SO_LINGER,
 284                       &linger, (socklen_t)sizeof(struct linger));
 285     WATCHPOINT_ASSERT(error == 0);
 286     if (error)
 287       return MEMCACHED_FAILURE;
 288   }
 289
 290   if (ptr->root->flags.tcp_nodelay)
 291   {
 292     int flag= 1;
 293     int error;
 294
 295     error= setsockopt(ptr->fd, IPPROTO_TCP, TCP_NODELAY,
 296                       &flag, (socklen_t)sizeof(int));
 297     WATCHPOINT_ASSERT(error == 0);
 298     if (error)
 299       return MEMCACHED_FAILURE;
 300   }
 301
 302   if (ptr->root->flags.tcp_keepalive)
 303   {
 304     int flag= 1;
 305     int error;
 306
 307     error= setsockopt(ptr->fd, SOL_SOCKET, SO_KEEPALIVE,
 308                       &flag, (socklen_t)sizeof(int));
 309     WATCHPOINT_ASSERT(error == 0);
 310     if (error)
 311       return MEMCACHED_FAILURE;
 312   }
 313
 314 #ifdef TCP_KEEPIDLE
 315   if (ptr->root->tcp_keepidle > 0)
 316   {
 317     int error;
 318
 319     error= setsockopt(ptr->fd, IPPROTO_TCP, TCP_KEEPIDLE,
 320                       &ptr->root->tcp_keepidle, (socklen_t)sizeof(int));
 321     WATCHPOINT_ASSERT(error == 0);
 322     if (error)
 323       return MEMCACHED_FAILURE;
 324   }
 325 #endif
 326
 327   if (ptr->root->send_size > 0)
 328   {
 329     int error;
 330
 331     error= setsockopt(ptr->fd, SOL_SOCKET, SO_SNDBUF,
 332                       &ptr->root->send_size, (socklen_t)sizeof(int));
 333     WATCHPOINT_ASSERT(error == 0);
 334     if (error)
 335       return MEMCACHED_FAILURE;
 336   }
 337
 338   if (ptr->root->recv_size > 0)
 339   {
 340     int error;
 341
 342     error= setsockopt(ptr->fd, SOL_SOCKET, SO_RCVBUF,
 343                       &ptr->root->recv_size, (socklen_t)sizeof(int));
 344     WATCHPOINT_ASSERT(error == 0);
 345     if (error)
 346       return MEMCACHED_FAILURE;
 347   }
 348
 349
 350   /* libmemcached will always use nonblocking IO to avoid write deadlocks */
 351   return set_socket_nonblocking(ptr);
 352 }
 353
 354 static memcached_return_t unix_socket_connect(memcached_server_st *ptr)
 355 {
 356 #ifndef WIN32
 357   WATCHPOINT_ASSERT(ptr->fd == -1);
 358
 359   if ((ptr->fd= socket(AF_UNIX, SOCK_STREAM, 0)) < 0)
 360   {
 361     return memcached_set_errno(*ptr, errno, NULL);
 362   }
 363
 364   struct sockaddr_un servAddr;
 365
 366   memset(&servAddr, 0, sizeof (struct sockaddr_un));
 367   servAddr.sun_family= AF_UNIX;
 368   strncpy(servAddr.sun_path, ptr->hostname, sizeof(servAddr.sun_path)); /* Copy filename */
 369
 370 test_connect:
 371   if (connect(ptr->fd,
 372               (struct sockaddr *)&servAddr,
 373               sizeof(servAddr)) < 0)
 374   {
 375     switch (errno)
 376     {
 377     case EINPROGRESS:
 378     case EALREADY:
 379     case EINTR:
 380       goto test_connect;
 381     case EISCONN: /* We were spinning waiting on connect */
 382       break;
 383     default:
 384       WATCHPOINT_ERRNO(errno);
 385       ptr->cached_errno= errno;
 386       return MEMCACHED_ERRNO;
 387     }
 388   }
 389
 390   WATCHPOINT_ASSERT(ptr->fd != INVALID_SOCKET);
 391
 392   return MEMCACHED_SUCCESS;
 393 #else
 394   (void)ptr;
 395   return MEMCACHED_NOT_SUPPORTED;
 396 #endif
 397 }
 398
