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 <sys/types.h>
  20 #include <netinet/in.h>
  21 #include <arpa/inet.h>
  22 #include <ctype.h>
  23 #include <errno.h>
  24 #include <fcntl.h>
  25 #include <net/if.h>
  26 #include <netdb.h>
  27 #include <netinet/tcp.h>
  28 #include <poll.h>
  29 #include <stddef.h>
  30 #include <stdio.h>
  31 #include <stdlib.h>
  32 #include <string.h>
  33 #include <sys/socket.h>
  34 #include <sys/stat.h>
  35 #include <sys/uio.h>
  36 #include <sys/un.h>
  37 #include <unistd.h>
  38 #include "openvswitch/dynamic-string.h"
  39 #include "ovs-thread.h"
  40 #include "packets.h"
  41 #include "openvswitch/poll-loop.h"
  42 #include "util.h"
  43 #include "openvswitch/vlog.h"
  44 #ifdef __linux__
  45 #include <linux/if_packet.h>
  46 #endif
  47 #ifdef HAVE_NETLINK
  48 #include "netlink-protocol.h"
  49 #include "netlink-socket.h"
  50 #endif
  51 #include "dns-resolve.h"
  52
  53 VLOG_DEFINE_THIS_MODULE(socket_util);
  54
  55 static int getsockopt_int(int fd, int level, int option, const char *optname,
  56                           int *valuep);
  57 static struct sockaddr_in *sin_cast(const struct sockaddr *);
  58 static struct sockaddr_in6 *sin6_cast(const struct sockaddr *);
  59 static const struct sockaddr *sa_cast(const struct sockaddr_storage *);
  60 static bool parse_sockaddr_components(struct sockaddr_storage *ss,
  61                                       char *host_s,
  62                                       const char *port_s,
  63                                       uint16_t default_port,
  64                                       const char *s,
  65                                       bool resolve_host);
  66
  67 /* Sets 'fd' to non-blocking mode.  Returns 0 if successful, otherwise a
  68  * positive errno value. */
  69 int
  70 set_nonblocking(int fd)
  71 {
  72 #ifndef _WIN32
  73     int flags = fcntl(fd, F_GETFL, 0);
  74     if (flags != -1) {
  75         if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) != -1) {
  76             return 0;
  77         } else {
  78             VLOG_ERR("fcntl(F_SETFL) failed: %s", ovs_strerror(errno));
  79             return errno;
  80         }
  81     } else {
  82         VLOG_ERR("fcntl(F_GETFL) failed: %s", ovs_strerror(errno));
  83         return errno;
  84     }
  85 #else
  86     unsigned long arg = 1;
  87     if (ioctlsocket(fd, FIONBIO, &arg)) {
  88         int error = sock_errno();
  89         VLOG_ERR("set_nonblocking failed: %s", sock_strerror(error));
  90         return error;
  91     }
  92     return 0;
  93 #endif
  94 }
  95
  96 void
  97 xset_nonblocking(int fd)
  98 {
  99     if (set_nonblocking(fd)) {
 100         exit(EXIT_FAILURE);
 101     }
 102 }
 103
 104 void
 105 setsockopt_tcp_nodelay(int fd)
 106 {
 107     int on = 1;
 108     int retval;
 109
 110     retval = setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, &on, sizeof on);
 111     if (retval) {
 112         retval = sock_errno();
 113         VLOG_ERR("setsockopt(TCP_NODELAY): %s", sock_strerror(retval));
 114     }
 115 }
 116
 117 /* Sets the DSCP value of socket 'fd' to 'dscp', which must be 63 or less.
 118  * 'family' must indicate the socket's address family (AF_INET or AF_INET6, to
 119  * do anything useful). */
 120 int
 121 set_dscp(int fd, int family, uint8_t dscp)
 122 {
 123     int retval;
 124     int val;
 125
 126 #ifdef _WIN32
 127     /* XXX: Consider using QoS2 APIs for Windows to set dscp. */
 128     return 0;
 129 #endif
 130
 131     if (dscp > 63) {
 132         return EINVAL;
 133     }
 134     val = dscp << 2;
 135
 136     switch (family) {
 137     case AF_INET:
 138         retval = setsockopt(fd, IPPROTO_IP, IP_TOS, &val, sizeof val);
 139         break;
 140
 141     case AF_INET6:
 142         retval = setsockopt(fd, IPPROTO_IPV6, IPV6_TCLASS, &val, sizeof val);
 143         break;
 144
 145     default:
 146         return ENOPROTOOPT;
 147     }
 148
 149     return retval ? sock_errno() : 0;
 150 }
 151
 152 /* Checks whether 'host_name' is an IPv4 or IPv6 address.  It is assumed
 153  * that 'host_name' is valid.  Returns false if it is IPv4 address, true if
 154  * it is IPv6 address. */
 155 bool
 156 addr_is_ipv6(const char *host_name)
 157 {
 158     return strchr(host_name, ':') != NULL;
 159 }
 160
 161 /* Translates 'host_name', which must be a string representation of an IP
 162  * address, into a numeric IP address in '*addr'.  Returns 0 if successful,
 163  * otherwise a positive errno value. */
 164 int
 165 lookup_ip(const char *host_name, struct in_addr *addr)
 166 {
 167     if (!ip_parse(host_name, &addr->s_addr)) {
 168         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 169         VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name);
 170         return ENOENT;
 171     }
 172     return 0;
 173 }
 174
 175 /* Translates 'host_name', which must be a string representation of an IPv6
 176  * address, into a numeric IPv6 address in '*addr'.  Returns 0 if successful,
 177  * otherwise a positive errno value. */
 178 int
 179 lookup_ipv6(const char *host_name, struct in6_addr *addr)
 180 {
 181     if (!ipv6_parse(host_name, addr)) {
 182         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 183         VLOG_ERR_RL(&rl, "\"%s\" is not a valid IPv6 address", host_name);
 184         return ENOENT;
 185     }
 186     return 0;
 187 }
 188
 189 /* Translates 'host_name', which must be a host name or a string representation
 190  * of an IP address, into a numeric IP address in '*addr'.  Returns 0 if
 191  * successful, otherwise a positive errno value.
