2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013 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/ioctl.h>
30 #include <sys/resource.h>
31 #include <sys/socket.h>
36 #include "dynamic-string.h"
37 #include "fatal-signal.h"
39 #include "poll-loop.h"
42 #if AF_PACKET && LINUX_DATAPATH
43 #include <linux/if_packet.h>
46 #include "netlink-protocol.h"
47 #include "netlink-socket.h"
50 VLOG_DEFINE_THIS_MODULE(socket_util);
52 /* #ifdefs make it a pain to maintain code: you have to try to build both ways.
53 * Thus, this file compiles all of the code regardless of the target, by
54 * writing "if (LINUX_DATAPATH)" instead of "#ifdef __linux__". */
55 #ifndef LINUX_DATAPATH
56 #define LINUX_DATAPATH 0
63 static int getsockopt_int(int fd, int level, int option, const char *optname,
66 /* Sets 'fd' to non-blocking mode. Returns 0 if successful, otherwise a
67 * positive errno value. */
69 set_nonblocking(int fd)
71 int flags = fcntl(fd, F_GETFL, 0);
73 if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) != -1) {
76 VLOG_ERR("fcntl(F_SETFL) failed: %s", ovs_strerror(errno));
80 VLOG_ERR("fcntl(F_GETFL) failed: %s", ovs_strerror(errno));
86 xset_nonblocking(int fd)
88 if (set_nonblocking(fd)) {
94 set_dscp(int fd, uint8_t dscp)
103 if (setsockopt(fd, IPPROTO_IP, IP_TOS, &val, sizeof val)) {
111 rlim_is_finite(rlim_t limit)
113 if (limit == RLIM_INFINITY) {
117 #ifdef RLIM_SAVED_CUR /* FreeBSD 8.0 lacks RLIM_SAVED_CUR. */
118 if (limit == RLIM_SAVED_CUR) {
123 #ifdef RLIM_SAVED_MAX /* FreeBSD 8.0 lacks RLIM_SAVED_MAX. */
124 if (limit == RLIM_SAVED_MAX) {
132 /* Returns the maximum valid FD value, plus 1. */
136 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
139 if (ovsthread_once_start(&once)) {
141 if (!getrlimit(RLIMIT_NOFILE, &r) && rlim_is_finite(r.rlim_cur)) {
142 max_fds = r.rlim_cur;
144 VLOG_WARN("failed to obtain fd limit, defaulting to 1024");
147 ovsthread_once_done(&once);
153 /* Translates 'host_name', which must be a string representation of an IP
154 * address, into a numeric IP address in '*addr'. Returns 0 if successful,
155 * otherwise a positive errno value. */
157 lookup_ip(const char *host_name, struct in_addr *addr)
159 if (!inet_aton(host_name, addr)) {
160 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
161 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name);
167 /* Translates 'host_name', which must be a string representation of an IPv6
168 * address, into a numeric IPv6 address in '*addr'. Returns 0 if successful,
169 * otherwise a positive errno value. */
171 lookup_ipv6(const char *host_name, struct in6_addr *addr)
173 if (inet_pton(AF_INET6, host_name, addr) != 1) {
174 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
175 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IPv6 address", host_name);
181 /* Translates 'host_name', which must be a host name or a string representation
182 * of an IP address, into a numeric IP address in '*addr'. Returns 0 if
183 * successful, otherwise a positive errno value.
185 * Most Open vSwitch code should not use this because it causes deadlocks:
186 * getaddrinfo() sends out a DNS request but that starts a new flow for which
187 * OVS must set up a flow, but it can't because it's waiting for a DNS reply.
