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>
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", strerror(errno));
80 VLOG_ERR("fcntl(F_GETFL) failed: %s", 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 int max_fds = -1;
139 if (!getrlimit(RLIMIT_NOFILE, &r) && rlim_is_finite(r.rlim_cur)) {
140 max_fds = r.rlim_cur;
142 VLOG_WARN("failed to obtain fd limit, defaulting to 1024");
149 /* Translates 'host_name', which must be a string representation of an IP
150 * address, into a numeric IP address in '*addr'. Returns 0 if successful,
151 * otherwise a positive errno value. */
153 lookup_ip(const char *host_name, struct in_addr *addr)
155 if (!inet_aton(host_name, addr)) {
156 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
157 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name);
163 /* Translates 'host_name', which must be a string representation of an IPv6
164 * address, into a numeric IPv6 address in '*addr'. Returns 0 if successful,
165 * otherwise a positive errno value. */
167 lookup_ipv6(const char *host_name, struct in6_addr *addr)
169 if (inet_pton(AF_INET6, host_name, addr) != 1) {
170 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
171 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IPv6 address", host_name);
177 /* Translates 'host_name', which must be a host name or a string representation
178 * of an IP address, into a numeric IP address in '*addr'. Returns 0 if
179 * successful, otherwise a positive errno value.
181 * Most Open vSwitch code should not use this because it causes deadlocks:
182 * gethostbyname() sends out a DNS request but that starts a new flow for which
183 * OVS must set up a flow, but it can't because it's waiting for a DNS reply.
184 * The synchronous lookup also delays other activity. (Of course we can solve
185 * this but it doesn't seem worthwhile quite yet.) */
187 lookup_hostname(const char *host_name, struct in_addr *addr)
191 if (inet_aton(host_name, addr)) {
195 h = gethostbyname(host_name);
197 *addr = *(struct in_addr *) h->h_addr;
201 return (h_errno == HOST_NOT_FOUND ? ENOENT
202 : h_errno == TRY_AGAIN ? EAGAIN
203 : h_errno == NO_RECOVERY ? EIO
204 : h_errno == NO_ADDRESS ? ENXIO
208 /* Returns the error condition associated with socket 'fd' and resets the
209 * socket's error status. */
211 get_socket_error(int fd)
215 if (getsockopt_int(fd, SOL_SOCKET, SO_ERROR, "SO_ERROR", &error)) {
222 check_connection_completion(int fd)
224 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
229 pfd.events = POLLOUT;
231 retval = poll(&pfd, 1, 0);
232 } while (retval < 0 && errno == EINTR);
234 if (pfd.revents & POLLERR) {
235 ssize_t n = send(fd, "", 1, MSG_DONTWAIT);
239 VLOG_ERR_RL(&rl, "poll return POLLERR but send succeeded");
244 } else if (retval < 0) {
245 VLOG_ERR_RL(&rl, "poll: %s", strerror(errno));
252 /* Drain all the data currently in the receive queue of a datagram socket (and
253 * possibly additional data). There is no way to know how many packets are in
254 * the receive queue, but we do know that the total number of bytes queued does
255 * not exceed the receive buffer size, so we pull packets until none are left
256 * or we've read that many bytes. */
262 rcvbuf = get_socket_rcvbuf(fd);
268 /* In Linux, specifying MSG_TRUNC in the flags argument causes the
269 * datagram length to be returned, even if that is longer than the
270 * buffer provided. Thus, we can use a 1-byte buffer to discard the
271 * incoming datagram and still be able to account how many bytes were
272 * removed from the receive buffer.