 192  *
 193  * Most Open vSwitch code should not use this because it causes deadlocks:
 194  * getaddrinfo() sends out a DNS request but that starts a new flow for which
 195  * OVS must set up a flow, but it can't because it's waiting for a DNS reply.
 196  * The synchronous lookup also delays other activity.  (Of course we can solve
 197  * this but it doesn't seem worthwhile quite yet.)  */
 198 int
 199 lookup_hostname(const char *host_name, struct in_addr *addr)
 200 {
 201     struct addrinfo *result;
 202     struct addrinfo hints;
 203
 204     if (ip_parse(host_name, &addr->s_addr)) {
 205         return 0;
 206     }
 207
 208     memset(&hints, 0, sizeof hints);
 209     hints.ai_family = AF_INET;
 210
 211     switch (getaddrinfo(host_name, NULL, &hints, &result)) {
 212     case 0:
 213         *addr = ALIGNED_CAST(struct sockaddr_in *,
 214                              result->ai_addr)->sin_addr;
 215         freeaddrinfo(result);
 216         return 0;
 217
 218 #ifdef EAI_ADDRFAMILY
 219     case EAI_ADDRFAMILY:
 220 #endif
 221     case EAI_NONAME:
 222     case EAI_SERVICE:
 223         return ENOENT;
 224
 225     case EAI_AGAIN:
 226         return EAGAIN;
 227
 228     case EAI_BADFLAGS:
 229     case EAI_FAMILY:
 230     case EAI_SOCKTYPE:
 231         return EINVAL;
 232
 233     case EAI_FAIL:
 234         return EIO;
 235
 236     case EAI_MEMORY:
 237         return ENOMEM;
 238
 239 #if defined (EAI_NODATA) && EAI_NODATA != EAI_NONAME
 240     case EAI_NODATA:
 241         return ENXIO;
 242 #endif
 243
 244 #ifdef EAI_SYSTEM
 245     case EAI_SYSTEM:
 246         return sock_errno();
 247 #endif
 248
 249     default:
 250         return EPROTO;
 251     }
 252 }
 253
 254 int
 255 check_connection_completion(int fd)
 256 {
 257     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
 258     struct pollfd pfd;
 259     int retval;
 260
 261     pfd.fd = fd;
 262     pfd.events = POLLOUT;
 263
 264 #ifndef _WIN32
 265     do {
 266         retval = poll(&pfd, 1, 0);
 267     } while (retval < 0 && errno == EINTR);
 268 #else
 269     fd_set wrset, exset;
 270     FD_ZERO(&wrset);
 271     FD_ZERO(&exset);
 272     FD_SET(fd, &exset);
 273     FD_SET(fd, &wrset);
 274     pfd.revents = 0;
 275     struct timeval tv = { 0, 0 };
 276     /* WSAPoll is broken on Windows, instead do a select */
 277     retval = select(0, NULL, &wrset, &exset, &tv);
 278     if (retval == 1) {
 279         if (FD_ISSET(fd, &wrset)) {
 280             pfd.revents |= pfd.events;
 281         }
 282         if (FD_ISSET(fd, &exset)) {
 283             pfd.revents |= POLLERR;
 284         }
 285     }
 286 #endif
 287     if (retval == 1) {
 288         if (pfd.revents & (POLLERR | POLLHUP)) {
 289             ssize_t n = send(fd, "", 1, 0);
 290             if (n < 0) {
 291                 return sock_errno();
 292             } else {
 293                 VLOG_ERR_RL(&rl, "poll return POLLERR but send succeeded");
 294                 return EPROTO;
 295             }
 296         }
 297         return 0;
 298     } else if (retval < 0) {
 299         VLOG_ERR_RL(&rl, "poll: %s", sock_strerror(sock_errno()));
 300         return errno;
 301     } else {
 302         return EAGAIN;
 303     }
 304 }
 305
 306 /* Returns the size of socket 'sock''s receive buffer (SO_RCVBUF), or a
 307  * negative errno value if an error occurs. */
 308 int
 309 get_socket_rcvbuf(int sock)
 310 {
 311     int rcvbuf;
 312     int error;
 313
 314     error = getsockopt_int(sock, SOL_SOCKET, SO_RCVBUF, "SO_RCVBUF", &rcvbuf);
 315     return error ? -error : rcvbuf;
 316 }
 317
 318 /* Reads and discards up to 'n' datagrams from 'fd', stopping as soon as no
 319  * more data can be immediately read.  ('fd' should therefore be in
 320  * non-blocking mode.)*/
 321 void
 322 drain_fd(int fd, size_t n_packets)
 323 {
 324     for (; n_packets > 0; n_packets--) {
 325         /* 'buffer' only needs to be 1 byte long in most circumstances.  This
 326          * size is defensive against the possibility that we someday want to
 327          * use a Linux tap device without TUN_NO_PI, in which case a buffer
 328          * smaller than sizeof(struct tun_pi) will give EINVAL on read. */
 329         char buffer[128];
 330         if (read(fd, buffer, sizeof buffer) <= 0) {
 331             break;
 332         }
 333     }
 334 }
 335
 336 ovs_be32
 337 guess_netmask(ovs_be32 ip_)
 338 {
 339     uint32_t ip = ntohl(ip_);
 340     return ((ip >> 31) == 0 ? htonl(0xff000000)   /* Class A */
 341             : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */
 342             : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */
 343             : htonl(0));                          /* ??? */
 344 }
 345
 346 static char *
 347 unbracket(char *s)
 348 {
 349     if (*s == '[') {
 350         s++;
 351
 352         char *end = strchr(s, '\0');
 353         if (end[-1] == ']') {
 354             end[-1] = '\0';
 355         }
 356     }
 357     return s;
 358 }
 359
 360 /* 'host_index' is 0 if the host precedes the port within 's', 1 otherwise. */
 361 static void
 362 inet_parse_tokens__(char *s, int host_index, char **hostp, char **portp)
 363 {
 364     char *colon = NULL;
 365     bool in_brackets = false;
 366     int n_colons = 0;
 367     for (char *p = s; *p; p++) {
 368         if (*p == '[') {
 369             in_brackets = true;
 370         } else if (*p == ']') {
 371             in_brackets = false;
 372         } else if (*p == ':' && !in_brackets) {
 373             n_colons++;
 374             colon = p;
 375         }
 376     }
 377
 378     *hostp = *portp = NULL;
 379     if (n_colons > 1) {
 380         *hostp = s;
 381     } else {
 382         char **tokens[2];
 383         tokens[host_index] = hostp;
 384         tokens[!host_index] = portp;
 385
 386         if (colon) {
 387             *colon = '\0';
 388             *tokens[1] = unbracket(colon + 1);
 389         }
 390         *tokens[0] = unbracket(s);
 391     }
 392 }
 393
 394 /* Parses 's', a string in the form "<host>[:<port>]", into its (required) host
 395  * and (optional) port components, and stores pointers to them in '*hostp' and
 396  * '*portp' respectively.  Always sets '*hostp' nonnull, although possibly to
 397  * an empty string.  Can set '*portp' to the null string.
