2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
18 #include "socket-util.h"
19 #include <arpa/inet.h>
29 #include <sys/ioctl.h>
30 #include <sys/resource.h>
31 #include <sys/socket.h>
36 #include "dynamic-string.h"
37 #include "fatal-signal.h"
38 #include "ovs-thread.h"
40 #include "poll-loop.h"
43 #if AF_PACKET && LINUX_DATAPATH
44 #include <linux/if_packet.h>
47 #include "netlink-protocol.h"
48 #include "netlink-socket.h"
51 VLOG_DEFINE_THIS_MODULE(socket_util);
53 /* #ifdefs make it a pain to maintain code: you have to try to build both ways.
54 * Thus, this file compiles all of the code regardless of the target, by
55 * writing "if (LINUX_DATAPATH)" instead of "#ifdef __linux__". */
56 #ifndef LINUX_DATAPATH
57 #define LINUX_DATAPATH 0
64 /* Maximum length of the sun_path member in a struct sockaddr_un, excluding
65 * space for a null terminator. */
66 #define MAX_UN_LEN (sizeof(((struct sockaddr_un *) 0)->sun_path) - 1)
68 static int getsockopt_int(int fd, int level, int option, const char *optname,
71 /* Sets 'fd' to non-blocking mode. Returns 0 if successful, otherwise a
72 * positive errno value. */
74 set_nonblocking(int fd)
77 int flags = fcntl(fd, F_GETFL, 0);
79 if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) != -1) {
82 VLOG_ERR("fcntl(F_SETFL) failed: %s", ovs_strerror(errno));
86 VLOG_ERR("fcntl(F_GETFL) failed: %s", ovs_strerror(errno));
90 unsigned long arg = 1;
91 if (ioctlsocket(fd, FIONBIO, &arg)) {
92 int error = sock_errno();
93 VLOG_ERR("set_nonblocking failed: %s", sock_strerror(error));
101 xset_nonblocking(int fd)
103 if (set_nonblocking(fd)) {
109 set_dscp(int fd, uint8_t dscp)
118 if (setsockopt(fd, IPPROTO_IP, IP_TOS, &val, sizeof val)) {
126 rlim_is_finite(rlim_t limit)
128 if (limit == RLIM_INFINITY) {
132 #ifdef RLIM_SAVED_CUR /* FreeBSD 8.0 lacks RLIM_SAVED_CUR. */
133 if (limit == RLIM_SAVED_CUR) {
138 #ifdef RLIM_SAVED_MAX /* FreeBSD 8.0 lacks RLIM_SAVED_MAX. */
139 if (limit == RLIM_SAVED_MAX) {
147 /* Returns the maximum valid FD value, plus 1. */
151 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
154 if (ovsthread_once_start(&once)) {
156 if (!getrlimit(RLIMIT_NOFILE, &r) && rlim_is_finite(r.rlim_cur)) {
157 max_fds = r.rlim_cur;
159 VLOG_WARN("failed to obtain fd limit, defaulting to 1024");
162 ovsthread_once_done(&once);
168 /* Translates 'host_name', which must be a string representation of an IP
169 * address, into a numeric IP address in '*addr'. Returns 0 if successful,
170 * otherwise a positive errno value. */
172 lookup_ip(const char *host_name, struct in_addr *addr)
174 if (!inet_aton(host_name, addr)) {
175 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
176 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name);
182 /* Translates 'host_name', which must be a string representation of an IPv6
183 * address, into a numeric IPv6 address in '*addr'. Returns 0 if successful,
184 * otherwise a positive errno value. */
186 lookup_ipv6(const char *host_name, struct in6_addr *addr)
188 if (inet_pton(AF_INET6, host_name, addr) != 1) {
189 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
190 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IPv6 address", host_name);
196 /* Translates 'host_name', which must be a host name or a string representation
197 * of an IP address, into a numeric IP address in '*addr'. Returns 0 if
198 * successful, otherwise a positive errno value.
200 * Most Open vSwitch code should not use this because it causes deadlocks:
201 * getaddrinfo() sends out a DNS request but that starts a new flow for which
202 * OVS must set up a flow, but it can't because it's waiting for a DNS reply.
