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)
76 int flags = fcntl(fd, F_GETFL, 0);
78 if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) != -1) {
81 VLOG_ERR("fcntl(F_SETFL) failed: %s", ovs_strerror(errno));
85 VLOG_ERR("fcntl(F_GETFL) failed: %s", ovs_strerror(errno));
91 xset_nonblocking(int fd)
93 if (set_nonblocking(fd)) {
99 set_dscp(int fd, uint8_t dscp)
108 if (setsockopt(fd, IPPROTO_IP, IP_TOS, &val, sizeof val)) {
116 rlim_is_finite(rlim_t limit)
118 if (limit == RLIM_INFINITY) {
122 #ifdef RLIM_SAVED_CUR /* FreeBSD 8.0 lacks RLIM_SAVED_CUR. */
123 if (limit == RLIM_SAVED_CUR) {
128 #ifdef RLIM_SAVED_MAX /* FreeBSD 8.0 lacks RLIM_SAVED_MAX. */
129 if (limit == RLIM_SAVED_MAX) {
137 /* Returns the maximum valid FD value, plus 1. */
141 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
144 if (ovsthread_once_start(&once)) {
146 if (!getrlimit(RLIMIT_NOFILE, &r) && rlim_is_finite(r.rlim_cur)) {
147 max_fds = r.rlim_cur;
149 VLOG_WARN("failed to obtain fd limit, defaulting to 1024");
152 ovsthread_once_done(&once);
158 /* Translates 'host_name', which must be a string representation of an IP
159 * address, into a numeric IP address in '*addr'. Returns 0 if successful,
160 * otherwise a positive errno value. */
162 lookup_ip(const char *host_name, struct in_addr *addr)
164 if (!inet_aton(host_name, addr)) {
165 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
166 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name);
172 /* Translates 'host_name', which must be a string representation of an IPv6
173 * address, into a numeric IPv6 address in '*addr'. Returns 0 if successful,
174 * otherwise a positive errno value. */
176 lookup_ipv6(const char *host_name, struct in6_addr *addr)
178 if (inet_pton(AF_INET6, host_name, addr) != 1) {
179 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
180 VLOG_ERR_RL(&rl, "\"%s\" is not a valid IPv6 address", host_name);
186 /* Translates 'host_name', which must be a host name or a string representation
187 * of an IP address, into a numeric IP address in '*addr'. Returns 0 if
188 * successful, otherwise a positive errno value.
190 * Most Open vSwitch code should not use this because it causes deadlocks:
191 * getaddrinfo() sends out a DNS request but that starts a new flow for which
192 * OVS must set up a flow, but it can't because it's waiting for a DNS reply.
193 * The synchronous lookup also delays other activity. (Of course we can solve
194 * this but it doesn't seem worthwhile quite yet.) */
196 lookup_hostname(const char *host_name, struct in_addr *addr)
198 struct addrinfo *result;
199 struct addrinfo hints;
201 if (inet_aton(host_name, addr)) {
205 memset(&hints, 0, sizeof hints);
206 hints.ai_family = AF_INET;
208 switch (getaddrinfo(host_name, NULL, &hints, &result)) {
210 *addr = ALIGNED_CAST(struct sockaddr_in *,
211 result->ai_addr)->sin_addr;
212 freeaddrinfo(result);
215 #ifdef EAI_ADDRFAMILY
250 check_connection_completion(int fd)
252 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
257 pfd.events = POLLOUT;
259 retval = poll(&pfd, 1, 0);
260 } while (retval < 0 && errno == EINTR);
262 if (pfd.revents & POLLERR) {
263 ssize_t n = send(fd, "", 1, MSG_DONTWAIT);
267 VLOG_ERR_RL(&rl, "poll return POLLERR but send succeeded");
272 } else if (retval < 0) {
273 VLOG_ERR_RL(&rl, "poll: %s", ovs_strerror(errno));
280 /* Drain all the data currently in the receive queue of a datagram socket (and
281 * possibly additional data). There is no way to know how many packets are in
282 * the receive queue, but we do know that the total number of bytes queued does
283 * not exceed the receive buffer size, so we pull packets until none are left
284 * or we've read that many bytes. */
290 rcvbuf = get_socket_rcvbuf(fd);
296 /* In Linux, specifying MSG_TRUNC in the flags argument causes the
297 * datagram length to be returned, even if that is longer than the
298 * buffer provided. Thus, we can use a 1-byte buffer to discard the
299 * incoming datagram and still be able to account how many bytes were
300 * removed from the receive buffer.