 398  *
 399  * Supports both IPv4 and IPv6.  IPv6 addresses may be quoted with square
 400  * brackets.  Resolves ambiguous cases that might represent an IPv6 address or
 401  * an IPv6 address and a port as representing just a host, e.g. "::1:2:3:4:80"
 402  * is a host but "[::1:2:3:4]:80" is a host and a port.
 403  *
 404  * Modifies 's' and points '*hostp' and '*portp' (if nonnull) into it.
 405  */
 406 void
 407 inet_parse_host_port_tokens(char *s, char **hostp, char **portp)
 408 {
 409     inet_parse_tokens__(s, 0, hostp, portp);
 410 }
 411
 412 /* Parses 's', a string in the form "<port>[:<host>]", into its port and host
 413  * components, and stores pointers to them in '*portp' and '*hostp'
 414  * respectively.  Either '*portp' and '*hostp' (but not both) can end up null.
 415  *
 416  * Supports both IPv4 and IPv6.  IPv6 addresses may be quoted with square
 417  * brackets.  Resolves ambiguous cases that might represent an IPv6 address or
 418  * an IPv6 address and a port as representing just a host, e.g. "::1:2:3:4:80"
 419  * is a host but "[::1:2:3:4]:80" is a host and a port.
 420  *
 421  * Modifies 's' and points '*hostp' and '*portp' (if nonnull) into it.
 422  */
 423 void
 424 inet_parse_port_host_tokens(char *s, char **portp, char **hostp)
 425 {
 426     inet_parse_tokens__(s, 1, hostp, portp);
 427 }
 428
 429 static bool
 430 parse_sockaddr_components_dns(struct sockaddr_storage *ss OVS_UNUSED,
 431                               char *host_s,
 432                               const char *port_s OVS_UNUSED,
 433                               uint16_t default_port OVS_UNUSED,
 434                               const char *s OVS_UNUSED)
 435 {
 436     char *tmp_host_s;
 437
 438     dns_resolve(host_s, &tmp_host_s);
 439     if (tmp_host_s != NULL) {
 440         parse_sockaddr_components(ss, tmp_host_s, port_s,
 441                                   default_port, s, false);
 442         free(tmp_host_s);
 443         return true;
 444     }
 445     return false;
 446 }
 447
 448 static bool
 449 parse_sockaddr_components(struct sockaddr_storage *ss,
 450                           char *host_s,
 451                           const char *port_s, uint16_t default_port,
 452                           const char *s,
 453                           bool resolve_host)
 454 {
 455     struct sockaddr_in *sin = sin_cast(sa_cast(ss));
 456     int port;
 457
 458     if (port_s && port_s[0]) {
 459         if (!str_to_int(port_s, 10, &port) || port < 0 || port > 65535) {
 460             VLOG_ERR("%s: bad port number \"%s\"", s, port_s);
 461             goto exit;
 462         }
 463     } else {
 464         port = default_port;
 465     }
 466
 467     memset(ss, 0, sizeof *ss);
 468     if (host_s && strchr(host_s, ':')) {
 469         struct sockaddr_in6 *sin6 = sin6_cast(sa_cast(ss));
 470         char *addr = strsep(&host_s, "%");
 471
 472         sin6->sin6_family = AF_INET6;
 473         sin6->sin6_port = htons(port);
 474         if (!addr || !*addr || !ipv6_parse(addr, &sin6->sin6_addr)) {
 475             goto exit;
 476         }
 477
 478 #ifdef HAVE_STRUCT_SOCKADDR_IN6_SIN6_SCOPE_ID
 479         char *scope = strsep(&host_s, "%");
 480         if (scope && *scope) {
 481             if (!scope[strspn(scope, "0123456789")]) {
 482                 sin6->sin6_scope_id = atoi(scope);
 483             } else {
 484                 sin6->sin6_scope_id = if_nametoindex(scope);
 485                 if (!sin6->sin6_scope_id) {
 486                     VLOG_ERR("%s: bad IPv6 scope \"%s\" (%s)",
 487                              s, scope, ovs_strerror(errno));
 488                     goto exit;
 489                 }
 490             }
 491         }
 492 #endif
 493     } else {
 494         sin->sin_family = AF_INET;
 495         sin->sin_port = htons(port);
 496         if (host_s && !ip_parse(host_s, &sin->sin_addr.s_addr)) {
 497             goto resolve;
 498         }
 499     }
 500
 501     return true;
 502
 503 resolve:
 504     if (resolve_host && parse_sockaddr_components_dns(ss, host_s, port_s,
 505                                                       default_port, s)) {
 506         return true;
 507     } else if (!resolve_host) {
 508         VLOG_ERR("%s: bad IP address \"%s\"", s, host_s);
 509     }
 510 exit:
 511     memset(ss, 0, sizeof *ss);
 512     return false;
 513 }
 514
 515 /* Parses 'target', which should be a string in the format "<host>[:<port>]".
