2 * Copyright (c) 2008, 2009, 2010, 2011, 2012 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>
30 #include <sys/resource.h>
31 #include <sys/socket.h>
35 #include "dynamic-string.h"
36 #include "fatal-signal.h"
38 #include "poll-loop.h"
41 #if AF_PACKET && __linux__
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)" instead of "#ifdef __linux__". */
64 static int getsockopt_int(int fd, int level, int option, const char *optname,
67 /* Sets 'fd' to non-blocking mode. Returns 0 if successful, otherwise a
68 * positive errno value. */
70 set_nonblocking(int fd)
72 int flags = fcntl(fd, F_GETFL, 0);
74 if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) != -1) {
77 VLOG_ERR("fcntl(F_SETFL) failed: %s", strerror(errno));
81 VLOG_ERR("fcntl(F_GETFL) failed: %s", strerror(errno));
87 xset_nonblocking(int fd)
89 if (set_nonblocking(fd)) {
95 set_dscp(int fd, uint8_t dscp)
102 if (setsockopt(fd, IPPROTO_IP, IP_TOS, &dscp, sizeof dscp)) {
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 * gethostbyname() 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 activty. (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)
190 if (inet_aton(host_name, addr)) {
194 h = gethostbyname(host_name);
196 *addr = *(struct in_addr *) h->h_addr;
200 return (h_errno == HOST_NOT_FOUND ? ENOENT
201 : h_errno == TRY_AGAIN ? EAGAIN
202 : h_errno == NO_RECOVERY ? EIO
203 : h_errno == NO_ADDRESS ? ENXIO
207 /* Returns the error condition associated with socket 'fd' and resets the
208 * socket's error status. */
210 get_socket_error(int fd)
214 if (getsockopt_int(fd, SOL_SOCKET, SO_ERROR, "SO_ERROR", &error)) {
221 check_connection_completion(int fd)
227 pfd.events = POLLOUT;
229 retval = poll(&pfd, 1, 0);
230 } while (retval < 0 && errno == EINTR);
232 return get_socket_error(fd);
233 } else if (retval < 0) {
234 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
235 VLOG_ERR_RL(&rl, "poll: %s", strerror(errno));
242 /* Drain all the data currently in the receive queue of a datagram socket (and
243 * possibly additional data). There is no way to know how many packets are in
244 * the receive queue, but we do know that the total number of bytes queued does
245 * not exceed the receive buffer size, so we pull packets until none are left
246 * or we've read that many bytes. */
252 rcvbuf = get_socket_rcvbuf(fd);
258 /* In Linux, specifying MSG_TRUNC in the flags argument causes the
259 * datagram length to be returned, even if that is longer than the
260 * buffer provided. Thus, we can use a 1-byte buffer to discard the
261 * incoming datagram and still be able to account how many bytes were
262 * removed from the receive buffer.
264 * On other Unix-like OSes, MSG_TRUNC has no effect in the flags
266 char buffer[LINUX ? 1 : 2048];
267 ssize_t n_bytes = recv(fd, buffer, sizeof buffer,
268 MSG_TRUNC | MSG_DONTWAIT);
269 if (n_bytes <= 0 || n_bytes >= rcvbuf) {
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) {
307 /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in
308 * '*un_len' the size of the sockaddr_un. */
310 make_sockaddr_un__(const char *name, struct sockaddr_un *un, socklen_t *un_len)
312 un->sun_family = AF_UNIX;
313 ovs_strzcpy(un->sun_path, name, sizeof un->sun_path);
314 *un_len = (offsetof(struct sockaddr_un, sun_path)
315 + strlen (un->sun_path) + 1);
318 /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in
319 * '*un_len' the size of the sockaddr_un.