302 * On other Unix-like OSes, MSG_TRUNC has no effect in the flags
304 char buffer[LINUX_DATAPATH ? 1 : 2048];
305 ssize_t n_bytes = recv(fd, buffer, sizeof buffer,
306 MSG_TRUNC | MSG_DONTWAIT);
307 if (n_bytes <= 0 || n_bytes >= rcvbuf) {
315 /* Returns the size of socket 'sock''s receive buffer (SO_RCVBUF), or a
316 * negative errno value if an error occurs. */
318 get_socket_rcvbuf(int sock)
323 error = getsockopt_int(sock, SOL_SOCKET, SO_RCVBUF, "SO_RCVBUF", &rcvbuf);
324 return error ? -error : rcvbuf;
327 /* Reads and discards up to 'n' datagrams from 'fd', stopping as soon as no
328 * more data can be immediately read. ('fd' should therefore be in
329 * non-blocking mode.)*/
331 drain_fd(int fd, size_t n_packets)
333 for (; n_packets > 0; n_packets--) {
334 /* 'buffer' only needs to be 1 byte long in most circumstances. This
335 * size is defensive against the possibility that we someday want to
336 * use a Linux tap device without TUN_NO_PI, in which case a buffer
337 * smaller than sizeof(struct tun_pi) will give EINVAL on read. */
339 if (read(fd, buffer, sizeof buffer) <= 0) {
345 /* Attempts to shorten 'name' by opening a file descriptor for the directory
346 * part of the name and indirecting through /proc/self/fd/<dirfd>/<basename>.
347 * On systems with Linux-like /proc, this works as long as <basename> isn't too
350 * On success, returns 0 and stores the short name in 'short_name' and a
351 * directory file descriptor to eventually be closed in '*dirfpd'. */
353 shorten_name_via_proc(const char *name, char short_name[MAX_UN_LEN + 1],
360 if (!LINUX_DATAPATH) {
364 dir = dir_name(name);
365 dirfd = open(dir, O_DIRECTORY | O_RDONLY);
367 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
370 VLOG_WARN_RL(&rl, "%s: open failed (%s)", dir, ovs_strerror(error));
377 base = base_name(name);
378 len = snprintf(short_name, MAX_UN_LEN + 1,
379 "/proc/self/fd/%d/%s", dirfd, base);
382 if (len >= 0 && len <= MAX_UN_LEN) {
391 /* Attempts to shorten 'name' by creating a symlink for the directory part of
392 * the name and indirecting through <symlink>/<basename>. This works on
393 * systems that support symlinks, as long as <basename> isn't too long.
395 * On success, returns 0 and stores the short name in 'short_name' and the
396 * symbolic link to eventually delete in 'linkname'. */
398 shorten_name_via_symlink(const char *name, char short_name[MAX_UN_LEN + 1],
399 char linkname[MAX_UN_LEN + 1])
401 char *abs, *dir, *base;
406 abs = abs_file_name(NULL, name);
408 base = base_name(abs);
411 tmpdir = getenv("TMPDIR");
412 if (tmpdir == NULL) {
416 for (i = 0; i < 1000; i++) {
419 len = snprintf(linkname, MAX_UN_LEN + 1,
420 "%s/ovs-un-c-%"PRIu32, tmpdir, random_uint32());
421 error = (len < 0 || len > MAX_UN_LEN ? ENAMETOOLONG
422 : symlink(dir, linkname) ? errno
424 if (error != EEXIST) {
432 fatal_signal_add_file_to_unlink(linkname);
434 len = snprintf(short_name, MAX_UN_LEN + 1, "%s/%s", linkname, base);
435 if (len < 0 || len > MAX_UN_LEN) {
436 fatal_signal_unlink_file_now(linkname);
437 error = ENAMETOOLONG;
450 /* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in
451 * '*un_len' the size of the sockaddr_un.