 516  * <host>, which is required, may be an IPv4 address or an IPv6 address
 517  * enclosed in square brackets.  If 'default_port' is nonnegative then <port>
 518  * is optional and defaults to 'default_port' (use 0 to make the kernel choose
 519  * an available port, although this isn't usually appropriate for active
 520  * connections).  If 'default_port' is negative, then <port> is required.
 521  * It resolves the host if 'resolve_host' is true.
 522  *
 523  * On success, returns true and stores the parsed remote address into '*ss'.
 524  * On failure, logs an error, stores zeros into '*ss', and returns false. */
 525 bool
 526 inet_parse_active(const char *target_, int default_port,
 527                   struct sockaddr_storage *ss, bool resolve_host)
 528 {
 529     char *target = xstrdup(target_);
 530     char *port, *host;
 531     bool ok;
 532
 533     inet_parse_host_port_tokens(target, &host, &port);
 534     if (!host) {
 535         VLOG_ERR("%s: host must be specified", target_);
 536         ok = false;
 537     } else if (!port && default_port < 0) {
 538         VLOG_ERR("%s: port must be specified", target_);
 539         ok = false;
 540     } else {
 541         ok = parse_sockaddr_components(ss, host, port, default_port,
 542                                        target_, resolve_host);
 543     }
 544     if (!ok) {
 545         memset(ss, 0, sizeof *ss);
 546     }
 547     free(target);
 548     return ok;
 549 }
 550
 551
 552 /* Opens a non-blocking IPv4 or IPv6 socket of the specified 'style' and
 553  * connects to 'target', which should be a string in the format
 554  * "<host>[:<port>]".  <host>, which is required, may be an IPv4 address or an
 555  * IPv6 address enclosed in square brackets.  If 'default_port' is nonnegative
 556  * then <port> is optional and defaults to 'default_port'.
 557  *
 558  * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
 559  *
 560  * On success, returns 0 (indicating connection complete) or EAGAIN (indicating
 561  * connection in progress), in which case the new file descriptor is stored
 562  * into '*fdp'.  On failure, returns a positive errno value other than EAGAIN
 563  * and stores -1 into '*fdp'.
 564  *
 565  * If 'ss' is non-null, then on success stores the target address into '*ss'.
 566  *
 567  * 'dscp' becomes the DSCP bits in the IP headers for the new connection.  It
 568  * should be in the range [0, 63] and will automatically be shifted to the
 569  * appropriately place in the IP tos field. */
 570 int
 571 inet_open_active(int style, const char *target, int default_port,
 572                  struct sockaddr_storage *ssp, int *fdp, uint8_t dscp)
 573 {
 574     struct sockaddr_storage ss;
 575     int fd = -1;
 576     int error;
 577
 578     /* Parse. */
 579     if (!inet_parse_active(target, default_port, &ss, true)) {
 580         error = EAFNOSUPPORT;
 581         goto exit;
 582     }
 583
 584     /* Create non-blocking socket. */
 585     fd = socket(ss.ss_family, style, 0);
 586     if (fd < 0) {
 587         error = sock_errno();
 588         VLOG_ERR("%s: socket: %s", target, sock_strerror(error));
 589         goto exit;
 590     }
 591     error = set_nonblocking(fd);
 592     if (error) {
 593         goto exit;
 594     }
 595
 596     /* The dscp bits must be configured before connect() to ensure that the
 597      * TOS field is set during the connection establishment.  If set after
 598      * connect(), the handshake SYN frames will be sent with a TOS of 0. */
 599     error = set_dscp(fd, ss.ss_family, dscp);
 600     if (error) {
 601         VLOG_ERR("%s: set_dscp: %s", target, sock_strerror(error));
 602         goto exit;
 603     }
 604
 605     /* Connect. */
 606     error = connect(fd, (struct sockaddr *) &ss, ss_length(&ss)) == 0
 607                     ? 0
 608                     : sock_errno();
 609     if (error == EINPROGRESS
 610 #ifdef _WIN32
 611         || error == WSAEALREADY || error == WSAEWOULDBLOCK
 612 #endif
 613         ) {
 614         error = EAGAIN;
 615     }
 616
 617 exit:
 618     if (error && error != EAGAIN) {
 619         if (ssp) {
 620             memset(ssp, 0, sizeof *ssp);
 621         }
 622         if (fd >= 0) {
 623             closesocket(fd);
 624             fd = -1;
 625         }
 626     } else {
 627         if (ssp) {
 628             *ssp = ss;
 629         }
 630     }
 631     *fdp = fd;
 632     return error;
 633 }
 634
 635 /* Parses 'target', which should be a string in the format "[<port>][:<host>]":
 636  *
 637  *      - If 'default_port' is -1, then <port> is required.  Otherwise, if
 638  *        <port> is omitted, then 'default_port' is used instead.
 639  *
 640  *      - If <port> (or 'default_port', if used) is 0, then no port is bound
 641  *        and the TCP/IP stack will select a port.
 642  *
 643  *      - <host> is optional.  If supplied, it may be an IPv4 address or an
 644  *        IPv6 address enclosed in square brackets.  If omitted, the IP address
 645  *        is wildcarded.
 646  *
 647  * If successful, stores the address into '*ss' and returns true; otherwise
 648  * zeros '*ss' and returns false. */
 649 bool
 650 inet_parse_passive(const char *target_, int default_port,
 651                    struct sockaddr_storage *ss)
 652 {
 653     char *target = xstrdup(target_);
 654     char *port, *host;
 655     bool ok;
 656
 657     inet_parse_port_host_tokens(target, &port, &host);
 658     if (!port && default_port < 0) {
 659         VLOG_ERR("%s: port must be specified", target_);
 660         ok = false;
 661     } else {
 662         ok = parse_sockaddr_components(ss, host, port, default_port,
 663                                        target_, true);
 664     }
 665     if (!ok) {
 666         memset(ss, 0, sizeof *ss);
 667     }
 668     free(target);
 669     return ok;
 670 }
 671
 672
 673 /* Opens a non-blocking IPv4 or IPv6 socket of the specified 'style', binds to
 674  * 'target', and listens for incoming connections.  Parses 'target' in the same
 675  * way was inet_parse_passive().