203 * The synchronous lookup also delays other activity. (Of course we can solve
204 * this but it doesn't seem worthwhile quite yet.) */
206 lookup_hostname(const char *host_name, struct in_addr *addr)
208 struct addrinfo *result;
209 struct addrinfo hints;
211 if (inet_aton(host_name, addr)) {
215 memset(&hints, 0, sizeof hints);
216 hints.ai_family = AF_INET;
218 switch (getaddrinfo(host_name, NULL, &hints, &result)) {
220 *addr = ALIGNED_CAST(struct sockaddr_in *,
221 result->ai_addr)->sin_addr;
222 freeaddrinfo(result);
225 #ifdef EAI_ADDRFAMILY
246 #if defined (EAI_NODATA) && EAI_NODATA != EAI_NONAME
262 check_connection_completion(int fd)
264 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
269 pfd.events = POLLOUT;
271 retval = poll(&pfd, 1, 0);
272 } while (retval < 0 && errno == EINTR);
274 if (pfd.revents & POLLERR) {
275 ssize_t n = send(fd, "", 1, MSG_DONTWAIT);
279 VLOG_ERR_RL(&rl, "poll return POLLERR but send succeeded");
284 } else if (retval < 0) {
285 VLOG_ERR_RL(&rl, "poll: %s", ovs_strerror(errno));
292 /* Drain all the data currently in the receive queue of a datagram socket (and
293 * possibly additional data). There is no way to know how many packets are in
294 * the receive queue, but we do know that the total number of bytes queued does
295 * not exceed the receive buffer size, so we pull packets until none are left
296 * or we've read that many bytes. */
302 rcvbuf = get_socket_rcvbuf(fd);
308 /* In Linux, specifying MSG_TRUNC in the flags argument causes the
309 * datagram length to be returned, even if that is longer than the
310 * buffer provided. Thus, we can use a 1-byte buffer to discard the
311 * incoming datagram and still be able to account how many bytes were
312 * removed from the receive buffer.
314 * On other Unix-like OSes, MSG_TRUNC has no effect in the flags
316 char buffer[LINUX_DATAPATH ? 1 : 2048];
317 ssize_t n_bytes = recv(fd, buffer, sizeof buffer,
318 MSG_TRUNC | MSG_DONTWAIT);
319 if (n_bytes <= 0 || n_bytes >= rcvbuf) {
327 /* Returns the size of socket 'sock''s receive buffer (SO_RCVBUF), or a
328 * negative errno value if an error occurs. */
330 get_socket_rcvbuf(int sock)
335 error = getsockopt_int(sock, SOL_SOCKET, SO_RCVBUF, "SO_RCVBUF", &rcvbuf);
336 return error ? -error : rcvbuf;
339 /* Reads and discards up to 'n' datagrams from 'fd', stopping as soon as no
340 * more data can be immediately read. ('fd' should therefore be in
341 * non-blocking mode.)*/
343 drain_fd(int fd, size_t n_packets)
345 for (; n_packets > 0; n_packets--) {
346 /* 'buffer' only needs to be 1 byte long in most circumstances. This
347 * size is defensive against the possibility that we someday want to
348 * use a Linux tap device without TUN_NO_PI, in which case a buffer
349 * smaller than sizeof(struct tun_pi) will give EINVAL on read. */
351 if (read(fd, buffer, sizeof buffer) <= 0) {
358 /* Attempts to shorten 'name' by opening a file descriptor for the directory
359 * part of the name and indirecting through /proc/self/fd/<dirfd>/<basename>.
360 * On systems with Linux-like /proc, this works as long as <basename> isn't too
363 * On success, returns 0 and stores the short name in 'short_name' and a
364 * directory file descriptor to eventually be closed in '*dirfpd'. */
366 shorten_name_via_proc(const char *name, char short_name[MAX_UN_LEN + 1],
373 if (!LINUX_DATAPATH) {
377 dir = dir_name(name);
378 dirfd = open(dir, O_DIRECTORY | O_RDONLY);
380 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
383 VLOG_WARN_RL(&rl, "%s: open failed (%s)", dir, ovs_strerror(error));
390 base = base_name(name);
391 len = snprintf(short_name, MAX_UN_LEN + 1,
392 "/proc/self/fd/%d/%s", dirfd, base);
395 if (len >= 0 && len <= MAX_UN_LEN) {
404 /* Attempts to shorten 'name' by creating a symlink for the directory part of
405 * the name and indirecting through <symlink>/<basename>. This works on
406 * systems that support symlinks, as long as <basename> isn't too long.