321 * Returns 0 on success, otherwise a positive errno value. On success,
322 * '*dirfdp' is either -1 or a nonnegative file descriptor that the caller
323 * should close after using '*un' to bind or connect. On failure, '*dirfdp' is
326 make_sockaddr_un(const char *name, struct sockaddr_un *un, socklen_t *un_len,
329 enum { MAX_UN_LEN = sizeof un->sun_path - 1 };
332 if (strlen(name) > MAX_UN_LEN) {
333 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
336 /* 'name' is too long to fit in a sockaddr_un, but we have a
337 * workaround for that on Linux: shorten it by opening a file
338 * descriptor for the directory part of the name and indirecting
339 * through /proc/self/fd/<dirfd>/<basename>. */
344 dir = dir_name(name);
345 base = base_name(name);
347 dirfd = open(dir, O_DIRECTORY | O_RDONLY);
354 short_name = xasprintf("/proc/self/fd/%d/%s", dirfd, base);
358 if (strlen(short_name) <= MAX_UN_LEN) {
359 make_sockaddr_un__(short_name, un, un_len);
367 VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum "
368 "%d bytes (even shortened)", name, MAX_UN_LEN);
370 /* 'name' is too long and we have no workaround. */
371 VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum "
372 "%d bytes", name, MAX_UN_LEN);
377 make_sockaddr_un__(name, un, un_len);
382 /* Binds Unix domain socket 'fd' to a file with permissions 0700. */
384 bind_unix_socket(int fd, struct sockaddr *sun, socklen_t sun_len)
386 /* According to _Unix Network Programming_, umask should affect bind(). */
387 mode_t old_umask = umask(0077);
388 int error = bind(fd, sun, sun_len) ? errno : 0;
393 /* Creates a Unix domain socket in the given 'style' (either SOCK_DGRAM or
394 * SOCK_STREAM) that is bound to '*bind_path' (if 'bind_path' is non-null) and
395 * connected to '*connect_path' (if 'connect_path' is non-null). If 'nonblock'
396 * is true, the socket is made non-blocking.
398 * Returns the socket's fd if successful, otherwise a negative errno value. */
400 make_unix_socket(int style, bool nonblock,
401 const char *bind_path, const char *connect_path)
406 fd = socket(PF_UNIX, style, 0);
411 /* Set nonblocking mode right away, if we want it. This prevents blocking
412 * in connect(), if connect_path != NULL. (In turn, that's a corner case:
413 * it will only happen if style is SOCK_STREAM or SOCK_SEQPACKET, and only
414 * if a backlog of un-accepted connections has built up in the kernel.) */
416 int flags = fcntl(fd, F_GETFL, 0);
421 if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) == -1) {
428 struct sockaddr_un un;
432 if (unlink(bind_path) && errno != ENOENT) {
433 VLOG_WARN("unlinking \"%s\": %s\n", bind_path, strerror(errno));
435 fatal_signal_add_file_to_unlink(bind_path);
437 error = make_sockaddr_un(bind_path, &un, &un_len, &dirfd);
439 error = bind_unix_socket(fd, (struct sockaddr *) &un, un_len);
450 struct sockaddr_un un;
454 error = make_sockaddr_un(connect_path, &un, &un_len, &dirfd);
456 && connect(fd, (struct sockaddr*) &un, un_len)
457 && errno != EINPROGRESS) {
471 if (error == EAGAIN) {
475 fatal_signal_unlink_file_now(bind_path);
482 get_unix_name_len(socklen_t sun_len)
484 return (sun_len >= offsetof(struct sockaddr_un, sun_path)
485 ? sun_len - offsetof(struct sockaddr_un, sun_path)
490 guess_netmask(ovs_be32 ip_)
492 uint32_t ip = ntohl(ip_);
493 return ((ip >> 31) == 0 ? htonl(0xff000000) /* Class A */
494 : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */
495 : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */
496 : htonl(0)); /* ??? */
499 /* Parses 'target', which should be a string in the format "<host>[:<port>]".
500 * <host> is required. If 'default_port' is nonzero then <port> is optional
501 * and defaults to 'default_port'.
503 * On success, returns true and stores the parsed remote address into '*sinp'.
