lib/socket-util.c

   1 /*
   2  * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014, 2015, 2016 Nicira, Inc.
   3  *
   4  * Licensed under the Apache License, Version 2.0 (the "License");
   5  * you may not use this file except in compliance with the License.
   6  * You may obtain a copy of the License at:
   7  *
   8  *     http://www.apache.org/licenses/LICENSE-2.0
   9  *
  10  * Unless required by applicable law or agreed to in writing, software
  11  * distributed under the License is distributed on an "AS IS" BASIS,
  12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13  * See the License for the specific language governing permissions and
  14  * limitations under the License.
  15  */
  16
  17 #include <config.h>
  18 #include "socket-util.h"
  19 #include <arpa/inet.h>
  20 #include <ctype.h>
  21 #include <errno.h>
  22 #include <fcntl.h>
  23 #include <net/if.h>
  24 #include <netdb.h>
  25 #include <netinet/tcp.h>
  26 #include <poll.h>
  27 #include <stddef.h>
  28 #include <stdio.h>
  29 #include <stdlib.h>
  30 #include <string.h>
  31 #include <sys/socket.h>
  32 #include <sys/stat.h>
  33 #include <sys/uio.h>
  34 #include <sys/un.h>
  35 #include <unistd.h>
  36 #include "openvswitch/dynamic-string.h"
  37 #include "ovs-thread.h"
  38 #include "packets.h"
  39 #include "poll-loop.h"
  40 #include "util.h"
  41 #include "openvswitch/vlog.h"
  42 #ifdef __linux__
  43 #include <linux/if_packet.h>
  44 #endif
  45 #ifdef HAVE_NETLINK
  46 #include "netlink-protocol.h"
  47 #include "netlink-socket.h"
  48 #endif
  49
  50 VLOG_DEFINE_THIS_MODULE(socket_util);
  51
  52 static int getsockopt_int(int fd, int level, int option, const char *optname,
  53                           int *valuep);
  54
  55 /* Sets 'fd' to non-blocking mode.  Returns 0 if successful, otherwise a
  56  * positive errno value. */
  57 int
  58 set_nonblocking(int fd)
  59 {
  60 #ifndef _WIN32
  61     int flags = fcntl(fd, F_GETFL, 0);
  62     if (flags != -1) {
  63         if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) != -1) {
  64             return 0;
  65         } else {
  66             VLOG_ERR("fcntl(F_SETFL) failed: %s", ovs_strerror(errno));
  67             return errno;
  68         }
  69     } else {
  70         VLOG_ERR("fcntl(F_GETFL) failed: %s", ovs_strerror(errno));
  71         return errno;
  72     }
  73 #else
  74     unsigned long arg = 1;
  75     if (ioctlsocket(fd, FIONBIO, &arg)) {
  76         int error = sock_errno();
  77         VLOG_ERR("set_nonblocking failed: %s", sock_strerror(error));
  78         return error;
  79     }
  80     return 0;
  81 #endif
  82 }
  83
  84 void
  85 xset_nonblocking(int fd)
  86 {
  87     if (set_nonblocking(fd)) {
  88         exit(EXIT_FAILURE);
  89     }
  90 }
  91
  92 void
  93 setsockopt_tcp_nodelay(int fd)
  94 {
  95     int on = 1;
  96     int retval;
  97
  98     retval = setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, &on, sizeof on);
  99     if (retval) {
 100         retval = sock_errno();
 101         VLOG_ERR("setsockopt(TCP_NODELAY): %s", sock_strerror(retval));
 102     }
 103 }
 104
 105 /* Sets the DSCP value of socket 'fd' to 'dscp', which must be 63 or less.
 106  * 'family' must indicate the socket's address family (AF_INET or AF_INET6, to
 107  * do anything useful). */
 108 int
 109 set_dscp(int fd, int family, uint8_t dscp)
 110 {
 111     int retval;
 112     int val;
 113
 114 #ifdef _WIN32
 115     /* XXX: Consider using QoS2 APIs for Windows to set dscp. */
 116     return 0;
 117 #endif
 118
 119     if (dscp > 63) {
 120         return EINVAL;
 121     }
 122     val = dscp << 2;
 123
 124     switch (family) {
 125     case AF_INET:
 126         retval = setsockopt(fd, IPPROTO_IP, IP_TOS, &val, sizeof val);
 127         break;
 128
 129     case AF_INET6:
 130         retval = setsockopt(fd, IPPROTO_IPV6, IPV6_TCLASS, &val, sizeof val);
 131         break;
 132
 133     default:
 134         return ENOPROTOOPT;
 135     }
 136
 137     return retval ? sock_errno() : 0;
 138 }
 139
 140 /* Checks whether 'host_name' is an IPv4 or IPv6 address.  It is assumed
 141  * that 'host_name' is valid.  Returns false if it is IPv4 address, true if
 142  * it is IPv6 address. */
 143 bool
 144 addr_is_ipv6(const char *host_name)
 145 {
 146     return strchr(host_name, ':') != NULL;
 147 }
 148
 149 /* Translates 'host_name', which must be a string representation of an IP
 150  * address, into a numeric IP address in '*addr'.  Returns 0 if successful,
 151  * otherwise a positive errno value. */
 152 int
 153 lookup_ip(const char *host_name, struct in_addr *addr)
 154 {
 155     if (!ip_parse(host_name, &addr->s_addr)) {
 156         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 157         VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name);
 158         return ENOENT;
 159     }
 160     return 0;
 161 }
 162
 163 /* Translates 'host_name', which must be a string representation of an IPv6
 164  * address, into a numeric IPv6 address in '*addr'.  Returns 0 if successful,
 165  * otherwise a positive errno value. */
 166 int
 167 lookup_ipv6(const char *host_name, struct in6_addr *addr)
 168 {
 169     if (!ipv6_parse(host_name, addr)) {
 170         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 171         VLOG_ERR_RL(&rl, "\"%s\" is not a valid IPv6 address", host_name);
 172         return ENOENT;
 173     }
 174     return 0;
 175 }
 176
 177 /* Translates 'host_name', which must be a host name or a string representation
 178  * of an IP address, into a numeric IP address in '*addr'.  Returns 0 if
 179  * successful, otherwise a positive errno value.
