2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
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:
8 * http://www.apache.org/licenses/LICENSE-2.0
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.
18 #include "socket-util.h"
19 #include <arpa/inet.h>
24 #include <netinet/tcp.h>
30 #include <sys/socket.h>
35 #include "dynamic-string.h"
36 #include "ovs-thread.h"
38 #include "poll-loop.h"
42 #include <linux/if_packet.h>
45 #include "netlink-protocol.h"
46 #include "netlink-socket.h"
49 VLOG_DEFINE_THIS_MODULE(socket_util);
51 static int getsockopt_int(int fd, int level, int option, const char *optname,
54 /* Sets 'fd' to non-blocking mode. Returns 0 if successful, otherwise a
55 * positive errno value. */
57 set_nonblocking(int fd)
60 int flags = fcntl(fd, F_GETFL, 0);
62 if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) != -1) {
65 VLOG_ERR("fcntl(F_SETFL) failed: %s", ovs_strerror(errno));
69 VLOG_ERR("fcntl(F_GETFL) failed: %s", ovs_strerror(errno));
73 unsigned long arg = 1;
74 if (ioctlsocket(fd, FIONBIO, &arg)) {
75 int error = sock_errno();
76 VLOG_ERR("set_nonblocking failed: %s", sock_strerror(error));
84 xset_nonblocking(int fd)
86 if (set_nonblocking(fd)) {
92 setsockopt_tcp_nodelay(int fd)
97 retval = setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, &on, sizeof on);
99 retval = sock_errno();
100 VLOG_ERR("setsockopt(TCP_NODELAY): %s", sock_strerror(retval));
105 set_dscp(int fd, uint8_t dscp)
111 /* XXX: Consider using QoS2 APIs for Windows to set dscp. */
119 /* Note: this function is used for both of IPv4 and IPv6 sockets */
122 if (setsockopt(fd, IPPROTO_IP, IP_TOS, &val, sizeof val)) {
123 if (sock_errno() != ENOPROTOOPT) {
129 if (setsockopt(fd, IPPROTO_IPV6, IPV6_TCLASS, &val, sizeof val)) {
130 if (sock_errno() != ENOPROTOOPT) {
143 /* Translates 'host_name', which must be a string representation of an IP
144 * address, into a numeric IP address in '*addr'. Returns 0 if successful,
145 * otherwise a positive errno value. */
147 lookup_ip(const char *host_name, struct in_addr *addr)
149 if (!inet_pton(AF_INET, host_name, addr)) {
150 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
151 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name);
157 /* Translates 'host_name', which must be a string representation of an IPv6
158 * address, into a numeric IPv6 address in '*addr'. Returns 0 if successful,
159 * otherwise a positive errno value. */
161 lookup_ipv6(const char *host_name, struct in6_addr *addr)
163 if (inet_pton(AF_INET6, host_name, addr) != 1) {
164 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
165 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IPv6 address", host_name);
171 /* Translates 'host_name', which must be a host name or a string representation
172 * of an IP address, into a numeric IP address in '*addr'. Returns 0 if
173 * successful, otherwise a positive errno value.
175 * Most Open vSwitch code should not use this because it causes deadlocks:
176 * getaddrinfo() sends out a DNS request but that starts a new flow for which
177 * OVS must set up a flow, but it can't because it's waiting for a DNS reply.