188 * The synchronous lookup also delays other activity. (Of course we can solve
189 * this but it doesn't seem worthwhile quite yet.) */
191 lookup_hostname(const char *host_name, struct in_addr *addr)
193 struct addrinfo *result;
194 struct addrinfo hints;
196 if (inet_aton(host_name, addr)) {
200 memset(&hints, 0, sizeof hints);
201 hints.ai_family = AF_INET;
203 switch (getaddrinfo(host_name, NULL, &hints, &result)) {
205 *addr = ALIGNED_CAST(struct sockaddr_in *,
206 result->ai_addr)->sin_addr;
207 freeaddrinfo(result);
210 #ifdef EAI_ADDRFAMILY
245 check_connection_completion(int fd)
247 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
252 pfd.events = POLLOUT;
254 retval = poll(&pfd, 1, 0);
255 } while (retval < 0 && errno == EINTR);
257 if (pfd.revents & POLLERR) {
258 ssize_t n = send(fd, "", 1, MSG_DONTWAIT);
262 VLOG_ERR_RL(&rl, "poll return POLLERR but send succeeded");
267 } else if (retval < 0) {
268 VLOG_ERR_RL(&rl, "poll: %s", ovs_strerror(errno));
275 /* Drain all the data currently in the receive queue of a datagram socket (and
276 * possibly additional data). There is no way to know how many packets are in
277 * the receive queue, but we do know that the total number of bytes queued does
278 * not exceed the receive buffer size, so we pull packets until none are left
279 * or we've read that many bytes. */
285 rcvbuf = get_socket_rcvbuf(fd);
291 /* In Linux, specifying MSG_TRUNC in the flags argument causes the
292 * datagram length to be returned, even if that is longer than the
293 * buffer provided. Thus, we can use a 1-byte buffer to discard the
294 * incoming datagram and still be able to account how many bytes were
295 * removed from the receive buffer.
297 * On other Unix-like OSes, MSG_TRUNC has no effect in the flags
299 char buffer[LINUX_DATAPATH ? 1 : 2048];
300 ssize_t n_bytes = recv(fd, buffer, sizeof buffer,
301 MSG_TRUNC | MSG_DONTWAIT);
302 if (n_bytes <= 0 || n_bytes >= rcvbuf) {
310 /* Returns the size of socket 'sock''s receive buffer (SO_RCVBUF), or a
311 * negative errno value if an error occurs. */
313 get_socket_rcvbuf(int sock)
318 error = getsockopt_int(sock, SOL_SOCKET, SO_RCVBUF, "SO_RCVBUF", &rcvbuf);
319 return error ? -error : rcvbuf;
322 /* Reads and discards up to 'n' datagrams from 'fd', stopping as soon as no
323 * more data can be immediately read. ('fd' should therefore be in
324 * non-blocking mode.)*/
326 drain_fd(int fd, size_t n_packets)
328 for (; n_packets > 0; n_packets--) {
329 /* 'buffer' only needs to be 1 byte long in most circumstances. This
330 * size is defensive against the possibility that we someday want to
331 * use a Linux tap device without TUN_NO_PI, in which case a buffer
332 * smaller than sizeof(struct tun_pi) will give EINVAL on read. */
334 if (read(fd, buffer, sizeof buffer) <= 0) {
340 /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in
341 * '*un_len' the size of the sockaddr_un. */
343 make_sockaddr_un__(const char *name, struct sockaddr_un *un, socklen_t *un_len)
345 un->sun_family = AF_UNIX;
346 ovs_strzcpy(un->sun_path, name, sizeof un->sun_path);
347 *un_len = (offsetof(struct sockaddr_un, sun_path)
348 + strlen (un->sun_path) + 1);
351 /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in
352 * '*un_len' the size of the sockaddr_un.
354 * Returns 0 on success, otherwise a positive errno value. On success,
355 * '*dirfdp' is either -1 or a nonnegative file descriptor that the caller
356 * should close after using '*un' to bind or connect. On failure, '*dirfdp' is
359 make_sockaddr_un(const char *name, struct sockaddr_un *un, socklen_t *un_len,
362 enum { MAX_UN_LEN = sizeof un->sun_path - 1 };
365 if (strlen(name) > MAX_UN_LEN) {
366 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
368 if (LINUX_DATAPATH) {
369 /* 'name' is too long to fit in a sockaddr_un, but we have a
370 * workaround for that on Linux: shorten it by opening a file
371 * descriptor for the directory part of the name and indirecting
372 * through /proc/self/fd/<dirfd>/<basename>. */
377 dir = dir_name(name);
378 base = base_name(name);
380 dirfd = open(dir, O_DIRECTORY | O_RDONLY);
387 short_name = xasprintf("/proc/self/fd/%d/%s", dirfd, base);
391 if (strlen(short_name) <= MAX_UN_LEN) {
392 make_sockaddr_un__(short_name, un, un_len);
400 VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum "
401 "%d bytes (even shortened)", name, MAX_UN_LEN);
403 /* 'name' is too long and we have no workaround. */
404 VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum "
405 "%d bytes", name, MAX_UN_LEN);
410 make_sockaddr_un__(name, un, un_len);
415 /* Binds Unix domain socket 'fd' to a file with permissions 0700. */
417 bind_unix_socket(int fd, struct sockaddr *sun, socklen_t sun_len)
419 /* According to _Unix Network Programming_, umask should affect bind(). */
420 mode_t old_umask = umask(0077);
421 int error = bind(fd, sun, sun_len) ? errno : 0;
426 /* Creates a Unix domain socket in the given 'style' (either SOCK_DGRAM or
427 * SOCK_STREAM) that is bound to '*bind_path' (if 'bind_path' is non-null) and
428 * connected to '*connect_path' (if 'connect_path' is non-null). If 'nonblock'
429 * is true, the socket is made non-blocking.
