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/resource.h>
30 #include <sys/socket.h>
35 #include "dynamic-string.h"
36 #include "fatal-signal.h"
38 #include "poll-loop.h"
41 #if AF_PACKET && LINUX_DATAPATH
42 #include <linux/if_packet.h>
45 #include "netlink-protocol.h"
46 #include "netlink-socket.h"
49 VLOG_DEFINE_THIS_MODULE(socket_util);
51 /* #ifdefs make it a pain to maintain code: you have to try to build both ways.
52 * Thus, this file compiles all of the code regardless of the target, by
53 * writing "if (LINUX_DATAPATH)" instead of "#ifdef __linux__". */
54 #ifndef LINUX_DATAPATH
55 #define LINUX_DATAPATH 0
62 static int getsockopt_int(int fd, int level, int option, const char *optname,
65 /* Sets 'fd' to non-blocking mode. Returns 0 if successful, otherwise a
66 * positive errno value. */
68 set_nonblocking(int fd)
70 int flags = fcntl(fd, F_GETFL, 0);
72 if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) != -1) {
75 VLOG_ERR("fcntl(F_SETFL) failed: %s", strerror(errno));
79 VLOG_ERR("fcntl(F_GETFL) failed: %s", strerror(errno));
85 xset_nonblocking(int fd)
87 if (set_nonblocking(fd)) {
93 set_dscp(int fd, uint8_t dscp)
102 if (setsockopt(fd, IPPROTO_IP, IP_TOS, &val, sizeof val)) {
110 rlim_is_finite(rlim_t limit)
112 if (limit == RLIM_INFINITY) {
116 #ifdef RLIM_SAVED_CUR /* FreeBSD 8.0 lacks RLIM_SAVED_CUR. */
117 if (limit == RLIM_SAVED_CUR) {
122 #ifdef RLIM_SAVED_MAX /* FreeBSD 8.0 lacks RLIM_SAVED_MAX. */
123 if (limit == RLIM_SAVED_MAX) {
131 /* Returns the maximum valid FD value, plus 1. */
135 static int max_fds = -1;
138 if (!getrlimit(RLIMIT_NOFILE, &r) && rlim_is_finite(r.rlim_cur)) {
139 max_fds = r.rlim_cur;
141 VLOG_WARN("failed to obtain fd limit, defaulting to 1024");
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 (!inet_aton(host_name, 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 (inet_pton(AF_INET6, host_name, addr) != 1) {
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 (inet_aton(host_name, addr)) {
195 memset(&hints, 0, sizeof hints);
196 hints.ai_family = AF_INET;
198 switch (getaddrinfo(host_name, NULL, &hints, &result)) {
200 *addr = ((struct sockaddr_in *) result->ai_addr)->sin_addr;
201 freeaddrinfo(result);
204 #ifdef EAI_ADDRFAMILY
239 check_connection_completion(int fd)
241 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
246 pfd.events = POLLOUT;
248 retval = poll(&pfd, 1, 0);
249 } while (retval < 0 && errno == EINTR);
251 if (pfd.revents & POLLERR) {
252 ssize_t n = send(fd, "", 1, MSG_DONTWAIT);
256 VLOG_ERR_RL(&rl, "poll return POLLERR but send succeeded");
261 } else if (retval < 0) {
262 VLOG_ERR_RL(&rl, "poll: %s", strerror(errno));
269 /* Drain all the data currently in the receive queue of a datagram socket (and
270 * possibly additional data). There is no way to know how many packets are in
271 * the receive queue, but we do know that the total number of bytes queued does
272 * not exceed the receive buffer size, so we pull packets until none are left
273 * or we've read that many bytes. */
279 rcvbuf = get_socket_rcvbuf(fd);
285 /* In Linux, specifying MSG_TRUNC in the flags argument causes the
286 * datagram length to be returned, even if that is longer than the
287 * buffer provided. Thus, we can use a 1-byte buffer to discard the
288 * incoming datagram and still be able to account how many bytes were
289 * removed from the receive buffer.