 180  *
 181  * Most Open vSwitch code should not use this because it causes deadlocks:
 182  * getaddrinfo() sends out a DNS request but that starts a new flow for which
 183  * OVS must set up a flow, but it can't because it's waiting for a DNS reply.
 184  * The synchronous lookup also delays other activity.  (Of course we can solve
 185  * this but it doesn't seem worthwhile quite yet.)  */
 186 int
 187 lookup_hostname(const char *host_name, struct in_addr *addr)
 188 {
 189     struct addrinfo *result;
 190     struct addrinfo hints;
 191
 192     if (ip_parse(host_name, &addr->s_addr)) {
 193         return 0;
 194     }
 195
 196     memset(&hints, 0, sizeof hints);
 197     hints.ai_family = AF_INET;
 198
 199     switch (getaddrinfo(host_name, NULL, &hints, &result)) {
 200     case 0:
 201         *addr = ALIGNED_CAST(struct sockaddr_in *,
 202                              result->ai_addr)->sin_addr;
 203         freeaddrinfo(result);
 204         return 0;
 205
 206 #ifdef EAI_ADDRFAMILY
 207     case EAI_ADDRFAMILY:
 208 #endif
 209     case EAI_NONAME:
 210     case EAI_SERVICE:
 211         return ENOENT;
 212
 213     case EAI_AGAIN:
 214         return EAGAIN;
 215
 216     case EAI_BADFLAGS:
 217     case EAI_FAMILY:
 218     case EAI_SOCKTYPE:
 219         return EINVAL;
 220
 221     case EAI_FAIL:
 222         return EIO;
 223
 224     case EAI_MEMORY:
 225         return ENOMEM;
 226
 227 #if defined (EAI_NODATA) && EAI_NODATA != EAI_NONAME
 228     case EAI_NODATA:
 229         return ENXIO;
 230 #endif
 231
 232 #ifdef EAI_SYSTEM
 233     case EAI_SYSTEM:
 234         return sock_errno();
 235 #endif
 236
 237     default:
 238         return EPROTO;
 239     }
 240 }
 241
 242 int
 243 check_connection_completion(int fd)
 244 {
 245     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
 246     struct pollfd pfd;
 247     int retval;
 248
 249     pfd.fd = fd;
 250     pfd.events = POLLOUT;
 251
 252 #ifndef _WIN32
 253     do {
 254         retval = poll(&pfd, 1, 0);
 255     } while (retval < 0 && errno == EINTR);
 256 #else
 257     fd_set wrset, exset;
 258     FD_ZERO(&wrset);
 259     FD_ZERO(&exset);
 260     FD_SET(fd, &exset);
 261     FD_SET(fd, &wrset);
 262     pfd.revents = 0;
 263     struct timeval tv = { 0, 0 };
 264     /* WSAPoll is broken on Windows, instead do a select */
 265     retval = select(0, NULL, &wrset, &exset, &tv);
 266     if (retval == 1) {
 267         if (FD_ISSET(fd, &wrset)) {
 268             pfd.revents |= pfd.events;
 269         }
 270         if (FD_ISSET(fd, &exset)) {
 271             pfd.revents |= POLLERR;
 272         }
 273     }
 274 #endif
 275     if (retval == 1) {
 276         if (pfd.revents & POLLERR) {
 277             ssize_t n = send(fd, "", 1, 0);
 278             if (n < 0) {
 279                 return sock_errno();
 280             } else {
 281                 VLOG_ERR_RL(&rl, "poll return POLLERR but send succeeded");
 282                 return EPROTO;
 283             }
 284         }
 285         return 0;
 286     } else if (retval < 0) {
 287         VLOG_ERR_RL(&rl, "poll: %s", sock_strerror(sock_errno()));
 288         return errno;
 289     } else {
 290         return EAGAIN;
 291     }
 292 }
 293
 294 /* Returns the size of socket 'sock''s receive buffer (SO_RCVBUF), or a
 295  * negative errno value if an error occurs. */
 296 int
 297 get_socket_rcvbuf(int sock)
 298 {
 299     int rcvbuf;
 300     int error;
 301
 302     error = getsockopt_int(sock, SOL_SOCKET, SO_RCVBUF, "SO_RCVBUF", &rcvbuf);
 303     return error ? -error : rcvbuf;
 304 }
 305
 306 /* Reads and discards up to 'n' datagrams from 'fd', stopping as soon as no
 307  * more data can be immediately read.  ('fd' should therefore be in
 308  * non-blocking mode.)*/
 309 void
 310 drain_fd(int fd, size_t n_packets)
 311 {
 312     for (; n_packets > 0; n_packets--) {
 313         /* 'buffer' only needs to be 1 byte long in most circumstances.  This