453 * Returns 0 on success, otherwise a positive errno value.
455 * Uses '*dirfdp' and 'linkname' to store references to data when the caller no
456 * longer needs to use 'un'. On success, freeing these references with
457 * free_sockaddr_un() is mandatory to avoid a leak; on failure, freeing them is
458 * unnecessary but harmless. */
460 make_sockaddr_un(const char *name, struct sockaddr_un *un, socklen_t *un_len,
461 int *dirfdp, char linkname[MAX_UN_LEN + 1])
463 char short_name[MAX_UN_LEN + 1];
467 if (strlen(name) > MAX_UN_LEN) {
468 /* 'name' is too long to fit in a sockaddr_un. Try a workaround. */
469 int error = shorten_name_via_proc(name, short_name, dirfdp);
470 if (error == ENAMETOOLONG) {
471 error = shorten_name_via_symlink(name, short_name, linkname);
474 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
476 VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum "
477 "%"PRIuSIZE" bytes", name, MAX_UN_LEN);
484 un->sun_family = AF_UNIX;
485 ovs_strzcpy(un->sun_path, name, sizeof un->sun_path);
486 *un_len = (offsetof(struct sockaddr_un, sun_path)
487 + strlen (un->sun_path) + 1);
491 /* Clean up after make_sockaddr_un(). */
493 free_sockaddr_un(int dirfd, const char *linkname)
499 fatal_signal_unlink_file_now(linkname);
503 /* Binds Unix domain socket 'fd' to a file with permissions 0700. */
505 bind_unix_socket(int fd, struct sockaddr *sun, socklen_t sun_len)
507 /* According to _Unix Network Programming_, umask should affect bind(). */
508 mode_t old_umask = umask(0077);
509 int error = bind(fd, sun, sun_len) ? errno : 0;
514 /* Creates a Unix domain socket in the given 'style' (either SOCK_DGRAM or
515 * SOCK_STREAM) that is bound to '*bind_path' (if 'bind_path' is non-null) and
516 * connected to '*connect_path' (if 'connect_path' is non-null). If 'nonblock'
517 * is true, the socket is made non-blocking.
519 * Returns the socket's fd if successful, otherwise a negative errno value. */
521 make_unix_socket(int style, bool nonblock,
522 const char *bind_path, const char *connect_path)
527 fd = socket(PF_UNIX, style, 0);
532 /* Set nonblocking mode right away, if we want it. This prevents blocking
533 * in connect(), if connect_path != NULL. (In turn, that's a corner case:
534 * it will only happen if style is SOCK_STREAM or SOCK_SEQPACKET, and only
535 * if a backlog of un-accepted connections has built up in the kernel.) */
537 error = set_nonblocking(fd);
544 char linkname[MAX_UN_LEN + 1];
545 struct sockaddr_un un;
549 if (unlink(bind_path) && errno != ENOENT) {
550 VLOG_WARN("unlinking \"%s\": %s\n",
551 bind_path, ovs_strerror(errno));
553 fatal_signal_add_file_to_unlink(bind_path);
555 error = make_sockaddr_un(bind_path, &un, &un_len, &dirfd, linkname);
557 error = bind_unix_socket(fd, (struct sockaddr *) &un, un_len);
559 free_sockaddr_un(dirfd, linkname);
567 char linkname[MAX_UN_LEN + 1];
568 struct sockaddr_un un;
572 error = make_sockaddr_un(connect_path, &un, &un_len, &dirfd, linkname);
574 && connect(fd, (struct sockaddr*) &un, un_len)
575 && errno != EINPROGRESS) {
578 free_sockaddr_un(dirfd, linkname);
588 if (error == EAGAIN) {
592 fatal_signal_unlink_file_now(bind_path);
599 get_unix_name_len(socklen_t sun_len)
601 return (sun_len >= offsetof(struct sockaddr_un, sun_path)
602 ? sun_len - offsetof(struct sockaddr_un, sun_path)
607 guess_netmask(ovs_be32 ip_)
609 uint32_t ip = ntohl(ip_);
610 return ((ip >> 31) == 0 ? htonl(0xff000000) /* Class A */
611 : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */
612 : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */
613 : htonl(0)); /* ??? */
616 /* This is like strsep() except:
618 * - The separator string is ":".