291 * On other Unix-like OSes, MSG_TRUNC has no effect in the flags
293 char buffer[LINUX_DATAPATH ? 1 : 2048];
294 ssize_t n_bytes = recv(fd, buffer, sizeof buffer,
295 MSG_TRUNC | MSG_DONTWAIT);
296 if (n_bytes <= 0 || n_bytes >= rcvbuf) {
304 /* Returns the size of socket 'sock''s receive buffer (SO_RCVBUF), or a
305 * negative errno value if an error occurs. */
307 get_socket_rcvbuf(int sock)
312 error = getsockopt_int(sock, SOL_SOCKET, SO_RCVBUF, "SO_RCVBUF", &rcvbuf);
313 return error ? -error : rcvbuf;
316 /* Reads and discards up to 'n' datagrams from 'fd', stopping as soon as no
317 * more data can be immediately read. ('fd' should therefore be in
318 * non-blocking mode.)*/
320 drain_fd(int fd, size_t n_packets)
322 for (; n_packets > 0; n_packets--) {
323 /* 'buffer' only needs to be 1 byte long in most circumstances. This
324 * size is defensive against the possibility that we someday want to
325 * use a Linux tap device without TUN_NO_PI, in which case a buffer
326 * smaller than sizeof(struct tun_pi) will give EINVAL on read. */
328 if (read(fd, buffer, sizeof buffer) <= 0) {
334 /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in
335 * '*un_len' the size of the sockaddr_un. */
337 make_sockaddr_un__(const char *name, struct sockaddr_un *un, socklen_t *un_len)
339 un->sun_family = AF_UNIX;
340 ovs_strzcpy(un->sun_path, name, sizeof un->sun_path);
341 *un_len = (offsetof(struct sockaddr_un, sun_path)
342 + strlen (un->sun_path) + 1);
345 /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in
346 * '*un_len' the size of the sockaddr_un.
348 * Returns 0 on success, otherwise a positive errno value. On success,
349 * '*dirfdp' is either -1 or a nonnegative file descriptor that the caller
350 * should close after using '*un' to bind or connect. On failure, '*dirfdp' is
353 make_sockaddr_un(const char *name, struct sockaddr_un *un, socklen_t *un_len,
356 enum { MAX_UN_LEN = sizeof un->sun_path - 1 };
359 if (strlen(name) > MAX_UN_LEN) {
360 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
362 if (LINUX_DATAPATH) {
363 /* 'name' is too long to fit in a sockaddr_un, but we have a
364 * workaround for that on Linux: shorten it by opening a file
365 * descriptor for the directory part of the name and indirecting
366 * through /proc/self/fd/<dirfd>/<basename>. */
371 dir = dir_name(name);
372 base = base_name(name);
374 dirfd = open(dir, O_DIRECTORY | O_RDONLY);
381 short_name = xasprintf("/proc/self/fd/%d/%s", dirfd, base);
385 if (strlen(short_name) <= MAX_UN_LEN) {
386 make_sockaddr_un__(short_name, un, un_len);
394 VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum "
395 "%d bytes (even shortened)", name, MAX_UN_LEN);
397 /* 'name' is too long and we have no workaround. */
398 VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum "
399 "%d bytes", name, MAX_UN_LEN);
404 make_sockaddr_un__(name, un, un_len);
409 /* Binds Unix domain socket 'fd' to a file with permissions 0700. */
411 bind_unix_socket(int fd, struct sockaddr *sun, socklen_t sun_len)
413 /* According to _Unix Network Programming_, umask should affect bind(). */
414 mode_t old_umask = umask(0077);
415 int error = bind(fd, sun, sun_len) ? errno : 0;
420 /* Creates a Unix domain socket in the given 'style' (either SOCK_DGRAM or
421 * SOCK_STREAM) that is bound to '*bind_path' (if 'bind_path' is non-null) and
422 * connected to '*connect_path' (if 'connect_path' is non-null). If 'nonblock'
423 * is true, the socket is made non-blocking.
