2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014, 2015 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"
40 #include "openvswitch/vlog.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));
104 /* Sets the DSCP value of socket 'fd' to 'dscp', which must be 63 or less.
105 * 'family' must indicate the socket's address family (AF_INET or AF_INET6, to
106 * do anything useful). */
108 set_dscp(int fd, int family, uint8_t dscp)
114 /* XXX: Consider using QoS2 APIs for Windows to set dscp. */
125 retval = setsockopt(fd, IPPROTO_IP, IP_TOS, &val, sizeof val);
129 retval = setsockopt(fd, IPPROTO_IPV6, IPV6_TCLASS, &val, sizeof val);
136 return retval ? sock_errno() : 0;
139 /* Translates 'host_name', which must be a string representation of an IP
140 * address, into a numeric IP address in '*addr'. Returns 0 if successful,
141 * otherwise a positive errno value. */
143 lookup_ip(const char *host_name, struct in_addr *addr)
145 if (!inet_pton(AF_INET, host_name, addr)) {
146 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
147 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name);
153 /* Translates 'host_name', which must be a string representation of an IPv6
154 * address, into a numeric IPv6 address in '*addr'. Returns 0 if successful,
155 * otherwise a positive errno value. */
157 lookup_ipv6(const char *host_name, struct in6_addr *addr)
159 if (inet_pton(AF_INET6, host_name, addr) != 1) {
160 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
161 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IPv6 address", host_name);
167 /* Translates 'host_name', which must be a host name or a string representation
168 * of an IP address, into a numeric IP address in '*addr'. Returns 0 if
169 * successful, otherwise a positive errno value.
171 * Most Open vSwitch code should not use this because it causes deadlocks:
172 * getaddrinfo() sends out a DNS request but that starts a new flow for which
173 * OVS must set up a flow, but it can't because it's waiting for a DNS reply.
174 * The synchronous lookup also delays other activity. (Of course we can solve
175 * this but it doesn't seem worthwhile quite yet.) */
177 lookup_hostname(const char *host_name, struct in_addr *addr)
179 struct addrinfo *result;
180 struct addrinfo hints;
182 if (inet_pton(AF_INET, host_name, addr)) {
186 memset(&hints, 0, sizeof hints);
187 hints.ai_family = AF_INET;
189 switch (getaddrinfo(host_name, NULL, &hints, &result)) {
191 *addr = ALIGNED_CAST(struct sockaddr_in *,
192 result->ai_addr)->sin_addr;
193 freeaddrinfo(result);
196 #ifdef EAI_ADDRFAMILY
217 #if defined (EAI_NODATA) && EAI_NODATA != EAI_NONAME
233 check_connection_completion(int fd)
235 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
240 pfd.events = POLLOUT;
244 retval = poll(&pfd, 1, 0);
245 } while (retval < 0 && errno == EINTR);
247 retval = WSAPoll(&pfd, 1, 0);
250 if (pfd.revents & POLLERR) {
251 ssize_t n = send(fd, "", 1, 0);
255 VLOG_ERR_RL(&rl, "poll return POLLERR but send succeeded");
260 } else if (retval < 0) {
261 VLOG_ERR_RL(&rl, "poll: %s", sock_strerror(sock_errno()));
268 /* Returns the size of socket 'sock''s receive buffer (SO_RCVBUF), or a
269 * negative errno value if an error occurs. */
271 get_socket_rcvbuf(int sock)
276 error = getsockopt_int(sock, SOL_SOCKET, SO_RCVBUF, "SO_RCVBUF", &rcvbuf);
277 return error ? -error : rcvbuf;
280 /* Reads and discards up to 'n' datagrams from 'fd', stopping as soon as no
281 * more data can be immediately read. ('fd' should therefore be in
282 * non-blocking mode.)*/
284 drain_fd(int fd, size_t n_packets)
286 for (; n_packets > 0; n_packets--) {
287 /* 'buffer' only needs to be 1 byte long in most circumstances. This
288 * size is defensive against the possibility that we someday want to
289 * use a Linux tap device without TUN_NO_PI, in which case a buffer
290 * smaller than sizeof(struct tun_pi) will give EINVAL on read. */
292 if (read(fd, buffer, sizeof buffer) <= 0) {
299 guess_netmask(ovs_be32 ip_)
301 uint32_t ip = ntohl(ip_);
302 return ((ip >> 31) == 0 ? htonl(0xff000000) /* Class A */
303 : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */
304 : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */
305 : htonl(0)); /* ??? */
308 /* This is like strsep() except:
310 * - The separator string is ":".
