/*
- * Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks.
+ * Copyright (c) 2008, 2009, 2010, 2011, 2012 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 <config.h>
#include "socket-util.h"
#include <arpa/inet.h>
+#include <assert.h>
#include <errno.h>
#include <fcntl.h>
#include <net/if.h>
#define O_DIRECTORY 0
#endif
+static int getsockopt_int(int fd, int level, int option, const char *optname,
+ int *valuep);
+
/* Sets 'fd' to non-blocking mode. Returns 0 if successful, otherwise a
* positive errno value. */
int
}
}
+static int
+set_dscp(int fd, uint8_t dscp)
+{
+ if (dscp > 63) {
+ return EINVAL;
+ }
+
+ dscp = dscp << 2;
+ if (setsockopt(fd, IPPROTO_IP, IP_TOS, &dscp, sizeof dscp)) {
+ return errno;
+ }
+
+ return 0;
+}
+
static bool
rlim_is_finite(rlim_t limit)
{
return 0;
}
+/* Translates 'host_name', which must be a host name or a string representation
+ * of an IP address, into a numeric IP address in '*addr'. Returns 0 if
+ * successful, otherwise a positive errno value.
+ *
+ * Most Open vSwitch code should not use this because it causes deadlocks:
+ * gethostbyname() sends out a DNS request but that starts a new flow for which
+ * OVS must set up a flow, but it can't because it's waiting for a DNS reply.
+ * The synchronous lookup also delays other activty. (Of course we can solve
+ * this but it doesn't seem worthwhile quite yet.) */
+int
+lookup_hostname(const char *host_name, struct in_addr *addr)
+{
+ struct hostent *h;
+
+ if (inet_aton(host_name, addr)) {
+ return 0;
+ }
+
+ h = gethostbyname(host_name);
+ if (h) {
+ *addr = *(struct in_addr *) h->h_addr;
+ return 0;
+ }
+
+ return (h_errno == HOST_NOT_FOUND ? ENOENT
+ : h_errno == TRY_AGAIN ? EAGAIN
+ : h_errno == NO_RECOVERY ? EIO
+ : h_errno == NO_ADDRESS ? ENXIO
+ : EINVAL);
+}
+
/* Returns the error condition associated with socket 'fd' and resets the
* socket's error status. */
int
get_socket_error(int fd)
{
int error;
- socklen_t len = sizeof(error);
- if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &error, &len) < 0) {
- static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
+
+ if (getsockopt_int(fd, SOL_SOCKET, SO_ERROR, "SO_ERROR", &error)) {
error = errno;
- VLOG_ERR_RL(&rl, "getsockopt(SO_ERROR): %s", strerror(error));
}
return error;
}
int
drain_rcvbuf(int fd)
{
- socklen_t rcvbuf_len;
- size_t rcvbuf;
+ int rcvbuf;
- rcvbuf_len = sizeof rcvbuf;
- if (getsockopt(fd, SOL_SOCKET, SO_RCVBUF, &rcvbuf, &rcvbuf_len) < 0) {
- static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
- VLOG_ERR_RL(&rl, "getsockopt(SO_RCVBUF) failed: %s", strerror(errno));
- return errno;
+ 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
return 0;
}
+/* Returns the size of socket 'sock''s receive buffer (SO_RCVBUF), or a
+ * negative errno value if an error occurs. */
+int
+get_socket_rcvbuf(int sock)
+{
+ int rcvbuf;
+ int error;
+
+ error = getsockopt_int(sock, SOL_SOCKET, SO_RCVBUF, "SO_RCVBUF", &rcvbuf);
+ return error ? -error : rcvbuf;
+}
+
/* Reads and discards up to 'n' datagrams from 'fd', stopping as soon as no
* more data can be immediately read. ('fd' should therefore be in
* non-blocking mode.)*/
error = EPROTO;
}
if (bind_path) {
- fatal_signal_remove_file_to_unlink(bind_path);
+ fatal_signal_unlink_file_now(bind_path);
}
close(fd);
return -error;
* and stores -1 into '*fdp'.
*
* If 'sinp' is non-null, then on success the target address is stored into
- * '*sinp'. */
+ * '*sinp'.