178 * The synchronous lookup also delays other activity. (Of course we can solve
179 * this but it doesn't seem worthwhile quite yet.) */
181 lookup_hostname(const char *host_name, struct in_addr *addr)
183 struct addrinfo *result;
184 struct addrinfo hints;
186 if (inet_pton(AF_INET, host_name, addr)) {
190 memset(&hints, 0, sizeof hints);
191 hints.ai_family = AF_INET;
193 switch (getaddrinfo(host_name, NULL, &hints, &result)) {
195 *addr = ALIGNED_CAST(struct sockaddr_in *,
196 result->ai_addr)->sin_addr;
197 freeaddrinfo(result);
200 #ifdef EAI_ADDRFAMILY
221 #if defined (EAI_NODATA) && EAI_NODATA != EAI_NONAME
237 check_connection_completion(int fd)
239 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
244 pfd.events = POLLOUT;
248 retval = poll(&pfd, 1, 0);
249 } while (retval < 0 && errno == EINTR);
251 retval = WSAPoll(&pfd, 1, 0);
254 if (pfd.revents & POLLERR) {
255 ssize_t n = send(fd, "", 1, 0);
259 VLOG_ERR_RL(&rl, "poll return POLLERR but send succeeded");
264 } else if (retval < 0) {
265 VLOG_ERR_RL(&rl, "poll: %s", sock_strerror(sock_errno()));
272 /* Returns the size of socket 'sock''s receive buffer (SO_RCVBUF), or a
273 * negative errno value if an error occurs. */
275 get_socket_rcvbuf(int sock)
280 error = getsockopt_int(sock, SOL_SOCKET, SO_RCVBUF, "SO_RCVBUF", &rcvbuf);
281 return error ? -error : rcvbuf;
284 /* Reads and discards up to 'n' datagrams from 'fd', stopping as soon as no
285 * more data can be immediately read. ('fd' should therefore be in
286 * non-blocking mode.)*/
288 drain_fd(int fd, size_t n_packets)
290 for (; n_packets > 0; n_packets--) {
291 /* 'buffer' only needs to be 1 byte long in most circumstances. This
292 * size is defensive against the possibility that we someday want to
293 * use a Linux tap device without TUN_NO_PI, in which case a buffer
294 * smaller than sizeof(struct tun_pi) will give EINVAL on read. */
296 if (read(fd, buffer, sizeof buffer) <= 0) {
303 guess_netmask(ovs_be32 ip_)
305 uint32_t ip = ntohl(ip_);
306 return ((ip >> 31) == 0 ? htonl(0xff000000) /* Class A */
307 : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */
308 : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */
309 : htonl(0)); /* ??? */
312 /* This is like strsep() except:
314 * - The separator string is ":".
316 * - Square brackets [] quote ":" separators and are removed from the
319 parse_bracketed_token(char **pp)
325 } else if (*p == '\0') {
328 } else if (*p == '[') {
330 char *end = start + strcspn(start, "]");
331 *pp = (*end == '\0' ? NULL
332 : end[1] == ':' ? end + 2
338 char *end = start + strcspn(start, ":");
339 *pp = *end == '\0' ? NULL : end + 1;
346 parse_sockaddr_components(struct sockaddr_storage *ss,
348 const char *port_s, uint16_t default_port,
351 struct sockaddr_in *sin = ALIGNED_CAST(struct sockaddr_in *, ss);
354 if (port_s && port_s[0]) {
355 if (!str_to_int(port_s, 10, &port) || port < 0 || port > 65535) {
356 VLOG_ERR("%s: bad port number \"%s\"", s, port_s);
362 memset(ss, 0, sizeof *ss);
363 if (strchr(host_s, ':')) {
364 struct sockaddr_in6 *sin6
365 = ALIGNED_CAST(struct sockaddr_in6 *, ss);
367 sin6->sin6_family = AF_INET6;
368 sin6->sin6_port = htons(port);
369 if (!inet_pton(AF_INET6, host_s, sin6->sin6_addr.s6_addr)) {
370 VLOG_ERR("%s: bad IPv6 address \"%s\"", s, host_s);
374 sin->sin_family = AF_INET;
375 sin->sin_port = htons(port);
376 if (!inet_pton(AF_INET, host_s, &sin->sin_addr.s_addr)) {
377 VLOG_ERR("%s: bad IPv4 address \"%s\"", s, host_s);
385 memset(ss, 0, sizeof *ss);
389 /* Parses 'target', which should be a string in the format "<host>[:<port>]".
390 * <host>, which is required, may be an IPv4 address or an IPv6 address
391 * enclosed in square brackets. If 'default_port' is nonzero then <port> is
392 * optional and defaults to 'default_port'.
394 * On success, returns true and stores the parsed remote address into '*ss'.
395 * On failure, logs an error, stores zeros into '*ss', and returns false. */
397 inet_parse_active(const char *target_, uint16_t default_port,
398 struct sockaddr_storage *ss)
400 char *target = xstrdup(target_);
407 host = parse_bracketed_token(&p);
408 port = parse_bracketed_token(&p);
410 VLOG_ERR("%s: host must be specified", target_);
412 } else if (!port && !default_port) {
413 VLOG_ERR("%s: port must be specified", target_);
416 ok = parse_sockaddr_components(ss, host, port, default_port, target_);
419 memset(ss, 0, sizeof *ss);
426 /* Opens a non-blocking IPv4 or IPv6 socket of the specified 'style' and
427 * connects to 'target', which should be a string in the format
428 * "<host>[:<port>]". <host>, which is required, may be an IPv4 address or an
429 * IPv6 address enclosed in square brackets. If 'default_port' is nonzero then
430 * <port> is optional and defaults to 'default_port'.
432 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
434 * On success, returns 0 (indicating connection complete) or EAGAIN (indicating
435 * connection in progress), in which case the new file descriptor is stored
436 * into '*fdp'. On failure, returns a positive errno value other than EAGAIN
437 * and stores -1 into '*fdp'.