504 * On failure, logs an error, stores zeros into '*sinp', and returns false. */
506 inet_parse_active(const char *target_, uint16_t default_port,
507 struct sockaddr_in *sinp)
509 char *target = xstrdup(target_);
510 char *save_ptr = NULL;
511 const char *host_name;
512 const char *port_string;
516 sinp->sin_family = AF_INET;
517 sinp->sin_port = htons(default_port);
520 host_name = strtok_r(target, ":", &save_ptr);
521 port_string = strtok_r(NULL, ":", &save_ptr);
523 VLOG_ERR("%s: bad peer name format", target_);
527 /* Look up IP, port. */
528 if (lookup_ip(host_name, &sinp->sin_addr)) {
531 if (port_string && atoi(port_string)) {
532 sinp->sin_port = htons(atoi(port_string));
533 } else if (!default_port) {
534 VLOG_ERR("%s: port number must be specified", target_);
542 memset(sinp, 0, sizeof *sinp);
548 /* Opens a non-blocking IPv4 socket of the specified 'style' and connects to
549 * 'target', which should be a string in the format "<host>[:<port>]". <host>
550 * is required. If 'default_port' is nonzero then <port> is optional and
551 * defaults to 'default_port'.
553 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
555 * On success, returns 0 (indicating connection complete) or EAGAIN (indicating
556 * connection in progress), in which case the new file descriptor is stored
557 * into '*fdp'. On failure, returns a positive errno value other than EAGAIN
558 * and stores -1 into '*fdp'.
560 * If 'sinp' is non-null, then on success the target address is stored into
563 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
564 * should be in the range [0, 63] and will automatically be shifted to the
565 * appropriately place in the IP tos field. */
567 inet_open_active(int style, const char *target, uint16_t default_port,
568 struct sockaddr_in *sinp, int *fdp, uint8_t dscp)
570 struct sockaddr_in sin;
575 if (!inet_parse_active(target, default_port, &sin)) {
576 error = EAFNOSUPPORT;
580 /* Create non-blocking socket. */
581 fd = socket(AF_INET, style, 0);
583 VLOG_ERR("%s: socket: %s", target, strerror(errno));
587 error = set_nonblocking(fd);
592 /* The dscp bits must be configured before connect() to ensure that the TOS
593 * field is set during the connection establishment. If set after
594 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
595 error = set_dscp(fd, dscp);
597 VLOG_ERR("%s: socket: %s", target, strerror(error));
602 error = connect(fd, (struct sockaddr *) &sin, sizeof sin) == 0 ? 0 : errno;
603 if (error == EINPROGRESS) {
608 if (!error || error == EAGAIN) {
612 } else if (fd >= 0) {
619 /* Parses 'target', which should be a string in the format "[<port>][:<ip>]":
621 * - If 'default_port' is -1, then <port> is required. Otherwise, if
622 * <port> is omitted, then 'default_port' is used instead.
624 * - If <port> (or 'default_port', if used) is 0, then no port is bound
625 * and the TCP/IP stack will select a port.
627 * - If <ip> is omitted then the IP address is wildcarded.
629 * If successful, stores the address into '*sinp' and returns true; otherwise
630 * zeros '*sinp' and returns false. */
632 inet_parse_passive(const char *target_, int default_port,
633 struct sockaddr_in *sinp)
635 char *target = xstrdup(target_);
636 char *string_ptr = target;
637 const char *host_name;
638 const char *port_string;
642 /* Address defaults. */
643 memset(sinp, 0, sizeof *sinp);
644 sinp->sin_family = AF_INET;
645 sinp->sin_addr.s_addr = htonl(INADDR_ANY);
646 sinp->sin_port = htons(default_port);
648 /* Parse optional port number. */
649 port_string = strsep(&string_ptr, ":");
650 if (port_string && str_to_int(port_string, 10, &port)) {
651 sinp->sin_port = htons(port);
652 } else if (default_port < 0) {
653 VLOG_ERR("%s: port number must be specified", target_);
657 /* Parse optional bind IP. */
658 host_name = strsep(&string_ptr, ":");
659 if (host_name && host_name[0] && lookup_ip(host_name, &sinp->sin_addr)) {
667 memset(sinp, 0, sizeof *sinp);
674 /* Opens a non-blocking IPv4 socket of the specified 'style', binds to
675 * 'target', and listens for incoming connections. Parses 'target' in the same
676 * way was inet_parse_passive().