312 * - Square brackets [] quote ":" separators and are removed from the
315 parse_bracketed_token(char **pp)
321 } else if (*p == '\0') {
324 } else if (*p == '[') {
326 char *end = start + strcspn(start, "]");
327 *pp = (*end == '\0' ? NULL
328 : end[1] == ':' ? end + 2
334 char *end = start + strcspn(start, ":");
335 *pp = *end == '\0' ? NULL : end + 1;
342 parse_sockaddr_components(struct sockaddr_storage *ss,
344 const char *port_s, uint16_t default_port,
347 struct sockaddr_in *sin = ALIGNED_CAST(struct sockaddr_in *, ss);
350 if (port_s && port_s[0]) {
351 if (!str_to_int(port_s, 10, &port) || port < 0 || port > 65535) {
352 VLOG_ERR("%s: bad port number \"%s\"", s, port_s);
358 memset(ss, 0, sizeof *ss);
359 if (strchr(host_s, ':')) {
360 struct sockaddr_in6 *sin6
361 = ALIGNED_CAST(struct sockaddr_in6 *, ss);
363 sin6->sin6_family = AF_INET6;
364 sin6->sin6_port = htons(port);
365 if (!inet_pton(AF_INET6, host_s, sin6->sin6_addr.s6_addr)) {
366 VLOG_ERR("%s: bad IPv6 address \"%s\"", s, host_s);
370 sin->sin_family = AF_INET;
371 sin->sin_port = htons(port);
372 if (!inet_pton(AF_INET, host_s, &sin->sin_addr.s_addr)) {
373 VLOG_ERR("%s: bad IPv4 address \"%s\"", s, host_s);
381 memset(ss, 0, sizeof *ss);
385 /* Parses 'target', which should be a string in the format "<host>[:<port>]".
386 * <host>, which is required, may be an IPv4 address or an IPv6 address
387 * enclosed in square brackets. If 'default_port' is nonzero then <port> is
388 * optional and defaults to 'default_port'.
390 * On success, returns true and stores the parsed remote address into '*ss'.
391 * On failure, logs an error, stores zeros into '*ss', and returns false. */
393 inet_parse_active(const char *target_, uint16_t default_port,
394 struct sockaddr_storage *ss)
396 char *target = xstrdup(target_);
403 host = parse_bracketed_token(&p);
404 port = parse_bracketed_token(&p);
406 VLOG_ERR("%s: host must be specified", target_);
408 } else if (!port && !default_port) {
409 VLOG_ERR("%s: port must be specified", target_);
412 ok = parse_sockaddr_components(ss, host, port, default_port, target_);
415 memset(ss, 0, sizeof *ss);
422 /* Opens a non-blocking IPv4 or IPv6 socket of the specified 'style' and
423 * connects to 'target', which should be a string in the format
424 * "<host>[:<port>]". <host>, which is required, may be an IPv4 address or an
425 * IPv6 address enclosed in square brackets. If 'default_port' is nonzero then
426 * <port> is optional and defaults to 'default_port'.
428 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
430 * On success, returns 0 (indicating connection complete) or EAGAIN (indicating
431 * connection in progress), in which case the new file descriptor is stored
432 * into '*fdp'. On failure, returns a positive errno value other than EAGAIN
433 * and stores -1 into '*fdp'.
435 * If 'ss' is non-null, then on success stores the target address into '*ss'.