425 * Returns the socket's fd if successful, otherwise a negative errno value. */
427 make_unix_socket(int style, bool nonblock,
428 const char *bind_path, const char *connect_path)
433 fd = socket(PF_UNIX, style, 0);
438 /* Set nonblocking mode right away, if we want it. This prevents blocking
439 * in connect(), if connect_path != NULL. (In turn, that's a corner case:
440 * it will only happen if style is SOCK_STREAM or SOCK_SEQPACKET, and only
441 * if a backlog of un-accepted connections has built up in the kernel.) */
443 error = set_nonblocking(fd);
450 struct sockaddr_un un;
454 if (unlink(bind_path) && errno != ENOENT) {
455 VLOG_WARN("unlinking \"%s\": %s\n", bind_path, strerror(errno));
457 fatal_signal_add_file_to_unlink(bind_path);
459 error = make_sockaddr_un(bind_path, &un, &un_len, &dirfd);
461 error = bind_unix_socket(fd, (struct sockaddr *) &un, un_len);
472 struct sockaddr_un un;
476 error = make_sockaddr_un(connect_path, &un, &un_len, &dirfd);
478 && connect(fd, (struct sockaddr*) &un, un_len)
479 && errno != EINPROGRESS) {
493 if (error == EAGAIN) {
497 fatal_signal_unlink_file_now(bind_path);
504 get_unix_name_len(socklen_t sun_len)
506 return (sun_len >= offsetof(struct sockaddr_un, sun_path)
507 ? sun_len - offsetof(struct sockaddr_un, sun_path)
512 guess_netmask(ovs_be32 ip_)
514 uint32_t ip = ntohl(ip_);
515 return ((ip >> 31) == 0 ? htonl(0xff000000) /* Class A */
516 : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */
517 : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */
518 : htonl(0)); /* ??? */
521 /* Parses 'target', which should be a string in the format "<host>[:<port>]".
522 * <host> is required. If 'default_port' is nonzero then <port> is optional
523 * and defaults to 'default_port'.
525 * On success, returns true and stores the parsed remote address into '*sinp'.
526 * On failure, logs an error, stores zeros into '*sinp', and returns false. */
528 inet_parse_active(const char *target_, uint16_t default_port,
529 struct sockaddr_in *sinp)
531 char *target = xstrdup(target_);
532 char *save_ptr = NULL;
533 const char *host_name;
534 const char *port_string;
538 sinp->sin_family = AF_INET;
539 sinp->sin_port = htons(default_port);
542 host_name = strtok_r(target, ":", &save_ptr);
543 port_string = strtok_r(NULL, ":", &save_ptr);
545 VLOG_ERR("%s: bad peer name format", target_);
549 /* Look up IP, port. */
550 if (lookup_ip(host_name, &sinp->sin_addr)) {
553 if (port_string && atoi(port_string)) {
554 sinp->sin_port = htons(atoi(port_string));
555 } else if (!default_port) {
556 VLOG_ERR("%s: port number must be specified", target_);
564 memset(sinp, 0, sizeof *sinp);
570 /* Opens a non-blocking IPv4 socket of the specified 'style' and connects to
571 * 'target', which should be a string in the format "<host>[:<port>]". <host>
572 * is required. If 'default_port' is nonzero then <port> is optional and
573 * defaults to 'default_port'.
575 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
577 * On success, returns 0 (indicating connection complete) or EAGAIN (indicating
578 * connection in progress), in which case the new file descriptor is stored
579 * into '*fdp'. On failure, returns a positive errno value other than EAGAIN
580 * and stores -1 into '*fdp'.