 314          * size is defensive against the possibility that we someday want to
 315          * use a Linux tap device without TUN_NO_PI, in which case a buffer
 316          * smaller than sizeof(struct tun_pi) will give EINVAL on read. */
 317         char buffer[128];
 318         if (read(fd, buffer, sizeof buffer) <= 0) {
 319             break;
 320         }
 321     }
 322 }
 323
 324 ovs_be32
 325 guess_netmask(ovs_be32 ip_)
 326 {
 327     uint32_t ip = ntohl(ip_);
 328     return ((ip >> 31) == 0 ? htonl(0xff000000)   /* Class A */
 329             : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */
 330             : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */
 331             : htonl(0));                          /* ??? */
 332 }
 333
 334 /* This is like strsep() except:
 335  *
 336  *    - The separator string is ":".
 337  *
 338  *    - Square brackets [] quote ":" separators and are removed from the
 339  *      tokens. */
 340 char *
 341 inet_parse_token(char **pp)
 342 {
 343     char *p = *pp;
 344
 345     if (p == NULL) {
 346         return NULL;
 347     } else if (*p == '\0') {
 348         *pp = NULL;
 349         return p;
 350     } else if (*p == '[') {
 351         char *start = p + 1;
 352         char *end = start + strcspn(start, "]");
 353         *pp = (*end == '\0' ? NULL
 354                : end[1] == ':' ? end + 2
 355                : end + 1);
 356         *end = '\0';
 357         return start;
 358     } else {
 359         char *start = p;
 360         char *end = start + strcspn(start, ":");
 361         *pp = *end == '\0' ? NULL : end + 1;
 362         *end = '\0';
 363         return start;
 364     }
 365 }
 366
 367 static bool
 368 parse_sockaddr_components(struct sockaddr_storage *ss,
 369                           char *host_s,
 370                           const char *port_s, uint16_t default_port,
 371                           const char *s)
 372 {
 373     struct sockaddr_in *sin = ALIGNED_CAST(struct sockaddr_in *, ss);
 374     int port;
 375
 376     if (port_s && port_s[0]) {
 377         if (!str_to_int(port_s, 10, &port) || port < 0 || port > 65535) {
 378             VLOG_ERR("%s: bad port number \"%s\"", s, port_s);
 379             goto exit;
 380         }
 381     } else {
 382         port = default_port;
 383     }
 384
 385     memset(ss, 0, sizeof *ss);
 386     if (host_s && strchr(host_s, ':')) {
 387         struct sockaddr_in6 *sin6
 388             = ALIGNED_CAST(struct sockaddr_in6 *, ss);
 389
 390         char *addr = strsep(&host_s, "%");
 391
 392         sin6->sin6_family = AF_INET6;
 393         sin6->sin6_port = htons(port);
 394         if (!addr || !*addr || !ipv6_parse(addr, &sin6->sin6_addr)) {
 395             VLOG_ERR("%s: bad IPv6 address \"%s\"", s, addr ? addr : "");
 396             goto exit;
 397         }
 398
 399 #ifdef HAVE_STRUCT_SOCKADDR_IN6_SIN6_SCOPE_ID
 400         char *scope = strsep(&host_s, "%");
 401         if (scope && *scope) {
 402             if (!scope[strspn(scope, "0123456789")]) {
 403                 sin6->sin6_scope_id = atoi(scope);
 404             } else {
 405                 sin6->sin6_scope_id = if_nametoindex(scope);
 406                 if (!sin6->sin6_scope_id) {
 407                     VLOG_ERR("%s: bad IPv6 scope \"%s\" (%s)",
 408                              s, scope, ovs_strerror(errno));
 409                     goto exit;
 410                 }
 411             }
 412         }
 413 #endif
 414     } else {
 415         sin->sin_family = AF_INET;
 416         sin->sin_port = htons(port);
 417         if (host_s && !ip_parse(host_s, &sin->sin_addr.s_addr)) {
 418             VLOG_ERR("%s: bad IPv4 address \"%s\"", s, host_s);
 419             goto exit;
 420         }
 421     }
 422
 423     return true;
 424
 425 exit:
 426     memset(ss, 0, sizeof *ss);
 427     return false;
 428 }
 429
 430 /* Parses 'target', which should be a string in the format "<host>[:<port>]".
