/*
- * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
+ * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014, 2015 Nicira, Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
#include <fcntl.h>
#include <net/if.h>
#include <netdb.h>
+#include <netinet/tcp.h>
#include <poll.h>
#include <stddef.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
-#include <sys/ioctl.h>
#include <sys/socket.h>
#include <sys/stat.h>
#include <sys/uio.h>
#include <sys/un.h>
#include <unistd.h>
#include "dynamic-string.h"
-#include "fatal-signal.h"
#include "ovs-thread.h"
#include "packets.h"
#include "poll-loop.h"
#include "util.h"
-#include "vlog.h"
-#if AF_PACKET && LINUX_DATAPATH
+#include "openvswitch/vlog.h"
+#ifdef __linux__
#include <linux/if_packet.h>
#endif
#ifdef HAVE_NETLINK
VLOG_DEFINE_THIS_MODULE(socket_util);
-/* #ifdefs make it a pain to maintain code: you have to try to build both ways.
- * Thus, this file compiles all of the code regardless of the target, by
- * writing "if (LINUX_DATAPATH)" instead of "#ifdef __linux__". */
-#ifndef LINUX_DATAPATH
-#define LINUX_DATAPATH 0
-#endif
-
-#ifndef O_DIRECTORY
-#define O_DIRECTORY 0
-#endif
-
-/* Maximum length of the sun_path member in a struct sockaddr_un, excluding
- * space for a null terminator. */
-#define MAX_UN_LEN (sizeof(((struct sockaddr_un *) 0)->sun_path) - 1)
-
static int getsockopt_int(int fd, int level, int option, const char *optname,
int *valuep);
}
}
+void
+setsockopt_tcp_nodelay(int fd)
+{
+ int on = 1;
+ int retval;
+
+ retval = setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, &on, sizeof on);
+ if (retval) {
+ retval = sock_errno();
+ VLOG_ERR("setsockopt(TCP_NODELAY): %s", sock_strerror(retval));
+ }
+}
+
+/* Sets the DSCP value of socket 'fd' to 'dscp', which must be 63 or less.
+ * 'family' must indicate the socket's address family (AF_INET or AF_INET6, to
+ * do anything useful). */
int
-set_dscp(int fd, uint8_t dscp)
+set_dscp(int fd, int family, uint8_t dscp)
{
+ int retval;
int val;
+#ifdef _WIN32
+ /* XXX: Consider using QoS2 APIs for Windows to set dscp. */
+ return 0;
+#endif
+
if (dscp > 63) {
return EINVAL;
}
-
val = dscp << 2;
- if (setsockopt(fd, IPPROTO_IP, IP_TOS, &val, sizeof val)) {
- return sock_errno();
+
+ switch (family) {
+ case AF_INET:
+ retval = setsockopt(fd, IPPROTO_IP, IP_TOS, &val, sizeof val);
+ break;
+
+ case AF_INET6:
+ retval = setsockopt(fd, IPPROTO_IPV6, IPV6_TCLASS, &val, sizeof val);
+ break;
+
+ default:
+ return ENOPROTOOPT;
}
- return 0;
+ return retval ? sock_errno() : 0;
+}
+
+/* Checks whether 'host_name' is an IPv4 or IPv6 address. It is assumed
+ * that 'host_name' is valid. Returns false if it is IPv4 address, true if
+ * it is IPv6 address. */
+bool
+addr_is_ipv6(const char *host_name)
+{
+ return strchr(host_name, ':') != NULL;
}
/* Translates 'host_name', which must be a string representation of an IP
int
lookup_ip(const char *host_name, struct in_addr *addr)
{
- if (!inet_pton(AF_INET, host_name, addr)) {
+ if (!ip_parse(host_name, &addr->s_addr)) {
static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
VLOG_ERR_RL(&rl, "\"%s\" is not a valid IP address", host_name);