582 * If 'sinp' is non-null, then on success the target address is stored into
585 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
586 * should be in the range [0, 63] and will automatically be shifted to the
587 * appropriately place in the IP tos field. */
589 inet_open_active(int style, const char *target, uint16_t default_port,
590 struct sockaddr_in *sinp, int *fdp, uint8_t dscp)
592 struct sockaddr_in sin;
597 if (!inet_parse_active(target, default_port, &sin)) {
598 error = EAFNOSUPPORT;
602 /* Create non-blocking socket. */
603 fd = socket(AF_INET, style, 0);
605 VLOG_ERR("%s: socket: %s", target, strerror(errno));
609 error = set_nonblocking(fd);
614 /* The dscp bits must be configured before connect() to ensure that the TOS
615 * field is set during the connection establishment. If set after
616 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
617 error = set_dscp(fd, dscp);
619 VLOG_ERR("%s: socket: %s", target, strerror(error));
624 error = connect(fd, (struct sockaddr *) &sin, sizeof sin) == 0 ? 0 : errno;
625 if (error == EINPROGRESS) {
630 if (!error || error == EAGAIN) {
634 } else if (fd >= 0) {
642 /* Parses 'target', which should be a string in the format "[<port>][:<ip>]":
644 * - If 'default_port' is -1, then <port> is required. Otherwise, if
645 * <port> is omitted, then 'default_port' is used instead.
647 * - If <port> (or 'default_port', if used) is 0, then no port is bound
648 * and the TCP/IP stack will select a port.
650 * - If <ip> is omitted then the IP address is wildcarded.
652 * If successful, stores the address into '*sinp' and returns true; otherwise
653 * zeros '*sinp' and returns false. */
655 inet_parse_passive(const char *target_, int default_port,
656 struct sockaddr_in *sinp)
658 char *target = xstrdup(target_);
659 char *string_ptr = target;
660 const char *host_name;
661 const char *port_string;
665 /* Address defaults. */
666 memset(sinp, 0, sizeof *sinp);
667 sinp->sin_family = AF_INET;
668 sinp->sin_addr.s_addr = htonl(INADDR_ANY);
669 sinp->sin_port = htons(default_port);
671 /* Parse optional port number. */
672 port_string = strsep(&string_ptr, ":");
673 if (port_string && str_to_int(port_string, 10, &port)) {
674 sinp->sin_port = htons(port);
675 } else if (default_port < 0) {
676 VLOG_ERR("%s: port number must be specified", target_);
680 /* Parse optional bind IP. */
681 host_name = strsep(&string_ptr, ":");
682 if (host_name && host_name[0] && lookup_ip(host_name, &sinp->sin_addr)) {
690 memset(sinp, 0, sizeof *sinp);
697 /* Opens a non-blocking IPv4 socket of the specified 'style', binds to
698 * 'target', and listens for incoming connections. Parses 'target' in the same
699 * way was inet_parse_passive().
701 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
703 * For TCP, the socket will have SO_REUSEADDR turned on.
705 * On success, returns a non-negative file descriptor. On failure, returns a
706 * negative errno value.
