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>
29 #include <sys/socket.h>
34 #include "dynamic-string.h"
35 #include "ovs-thread.h"
37 #include "poll-loop.h"
41 #include <linux/if_packet.h>
44 #include "netlink-protocol.h"
45 #include "netlink-socket.h"
48 VLOG_DEFINE_THIS_MODULE(socket_util);
50 static int getsockopt_int(int fd, int level, int option, const char *optname,
53 /* Sets 'fd' to non-blocking mode. Returns 0 if successful, otherwise a
54 * positive errno value. */
56 set_nonblocking(int fd)
59 int flags = fcntl(fd, F_GETFL, 0);
61 if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) != -1) {
64 VLOG_ERR("fcntl(F_SETFL) failed: %s", ovs_strerror(errno));
68 VLOG_ERR("fcntl(F_GETFL) failed: %s", ovs_strerror(errno));
72 unsigned long arg = 1;
73 if (ioctlsocket(fd, FIONBIO, &arg)) {
74 int error = sock_errno();
75 VLOG_ERR("set_nonblocking failed: %s", sock_strerror(error));
83 xset_nonblocking(int fd)
85 if (set_nonblocking(fd)) {
91 set_dscp(int fd, uint8_t dscp)
97 /* XXX: Consider using QoS2 APIs for Windows to set dscp. */
105 /* Note: this function is used for both of IPv4 and IPv6 sockets */
108 if (setsockopt(fd, IPPROTO_IP, IP_TOS, &val, sizeof val)) {
109 if (sock_errno() != ENOPROTOOPT) {
115 if (setsockopt(fd, IPPROTO_IPV6, IPV6_TCLASS, &val, sizeof val)) {
116 if (sock_errno() != ENOPROTOOPT) {
129 /* Translates 'host_name', which must be a string representation of an IP
130 * address, into a numeric IP address in '*addr'. Returns 0 if successful,
131 * otherwise a positive errno value. */
133 lookup_ip(const char *host_name, struct in_addr *addr)
135 if (!inet_pton(AF_INET, host_name, addr)) {
136 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
137 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name);
143 /* Translates 'host_name', which must be a string representation of an IPv6
144 * address, into a numeric IPv6 address in '*addr'. Returns 0 if successful,
145 * otherwise a positive errno value. */
147 lookup_ipv6(const char *host_name, struct in6_addr *addr)
149 if (inet_pton(AF_INET6, host_name, addr) != 1) {
150 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
151 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IPv6 address", host_name);
157 /* Translates 'host_name', which must be a host name or a string representation
158 * of an IP address, into a numeric IP address in '*addr'. Returns 0 if
159 * successful, otherwise a positive errno value.
161 * Most Open vSwitch code should not use this because it causes deadlocks:
162 * getaddrinfo() sends out a DNS request but that starts a new flow for which
163 * OVS must set up a flow, but it can't because it's waiting for a DNS reply.
164 * The synchronous lookup also delays other activity. (Of course we can solve
165 * this but it doesn't seem worthwhile quite yet.) */
167 lookup_hostname(const char *host_name, struct in_addr *addr)
169 struct addrinfo *result;
170 struct addrinfo hints;
172 if (inet_pton(AF_INET, host_name, addr)) {
176 memset(&hints, 0, sizeof hints);
177 hints.ai_family = AF_INET;
179 switch (getaddrinfo(host_name, NULL, &hints, &result)) {
181 *addr = ALIGNED_CAST(struct sockaddr_in *,
182 result->ai_addr)->sin_addr;
183 freeaddrinfo(result);
186 #ifdef EAI_ADDRFAMILY
207 #if defined (EAI_NODATA) && EAI_NODATA != EAI_NONAME
223 check_connection_completion(int fd)
225 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
230 pfd.events = POLLOUT;
234 retval = poll(&pfd, 1, 0);
235 } while (retval < 0 && errno == EINTR);
237 retval = WSAPoll(&pfd, 1, 0);
240 if (pfd.revents & POLLERR) {
241 ssize_t n = send(fd, "", 1, 0);
245 VLOG_ERR_RL(&rl, "poll return POLLERR but send succeeded");
250 } else if (retval < 0) {
251 VLOG_ERR_RL(&rl, "poll: %s", sock_strerror(sock_errno()));
258 /* Returns the size of socket 'sock''s receive buffer (SO_RCVBUF), or a
259 * negative errno value if an error occurs. */
261 get_socket_rcvbuf(int sock)
266 error = getsockopt_int(sock, SOL_SOCKET, SO_RCVBUF, "SO_RCVBUF", &rcvbuf);
267 return error ? -error : rcvbuf;
270 /* Reads and discards up to 'n' datagrams from 'fd', stopping as soon as no
271 * more data can be immediately read. ('fd' should therefore be in
272 * non-blocking mode.)*/
274 drain_fd(int fd, size_t n_packets)
276 for (; n_packets > 0; n_packets--) {
277 /* 'buffer' only needs to be 1 byte long in most circumstances. This
278 * size is defensive against the possibility that we someday want to
279 * use a Linux tap device without TUN_NO_PI, in which case a buffer
280 * smaller than sizeof(struct tun_pi) will give EINVAL on read. */
282 if (read(fd, buffer, sizeof buffer) <= 0) {
289 guess_netmask(ovs_be32 ip_)
291 uint32_t ip = ntohl(ip_);
292 return ((ip >> 31) == 0 ? htonl(0xff000000) /* Class A */
293 : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */
294 : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */
295 : htonl(0)); /* ??? */
298 /* This is like strsep() except:
300 * - The separator string is ":".