 431  * <host>, which is required, may be an IPv4 address or an IPv6 address
 432  * enclosed in square brackets.  If 'default_port' is nonzero then <port> is
 433  * optional and defaults to 'default_port'.
 434  *
 435  * On success, returns true and stores the parsed remote address into '*ss'.
 436  * On failure, logs an error, stores zeros into '*ss', and returns false. */
 437 bool
 438 inet_parse_active(const char *target_, uint16_t default_port,
 439                   struct sockaddr_storage *ss)
 440 {
 441     char *target = xstrdup(target_);
 442     const char *port;
 443     char *host;
 444     char *p;
 445     bool ok;
 446
 447     p = target;
 448     host = inet_parse_token(&p);
 449     port = inet_parse_token(&p);
 450     if (!host) {
 451         VLOG_ERR("%s: host must be specified", target_);
 452         ok = false;
 453     } else if (!port && !default_port) {
 454         VLOG_ERR("%s: port must be specified", target_);
 455         ok = false;
 456     } else {
 457         ok = parse_sockaddr_components(ss, host, port, default_port, target_);
 458     }
 459     if (!ok) {
 460         memset(ss, 0, sizeof *ss);
 461     }
 462     free(target);
 463     return ok;
 464 }
 465
 466
 467 /* Opens a non-blocking IPv4 or IPv6 socket of the specified 'style' and
 468  * connects to 'target', which should be a string in the format
 469  * "<host>[:<port>]".  <host>, which is required, may be an IPv4 address or an
 470  * IPv6 address enclosed in square brackets.  If 'default_port' is nonzero then
 471  * <port> is optional and defaults to 'default_port'.
 472  *
 473  * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
 474  *
 475  * On success, returns 0 (indicating connection complete) or EAGAIN (indicating
 476  * connection in progress), in which case the new file descriptor is stored
 477  * into '*fdp'.  On failure, returns a positive errno value other than EAGAIN
 478  * and stores -1 into '*fdp'.
 479  *
 480  * If 'ss' is non-null, then on success stores the target address into '*ss'.
 481  *
 482  * 'dscp' becomes the DSCP bits in the IP headers for the new connection.  It
 483  * should be in the range [0, 63] and will automatically be shifted to the
 484  * appropriately place in the IP tos field. */
 485 int
 486 inet_open_active(int style, const char *target, uint16_t default_port,
 487                  struct sockaddr_storage *ssp, int *fdp, uint8_t dscp)
 488 {
 489     struct sockaddr_storage ss;
 490     int fd = -1;
 491     int error;
 492
 493     /* Parse. */
 494     if (!inet_parse_active(target, default_port, &ss)) {
 495         error = EAFNOSUPPORT;
 496         goto exit;
 497     }
 498
 499     /* Create non-blocking socket. */
 500     fd = socket(ss.ss_family, style, 0);
 501     if (fd < 0) {
 502         error = sock_errno();
 503         VLOG_ERR("%s: socket: %s", target, sock_strerror(error));
 504         goto exit;
 505     }
 506     error = set_nonblocking(fd);
 507     if (error) {
 508         goto exit;
 509     }
 510
 511     /* The dscp bits must be configured before connect() to ensure that the
 512      * TOS field is set during the connection establishment.  If set after
 513      * connect(), the handshake SYN frames will be sent with a TOS of 0. */
 514     error = set_dscp(fd, ss.ss_family, dscp);
 515     if (error) {
 516         VLOG_ERR("%s: set_dscp: %s", target, sock_strerror(error));
 517         goto exit;
 518     }
 519
 520     /* Connect. */
 521     error = connect(fd, (struct sockaddr *) &ss, ss_length(&ss)) == 0
 522                     ? 0
 523                     : sock_errno();
 524     if (error == EINPROGRESS
 525 #ifdef _WIN32
 526         || error == WSAEALREADY || error == WSAEWOULDBLOCK
 527 #endif
 528         ) {
 529         error = EAGAIN;
 530     }
 531
 532 exit:
 533     if (error && error != EAGAIN) {
 534         if (ssp) {
 535             memset(ssp, 0, sizeof *ssp);
 536         }
 537         if (fd >= 0) {
 538             closesocket(fd);
 539             fd = -1;
 540         }
 541     } else {
 542         if (ssp) {
 543             *ssp = ss;
 544         }
 545     }
 546     *fdp = fd;
 547     return error;
 548 }
 549
 550 /* Parses 'target', which should be a string in the format "[<port>][:<host>]":
 551  *
 552  *      - If 'default_port' is -1, then <port> is required.  Otherwise, if
 553  *        <port> is omitted, then 'default_port' is used instead.
 554  *
 555  *      - If <port> (or 'default_port', if used) is 0, then no port is bound
 556  *        and the TCP/IP stack will select a port.
 557  *
 558  *      - <host> is optional.  If supplied, it may be an IPv4 address or an
 559  *        IPv6 address enclosed in square brackets.  If omitted, the IP address
 560  *        is wildcarded.