678 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
680 * For TCP, the socket will have SO_REUSEADDR turned on.
682 * On success, returns a non-negative file descriptor. On failure, returns a
683 * negative errno value.
685 * If 'sinp' is non-null, then on success the bound address is stored into
688 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
689 * should be in the range [0, 63] and will automatically be shifted to the
690 * appropriately place in the IP tos field. */
692 inet_open_passive(int style, const char *target, int default_port,
693 struct sockaddr_in *sinp, uint8_t dscp)
695 struct sockaddr_in sin;
697 unsigned int yes = 1;
699 if (!inet_parse_passive(target, default_port, &sin)) {
700 return -EAFNOSUPPORT;
703 /* Create non-blocking socket, set SO_REUSEADDR. */
704 fd = socket(AF_INET, style, 0);
707 VLOG_ERR("%s: socket: %s", target, strerror(error));
710 error = set_nonblocking(fd);
714 if (style == SOCK_STREAM
715 && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) {
717 VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s", target, strerror(error));
722 if (bind(fd, (struct sockaddr *) &sin, sizeof sin) < 0) {
724 VLOG_ERR("%s: bind: %s", target, strerror(error));
728 /* The dscp bits must be configured before connect() to ensure that the TOS
729 * field is set during the connection establishment. If set after
730 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
731 error = set_dscp(fd, dscp);
733 VLOG_ERR("%s: socket: %s", target, strerror(error));
738 if (style == SOCK_STREAM && listen(fd, 10) < 0) {
740 VLOG_ERR("%s: listen: %s", target, strerror(error));
745 socklen_t sin_len = sizeof sin;
746 if (getsockname(fd, (struct sockaddr *) &sin, &sin_len) < 0){
748 VLOG_ERR("%s: getsockname: %s", target, strerror(error));
751 if (sin.sin_family != AF_INET || sin_len != sizeof sin) {
752 error = EAFNOSUPPORT;
753 VLOG_ERR("%s: getsockname: invalid socket name", target);
766 /* Returns a readable and writable fd for /dev/null, if successful, otherwise
767 * a negative errno value. The caller must not close the returned fd (because
768 * the same fd will be handed out to subsequent callers). */
772 static int null_fd = -1;
774 null_fd = open("/dev/null", O_RDWR);
777 VLOG_ERR("could not open /dev/null: %s", strerror(error));
785 read_fully(int fd, void *p_, size_t size, size_t *bytes_read)
791 ssize_t retval = read(fd, p, size);
793 *bytes_read += retval;
796 } else if (retval == 0) {
798 } else if (errno != EINTR) {
806 write_fully(int fd, const void *p_, size_t size, size_t *bytes_written)
808 const uint8_t *p = p_;
812 ssize_t retval = write(fd, p, size);
814 *bytes_written += retval;
817 } else if (retval == 0) {
818 VLOG_WARN("write returned 0");
820 } else if (errno != EINTR) {
827 /* Given file name 'file_name', fsyncs the directory in which it is contained.