437 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
438 * should be in the range [0, 63] and will automatically be shifted to the
439 * appropriately place in the IP tos field. */
441 inet_open_active(int style, const char *target, uint16_t default_port,
442 struct sockaddr_storage *ssp, int *fdp, uint8_t dscp)
444 struct sockaddr_storage ss;
449 if (!inet_parse_active(target, default_port, &ss)) {
450 error = EAFNOSUPPORT;
454 /* Create non-blocking socket. */
455 fd = socket(ss.ss_family, style, 0);
457 error = sock_errno();
458 VLOG_ERR("%s: socket: %s", target, sock_strerror(error));
461 error = set_nonblocking(fd);
466 /* The dscp bits must be configured before connect() to ensure that the
467 * TOS field is set during the connection establishment. If set after
468 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
469 error = set_dscp(fd, ss.ss_family, dscp);
471 VLOG_ERR("%s: set_dscp: %s", target, sock_strerror(error));
476 error = connect(fd, (struct sockaddr *) &ss, ss_length(&ss)) == 0
479 if (error == EINPROGRESS
481 || error == WSAEALREADY || error == WSAEWOULDBLOCK
488 if (error && error != EAGAIN) {
490 memset(ssp, 0, sizeof *ssp);
505 /* Parses 'target', which should be a string in the format "[<port>][:<host>]":
507 * - If 'default_port' is -1, then <port> is required. Otherwise, if
508 * <port> is omitted, then 'default_port' is used instead.
510 * - If <port> (or 'default_port', if used) is 0, then no port is bound
511 * and the TCP/IP stack will select a port.
513 * - <host> is optional. If supplied, it may be an IPv4 address or an
514 * IPv6 address enclosed in square brackets. If omitted, the IP address
517 * If successful, stores the address into '*ss' and returns true; otherwise
518 * zeros '*ss' and returns false. */
520 inet_parse_passive(const char *target_, int default_port,
521 struct sockaddr_storage *ss)
523 char *target = xstrdup(target_);
530 port = parse_bracketed_token(&p);
531 host = parse_bracketed_token(&p);
532 if (!port && default_port < 0) {
533 VLOG_ERR("%s: port must be specified", target_);
536 ok = parse_sockaddr_components(ss, host ? host : "0.0.0.0",
537 port, default_port, target_);
540 memset(ss, 0, sizeof *ss);
547 /* Opens a non-blocking IPv4 or IPv6 socket of the specified 'style', binds to
548 * 'target', and listens for incoming connections. Parses 'target' in the same
549 * way was inet_parse_passive().
551 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
553 * For TCP, the socket will have SO_REUSEADDR turned on.
555 * On success, returns a non-negative file descriptor. On failure, returns a
556 * negative errno value.
558 * If 'ss' is non-null, then on success stores the bound address into '*ss'.
560 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
561 * should be in the range [0, 63] and will automatically be shifted to the
562 * appropriately place in the IP tos field.
564 * If 'kernel_print_port' is true and the port is dynamically assigned by
565 * the kernel, print the chosen port. */
567 inet_open_passive(int style, const char *target, int default_port,
568 struct sockaddr_storage *ssp, uint8_t dscp,
569 bool kernel_print_port)
571 bool kernel_chooses_port;
572 struct sockaddr_storage ss;
574 unsigned int yes = 1;
576 if (!inet_parse_passive(target, default_port, &ss)) {
577 return -EAFNOSUPPORT;
579 kernel_chooses_port = ss_get_port(&ss) == 0;
581 /* Create non-blocking socket, set SO_REUSEADDR. */
582 fd = socket(ss.ss_family, style, 0);
584 error = sock_errno();
585 VLOG_ERR("%s: socket: %s", target, sock_strerror(error));
588 error = set_nonblocking(fd);
592 if (style == SOCK_STREAM
593 && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) {
594 error = sock_errno();
595 VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s",
596 target, sock_strerror(error));
601 if (bind(fd, (struct sockaddr *) &ss, ss_length(&ss)) < 0) {
602 error = sock_errno();
603 VLOG_ERR("%s: bind: %s", target, sock_strerror(error));
607 /* The dscp bits must be configured before connect() to ensure that the TOS
608 * field is set during the connection establishment. If set after
609 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
610 error = set_dscp(fd, ss.ss_family, dscp);
612 VLOG_ERR("%s: set_dscp: %s", target, sock_strerror(error));
617 if (style == SOCK_STREAM && listen(fd, 10) < 0) {
618 error = sock_errno();
619 VLOG_ERR("%s: listen: %s", target, sock_strerror(error));
623 if (ssp || kernel_chooses_port) {
624 socklen_t ss_len = sizeof ss;
625 if (getsockname(fd, (struct sockaddr *) &ss, &ss_len) < 0) {
626 error = sock_errno();
627 VLOG_ERR("%s: getsockname: %s", target, sock_strerror(error));
630 if (kernel_chooses_port && kernel_print_port) {
631 VLOG_INFO("%s: listening on port %"PRIu16,
632 target, ss_get_port(&ss));
643 memset(ssp, 0, sizeof *ssp);
650 read_fully(int fd, void *p_, size_t size, size_t *bytes_read)
656 ssize_t retval = read(fd, p, size);
658 *bytes_read += retval;
661 } else if (retval == 0) {
663 } else if (errno != EINTR) {
671 write_fully(int fd, const void *p_, size_t size, size_t *bytes_written)
673 const uint8_t *p = p_;
677 ssize_t retval = write(fd, p, size);
679 *bytes_written += retval;
682 } else if (retval == 0) {
683 VLOG_WARN("write returned 0");
685 } else if (errno != EINTR) {
692 /* Given file name 'file_name', fsyncs the directory in which it is contained.