 561  *
 562  * If successful, stores the address into '*ss' and returns true; otherwise
 563  * zeros '*ss' and returns false. */
 564 bool
 565 inet_parse_passive(const char *target_, int default_port,
 566                    struct sockaddr_storage *ss)
 567 {
 568     char *target = xstrdup(target_);
 569     const char *port;
 570     char *host;
 571     char *p;
 572     bool ok;
 573
 574     p = target;
 575     port = inet_parse_token(&p);
 576     host = inet_parse_token(&p);
 577     if (!port && default_port < 0) {
 578         VLOG_ERR("%s: port must be specified", target_);
 579         ok = false;
 580     } else {
 581         ok = parse_sockaddr_components(ss, host, port, default_port, target_);
 582     }
 583     if (!ok) {
 584         memset(ss, 0, sizeof *ss);
 585     }
 586     free(target);
 587     return ok;
 588 }
 589
 590
 591 /* Opens a non-blocking IPv4 or IPv6 socket of the specified 'style', binds to
 592  * 'target', and listens for incoming connections.  Parses 'target' in the same
 593  * way was inet_parse_passive().
 594  *
 595  * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
 596  *
 597  * For TCP, the socket will have SO_REUSEADDR turned on.
 598  *
 599  * On success, returns a non-negative file descriptor.  On failure, returns a
 600  * negative errno value.
 601  *
 602  * If 'ss' is non-null, then on success stores the bound address into '*ss'.
 603  *
 604  * 'dscp' becomes the DSCP bits in the IP headers for the new connection.  It
 605  * should be in the range [0, 63] and will automatically be shifted to the
 606  * appropriately place in the IP tos field.
 607  *
 608  * If 'kernel_print_port' is true and the port is dynamically assigned by
 609  * the kernel, print the chosen port. */
 610 int
 611 inet_open_passive(int style, const char *target, int default_port,
 612                   struct sockaddr_storage *ssp, uint8_t dscp,
 613                   bool kernel_print_port)
 614 {
 615     bool kernel_chooses_port;
 616     struct sockaddr_storage ss;
 617     int fd = 0, error;
 618     unsigned int yes = 1;
 619
 620     if (!inet_parse_passive(target, default_port, &ss)) {
 621         return -EAFNOSUPPORT;
 622     }
 623     kernel_chooses_port = ss_get_port(&ss) == 0;
 624
 625     /* Create non-blocking socket, set SO_REUSEADDR. */
 626     fd = socket(ss.ss_family, style, 0);
 627     if (fd < 0) {
 628         error = sock_errno();
 629         VLOG_ERR("%s: socket: %s", target, sock_strerror(error));
 630         return -error;
 631     }
 632     error = set_nonblocking(fd);
 633     if (error) {
 634         goto error;
 635     }
 636     if (style == SOCK_STREAM
 637         && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) {
 638         error = sock_errno();
 639         VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s",
 640                  target, sock_strerror(error));
 641         goto error;
 642     }
 643
 644     /* Bind. */
 645     if (bind(fd, (struct sockaddr *) &ss, ss_length(&ss)) < 0) {
 646         error = sock_errno();
 647         VLOG_ERR("%s: bind: %s", target, sock_strerror(error));
 648         goto error;
 649     }
 650
 651     /* The dscp bits must be configured before connect() to ensure that the TOS
 652      * field is set during the connection establishment.  If set after
 653      * connect(), the handshake SYN frames will be sent with a TOS of 0. */
 654     error = set_dscp(fd, ss.ss_family, dscp);
 655     if (error) {
 656         VLOG_ERR("%s: set_dscp: %s", target, sock_strerror(error));
 657         goto error;
 658     }
 659
 660     /* Listen. */
 661     if (style == SOCK_STREAM && listen(fd, 10) < 0) {
 662         error = sock_errno();
 663         VLOG_ERR("%s: listen: %s", target, sock_strerror(error));
 664         goto error;
 665     }
 666
 667     if (ssp || kernel_chooses_port) {
 668         socklen_t ss_len = sizeof ss;
 669         if (getsockname(fd, (struct sockaddr *) &ss, &ss_len) < 0) {
 670             error = sock_errno();
 671             VLOG_ERR("%s: getsockname: %s", target, sock_strerror(error));
 672             goto error;
 673         }
 674         if (kernel_chooses_port && kernel_print_port) {
 675             VLOG_INFO("%s: listening on port %"PRIu16,
 676                       target, ss_get_port(&ss));
 677         }
 678         if (ssp) {
 679             *ssp = ss;
 680         }
 681     }
 682
 683     return fd;
 684
 685 error:
 686     if (ssp) {
 687         memset(ssp, 0, sizeof *ssp);
 688     }
 689     closesocket(fd);
 690     return -error;
 691 }
 692
 693 int
 694 read_fully(int fd, void *p_, size_t size, size_t *bytes_read)
 695 {
 696     uint8_t *p = p_;
 697
 698     *bytes_read = 0;
 699     while (size > 0) {
 700         ssize_t retval = read(fd, p, size);
 701         if (retval > 0) {
 702             *bytes_read += retval;
 703             size -= retval;
 704             p += retval;
 705         } else if (retval == 0) {
 706             return EOF;
 707         } else if (errno != EINTR) {
 708             return errno;
 709         }
 710     }
 711     return 0;
 712 }
 713
 714 int
 715 write_fully(int fd, const void *p_, size_t size, size_t *bytes_written)
 716 {
 717     const uint8_t *p = p_;
 718
 719     *bytes_written = 0;
 720     while (size > 0) {
 721         ssize_t retval = write(fd, p, size);
 722         if (retval > 0) {
 723             *bytes_written += retval;
 724             size -= retval;
 725             p += retval;
 726         } else if (retval == 0) {
 727             VLOG_WARN("write returned 0");
 728             return EPROTO;
 729         } else if (errno != EINTR) {
 730             return errno;
 731         }
 732     }
 733     return 0;
 734 }
 735
 736 /* Given file name 'file_name', fsyncs the directory in which it is contained.