274 * On other Unix-like OSes, MSG_TRUNC has no effect in the flags
276 char buffer[LINUX_DATAPATH ? 1 : 2048];
277 ssize_t n_bytes = recv(fd, buffer, sizeof buffer,
278 MSG_TRUNC | MSG_DONTWAIT);
279 if (n_bytes <= 0 || n_bytes >= rcvbuf) {
287 /* Returns the size of socket 'sock''s receive buffer (SO_RCVBUF), or a
288 * negative errno value if an error occurs. */
290 get_socket_rcvbuf(int sock)
295 error = getsockopt_int(sock, SOL_SOCKET, SO_RCVBUF, "SO_RCVBUF", &rcvbuf);
296 return error ? -error : rcvbuf;
299 /* Reads and discards up to 'n' datagrams from 'fd', stopping as soon as no
300 * more data can be immediately read. ('fd' should therefore be in
301 * non-blocking mode.)*/
303 drain_fd(int fd, size_t n_packets)
305 for (; n_packets > 0; n_packets--) {
306 /* 'buffer' only needs to be 1 byte long in most circumstances. This
307 * size is defensive against the possibility that we someday want to
308 * use a Linux tap device without TUN_NO_PI, in which case a buffer
309 * smaller than sizeof(struct tun_pi) will give EINVAL on read. */
311 if (read(fd, buffer, sizeof buffer) <= 0) {
317 /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in
318 * '*un_len' the size of the sockaddr_un. */
320 make_sockaddr_un__(const char *name, struct sockaddr_un *un, socklen_t *un_len)
322 un->sun_family = AF_UNIX;
323 ovs_strzcpy(un->sun_path, name, sizeof un->sun_path);
324 *un_len = (offsetof(struct sockaddr_un, sun_path)
325 + strlen (un->sun_path) + 1);
328 /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in
329 * '*un_len' the size of the sockaddr_un.
331 * Returns 0 on success, otherwise a positive errno value. On success,
332 * '*dirfdp' is either -1 or a nonnegative file descriptor that the caller
333 * should close after using '*un' to bind or connect. On failure, '*dirfdp' is
336 make_sockaddr_un(const char *name, struct sockaddr_un *un, socklen_t *un_len,
339 enum { MAX_UN_LEN = sizeof un->sun_path - 1 };
342 if (strlen(name) > MAX_UN_LEN) {
343 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
345 if (LINUX_DATAPATH) {
346 /* 'name' is too long to fit in a sockaddr_un, but we have a
347 * workaround for that on Linux: shorten it by opening a file
348 * descriptor for the directory part of the name and indirecting
349 * through /proc/self/fd/<dirfd>/<basename>. */
354 dir = dir_name(name);
355 base = base_name(name);
357 dirfd = open(dir, O_DIRECTORY | O_RDONLY);
364 short_name = xasprintf("/proc/self/fd/%d/%s", dirfd, base);
368 if (strlen(short_name) <= MAX_UN_LEN) {
369 make_sockaddr_un__(short_name, un, un_len);
377 VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum "
378 "%d bytes (even shortened)", name, MAX_UN_LEN);
380 /* 'name' is too long and we have no workaround. */
381 VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum "
382 "%d bytes", name, MAX_UN_LEN);
387 make_sockaddr_un__(name, un, un_len);
392 /* Binds Unix domain socket 'fd' to a file with permissions 0700. */
394 bind_unix_socket(int fd, struct sockaddr *sun, socklen_t sun_len)
396 /* According to _Unix Network Programming_, umask should affect bind(). */
397 mode_t old_umask = umask(0077);
398 int error = bind(fd, sun, sun_len) ? errno : 0;
403 /* Creates a Unix domain socket in the given 'style' (either SOCK_DGRAM or
404 * SOCK_STREAM) that is bound to '*bind_path' (if 'bind_path' is non-null) and
405 * connected to '*connect_path' (if 'connect_path' is non-null). If 'nonblock'
406 * is true, the socket is made non-blocking.
408 * Returns the socket's fd if successful, otherwise a negative errno value. */
410 make_unix_socket(int style, bool nonblock,
411 const char *bind_path, const char *connect_path)
416 fd = socket(PF_UNIX, style, 0);
421 /* Set nonblocking mode right away, if we want it. This prevents blocking
422 * in connect(), if connect_path != NULL. (In turn, that's a corner case:
423 * it will only happen if style is SOCK_STREAM or SOCK_SEQPACKET, and only
424 * if a backlog of un-accepted connections has built up in the kernel.) */
426 int flags = fcntl(fd, F_GETFL, 0);
431 if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) == -1) {
438 struct sockaddr_un un;
442 if (unlink(bind_path) && errno != ENOENT) {
443 VLOG_WARN("unlinking \"%s\": %s\n", bind_path, strerror(errno));
445 fatal_signal_add_file_to_unlink(bind_path);
447 error = make_sockaddr_un(bind_path, &un, &un_len, &dirfd);
449 error = bind_unix_socket(fd, (struct sockaddr *) &un, un_len);
460 struct sockaddr_un un;
464 error = make_sockaddr_un(connect_path, &un, &un_len, &dirfd);
466 && connect(fd, (struct sockaddr*) &un, un_len)
467 && errno != EINPROGRESS) {
481 if (error == EAGAIN) {
485 fatal_signal_unlink_file_now(bind_path);
492 get_unix_name_len(socklen_t sun_len)
494 return (sun_len >= offsetof(struct sockaddr_un, sun_path)
495 ? sun_len - offsetof(struct sockaddr_un, sun_path)
500 guess_netmask(ovs_be32 ip_)
502 uint32_t ip = ntohl(ip_);
503 return ((ip >> 31) == 0 ? htonl(0xff000000) /* Class A */
504 : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */
505 : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */
506 : htonl(0)); /* ??? */
509 /* Parses 'target', which should be a string in the format "<host>[:<port>]".