 676  *
 677  * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
 678  *
 679  * For TCP, the socket will have SO_REUSEADDR turned on.
 680  *
 681  * On success, returns a non-negative file descriptor.  On failure, returns a
 682  * negative errno value.
 683  *
 684  * If 'ss' is non-null, then on success stores the bound address into '*ss'.
 685  *
 686  * 'dscp' becomes the DSCP bits in the IP headers for the new connection.  It
 687  * should be in the range [0, 63] and will automatically be shifted to the
 688  * appropriately place in the IP tos field.
 689  *
 690  * If 'kernel_print_port' is true and the port is dynamically assigned by
 691  * the kernel, print the chosen port. */
 692 int
 693 inet_open_passive(int style, const char *target, int default_port,
 694                   struct sockaddr_storage *ssp, uint8_t dscp,
 695                   bool kernel_print_port)
 696 {
 697     bool kernel_chooses_port;
 698     struct sockaddr_storage ss;
 699     int fd = 0, error;
 700     unsigned int yes = 1;
 701
 702     if (!inet_parse_passive(target, default_port, &ss)) {
 703         return -EAFNOSUPPORT;
 704     }
 705     kernel_chooses_port = ss_get_port(&ss) == 0;
 706
 707     /* Create non-blocking socket, set SO_REUSEADDR. */
 708     fd = socket(ss.ss_family, style, 0);
 709     if (fd < 0) {
 710         error = sock_errno();
 711         VLOG_ERR("%s: socket: %s", target, sock_strerror(error));
 712         return -error;
 713     }
 714     error = set_nonblocking(fd);
 715     if (error) {
 716         goto error;
 717     }
 718     if (style == SOCK_STREAM
 719         && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) {
 720         error = sock_errno();
 721         VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s",
 722                  target, sock_strerror(error));
 723         goto error;
 724     }
 725
 726     /* Bind. */
 727     if (bind(fd, (struct sockaddr *) &ss, ss_length(&ss)) < 0) {
 728         error = sock_errno();
 729         static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
 730         VLOG_ERR_RL(&rl, "%s: bind: %s", target, sock_strerror(error));
 731         goto error;
 732     }
 733
 734     /* The dscp bits must be configured before connect() to ensure that the TOS
 735      * field is set during the connection establishment.  If set after
 736      * connect(), the handshake SYN frames will be sent with a TOS of 0. */
 737     error = set_dscp(fd, ss.ss_family, dscp);
 738     if (error) {
 739         VLOG_ERR("%s: set_dscp: %s", target, sock_strerror(error));
 740         goto error;
 741     }
 742
 743     /* Listen. */
 744     if (style == SOCK_STREAM && listen(fd, 10) < 0) {
 745         error = sock_errno();
 746         VLOG_ERR("%s: listen: %s", target, sock_strerror(error));
 747         goto error;
 748     }
 749
 750     if (ssp || kernel_chooses_port) {
 751         socklen_t ss_len = sizeof ss;
 752         if (getsockname(fd, (struct sockaddr *) &ss, &ss_len) < 0) {
 753             error = sock_errno();
 754             VLOG_ERR("%s: getsockname: %s", target, sock_strerror(error));
 755             goto error;
 756         }
 757         if (kernel_chooses_port && kernel_print_port) {
 758             VLOG_INFO("%s: listening on port %"PRIu16,
 759                       target, ss_get_port(&ss));
 760         }
 761         if (ssp) {
 762             *ssp = ss;
 763         }
 764     }
 765
 766     return fd;
 767
 768 error:
 769     if (ssp) {
 770         memset(ssp, 0, sizeof *ssp);
 771     }
 772     closesocket(fd);
 773     return -error;
 774 }
 775
 776 /* Parses 'target', which may be an IPv4 address or an IPv6 address
 777  * enclosed in square brackets.
 778  *
 779  * On success, returns true and stores the parsed remote address into '*ss'.
 780  * On failure, logs an error, stores zeros into '*ss', and returns false. */
 781 bool
 782 inet_parse_address(const char *target_, struct sockaddr_storage *ss)
 783 {
 784     char *target = xstrdup(target_);
 785     char *host = unbracket(target);
 786     bool ok = parse_sockaddr_components(ss, host, NULL, 0, target_, false);
 787     if (!ok) {
 788         memset(ss, 0, sizeof *ss);
 789     }
 790     free(target);
 791     return ok;
 792 }
 793
 794 int
 795 read_fully(int fd, void *p_, size_t size, size_t *bytes_read)
 796 {
 797     uint8_t *p = p_;
 798
 799     *bytes_read = 0;
 800     while (size > 0) {
 801         ssize_t retval = read(fd, p, size);
 802         if (retval > 0) {
 803             *bytes_read += retval;
 804             size -= retval;
 805             p += retval;
 806         } else if (retval == 0) {
 807             return EOF;
 808         } else if (errno != EINTR) {
 809             return errno;
 810         }
 811     }
 812     return 0;
 813 }
 814
 815 int
 816 write_fully(int fd, const void *p_, size_t size, size_t *bytes_written)
 817 {
 818     const uint8_t *p = p_;
 819
 820     *bytes_written = 0;
 821     while (size > 0) {
 822         ssize_t retval = write(fd, p, size);
 823         if (retval > 0) {
 824             *bytes_written += retval;
 825             size -= retval;
 826             p += retval;
 827         } else if (retval == 0) {
 828             VLOG_WARN("write returned 0");
 829             return EPROTO;
 830         } else if (errno != EINTR) {
 831             return errno;
 832         }
 833     }
 834     return 0;
 835 }
 836
 837 /* Given file name 'file_name', fsyncs the directory in which it is contained.