302 * - Square brackets [] quote ":" separators and are removed from the
305 parse_bracketed_token(char **pp)
311 } else if (*p == '\0') {
314 } else if (*p == '[') {
316 char *end = start + strcspn(start, "]");
317 *pp = (*end == '\0' ? NULL
318 : end[1] == ':' ? end + 2
324 char *end = start + strcspn(start, ":");
325 *pp = *end == '\0' ? NULL : end + 1;
332 parse_sockaddr_components(struct sockaddr_storage *ss,
334 const char *port_s, uint16_t default_port,
337 struct sockaddr_in *sin = ALIGNED_CAST(struct sockaddr_in *, ss);
340 if (port_s && port_s[0]) {
341 if (!str_to_int(port_s, 10, &port) || port < 0 || port > 65535) {
342 VLOG_ERR("%s: bad port number \"%s\"", s, port_s);
348 memset(ss, 0, sizeof *ss);
349 if (strchr(host_s, ':')) {
350 struct sockaddr_in6 *sin6
351 = ALIGNED_CAST(struct sockaddr_in6 *, ss);
353 sin6->sin6_family = AF_INET6;
354 sin6->sin6_port = htons(port);
355 if (!inet_pton(AF_INET6, host_s, sin6->sin6_addr.s6_addr)) {
356 VLOG_ERR("%s: bad IPv6 address \"%s\"", s, host_s);
360 sin->sin_family = AF_INET;
361 sin->sin_port = htons(port);
362 if (!inet_pton(AF_INET, host_s, &sin->sin_addr.s_addr)) {
363 VLOG_ERR("%s: bad IPv4 address \"%s\"", s, host_s);
371 memset(ss, 0, sizeof *ss);
375 /* Parses 'target', which should be a string in the format "<host>[:<port>]".
376 * <host>, which is required, may be an IPv4 address or an IPv6 address
377 * enclosed in square brackets. If 'default_port' is nonzero then <port> is
378 * optional and defaults to 'default_port'.
380 * On success, returns true and stores the parsed remote address into '*ss'.
381 * On failure, logs an error, stores zeros into '*ss', and returns false. */
383 inet_parse_active(const char *target_, uint16_t default_port,
384 struct sockaddr_storage *ss)
386 char *target = xstrdup(target_);
393 host = parse_bracketed_token(&p);
394 port = parse_bracketed_token(&p);
396 VLOG_ERR("%s: host must be specified", target_);
398 } else if (!port && !default_port) {
399 VLOG_ERR("%s: port must be specified", target_);
402 ok = parse_sockaddr_components(ss, host, port, default_port, target_);
405 memset(ss, 0, sizeof *ss);
412 /* Opens a non-blocking IPv4 or IPv6 socket of the specified 'style' and
413 * connects to 'target', which should be a string in the format
414 * "<host>[:<port>]". <host>, which is required, may be an IPv4 address or an
415 * IPv6 address enclosed in square brackets. If 'default_port' is nonzero then
416 * <port> is optional and defaults to 'default_port'.
418 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
420 * On success, returns 0 (indicating connection complete) or EAGAIN (indicating
421 * connection in progress), in which case the new file descriptor is stored
422 * into '*fdp'. On failure, returns a positive errno value other than EAGAIN
423 * and stores -1 into '*fdp'.
425 * If 'ss' is non-null, then on success stores the target address into '*ss'.