return ENOENT;
int
lookup_ipv6(const char *host_name, struct in6_addr *addr)
{
- if (inet_pton(AF_INET6, host_name, addr) != 1) {
+ if (!ipv6_parse(host_name, addr)) {
static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
VLOG_ERR_RL(&rl, "\"%s\" is not a valid IPv6 address", host_name);
return ENOENT;
struct addrinfo *result;
struct addrinfo hints;
- if (inet_pton(AF_INET, host_name, addr)) {
+ if (ip_parse(host_name, &addr->s_addr)) {
return 0;
}
}
}
-#ifndef _WIN32
-/* Drain all the data currently in the receive queue of a datagram socket (and
- * possibly additional data). There is no way to know how many packets are in
- * the receive queue, but we do know that the total number of bytes queued does
- * not exceed the receive buffer size, so we pull packets until none are left
- * or we've read that many bytes. */
-int
-drain_rcvbuf(int fd)
-{
- int rcvbuf;
-
- rcvbuf = get_socket_rcvbuf(fd);
- if (rcvbuf < 0) {
- return -rcvbuf;
- }
-
- while (rcvbuf > 0) {
- /* In Linux, specifying MSG_TRUNC in the flags argument causes the
- * datagram length to be returned, even if that is longer than the
- * buffer provided. Thus, we can use a 1-byte buffer to discard the
- * incoming datagram and still be able to account how many bytes were
- * removed from the receive buffer.
- *
- * On other Unix-like OSes, MSG_TRUNC has no effect in the flags
- * argument. */
- char buffer[LINUX_DATAPATH ? 1 : 2048];
- ssize_t n_bytes = recv(fd, buffer, sizeof buffer,
- MSG_TRUNC | MSG_DONTWAIT);
- if (n_bytes <= 0 || n_bytes >= rcvbuf) {
- break;
- }
- rcvbuf -= n_bytes;
- }
- return 0;
-}
-#endif
-
/* Returns the size of socket 'sock''s receive buffer (SO_RCVBUF), or a
* negative errno value if an error occurs. */
int
}
}
-#ifndef _WIN32
-/* Attempts to shorten 'name' by opening a file descriptor for the directory
- * part of the name and indirecting through /proc/self/fd/<dirfd>/<basename>.
- * On systems with Linux-like /proc, this works as long as <basename> isn't too
- * long.
- *
- * On success, returns 0 and stores the short name in 'short_name' and a
- * directory file descriptor to eventually be closed in '*dirfpd'. */
-static int
-shorten_name_via_proc(const char *name, char short_name[MAX_UN_LEN + 1],
- int *dirfdp)
-{
- char *dir, *base;
- int dirfd;
- int len;
-
- if (!LINUX_DATAPATH) {
- return ENAMETOOLONG;
- }
-
- dir = dir_name(name);
- dirfd = open(dir, O_DIRECTORY | O_RDONLY);
- if (dirfd < 0) {
- static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
- int error = errno;
-
- VLOG_WARN_RL(&rl, "%s: open failed (%s)", dir, ovs_strerror(error));
- free(dir);
-
- return error;
- }
- free(dir);
-
- base = base_name(name);
- len = snprintf(short_name, MAX_UN_LEN + 1,
- "/proc/self/fd/%d/%s", dirfd, base);
- free(base);
-
- if (len >= 0 && len <= MAX_UN_LEN) {
- *dirfdp = dirfd;
- return 0;
- } else {
- close(dirfd);
- return ENAMETOOLONG;
- }
-}
-
-/* Attempts to shorten 'name' by creating a symlink for the directory part of
- * the name and indirecting through <symlink>/<basename>. This works on
- * systems that support symlinks, as long as <basename> isn't too long.