431 * Returns the socket's fd if successful, otherwise a negative errno value. */
433 make_unix_socket(int style, bool nonblock,
434 const char *bind_path, const char *connect_path)
439 fd = socket(PF_UNIX, style, 0);
444 /* Set nonblocking mode right away, if we want it. This prevents blocking
445 * in connect(), if connect_path != NULL. (In turn, that's a corner case:
446 * it will only happen if style is SOCK_STREAM or SOCK_SEQPACKET, and only
447 * if a backlog of un-accepted connections has built up in the kernel.) */
449 error = set_nonblocking(fd);
456 struct sockaddr_un un;
460 if (unlink(bind_path) && errno != ENOENT) {
461 VLOG_WARN("unlinking \"%s\": %s\n",
462 bind_path, ovs_strerror(errno));
464 fatal_signal_add_file_to_unlink(bind_path);
466 error = make_sockaddr_un(bind_path, &un, &un_len, &dirfd);
468 error = bind_unix_socket(fd, (struct sockaddr *) &un, un_len);
479 struct sockaddr_un un;
483 error = make_sockaddr_un(connect_path, &un, &un_len, &dirfd);
485 && connect(fd, (struct sockaddr*) &un, un_len)
486 && errno != EINPROGRESS) {
500 if (error == EAGAIN) {
504 fatal_signal_unlink_file_now(bind_path);
511 get_unix_name_len(socklen_t sun_len)
513 return (sun_len >= offsetof(struct sockaddr_un, sun_path)
514 ? sun_len - offsetof(struct sockaddr_un, sun_path)
519 guess_netmask(ovs_be32 ip_)
521 uint32_t ip = ntohl(ip_);
522 return ((ip >> 31) == 0 ? htonl(0xff000000) /* Class A */
523 : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */
524 : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */
525 : htonl(0)); /* ??? */
528 /* Parses 'target', which should be a string in the format "<host>[:<port>]".
529 * <host> is required. If 'default_port' is nonzero then <port> is optional
530 * and defaults to 'default_port'.
532 * On success, returns true and stores the parsed remote address into '*sinp'.
533 * On failure, logs an error, stores zeros into '*sinp', and returns false. */
535 inet_parse_active(const char *target_, uint16_t default_port,
536 struct sockaddr_in *sinp)
538 char *target = xstrdup(target_);
539 char *save_ptr = NULL;
540 const char *host_name;
541 const char *port_string;
545 sinp->sin_family = AF_INET;
546 sinp->sin_port = htons(default_port);
549 host_name = strtok_r(target, ":", &save_ptr);
550 port_string = strtok_r(NULL, ":", &save_ptr);
552 VLOG_ERR("%s: bad peer name format", target_);
556 /* Look up IP, port. */
557 if (lookup_ip(host_name, &sinp->sin_addr)) {
560 if (port_string && atoi(port_string)) {
561 sinp->sin_port = htons(atoi(port_string));
562 } else if (!default_port) {
563 VLOG_ERR("%s: port number must be specified", target_);
571 memset(sinp, 0, sizeof *sinp);
577 /* Opens a non-blocking IPv4 socket of the specified 'style' and connects to
578 * 'target', which should be a string in the format "<host>[:<port>]". <host>
579 * is required. If 'default_port' is nonzero then <port> is optional and
580 * defaults to 'default_port'.