693 * Returns 0 if successful, otherwise a positive errno value. */
695 fsync_parent_dir(const char *file_name)
702 dir = dir_name(file_name);
703 fd = open(dir, O_RDONLY);
706 if (errno == EINVAL || errno == EROFS) {
707 /* This directory does not support synchronization. Not
708 * really an error. */
711 VLOG_ERR("%s: fsync failed (%s)", dir, ovs_strerror(error));
717 VLOG_ERR("%s: open failed (%s)", dir, ovs_strerror(error));
725 /* Obtains the modification time of the file named 'file_name' to the greatest
726 * supported precision. If successful, stores the mtime in '*mtime' and
727 * returns 0. On error, returns a positive errno value and stores zeros in
730 get_mtime(const char *file_name, struct timespec *mtime)
734 if (!stat(file_name, &s)) {
735 mtime->tv_sec = s.st_mtime;
737 #if HAVE_STRUCT_STAT_ST_MTIM_TV_NSEC
738 mtime->tv_nsec = s.st_mtim.tv_nsec;
739 #elif HAVE_STRUCT_STAT_ST_MTIMENSEC
740 mtime->tv_nsec = s.st_mtimensec;
747 mtime->tv_sec = mtime->tv_nsec = 0;
753 getsockopt_int(int fd, int level, int option, const char *optname, int *valuep)
755 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
761 if (getsockopt(fd, level, option, &value, &len)) {
762 error = sock_errno();
763 VLOG_ERR_RL(&rl, "getsockopt(%s): %s", optname, sock_strerror(error));
764 } else if (len != sizeof value) {
766 VLOG_ERR_RL(&rl, "getsockopt(%s): value is %u bytes (expected %"PRIuSIZE")",
767 optname, (unsigned int) len, sizeof value);
772 *valuep = error ? 0 : value;
777 describe_sockaddr(struct ds *string, int fd,
778 int (*getaddr)(int, struct sockaddr *, socklen_t *))
780 struct sockaddr_storage ss;
781 socklen_t len = sizeof ss;
783 if (!getaddr(fd, (struct sockaddr *) &ss, &len)) {
784 if (ss.ss_family == AF_INET || ss.ss_family == AF_INET6) {
785 char addrbuf[SS_NTOP_BUFSIZE];
787 ds_put_format(string, "%s:%"PRIu16,
788 ss_format_address(&ss, addrbuf, sizeof addrbuf),
791 } else if (ss.ss_family == AF_UNIX) {
792 struct sockaddr_un sun;
796 memcpy(&sun, &ss, sizeof sun);
797 maxlen = len - offsetof(struct sockaddr_un, sun_path);
798 null = memchr(sun.sun_path, '\0', maxlen);
799 ds_put_buffer(string, sun.sun_path,
800 null ? null - sun.sun_path : maxlen);
804 else if (ss.ss_family == AF_NETLINK) {
807 /* SO_PROTOCOL was introduced in 2.6.32. Support it regardless of the version
808 * of the Linux kernel headers in use at build time. */
810 #define SO_PROTOCOL 38
813 if (!getsockopt_int(fd, SOL_SOCKET, SO_PROTOCOL, "SO_PROTOCOL",
817 ds_put_cstr(string, "NETLINK_ROUTE");
820 case NETLINK_GENERIC:
821 ds_put_cstr(string, "NETLINK_GENERIC");
825 ds_put_format(string, "AF_NETLINK family %d", protocol);
829 ds_put_cstr(string, "AF_NETLINK");
834 else if (ss.ss_family == AF_PACKET) {
835 struct sockaddr_ll sll;
837 memcpy(&sll, &ss, sizeof sll);
838 ds_put_cstr(string, "AF_PACKET");
839 if (sll.sll_ifindex) {
842 if (if_indextoname(sll.sll_ifindex, name)) {
843 ds_put_format(string, "(%s)", name);
845 ds_put_format(string, "(ifindex=%d)", sll.sll_ifindex);