408 * On success, returns 0 and stores the short name in 'short_name' and the
409 * symbolic link to eventually delete in 'linkname'. */
411 shorten_name_via_symlink(const char *name, char short_name[MAX_UN_LEN + 1],
412 char linkname[MAX_UN_LEN + 1])
414 char *abs, *dir, *base;
419 abs = abs_file_name(NULL, name);
421 base = base_name(abs);
424 tmpdir = getenv("TMPDIR");
425 if (tmpdir == NULL) {
429 for (i = 0; i < 1000; i++) {
432 len = snprintf(linkname, MAX_UN_LEN + 1,
433 "%s/ovs-un-c-%"PRIu32, tmpdir, random_uint32());
434 error = (len < 0 || len > MAX_UN_LEN ? ENAMETOOLONG
435 : symlink(dir, linkname) ? errno
437 if (error != EEXIST) {
445 fatal_signal_add_file_to_unlink(linkname);
447 len = snprintf(short_name, MAX_UN_LEN + 1, "%s/%s", linkname, base);
448 if (len < 0 || len > MAX_UN_LEN) {
449 fatal_signal_unlink_file_now(linkname);
450 error = ENAMETOOLONG;
463 /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in
464 * '*un_len' the size of the sockaddr_un.
466 * Returns 0 on success, otherwise a positive errno value.
468 * Uses '*dirfdp' and 'linkname' to store references to data when the caller no
469 * longer needs to use 'un'. On success, freeing these references with
470 * free_sockaddr_un() is mandatory to avoid a leak; on failure, freeing them is
471 * unnecessary but harmless. */
473 make_sockaddr_un(const char *name, struct sockaddr_un *un, socklen_t *un_len,
474 int *dirfdp, char linkname[MAX_UN_LEN + 1])
476 char short_name[MAX_UN_LEN + 1];
480 if (strlen(name) > MAX_UN_LEN) {
481 /* 'name' is too long to fit in a sockaddr_un. Try a workaround. */
482 int error = shorten_name_via_proc(name, short_name, dirfdp);
483 if (error == ENAMETOOLONG) {
484 error = shorten_name_via_symlink(name, short_name, linkname);
487 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
489 VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum "
490 "%"PRIuSIZE" bytes", name, MAX_UN_LEN);
497 un->sun_family = AF_UNIX;
498 ovs_strzcpy(un->sun_path, name, sizeof un->sun_path);
499 *un_len = (offsetof(struct sockaddr_un, sun_path)
500 + strlen (un->sun_path) + 1);
504 /* Clean up after make_sockaddr_un(). */
506 free_sockaddr_un(int dirfd, const char *linkname)
512 fatal_signal_unlink_file_now(linkname);
516 /* Binds Unix domain socket 'fd' to a file with permissions 0700. */
518 bind_unix_socket(int fd, struct sockaddr *sun, socklen_t sun_len)
520 /* According to _Unix Network Programming_, umask should affect bind(). */
521 mode_t old_umask = umask(0077);
522 int error = bind(fd, sun, sun_len) ? errno : 0;
527 /* Creates a Unix domain socket in the given 'style' (either SOCK_DGRAM or
528 * SOCK_STREAM) that is bound to '*bind_path' (if 'bind_path' is non-null) and
529 * connected to '*connect_path' (if 'connect_path' is non-null). If 'nonblock'
530 * is true, the socket is made non-blocking.