708 * If 'sinp' is non-null, then on success the bound address is stored into
711 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
712 * should be in the range [0, 63] and will automatically be shifted to the
713 * appropriately place in the IP tos field. */
715 inet_open_passive(int style, const char *target, int default_port,
716 struct sockaddr_in *sinp, uint8_t dscp)
718 bool kernel_chooses_port;
719 struct sockaddr_in sin;
721 unsigned int yes = 1;
723 if (!inet_parse_passive(target, default_port, &sin)) {
724 return -EAFNOSUPPORT;
727 /* Create non-blocking socket, set SO_REUSEADDR. */
728 fd = socket(AF_INET, style, 0);
731 VLOG_ERR("%s: socket: %s", target, strerror(error));
734 error = set_nonblocking(fd);
738 if (style == SOCK_STREAM
739 && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) {
741 VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s", target, strerror(error));
746 if (bind(fd, (struct sockaddr *) &sin, sizeof sin) < 0) {
748 VLOG_ERR("%s: bind: %s", target, strerror(error));
752 /* The dscp bits must be configured before connect() to ensure that the TOS
753 * field is set during the connection establishment. If set after
754 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
755 error = set_dscp(fd, dscp);
757 VLOG_ERR("%s: socket: %s", target, strerror(error));
762 if (style == SOCK_STREAM && listen(fd, 10) < 0) {
764 VLOG_ERR("%s: listen: %s", target, strerror(error));
768 kernel_chooses_port = sin.sin_port == htons(0);
769 if (sinp || kernel_chooses_port) {
770 socklen_t sin_len = sizeof sin;
771 if (getsockname(fd, (struct sockaddr *) &sin, &sin_len) < 0) {
773 VLOG_ERR("%s: getsockname: %s", target, strerror(error));
776 if (sin.sin_family != AF_INET || sin_len != sizeof sin) {
777 error = EAFNOSUPPORT;
778 VLOG_ERR("%s: getsockname: invalid socket name", target);
784 if (kernel_chooses_port) {
785 VLOG_INFO("%s: listening on port %"PRIu16,
786 target, ntohs(sin.sin_port));
797 /* Returns a readable and writable fd for /dev/null, if successful, otherwise
798 * a negative errno value. The caller must not close the returned fd (because
799 * the same fd will be handed out to subsequent callers). */
803 static int null_fd = -1;
805 null_fd = open("/dev/null", O_RDWR);
808 VLOG_ERR("could not open /dev/null: %s", strerror(error));
816 read_fully(int fd, void *p_, size_t size, size_t *bytes_read)
822 ssize_t retval = read(fd, p, size);
824 *bytes_read += retval;
827 } else if (retval == 0) {
829 } else if (errno != EINTR) {
837 write_fully(int fd, const void *p_, size_t size, size_t *bytes_written)
839 const uint8_t *p = p_;
843 ssize_t retval = write(fd, p, size);
845 *bytes_written += retval;
848 } else if (retval == 0) {
849 VLOG_WARN("write returned 0");
851 } else if (errno != EINTR) {
858 /* Given file name 'file_name', fsyncs the directory in which it is contained.
859 * Returns 0 if successful, otherwise a positive errno value. */
861 fsync_parent_dir(const char *file_name)
867 dir = dir_name(file_name);
868 fd = open(dir, O_RDONLY);
871 if (errno == EINVAL || errno == EROFS) {
872 /* This directory does not support synchronization. Not
873 * really an error. */
876 VLOG_ERR("%s: fsync failed (%s)", dir, strerror(error));
882 VLOG_ERR("%s: open failed (%s)", dir, strerror(error));
889 /* Obtains the modification time of the file named 'file_name' to the greatest
890 * supported precision. If successful, stores the mtime in '*mtime' and
891 * returns 0. On error, returns a positive errno value and stores zeros in
894 get_mtime(const char *file_name, struct timespec *mtime)
898 if (!stat(file_name, &s)) {
899 mtime->tv_sec = s.st_mtime;
901 #if HAVE_STRUCT_STAT_ST_MTIM_TV_NSEC
902 mtime->tv_nsec = s.st_mtim.tv_nsec;
903 #elif HAVE_STRUCT_STAT_ST_MTIMENSEC
904 mtime->tv_nsec = s.st_mtimensec;