+ *
+ * '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. */
int
inet_open_active(int style, const char *target, uint16_t default_port,
- struct sockaddr_in *sinp, int *fdp)
+ struct sockaddr_in *sinp, int *fdp, uint8_t dscp)
{
struct sockaddr_in sin;
int fd = -1;
}
error = set_nonblocking(fd);
if (error) {
- goto exit_close;
+ 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);
+ if (error) {
+ VLOG_ERR("%s: socket: %s", target, strerror(error));
+ goto exit;
}
/* Connect. */
error = connect(fd, (struct sockaddr *) &sin, sizeof sin) == 0 ? 0 : errno;
if (error == EINPROGRESS) {
error = EAGAIN;
- } else if (error && error != EAGAIN) {
- goto exit_close;
}
- /* Success: error is 0 or EAGAIN. */
- goto exit;
-
-exit_close:
- close(fd);
exit:
if (!error || error == EAGAIN) {
if (sinp) {
*sinp = sin;
}
- *fdp = fd;
- } else {
- *fdp = -1;
+ } else if (fd >= 0) {
+ close(fd);
}
+ *fdp = fd;
return error;
}
-/* Opens a non-blocking IPv4 socket of the specified 'style', binds to
- * 'target', and listens for incoming connections. 'target' should be a string
- * in the format "[<port>][:<ip>]":
+/* Parses 'target', which should be a string in the format "[<port>][:<ip>]":
*
* - If 'default_port' is -1, then <port> is required. Otherwise, if
* <port> is omitted, then 'default_port' is used instead.
*
* - If <ip> is omitted then the IP address is wildcarded.
*
- * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
- *
- * For TCP, the socket will have SO_REUSEADDR turned on.
- *
- * On success, returns a non-negative file descriptor. On failure, returns a
- * negative errno value.
- *
- * If 'sinp' is non-null, then on success the bound address is stored into
- * '*sinp'. */
-int
-inet_open_passive(int style, const char *target_, int default_port,
- struct sockaddr_in *sinp)
+ * If successful, stores the address into '*sinp' and returns true; otherwise
+ * zeros '*sinp' and returns false. */
+bool
+inet_parse_passive(const char *target_, int default_port,
+ struct sockaddr_in *sinp)
{
char *target = xstrdup(target_);
char *string_ptr = target;
- struct sockaddr_in sin;
const char *host_name;
const char *port_string;
- int fd = 0, error, port;
- unsigned int yes = 1;
+ bool ok = false;
+ int port;
/* Address defaults. */
- memset(&sin, 0, sizeof sin);
- sin.sin_family = AF_INET;
- sin.sin_addr.s_addr = htonl(INADDR_ANY);
- sin.sin_port = htons(default_port);
+ memset(sinp, 0, sizeof *sinp);
+ sinp->sin_family = AF_INET;
+ sinp->sin_addr.s_addr = htonl(INADDR_ANY);
+ sinp->sin_port = htons(default_port);
/* Parse optional port number. */
port_string = strsep(&string_ptr, ":");
if (port_string && str_to_int(port_string, 10, &port)) {
- sin.sin_port = htons(port);
+ sinp->sin_port = htons(port);
} else if (default_port < 0) {
VLOG_ERR("%s: port number must be specified", target_);
- error = EAFNOSUPPORT;
goto exit;
}
/* Parse optional bind IP. */
host_name = strsep(&string_ptr, ":");
- if (host_name && host_name[0]) {
- error = lookup_ip(host_name, &sin.sin_addr);
- if (error) {
- goto exit;
- }
+ if (host_name && host_name[0] && lookup_ip(host_name, &sinp->sin_addr)) {
+ goto exit;
+ }
+
+ ok = true;
+
+exit:
+ if (!ok) {
+ memset(sinp, 0, sizeof *sinp);
+ }
+ free(target);
+ return ok;
+}
+
+
+/* Opens a non-blocking IPv4 socket of the specified 'style', binds to
+ * 'target', and listens for incoming connections. Parses 'target' in the same
+ * way was inet_parse_passive().
+ *
+ * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP).
+ *
+ * For TCP, the socket will have SO_REUSEADDR turned on.
+ *
+ * On success, returns a non-negative file descriptor. On failure, returns a
+ * negative errno value.
+ *
+ * If 'sinp' is non-null, then on success the bound address is stored into
+ * '*sinp'.