620 * - Square brackets [] quote ":" separators and are removed from the
623 parse_bracketed_token(char **pp)
629 } else if (*p == '\0') {
632 } else if (*p == '[') {
634 char *end = start + strcspn(start, "]");
635 *pp = (*end == '\0' ? NULL
636 : end[1] == ':' ? end + 2
642 char *end = start + strcspn(start, ":");
643 *pp = *end == '\0' ? NULL : end + 1;
650 parse_sockaddr_components(struct sockaddr_storage *ss,
652 const char *port_s, uint16_t default_port,
655 struct sockaddr_in *sin = ALIGNED_CAST(struct sockaddr_in *, ss);
658 if (port_s && port_s[0]) {
659 if (!str_to_int(port_s, 10, &port) || port < 0 || port > 65535) {
660 VLOG_ERR("%s: bad port number \"%s\"", s, port_s);
666 memset(ss, 0, sizeof *ss);
667 if (strchr(host_s, ':')) {
668 struct sockaddr_in6 *sin6
669 = ALIGNED_CAST(struct sockaddr_in6 *, ss);
671 sin6->sin6_family = AF_INET6;
672 sin6->sin6_port = htons(port);
673 if (!inet_pton(AF_INET6, host_s, sin6->sin6_addr.s6_addr)) {
674 VLOG_ERR("%s: bad IPv6 address \"%s\"", s, host_s);
678 sin->sin_family = AF_INET;
679 sin->sin_port = htons(port);
680 if (!inet_pton(AF_INET, host_s, &sin->sin_addr.s_addr)) {
681 VLOG_ERR("%s: bad IPv4 address \"%s\"", s, host_s);
689 memset(ss, 0, sizeof *ss);
693 /* Parses 'target', which should be a string in the format "<host>[:<port>]".
694 * <host>, which is required, may be an IPv4 address or an IPv6 address
695 * enclosed in square brackets. If 'default_port' is nonzero then <port> is
696 * optional and defaults to 'default_port'.
698 * On success, returns true and stores the parsed remote address into '*ss'.
699 * On failure, logs an error, stores zeros into '*ss', and returns false. */
701 inet_parse_active(const char *target_, uint16_t default_port,
702 struct sockaddr_storage *ss)
704 char *target = xstrdup(target_);
711 host = parse_bracketed_token(&p);
712 port = parse_bracketed_token(&p);
714 VLOG_ERR("%s: host must be specified", target_);
716 } else if (!port && !default_port) {
717 VLOG_ERR("%s: port must be specified", target_);
720 ok = parse_sockaddr_components(ss, host, port, default_port, target_);
723 memset(ss, 0, sizeof *ss);
730 /* Opens a non-blocking IPv4 or IPv6 socket of the specified 'style' and
731 * connects to 'target', which should be a string in the format
732 * "<host>[:<port>]". <host>, which is required, may be an IPv4 address or an
733 * IPv6 address enclosed in square brackets. If 'default_port' is nonzero then
734 * <port> is optional and defaults to 'default_port'.