828 * Returns 0 if successful, otherwise a positive errno value. */
830 fsync_parent_dir(const char *file_name)
836 dir = dir_name(file_name);
837 fd = open(dir, O_RDONLY);
840 if (errno == EINVAL || errno == EROFS) {
841 /* This directory does not support synchronization. Not
842 * really an error. */
845 VLOG_ERR("%s: fsync failed (%s)", dir, strerror(error));
851 VLOG_ERR("%s: open failed (%s)", dir, strerror(error));
858 /* Obtains the modification time of the file named 'file_name' to the greatest
859 * supported precision. If successful, stores the mtime in '*mtime' and
860 * returns 0. On error, returns a positive errno value and stores zeros in
863 get_mtime(const char *file_name, struct timespec *mtime)
867 if (!stat(file_name, &s)) {
868 mtime->tv_sec = s.st_mtime;
870 #if HAVE_STRUCT_STAT_ST_MTIM_TV_NSEC
871 mtime->tv_nsec = s.st_mtim.tv_nsec;
872 #elif HAVE_STRUCT_STAT_ST_MTIMENSEC
873 mtime->tv_nsec = s.st_mtimensec;
880 mtime->tv_sec = mtime->tv_nsec = 0;
889 VLOG_FATAL("failed to create pipe (%s)", strerror(errno));
894 xsocketpair(int domain, int type, int protocol, int fds[2])
896 if (socketpair(domain, type, protocol, fds)) {
897 VLOG_FATAL("failed to create socketpair (%s)", strerror(errno));
902 getsockopt_int(int fd, int level, int option, const char *optname, int *valuep)
904 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
910 if (getsockopt(fd, level, option, &value, &len)) {
912 VLOG_ERR_RL(&rl, "getsockopt(%s): %s", optname, strerror(error));
913 } else if (len != sizeof value) {
915 VLOG_ERR_RL(&rl, "getsockopt(%s): value is %u bytes (expected %zu)",
916 optname, (unsigned int) len, sizeof value);
921 *valuep = error ? 0 : value;
926 describe_sockaddr(struct ds *string, int fd,
927 int (*getaddr)(int, struct sockaddr *, socklen_t *))
929 struct sockaddr_storage ss;
930 socklen_t len = sizeof ss;
932 if (!getaddr(fd, (struct sockaddr *) &ss, &len)) {
933 if (ss.ss_family == AF_INET) {
934 struct sockaddr_in sin;
936 memcpy(&sin, &ss, sizeof sin);
937 ds_put_format(string, IP_FMT":%"PRIu16,
938 IP_ARGS(&sin.sin_addr.s_addr), ntohs(sin.sin_port));
939 } else if (ss.ss_family == AF_UNIX) {
940 struct sockaddr_un sun;
944 memcpy(&sun, &ss, sizeof sun);
945 maxlen = len - offsetof(struct sockaddr_un, sun_path);
946 null = memchr(sun.sun_path, '\0', maxlen);
947 ds_put_buffer(string, sun.sun_path,
948 null ? null - sun.sun_path : maxlen);
951 else if (ss.ss_family == AF_NETLINK) {
954 /* SO_PROTOCOL was introduced in 2.6.32. Support it regardless of the version
955 * of the Linux kernel headers in use at build time. */
957 #define SO_PROTOCOL 38
960 if (!getsockopt_int(fd, SOL_SOCKET, SO_PROTOCOL, "SO_PROTOCOL",
964 ds_put_cstr(string, "NETLINK_ROUTE");
967 case NETLINK_GENERIC:
968 ds_put_cstr(string, "NETLINK_GENERIC");
972 ds_put_format(string, "AF_NETLINK family %d", protocol);
976 ds_put_cstr(string, "AF_NETLINK");
980 #if AF_PACKET && __linux__
981 else if (ss.ss_family == AF_PACKET) {
982 struct sockaddr_ll sll;
984 memcpy(&sll, &ss, sizeof sll);
985 ds_put_cstr(string, "AF_PACKET");
986 if (sll.sll_ifindex) {
989 if (if_indextoname(sll.sll_ifindex, name)) {
990 ds_put_format(string, "(%s)", name);
992 ds_put_format(string, "(ifindex=%d)", sll.sll_ifindex);
995 if (sll.sll_protocol) {
996 ds_put_format(string, "(protocol=0x%"PRIu16")",
997 ntohs(sll.sll_protocol));
1001 else if (ss.ss_family == AF_UNSPEC) {
1002 ds_put_cstr(string, "AF_UNSPEC");
1004 ds_put_format(string, "AF_%d", (int) ss.ss_family);
1012 put_fd_filename(struct ds *string, int fd)
1018 linkname = xasprintf("/proc/self/fd/%d", fd);
1019 n = readlink(linkname, buf, sizeof buf);
1021 ds_put_char(string, ' ');
1022 ds_put_buffer(string, buf, n);
1023 if (n > sizeof buf) {
1024 ds_put_cstr(string, "...");
1031 /* Returns a malloc()'d string describing 'fd', for use in logging. */
1039 if (fstat(fd, &s)) {
1040 ds_put_format(&string, "fstat failed (%s)", strerror(errno));
1041 } else if (S_ISSOCK(s.st_mode)) {
1042 describe_sockaddr(&string, fd, getsockname);
1043 ds_put_cstr(&string, "<->");
1044 describe_sockaddr(&string, fd, getpeername);
1046 ds_put_cstr(&string, (isatty(fd) ? "tty"
1047 : S_ISDIR(s.st_mode) ? "directory"
1048 : S_ISCHR(s.st_mode) ? "character device"
1049 : S_ISBLK(s.st_mode) ? "block device"
1050 : S_ISREG(s.st_mode) ? "file"
1051 : S_ISFIFO(s.st_mode) ? "FIFO"
1052 : S_ISLNK(s.st_mode) ? "symbolic link"
1055 put_fd_filename(&string, fd);
1058 return ds_steal_cstr(&string);
1061 /* Returns the total of the 'iov_len' members of the 'n_iovs' in 'iovs'.
