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 "openvswitch/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 /* Checks whether 'host_name' is an IPv4 or IPv6 address. It is assumed
140 * that 'host_name' is valid. Returns false if it is IPv4 address, true if
141 * it is IPv6 address. */
143 addr_is_ipv6(const char *host_name)
145 return strchr(host_name, ':') != NULL;
148 /* Translates 'host_name', which must be a string representation of an IP
149 * address, into a numeric IP address in '*addr'. Returns 0 if successful,
150 * otherwise a positive errno value. */
152 lookup_ip(const char *host_name, struct in_addr *addr)
154 if (!ip_parse(host_name, &addr->s_addr)) {
155 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
156 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name);
162 /* Translates 'host_name', which must be a string representation of an IPv6
163 * address, into a numeric IPv6 address in '*addr'. Returns 0 if successful,
164 * otherwise a positive errno value. */
166 lookup_ipv6(const char *host_name, struct in6_addr *addr)
168 if (!ipv6_parse(host_name, addr)) {
169 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
170 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IPv6 address", host_name);
176 /* Translates 'host_name', which must be a host name or a string representation
177 * of an IP address, into a numeric IP address in '*addr'. Returns 0 if
178 * successful, otherwise a positive errno value.
180 * Most Open vSwitch code should not use this because it causes deadlocks:
181 * getaddrinfo() sends out a DNS request but that starts a new flow for which
182 * OVS must set up a flow, but it can't because it's waiting for a DNS reply.
183 * The synchronous lookup also delays other activity. (Of course we can solve
184 * this but it doesn't seem worthwhile quite yet.) */
186 lookup_hostname(const char *host_name, struct in_addr *addr)
188 struct addrinfo *result;
189 struct addrinfo hints;
191 if (ip_parse(host_name, &addr->s_addr)) {
195 memset(&hints, 0, sizeof hints);
196 hints.ai_family = AF_INET;
198 switch (getaddrinfo(host_name, NULL, &hints, &result)) {
200 *addr = ALIGNED_CAST(struct sockaddr_in *,
201 result->ai_addr)->sin_addr;
202 freeaddrinfo(result);
205 #ifdef EAI_ADDRFAMILY
226 #if defined (EAI_NODATA) && EAI_NODATA != EAI_NONAME
242 check_connection_completion(int fd)
244 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
249 pfd.events = POLLOUT;
253 retval = poll(&pfd, 1, 0);
254 } while (retval < 0 && errno == EINTR);
256 retval = WSAPoll(&pfd, 1, 0);
259 if (pfd.revents & POLLERR) {
260 ssize_t n = send(fd, "", 1, 0);
264 VLOG_ERR_RL(&rl, "poll return POLLERR but send succeeded");
269 } else if (retval < 0) {
270 VLOG_ERR_RL(&rl, "poll: %s", sock_strerror(sock_errno()));
277 /* Returns the size of socket 'sock''s receive buffer (SO_RCVBUF), or a
278 * negative errno value if an error occurs. */
280 get_socket_rcvbuf(int sock)
285 error = getsockopt_int(sock, SOL_SOCKET, SO_RCVBUF, "SO_RCVBUF", &rcvbuf);
286 return error ? -error : rcvbuf;
289 /* Reads and discards up to 'n' datagrams from 'fd', stopping as soon as no
290 * more data can be immediately read. ('fd' should therefore be in
291 * non-blocking mode.)*/
293 drain_fd(int fd, size_t n_packets)
295 for (; n_packets > 0; n_packets--) {
296 /* 'buffer' only needs to be 1 byte long in most circumstances. This
297 * size is defensive against the possibility that we someday want to
298 * use a Linux tap device without TUN_NO_PI, in which case a buffer
299 * smaller than sizeof(struct tun_pi) will give EINVAL on read. */
301 if (read(fd, buffer, sizeof buffer) <= 0) {
308 guess_netmask(ovs_be32 ip_)
310 uint32_t ip = ntohl(ip_);
311 return ((ip >> 31) == 0 ? htonl(0xff000000) /* Class A */
312 : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */
313 : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */
314 : htonl(0)); /* ??? */
317 /* This is like strsep() except:
319 * - The separator string is ":".