510 * <host> is required. If 'default_port' is nonzero then <port> is optional
511 * and defaults to 'default_port'.
513 * On success, returns true and stores the parsed remote address into '*sinp'.
514 * On failure, logs an error, stores zeros into '*sinp', and returns false. */
516 inet_parse_active(const char *target_, uint16_t default_port,
517 struct sockaddr_in *sinp)
519 char *target = xstrdup(target_);
520 char *save_ptr = NULL;
521 const char *host_name;
522 const char *port_string;
526 sinp->sin_family = AF_INET;
527 sinp->sin_port = htons(default_port);
530 host_name = strtok_r(target, ":", &save_ptr);
531 port_string = strtok_r(NULL, ":", &save_ptr);
533 VLOG_ERR("%s: bad peer name format", target_);
537 /* Look up IP, port. */
538 if (lookup_ip(host_name, &sinp->sin_addr)) {
541 if (port_string && atoi(port_string)) {
542 sinp->sin_port = htons(atoi(port_string));
543 } else if (!default_port) {
544 VLOG_ERR("%s: port number must be specified", target_);
552 memset(sinp, 0, sizeof *sinp);
558 /* Opens a non-blocking IPv4 socket of the specified 'style' and connects to
559 * 'target', which should be a string in the format "<host>[:<port>]". <host>
560 * is required. If 'default_port' is nonzero then <port> is optional and
561 * defaults to 'default_port'.
563 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
565 * On success, returns 0 (indicating connection complete) or EAGAIN (indicating
566 * connection in progress), in which case the new file descriptor is stored
567 * into '*fdp'. On failure, returns a positive errno value other than EAGAIN
568 * and stores -1 into '*fdp'.
570 * If 'sinp' is non-null, then on success the target address is stored into
573 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
574 * should be in the range [0, 63] and will automatically be shifted to the
575 * appropriately place in the IP tos field. */
577 inet_open_active(int style, const char *target, uint16_t default_port,
578 struct sockaddr_in *sinp, int *fdp, uint8_t dscp)
580 struct sockaddr_in sin;
585 if (!inet_parse_active(target, default_port, &sin)) {
586 error = EAFNOSUPPORT;
590 /* Create non-blocking socket. */
591 fd = socket(AF_INET, style, 0);
593 VLOG_ERR("%s: socket: %s", target, strerror(errno));
597 error = set_nonblocking(fd);
602 /* The dscp bits must be configured before connect() to ensure that the TOS
603 * field is set during the connection establishment. If set after
604 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
605 error = set_dscp(fd, dscp);
607 VLOG_ERR("%s: socket: %s", target, strerror(error));
612 error = connect(fd, (struct sockaddr *) &sin, sizeof sin) == 0 ? 0 : errno;
613 if (error == EINPROGRESS) {
618 if (!error || error == EAGAIN) {
622 } else if (fd >= 0) {
630 /* Parses 'target', which should be a string in the format "[<port>][:<ip>]":
632 * - If 'default_port' is -1, then <port> is required. Otherwise, if
633 * <port> is omitted, then 'default_port' is used instead.
635 * - If <port> (or 'default_port', if used) is 0, then no port is bound
636 * and the TCP/IP stack will select a port.
638 * - If <ip> is omitted then the IP address is wildcarded.