- *
- * On success, returns 0 and stores the short name in 'short_name' and the
- * symbolic link to eventually delete in 'linkname'. */
-static int
-shorten_name_via_symlink(const char *name, char short_name[MAX_UN_LEN + 1],
- char linkname[MAX_UN_LEN + 1])
-{
- char *abs, *dir, *base;
- const char *tmpdir;
- int error;
- int i;
-
- abs = abs_file_name(NULL, name);
- dir = dir_name(abs);
- base = base_name(abs);
- free(abs);
-
- tmpdir = getenv("TMPDIR");
- if (tmpdir == NULL) {
- tmpdir = "/tmp";
- }
-
- for (i = 0; i < 1000; i++) {
- int len;
-
- len = snprintf(linkname, MAX_UN_LEN + 1,
- "%s/ovs-un-c-%"PRIu32, tmpdir, random_uint32());
- error = (len < 0 || len > MAX_UN_LEN ? ENAMETOOLONG
- : symlink(dir, linkname) ? errno
- : 0);
- if (error != EEXIST) {
- break;
- }
- }
-
- if (!error) {
- int len;
-
- fatal_signal_add_file_to_unlink(linkname);
-
- len = snprintf(short_name, MAX_UN_LEN + 1, "%s/%s", linkname, base);
- if (len < 0 || len > MAX_UN_LEN) {
- fatal_signal_unlink_file_now(linkname);
- error = ENAMETOOLONG;
- }
- }
-
- if (error) {
- linkname[0] = '\0';
- }
- free(dir);
- free(base);
-
- return error;
-}
-
-/* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in
- * '*un_len' the size of the sockaddr_un.
- *
- * Returns 0 on success, otherwise a positive errno value.
- *
- * Uses '*dirfdp' and 'linkname' to store references to data when the caller no
- * longer needs to use 'un'. On success, freeing these references with
- * free_sockaddr_un() is mandatory to avoid a leak; on failure, freeing them is
- * unnecessary but harmless. */
-static int
-make_sockaddr_un(const char *name, struct sockaddr_un *un, socklen_t *un_len,
- int *dirfdp, char linkname[MAX_UN_LEN + 1])
-{
- char short_name[MAX_UN_LEN + 1];
-
- *dirfdp = -1;
- linkname[0] = '\0';
- if (strlen(name) > MAX_UN_LEN) {
- /* 'name' is too long to fit in a sockaddr_un. Try a workaround. */
- int error = shorten_name_via_proc(name, short_name, dirfdp);
- if (error == ENAMETOOLONG) {
- error = shorten_name_via_symlink(name, short_name, linkname);
- }
- if (error) {
- static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
-
- VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum "
- "%"PRIuSIZE" bytes", name, MAX_UN_LEN);
- return error;
- }
-
- name = short_name;
- }
-
- un->sun_family = AF_UNIX;
- ovs_strzcpy(un->sun_path, name, sizeof un->sun_path);
- *un_len = (offsetof(struct sockaddr_un, sun_path)
- + strlen (un->sun_path) + 1);
- return 0;
-}
-
-/* Clean up after make_sockaddr_un(). */
-static void
-free_sockaddr_un(int dirfd, const char *linkname)
-{
- if (dirfd >= 0) {
- close(dirfd);
- }
- if (linkname[0]) {
- fatal_signal_unlink_file_now(linkname);
- }
-}
-
-/* Binds Unix domain socket 'fd' to a file with permissions 0700. */
-static int
-bind_unix_socket(int fd, struct sockaddr *sun, socklen_t sun_len)
-{
- /* According to _Unix Network Programming_, umask should affect bind(). */
- mode_t old_umask = umask(0077);
- int error = bind(fd, sun, sun_len) ? errno : 0;
- umask(old_umask);
- return error;
-}
-
-/* Creates a Unix domain socket in the given 'style' (either SOCK_DGRAM or
- * SOCK_STREAM) that is bound to '*bind_path' (if 'bind_path' is non-null) and
- * connected to '*connect_path' (if 'connect_path' is non-null). If 'nonblock'
- * is true, the socket is made non-blocking.