532 * Returns the socket's fd if successful, otherwise a negative errno value. */
534 make_unix_socket(int style, bool nonblock,
535 const char *bind_path, const char *connect_path)
540 fd = socket(PF_UNIX, style, 0);
545 /* Set nonblocking mode right away, if we want it. This prevents blocking
546 * in connect(), if connect_path != NULL. (In turn, that's a corner case:
547 * it will only happen if style is SOCK_STREAM or SOCK_SEQPACKET, and only
548 * if a backlog of un-accepted connections has built up in the kernel.) */
550 error = set_nonblocking(fd);
557 char linkname[MAX_UN_LEN + 1];
558 struct sockaddr_un un;
562 if (unlink(bind_path) && errno != ENOENT) {
563 VLOG_WARN("unlinking \"%s\": %s\n",
564 bind_path, ovs_strerror(errno));
566 fatal_signal_add_file_to_unlink(bind_path);
568 error = make_sockaddr_un(bind_path, &un, &un_len, &dirfd, linkname);
570 error = bind_unix_socket(fd, (struct sockaddr *) &un, un_len);
572 free_sockaddr_un(dirfd, linkname);
580 char linkname[MAX_UN_LEN + 1];
581 struct sockaddr_un un;
585 error = make_sockaddr_un(connect_path, &un, &un_len, &dirfd, linkname);
587 && connect(fd, (struct sockaddr*) &un, un_len)
588 && errno != EINPROGRESS) {
591 free_sockaddr_un(dirfd, linkname);
601 if (error == EAGAIN) {
605 fatal_signal_unlink_file_now(bind_path);
612 get_unix_name_len(socklen_t sun_len)
614 return (sun_len >= offsetof(struct sockaddr_un, sun_path)
615 ? sun_len - offsetof(struct sockaddr_un, sun_path)
621 guess_netmask(ovs_be32 ip_)
623 uint32_t ip = ntohl(ip_);
624 return ((ip >> 31) == 0 ? htonl(0xff000000) /* Class A */
625 : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */
626 : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */
627 : htonl(0)); /* ??? */
630 /* This is like strsep() except:
632 * - The separator string is ":".
634 * - Square brackets [] quote ":" separators and are removed from the
637 parse_bracketed_token(char **pp)
643 } else if (*p == '\0') {
646 } else if (*p == '[') {
648 char *end = start + strcspn(start, "]");
649 *pp = (*end == '\0' ? NULL
650 : end[1] == ':' ? end + 2
656 char *end = start + strcspn(start, ":");
657 *pp = *end == '\0' ? NULL : end + 1;
664 parse_sockaddr_components(struct sockaddr_storage *ss,
666 const char *port_s, uint16_t default_port,
669 struct sockaddr_in *sin = ALIGNED_CAST(struct sockaddr_in *, ss);
672 if (port_s && port_s[0]) {
673 if (!str_to_int(port_s, 10, &port) || port < 0 || port > 65535) {
674 VLOG_ERR("%s: bad port number \"%s\"", s, port_s);
680 memset(ss, 0, sizeof *ss);
681 if (strchr(host_s, ':')) {
682 struct sockaddr_in6 *sin6
683 = ALIGNED_CAST(struct sockaddr_in6 *, ss);
685 sin6->sin6_family = AF_INET6;
686 sin6->sin6_port = htons(port);
687 if (!inet_pton(AF_INET6, host_s, sin6->sin6_addr.s6_addr)) {
688 VLOG_ERR("%s: bad IPv6 address \"%s\"", s, host_s);
692 sin->sin_family = AF_INET;
693 sin->sin_port = htons(port);
694 if (!inet_pton(AF_INET, host_s, &sin->sin_addr.s_addr)) {
695 VLOG_ERR("%s: bad IPv4 address \"%s\"", s, host_s);
703 memset(ss, 0, sizeof *ss);
707 /* Parses 'target', which should be a string in the format "<host>[:<port>]".
708 * <host>, which is required, may be an IPv4 address or an IPv6 address
709 * enclosed in square brackets. If 'default_port' is nonzero then <port> is
710 * optional and defaults to 'default_port'.
712 * On success, returns true and stores the parsed remote address into '*ss'.