640 * If successful, stores the address into '*sinp' and returns true; otherwise
641 * zeros '*sinp' and returns false. */
643 inet_parse_passive(const char *target_, int default_port,
644 struct sockaddr_in *sinp)
646 char *target = xstrdup(target_);
647 char *string_ptr = target;
648 const char *host_name;
649 const char *port_string;
653 /* Address defaults. */
654 memset(sinp, 0, sizeof *sinp);
655 sinp->sin_family = AF_INET;
656 sinp->sin_addr.s_addr = htonl(INADDR_ANY);
657 sinp->sin_port = htons(default_port);
659 /* Parse optional port number. */
660 port_string = strsep(&string_ptr, ":");
661 if (port_string && str_to_int(port_string, 10, &port)) {
662 sinp->sin_port = htons(port);
663 } else if (default_port < 0) {
664 VLOG_ERR("%s: port number must be specified", target_);
668 /* Parse optional bind IP. */
669 host_name = strsep(&string_ptr, ":");
670 if (host_name && host_name[0] && lookup_ip(host_name, &sinp->sin_addr)) {
678 memset(sinp, 0, sizeof *sinp);
685 /* Opens a non-blocking IPv4 socket of the specified 'style', binds to
686 * 'target', and listens for incoming connections. Parses 'target' in the same
687 * way was inet_parse_passive().
689 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
691 * For TCP, the socket will have SO_REUSEADDR turned on.
693 * On success, returns a non-negative file descriptor. On failure, returns a
694 * negative errno value.
696 * If 'sinp' is non-null, then on success the bound address is stored into
699 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
700 * should be in the range [0, 63] and will automatically be shifted to the
701 * appropriately place in the IP tos field. */
703 inet_open_passive(int style, const char *target, int default_port,
704 struct sockaddr_in *sinp, uint8_t dscp)
706 struct sockaddr_in sin;
708 unsigned int yes = 1;
710 if (!inet_parse_passive(target, default_port, &sin)) {
711 return -EAFNOSUPPORT;
714 /* Create non-blocking socket, set SO_REUSEADDR. */
715 fd = socket(AF_INET, style, 0);
718 VLOG_ERR("%s: socket: %s", target, strerror(error));
721 error = set_nonblocking(fd);
725 if (style == SOCK_STREAM
726 && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) {
728 VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s", target, strerror(error));
733 if (bind(fd, (struct sockaddr *) &sin, sizeof sin) < 0) {
735 VLOG_ERR("%s: bind: %s", target, strerror(error));
739 /* The dscp bits must be configured before connect() to ensure that the TOS
740 * field is set during the connection establishment. If set after
741 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
742 error = set_dscp(fd, dscp);
744 VLOG_ERR("%s: socket: %s", target, strerror(error));
749 if (style == SOCK_STREAM && listen(fd, 10) < 0) {
751 VLOG_ERR("%s: listen: %s", target, strerror(error));
756 socklen_t sin_len = sizeof sin;
757 if (getsockname(fd, (struct sockaddr *) &sin, &sin_len) < 0){
759 VLOG_ERR("%s: getsockname: %s", target, strerror(error));
762 if (sin.sin_family != AF_INET || sin_len != sizeof sin) {
763 error = EAFNOSUPPORT;
764 VLOG_ERR("%s: getsockname: invalid socket name", target);
777 /* Returns a readable and writable fd for /dev/null, if successful, otherwise
778 * a negative errno value. The caller must not close the returned fd (because
779 * the same fd will be handed out to subsequent callers). */
783 static int null_fd = -1;
785 null_fd = open("/dev/null", O_RDWR);
788 VLOG_ERR("could not open /dev/null: %s", strerror(error));
796 read_fully(int fd, void *p_, size_t size, size_t *bytes_read)
802 ssize_t retval = read(fd, p, size);
804 *bytes_read += retval;
807 } else if (retval == 0) {
809 } else if (errno != EINTR) {
817 write_fully(int fd, const void *p_, size_t size, size_t *bytes_written)
819 const uint8_t *p = p_;
823 ssize_t retval = write(fd, p, size);
825 *bytes_written += retval;
828 } else if (retval == 0) {
829 VLOG_WARN("write returned 0");
831 } else if (errno != EINTR) {
838 /* Given file name 'file_name', fsyncs the directory in which it is contained.