321 * - Square brackets [] quote ":" separators and are removed from the
324 parse_bracketed_token(char **pp)
330 } else if (*p == '\0') {
333 } else if (*p == '[') {
335 char *end = start + strcspn(start, "]");
336 *pp = (*end == '\0' ? NULL
337 : end[1] == ':' ? end + 2
343 char *end = start + strcspn(start, ":");
344 *pp = *end == '\0' ? NULL : end + 1;
351 parse_sockaddr_components(struct sockaddr_storage *ss,
353 const char *port_s, uint16_t default_port,
356 struct sockaddr_in *sin = ALIGNED_CAST(struct sockaddr_in *, ss);
359 if (port_s && port_s[0]) {
360 if (!str_to_int(port_s, 10, &port) || port < 0 || port > 65535) {
361 VLOG_ERR("%s: bad port number \"%s\"", s, port_s);
368 memset(ss, 0, sizeof *ss);
369 if (strchr(host_s, ':')) {
370 struct sockaddr_in6 *sin6
371 = ALIGNED_CAST(struct sockaddr_in6 *, ss);
373 sin6->sin6_family = AF_INET6;
374 sin6->sin6_port = htons(port);
375 if (!ipv6_parse(host_s, &sin6->sin6_addr)) {
376 VLOG_ERR("%s: bad IPv6 address \"%s\"", s, host_s);
380 sin->sin_family = AF_INET;
381 sin->sin_port = htons(port);
382 if (!ip_parse(host_s, &sin->sin_addr.s_addr)) {
383 VLOG_ERR("%s: bad IPv4 address \"%s\"", s, host_s);
391 memset(ss, 0, sizeof *ss);
395 /* Parses 'target', which should be a string in the format "<host>[:<port>]".
396 * <host>, which is required, may be an IPv4 address or an IPv6 address
397 * enclosed in square brackets. If 'default_port' is nonzero then <port> is
398 * optional and defaults to 'default_port'.
400 * On success, returns true and stores the parsed remote address into '*ss'.
401 * On failure, logs an error, stores zeros into '*ss', and returns false. */
403 inet_parse_active(const char *target_, uint16_t default_port,
404 struct sockaddr_storage *ss)
406 char *target = xstrdup(target_);
413 host = parse_bracketed_token(&p);
414 port = parse_bracketed_token(&p);
416 VLOG_ERR("%s: host must be specified", target_);
418 } else if (!port && !default_port) {
419 VLOG_ERR("%s: port must be specified", target_);
422 ok = parse_sockaddr_components(ss, host, port, default_port, target_);
425 memset(ss, 0, sizeof *ss);
432 /* Opens a non-blocking IPv4 or IPv6 socket of the specified 'style' and
433 * connects to 'target', which should be a string in the format
434 * "<host>[:<port>]". <host>, which is required, may be an IPv4 address or an
435 * IPv6 address enclosed in square brackets. If 'default_port' is nonzero then
436 * <port> is optional and defaults to 'default_port'.
438 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
440 * On success, returns 0 (indicating connection complete) or EAGAIN (indicating
441 * connection in progress), in which case the new file descriptor is stored
442 * into '*fdp'. On failure, returns a positive errno value other than EAGAIN
443 * and stores -1 into '*fdp'.
445 * If 'ss' is non-null, then on success stores the target address into '*ss'.
447 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
448 * should be in the range [0, 63] and will automatically be shifted to the
449 * appropriately place in the IP tos field. */
451 inet_open_active(int style, const char *target, uint16_t default_port,
452 struct sockaddr_storage *ssp, int *fdp, uint8_t dscp)
454 struct sockaddr_storage ss;
459 if (!inet_parse_active(target, default_port, &ss)) {
460 error = EAFNOSUPPORT;
464 /* Create non-blocking socket. */
465 fd = socket(ss.ss_family, style, 0);
467 error = sock_errno();
468 VLOG_ERR("%s: socket: %s", target, sock_strerror(error));
471 error = set_nonblocking(fd);
476 /* The dscp bits must be configured before connect() to ensure that the
477 * TOS field is set during the connection establishment. If set after
478 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
479 error = set_dscp(fd, ss.ss_family, dscp);
481 VLOG_ERR("%s: set_dscp: %s", target, sock_strerror(error));
486 error = connect(fd, (struct sockaddr *) &ss, ss_length(&ss)) == 0
489 if (error == EINPROGRESS
491 || error == WSAEALREADY || error == WSAEWOULDBLOCK
498 if (error && error != EAGAIN) {
500 memset(ssp, 0, sizeof *ssp);
515 /* Parses 'target', which should be a string in the format "[<port>][:<host>]":
517 * - If 'default_port' is -1, then <port> is required. Otherwise, if
518 * <port> is omitted, then 'default_port' is used instead.