439 * If 'ss' is non-null, then on success stores the target address into '*ss'.
441 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
442 * should be in the range [0, 63] and will automatically be shifted to the
443 * appropriately place in the IP tos field. */
445 inet_open_active(int style, const char *target, uint16_t default_port,
446 struct sockaddr_storage *ssp, int *fdp, uint8_t dscp)
448 struct sockaddr_storage ss;
453 if (!inet_parse_active(target, default_port, &ss)) {
454 error = EAFNOSUPPORT;
458 /* Create non-blocking socket. */
459 fd = socket(ss.ss_family, style, 0);
461 error = sock_errno();
462 VLOG_ERR("%s: socket: %s", target, sock_strerror(error));
465 error = set_nonblocking(fd);
470 /* The dscp bits must be configured before connect() to ensure that the
471 * TOS field is set during the connection establishment. If set after
472 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
473 error = set_dscp(fd, dscp);
475 VLOG_ERR("%s: set_dscp: %s", target, sock_strerror(error));
480 error = connect(fd, (struct sockaddr *) &ss, ss_length(&ss)) == 0
483 if (error == EINPROGRESS
485 || error == WSAEALREADY || error == WSAEWOULDBLOCK
492 if (error && error != EAGAIN) {
494 memset(ssp, 0, sizeof *ssp);
509 /* Parses 'target', which should be a string in the format "[<port>][:<host>]":
511 * - If 'default_port' is -1, then <port> is required. Otherwise, if
512 * <port> is omitted, then 'default_port' is used instead.
514 * - If <port> (or 'default_port', if used) is 0, then no port is bound
515 * and the TCP/IP stack will select a port.
517 * - <host> is optional. If supplied, it may be an IPv4 address or an
518 * IPv6 address enclosed in square brackets. If omitted, the IP address
521 * If successful, stores the address into '*ss' and returns true; otherwise
522 * zeros '*ss' and returns false. */
524 inet_parse_passive(const char *target_, int default_port,
525 struct sockaddr_storage *ss)
527 char *target = xstrdup(target_);
534 port = parse_bracketed_token(&p);
535 host = parse_bracketed_token(&p);
536 if (!port && default_port < 0) {
537 VLOG_ERR("%s: port must be specified", target_);
540 ok = parse_sockaddr_components(ss, host ? host : "0.0.0.0",
541 port, default_port, target_);
544 memset(ss, 0, sizeof *ss);
551 /* Opens a non-blocking IPv4 or IPv6 socket of the specified 'style', binds to
552 * 'target', and listens for incoming connections. Parses 'target' in the same
553 * way was inet_parse_passive().
555 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
557 * For TCP, the socket will have SO_REUSEADDR turned on.
559 * On success, returns a non-negative file descriptor. On failure, returns a
560 * negative errno value.
562 * If 'ss' is non-null, then on success stores the bound address into '*ss'.
564 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
565 * should be in the range [0, 63] and will automatically be shifted to the
566 * appropriately place in the IP tos field.
568 * If 'kernel_print_port' is true and the port is dynamically assigned by
569 * the kernel, print the chosen port. */
571 inet_open_passive(int style, const char *target, int default_port,
572 struct sockaddr_storage *ssp, uint8_t dscp,
573 bool kernel_print_port)
575 bool kernel_chooses_port;
576 struct sockaddr_storage ss;
578 unsigned int yes = 1;
580 if (!inet_parse_passive(target, default_port, &ss)) {
581 return -EAFNOSUPPORT;
583 kernel_chooses_port = ss_get_port(&ss) == 0;
585 /* Create non-blocking socket, set SO_REUSEADDR. */
586 fd = socket(ss.ss_family, style, 0);
588 error = sock_errno();
589 VLOG_ERR("%s: socket: %s", target, sock_strerror(error));
592 error = set_nonblocking(fd);
596 if (style == SOCK_STREAM
597 && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) {
598 error = sock_errno();
599 VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s",
600 target, sock_strerror(error));
605 if (bind(fd, (struct sockaddr *) &ss, ss_length(&ss)) < 0) {
606 error = sock_errno();
607 VLOG_ERR("%s: bind: %s", target, sock_strerror(error));
611 /* The dscp bits must be configured before connect() to ensure that the TOS
612 * field is set during the connection establishment. If set after
613 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
614 error = set_dscp(fd, dscp);
616 VLOG_ERR("%s: set_dscp: %s", target, sock_strerror(error));
621 if (style == SOCK_STREAM && listen(fd, 10) < 0) {
622 error = sock_errno();
623 VLOG_ERR("%s: listen: %s", target, sock_strerror(error));
627 if (ssp || kernel_chooses_port) {
628 socklen_t ss_len = sizeof ss;
629 if (getsockname(fd, (struct sockaddr *) &ss, &ss_len) < 0) {
630 error = sock_errno();
631 VLOG_ERR("%s: getsockname: %s", target, sock_strerror(error));
634 if (kernel_chooses_port && kernel_print_port) {
635 VLOG_INFO("%s: listening on port %"PRIu16,
636 target, ss_get_port(&ss));
647 memset(ssp, 0, sizeof *ssp);
654 read_fully(int fd, void *p_, size_t size, size_t *bytes_read)
660 ssize_t retval = read(fd, p, size);
662 *bytes_read += retval;
665 } else if (retval == 0) {
667 } else if (errno != EINTR) {
675 write_fully(int fd, const void *p_, size_t size, size_t *bytes_written)
677 const uint8_t *p = p_;
681 ssize_t retval = write(fd, p, size);
683 *bytes_written += retval;
686 } else if (retval == 0) {
687 VLOG_WARN("write returned 0");
689 } else if (errno != EINTR) {
696 /* Given file name 'file_name', fsyncs the directory in which it is contained.