 737  * Returns 0 if successful, otherwise a positive errno value. */
 738 int
 739 fsync_parent_dir(const char *file_name)
 740 {
 741     int error = 0;
 742 #ifndef _WIN32
 743     char *dir;
 744     int fd;
 745
 746     dir = dir_name(file_name);
 747     fd = open(dir, O_RDONLY);
 748     if (fd >= 0) {
 749         if (fsync(fd)) {
 750             if (errno == EINVAL || errno == EROFS) {
 751                 /* This directory does not support synchronization.  Not
 752                  * really an error. */
 753             } else {
 754                 error = errno;
 755                 VLOG_ERR("%s: fsync failed (%s)", dir, ovs_strerror(error));
 756             }
 757         }
 758         close(fd);
 759     } else {
 760         error = errno;
 761         VLOG_ERR("%s: open failed (%s)", dir, ovs_strerror(error));
 762     }
 763     free(dir);
 764 #endif
 765
 766     return error;
 767 }
 768
 769 /* Obtains the modification time of the file named 'file_name' to the greatest
 770  * supported precision.  If successful, stores the mtime in '*mtime' and
 771  * returns 0.  On error, returns a positive errno value and stores zeros in
 772  * '*mtime'. */
 773 int
 774 get_mtime(const char *file_name, struct timespec *mtime)
 775 {
 776     struct stat s;
 777
 778     if (!stat(file_name, &s)) {
 779         mtime->tv_sec = s.st_mtime;
 780
 781 #if HAVE_STRUCT_STAT_ST_MTIM_TV_NSEC
 782         mtime->tv_nsec = s.st_mtim.tv_nsec;
 783 #elif HAVE_STRUCT_STAT_ST_MTIMENSEC
 784         mtime->tv_nsec = s.st_mtimensec;
 785 #else
 786         mtime->tv_nsec = 0;
 787 #endif
 788
 789         return 0;
 790     } else {
 791         mtime->tv_sec = mtime->tv_nsec = 0;
 792         return errno;
 793     }
 794 }
 795
 796 static int
 797 getsockopt_int(int fd, int level, int option, const char *optname, int *valuep)
 798 {
 799     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
 800     socklen_t len;
 801     int value;
 802     int error;
 803
 804     len = sizeof value;
 805     if (getsockopt(fd, level, option, &value, &len)) {
 806         error = sock_errno();
 807         VLOG_ERR_RL(&rl, "getsockopt(%s): %s", optname, sock_strerror(error));
 808     } else if (len != sizeof value) {
 809         error = EINVAL;
 810         VLOG_ERR_RL(&rl, "getsockopt(%s): value is %u bytes (expected %"PRIuSIZE")",
 811                     optname, (unsigned int) len, sizeof value);
 812     } else {
 813         error = 0;
 814     }
 815
 816     *valuep = error ? 0 : value;
 817     return error;
 818 }
 819
 820 static void
 821 describe_sockaddr(struct ds *string, int fd,
 822                   int (*getaddr)(int, struct sockaddr *, socklen_t *))
 823 {
 824     struct sockaddr_storage ss;
 825     socklen_t len = sizeof ss;
 826
 827     if (!getaddr(fd, (struct sockaddr *) &ss, &len)) {
 828         if (ss.ss_family == AF_INET || ss.ss_family == AF_INET6) {
 829             ss_format_address(&ss, string);
 830             ds_put_format(string, ":%"PRIu16, ss_get_port(&ss));
 831 #ifndef _WIN32
 832         } else if (ss.ss_family == AF_UNIX) {
 833             struct sockaddr_un sun;
 834             const char *null;
 835             size_t maxlen;
 836
 837             memcpy(&sun, &ss, sizeof sun);
 838             maxlen = len - offsetof(struct sockaddr_un, sun_path);
 839             null = memchr(sun.sun_path, '\0', maxlen);
 840             ds_put_buffer(string, sun.sun_path,
 841                           null ? null - sun.sun_path : maxlen);
 842 #endif
 843         }
 844 #ifdef HAVE_NETLINK
 845         else if (ss.ss_family == AF_NETLINK) {
 846             int protocol;
 847
 848 /* SO_PROTOCOL was introduced in 2.6.32.  Support it regardless of the version
 849  * of the Linux kernel headers in use at build time. */
 850 #ifndef SO_PROTOCOL
 851 #define SO_PROTOCOL 38
 852 #endif
 853
 854             if (!getsockopt_int(fd, SOL_SOCKET, SO_PROTOCOL, "SO_PROTOCOL",
 855                                 &protocol)) {
 856                 switch (protocol) {
 857                 case NETLINK_ROUTE:
 858                     ds_put_cstr(string, "NETLINK_ROUTE");
 859                     break;
 860
 861                 case NETLINK_GENERIC:
 862                     ds_put_cstr(string, "NETLINK_GENERIC");
 863                     break;
 864
 865                 default:
 866                     ds_put_format(string, "AF_NETLINK family %d", protocol);
 867                     break;
 868                 }
 869             } else {
 870                 ds_put_cstr(string, "AF_NETLINK");
 871             }
 872         }
 873 #endif
 874 #if __linux__
 875         else if (ss.ss_family == AF_PACKET) {
 876             struct sockaddr_ll sll;
 877
 878             memcpy(&sll, &ss, sizeof sll);
 879             ds_put_cstr(string, "AF_PACKET");
 880             if (sll.sll_ifindex) {
 881                 char name[IFNAMSIZ];
 882
 883                 if (if_indextoname(sll.sll_ifindex, name)) {
 884                     ds_put_format(string, "(%s)", name);
 885                 } else {
 886                     ds_put_format(string, "(ifindex=%d)", sll.sll_ifindex);
 887                 }
 888             }
 889             if (sll.sll_protocol) {
 890                 ds_put_format(string, "(protocol=0x%"PRIu16")",
 891                               ntohs(sll.sll_protocol));
 892             }
 893         }
 894 #endif
 895         else if (ss.ss_family == AF_UNSPEC) {
 896             ds_put_cstr(string, "AF_UNSPEC");
 897         } else {
 898             ds_put_format(string, "AF_%d", (int) ss.ss_family);
 899         }
 900     }
 901 }
 902
 903
 904 #ifdef __linux__
 905 static void
 906 put_fd_filename(struct ds *string, int fd)
 907 {
 908     char buf[1024];
 909     char *linkname;
 910     int n;
 911
 912     linkname = xasprintf("/proc/self/fd/%d", fd);
 913     n = readlink(linkname, buf, sizeof buf);
 914     if (n > 0) {
 915         ds_put_char(string, ' ');
 916         ds_put_buffer(string, buf, n);
 917         if (n > sizeof buf) {
 918             ds_put_cstr(string, "...");
 919         }
 920     }
 921     free(linkname);
 922 }
 923 #endif
 924
 925 /* Returns a malloc()'d string describing 'fd', for use in logging. */
 926 char *
 927 describe_fd(int fd)
 928 {
 929     struct ds string;
 930     struct stat s;
 931
 932     ds_init(&string);
 933 #ifndef _WIN32
 934     if (fstat(fd, &s)) {
 935         ds_put_format(&string, "fstat failed (%s)", ovs_strerror(errno));
 936     } else if (S_ISSOCK(s.st_mode)) {
 937         describe_sockaddr(&string, fd, getsockname);
 938         ds_put_cstr(&string, "<->");
 939         describe_sockaddr(&string, fd, getpeername);
 940     } else {
 941         ds_put_cstr(&string, (isatty(fd) ? "tty"
 942                               : S_ISDIR(s.st_mode) ? "directory"
 943                               : S_ISCHR(s.st_mode) ? "character device"
 944                               : S_ISBLK(s.st_mode) ? "block device"
 945                               : S_ISREG(s.st_mode) ? "file"
 946                               : S_ISFIFO(s.st_mode) ? "FIFO"
 947                               : S_ISLNK(s.st_mode) ? "symbolic link"
 948                               : "unknown"));
 949 #ifdef __linux__
 950         put_fd_filename(&string, fd);
 951 #endif
 952     }
 953 #else
 954     ds_put_format(&string,"file descriptor");
 955 #endif /* _WIN32 */
 956     return ds_steal_cstr(&string);
 957 }
 958
 959 \f
 960 /* sockaddr_storage helpers. */
 961
 962 /* Returns the IPv4 or IPv6 port in 'ss'. */
 963 uint16_t
 964 ss_get_port(const struct sockaddr_storage *ss)
 965 {
 966     if (ss->ss_family == AF_INET) {
 967         const struct sockaddr_in *sin
 968             = ALIGNED_CAST(const struct sockaddr_in *, ss);
 969         return ntohs(sin->sin_port);
 970     } else if (ss->ss_family == AF_INET6) {
 971         const struct sockaddr_in6 *sin6
 972             = ALIGNED_CAST(const struct sockaddr_in6 *, ss);
 973         return ntohs(sin6->sin6_port);
 974     } else {
 975         OVS_NOT_REACHED();
 976     }
 977 }
 978
 979 /* Returns true if 'name' is safe to include inside a network address field.