520 * - If <port> (or 'default_port', if used) is 0, then no port is bound
521 * and the TCP/IP stack will select a port.
523 * - <host> is optional. If supplied, it may be an IPv4 address or an
524 * IPv6 address enclosed in square brackets. If omitted, the IP address
527 * If successful, stores the address into '*ss' and returns true; otherwise
528 * zeros '*ss' and returns false. */
530 inet_parse_passive(const char *target_, int default_port,
531 struct sockaddr_storage *ss)
533 char *target = xstrdup(target_);
540 port = parse_bracketed_token(&p);
541 host = parse_bracketed_token(&p);
542 if (!port && default_port < 0) {
543 VLOG_ERR("%s: port must be specified", target_);
546 ok = parse_sockaddr_components(ss, host ? host : "0.0.0.0",
547 port, default_port, target_);
550 memset(ss, 0, sizeof *ss);
557 /* Opens a non-blocking IPv4 or IPv6 socket of the specified 'style', binds to
558 * 'target', and listens for incoming connections. Parses 'target' in the same
559 * way was inet_parse_passive().
561 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
563 * For TCP, the socket will have SO_REUSEADDR turned on.
565 * On success, returns a non-negative file descriptor. On failure, returns a
566 * negative errno value.
568 * If 'ss' is non-null, then on success stores the bound address into '*ss'.
570 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
571 * should be in the range [0, 63] and will automatically be shifted to the
572 * appropriately place in the IP tos field.
574 * If 'kernel_print_port' is true and the port is dynamically assigned by
575 * the kernel, print the chosen port. */
577 inet_open_passive(int style, const char *target, int default_port,
578 struct sockaddr_storage *ssp, uint8_t dscp,
579 bool kernel_print_port)
581 bool kernel_chooses_port;
582 struct sockaddr_storage ss;
584 unsigned int yes = 1;
586 if (!inet_parse_passive(target, default_port, &ss)) {
587 return -EAFNOSUPPORT;
589 kernel_chooses_port = ss_get_port(&ss) == 0;
591 /* Create non-blocking socket, set SO_REUSEADDR. */
592 fd = socket(ss.ss_family, style, 0);
594 error = sock_errno();
595 VLOG_ERR("%s: socket: %s", target, sock_strerror(error));
598 error = set_nonblocking(fd);
602 if (style == SOCK_STREAM
603 && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) {
604 error = sock_errno();
605 VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s",
606 target, sock_strerror(error));
611 if (bind(fd, (struct sockaddr *) &ss, ss_length(&ss)) < 0) {
612 error = sock_errno();
613 VLOG_ERR("%s: bind: %s", target, sock_strerror(error));
617 /* The dscp bits must be configured before connect() to ensure that the TOS
618 * field is set during the connection establishment. If set after
619 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
620 error = set_dscp(fd, ss.ss_family, dscp);
622 VLOG_ERR("%s: set_dscp: %s", target, sock_strerror(error));
627 if (style == SOCK_STREAM && listen(fd, 10) < 0) {
628 error = sock_errno();
629 VLOG_ERR("%s: listen: %s", target, sock_strerror(error));
633 if (ssp || kernel_chooses_port) {
634 socklen_t ss_len = sizeof ss;
635 if (getsockname(fd, (struct sockaddr *) &ss, &ss_len) < 0) {
636 error = sock_errno();
637 VLOG_ERR("%s: getsockname: %s", target, sock_strerror(error));
640 if (kernel_chooses_port && kernel_print_port) {
641 VLOG_INFO("%s: listening on port %"PRIu16,
642 target, ss_get_port(&ss));
653 memset(ssp, 0, sizeof *ssp);
660 read_fully(int fd, void *p_, size_t size, size_t *bytes_read)
666 ssize_t retval = read(fd, p, size);
668 *bytes_read += retval;
671 } else if (retval == 0) {
673 } else if (errno != EINTR) {
681 write_fully(int fd, const void *p_, size_t size, size_t *bytes_written)
683 const uint8_t *p = p_;
687 ssize_t retval = write(fd, p, size);
689 *bytes_written += retval;
692 } else if (retval == 0) {
693 VLOG_WARN("write returned 0");
695 } else if (errno != EINTR) {
702 /* Given file name 'file_name', fsyncs the directory in which it is contained.