697 * Returns 0 if successful, otherwise a positive errno value. */
699 fsync_parent_dir(const char *file_name)
706 dir = dir_name(file_name);
707 fd = open(dir, O_RDONLY);
710 if (errno == EINVAL || errno == EROFS) {
711 /* This directory does not support synchronization. Not
712 * really an error. */
715 VLOG_ERR("%s: fsync failed (%s)", dir, ovs_strerror(error));
721 VLOG_ERR("%s: open failed (%s)", dir, ovs_strerror(error));
729 /* Obtains the modification time of the file named 'file_name' to the greatest
730 * supported precision. If successful, stores the mtime in '*mtime' and
731 * returns 0. On error, returns a positive errno value and stores zeros in
734 get_mtime(const char *file_name, struct timespec *mtime)
738 if (!stat(file_name, &s)) {
739 mtime->tv_sec = s.st_mtime;
741 #if HAVE_STRUCT_STAT_ST_MTIM_TV_NSEC
742 mtime->tv_nsec = s.st_mtim.tv_nsec;
743 #elif HAVE_STRUCT_STAT_ST_MTIMENSEC
744 mtime->tv_nsec = s.st_mtimensec;
751 mtime->tv_sec = mtime->tv_nsec = 0;
757 getsockopt_int(int fd, int level, int option, const char *optname, int *valuep)
759 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
765 if (getsockopt(fd, level, option, &value, &len)) {
766 error = sock_errno();
767 VLOG_ERR_RL(&rl, "getsockopt(%s): %s", optname, sock_strerror(error));
768 } else if (len != sizeof value) {
770 VLOG_ERR_RL(&rl, "getsockopt(%s): value is %u bytes (expected %"PRIuSIZE")",
771 optname, (unsigned int) len, sizeof value);
776 *valuep = error ? 0 : value;
781 describe_sockaddr(struct ds *string, int fd,
782 int (*getaddr)(int, struct sockaddr *, socklen_t *))
784 struct sockaddr_storage ss;
785 socklen_t len = sizeof ss;
787 if (!getaddr(fd, (struct sockaddr *) &ss, &len)) {
788 if (ss.ss_family == AF_INET || ss.ss_family == AF_INET6) {
789 char addrbuf[SS_NTOP_BUFSIZE];
791 ds_put_format(string, "%s:%"PRIu16,
792 ss_format_address(&ss, addrbuf, sizeof addrbuf),
795 } else if (ss.ss_family == AF_UNIX) {
796 struct sockaddr_un sun;
800 memcpy(&sun, &ss, sizeof sun);
801 maxlen = len - offsetof(struct sockaddr_un, sun_path);
802 null = memchr(sun.sun_path, '\0', maxlen);
803 ds_put_buffer(string, sun.sun_path,
804 null ? null - sun.sun_path : maxlen);
808 else if (ss.ss_family == AF_NETLINK) {
811 /* SO_PROTOCOL was introduced in 2.6.32. Support it regardless of the version
812 * of the Linux kernel headers in use at build time. */
814 #define SO_PROTOCOL 38
817 if (!getsockopt_int(fd, SOL_SOCKET, SO_PROTOCOL, "SO_PROTOCOL",
821 ds_put_cstr(string, "NETLINK_ROUTE");
824 case NETLINK_GENERIC:
825 ds_put_cstr(string, "NETLINK_GENERIC");
829 ds_put_format(string, "AF_NETLINK family %d", protocol);
833 ds_put_cstr(string, "AF_NETLINK");
838 else if (ss.ss_family == AF_PACKET) {
839 struct sockaddr_ll sll;
841 memcpy(&sll, &ss, sizeof sll);
842 ds_put_cstr(string, "AF_PACKET");
843 if (sll.sll_ifindex) {
846 if (if_indextoname(sll.sll_ifindex, name)) {
847 ds_put_format(string, "(%s)", name);
849 ds_put_format(string, "(ifindex=%d)", sll.sll_ifindex);