839 * Returns 0 if successful, otherwise a positive errno value. */
841 fsync_parent_dir(const char *file_name)
847 dir = dir_name(file_name);
848 fd = open(dir, O_RDONLY);
851 if (errno == EINVAL || errno == EROFS) {
852 /* This directory does not support synchronization. Not
853 * really an error. */
856 VLOG_ERR("%s: fsync failed (%s)", dir, strerror(error));
862 VLOG_ERR("%s: open failed (%s)", dir, strerror(error));
869 /* Obtains the modification time of the file named 'file_name' to the greatest
870 * supported precision. If successful, stores the mtime in '*mtime' and
871 * returns 0. On error, returns a positive errno value and stores zeros in
874 get_mtime(const char *file_name, struct timespec *mtime)
878 if (!stat(file_name, &s)) {
879 mtime->tv_sec = s.st_mtime;
881 #if HAVE_STRUCT_STAT_ST_MTIM_TV_NSEC
882 mtime->tv_nsec = s.st_mtim.tv_nsec;
883 #elif HAVE_STRUCT_STAT_ST_MTIMENSEC
884 mtime->tv_nsec = s.st_mtimensec;
891 mtime->tv_sec = mtime->tv_nsec = 0;
900 VLOG_FATAL("failed to create pipe (%s)", strerror(errno));
905 xpipe_nonblocking(int fds[2])
908 xset_nonblocking(fds[0]);
909 xset_nonblocking(fds[1]);
913 xsocketpair(int domain, int type, int protocol, int fds[2])
915 if (socketpair(domain, type, protocol, fds)) {
916 VLOG_FATAL("failed to create socketpair (%s)", strerror(errno));
921 getsockopt_int(int fd, int level, int option, const char *optname, int *valuep)
923 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
929 if (getsockopt(fd, level, option, &value, &len)) {
931 VLOG_ERR_RL(&rl, "getsockopt(%s): %s", optname, strerror(error));
932 } else if (len != sizeof value) {
934 VLOG_ERR_RL(&rl, "getsockopt(%s): value is %u bytes (expected %zu)",
935 optname, (unsigned int) len, sizeof value);
940 *valuep = error ? 0 : value;
945 describe_sockaddr(struct ds *string, int fd,
946 int (*getaddr)(int, struct sockaddr *, socklen_t *))
948 struct sockaddr_storage ss;
949 socklen_t len = sizeof ss;
951 if (!getaddr(fd, (struct sockaddr *) &ss, &len)) {
952 if (ss.ss_family == AF_INET) {
953 struct sockaddr_in sin;
955 memcpy(&sin, &ss, sizeof sin);
956 ds_put_format(string, IP_FMT":%"PRIu16,
957 IP_ARGS(&sin.sin_addr.s_addr), ntohs(sin.sin_port));
958 } else if (ss.ss_family == AF_UNIX) {
959 struct sockaddr_un sun;
963 memcpy(&sun, &ss, sizeof sun);
964 maxlen = len - offsetof(struct sockaddr_un, sun_path);
965 null = memchr(sun.sun_path, '\0', maxlen);
966 ds_put_buffer(string, sun.sun_path,
967 null ? null - sun.sun_path : maxlen);
970 else if (ss.ss_family == AF_NETLINK) {
973 /* SO_PROTOCOL was introduced in 2.6.32. Support it regardless of the version
974 * of the Linux kernel headers in use at build time. */
976 #define SO_PROTOCOL 38
979 if (!getsockopt_int(fd, SOL_SOCKET, SO_PROTOCOL, "SO_PROTOCOL",
983 ds_put_cstr(string, "NETLINK_ROUTE");
986 case NETLINK_GENERIC:
987 ds_put_cstr(string, "NETLINK_GENERIC");
991 ds_put_format(string, "AF_NETLINK family %d", protocol);
995 ds_put_cstr(string, "AF_NETLINK");
999 #if AF_PACKET && LINUX_DATAPATH
1000 else if (ss.ss_family == AF_PACKET) {
1001 struct sockaddr_ll sll;
1003 memcpy(&sll, &ss, sizeof sll);
1004 ds_put_cstr(string, "AF_PACKET");
1005 if (sll.sll_ifindex) {
1006 char name[IFNAMSIZ];
1008 if (if_indextoname(sll.sll_ifindex, name)) {