713 * On failure, logs an error, stores zeros into '*ss', and returns false. */
715 inet_parse_active(const char *target_, uint16_t default_port,
716 struct sockaddr_storage *ss)
718 char *target = xstrdup(target_);
725 host = parse_bracketed_token(&p);
726 port = parse_bracketed_token(&p);
728 VLOG_ERR("%s: host must be specified", target_);
730 } else if (!port && !default_port) {
731 VLOG_ERR("%s: port must be specified", target_);
734 ok = parse_sockaddr_components(ss, host, port, default_port, target_);
737 memset(ss, 0, sizeof *ss);
744 /* Opens a non-blocking IPv4 or IPv6 socket of the specified 'style' and
745 * connects to 'target', which should be a string in the format
746 * "<host>[:<port>]". <host>, which is required, may be an IPv4 address or an
747 * IPv6 address enclosed in square brackets. If 'default_port' is nonzero then
748 * <port> is optional and defaults to 'default_port'.
750 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
752 * On success, returns 0 (indicating connection complete) or EAGAIN (indicating
753 * connection in progress), in which case the new file descriptor is stored
754 * into '*fdp'. On failure, returns a positive errno value other than EAGAIN
755 * and stores -1 into '*fdp'.
757 * If 'ss' is non-null, then on success stores the target address into '*ss'.
759 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
760 * should be in the range [0, 63] and will automatically be shifted to the
761 * appropriately place in the IP tos field. */
763 inet_open_active(int style, const char *target, uint16_t default_port,
764 struct sockaddr_storage *ssp, int *fdp, uint8_t dscp)
766 struct sockaddr_storage ss;
771 if (!inet_parse_active(target, default_port, &ss)) {
772 error = EAFNOSUPPORT;
776 /* Create non-blocking socket. */
777 fd = socket(ss.ss_family, style, 0);
779 error = sock_errno();
780 VLOG_ERR("%s: socket: %s", target, sock_strerror(error));
783 error = set_nonblocking(fd);
788 /* The dscp bits must be configured before connect() to ensure that the
789 * TOS field is set during the connection establishment. If set after
790 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
791 error = set_dscp(fd, dscp);
793 VLOG_ERR("%s: socket: %s", target, sock_strerror(error));
798 error = connect(fd, (struct sockaddr *) &ss, ss_length(&ss)) == 0
801 if (error == EINPROGRESS
803 || error == WSAEALREADY || error == WSAEWOULDBLOCK
810 if (error && error != EAGAIN) {
812 memset(ssp, 0, sizeof *ssp);
827 /* Parses 'target', which should be a string in the format "[<port>][:<host>]":
829 * - If 'default_port' is -1, then <port> is required. Otherwise, if
830 * <port> is omitted, then 'default_port' is used instead.
832 * - If <port> (or 'default_port', if used) is 0, then no port is bound
833 * and the TCP/IP stack will select a port.
835 * - <host> is optional. If supplied, it may be an IPv4 address or an
836 * IPv6 address enclosed in square brackets. If omitted, the IP address
839 * If successful, stores the address into '*ss' and returns true; otherwise
840 * zeros '*ss' and returns false. */
842 inet_parse_passive(const char *target_, int default_port,
843 struct sockaddr_storage *ss)
845 char *target = xstrdup(target_);
852 port = parse_bracketed_token(&p);
853 host = parse_bracketed_token(&p);
854 if (!port && default_port < 0) {
855 VLOG_ERR("%s: port must be specified", target_);
858 ok = parse_sockaddr_components(ss, host ? host : "0.0.0.0",
859 port, default_port, target_);
862 memset(ss, 0, sizeof *ss);
869 /* Opens a non-blocking IPv4 or IPv6 socket of the specified 'style', binds to
870 * 'target', and listens for incoming connections. Parses 'target' in the same
871 * way was inet_parse_passive().
873 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
875 * For TCP, the socket will have SO_REUSEADDR turned on.
877 * On success, returns a non-negative file descriptor. On failure, returns a
878 * negative errno value.
880 * If 'ss' is non-null, then on success stores the bound address into '*ss'.