1062 * The caller must ensure that the total does not exceed SIZE_MAX. */
1064 iovec_len(const struct iovec iovs[], size_t n_iovs)
1069 for (i = 0; i < n_iovs; i++) {
1070 len += iovs[i].iov_len;
1075 /* Returns true if all of the 'n_iovs' iovecs in 'iovs' have length zero. */
1077 iovec_is_empty(const struct iovec iovs[], size_t n_iovs)
1081 for (i = 0; i < n_iovs; i++) {
1082 if (iovs[i].iov_len) {
1089 /* Sends the 'n_iovs' iovecs of data in 'iovs' and the 'n_fds' file descriptors
1090 * in 'fds' on Unix domain socket 'sock'. Returns the number of bytes
1091 * successfully sent or -1 if an error occurred. On error, sets errno
1094 send_iovec_and_fds(int sock,
1095 const struct iovec *iovs, size_t n_iovs,
1096 const int fds[], size_t n_fds)
1102 char control[CMSG_SPACE(SOUTIL_MAX_FDS * sizeof *fds)];
1106 assert(!iovec_is_empty(iovs, n_iovs));
1107 assert(n_fds <= SOUTIL_MAX_FDS);
1109 memset(&cmsg, 0, sizeof cmsg);
1110 cmsg.cm.cmsg_len = CMSG_LEN(n_fds * sizeof *fds);
1111 cmsg.cm.cmsg_level = SOL_SOCKET;
1112 cmsg.cm.cmsg_type = SCM_RIGHTS;
1113 memcpy(CMSG_DATA(&cmsg.cm), fds, n_fds * sizeof *fds);
1115 msg.msg_name = NULL;
1116 msg.msg_namelen = 0;
1117 msg.msg_iov = (struct iovec *) iovs;
1118 msg.msg_iovlen = n_iovs;
1119 msg.msg_control = &cmsg.cm;
1120 msg.msg_controllen = CMSG_SPACE(n_fds * sizeof *fds);
1123 return sendmsg(sock, &msg, 0);
1125 return writev(sock, iovs, n_iovs);
1129 /* Sends the 'n_iovs' iovecs of data in 'iovs' and the 'n_fds' file descriptors
1130 * in 'fds' on Unix domain socket 'sock'. If 'skip_bytes' is nonzero, then the
1131 * first 'skip_bytes' of data in the iovecs are not sent, and none of the file
1132 * descriptors are sent. The function continues to retry sending until an
1133 * error (other than EINTR) occurs or all the data and fds are sent.
1135 * Returns 0 if all the data and fds were successfully sent, otherwise a
1136 * positive errno value. Regardless of success, stores the number of bytes
1137 * sent (always at least 'skip_bytes') in '*bytes_sent'. (If at least one byte
1138 * is sent, then all the fds have been sent.)