- *
- * Returns the socket's fd if successful, otherwise a negative errno value. */
-int
-make_unix_socket(int style, bool nonblock,
- const char *bind_path, const char *connect_path)
-{
- int error;
- int fd;
-
- fd = socket(PF_UNIX, style, 0);
- if (fd < 0) {
- return -errno;
- }
-
- /* Set nonblocking mode right away, if we want it. This prevents blocking
- * in connect(), if connect_path != NULL. (In turn, that's a corner case:
- * it will only happen if style is SOCK_STREAM or SOCK_SEQPACKET, and only
- * if a backlog of un-accepted connections has built up in the kernel.) */
- if (nonblock) {
- error = set_nonblocking(fd);
- if (error) {
- goto error;
- }
- }
-
- if (bind_path) {
- char linkname[MAX_UN_LEN + 1];
- struct sockaddr_un un;
- socklen_t un_len;
- int dirfd;
-
- if (unlink(bind_path) && errno != ENOENT) {
- VLOG_WARN("unlinking \"%s\": %s\n",
- bind_path, ovs_strerror(errno));
- }
- fatal_signal_add_file_to_unlink(bind_path);
-
- error = make_sockaddr_un(bind_path, &un, &un_len, &dirfd, linkname);
- if (!error) {
- error = bind_unix_socket(fd, (struct sockaddr *) &un, un_len);
- }
- free_sockaddr_un(dirfd, linkname);
-
- if (error) {
- goto error;
- }
- }
-
- if (connect_path) {
- char linkname[MAX_UN_LEN + 1];
- struct sockaddr_un un;
- socklen_t un_len;
- int dirfd;
-
- error = make_sockaddr_un(connect_path, &un, &un_len, &dirfd, linkname);
- if (!error
- && connect(fd, (struct sockaddr*) &un, un_len)
- && errno != EINPROGRESS) {
- error = errno;
- }
- free_sockaddr_un(dirfd, linkname);
-
- if (error) {
- goto error;
- }
- }
-
- return fd;
-
-error:
- if (error == EAGAIN) {
- error = EPROTO;
- }
- if (bind_path) {
- fatal_signal_unlink_file_now(bind_path);
- }
- close(fd);
- return -error;
-}
-
-int
-get_unix_name_len(socklen_t sun_len)
-{
- return (sun_len >= offsetof(struct sockaddr_un, sun_path)
- ? sun_len - offsetof(struct sockaddr_un, sun_path)
- : 0);
-}
-#endif /* _WIN32 */
-
ovs_be32
guess_netmask(ovs_be32 ip_)
{
sin6->sin6_family = AF_INET6;
sin6->sin6_port = htons(port);
- if (!inet_pton(AF_INET6, host_s, sin6->sin6_addr.s6_addr)) {
+ if (!ipv6_parse(host_s, &sin6->sin6_addr)) {
VLOG_ERR("%s: bad IPv6 address \"%s\"", s, host_s);
goto exit;
}
} else {
sin->sin_family = AF_INET;
sin->sin_port = htons(port);
- if (!inet_pton(AF_INET, host_s, &sin->sin_addr.s_addr)) {
+ if (!ip_parse(host_s, &sin->sin_addr.s_addr)) {
VLOG_ERR("%s: bad IPv4 address \"%s\"", s, host_s);
goto exit;
}
/* The dscp bits must be configured before connect() to ensure that the
* TOS field is set during the connection establishment. If set after
* connect(), the handshake SYN frames will be sent with a TOS of 0. */
- error = set_dscp(fd, dscp);
+ error = set_dscp(fd, ss.ss_family, dscp);
if (error) {
- VLOG_ERR("%s: socket: %s", target, sock_strerror(error));
+ VLOG_ERR("%s: set_dscp: %s", target, sock_strerror(error));
goto exit;
}
*
* 'dscp' becomes the DSCP bits in the IP headers for the new connection. It
* should be in the range [0, 63] and will automatically be shifted to the
- * appropriately place in the IP tos field. */
+ * appropriately place in the IP tos field.