882 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
883 * should be in the range [0, 63] and will automatically be shifted to the
884 * appropriately place in the IP tos field. */
886 inet_open_passive(int style, const char *target, int default_port,
887 struct sockaddr_storage *ssp, uint8_t dscp)
889 bool kernel_chooses_port;
890 struct sockaddr_storage ss;
892 unsigned int yes = 1;
894 if (!inet_parse_passive(target, default_port, &ss)) {
895 return -EAFNOSUPPORT;
897 kernel_chooses_port = ss_get_port(&ss) == 0;
899 /* Create non-blocking socket, set SO_REUSEADDR. */
900 fd = socket(ss.ss_family, style, 0);
902 error = sock_errno();
903 VLOG_ERR("%s: socket: %s", target, sock_strerror(error));
906 error = set_nonblocking(fd);
910 if (style == SOCK_STREAM
911 && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) {
912 error = sock_errno();
913 VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s",
914 target, sock_strerror(error));
919 if (bind(fd, (struct sockaddr *) &ss, ss_length(&ss)) < 0) {
920 error = sock_errno();
921 VLOG_ERR("%s: bind: %s", target, sock_strerror(error));
925 /* The dscp bits must be configured before connect() to ensure that the TOS
926 * field is set during the connection establishment. If set after
927 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
928 error = set_dscp(fd, dscp);
930 VLOG_ERR("%s: socket: %s", target, sock_strerror(error));
935 if (style == SOCK_STREAM && listen(fd, 10) < 0) {
936 error = sock_errno();
937 VLOG_ERR("%s: listen: %s", target, sock_strerror(error));
941 if (ssp || kernel_chooses_port) {
942 socklen_t ss_len = sizeof ss;
943 if (getsockname(fd, (struct sockaddr *) &ss, &ss_len) < 0) {
944 error = sock_errno();
945 VLOG_ERR("%s: getsockname: %s", target, sock_strerror(error));
948 if (kernel_chooses_port) {
949 VLOG_INFO("%s: listening on port %"PRIu16,
950 target, ss_get_port(&ss));
961 memset(ssp, 0, sizeof *ssp);
967 /* Returns a readable and writable fd for /dev/null, if successful, otherwise
968 * a negative errno value. The caller must not close the returned fd (because
969 * the same fd will be handed out to subsequent callers). */
973 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
976 if (ovsthread_once_start(&once)) {
977 null_fd = open("/dev/null", O_RDWR);
980 VLOG_ERR("could not open /dev/null: %s", ovs_strerror(error));
983 ovsthread_once_done(&once);
990 read_fully(int fd, void *p_, size_t size, size_t *bytes_read)
996 ssize_t retval = read(fd, p, size);
998 *bytes_read += retval;
1001 } else if (retval == 0) {
1003 } else if (errno != EINTR) {
1011 write_fully(int fd, const void *p_, size_t size, size_t *bytes_written)
1013 const uint8_t *p = p_;
1017 ssize_t retval = write(fd, p, size);
1019 *bytes_written += retval;
1022 } else if (retval == 0) {
1023 VLOG_WARN("write returned 0");
1025 } else if (errno != EINTR) {
1032 /* Given file name 'file_name', fsyncs the directory in which it is contained.
1033 * Returns 0 if successful, otherwise a positive errno value. */
1035 fsync_parent_dir(const char *file_name)
1042 dir = dir_name(file_name);
1043 fd = open(dir, O_RDONLY);
1046 if (errno == EINVAL || errno == EROFS) {
1047 /* This directory does not support synchronization. Not
1048 * really an error. */
1051 VLOG_ERR("%s: fsync failed (%s)", dir, ovs_strerror(error));
1057 VLOG_ERR("%s: open failed (%s)", dir, ovs_strerror(error));
1065 /* Obtains the modification time of the file named 'file_name' to the greatest