736 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
738 * On success, returns 0 (indicating connection complete) or EAGAIN (indicating
739 * connection in progress), in which case the new file descriptor is stored
740 * into '*fdp'. On failure, returns a positive errno value other than EAGAIN
741 * and stores -1 into '*fdp'.
743 * If 'ss' is non-null, then on success stores the target address into '*ss'.
745 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
746 * should be in the range [0, 63] and will automatically be shifted to the
747 * appropriately place in the IP tos field. */
749 inet_open_active(int style, const char *target, uint16_t default_port,
750 struct sockaddr_storage *ssp, int *fdp, uint8_t dscp)
752 struct sockaddr_storage ss;
757 if (!inet_parse_active(target, default_port, &ss)) {
758 error = EAFNOSUPPORT;
762 /* Create non-blocking socket. */
763 fd = socket(ss.ss_family, style, 0);
765 VLOG_ERR("%s: socket: %s", target, ovs_strerror(errno));
769 error = set_nonblocking(fd);
774 /* The dscp bits must be configured before connect() to ensure that the
775 * TOS field is set during the connection establishment. If set after
776 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
777 error = set_dscp(fd, dscp);
779 VLOG_ERR("%s: socket: %s", target, ovs_strerror(error));
784 error = connect(fd, (struct sockaddr *) &ss, ss_length(&ss)) == 0
787 if (error == EINPROGRESS) {
792 if (error && error != EAGAIN) {
794 memset(ssp, 0, sizeof *ssp);
809 /* Parses 'target', which should be a string in the format "[<port>][:<host>]":
811 * - If 'default_port' is -1, then <port> is required. Otherwise, if
812 * <port> is omitted, then 'default_port' is used instead.
814 * - If <port> (or 'default_port', if used) is 0, then no port is bound
815 * and the TCP/IP stack will select a port.
817 * - <host> is optional. If supplied, it may be an IPv4 address or an
818 * IPv6 address enclosed in square brackets. If omitted, the IP address
821 * If successful, stores the address into '*ss' and returns true; otherwise
822 * zeros '*ss' and returns false. */
824 inet_parse_passive(const char *target_, int default_port,
825 struct sockaddr_storage *ss)
827 char *target = xstrdup(target_);
834 port = parse_bracketed_token(&p);
835 host = parse_bracketed_token(&p);
836 if (!port && default_port < 0) {
837 VLOG_ERR("%s: port must be specified", target_);
840 ok = parse_sockaddr_components(ss, host ? host : "0.0.0.0",
841 port, default_port, target_);
844 memset(ss, 0, sizeof *ss);
851 /* Opens a non-blocking IPv4 or IPv6 socket of the specified 'style', binds to
852 * 'target', and listens for incoming connections. Parses 'target' in the same
853 * way was inet_parse_passive().
855 * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
857 * For TCP, the socket will have SO_REUSEADDR turned on.
859 * On success, returns a non-negative file descriptor. On failure, returns a
860 * negative errno value.
862 * If 'ss' is non-null, then on success stores the bound address into '*ss'.