911 mtime->tv_sec = mtime->tv_nsec = 0;
920 VLOG_FATAL("failed to create pipe (%s)", strerror(errno));
925 xpipe_nonblocking(int fds[2])
928 xset_nonblocking(fds[0]);
929 xset_nonblocking(fds[1]);
933 xsocketpair(int domain, int type, int protocol, int fds[2])
935 if (socketpair(domain, type, protocol, fds)) {
936 VLOG_FATAL("failed to create socketpair (%s)", strerror(errno));
941 getsockopt_int(int fd, int level, int option, const char *optname, int *valuep)
943 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
949 if (getsockopt(fd, level, option, &value, &len)) {
951 VLOG_ERR_RL(&rl, "getsockopt(%s): %s", optname, strerror(error));
952 } else if (len != sizeof value) {
954 VLOG_ERR_RL(&rl, "getsockopt(%s): value is %u bytes (expected %zu)",
955 optname, (unsigned int) len, sizeof value);
960 *valuep = error ? 0 : value;
965 describe_sockaddr(struct ds *string, int fd,
966 int (*getaddr)(int, struct sockaddr *, socklen_t *))
968 struct sockaddr_storage ss;
969 socklen_t len = sizeof ss;
971 if (!getaddr(fd, (struct sockaddr *) &ss, &len)) {
972 if (ss.ss_family == AF_INET) {
973 struct sockaddr_in sin;
975 memcpy(&sin, &ss, sizeof sin);
976 ds_put_format(string, IP_FMT":%"PRIu16,
977 IP_ARGS(sin.sin_addr.s_addr), ntohs(sin.sin_port));
978 } else if (ss.ss_family == AF_UNIX) {
979 struct sockaddr_un sun;
983 memcpy(&sun, &ss, sizeof sun);
984 maxlen = len - offsetof(struct sockaddr_un, sun_path);
985 null = memchr(sun.sun_path, '\0', maxlen);
986 ds_put_buffer(string, sun.sun_path,
987 null ? null - sun.sun_path : maxlen);
990 else if (ss.ss_family == AF_NETLINK) {
993 /* SO_PROTOCOL was introduced in 2.6.32. Support it regardless of the version
994 * of the Linux kernel headers in use at build time. */
996 #define SO_PROTOCOL 38
999 if (!getsockopt_int(fd, SOL_SOCKET, SO_PROTOCOL, "SO_PROTOCOL",
1003 ds_put_cstr(string, "NETLINK_ROUTE");
1006 case NETLINK_GENERIC:
1007 ds_put_cstr(string, "NETLINK_GENERIC");
1011 ds_put_format(string, "AF_NETLINK family %d", protocol);
1015 ds_put_cstr(string, "AF_NETLINK");
1019 #if AF_PACKET && LINUX_DATAPATH
1020 else if (ss.ss_family == AF_PACKET) {
1021 struct sockaddr_ll sll;
1023 memcpy(&sll, &ss, sizeof sll);
1024 ds_put_cstr(string, "AF_PACKET");
1025 if (sll.sll_ifindex) {
1026 char name[IFNAMSIZ];
1028 if (if_indextoname(sll.sll_ifindex, name)) {
1029 ds_put_format(string, "(%s)", name);
1031 ds_put_format(string, "(ifindex=%d)", sll.sll_ifindex);
1034 if (sll.sll_protocol) {
1035 ds_put_format(string, "(protocol=0x%"PRIu16")",
1036 ntohs(sll.sll_protocol));
1040 else if (ss.ss_family == AF_UNSPEC) {
1041 ds_put_cstr(string, "AF_UNSPEC");
1043 ds_put_format(string, "AF_%d", (int) ss.ss_family);
1049 #ifdef LINUX_DATAPATH
1051 put_fd_filename(struct ds *string, int fd)
1057 linkname = xasprintf("/proc/self/fd/%d", fd);
1058 n = readlink(linkname, buf, sizeof buf);
1060 ds_put_char(string, ' ');
1061 ds_put_buffer(string, buf, n);
1062 if (n > sizeof buf) {
1063 ds_put_cstr(string, "...");
1070 /* Returns a malloc()'d string describing 'fd', for use in logging. */
1078 if (fstat(fd, &s)) {
1079 ds_put_format(&string, "fstat failed (%s)", strerror(errno));
1080 } else if (S_ISSOCK(s.st_mode)) {
1081 describe_sockaddr(&string, fd, getsockname);
1082 ds_put_cstr(&string, "<->");
1083 describe_sockaddr(&string, fd, getpeername);
1085 ds_put_cstr(&string, (isatty(fd) ? "tty"
1086 : S_ISDIR(s.st_mode) ? "directory"
1087 : S_ISCHR(s.st_mode) ? "character device"
1088 : S_ISBLK(s.st_mode) ? "block device"
1089 : S_ISREG(s.st_mode) ? "file"
1090 : S_ISFIFO(s.st_mode) ? "FIFO"
1091 : S_ISLNK(s.st_mode) ? "symbolic link"
1093 #ifdef LINUX_DATAPATH
1094 put_fd_filename(&string, fd);
1097 return ds_steal_cstr(&string);
1100 /* Returns the total of the 'iov_len' members of the 'n_iovs' in 'iovs'.