1066 * supported precision. If successful, stores the mtime in '*mtime' and
1067 * returns 0. On error, returns a positive errno value and stores zeros in
1070 get_mtime(const char *file_name, struct timespec *mtime)
1074 if (!stat(file_name, &s)) {
1075 mtime->tv_sec = s.st_mtime;
1077 #if HAVE_STRUCT_STAT_ST_MTIM_TV_NSEC
1078 mtime->tv_nsec = s.st_mtim.tv_nsec;
1079 #elif HAVE_STRUCT_STAT_ST_MTIMENSEC
1080 mtime->tv_nsec = s.st_mtimensec;
1087 mtime->tv_sec = mtime->tv_nsec = 0;
1096 VLOG_FATAL("failed to create pipe (%s)", ovs_strerror(errno));
1101 xpipe_nonblocking(int fds[2])
1104 xset_nonblocking(fds[0]);
1105 xset_nonblocking(fds[1]);
1109 getsockopt_int(int fd, int level, int option, const char *optname, int *valuep)
1111 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
1117 if (getsockopt(fd, level, option, &value, &len)) {
1118 error = sock_errno();
1119 VLOG_ERR_RL(&rl, "getsockopt(%s): %s", optname, sock_strerror(error));
1120 } else if (len != sizeof value) {
1122 VLOG_ERR_RL(&rl, "getsockopt(%s): value is %u bytes (expected %"PRIuSIZE")",
1123 optname, (unsigned int) len, sizeof value);
1128 *valuep = error ? 0 : value;
1133 describe_sockaddr(struct ds *string, int fd,
1134 int (*getaddr)(int, struct sockaddr *, socklen_t *))
1136 struct sockaddr_storage ss;
1137 socklen_t len = sizeof ss;
1139 if (!getaddr(fd, (struct sockaddr *) &ss, &len)) {
1140 if (ss.ss_family == AF_INET || ss.ss_family == AF_INET6) {
1141 char addrbuf[SS_NTOP_BUFSIZE];
1143 ds_put_format(string, "%s:%"PRIu16,
1144 ss_format_address(&ss, addrbuf, sizeof addrbuf),
1147 } else if (ss.ss_family == AF_UNIX) {
1148 struct sockaddr_un sun;
1152 memcpy(&sun, &ss, sizeof sun);
1153 maxlen = len - offsetof(struct sockaddr_un, sun_path);
1154 null = memchr(sun.sun_path, '\0', maxlen);
1155 ds_put_buffer(string, sun.sun_path,
1156 null ? null - sun.sun_path : maxlen);
1160 else if (ss.ss_family == AF_NETLINK) {
1163 /* SO_PROTOCOL was introduced in 2.6.32. Support it regardless of the version
1164 * of the Linux kernel headers in use at build time. */
1166 #define SO_PROTOCOL 38
1169 if (!getsockopt_int(fd, SOL_SOCKET, SO_PROTOCOL, "SO_PROTOCOL",
1173 ds_put_cstr(string, "NETLINK_ROUTE");
1176 case NETLINK_GENERIC:
1177 ds_put_cstr(string, "NETLINK_GENERIC");
1181 ds_put_format(string, "AF_NETLINK family %d", protocol);
1185 ds_put_cstr(string, "AF_NETLINK");
1189 #if AF_PACKET && LINUX_DATAPATH
1190 else if (ss.ss_family == AF_PACKET) {
1191 struct sockaddr_ll sll;
1193 memcpy(&sll, &ss, sizeof sll);
1194 ds_put_cstr(string, "AF_PACKET");
1195 if (sll.sll_ifindex) {
1196 char name[IFNAMSIZ];
1198 if (if_indextoname(sll.sll_ifindex, name)) {
1199 ds_put_format(string, "(%s)", name);
1201 ds_put_format(string, "(ifindex=%d)", sll.sll_ifindex);
1204 if (sll.sll_protocol) {
1205 ds_put_format(string, "(protocol=0x%"PRIu16")",
1206 ntohs(sll.sll_protocol));
1210 else if (ss.ss_family == AF_UNSPEC) {
1211 ds_put_cstr(string, "AF_UNSPEC");
1213 ds_put_format(string, "AF_%d", (int) ss.ss_family);
1219 #ifdef LINUX_DATAPATH
1221 put_fd_filename(struct ds *string, int fd)
1227 linkname = xasprintf("/proc/self/fd/%d", fd);
1228 n = readlink(linkname, buf, sizeof buf);
1230 ds_put_char(string, ' ');
1231 ds_put_buffer(string, buf, n);
1232 if (n > sizeof buf) {
1233 ds_put_cstr(string, "...");
1240 /* Returns a malloc()'d string describing 'fd', for use in logging. */
1249 if (fstat(fd, &s)) {