582 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
584 * On success, returns 0 (indicating connection complete) or EAGAIN (indicating
585 * connection in progress), in which case the new file descriptor is stored
586 * into '*fdp'. On failure, returns a positive errno value other than EAGAIN
587 * and stores -1 into '*fdp'.
589 * If 'sinp' is non-null, then on success the target address is stored into
592 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
593 * should be in the range [0, 63] and will automatically be shifted to the
594 * appropriately place in the IP tos field. */
596 inet_open_active(int style, const char *target, uint16_t default_port,
597 struct sockaddr_in *sinp, int *fdp, uint8_t dscp)
599 struct sockaddr_in sin;
604 if (!inet_parse_active(target, default_port, &sin)) {
605 error = EAFNOSUPPORT;
609 /* Create non-blocking socket. */
610 fd = socket(AF_INET, style, 0);
612 VLOG_ERR("%s: socket: %s", target, ovs_strerror(errno));
616 error = set_nonblocking(fd);
621 /* The dscp bits must be configured before connect() to ensure that the TOS
622 * field is set during the connection establishment. If set after
623 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
624 error = set_dscp(fd, dscp);
626 VLOG_ERR("%s: socket: %s", target, ovs_strerror(error));
631 error = connect(fd, (struct sockaddr *) &sin, sizeof sin) == 0 ? 0 : errno;
632 if (error == EINPROGRESS) {
637 if (!error || error == EAGAIN) {
641 } else if (fd >= 0) {
649 /* Parses 'target', which should be a string in the format "[<port>][:<ip>]":
651 * - If 'default_port' is -1, then <port> is required. Otherwise, if
652 * <port> is omitted, then 'default_port' is used instead.
654 * - If <port> (or 'default_port', if used) is 0, then no port is bound
655 * and the TCP/IP stack will select a port.
657 * - If <ip> is omitted then the IP address is wildcarded.
659 * If successful, stores the address into '*sinp' and returns true; otherwise
660 * zeros '*sinp' and returns false. */
662 inet_parse_passive(const char *target_, int default_port,
663 struct sockaddr_in *sinp)
665 char *target = xstrdup(target_);
666 char *string_ptr = target;
667 const char *host_name;
668 const char *port_string;
672 /* Address defaults. */
673 memset(sinp, 0, sizeof *sinp);
674 sinp->sin_family = AF_INET;
675 sinp->sin_addr.s_addr = htonl(INADDR_ANY);
676 sinp->sin_port = htons(default_port);
678 /* Parse optional port number. */
679 port_string = strsep(&string_ptr, ":");
680 if (port_string && str_to_int(port_string, 10, &port)) {
681 sinp->sin_port = htons(port);
682 } else if (default_port < 0) {
683 VLOG_ERR("%s: port number must be specified", target_);
687 /* Parse optional bind IP. */
688 host_name = strsep(&string_ptr, ":");
689 if (host_name && host_name[0] && lookup_ip(host_name, &sinp->sin_addr)) {
697 memset(sinp, 0, sizeof *sinp);
704 /* Opens a non-blocking IPv4 socket of the specified 'style', binds to
705 * 'target', and listens for incoming connections. Parses 'target' in the same
706 * way was inet_parse_passive().
708 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
710 * For TCP, the socket will have SO_REUSEADDR turned on.
712 * On success, returns a non-negative file descriptor. On failure, returns a
713 * negative errno value.