1009 ds_put_format(string, "(%s)", name);
1011 ds_put_format(string, "(ifindex=%d)", sll.sll_ifindex);
1014 if (sll.sll_protocol) {
1015 ds_put_format(string, "(protocol=0x%"PRIu16")",
1016 ntohs(sll.sll_protocol));
1020 else if (ss.ss_family == AF_UNSPEC) {
1021 ds_put_cstr(string, "AF_UNSPEC");
1023 ds_put_format(string, "AF_%d", (int) ss.ss_family);
1029 #ifdef LINUX_DATAPATH
1031 put_fd_filename(struct ds *string, int fd)
1037 linkname = xasprintf("/proc/self/fd/%d", fd);
1038 n = readlink(linkname, buf, sizeof buf);
1040 ds_put_char(string, ' ');
1041 ds_put_buffer(string, buf, n);
1042 if (n > sizeof buf) {
1043 ds_put_cstr(string, "...");
1050 /* Returns a malloc()'d string describing 'fd', for use in logging. */
1058 if (fstat(fd, &s)) {
1059 ds_put_format(&string, "fstat failed (%s)", strerror(errno));
1060 } else if (S_ISSOCK(s.st_mode)) {
1061 describe_sockaddr(&string, fd, getsockname);
1062 ds_put_cstr(&string, "<->");
1063 describe_sockaddr(&string, fd, getpeername);
1065 ds_put_cstr(&string, (isatty(fd) ? "tty"
1066 : S_ISDIR(s.st_mode) ? "directory"
1067 : S_ISCHR(s.st_mode) ? "character device"
1068 : S_ISBLK(s.st_mode) ? "block device"
1069 : S_ISREG(s.st_mode) ? "file"
1070 : S_ISFIFO(s.st_mode) ? "FIFO"
1071 : S_ISLNK(s.st_mode) ? "symbolic link"
1073 #ifdef LINUX_DATAPATH
1074 put_fd_filename(&string, fd);
1077 return ds_steal_cstr(&string);
1080 /* Returns the total of the 'iov_len' members of the 'n_iovs' in 'iovs'.
1081 * The caller must ensure that the total does not exceed SIZE_MAX. */
1083 iovec_len(const struct iovec iovs[], size_t n_iovs)
1088 for (i = 0; i < n_iovs; i++) {
1089 len += iovs[i].iov_len;
1094 /* Returns true if all of the 'n_iovs' iovecs in 'iovs' have length zero. */
1096 iovec_is_empty(const struct iovec iovs[], size_t n_iovs)
1100 for (i = 0; i < n_iovs; i++) {
1101 if (iovs[i].iov_len) {
1108 /* Sends the 'n_iovs' iovecs of data in 'iovs' and the 'n_fds' file descriptors
1109 * in 'fds' on Unix domain socket 'sock'. Returns the number of bytes
1110 * successfully sent or -1 if an error occurred. On error, sets errno
1113 send_iovec_and_fds(int sock,
1114 const struct iovec *iovs, size_t n_iovs,
1115 const int fds[], size_t n_fds)
1121 char control[CMSG_SPACE(SOUTIL_MAX_FDS * sizeof *fds)];
1125 assert(!iovec_is_empty(iovs, n_iovs));
1126 assert(n_fds <= SOUTIL_MAX_FDS);
1128 memset(&cmsg, 0, sizeof cmsg);
1129 cmsg.cm.cmsg_len = CMSG_LEN(n_fds * sizeof *fds);
1130 cmsg.cm.cmsg_level = SOL_SOCKET;
1131 cmsg.cm.cmsg_type = SCM_RIGHTS;
1132 memcpy(CMSG_DATA(&cmsg.cm), fds, n_fds * sizeof *fds);
1134 msg.msg_name = NULL;
1135 msg.msg_namelen = 0;
1136 msg.msg_iov = (struct iovec *) iovs;
1137 msg.msg_iovlen = n_iovs;
1138 msg.msg_control = &cmsg.cm;
1139 msg.msg_controllen = CMSG_SPACE(n_fds * sizeof *fds);
1142 return sendmsg(sock, &msg, 0);
1144 return writev(sock, iovs, n_iovs);
1148 /* Sends the 'n_iovs' iovecs of data in 'iovs' and the 'n_fds' file descriptors
1149 * in 'fds' on Unix domain socket 'sock'. If 'skip_bytes' is nonzero, then the
1150 * first 'skip_bytes' of data in the iovecs are not sent, and none of the file
1151 * descriptors are sent. The function continues to retry sending until an
1152 * error (other than EINTR) occurs or all the data and fds are sent.