864 * 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
865 * should be in the range [0, 63] and will automatically be shifted to the
866 * appropriately place in the IP tos field. */
868 inet_open_passive(int style, const char *target, int default_port,
869 struct sockaddr_storage *ssp, uint8_t dscp)
871 bool kernel_chooses_port;
872 struct sockaddr_storage ss;
874 unsigned int yes = 1;
876 if (!inet_parse_passive(target, default_port, &ss)) {
877 return -EAFNOSUPPORT;
879 kernel_chooses_port = ss_get_port(&ss) == 0;
881 /* Create non-blocking socket, set SO_REUSEADDR. */
882 fd = socket(ss.ss_family, style, 0);
885 VLOG_ERR("%s: socket: %s", target, ovs_strerror(error));
888 error = set_nonblocking(fd);
892 if (style == SOCK_STREAM
893 && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) {
895 VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s",
896 target, ovs_strerror(error));
901 if (bind(fd, (struct sockaddr *) &ss, ss_length(&ss)) < 0) {
903 VLOG_ERR("%s: bind: %s", target, ovs_strerror(error));
907 /* The dscp bits must be configured before connect() to ensure that the TOS
908 * field is set during the connection establishment. If set after
909 * connect(), the handshake SYN frames will be sent with a TOS of 0. */
910 error = set_dscp(fd, dscp);
912 VLOG_ERR("%s: socket: %s", target, ovs_strerror(error));
917 if (style == SOCK_STREAM && listen(fd, 10) < 0) {
919 VLOG_ERR("%s: listen: %s", target, ovs_strerror(error));
923 if (ssp || kernel_chooses_port) {
924 socklen_t ss_len = sizeof ss;
925 if (getsockname(fd, (struct sockaddr *) &ss, &ss_len) < 0) {
927 VLOG_ERR("%s: getsockname: %s", target, ovs_strerror(error));
930 if (kernel_chooses_port) {
931 VLOG_INFO("%s: listening on port %"PRIu16,
932 target, ss_get_port(&ss));
943 memset(ssp, 0, sizeof *ssp);
949 /* Returns a readable and writable fd for /dev/null, if successful, otherwise
950 * a negative errno value. The caller must not close the returned fd (because
951 * the same fd will be handed out to subsequent callers). */
955 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
958 if (ovsthread_once_start(&once)) {
959 null_fd = open("/dev/null", O_RDWR);
962 VLOG_ERR("could not open /dev/null: %s", ovs_strerror(error));
965 ovsthread_once_done(&once);
972 read_fully(int fd, void *p_, size_t size, size_t *bytes_read)
978 ssize_t retval = read(fd, p, size);
980 *bytes_read += retval;
983 } else if (retval == 0) {
985 } else if (errno != EINTR) {
993 write_fully(int fd, const void *p_, size_t size, size_t *bytes_written)
995 const uint8_t *p = p_;
999 ssize_t retval = write(fd, p, size);
1001 *bytes_written += retval;
1004 } else if (retval == 0) {
1005 VLOG_WARN("write returned 0");
1007 } else if (errno != EINTR) {
1014 /* Given file name 'file_name', fsyncs the directory in which it is contained.
1015 * Returns 0 if successful, otherwise a positive errno value. */
1017 fsync_parent_dir(const char *file_name)
1023 dir = dir_name(file_name);
1024 fd = open(dir, O_RDONLY);
1027 if (errno == EINVAL || errno == EROFS) {
1028 /* This directory does not support synchronization. Not
1029 * really an error. */
1032 VLOG_ERR("%s: fsync failed (%s)", dir, ovs_strerror(error));
1038 VLOG_ERR("%s: open failed (%s)", dir, ovs_strerror(error));
1045 /* Obtains the modification time of the file named 'file_name' to the greatest