703 * Returns 0 if successful, otherwise a positive errno value. */
705 fsync_parent_dir(const char *file_name)
712 dir = dir_name(file_name);
713 fd = open(dir, O_RDONLY);
716 if (errno == EINVAL || errno == EROFS) {
717 /* This directory does not support synchronization. Not
718 * really an error. */
721 VLOG_ERR("%s: fsync failed (%s)", dir, ovs_strerror(error));
727 VLOG_ERR("%s: open failed (%s)", dir, ovs_strerror(error));
735 /* Obtains the modification time of the file named 'file_name' to the greatest
736 * supported precision. If successful, stores the mtime in '*mtime' and
737 * returns 0. On error, returns a positive errno value and stores zeros in
740 get_mtime(const char *file_name, struct timespec *mtime)
744 if (!stat(file_name, &s)) {
745 mtime->tv_sec = s.st_mtime;
747 #if HAVE_STRUCT_STAT_ST_MTIM_TV_NSEC
748 mtime->tv_nsec = s.st_mtim.tv_nsec;
749 #elif HAVE_STRUCT_STAT_ST_MTIMENSEC
750 mtime->tv_nsec = s.st_mtimensec;
757 mtime->tv_sec = mtime->tv_nsec = 0;
763 getsockopt_int(int fd, int level, int option, const char *optname, int *valuep)
765 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
771 if (getsockopt(fd, level, option, &value, &len)) {
772 error = sock_errno();
773 VLOG_ERR_RL(&rl, "getsockopt(%s): %s", optname, sock_strerror(error));
774 } else if (len != sizeof value) {
776 VLOG_ERR_RL(&rl, "getsockopt(%s): value is %u bytes (expected %"PRIuSIZE")",
777 optname, (unsigned int) len, sizeof value);
782 *valuep = error ? 0 : value;
787 describe_sockaddr(struct ds *string, int fd,
788 int (*getaddr)(int, struct sockaddr *, socklen_t *))
790 struct sockaddr_storage ss;
791 socklen_t len = sizeof ss;
793 if (!getaddr(fd, (struct sockaddr *) &ss, &len)) {
794 if (ss.ss_family == AF_INET || ss.ss_family == AF_INET6) {
795 char addrbuf[SS_NTOP_BUFSIZE];
797 ds_put_format(string, "%s:%"PRIu16,
798 ss_format_address(&ss, addrbuf, sizeof addrbuf),
801 } else if (ss.ss_family == AF_UNIX) {
802 struct sockaddr_un sun;
806 memcpy(&sun, &ss, sizeof sun);
807 maxlen = len - offsetof(struct sockaddr_un, sun_path);
808 null = memchr(sun.sun_path, '\0', maxlen);
809 ds_put_buffer(string, sun.sun_path,
810 null ? null - sun.sun_path : maxlen);
814 else if (ss.ss_family == AF_NETLINK) {
817 /* SO_PROTOCOL was introduced in 2.6.32. Support it regardless of the version
818 * of the Linux kernel headers in use at build time. */
820 #define SO_PROTOCOL 38
823 if (!getsockopt_int(fd, SOL_SOCKET, SO_PROTOCOL, "SO_PROTOCOL",
827 ds_put_cstr(string, "NETLINK_ROUTE");
830 case NETLINK_GENERIC:
831 ds_put_cstr(string, "NETLINK_GENERIC");
835 ds_put_format(string, "AF_NETLINK family %d", protocol);
839 ds_put_cstr(string, "AF_NETLINK");
844 else if (ss.ss_family == AF_PACKET) {
845 struct sockaddr_ll sll;
847 memcpy(&sll, &ss, sizeof sll);
848 ds_put_cstr(string, "AF_PACKET");
849 if (sll.sll_ifindex) {
852 if (if_indextoname(sll.sll_ifindex, name)) {
853 ds_put_format(string, "(%s)", name);
855 ds_put_format(string, "(ifindex=%d)", sll.sll_ifindex);
858 if (sll.sll_protocol) {