427 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
428 * should be in the range [0, 63] and will automatically be shifted to the
429 * appropriately place in the IP tos field. */
431 inet_open_active(int style, const char *target, uint16_t default_port,
432 struct sockaddr_storage *ssp, int *fdp, uint8_t dscp)
434 struct sockaddr_storage ss;
439 if (!inet_parse_active(target, default_port, &ss)) {
440 error = EAFNOSUPPORT;
444 /* Create non-blocking socket. */
445 fd = socket(ss.ss_family, style, 0);
447 error = sock_errno();
448 VLOG_ERR("%s: socket: %s", target, sock_strerror(error));
451 error = set_nonblocking(fd);
456 /* The dscp bits must be configured before connect() to ensure that the
457 * TOS field is set during the connection establishment. If set after
458 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
459 error = set_dscp(fd, dscp);
461 VLOG_ERR("%s: set_dscp: %s", target, sock_strerror(error));
466 error = connect(fd, (struct sockaddr *) &ss, ss_length(&ss)) == 0
469 if (error == EINPROGRESS
471 || error == WSAEALREADY || error == WSAEWOULDBLOCK
478 if (error && error != EAGAIN) {
480 memset(ssp, 0, sizeof *ssp);
495 /* Parses 'target', which should be a string in the format "[<port>][:<host>]":
497 * - If 'default_port' is -1, then <port> is required. Otherwise, if
498 * <port> is omitted, then 'default_port' is used instead.
500 * - If <port> (or 'default_port', if used) is 0, then no port is bound
501 * and the TCP/IP stack will select a port.
503 * - <host> is optional. If supplied, it may be an IPv4 address or an
504 * IPv6 address enclosed in square brackets. If omitted, the IP address
507 * If successful, stores the address into '*ss' and returns true; otherwise
508 * zeros '*ss' and returns false. */
510 inet_parse_passive(const char *target_, int default_port,
511 struct sockaddr_storage *ss)
513 char *target = xstrdup(target_);
520 port = parse_bracketed_token(&p);
521 host = parse_bracketed_token(&p);
522 if (!port && default_port < 0) {
523 VLOG_ERR("%s: port must be specified", target_);
526 ok = parse_sockaddr_components(ss, host ? host : "0.0.0.0",
527 port, default_port, target_);
530 memset(ss, 0, sizeof *ss);
537 /* Opens a non-blocking IPv4 or IPv6 socket of the specified 'style', binds to
538 * 'target', and listens for incoming connections. Parses 'target' in the same
539 * way was inet_parse_passive().
541 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
543 * For TCP, the socket will have SO_REUSEADDR turned on.
545 * On success, returns a non-negative file descriptor. On failure, returns a
546 * negative errno value.
548 * If 'ss' is non-null, then on success stores the bound address into '*ss'.
550 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
551 * should be in the range [0, 63] and will automatically be shifted to the
552 * appropriately place in the IP tos field.
554 * If 'kernel_print_port' is true and the port is dynamically assigned by
555 * the kernel, print the chosen port. */
557 inet_open_passive(int style, const char *target, int default_port,
558 struct sockaddr_storage *ssp, uint8_t dscp,
559 bool kernel_print_port)
561 bool kernel_chooses_port;
562 struct sockaddr_storage ss;
564 unsigned int yes = 1;
566 if (!inet_parse_passive(target, default_port, &ss)) {
567 return -EAFNOSUPPORT;
569 kernel_chooses_port = ss_get_port(&ss) == 0;
571 /* Create non-blocking socket, set SO_REUSEADDR. */
572 fd = socket(ss.ss_family, style, 0);
574 error = sock_errno();
575 VLOG_ERR("%s: socket: %s", target, sock_strerror(error));
578 error = set_nonblocking(fd);
582 if (style == SOCK_STREAM
583 && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) {
584 error = sock_errno();
585 VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s",
586 target, sock_strerror(error));
591 if (bind(fd, (struct sockaddr *) &ss, ss_length(&ss)) < 0) {
592 error = sock_errno();
593 VLOG_ERR("%s: bind: %s", target, sock_strerror(error));
597 /* The dscp bits must be configured before connect() to ensure that the TOS
598 * field is set during the connection establishment. If set after
599 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
600 error = set_dscp(fd, dscp);
602 VLOG_ERR("%s: set_dscp: %s", target, sock_strerror(error));
607 if (style == SOCK_STREAM && listen(fd, 10) < 0) {
608 error = sock_errno();
609 VLOG_ERR("%s: listen: %s", target, sock_strerror(error));
613 if (ssp || kernel_chooses_port) {
614 socklen_t ss_len = sizeof ss;
615 if (getsockname(fd, (struct sockaddr *) &ss, &ss_len) < 0) {
616 error = sock_errno();
617 VLOG_ERR("%s: getsockname: %s", target, sock_strerror(error));
620 if (kernel_chooses_port && kernel_print_port) {
621 VLOG_INFO("%s: listening on port %"PRIu16,
622 target, ss_get_port(&ss));
633 memset(ssp, 0, sizeof *ssp);
640 read_fully(int fd, void *p_, size_t size, size_t *bytes_read)
646 ssize_t retval = read(fd, p, size);
648 *bytes_read += retval;
651 } else if (retval == 0) {
653 } else if (errno != EINTR) {
661 write_fully(int fd, const void *p_, size_t size, size_t *bytes_written)
663 const uint8_t *p = p_;
667 ssize_t retval = write(fd, p, size);
669 *bytes_written += retval;
672 } else if (retval == 0) {
673 VLOG_WARN("write returned 0");
675 } else if (errno != EINTR) {
682 /* Given file name 'file_name', fsyncs the directory in which it is contained.