 838  * Returns 0 if successful, otherwise a positive errno value. */
 839 int
 840 fsync_parent_dir(const char *file_name)
 841 {
 842     int error = 0;
 843 #ifndef _WIN32
 844     char *dir;
 845     int fd;
 846
 847     dir = dir_name(file_name);
 848     fd = open(dir, O_RDONLY);
 849     if (fd >= 0) {
 850         if (fsync(fd)) {
 851             if (errno == EINVAL || errno == EROFS) {
 852                 /* This directory does not support synchronization.  Not
 853                  * really an error. */
 854             } else {
 855                 error = errno;
 856                 VLOG_ERR("%s: fsync failed (%s)", dir, ovs_strerror(error));
 857             }
 858         }
 859         close(fd);
 860     } else {
 861         error = errno;
 862         VLOG_ERR("%s: open failed (%s)", dir, ovs_strerror(error));
 863     }
 864     free(dir);
 865 #endif
 866
 867     return error;
 868 }
 869
 870 /* Obtains the modification time of the file named 'file_name' to the greatest
 871  * supported precision.  If successful, stores the mtime in '*mtime' and
 872  * returns 0.  On error, returns a positive errno value and stores zeros in
 873  * '*mtime'. */
 874 int
 875 get_mtime(const char *file_name, struct timespec *mtime)
 876 {
 877     struct stat s;
 878
 879     if (!stat(file_name, &s)) {
 880         mtime->tv_sec = s.st_mtime;
 881
 882 #if HAVE_STRUCT_STAT_ST_MTIM_TV_NSEC
 883         mtime->tv_nsec = s.st_mtim.tv_nsec;
 884 #elif HAVE_STRUCT_STAT_ST_MTIMENSEC
 885         mtime->tv_nsec = s.st_mtimensec;
 886 #else
 887         mtime->tv_nsec = 0;
 888 #endif
 889
 890         return 0;
 891     } else {
 892         mtime->tv_sec = mtime->tv_nsec = 0;
 893         return errno;
 894     }
 895 }
 896
 897 static int
 898 getsockopt_int(int fd, int level, int option, const char *optname, int *valuep)
 899 {
 900     static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
 901     socklen_t len;
 902     int value;
 903     int error;
 904
 905     len = sizeof value;
 906     if (getsockopt(fd, level, option, &value, &len)) {
 907         error = sock_errno();
 908         VLOG_ERR_RL(&rl, "getsockopt(%s): %s", optname, sock_strerror(error));
 909     } else if (len != sizeof value) {
 910         error = EINVAL;
 911         VLOG_ERR_RL(&rl, "getsockopt(%s): value is %u bytes (expected %"PRIuSIZE")",
 912                     optname, (unsigned int) len, sizeof value);
 913     } else {
 914         error = 0;
 915     }
 916
 917     *valuep = error ? 0 : value;
 918     return error;
 919 }
 920
 921 static void
 922 describe_sockaddr(struct ds *string, int fd,
 923                   int (*getaddr)(int, struct sockaddr *, socklen_t *))
 924 {
 925     struct sockaddr_storage ss;
 926     socklen_t len = sizeof ss;
 927
 928     if (!getaddr(fd, (struct sockaddr *) &ss, &len)) {
 929         if (ss.ss_family == AF_INET || ss.ss_family == AF_INET6) {
 930             ss_format_address(&ss, string);
 931             ds_put_format(string, ":%"PRIu16, ss_get_port(&ss));
 932 #ifndef _WIN32
 933         } else if (ss.ss_family == AF_UNIX) {
 934             struct sockaddr_un sun;
 935             const char *null;
 936             size_t maxlen;
 937
 938             memcpy(&sun, &ss, sizeof sun);
 939             maxlen = len - offsetof(struct sockaddr_un, sun_path);
 940             null = memchr(sun.sun_path, '\0', maxlen);
 941             ds_put_buffer(string, sun.sun_path,
 942                           null ? null - sun.sun_path : maxlen);
 943 #endif
 944         }
 945 #ifdef HAVE_NETLINK
 946         else if (ss.ss_family == AF_NETLINK) {
 947             int protocol;
 948
 949 /* SO_PROTOCOL was introduced in 2.6.32.  Support it regardless of the version
 950  * of the Linux kernel headers in use at build time. */
 951 #ifndef SO_PROTOCOL
 952 #define SO_PROTOCOL 38
 953 #endif
 954
 955             if (!getsockopt_int(fd, SOL_SOCKET, SO_PROTOCOL, "SO_PROTOCOL",
 956                                 &protocol)) {
 957                 switch (protocol) {
 958                 case NETLINK_ROUTE:
 959                     ds_put_cstr(string, "NETLINK_ROUTE");
 960                     break;
 961
 962                 case NETLINK_GENERIC:
 963                     ds_put_cstr(string, "NETLINK_GENERIC");
 964                     break;
 965
 966                 default:
 967                     ds_put_format(string, "AF_NETLINK family %d", protocol);
 968                     break;
 969                 }
 970             } else {
 971                 ds_put_cstr(string, "AF_NETLINK");
 972             }
 973         }
 974 #endif
 975 #if __linux__
 976         else if (ss.ss_family == AF_PACKET) {
 977             struct sockaddr_ll sll;
 978
 979             memcpy(&sll, &ss, sizeof sll);
 980             ds_put_cstr(string, "AF_PACKET");
 981             if (sll.sll_ifindex) {
 982                 char name[IFNAMSIZ];
 983
 984                 if (if_indextoname(sll.sll_ifindex, name)) {
 985                     ds_put_format(string, "(%s)", name);
 986                 } else {
 987                     ds_put_format(string, "(ifindex=%d)", sll.sll_ifindex);
 988                 }
 989             }
 990             if (sll.sll_protocol) {
 991                 ds_put_format(string, "(protocol=0x%"PRIu16")",