859 ds_put_format(string, "(protocol=0x%"PRIu16")",
860 ntohs(sll.sll_protocol));
864 else if (ss.ss_family == AF_UNSPEC) {
865 ds_put_cstr(string, "AF_UNSPEC");
867 ds_put_format(string, "AF_%d", (int) ss.ss_family);
875 put_fd_filename(struct ds *string, int fd)
881 linkname = xasprintf("/proc/self/fd/%d", fd);
882 n = readlink(linkname, buf, sizeof buf);
884 ds_put_char(string, ' ');
885 ds_put_buffer(string, buf, n);
886 if (n > sizeof buf) {
887 ds_put_cstr(string, "...");
894 /* Returns a malloc()'d string describing 'fd', for use in logging. */
904 ds_put_format(&string, "fstat failed (%s)", ovs_strerror(errno));
905 } else if (S_ISSOCK(s.st_mode)) {
906 describe_sockaddr(&string, fd, getsockname);
907 ds_put_cstr(&string, "<->");
908 describe_sockaddr(&string, fd, getpeername);
910 ds_put_cstr(&string, (isatty(fd) ? "tty"
911 : S_ISDIR(s.st_mode) ? "directory"
912 : S_ISCHR(s.st_mode) ? "character device"
913 : S_ISBLK(s.st_mode) ? "block device"
914 : S_ISREG(s.st_mode) ? "file"
915 : S_ISFIFO(s.st_mode) ? "FIFO"
916 : S_ISLNK(s.st_mode) ? "symbolic link"
919 put_fd_filename(&string, fd);
923 ds_put_format(&string,"file descriptor");
925 return ds_steal_cstr(&string);
929 /* sockaddr_storage helpers. */
931 /* Returns the IPv4 or IPv6 port in 'ss'. */
933 ss_get_port(const struct sockaddr_storage *ss)
935 if (ss->ss_family == AF_INET) {
936 const struct sockaddr_in *sin
937 = ALIGNED_CAST(const struct sockaddr_in *, ss);
938 return ntohs(sin->sin_port);
939 } else if (ss->ss_family == AF_INET6) {
940 const struct sockaddr_in6 *sin6
941 = ALIGNED_CAST(const struct sockaddr_in6 *, ss);
942 return ntohs(sin6->sin6_port);
948 /* Formats the IPv4 or IPv6 address in 'ss' into the 'bufsize' bytes in 'buf'.
949 * If 'ss' is an IPv6 address, puts square brackets around the address.
950 * 'bufsize' should be at least SS_NTOP_BUFSIZE.
954 ss_format_address(const struct sockaddr_storage *ss,
955 char *buf, size_t bufsize)
957 ovs_assert(bufsize >= SS_NTOP_BUFSIZE);
958 if (ss->ss_family == AF_INET) {
959 const struct sockaddr_in *sin
960 = ALIGNED_CAST(const struct sockaddr_in *, ss);
962 snprintf(buf, bufsize, IP_FMT, IP_ARGS(sin->sin_addr.s_addr));
963 } else if (ss->ss_family == AF_INET6) {
964 const struct sockaddr_in6 *sin6
965 = ALIGNED_CAST(const struct sockaddr_in6 *, ss);
968 inet_ntop(AF_INET6, sin6->sin6_addr.s6_addr, buf + 1, bufsize - 1);
969 strcpy(strchr(buf, '\0'), "]");
978 ss_length(const struct sockaddr_storage *ss)
980 switch (ss->ss_family) {
982 return sizeof(struct sockaddr_in);
985 return sizeof(struct sockaddr_in6);
992 /* For Windows socket calls, 'errno' is not set. One has to call
993 * WSAGetLastError() to get the error number and then pass it to
994 * this function to get the correct error string.
996 * ovs_strerror() calls strerror_r() and would not get the correct error
997 * string for Windows sockets, but is good for POSIX. */
999 sock_strerror(int error)
1002 return ovs_format_message(error);
1004 return ovs_strerror(error);