 399 static memcached_return_t network_connect(memcached_server_st *ptr)
 400 {
 401   bool timeout_error_occured= false;
 402
 403   WATCHPOINT_ASSERT(ptr->fd == INVALID_SOCKET);
 404   WATCHPOINT_ASSERT(ptr->cursor_active == 0);
 405
 406   if (not ptr->address_info)
 407   {
 408     memcached_return_t rc;
 409     uint32_t counter= 5;
 410     while (--counter)
 411     {
 412       if ((rc= set_hostinfo(ptr)) != MEMCACHED_TIMEOUT)
 413         break;
 414
 415 #ifndef WIN32
 416       struct timespec dream, rem;
 417
 418       dream.tv_nsec= 1000;
 419       dream.tv_sec= 0;
 420
 421       nanosleep(&dream, &rem);
 422 #endif
 423     }
 424
 425     if (memcached_failed(rc))
 426       return rc;
 427   }
 428
 429   /* Create the socket */
 430   while (ptr->address_info_next && ptr->fd == INVALID_SOCKET)
 431   {
 432     /* Memcache server does not support IPV6 in udp mode, so skip if not ipv4 */
 433     if (ptr->type == MEMCACHED_CONNECTION_UDP && ptr->address_info_next->ai_family != AF_INET)
 434     {
 435       ptr->address_info_next= ptr->address_info_next->ai_next;
 436       continue;
 437     }
 438
 439     if ((ptr->fd= socket(ptr->address_info_next->ai_family,
 440                          ptr->address_info_next->ai_socktype,
 441                          ptr->address_info_next->ai_protocol)) < 0)
 442     {
 443       return memcached_set_errno(*ptr, get_socket_errno(), NULL);
 444     }
 445
 446     (void)set_socket_options(ptr);
 447
 448     /* connect to server */
 449     if ((connect(ptr->fd, ptr->address_info_next->ai_addr, ptr->address_info_next->ai_addrlen) != SOCKET_ERROR))
 450     {
 451       break; // Success
 452     }
 453
 454     /* An error occurred */
 455     ptr->cached_errno= get_socket_errno();
 456     switch (ptr->cached_errno)
 457     {
 458     case EWOULDBLOCK:
 459     case EINPROGRESS: // nonblocking mode - first return
 460     case EALREADY: // nonblocking mode - subsequent returns
 461       {
 462         memcached_return_t rc;
 463         rc= connect_poll(ptr);
 464
 465         if (rc == MEMCACHED_TIMEOUT)
 466           timeout_error_occured= true;
 467
 468         if (rc == MEMCACHED_SUCCESS)
 469           break;
 470       }
 471
 472     case EISCONN: // we are connected :-)
 473       break;
 474
 475     case EINTR: // Special case, we retry ai_addr
 476       (void)closesocket(ptr->fd);
 477       ptr->fd= INVALID_SOCKET;
 478       continue;
 479
 480     default:
 481       (void)closesocket(ptr->fd);
 482       ptr->fd= INVALID_SOCKET;
 483       ptr->address_info_next= ptr->address_info_next->ai_next;
 484       break;
 485     }
 486   }
 487
 488   if (ptr->fd == INVALID_SOCKET)
 489   {
 490     WATCHPOINT_STRING("Never got a good file descriptor");
 491
 492     /* Failed to connect. schedule next retry */
 493     if (ptr->root->retry_timeout)
 494     {
 495       struct timeval next_time;
 496
 497       if (gettimeofday(&next_time, NULL) == 0)
 498         ptr->next_retry= next_time.tv_sec + ptr->root->retry_timeout;
 499     }
 500
 501     if (timeout_error_occured)
 502       return MEMCACHED_TIMEOUT;
 503
 504     return MEMCACHED_ERRNO; /* The last error should be from connect() */
 505   }
 506