 992                               ntohs(sll.sll_protocol));
 993             }
 994         }
 995 #endif
 996         else if (ss.ss_family == AF_UNSPEC) {
 997             ds_put_cstr(string, "AF_UNSPEC");
 998         } else {
 999             ds_put_format(string, "AF_%d", (int) ss.ss_family);
1000         }
1001     }
1002 }
1003
1004
1005 #ifdef __linux__
1006 static void
1007 put_fd_filename(struct ds *string, int fd)
1008 {
1009     char buf[1024];
1010     char *linkname;
1011     int n;
1012
1013     linkname = xasprintf("/proc/self/fd/%d", fd);
1014     n = readlink(linkname, buf, sizeof buf);
1015     if (n > 0) {
1016         ds_put_char(string, ' ');
1017         ds_put_buffer(string, buf, n);
1018         if (n > sizeof buf) {
1019             ds_put_cstr(string, "...");
1020         }
1021     }
1022     free(linkname);
1023 }
1024 #endif
1025
1026 /* Returns a malloc()'d string describing 'fd', for use in logging. */
1027 char *
1028 describe_fd(int fd)
1029 {
1030     struct ds string;
1031     struct stat s;
1032
1033     ds_init(&string);
1034 #ifndef _WIN32
1035     if (fstat(fd, &s)) {
1036         ds_put_format(&string, "fstat failed (%s)", ovs_strerror(errno));
1037     } else if (S_ISSOCK(s.st_mode)) {
1038         describe_sockaddr(&string, fd, getsockname);
1039         ds_put_cstr(&string, "<->");
1040         describe_sockaddr(&string, fd, getpeername);
1041     } else {
1042         ds_put_cstr(&string, (isatty(fd) ? "tty"
1043                               : S_ISDIR(s.st_mode) ? "directory"
1044                               : S_ISCHR(s.st_mode) ? "character device"
1045                               : S_ISBLK(s.st_mode) ? "block device"
1046                               : S_ISREG(s.st_mode) ? "file"
1047                               : S_ISFIFO(s.st_mode) ? "FIFO"
1048                               : S_ISLNK(s.st_mode) ? "symbolic link"
1049                               : "unknown"));
1050 #ifdef __linux__
1051         put_fd_filename(&string, fd);
1052 #endif
1053     }
1054 #else
1055     ds_put_format(&string,"file descriptor");
1056 #endif /* _WIN32 */
1057     return ds_steal_cstr(&string);
1058 }
1059 \f
1060 /* sockaddr helpers. */
1061
1062 static struct sockaddr_in *
1063 sin_cast(const struct sockaddr *sa)
1064 {
1065     return ALIGNED_CAST(struct sockaddr_in *, sa);
1066 }
1067
1068 static struct sockaddr_in6 *
1069 sin6_cast(const struct sockaddr *sa)
1070 {
1071     return ALIGNED_CAST(struct sockaddr_in6 *, sa);
1072 }
1073
1074 /* Returns true if 'sa' represents an IPv4 or IPv6 address, false otherwise. */
1075 bool
1076 sa_is_ip(const struct sockaddr *sa)
1077 {
1078     return sa->sa_family == AF_INET || sa->sa_family == AF_INET6;
1079 }
1080
1081 /* Returns the IPv4 or IPv6 address in 'sa'.  Returns IPv4 addresses as
1082  * v6-mapped. */
1083 struct in6_addr
1084 sa_get_address(const struct sockaddr *sa)
1085 {
1086     ovs_assert(sa_is_ip(sa));
1087     return (sa->sa_family == AF_INET
1088             ? in6_addr_mapped_ipv4(sin_cast(sa)->sin_addr.s_addr)
1089             : sin6_cast(sa)->sin6_addr);
1090 }
1091
1092 /* Returns the IPv4 or IPv6 port in 'sa'. */
1093 uint16_t
1094 sa_get_port(const struct sockaddr *sa)
1095 {
1096     ovs_assert(sa_is_ip(sa));
1097     return ntohs(sa->sa_family == AF_INET
1098                  ? sin_cast(sa)->sin_port
1099                  : sin6_cast(sa)->sin6_port);
1100 }
1101
1102 /* Returns true if 'name' is safe to include inside a network address field.
1103  * We want to avoid names that include confusing punctuation, etc. */
1104 static bool OVS_UNUSED
1105 is_safe_name(const char *name)
1106 {
1107     if (!name[0] || isdigit((unsigned char) name[0])) {
1108         return false;
1109     }
1110     for (const char *p = name; *p; p++) {
1111         if (!isalnum((unsigned char) *p) && *p != '-' && *p != '_') {
1112             return false;
1113         }
1114     }
1115     return true;
1116 }
1117
1118 static void
1119 sa_format_address__(const struct sockaddr *sa,
1120                     const char *lbrack, const char *rbrack,
1121                     struct ds *s)
1122 {
1123     ovs_assert(sa_is_ip(sa));
1124     if (sa->sa_family == AF_INET) {
1125         ds_put_format(s, IP_FMT, IP_ARGS(sin_cast(sa)->sin_addr.s_addr));
1126     } else {
1127         const struct sockaddr_in6 *sin6 = sin6_cast(sa);
1128
1129         ds_put_cstr(s, lbrack);
1130         ds_reserve(s, s->length + INET6_ADDRSTRLEN);
1131         char *tail = &s->string[s->length];
1132         inet_ntop(AF_INET6, sin6->sin6_addr.s6_addr, tail, INET6_ADDRSTRLEN);
1133         s->length += strlen(tail);
1134
1135 #ifdef HAVE_STRUCT_SOCKADDR_IN6_SIN6_SCOPE_ID
1136         uint32_t scope = sin6->sin6_scope_id;
1137         if (scope) {
1138             char namebuf[IF_NAMESIZE];
1139             char *name = if_indextoname(scope, namebuf);
1140             ds_put_char(s, '%');
1141             if (name && is_safe_name(name)) {
1142                 ds_put_cstr(s, name);
1143             } else {
1144                 ds_put_format(s, "%"PRIu32, scope);
1145             }
1146         }
1147 #endif
1148
1149         ds_put_cstr(s, rbrack);
1150     }
1151 }
1152
1153 /* Formats the IPv4 or IPv6 address in 'sa' into 's'.  If 'sa' is an IPv6
1154  * address, puts square brackets around the address. */