715 * If 'sinp' is non-null, then on success the bound address is stored into
718 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
719 * should be in the range [0, 63] and will automatically be shifted to the
720 * appropriately place in the IP tos field. */
722 inet_open_passive(int style, const char *target, int default_port,
723 struct sockaddr_in *sinp, uint8_t dscp)
725 bool kernel_chooses_port;
726 struct sockaddr_in sin;
728 unsigned int yes = 1;
730 if (!inet_parse_passive(target, default_port, &sin)) {
731 return -EAFNOSUPPORT;
734 /* Create non-blocking socket, set SO_REUSEADDR. */
735 fd = socket(AF_INET, style, 0);
738 VLOG_ERR("%s: socket: %s", target, ovs_strerror(error));
741 error = set_nonblocking(fd);
745 if (style == SOCK_STREAM
746 && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) {
748 VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s",
749 target, ovs_strerror(error));
754 if (bind(fd, (struct sockaddr *) &sin, sizeof sin) < 0) {
756 VLOG_ERR("%s: bind: %s", target, ovs_strerror(error));
760 /* The dscp bits must be configured before connect() to ensure that the TOS
761 * field is set during the connection establishment. If set after
762 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
763 error = set_dscp(fd, dscp);
765 VLOG_ERR("%s: socket: %s", target, ovs_strerror(error));
770 if (style == SOCK_STREAM && listen(fd, 10) < 0) {
772 VLOG_ERR("%s: listen: %s", target, ovs_strerror(error));
776 kernel_chooses_port = sin.sin_port == htons(0);
777 if (sinp || kernel_chooses_port) {
778 socklen_t sin_len = sizeof sin;
779 if (getsockname(fd, (struct sockaddr *) &sin, &sin_len) < 0) {
781 VLOG_ERR("%s: getsockname: %s", target, ovs_strerror(error));
784 if (sin.sin_family != AF_INET || sin_len != sizeof sin) {
785 error = EAFNOSUPPORT;
786 VLOG_ERR("%s: getsockname: invalid socket name", target);
792 if (kernel_chooses_port) {
793 VLOG_INFO("%s: listening on port %"PRIu16,
794 target, ntohs(sin.sin_port));
805 /* Returns a readable and writable fd for /dev/null, if successful, otherwise
806 * a negative errno value. The caller must not close the returned fd (because
807 * the same fd will be handed out to subsequent callers). */
811 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
814 if (ovsthread_once_start(&once)) {
815 null_fd = open("/dev/null", O_RDWR);
818 VLOG_ERR("could not open /dev/null: %s", ovs_strerror(error));
821 ovsthread_once_done(&once);
828 read_fully(int fd, void *p_, size_t size, size_t *bytes_read)
834 ssize_t retval = read(fd, p, size);
836 *bytes_read += retval;
839 } else if (retval == 0) {
841 } else if (errno != EINTR) {
849 write_fully(int fd, const void *p_, size_t size, size_t *bytes_written)
851 const uint8_t *p = p_;
855 ssize_t retval = write(fd, p, size);
857 *bytes_written += retval;
860 } else if (retval == 0) {
861 VLOG_WARN("write returned 0");
863 } else if (errno != EINTR) {
870 /* Given file name 'file_name', fsyncs the directory in which it is contained.
871 * Returns 0 if successful, otherwise a positive errno value. */
873 fsync_parent_dir(const char *file_name)
879 dir = dir_name(file_name);
880 fd = open(dir, O_RDONLY);
883 if (errno == EINVAL || errno == EROFS) {
884 /* This directory does not support synchronization. Not
885 * really an error. */
888 VLOG_ERR("%s: fsync failed (%s)", dir, ovs_strerror(error));
894 VLOG_ERR("%s: open failed (%s)", dir, ovs_strerror(error));
901 /* Obtains the modification time of the file named 'file_name' to the greatest
902 * supported precision. If successful, stores the mtime in '*mtime' and
903 * returns 0. On error, returns a positive errno value and stores zeros in
906 get_mtime(const char *file_name, struct timespec *mtime)
910 if (!stat(file_name, &s)) {
911 mtime->tv_sec = s.st_mtime;
913 #if HAVE_STRUCT_STAT_ST_MTIM_TV_NSEC
914 mtime->tv_nsec = s.st_mtim.tv_nsec;