+ *
+ * If 'kernel_print_port' is true and the port is dynamically assigned by
+ * the kernel, print the chosen port. */
int
inet_open_passive(int style, const char *target, int default_port,
- struct sockaddr_storage *ssp, uint8_t dscp)
+ struct sockaddr_storage *ssp, uint8_t dscp,
+ bool kernel_print_port)
{
bool kernel_chooses_port;
struct sockaddr_storage ss;
/* The dscp bits must be configured before connect() to ensure that the TOS
* field is set during the connection establishment. If set after
* connect(), the handshake SYN frames will be sent with a TOS of 0. */
- error = set_dscp(fd, dscp);
+ error = set_dscp(fd, ss.ss_family, dscp);
if (error) {
- VLOG_ERR("%s: socket: %s", target, sock_strerror(error));
+ VLOG_ERR("%s: set_dscp: %s", target, sock_strerror(error));
goto error;
}
VLOG_ERR("%s: getsockname: %s", target, sock_strerror(error));
goto error;
}
- if (kernel_chooses_port) {
+ if (kernel_chooses_port && kernel_print_port) {
VLOG_INFO("%s: listening on port %"PRIu16,
target, ss_get_port(&ss));
}
}
}
-void
-xpipe(int fds[2])
-{
- if (pipe(fds)) {
- VLOG_FATAL("failed to create pipe (%s)", ovs_strerror(errno));
- }
-}
-
-void
-xpipe_nonblocking(int fds[2])
-{
- xpipe(fds);
- xset_nonblocking(fds[0]);
- xset_nonblocking(fds[1]);
-}
-
static int
getsockopt_int(int fd, int level, int option, const char *optname, int *valuep)
{
}
}
#endif
-#if AF_PACKET && LINUX_DATAPATH
+#if __linux__
else if (ss.ss_family == AF_PACKET) {
struct sockaddr_ll sll;
}
-#ifdef LINUX_DATAPATH
+#ifdef __linux__
static void
put_fd_filename(struct ds *string, int fd)
{
: S_ISFIFO(s.st_mode) ? "FIFO"
: S_ISLNK(s.st_mode) ? "symbolic link"
: "unknown"));
-#ifdef LINUX_DATAPATH
+#ifdef __linux__
put_fd_filename(&string, fd);
#endif
}
return ds_steal_cstr(&string);
}
-#ifndef _WIN32
-/* Calls ioctl() on an AF_INET sock, passing the specified 'command' and
- * 'arg'. Returns 0 if successful, otherwise a positive errno value. */
-int
-af_inet_ioctl(unsigned long int command, const void *arg)
-{
- static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
- static int sock;
-
- if (ovsthread_once_start(&once)) {
- sock = socket(AF_INET, SOCK_DGRAM, 0);
- if (sock < 0) {
- int error = sock_errno();
- VLOG_ERR("failed to create inet socket: %s", sock_strerror(error));
- sock = -error;
- }
- ovsthread_once_done(&once);
- }
-
- return (sock < 0 ? -sock
- : ioctl(sock, command, arg) == -1 ? errno
- : 0);
-}
-
-int
-af_inet_ifreq_ioctl(const char *name, struct ifreq *ifr, unsigned long int cmd,
- const char *cmd_name)
-{
- int error;
-
- ovs_strzcpy(ifr->ifr_name, name, sizeof ifr->ifr_name);
- error = af_inet_ioctl(cmd, ifr);
- if (error) {
- static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20);
- VLOG_DBG_RL(&rl, "%s: ioctl(%s) failed: %s", name, cmd_name,
- ovs_strerror(error));
- }
- return error;
-}
-#endif
\f
/* sockaddr_storage helpers. */