 507   return MEMCACHED_SUCCESS; /* The last error should be from connect() */
 508 }
 509
 510 void set_last_disconnected_host(memcached_server_write_instance_st ptr)
 511 {
 512   // const_cast
 513   memcached_st *root= (memcached_st *)ptr->root;
 514
 515 #if 0
 516   WATCHPOINT_STRING(ptr->hostname);
 517   WATCHPOINT_NUMBER(ptr->port);
 518   WATCHPOINT_ERRNO(ptr->cached_errno);
 519 #endif
 520   if (root->last_disconnected_server)
 521     memcached_server_free(root->last_disconnected_server);
 522   root->last_disconnected_server= memcached_server_clone(NULL, ptr);
 523 }
 524
 525 memcached_return_t memcached_connect(memcached_server_write_instance_st ptr)
 526 {
 527   memcached_return_t rc= MEMCACHED_NO_SERVERS;
 528
 529   if (ptr->fd != INVALID_SOCKET)
 530     return MEMCACHED_SUCCESS;
 531
 532   LIBMEMCACHED_MEMCACHED_CONNECT_START();
 533
 534   /* both retry_timeout and server_failure_limit must be set in order to delay retrying a server on error. */
 535   WATCHPOINT_ASSERT(ptr->root);
 536   if (ptr->root->retry_timeout && ptr->next_retry)
 537   {
 538     struct timeval curr_time;
 539
 540     gettimeofday(&curr_time, NULL);
 541
 542     // We should optimize this to remove the allocation if the server was
 543     // the last server to die
 544     if (ptr->next_retry > curr_time.tv_sec)
 545     {
 546       set_last_disconnected_host(ptr);
 547
 548       return MEMCACHED_SERVER_MARKED_DEAD;
 549     }
 550   }
 551
 552   // If we are over the counter failure, we just fail. Reject host only
 553   // works if you have a set number of failures.
 554   if (ptr->root->server_failure_limit && ptr->server_failure_counter >= ptr->root->server_failure_limit)
 555   {
 556     set_last_disconnected_host(ptr);
 557
 558     // @todo fix this by fixing behavior to no longer make use of
 559     // memcached_st
 560     if (_is_auto_eject_host(ptr->root))
 561     {
 562       run_distribution((memcached_st *)ptr->root);
 563     }
 564
 565     return MEMCACHED_SERVER_MARKED_DEAD;
 566   }
 567
 568   /* We need to clean up the multi startup piece */
 569   switch (ptr->type)
 570   {
 571   case MEMCACHED_CONNECTION_UNKNOWN:
 572     WATCHPOINT_ASSERT(0);
 573     rc= MEMCACHED_NOT_SUPPORTED;
 574     break;
 575   case MEMCACHED_CONNECTION_UDP:
 576   case MEMCACHED_CONNECTION_TCP:
 577     rc= network_connect(ptr);
 578 #ifdef LIBMEMCACHED_WITH_SASL_SUPPORT
 579     if (ptr->fd != INVALID_SOCKET && ptr->root->sasl.callbacks)
 580     {
 581       rc= memcached_sasl_authenticate_connection(ptr);
 582       if (memcached_failed(rc))
 583       {
 584         (void)closesocket(ptr->fd);
 585         ptr->fd= INVALID_SOCKET;
 586       }
 587     }
 588 #endif
 589     break;
 590   case MEMCACHED_CONNECTION_UNIX_SOCKET:
 591     rc= unix_socket_connect(ptr);
 592     break;
 593   case MEMCACHED_CONNECTION_MAX:
 594   default:
 595     WATCHPOINT_ASSERT(0);
 596   }
 597
 598   if (rc == MEMCACHED_SUCCESS)
 599   {
 600     ptr->server_failure_counter= 0;
 601     ptr->next_retry= 0;
 602   }
 603   else
 604   {
 605     ptr->server_failure_counter++;
 606
 607     set_last_disconnected_host(ptr);
 608   }
 609
 610   LIBMEMCACHED_MEMCACHED_CONNECT_END();
 611
 612   return rc;
 613 }