915 #elif HAVE_STRUCT_STAT_ST_MTIMENSEC
916 mtime->tv_nsec = s.st_mtimensec;
923 mtime->tv_sec = mtime->tv_nsec = 0;
932 VLOG_FATAL("failed to create pipe (%s)", ovs_strerror(errno));
937 xpipe_nonblocking(int fds[2])
940 xset_nonblocking(fds[0]);
941 xset_nonblocking(fds[1]);
945 xsocketpair(int domain, int type, int protocol, int fds[2])
947 if (socketpair(domain, type, protocol, fds)) {
948 VLOG_FATAL("failed to create socketpair (%s)", ovs_strerror(errno));
953 getsockopt_int(int fd, int level, int option, const char *optname, int *valuep)
955 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
961 if (getsockopt(fd, level, option, &value, &len)) {
963 VLOG_ERR_RL(&rl, "getsockopt(%s): %s", optname, ovs_strerror(error));
964 } else if (len != sizeof value) {
966 VLOG_ERR_RL(&rl, "getsockopt(%s): value is %u bytes (expected %zu)",
967 optname, (unsigned int) len, sizeof value);
972 *valuep = error ? 0 : value;
977 describe_sockaddr(struct ds *string, int fd,
978 int (*getaddr)(int, struct sockaddr *, socklen_t *))
980 struct sockaddr_storage ss;
981 socklen_t len = sizeof ss;
983 if (!getaddr(fd, (struct sockaddr *) &ss, &len)) {
984 if (ss.ss_family == AF_INET) {
985 struct sockaddr_in sin;
987 memcpy(&sin, &ss, sizeof sin);
988 ds_put_format(string, IP_FMT":%"PRIu16,
989 IP_ARGS(sin.sin_addr.s_addr), ntohs(sin.sin_port));
990 } else if (ss.ss_family == AF_UNIX) {
991 struct sockaddr_un sun;
995 memcpy(&sun, &ss, sizeof sun);
996 maxlen = len - offsetof(struct sockaddr_un, sun_path);
997 null = memchr(sun.sun_path, '\0', maxlen);
998 ds_put_buffer(string, sun.sun_path,
999 null ? null - sun.sun_path : maxlen);
1002 else if (ss.ss_family == AF_NETLINK) {
1005 /* SO_PROTOCOL was introduced in 2.6.32. Support it regardless of the version
1006 * of the Linux kernel headers in use at build time. */
1008 #define SO_PROTOCOL 38
1011 if (!getsockopt_int(fd, SOL_SOCKET, SO_PROTOCOL, "SO_PROTOCOL",
1015 ds_put_cstr(string, "NETLINK_ROUTE");
1018 case NETLINK_GENERIC:
1019 ds_put_cstr(string, "NETLINK_GENERIC");
1023 ds_put_format(string, "AF_NETLINK family %d", protocol);
1027 ds_put_cstr(string, "AF_NETLINK");
1031 #if AF_PACKET && LINUX_DATAPATH
1032 else if (ss.ss_family == AF_PACKET) {
1033 struct sockaddr_ll sll;
1035 memcpy(&sll, &ss, sizeof sll);
1036 ds_put_cstr(string, "AF_PACKET");
1037 if (sll.sll_ifindex) {
1038 char name[IFNAMSIZ];
1040 if (if_indextoname(sll.sll_ifindex, name)) {
1041 ds_put_format(string, "(%s)", name);
1043 ds_put_format(string, "(ifindex=%d)", sll.sll_ifindex);
1046 if (sll.sll_protocol) {
1047 ds_put_format(string, "(protocol=0x%"PRIu16")",
1048 ntohs(sll.sll_protocol));
1052 else if (ss.ss_family == AF_UNSPEC) {
1053 ds_put_cstr(string, "AF_UNSPEC");
1055 ds_put_format(string, "AF_%d", (int) ss.ss_family);
1061 #ifdef LINUX_DATAPATH
1063 put_fd_filename(struct ds *string, int fd)
1069 linkname = xasprintf("/proc/self/fd/%d", fd);
1070 n = readlink(linkname, buf, sizeof buf);
1072 ds_put_char(string, ' ');
1073 ds_put_buffer(string, buf, n);
1074 if (n > sizeof buf) {
1075 ds_put_cstr(string, "...");
1082 /* Returns a malloc()'d string describing 'fd', for use in logging. */
1090 if (fstat(fd, &s)) {
1091 ds_put_format(&string, "fstat failed (%s)", ovs_strerror(errno));
1092 } else if (S_ISSOCK(s.st_mode)) {
1093 describe_sockaddr(&string, fd, getsockname);
1094 ds_put_cstr(&string, "<->");
1095 describe_sockaddr(&string, fd, getpeername);
1097 ds_put_cstr(&string, (isatty(fd) ? "tty"
1098 : S_ISDIR(s.st_mode) ? "directory"
1099 : S_ISCHR(s.st_mode) ? "character device"
1100 : S_ISBLK(s.st_mode) ? "block device"
1101 : S_ISREG(s.st_mode) ? "file"
1102 : S_ISFIFO(s.st_mode) ? "FIFO"
1103 : S_ISLNK(s.st_mode) ? "symbolic link"
1105 #ifdef LINUX_DATAPATH
1106 put_fd_filename(&string, fd);
1109 return ds_steal_cstr(&string);
1112 /* Returns the total of the 'iov_len' members of the 'n_iovs' in 'iovs'.