1250 ds_put_format(&string, "fstat failed (%s)", ovs_strerror(errno));
1251 } else if (S_ISSOCK(s.st_mode)) {
1252 describe_sockaddr(&string, fd, getsockname);
1253 ds_put_cstr(&string, "<->");
1254 describe_sockaddr(&string, fd, getpeername);
1256 ds_put_cstr(&string, (isatty(fd) ? "tty"
1257 : S_ISDIR(s.st_mode) ? "directory"
1258 : S_ISCHR(s.st_mode) ? "character device"
1259 : S_ISBLK(s.st_mode) ? "block device"
1260 : S_ISREG(s.st_mode) ? "file"
1261 : S_ISFIFO(s.st_mode) ? "FIFO"
1262 : S_ISLNK(s.st_mode) ? "symbolic link"
1264 #ifdef LINUX_DATAPATH
1265 put_fd_filename(&string, fd);
1269 ds_put_format(&string,"file descriptor");
1271 return ds_steal_cstr(&string);
1275 /* Calls ioctl() on an AF_INET sock, passing the specified 'command' and
1276 * 'arg'. Returns 0 if successful, otherwise a positive errno value. */
1278 af_inet_ioctl(unsigned long int command, const void *arg)
1280 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
1283 if (ovsthread_once_start(&once)) {
1284 sock = socket(AF_INET, SOCK_DGRAM, 0);
1286 int error = sock_errno();
1287 VLOG_ERR("failed to create inet socket: %s", sock_strerror(error));
1290 ovsthread_once_done(&once);
1293 return (sock < 0 ? -sock
1294 : ioctl(sock, command, arg) == -1 ? errno
1299 af_inet_ifreq_ioctl(const char *name, struct ifreq *ifr, unsigned long int cmd,
1300 const char *cmd_name)
1304 ovs_strzcpy(ifr->ifr_name, name, sizeof ifr->ifr_name);
1305 error = af_inet_ioctl(cmd, ifr);
1307 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1308 VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name,
1309 ovs_strerror(error));
1315 /* sockaddr_storage helpers. */
1317 /* Returns the IPv4 or IPv6 port in 'ss'. */
1319 ss_get_port(const struct sockaddr_storage *ss)
1321 if (ss->ss_family == AF_INET) {
1322 const struct sockaddr_in *sin
1323 = ALIGNED_CAST(const struct sockaddr_in *, ss);
1324 return ntohs(sin->sin_port);
1325 } else if (ss->ss_family == AF_INET6) {
1326 const struct sockaddr_in6 *sin6
1327 = ALIGNED_CAST(const struct sockaddr_in6 *, ss);
1328 return ntohs(sin6->sin6_port);
1334 /* Formats the IPv4 or IPv6 address in 'ss' into the 'bufsize' bytes in 'buf'.
1335 * If 'ss' is an IPv6 address, puts square brackets around the address.
1336 * 'bufsize' should be at least SS_NTOP_BUFSIZE.
1340 ss_format_address(const struct sockaddr_storage *ss,
1341 char *buf, size_t bufsize)
1343 ovs_assert(bufsize >= SS_NTOP_BUFSIZE);
1344 if (ss->ss_family == AF_INET) {
1345 const struct sockaddr_in *sin
1346 = ALIGNED_CAST(const struct sockaddr_in *, ss);
1348 snprintf(buf, bufsize, IP_FMT, IP_ARGS(sin->sin_addr.s_addr));
1349 } else if (ss->ss_family == AF_INET6) {
1350 const struct sockaddr_in6 *sin6
1351 = ALIGNED_CAST(const struct sockaddr_in6 *, ss);
1354 inet_ntop(AF_INET6, sin6->sin6_addr.s6_addr, buf + 1, bufsize - 1);
1355 strcpy(strchr(buf, '\0'), "]");
1364 ss_length(const struct sockaddr_storage *ss)
1366 switch (ss->ss_family) {
1368 return sizeof(struct sockaddr_in);
1371 return sizeof(struct sockaddr_in6);
1378 /* For Windows socket calls, 'errno' is not set. One has to call
1379 * WSAGetLastError() to get the error number and then pass it to
1380 * this function to get the correct error string.
1382 * ovs_strerror() calls strerror_r() and would not get the correct error
1383 * string for Windows sockets, but is good for POSIX. */
1385 sock_strerror(int error)
1388 return ovs_format_message(error);
1390 return ovs_strerror(error);