848 if (sll.sll_protocol) {
849 ds_put_format(string, "(protocol=0x%"PRIu16")",
850 ntohs(sll.sll_protocol));
854 else if (ss.ss_family == AF_UNSPEC) {
855 ds_put_cstr(string, "AF_UNSPEC");
857 ds_put_format(string, "AF_%d", (int) ss.ss_family);
865 put_fd_filename(struct ds *string, int fd)
871 linkname = xasprintf("/proc/self/fd/%d", fd);
872 n = readlink(linkname, buf, sizeof buf);
874 ds_put_char(string, ' ');
875 ds_put_buffer(string, buf, n);
876 if (n > sizeof buf) {
877 ds_put_cstr(string, "...");
884 /* Returns a malloc()'d string describing 'fd', for use in logging. */
894 ds_put_format(&string, "fstat failed (%s)", ovs_strerror(errno));
895 } else if (S_ISSOCK(s.st_mode)) {
896 describe_sockaddr(&string, fd, getsockname);
897 ds_put_cstr(&string, "<->");
898 describe_sockaddr(&string, fd, getpeername);
900 ds_put_cstr(&string, (isatty(fd) ? "tty"
901 : S_ISDIR(s.st_mode) ? "directory"
902 : S_ISCHR(s.st_mode) ? "character device"
903 : S_ISBLK(s.st_mode) ? "block device"
904 : S_ISREG(s.st_mode) ? "file"
905 : S_ISFIFO(s.st_mode) ? "FIFO"
906 : S_ISLNK(s.st_mode) ? "symbolic link"
909 put_fd_filename(&string, fd);
913 ds_put_format(&string,"file descriptor");
915 return ds_steal_cstr(&string);
919 /* sockaddr_storage helpers. */
921 /* Returns the IPv4 or IPv6 port in 'ss'. */
923 ss_get_port(const struct sockaddr_storage *ss)
925 if (ss->ss_family == AF_INET) {
926 const struct sockaddr_in *sin
927 = ALIGNED_CAST(const struct sockaddr_in *, ss);
928 return ntohs(sin->sin_port);
929 } else if (ss->ss_family == AF_INET6) {
930 const struct sockaddr_in6 *sin6
931 = ALIGNED_CAST(const struct sockaddr_in6 *, ss);
932 return ntohs(sin6->sin6_port);
938 /* Formats the IPv4 or IPv6 address in 'ss' into the 'bufsize' bytes in 'buf'.
939 * If 'ss' is an IPv6 address, puts square brackets around the address.
940 * 'bufsize' should be at least SS_NTOP_BUFSIZE.
944 ss_format_address(const struct sockaddr_storage *ss,
945 char *buf, size_t bufsize)
947 ovs_assert(bufsize >= SS_NTOP_BUFSIZE);
948 if (ss->ss_family == AF_INET) {
949 const struct sockaddr_in *sin
950 = ALIGNED_CAST(const struct sockaddr_in *, ss);
952 snprintf(buf, bufsize, IP_FMT, IP_ARGS(sin->sin_addr.s_addr));
953 } else if (ss->ss_family == AF_INET6) {
954 const struct sockaddr_in6 *sin6
955 = ALIGNED_CAST(const struct sockaddr_in6 *, ss);
958 inet_ntop(AF_INET6, sin6->sin6_addr.s6_addr, buf + 1, bufsize - 1);
959 strcpy(strchr(buf, '\0'), "]");
968 ss_length(const struct sockaddr_storage *ss)
970 switch (ss->ss_family) {
972 return sizeof(struct sockaddr_in);
975 return sizeof(struct sockaddr_in6);
982 /* For Windows socket calls, 'errno' is not set. One has to call
983 * WSAGetLastError() to get the error number and then pass it to
984 * this function to get the correct error string.
986 * ovs_strerror() calls strerror_r() and would not get the correct error
987 * string for Windows sockets, but is good for POSIX. */
989 sock_strerror(int error)
992 return ovs_format_message(error);
994 return ovs_strerror(error);