1113 * The caller must ensure that the total does not exceed SIZE_MAX. */
1115 iovec_len(const struct iovec iovs[], size_t n_iovs)
1120 for (i = 0; i < n_iovs; i++) {
1121 len += iovs[i].iov_len;
1126 /* Returns true if all of the 'n_iovs' iovecs in 'iovs' have length zero. */
1128 iovec_is_empty(const struct iovec iovs[], size_t n_iovs)
1132 for (i = 0; i < n_iovs; i++) {
1133 if (iovs[i].iov_len) {
1140 /* Sends the 'n_iovs' iovecs of data in 'iovs' and the 'n_fds' file descriptors
1141 * in 'fds' on Unix domain socket 'sock'. Returns the number of bytes
1142 * successfully sent or -1 if an error occurred. On error, sets errno
1145 send_iovec_and_fds(int sock,
1146 const struct iovec *iovs, size_t n_iovs,
1147 const int fds[], size_t n_fds)
1149 ovs_assert(sock >= 0);
1153 char control[CMSG_SPACE(SOUTIL_MAX_FDS * sizeof *fds)];
1157 ovs_assert(!iovec_is_empty(iovs, n_iovs));
1158 ovs_assert(n_fds <= SOUTIL_MAX_FDS);
1160 memset(&cmsg, 0, sizeof cmsg);
1161 cmsg.cm.cmsg_len = CMSG_LEN(n_fds * sizeof *fds);
1162 cmsg.cm.cmsg_level = SOL_SOCKET;
1163 cmsg.cm.cmsg_type = SCM_RIGHTS;
1164 memcpy(CMSG_DATA(&cmsg.cm), fds, n_fds * sizeof *fds);
1166 msg.msg_name = NULL;
1167 msg.msg_namelen = 0;
1168 msg.msg_iov = CONST_CAST(struct iovec *, iovs);
1169 msg.msg_iovlen = n_iovs;
1170 msg.msg_control = &cmsg.cm;
1171 msg.msg_controllen = CMSG_SPACE(n_fds * sizeof *fds);
1174 return sendmsg(sock, &msg, 0);
1176 return writev(sock, iovs, n_iovs);
1180 /* Sends the 'n_iovs' iovecs of data in 'iovs' and the 'n_fds' file descriptors
1181 * in 'fds' on Unix domain socket 'sock'. If 'skip_bytes' is nonzero, then the
1182 * first 'skip_bytes' of data in the iovecs are not sent, and none of the file
1183 * descriptors are sent. The function continues to retry sending until an
1184 * error (other than EINTR) occurs or all the data and fds are sent.
1186 * Returns 0 if all the data and fds were successfully sent, otherwise a
1187 * positive errno value. Regardless of success, stores the number of bytes
1188 * sent (always at least 'skip_bytes') in '*bytes_sent'. (If at least one byte
1189 * is sent, then all the fds have been sent.)
1191 * 'skip_bytes' must be less than or equal to iovec_len(iovs, n_iovs). */
1193 send_iovec_and_fds_fully(int sock,
1194 const struct iovec iovs[], size_t n_iovs,
1195 const int fds[], size_t n_fds,
1196 size_t skip_bytes, size_t *bytes_sent)
1199 while (n_iovs > 0) {
1203 retval = skip_bytes;
1205 } else if (!*bytes_sent) {
1206 retval = send_iovec_and_fds(sock, iovs, n_iovs, fds, n_fds);
1208 retval = writev(sock, iovs, n_iovs);
1212 *bytes_sent += retval;
1213 while (retval > 0) {
1214 const uint8_t *base = iovs->iov_base;
1215 size_t len = iovs->iov_len;
1221 error = write_fully(sock, base + retval, len - retval,
1223 *bytes_sent += sent;
1233 } else if (retval == 0) {
1234 if (iovec_is_empty(iovs, n_iovs)) {
1237 VLOG_WARN("send returned 0");
1239 } else if (errno != EINTR) {
1247 /* Sends the 'n_iovs' iovecs of data in 'iovs' and the 'n_fds' file descriptors
1248 * in 'fds' on Unix domain socket 'sock'. The function continues to retry
1249 * sending until an error (other than EAGAIN or EINTR) occurs or all the data
1250 * and fds are sent. Upon EAGAIN, the function blocks until the socket is
1251 * ready for more data.