+ *
+ * '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. */
+int
+inet_open_passive(int style, const char *target, int default_port,
+ struct sockaddr_in *sinp, uint8_t dscp)
+{
+ struct sockaddr_in sin;
+ int fd = 0, error;
+ unsigned int yes = 1;
+
+ if (!inet_parse_passive(target, default_port, &sin)) {
+ return -EAFNOSUPPORT;
}
/* Create non-blocking socket, set SO_REUSEADDR. */
fd = socket(AF_INET, style, 0);
if (fd < 0) {
error = errno;
- VLOG_ERR("%s: socket: %s", target_, strerror(error));
- goto exit;
+ VLOG_ERR("%s: socket: %s", target, strerror(error));
+ return -error;
}
error = set_nonblocking(fd);
if (error) {
- goto exit_close;
+ goto error;
}
if (style == SOCK_STREAM
&& setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof yes) < 0) {
error = errno;
- VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s", target_, strerror(error));
- goto exit_close;
+ VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s", target, strerror(error));
+ goto error;
}
/* Bind. */
if (bind(fd, (struct sockaddr *) &sin, sizeof sin) < 0) {
error = errno;
- VLOG_ERR("%s: bind: %s", target_, strerror(error));
- goto exit_close;
+ VLOG_ERR("%s: bind: %s", target, strerror(error));
+ goto error;
+ }
+
+ /* 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);
+ if (error) {
+ VLOG_ERR("%s: socket: %s", target, strerror(error));
+ goto error;
}
/* Listen. */
- if (listen(fd, 10) < 0) {
+ if (style == SOCK_STREAM && listen(fd, 10) < 0) {
error = errno;
- VLOG_ERR("%s: listen: %s", target_, strerror(error));
- goto exit_close;
+ VLOG_ERR("%s: listen: %s", target, strerror(error));
+ goto error;
}
if (sinp) {
socklen_t sin_len = sizeof sin;
if (getsockname(fd, (struct sockaddr *) &sin, &sin_len) < 0){
error = errno;
- VLOG_ERR("%s: getsockname: %s", target_, strerror(error));
- goto exit_close;
+ VLOG_ERR("%s: getsockname: %s", target, strerror(error));
+ goto error;
}
if (sin.sin_family != AF_INET || sin_len != sizeof sin) {
- VLOG_ERR("%s: getsockname: invalid socket name", target_);
- goto exit_close;
+ error = EAFNOSUPPORT;
+ VLOG_ERR("%s: getsockname: invalid socket name", target);
+ goto error;
}
*sinp = sin;
}
- error = 0;
- goto exit;
+ return fd;
-exit_close:
+error:
close(fd);
-exit:
- free(target);
- return error ? -error : fd;
+ return -error;
}
/* Returns a readable and writable fd for /dev/null, if successful, otherwise
}
static int
-getsockopt_int(int fd, int level, int optname, int *valuep)
+getsockopt_int(int fd, int level, int option, const char *optname, int *valuep)
{
- socklen_t len = sizeof *valuep;
+ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10);
+ socklen_t len;
+ int value;
+ int error;
- return (getsockopt(fd, level, optname, valuep, &len) ? errno
- : len == sizeof *valuep ? 0
- : EINVAL);
+ len = sizeof value;
+ if (getsockopt(fd, level, option, &value, &len)) {
+ error = errno;
+ VLOG_ERR_RL(&rl, "getsockopt(%s): %s", optname, strerror(error));
+ } else if (len != sizeof value) {
+ error = EINVAL;
+ VLOG_ERR_RL(&rl, "getsockopt(%s): value is %u bytes (expected %zu)",
+ optname, (unsigned int) len, sizeof value);
+ } else {
+ error = 0;
+ }
+
+ *valuep = error ? 0 : value;
+ return error;
}
static void
#define SO_PROTOCOL 38
#endif
- if (!getsockopt_int(fd, SOL_SOCKET, SO_PROTOCOL, &protocol)) {
+ if (!getsockopt_int(fd, SOL_SOCKET, SO_PROTOCOL, "SO_PROTOCOL",
+ &protocol)) {
switch (protocol) {
case NETLINK_ROUTE:
ds_put_cstr(string, "NETLINK_ROUTE");