852 if (sll.sll_protocol) {
853 ds_put_format(string, "(protocol=0x%"PRIu16")",
854 ntohs(sll.sll_protocol));
858 else if (ss.ss_family == AF_UNSPEC) {
859 ds_put_cstr(string, "AF_UNSPEC");
861 ds_put_format(string, "AF_%d", (int) ss.ss_family);
869 put_fd_filename(struct ds *string, int fd)
875 linkname = xasprintf("/proc/self/fd/%d", fd);
876 n = readlink(linkname, buf, sizeof buf);
878 ds_put_char(string, ' ');
879 ds_put_buffer(string, buf, n);
880 if (n > sizeof buf) {
881 ds_put_cstr(string, "...");
888 /* Returns a malloc()'d string describing 'fd', for use in logging. */
898 ds_put_format(&string, "fstat failed (%s)", ovs_strerror(errno));
899 } else if (S_ISSOCK(s.st_mode)) {
900 describe_sockaddr(&string, fd, getsockname);
901 ds_put_cstr(&string, "<->");
902 describe_sockaddr(&string, fd, getpeername);
904 ds_put_cstr(&string, (isatty(fd) ? "tty"
905 : S_ISDIR(s.st_mode) ? "directory"
906 : S_ISCHR(s.st_mode) ? "character device"
907 : S_ISBLK(s.st_mode) ? "block device"
908 : S_ISREG(s.st_mode) ? "file"
909 : S_ISFIFO(s.st_mode) ? "FIFO"
910 : S_ISLNK(s.st_mode) ? "symbolic link"
913 put_fd_filename(&string, fd);
917 ds_put_format(&string,"file descriptor");
919 return ds_steal_cstr(&string);
923 /* sockaddr_storage helpers. */
925 /* Returns the IPv4 or IPv6 port in 'ss'. */
927 ss_get_port(const struct sockaddr_storage *ss)
929 if (ss->ss_family == AF_INET) {
930 const struct sockaddr_in *sin
931 = ALIGNED_CAST(const struct sockaddr_in *, ss);
932 return ntohs(sin->sin_port);
933 } else if (ss->ss_family == AF_INET6) {
934 const struct sockaddr_in6 *sin6
935 = ALIGNED_CAST(const struct sockaddr_in6 *, ss);
936 return ntohs(sin6->sin6_port);
942 /* Formats the IPv4 or IPv6 address in 'ss' into the 'bufsize' bytes in 'buf'.
943 * If 'ss' is an IPv6 address, puts square brackets around the address.
944 * 'bufsize' should be at least SS_NTOP_BUFSIZE.
948 ss_format_address(const struct sockaddr_storage *ss,
949 char *buf, size_t bufsize)
951 ovs_assert(bufsize >= SS_NTOP_BUFSIZE);
952 if (ss->ss_family == AF_INET) {
953 const struct sockaddr_in *sin
954 = ALIGNED_CAST(const struct sockaddr_in *, ss);
956 snprintf(buf, bufsize, IP_FMT, IP_ARGS(sin->sin_addr.s_addr));
957 } else if (ss->ss_family == AF_INET6) {
958 const struct sockaddr_in6 *sin6
959 = ALIGNED_CAST(const struct sockaddr_in6 *, ss);
962 inet_ntop(AF_INET6, sin6->sin6_addr.s6_addr, buf + 1, bufsize - 1);
963 strcpy(strchr(buf, '\0'), "]");
972 ss_length(const struct sockaddr_storage *ss)
974 switch (ss->ss_family) {
976 return sizeof(struct sockaddr_in);
979 return sizeof(struct sockaddr_in6);
986 /* For Windows socket calls, 'errno' is not set. One has to call
987 * WSAGetLastError() to get the error number and then pass it to
988 * this function to get the correct error string.
990 * ovs_strerror() calls strerror_r() and would not get the correct error
991 * string for Windows sockets, but is good for POSIX. */
993 sock_strerror(int error)
996 return ovs_format_message(error);
998 return ovs_strerror(error);