1253 * Returns 0 if all the data and fds were successfully sent, otherwise a
1254 * positive errno value. */
1256 send_iovec_and_fds_fully_block(int sock,
1257 const struct iovec iovs[], size_t n_iovs,
1258 const int fds[], size_t n_fds)
1265 error = send_iovec_and_fds_fully(sock, iovs, n_iovs,
1266 fds, n_fds, sent, &sent);
1267 if (error != EAGAIN) {
1270 poll_fd_wait(sock, POLLOUT);
1275 /* Attempts to receive from Unix domain socket 'sock' up to 'size' bytes of
1276 * data into 'data' and up to SOUTIL_MAX_FDS file descriptors into 'fds'.
1278 * - Upon success, returns the number of bytes of data copied into 'data'
1279 * and stores the number of received file descriptors into '*n_fdsp'.
1281 * - On failure, returns a negative errno value and stores 0 in
1284 * - On EOF, returns 0 and stores 0 in '*n_fdsp'. */
1286 recv_data_and_fds(int sock,
1287 void *data, size_t size,
1288 int fds[SOUTIL_MAX_FDS], size_t *n_fdsp)
1292 char control[CMSG_SPACE(SOUTIL_MAX_FDS * sizeof *fds)];
1304 iov.iov_base = data;
1307 msg.msg_name = NULL;
1308 msg.msg_namelen = 0;
1311 msg.msg_control = &cmsg.cm;
1312 msg.msg_controllen = sizeof cmsg.control;
1315 retval = recvmsg(sock, &msg, 0);
1316 } while (retval < 0 && errno == EINTR);
1318 return retval < 0 ? -errno : 0;
1321 for (p = CMSG_FIRSTHDR(&msg); p; p = CMSG_NXTHDR(&msg, p)) {
1322 if (p->cmsg_level != SOL_SOCKET || p->cmsg_type != SCM_RIGHTS) {
1323 VLOG_ERR("unexpected control message %d:%d",
1324 p->cmsg_level, p->cmsg_type);
1326 } else if (*n_fdsp) {
1327 VLOG_ERR("multiple SCM_RIGHTS received");
1330 size_t n_fds = (p->cmsg_len - CMSG_LEN(0)) / sizeof *fds;
1331 const int *fds_data = ALIGNED_CAST(const int *, CMSG_DATA(p));
1333 ovs_assert(n_fds > 0);
1334 if (n_fds > SOUTIL_MAX_FDS) {
1335 VLOG_ERR("%zu fds received but only %d supported",
1336 n_fds, SOUTIL_MAX_FDS);
1337 for (i = 0; i < n_fds; i++) {
1344 memcpy(fds, fds_data, n_fds * sizeof *fds);
1351 for (i = 0; i < *n_fdsp; i++) {
1358 /* Calls ioctl() on an AF_INET sock, passing the specified 'command' and
1359 * 'arg'. Returns 0 if successful, otherwise a positive errno value. */
1361 af_inet_ioctl(unsigned long int command, const void *arg)
1363 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
1366 if (ovsthread_once_start(&once)) {
1367 sock = socket(AF_INET, SOCK_DGRAM, 0);
1370 VLOG_ERR("failed to create inet socket: %s", ovs_strerror(errno));
1372 ovsthread_once_done(&once);
1375 return (sock < 0 ? -sock
1376 : ioctl(sock, command, arg) == -1 ? errno
1381 af_inet_ifreq_ioctl(const char *name, struct ifreq *ifr, unsigned long int cmd,
1382 const char *cmd_name)
1386 ovs_strzcpy(ifr->ifr_name, name, sizeof ifr->ifr_name);
1387 error = af_inet_ioctl(cmd, ifr);
1389 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1390 VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name,
1391 ovs_strerror(error));