1101 * The caller must ensure that the total does not exceed SIZE_MAX. */
1103 iovec_len(const struct iovec iovs[], size_t n_iovs)
1108 for (i = 0; i < n_iovs; i++) {
1109 len += iovs[i].iov_len;
1114 /* Returns true if all of the 'n_iovs' iovecs in 'iovs' have length zero. */
1116 iovec_is_empty(const struct iovec iovs[], size_t n_iovs)
1120 for (i = 0; i < n_iovs; i++) {
1121 if (iovs[i].iov_len) {
1128 /* Sends the 'n_iovs' iovecs of data in 'iovs' and the 'n_fds' file descriptors
1129 * in 'fds' on Unix domain socket 'sock'. Returns the number of bytes
1130 * successfully sent or -1 if an error occurred. On error, sets errno
1133 send_iovec_and_fds(int sock,
1134 const struct iovec *iovs, size_t n_iovs,
1135 const int fds[], size_t n_fds)
1137 ovs_assert(sock >= 0);
1141 char control[CMSG_SPACE(SOUTIL_MAX_FDS * sizeof *fds)];
1145 ovs_assert(!iovec_is_empty(iovs, n_iovs));
1146 ovs_assert(n_fds <= SOUTIL_MAX_FDS);
1148 memset(&cmsg, 0, sizeof cmsg);
1149 cmsg.cm.cmsg_len = CMSG_LEN(n_fds * sizeof *fds);
1150 cmsg.cm.cmsg_level = SOL_SOCKET;
1151 cmsg.cm.cmsg_type = SCM_RIGHTS;
1152 memcpy(CMSG_DATA(&cmsg.cm), fds, n_fds * sizeof *fds);
1154 msg.msg_name = NULL;
1155 msg.msg_namelen = 0;
1156 msg.msg_iov = CONST_CAST(struct iovec *, iovs);
1157 msg.msg_iovlen = n_iovs;
1158 msg.msg_control = &cmsg.cm;
1159 msg.msg_controllen = CMSG_SPACE(n_fds * sizeof *fds);
1162 return sendmsg(sock, &msg, 0);
1164 return writev(sock, iovs, n_iovs);
1168 /* Sends the 'n_iovs' iovecs of data in 'iovs' and the 'n_fds' file descriptors
1169 * in 'fds' on Unix domain socket 'sock'. If 'skip_bytes' is nonzero, then the
1170 * first 'skip_bytes' of data in the iovecs are not sent, and none of the file
1171 * descriptors are sent. The function continues to retry sending until an
1172 * error (other than EINTR) occurs or all the data and fds are sent.
1174 * Returns 0 if all the data and fds were successfully sent, otherwise a
1175 * positive errno value. Regardless of success, stores the number of bytes
1176 * sent (always at least 'skip_bytes') in '*bytes_sent'. (If at least one byte
1177 * is sent, then all the fds have been sent.)