1140 * 'skip_bytes' must be less than or equal to iovec_len(iovs, n_iovs). */
1142 send_iovec_and_fds_fully(int sock,
1143 const struct iovec iovs[], size_t n_iovs,
1144 const int fds[], size_t n_fds,
1145 size_t skip_bytes, size_t *bytes_sent)
1148 while (n_iovs > 0) {
1152 retval = skip_bytes;
1154 } else if (!*bytes_sent) {
1155 retval = send_iovec_and_fds(sock, iovs, n_iovs, fds, n_fds);
1157 retval = writev(sock, iovs, n_iovs);
1161 *bytes_sent += retval;
1162 while (retval > 0) {
1163 const uint8_t *base = iovs->iov_base;
1164 size_t len = iovs->iov_len;
1170 error = write_fully(sock, base + retval, len - retval,
1172 *bytes_sent += sent;
1182 } else if (retval == 0) {
1183 if (iovec_is_empty(iovs, n_iovs)) {
1186 VLOG_WARN("send returned 0");
1188 } else if (errno != EINTR) {
1196 /* Sends the 'n_iovs' iovecs of data in 'iovs' and the 'n_fds' file descriptors
1197 * in 'fds' on Unix domain socket 'sock'. The function continues to retry
1198 * sending until an error (other than EAGAIN or EINTR) occurs or all the data
1199 * and fds are sent. Upon EAGAIN, the function blocks until the socket is
1200 * ready for more data.
1202 * Returns 0 if all the data and fds were successfully sent, otherwise a
1203 * positive errno value. */
1205 send_iovec_and_fds_fully_block(int sock,
1206 const struct iovec iovs[], size_t n_iovs,
1207 const int fds[], size_t n_fds)
1214 error = send_iovec_and_fds_fully(sock, iovs, n_iovs,
1215 fds, n_fds, sent, &sent);
1216 if (error != EAGAIN) {
1219 poll_fd_wait(sock, POLLOUT);
1224 /* Attempts to receive from Unix domain socket 'sock' up to 'size' bytes of
1225 * data into 'data' and up to SOUTIL_MAX_FDS file descriptors into 'fds'.
1227 * - Upon success, returns the number of bytes of data copied into 'data'
1228 * and stores the number of received file descriptors into '*n_fdsp'.
1230 * - On failure, returns a negative errno value and stores 0 in
1233 * - On EOF, returns 0 and stores 0 in '*n_fdsp'. */
1235 recv_data_and_fds(int sock,
1236 void *data, size_t size,
1237 int fds[SOUTIL_MAX_FDS], size_t *n_fdsp)
1241 char control[CMSG_SPACE(SOUTIL_MAX_FDS * sizeof *fds)];
1253 iov.iov_base = data;
1256 msg.msg_name = NULL;
1257 msg.msg_namelen = 0;
1260 msg.msg_control = &cmsg.cm;
1261 msg.msg_controllen = sizeof cmsg.control;
1264 retval = recvmsg(sock, &msg, 0);
1265 } while (retval < 0 && errno == EINTR);
1267 return retval < 0 ? -errno : 0;
1270 for (p = CMSG_FIRSTHDR(&msg); p; p = CMSG_NXTHDR(&msg, p)) {
1271 if (p->cmsg_level != SOL_SOCKET || p->cmsg_type != SCM_RIGHTS) {
1272 VLOG_ERR("unexpected control message %d:%d",
1273 p->cmsg_level, p->cmsg_type);
1275 } else if (*n_fdsp) {
1276 VLOG_ERR("multiple SCM_RIGHTS received");
1279 size_t n_fds = (p->cmsg_len - CMSG_LEN(0)) / sizeof *fds;
1280 const int *fds_data = (const int *) CMSG_DATA(p);
1283 if (n_fds > SOUTIL_MAX_FDS) {
1284 VLOG_ERR("%zu fds received but only %d supported",
1285 n_fds, SOUTIL_MAX_FDS);
1286 for (i = 0; i < n_fds; i++) {
1293 memcpy(fds, fds_data, n_fds * sizeof *fds);
1300 for (i = 0; i < *n_fdsp; i++) {