1155 void
1156 sa_format_address(const struct sockaddr *sa, struct ds *s)
1157 {
1158     sa_format_address__(sa, "[", "]", s);
1159 }
1160
1161 /* Formats the IPv4 or IPv6 address in 'sa' into 's'.  Does not add square
1162  * brackets around IPv6 addresses. */
1163 void
1164 sa_format_address_nobracks(const struct sockaddr *sa, struct ds *s)
1165 {
1166     sa_format_address__(sa, "", "", s);
1167 }
1168
1169 size_t
1170 sa_length(const struct sockaddr *sa)
1171 {
1172     switch (sa->sa_family) {
1173     case AF_INET:
1174         return sizeof(struct sockaddr_in);
1175
1176     case AF_INET6:
1177         return sizeof(struct sockaddr_in6);
1178
1179     default:
1180         OVS_NOT_REACHED();
1181     }
1182 }
1183 \f
1184 /* sockaddr_storage helpers.  */
1185
1186 static const struct sockaddr *
1187 sa_cast(const struct sockaddr_storage *ss)
1188 {
1189     return ALIGNED_CAST(const struct sockaddr *, ss);
1190 }
1191
1192 bool
1193 ss_is_ip(const struct sockaddr_storage *ss)
1194 {
1195     return sa_is_ip(sa_cast(ss));
1196 }
1197
1198 uint16_t
1199 ss_get_port(const struct sockaddr_storage *ss)
1200 {
1201     return sa_get_port(sa_cast(ss));
1202 }
1203
1204 struct in6_addr
1205 ss_get_address(const struct sockaddr_storage *ss)
1206 {
1207     return sa_get_address(sa_cast(ss));
1208 }
1209
1210 void
1211 ss_format_address(const struct sockaddr_storage *ss, struct ds *s)
1212 {
1213     sa_format_address(sa_cast(ss), s);
1214 }
1215
1216 void
1217 ss_format_address_nobracks(const struct sockaddr_storage *ss, struct ds *s)
1218 {
1219     sa_format_address_nobracks(sa_cast(ss), s);
1220 }
1221
1222 size_t
1223 ss_length(const struct sockaddr_storage *ss)
1224 {
1225     return sa_length(sa_cast(ss));
1226 }
1227 \f
1228 /* For Windows socket calls, 'errno' is not set.  One has to call
1229  * WSAGetLastError() to get the error number and then pass it to
1230  * this function to get the correct error string.
1231  *
1232  * ovs_strerror() calls strerror_r() and would not get the correct error
1233  * string for Windows sockets, but is good for POSIX. */
1234 const char *
1235 sock_strerror(int error)
1236 {
1237 #ifdef _WIN32
1238     return ovs_format_message(error);
1239 #else
1240     return ovs_strerror(error);
1241 #endif
1242 }
1243 \f
1244 #ifdef __linux__
1245 static int
1246 emulate_sendmmsg(int fd, struct mmsghdr *msgs, unsigned int n,
1247                  unsigned int flags)
1248 {
1249     for (unsigned int i = 0; i < n; i++) {
1250         ssize_t retval = sendmsg(fd, &msgs[i].msg_hdr, flags);
1251         if (retval < 0) {
1252             return i ? i : retval;
1253         }
1254         msgs[i].msg_len = retval;
1255     }
1256     return n;
1257 }
1258
1259 #ifndef HAVE_SENDMMSG
1260 int
1261 sendmmsg(int fd, struct mmsghdr *msgs, unsigned int n, unsigned int flags)
1262 {
1263     return emulate_sendmmsg(fd, msgs, n, flags);
1264 }
1265 #else
1266 /* sendmmsg was redefined in lib/socket-util.c, should undef sendmmsg here
1267  * to avoid recursion */
1268 #undef sendmmsg
1269 int
1270 wrap_sendmmsg(int fd, struct mmsghdr *msgs, unsigned int n, unsigned int flags)
1271 {
1272     static bool sendmmsg_broken = false;
1273     if (!sendmmsg_broken) {
1274         int save_errno = errno;
1275         int retval = sendmmsg(fd, msgs, n, flags);
1276         if (retval >= 0 || errno != ENOSYS) {
1277             return retval;
1278         }
1279         sendmmsg_broken = true;
1280         errno = save_errno;
1281     }
1282     return emulate_sendmmsg(fd, msgs, n, flags);
1283 }
1284 #endif
1285
1286 static int
1287 emulate_recvmmsg(int fd, struct mmsghdr *msgs, unsigned int n,
1288                  int flags, struct timespec *timeout OVS_UNUSED)
1289 {
1290     ovs_assert(!timeout);       /* XXX not emulated */
1291
1292     bool waitforone = flags & MSG_WAITFORONE;
1293     flags &= ~MSG_WAITFORONE;
1294
1295     for (unsigned int i = 0; i < n; i++) {
1296         ssize_t retval = recvmsg(fd, &msgs[i].msg_hdr, flags);
1297         if (retval < 0) {
1298             return i ? i : retval;
1299         }
1300         msgs[i].msg_len = retval;
1301
1302         if (waitforone) {
1303             flags |= MSG_DONTWAIT;
1304         }
1305     }
1306     return n;
1307 }
1308
1309 #ifndef HAVE_SENDMMSG
1310 int
1311 recvmmsg(int fd, struct mmsghdr *msgs, unsigned int n,
1312          int flags, struct timespec *timeout)
1313 {
1314     return emulate_recvmmsg(fd, msgs, n, flags, timeout);
1315 }
1316 #else
1317 /* recvmmsg was redefined in lib/socket-util.c, should undef recvmmsg here
1318  * to avoid recursion */
1319 #undef recvmmsg
1320 int
1321 wrap_recvmmsg(int fd, struct mmsghdr *msgs, unsigned int n,
1322               int flags, struct timespec *timeout)
1323 {
1324     ovs_assert(!timeout);       /* XXX not emulated */
1325
1326     static bool recvmmsg_broken = false;
1327     if (!recvmmsg_broken) {
1328         int save_errno = errno;
1329         int retval = recvmmsg(fd, msgs, n, flags, timeout);
1330         if (retval >= 0 || errno != ENOSYS) {
1331             return retval;
1332         }
1333         recvmmsg_broken = true;
1334         errno = save_errno;
1335     }
1336     return emulate_recvmmsg(fd, msgs, n, flags, timeout);
1337 }
1338 #endif
1339 #endif /* __linux__ */