1154 * Returns 0 if all the data and fds were successfully sent, otherwise a
1155 * positive errno value. Regardless of success, stores the number of bytes
1156 * sent (always at least 'skip_bytes') in '*bytes_sent'. (If at least one byte
1157 * is sent, then all the fds have been sent.)
1159 * 'skip_bytes' must be less than or equal to iovec_len(iovs, n_iovs). */
1161 send_iovec_and_fds_fully(int sock,
1162 const struct iovec iovs[], size_t n_iovs,
1163 const int fds[], size_t n_fds,
1164 size_t skip_bytes, size_t *bytes_sent)
1167 while (n_iovs > 0) {
1171 retval = skip_bytes;
1173 } else if (!*bytes_sent) {
1174 retval = send_iovec_and_fds(sock, iovs, n_iovs, fds, n_fds);
1176 retval = writev(sock, iovs, n_iovs);
1180 *bytes_sent += retval;
1181 while (retval > 0) {
1182 const uint8_t *base = iovs->iov_base;
1183 size_t len = iovs->iov_len;
1189 error = write_fully(sock, base + retval, len - retval,
1191 *bytes_sent += sent;
1201 } else if (retval == 0) {
1202 if (iovec_is_empty(iovs, n_iovs)) {
1205 VLOG_WARN("send returned 0");
1207 } else if (errno != EINTR) {
1215 /* Sends the 'n_iovs' iovecs of data in 'iovs' and the 'n_fds' file descriptors
1216 * in 'fds' on Unix domain socket 'sock'. The function continues to retry
1217 * sending until an error (other than EAGAIN or EINTR) occurs or all the data
1218 * and fds are sent. Upon EAGAIN, the function blocks until the socket is
1219 * ready for more data.
1221 * Returns 0 if all the data and fds were successfully sent, otherwise a
1222 * positive errno value. */
1224 send_iovec_and_fds_fully_block(int sock,
1225 const struct iovec iovs[], size_t n_iovs,
1226 const int fds[], size_t n_fds)
1233 error = send_iovec_and_fds_fully(sock, iovs, n_iovs,
1234 fds, n_fds, sent, &sent);
1235 if (error != EAGAIN) {
1238 poll_fd_wait(sock, POLLOUT);
1243 /* Attempts to receive from Unix domain socket 'sock' up to 'size' bytes of
1244 * data into 'data' and up to SOUTIL_MAX_FDS file descriptors into 'fds'.
1246 * - Upon success, returns the number of bytes of data copied into 'data'
1247 * and stores the number of received file descriptors into '*n_fdsp'.
1249 * - On failure, returns a negative errno value and stores 0 in
1252 * - On EOF, returns 0 and stores 0 in '*n_fdsp'. */
1254 recv_data_and_fds(int sock,
1255 void *data, size_t size,
1256 int fds[SOUTIL_MAX_FDS], size_t *n_fdsp)
1260 char control[CMSG_SPACE(SOUTIL_MAX_FDS * sizeof *fds)];
1272 iov.iov_base = data;
1275 msg.msg_name = NULL;
1276 msg.msg_namelen = 0;
1279 msg.msg_control = &cmsg.cm;
1280 msg.msg_controllen = sizeof cmsg.control;
1283 retval = recvmsg(sock, &msg, 0);
1284 } while (retval < 0 && errno == EINTR);
1286 return retval < 0 ? -errno : 0;
1289 for (p = CMSG_FIRSTHDR(&msg); p; p = CMSG_NXTHDR(&msg, p)) {
1290 if (p->cmsg_level != SOL_SOCKET || p->cmsg_type != SCM_RIGHTS) {
1291 VLOG_ERR("unexpected control message %d:%d",
1292 p->cmsg_level, p->cmsg_type);
1294 } else if (*n_fdsp) {
1295 VLOG_ERR("multiple SCM_RIGHTS received");
1298 size_t n_fds = (p->cmsg_len - CMSG_LEN(0)) / sizeof *fds;
1299 const int *fds_data = (const int *) CMSG_DATA(p);
1302 if (n_fds > SOUTIL_MAX_FDS) {
1303 VLOG_ERR("%zu fds received but only %d supported",
1304 n_fds, SOUTIL_MAX_FDS);
1305 for (i = 0; i < n_fds; i++) {
1312 memcpy(fds, fds_data, n_fds * sizeof *fds);
1319 for (i = 0; i < *n_fdsp; i++) {