1179 * 'skip_bytes' must be less than or equal to iovec_len(iovs, n_iovs). */
1181 send_iovec_and_fds_fully(int sock,
1182 const struct iovec iovs[], size_t n_iovs,
1183 const int fds[], size_t n_fds,
1184 size_t skip_bytes, size_t *bytes_sent)
1187 while (n_iovs > 0) {
1191 retval = skip_bytes;
1193 } else if (!*bytes_sent) {
1194 retval = send_iovec_and_fds(sock, iovs, n_iovs, fds, n_fds);
1196 retval = writev(sock, iovs, n_iovs);
1200 *bytes_sent += retval;
1201 while (retval > 0) {
1202 const uint8_t *base = iovs->iov_base;
1203 size_t len = iovs->iov_len;
1209 error = write_fully(sock, base + retval, len - retval,
1211 *bytes_sent += sent;
1221 } else if (retval == 0) {
1222 if (iovec_is_empty(iovs, n_iovs)) {
1225 VLOG_WARN("send returned 0");
1227 } else if (errno != EINTR) {
1235 /* Sends the 'n_iovs' iovecs of data in 'iovs' and the 'n_fds' file descriptors
1236 * in 'fds' on Unix domain socket 'sock'. The function continues to retry
1237 * sending until an error (other than EAGAIN or EINTR) occurs or all the data
1238 * and fds are sent. Upon EAGAIN, the function blocks until the socket is
1239 * ready for more data.
1241 * Returns 0 if all the data and fds were successfully sent, otherwise a
1242 * positive errno value. */
1244 send_iovec_and_fds_fully_block(int sock,
1245 const struct iovec iovs[], size_t n_iovs,
1246 const int fds[], size_t n_fds)
1253 error = send_iovec_and_fds_fully(sock, iovs, n_iovs,
1254 fds, n_fds, sent, &sent);
1255 if (error != EAGAIN) {
1258 poll_fd_wait(sock, POLLOUT);
1263 /* Attempts to receive from Unix domain socket 'sock' up to 'size' bytes of
1264 * data into 'data' and up to SOUTIL_MAX_FDS file descriptors into 'fds'.
1266 * - Upon success, returns the number of bytes of data copied into 'data'
1267 * and stores the number of received file descriptors into '*n_fdsp'.
1269 * - On failure, returns a negative errno value and stores 0 in
1272 * - On EOF, returns 0 and stores 0 in '*n_fdsp'. */
1274 recv_data_and_fds(int sock,
1275 void *data, size_t size,
1276 int fds[SOUTIL_MAX_FDS], size_t *n_fdsp)
1280 char control[CMSG_SPACE(SOUTIL_MAX_FDS * sizeof *fds)];
1292 iov.iov_base = data;
1295 msg.msg_name = NULL;
1296 msg.msg_namelen = 0;
1299 msg.msg_control = &cmsg.cm;
1300 msg.msg_controllen = sizeof cmsg.control;
1303 retval = recvmsg(sock, &msg, 0);
1304 } while (retval < 0 && errno == EINTR);
1306 return retval < 0 ? -errno : 0;
1309 for (p = CMSG_FIRSTHDR(&msg); p; p = CMSG_NXTHDR(&msg, p)) {
1310 if (p->cmsg_level != SOL_SOCKET || p->cmsg_type != SCM_RIGHTS) {
1311 VLOG_ERR("unexpected control message %d:%d",
1312 p->cmsg_level, p->cmsg_type);
1314 } else if (*n_fdsp) {
1315 VLOG_ERR("multiple SCM_RIGHTS received");
1318 size_t n_fds = (p->cmsg_len - CMSG_LEN(0)) / sizeof *fds;
1319 const int *fds_data = (const int *) CMSG_DATA(p);
1321 ovs_assert(n_fds > 0);
1322 if (n_fds > SOUTIL_MAX_FDS) {
1323 VLOG_ERR("%zu fds received but only %d supported",
1324 n_fds, SOUTIL_MAX_FDS);
1325 for (i = 0; i < n_fds; i++) {
1332 memcpy(fds, fds_data, n_fds * sizeof *fds);
1339 for (i = 0; i < *n_fdsp; i++) {