683 * Returns 0 if successful, otherwise a positive errno value. */
685 fsync_parent_dir(const char *file_name)
692 dir = dir_name(file_name);
693 fd = open(dir, O_RDONLY);
696 if (errno == EINVAL || errno == EROFS) {
697 /* This directory does not support synchronization. Not
698 * really an error. */
701 VLOG_ERR("%s: fsync failed (%s)", dir, ovs_strerror(error));
707 VLOG_ERR("%s: open failed (%s)", dir, ovs_strerror(error));
715 /* Obtains the modification time of the file named 'file_name' to the greatest
716 * supported precision. If successful, stores the mtime in '*mtime' and
717 * returns 0. On error, returns a positive errno value and stores zeros in
720 get_mtime(const char *file_name, struct timespec *mtime)
724 if (!stat(file_name, &s)) {
725 mtime->tv_sec = s.st_mtime;
727 #if HAVE_STRUCT_STAT_ST_MTIM_TV_NSEC
728 mtime->tv_nsec = s.st_mtim.tv_nsec;
729 #elif HAVE_STRUCT_STAT_ST_MTIMENSEC
730 mtime->tv_nsec = s.st_mtimensec;
737 mtime->tv_sec = mtime->tv_nsec = 0;
743 getsockopt_int(int fd, int level, int option, const char *optname, int *valuep)
745 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
751 if (getsockopt(fd, level, option, &value, &len)) {
752 error = sock_errno();
753 VLOG_ERR_RL(&rl, "getsockopt(%s): %s", optname, sock_strerror(error));
754 } else if (len != sizeof value) {
756 VLOG_ERR_RL(&rl, "getsockopt(%s): value is %u bytes (expected %"PRIuSIZE")",
757 optname, (unsigned int) len, sizeof value);
762 *valuep = error ? 0 : value;
767 describe_sockaddr(struct ds *string, int fd,
768 int (*getaddr)(int, struct sockaddr *, socklen_t *))
770 struct sockaddr_storage ss;
771 socklen_t len = sizeof ss;
773 if (!getaddr(fd, (struct sockaddr *) &ss, &len)) {
774 if (ss.ss_family == AF_INET || ss.ss_family == AF_INET6) {
775 char addrbuf[SS_NTOP_BUFSIZE];
777 ds_put_format(string, "%s:%"PRIu16,
778 ss_format_address(&ss, addrbuf, sizeof addrbuf),
781 } else if (ss.ss_family == AF_UNIX) {
782 struct sockaddr_un sun;
786 memcpy(&sun, &ss, sizeof sun);
787 maxlen = len - offsetof(struct sockaddr_un, sun_path);
788 null = memchr(sun.sun_path, '\0', maxlen);
789 ds_put_buffer(string, sun.sun_path,
790 null ? null - sun.sun_path : maxlen);
794 else if (ss.ss_family == AF_NETLINK) {
797 /* SO_PROTOCOL was introduced in 2.6.32. Support it regardless of the version
798 * of the Linux kernel headers in use at build time. */
800 #define SO_PROTOCOL 38
803 if (!getsockopt_int(fd, SOL_SOCKET, SO_PROTOCOL, "SO_PROTOCOL",
807 ds_put_cstr(string, "NETLINK_ROUTE");
810 case NETLINK_GENERIC:
811 ds_put_cstr(string, "NETLINK_GENERIC");
815 ds_put_format(string, "AF_NETLINK family %d", protocol);
819 ds_put_cstr(string, "AF_NETLINK");
824 else if (ss.ss_family == AF_PACKET) {
825 struct sockaddr_ll sll;
827 memcpy(&sll, &ss, sizeof sll);
828 ds_put_cstr(string, "AF_PACKET");
829 if (sll.sll_ifindex) {
832 if (if_indextoname(sll.sll_ifindex, name)) {
833 ds_put_format(string, "(%s)", name);
835 ds_put_format(string, "(ifindex=%d)", sll.sll_ifindex);