1046 * supported precision. If successful, stores the mtime in '*mtime' and
1047 * returns 0. On error, returns a positive errno value and stores zeros in
1050 get_mtime(const char *file_name, struct timespec *mtime)
1054 if (!stat(file_name, &s)) {
1055 mtime->tv_sec = s.st_mtime;
1057 #if HAVE_STRUCT_STAT_ST_MTIM_TV_NSEC
1058 mtime->tv_nsec = s.st_mtim.tv_nsec;
1059 #elif HAVE_STRUCT_STAT_ST_MTIMENSEC
1060 mtime->tv_nsec = s.st_mtimensec;
1067 mtime->tv_sec = mtime->tv_nsec = 0;
1076 VLOG_FATAL("failed to create pipe (%s)", ovs_strerror(errno));
1081 xpipe_nonblocking(int fds[2])
1084 xset_nonblocking(fds[0]);
1085 xset_nonblocking(fds[1]);
1089 getsockopt_int(int fd, int level, int option, const char *optname, int *valuep)
1091 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
1097 if (getsockopt(fd, level, option, &value, &len)) {
1099 VLOG_ERR_RL(&rl, "getsockopt(%s): %s", optname, ovs_strerror(error));
1100 } else if (len != sizeof value) {
1102 VLOG_ERR_RL(&rl, "getsockopt(%s): value is %u bytes (expected %"PRIuSIZE")",
1103 optname, (unsigned int) len, sizeof value);
1108 *valuep = error ? 0 : value;
1113 describe_sockaddr(struct ds *string, int fd,
1114 int (*getaddr)(int, struct sockaddr *, socklen_t *))
1116 struct sockaddr_storage ss;
1117 socklen_t len = sizeof ss;
1119 if (!getaddr(fd, (struct sockaddr *) &ss, &len)) {
1120 if (ss.ss_family == AF_INET || ss.ss_family == AF_INET6) {
1121 char addrbuf[SS_NTOP_BUFSIZE];
1123 ds_put_format(string, "%s:%"PRIu16,
1124 ss_format_address(&ss, addrbuf, sizeof addrbuf),
1126 } else if (ss.ss_family == AF_UNIX) {
1127 struct sockaddr_un sun;
1131 memcpy(&sun, &ss, sizeof sun);
1132 maxlen = len - offsetof(struct sockaddr_un, sun_path);
1133 null = memchr(sun.sun_path, '\0', maxlen);
1134 ds_put_buffer(string, sun.sun_path,
1135 null ? null - sun.sun_path : maxlen);
1138 else if (ss.ss_family == AF_NETLINK) {
1141 /* SO_PROTOCOL was introduced in 2.6.32. Support it regardless of the version
1142 * of the Linux kernel headers in use at build time. */
1144 #define SO_PROTOCOL 38
1147 if (!getsockopt_int(fd, SOL_SOCKET, SO_PROTOCOL, "SO_PROTOCOL",
1151 ds_put_cstr(string, "NETLINK_ROUTE");
1154 case NETLINK_GENERIC:
1155 ds_put_cstr(string, "NETLINK_GENERIC");
1159 ds_put_format(string, "AF_NETLINK family %d", protocol);
1163 ds_put_cstr(string, "AF_NETLINK");
1167 #if AF_PACKET && LINUX_DATAPATH
1168 else if (ss.ss_family == AF_PACKET) {
1169 struct sockaddr_ll sll;
1171 memcpy(&sll, &ss, sizeof sll);
1172 ds_put_cstr(string, "AF_PACKET");
1173 if (sll.sll_ifindex) {
1174 char name[IFNAMSIZ];
1176 if (if_indextoname(sll.sll_ifindex, name)) {
1177 ds_put_format(string, "(%s)", name);
1179 ds_put_format(string, "(ifindex=%d)", sll.sll_ifindex);
1182 if (sll.sll_protocol) {
1183 ds_put_format(string, "(protocol=0x%"PRIu16")",
1184 ntohs(sll.sll_protocol));
1188 else if (ss.ss_family == AF_UNSPEC) {
1189 ds_put_cstr(string, "AF_UNSPEC");
1191 ds_put_format(string, "AF_%d", (int) ss.ss_family);
1197 #ifdef LINUX_DATAPATH
1199 put_fd_filename(struct ds *string, int fd)
1205 linkname = xasprintf("/proc/self/fd/%d", fd);
1206 n = readlink(linkname, buf, sizeof buf);
1208 ds_put_char(string, ' ');
1209 ds_put_buffer(string, buf, n);
1210 if (n > sizeof buf) {
1211 ds_put_cstr(string, "...");
1218 /* Returns a malloc()'d string describing 'fd', for use in logging. */
1226 if (fstat(fd, &s)) {