 980  * We want to avoid names that include confusing punctuation, etc. */
 981 static bool OVS_UNUSED
 982 is_safe_name(const char *name)
 983 {
 984     if (!name[0] || isdigit((unsigned char) name[0])) {
 985         return false;
 986     }
 987     for (const char *p = name; *p; p++) {
 988         if (!isalnum((unsigned char) *p) && *p != '-' && *p != '_') {
 989             return false;
 990         }
 991     }
 992     return true;
 993 }
 994
 995 /* Formats the IPv4 or IPv6 address in 'ss' into 's'.  If 'ss' is an IPv6
 996  * address, puts square brackets around the address.  'bufsize' should be at
 997  * least SS_NTOP_BUFSIZE. */
 998 void
 999 ss_format_address(const struct sockaddr_storage *ss, struct ds *s)
1000 {
1001     if (ss->ss_family == AF_INET) {
1002         const struct sockaddr_in *sin
1003             = ALIGNED_CAST(const struct sockaddr_in *, ss);
1004
1005         ds_put_format(s, IP_FMT, IP_ARGS(sin->sin_addr.s_addr));
1006     } else if (ss->ss_family == AF_INET6) {
1007         const struct sockaddr_in6 *sin6
1008             = ALIGNED_CAST(const struct sockaddr_in6 *, ss);
1009
1010         ds_put_char(s, '[');
1011         ds_reserve(s, s->length + INET6_ADDRSTRLEN);
1012         char *tail = &s->string[s->length];
1013         inet_ntop(AF_INET6, sin6->sin6_addr.s6_addr, tail, INET6_ADDRSTRLEN);
1014         s->length += strlen(tail);
1015
1016 #ifdef HAVE_STRUCT_SOCKADDR_IN6_SIN6_SCOPE_ID
1017         uint32_t scope = sin6->sin6_scope_id;
1018         if (scope) {
1019             char namebuf[IF_NAMESIZE];
1020             char *name = if_indextoname(scope, namebuf);
1021             ds_put_char(s, '%');
1022             if (name && is_safe_name(name)) {
1023                 ds_put_cstr(s, name);
1024             } else {
1025                 ds_put_format(s, "%"PRIu32, scope);
1026             }
1027         }
1028 #endif
1029
1030         ds_put_char(s, ']');
1031     } else {
1032         OVS_NOT_REACHED();
1033     }
1034 }
1035
1036 size_t
1037 ss_length(const struct sockaddr_storage *ss)
1038 {
1039     switch (ss->ss_family) {
1040     case AF_INET:
1041         return sizeof(struct sockaddr_in);
1042
1043     case AF_INET6:
1044         return sizeof(struct sockaddr_in6);
1045
1046     default:
1047         OVS_NOT_REACHED();
1048     }
1049 }
1050
1051 /* For Windows socket calls, 'errno' is not set.  One has to call
1052  * WSAGetLastError() to get the error number and then pass it to
1053  * this function to get the correct error string.
1054  *
1055  * ovs_strerror() calls strerror_r() and would not get the correct error
1056  * string for Windows sockets, but is good for POSIX. */
1057 const char *
1058 sock_strerror(int error)
1059 {
1060 #ifdef _WIN32
1061     return ovs_format_message(error);
1062 #else
1063     return ovs_strerror(error);
1064 #endif
1065 }
1066 \f
1067 #ifndef _WIN32 //Avoid using sendmsg on Windows entirely
1068 static int
1069 emulate_sendmmsg(int fd, struct mmsghdr *msgs, unsigned int n,
1070                  unsigned int flags)
1071 {
1072     for (unsigned int i = 0; i < n; i++) {
1073         ssize_t retval = sendmsg(fd, &msgs[i].msg_hdr, flags);
1074         if (retval < 0) {
1075             return i ? i : retval;
1076         }
1077         msgs[i].msg_len = retval;
1078     }
1079     return n;
1080 }
1081
1082 #ifndef HAVE_SENDMMSG
1083 int
1084 sendmmsg(int fd, struct mmsghdr *msgs, unsigned int n, unsigned int flags)
1085 {
1086     return emulate_sendmmsg(fd, msgs, n, flags);
1087 }
1088 #else
1089 /* sendmmsg was redefined in lib/socket-util.c, should undef sendmmsg here
1090  * to avoid recursion */
1091 #undef sendmmsg
1092 int
1093 wrap_sendmmsg(int fd, struct mmsghdr *msgs, unsigned int n, unsigned int flags)
1094 {
1095     static bool sendmmsg_broken = false;
1096     if (!sendmmsg_broken) {
1097         int save_errno = errno;
1098         int retval = sendmmsg(fd, msgs, n, flags);
1099         if (retval >= 0 || errno != ENOSYS) {
1100             return retval;
1101         }
1102         sendmmsg_broken = true;
1103         errno = save_errno;
1104     }
1105     return emulate_sendmmsg(fd, msgs, n, flags);
1106 }
1107 #endif
1108 #endif