838 if (sll.sll_protocol) {
839 ds_put_format(string, "(protocol=0x%"PRIu16")",
840 ntohs(sll.sll_protocol));
844 else if (ss.ss_family == AF_UNSPEC) {
845 ds_put_cstr(string, "AF_UNSPEC");
847 ds_put_format(string, "AF_%d", (int) ss.ss_family);
855 put_fd_filename(struct ds *string, int fd)
861 linkname = xasprintf("/proc/self/fd/%d", fd);
862 n = readlink(linkname, buf, sizeof buf);
864 ds_put_char(string, ' ');
865 ds_put_buffer(string, buf, n);
866 if (n > sizeof buf) {
867 ds_put_cstr(string, "...");
874 /* Returns a malloc()'d string describing 'fd', for use in logging. */
884 ds_put_format(&string, "fstat failed (%s)", ovs_strerror(errno));
885 } else if (S_ISSOCK(s.st_mode)) {
886 describe_sockaddr(&string, fd, getsockname);
887 ds_put_cstr(&string, "<->");
888 describe_sockaddr(&string, fd, getpeername);
890 ds_put_cstr(&string, (isatty(fd) ? "tty"
891 : S_ISDIR(s.st_mode) ? "directory"
892 : S_ISCHR(s.st_mode) ? "character device"
893 : S_ISBLK(s.st_mode) ? "block device"
894 : S_ISREG(s.st_mode) ? "file"
895 : S_ISFIFO(s.st_mode) ? "FIFO"
896 : S_ISLNK(s.st_mode) ? "symbolic link"
899 put_fd_filename(&string, fd);
903 ds_put_format(&string,"file descriptor");
905 return ds_steal_cstr(&string);
909 /* sockaddr_storage helpers. */
911 /* Returns the IPv4 or IPv6 port in 'ss'. */
913 ss_get_port(const struct sockaddr_storage *ss)
915 if (ss->ss_family == AF_INET) {
916 const struct sockaddr_in *sin
917 = ALIGNED_CAST(const struct sockaddr_in *, ss);
918 return ntohs(sin->sin_port);
919 } else if (ss->ss_family == AF_INET6) {
920 const struct sockaddr_in6 *sin6
921 = ALIGNED_CAST(const struct sockaddr_in6 *, ss);
922 return ntohs(sin6->sin6_port);
928 /* Formats the IPv4 or IPv6 address in 'ss' into the 'bufsize' bytes in 'buf'.
929 * If 'ss' is an IPv6 address, puts square brackets around the address.
930 * 'bufsize' should be at least SS_NTOP_BUFSIZE.
934 ss_format_address(const struct sockaddr_storage *ss,
935 char *buf, size_t bufsize)
937 ovs_assert(bufsize >= SS_NTOP_BUFSIZE);
938 if (ss->ss_family == AF_INET) {
939 const struct sockaddr_in *sin
940 = ALIGNED_CAST(const struct sockaddr_in *, ss);
942 snprintf(buf, bufsize, IP_FMT, IP_ARGS(sin->sin_addr.s_addr));
943 } else if (ss->ss_family == AF_INET6) {
944 const struct sockaddr_in6 *sin6
945 = ALIGNED_CAST(const struct sockaddr_in6 *, ss);
948 inet_ntop(AF_INET6, sin6->sin6_addr.s6_addr, buf + 1, bufsize - 1);
949 strcpy(strchr(buf, '\0'), "]");
958 ss_length(const struct sockaddr_storage *ss)
960 switch (ss->ss_family) {
962 return sizeof(struct sockaddr_in);
965 return sizeof(struct sockaddr_in6);
972 /* For Windows socket calls, 'errno' is not set. One has to call
973 * WSAGetLastError() to get the error number and then pass it to
974 * this function to get the correct error string.
976 * ovs_strerror() calls strerror_r() and would not get the correct error
977 * string for Windows sockets, but is good for POSIX. */
979 sock_strerror(int error)
982 return ovs_format_message(error);
984 return ovs_strerror(error);