1227 ds_put_format(&string, "fstat failed (%s)", ovs_strerror(errno));
1228 } else if (S_ISSOCK(s.st_mode)) {
1229 describe_sockaddr(&string, fd, getsockname);
1230 ds_put_cstr(&string, "<->");
1231 describe_sockaddr(&string, fd, getpeername);
1233 ds_put_cstr(&string, (isatty(fd) ? "tty"
1234 : S_ISDIR(s.st_mode) ? "directory"
1235 : S_ISCHR(s.st_mode) ? "character device"
1236 : S_ISBLK(s.st_mode) ? "block device"
1237 : S_ISREG(s.st_mode) ? "file"
1238 : S_ISFIFO(s.st_mode) ? "FIFO"
1239 : S_ISLNK(s.st_mode) ? "symbolic link"
1241 #ifdef LINUX_DATAPATH
1242 put_fd_filename(&string, fd);
1245 return ds_steal_cstr(&string);
1248 /* Calls ioctl() on an AF_INET sock, passing the specified 'command' and
1249 * 'arg'. Returns 0 if successful, otherwise a positive errno value. */
1251 af_inet_ioctl(unsigned long int command, const void *arg)
1253 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
1256 if (ovsthread_once_start(&once)) {
1257 sock = socket(AF_INET, SOCK_DGRAM, 0);
1260 VLOG_ERR("failed to create inet socket: %s", ovs_strerror(errno));
1262 ovsthread_once_done(&once);
1265 return (sock < 0 ? -sock
1266 : ioctl(sock, command, arg) == -1 ? errno
1271 af_inet_ifreq_ioctl(const char *name, struct ifreq *ifr, unsigned long int cmd,
1272 const char *cmd_name)
1276 ovs_strzcpy(ifr->ifr_name, name, sizeof ifr->ifr_name);
1277 error = af_inet_ioctl(cmd, ifr);
1279 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
1280 VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name,
1281 ovs_strerror(error));
1286 /* sockaddr_storage helpers. */
1288 /* Returns the IPv4 or IPv6 port in 'ss'. */
1290 ss_get_port(const struct sockaddr_storage *ss)
1292 if (ss->ss_family == AF_INET) {
1293 const struct sockaddr_in *sin
1294 = ALIGNED_CAST(const struct sockaddr_in *, ss);
1295 return ntohs(sin->sin_port);
1296 } else if (ss->ss_family == AF_INET6) {
1297 const struct sockaddr_in6 *sin6
1298 = ALIGNED_CAST(const struct sockaddr_in6 *, ss);
1299 return ntohs(sin6->sin6_port);
1305 /* Formats the IPv4 or IPv6 address in 'ss' into the 'bufsize' bytes in 'buf'.
1306 * If 'ss' is an IPv6 address, puts square brackets around the address.
1307 * 'bufsize' should be at least SS_NTOP_BUFSIZE.
1311 ss_format_address(const struct sockaddr_storage *ss,
1312 char *buf, size_t bufsize)
1314 ovs_assert(bufsize >= SS_NTOP_BUFSIZE);
1315 if (ss->ss_family == AF_INET) {
1316 const struct sockaddr_in *sin
1317 = ALIGNED_CAST(const struct sockaddr_in *, ss);
1319 snprintf(buf, bufsize, IP_FMT, IP_ARGS(sin->sin_addr.s_addr));
1320 } else if (ss->ss_family == AF_INET6) {
1321 const struct sockaddr_in6 *sin6
1322 = ALIGNED_CAST(const struct sockaddr_in6 *, ss);
1325 inet_ntop(AF_INET6, sin6->sin6_addr.s6_addr, buf + 1, bufsize - 1);
1326 strcpy(strchr(buf, '\0'), "]");
1335 ss_length(const struct sockaddr_storage *ss)
1337 switch (ss->ss_family) {
1339 return sizeof(struct sockaddr_in);
1342 return sizeof(struct sockaddr_in6);
1349 /* For Windows socket calls, 'errno' is not set. One has to call
1350 * WSAGetLastError() to get the error number and then pass it to
1351 * this function to get the correct error string.
1353 * ovs_strerror() calls strerror_r() and would not get the correct error
1354 * string for Windows sockets, but is good for POSIX. */
1356 sock_strerror(int error)
1359 return ovs_format_message(error);
1361 return ovs_strerror(error);