X-Git-Url: http://git.cascardo.eti.br/?a=blobdiff_plain;f=lib%2Fsocket-util.c;h=6f959b271a785b2daf98bc188e5c47d6c6e0f916;hb=968eec593cc61690c9e0ed97450c4889258381af;hp=3af74a48d9dd5ce400b7ccb2fa730780474093e5;hpb=3762274e6359f4afe04107851f4c71347fa0afa0;p=cascardo%2Fovs.git diff --git a/lib/socket-util.c b/lib/socket-util.c index 3af74a48d..6f959b271 100644 --- a/lib/socket-util.c +++ b/lib/socket-util.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2008, 2009, 2010 Nicira Networks. + * 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. @@ -19,153 +19,271 @@ #include #include #include +#include #include +#include #include #include #include #include #include -#include #include #include +#include #include #include -#include "fatal-signal.h" +#include "dynamic-string.h" +#include "ovs-thread.h" +#include "packets.h" +#include "poll-loop.h" #include "util.h" +#include "openvswitch/vlog.h" +#ifdef __linux__ +#include +#endif +#ifdef HAVE_NETLINK +#include "netlink-protocol.h" +#include "netlink-socket.h" +#endif -#include "vlog.h" -#define THIS_MODULE VLM_socket_util +VLOG_DEFINE_THIS_MODULE(socket_util); + +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 set_nonblocking(int fd) { +#ifndef _WIN32 int flags = fcntl(fd, F_GETFL, 0); if (flags != -1) { if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) != -1) { return 0; } else { - VLOG_ERR("fcntl(F_SETFL) failed: %s", strerror(errno)); + VLOG_ERR("fcntl(F_SETFL) failed: %s", ovs_strerror(errno)); return errno; } } else { - VLOG_ERR("fcntl(F_GETFL) failed: %s", strerror(errno)); + VLOG_ERR("fcntl(F_GETFL) failed: %s", ovs_strerror(errno)); return errno; } +#else + unsigned long arg = 1; + if (ioctlsocket(fd, FIONBIO, &arg)) { + int error = sock_errno(); + VLOG_ERR("set_nonblocking failed: %s", sock_strerror(error)); + return error; + } + return 0; +#endif +} + +void +xset_nonblocking(int fd) +{ + if (set_nonblocking(fd)) { + exit(EXIT_FAILURE); + } +} + +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)); + } } -/* Returns the maximum valid FD value, plus 1. */ +/* 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 -get_max_fds(void) +set_dscp(int fd, int family, uint8_t dscp) { - static int max_fds = -1; - if (max_fds < 0) { - struct rlimit r; - if (!getrlimit(RLIMIT_NOFILE, &r) - && r.rlim_cur != RLIM_INFINITY - && r.rlim_cur != RLIM_SAVED_MAX - && r.rlim_cur != RLIM_SAVED_CUR) { - max_fds = r.rlim_cur; - } else { - VLOG_WARN("failed to obtain fd limit, defaulting to 1024"); - max_fds = 1024; - } + 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; + + 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 max_fds; + + 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 * address, into a numeric IP address in '*addr'. Returns 0 if successful, * otherwise a positive errno value. */ int -lookup_ip(const char *host_name, struct in_addr *addr) +lookup_ip(const char *host_name, struct in_addr *addr) { - if (!inet_aton(host_name, addr)) { - struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); + 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; } return 0; } -/* Returns the error condition associated with socket 'fd' and resets the - * socket's error status. */ +/* Translates 'host_name', which must be a string representation of an IPv6 + * address, into a numeric IPv6 address in '*addr'. Returns 0 if successful, + * otherwise a positive errno value. */ int -get_socket_error(int fd) +lookup_ipv6(const char *host_name, struct in6_addr *addr) { - int error; - socklen_t len = sizeof(error); - if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &error, &len) < 0) { - struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10); - error = errno; - VLOG_ERR_RL(&rl, "getsockopt(SO_ERROR): %s", strerror(error)); + 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; + } + 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: + * getaddrinfo() 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 activity. (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 addrinfo *result; + struct addrinfo hints; + + if (ip_parse(host_name, &addr->s_addr)) { + return 0; + } + + memset(&hints, 0, sizeof hints); + hints.ai_family = AF_INET; + + switch (getaddrinfo(host_name, NULL, &hints, &result)) { + case 0: + *addr = ALIGNED_CAST(struct sockaddr_in *, + result->ai_addr)->sin_addr; + freeaddrinfo(result); + return 0; + +#ifdef EAI_ADDRFAMILY + case EAI_ADDRFAMILY: +#endif + case EAI_NONAME: + case EAI_SERVICE: + return ENOENT; + + case EAI_AGAIN: + return EAGAIN; + + case EAI_BADFLAGS: + case EAI_FAMILY: + case EAI_SOCKTYPE: + return EINVAL; + + case EAI_FAIL: + return EIO; + + case EAI_MEMORY: + return ENOMEM; + +#if defined (EAI_NODATA) && EAI_NODATA != EAI_NONAME + case EAI_NODATA: + return ENXIO; +#endif + +#ifdef EAI_SYSTEM + case EAI_SYSTEM: + return sock_errno(); +#endif + + default: + return EPROTO; } - return error; } int -check_connection_completion(int fd) +check_connection_completion(int fd) { + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10); struct pollfd pfd; int retval; pfd.fd = fd; pfd.events = POLLOUT; + +#ifndef _WIN32 do { retval = poll(&pfd, 1, 0); } while (retval < 0 && errno == EINTR); +#else + retval = WSAPoll(&pfd, 1, 0); +#endif if (retval == 1) { - return get_socket_error(fd); + if (pfd.revents & POLLERR) { + ssize_t n = send(fd, "", 1, 0); + if (n < 0) { + return sock_errno(); + } else { + VLOG_ERR_RL(&rl, "poll return POLLERR but send succeeded"); + return EPROTO; + } + } + return 0; } else if (retval < 0) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10); - VLOG_ERR_RL(&rl, "poll: %s", strerror(errno)); + VLOG_ERR_RL(&rl, "poll: %s", sock_strerror(sock_errno())); return errno; } else { return EAGAIN; } } -/* 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. */ +/* Returns the size of socket 'sock''s receive buffer (SO_RCVBUF), or a + * negative errno value if an error occurs. */ int -drain_rcvbuf(int fd) +get_socket_rcvbuf(int sock) { - socklen_t rcvbuf_len; - size_t rcvbuf; + int rcvbuf; + int error; - 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; - } - 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. */ -#ifdef __linux__ -#define BUFFER_SIZE 1 -#else -#define BUFFER_SIZE 2048 -#endif - char buffer[BUFFER_SIZE]; - 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; + 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 @@ -186,167 +304,135 @@ drain_fd(int fd, size_t n_packets) } } -/* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in - * '*un_len' the size of the sockaddr_un. */ -static void -make_sockaddr_un(const char *name, struct sockaddr_un* un, socklen_t *un_len) +ovs_be32 +guess_netmask(ovs_be32 ip_) { - un->sun_family = AF_UNIX; - strncpy(un->sun_path, name, sizeof un->sun_path); - un->sun_path[sizeof un->sun_path - 1] = '\0'; - *un_len = (offsetof(struct sockaddr_un, sun_path) - + strlen (un->sun_path) + 1); + uint32_t ip = ntohl(ip_); + return ((ip >> 31) == 0 ? htonl(0xff000000) /* Class A */ + : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */ + : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */ + : htonl(0)); /* ??? */ } -/* 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. If 'passcred' is true, the socket - * is configured to receive SCM_CREDENTIALS control messages. +/* This is like strsep() except: * - * Returns the socket's fd if successful, otherwise a negative errno value. */ -int -make_unix_socket(int style, bool nonblock, bool passcred OVS_UNUSED, - const char *bind_path, const char *connect_path) + * - The separator string is ":". + * + * - Square brackets [] quote ":" separators and are removed from the + * tokens. */ +static char * +parse_bracketed_token(char **pp) { - int error; - int fd; - - fd = socket(PF_UNIX, style, 0); - if (fd < 0) { - return -errno; + char *p = *pp; + + if (p == NULL) { + return NULL; + } else if (*p == '\0') { + *pp = NULL; + return p; + } else if (*p == '[') { + char *start = p + 1; + char *end = start + strcspn(start, "]"); + *pp = (*end == '\0' ? NULL + : end[1] == ':' ? end + 2 + : end + 1); + *end = '\0'; + return start; + } else { + char *start = p; + char *end = start + strcspn(start, ":"); + *pp = *end == '\0' ? NULL : end + 1; + *end = '\0'; + return start; } +} - /* 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) { - int flags = fcntl(fd, F_GETFL, 0); - if (flags == -1) { - goto error; - } - if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) == -1) { - goto error; - } - } +static bool +parse_sockaddr_components(struct sockaddr_storage *ss, + const char *host_s, + const char *port_s, uint16_t default_port, + const char *s) +{ + struct sockaddr_in *sin = ALIGNED_CAST(struct sockaddr_in *, ss); + int port; - if (bind_path) { - struct sockaddr_un un; - socklen_t un_len; - make_sockaddr_un(bind_path, &un, &un_len); - if (unlink(un.sun_path) && errno != ENOENT) { - VLOG_WARN("unlinking \"%s\": %s\n", un.sun_path, strerror(errno)); - } - fatal_signal_add_file_to_unlink(bind_path); - if (bind(fd, (struct sockaddr*) &un, un_len) - || fchmod(fd, S_IRWXU)) { - goto error; + if (port_s && port_s[0]) { + if (!str_to_int(port_s, 10, &port) || port < 0 || port > 65535) { + VLOG_ERR("%s: bad port number \"%s\"", s, port_s); } + } else { + port = default_port; } - if (connect_path) { - struct sockaddr_un un; - socklen_t un_len; - make_sockaddr_un(connect_path, &un, &un_len); - if (connect(fd, (struct sockaddr*) &un, un_len) - && errno != EINPROGRESS) { - printf("connect failed with %s\n", strerror(errno)); - goto error; - } - } + memset(ss, 0, sizeof *ss); + if (strchr(host_s, ':')) { + struct sockaddr_in6 *sin6 + = ALIGNED_CAST(struct sockaddr_in6 *, ss); -#ifdef SCM_CREDENTIALS - if (passcred) { - int enable = 1; - if (setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &enable, sizeof(enable))) { - goto error; + sin6->sin6_family = AF_INET6; + sin6->sin6_port = htons(port); + 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 (!ip_parse(host_s, &sin->sin_addr.s_addr)) { + VLOG_ERR("%s: bad IPv4 address \"%s\"", s, host_s); + goto exit; } } -#endif - - return fd; - -error: - error = errno == EAGAIN ? EPROTO : errno; - if (bind_path) { - fatal_signal_remove_file_to_unlink(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); -} + return true; -uint32_t -guess_netmask(uint32_t ip) -{ - ip = ntohl(ip); - return ((ip >> 31) == 0 ? htonl(0xff000000) /* Class A */ - : (ip >> 30) == 2 ? htonl(0xffff0000) /* Class B */ - : (ip >> 29) == 6 ? htonl(0xffffff00) /* Class C */ - : htonl(0)); /* ??? */ +exit: + memset(ss, 0, sizeof *ss); + return false; } /* Parses 'target', which should be a string in the format "[:]". - * is required. If 'default_port' is nonzero then is optional - * and defaults to 'default_port'. + * , which is required, may be an IPv4 address or an IPv6 address + * enclosed in square brackets. If 'default_port' is nonzero then is + * optional and defaults to 'default_port'. * - * On success, returns true and stores the parsed remote address into '*sinp'. - * On failure, logs an error, stores zeros into '*sinp', and returns false. */ + * On success, returns true and stores the parsed remote address into '*ss'. + * On failure, logs an error, stores zeros into '*ss', and returns false. */ bool inet_parse_active(const char *target_, uint16_t default_port, - struct sockaddr_in *sinp) + struct sockaddr_storage *ss) { char *target = xstrdup(target_); - char *save_ptr = NULL; - const char *host_name; - const char *port_string; - bool ok = false; - - /* Defaults. */ - sinp->sin_family = AF_INET; - sinp->sin_port = htons(default_port); - - /* Tokenize. */ - host_name = strtok_r(target, ":", &save_ptr); - port_string = strtok_r(NULL, ":", &save_ptr); - if (!host_name) { - VLOG_ERR("%s: bad peer name format", target_); - goto exit; - } - - /* Look up IP, port. */ - if (lookup_ip(host_name, &sinp->sin_addr)) { - goto exit; - } - if (port_string && atoi(port_string)) { - sinp->sin_port = htons(atoi(port_string)); - } else if (!default_port) { - VLOG_ERR("%s: port number must be specified", target_); - goto exit; + const char *port; + const char *host; + char *p; + bool ok; + + p = target; + host = parse_bracketed_token(&p); + port = parse_bracketed_token(&p); + if (!host) { + VLOG_ERR("%s: host must be specified", target_); + ok = false; + } else if (!port && !default_port) { + VLOG_ERR("%s: port must be specified", target_); + ok = false; + } else { + ok = parse_sockaddr_components(ss, host, port, default_port, target_); } - - ok = true; - -exit: if (!ok) { - memset(sinp, 0, sizeof *sinp); + memset(ss, 0, sizeof *ss); } free(target); return ok; } -/* Opens a non-blocking IPv4 socket of the specified 'style' and connects to - * 'target', which should be a string in the format "[:]". - * is required. If 'default_port' is nonzero then is optional and - * defaults to 'default_port'. + +/* Opens a non-blocking IPv4 or IPv6 socket of the specified 'style' and + * connects to 'target', which should be a string in the format + * "[:]". , which is required, may be an IPv4 address or an + * IPv6 address enclosed in square brackets. If 'default_port' is nonzero then + * is optional and defaults to 'default_port'. * * 'style' should be SOCK_STREAM (for TCP) or SOCK_DGRAM (for UDP). * @@ -355,62 +441,77 @@ exit: * into '*fdp'. On failure, returns a positive errno value other than EAGAIN * and stores -1 into '*fdp'. * - * If 'sinp' is non-null, then on success the target address is stored into - * '*sinp'. */ + * If 'ss' is non-null, then on success stores the target address into '*ss'. + * + * '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_storage *ssp, int *fdp, uint8_t dscp) { - struct sockaddr_in sin; + struct sockaddr_storage ss; int fd = -1; int error; /* Parse. */ - if (!inet_parse_active(target, default_port, &sin)) { + if (!inet_parse_active(target, default_port, &ss)) { error = EAFNOSUPPORT; goto exit; } /* Create non-blocking socket. */ - fd = socket(AF_INET, style, 0); + fd = socket(ss.ss_family, style, 0); if (fd < 0) { - VLOG_ERR("%s: socket: %s", target, strerror(errno)); - error = errno; + error = sock_errno(); + VLOG_ERR("%s: socket: %s", target, sock_strerror(error)); goto exit; } 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, ss.ss_family, dscp); + if (error) { + VLOG_ERR("%s: set_dscp: %s", target, sock_strerror(error)); + goto exit; } /* Connect. */ - error = connect(fd, (struct sockaddr *) &sin, sizeof sin) == 0 ? 0 : errno; - if (error == EINPROGRESS) { + error = connect(fd, (struct sockaddr *) &ss, ss_length(&ss)) == 0 + ? 0 + : sock_errno(); + if (error == EINPROGRESS +#ifdef _WIN32 + || error == WSAEALREADY || error == WSAEWOULDBLOCK +#endif + ) { 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; + if (error && error != EAGAIN) { + if (ssp) { + memset(ssp, 0, sizeof *ssp); + } + if (fd >= 0) { + closesocket(fd); + fd = -1; } - *fdp = fd; } else { - *fdp = -1; + if (ssp) { + *ssp = ss; + } } + *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 "[][:]": +/* Parses 'target', which should be a string in the format "[][:]": * * - If 'default_port' is -1, then is required. Otherwise, if * is omitted, then 'default_port' is used instead. @@ -418,7 +519,43 @@ exit: * - If (or 'default_port', if used) is 0, then no port is bound * and the TCP/IP stack will select a port. * - * - If is omitted then the IP address is wildcarded. + * - is optional. If supplied, it may be an IPv4 address or an + * IPv6 address enclosed in square brackets. If omitted, the IP address + * is wildcarded. + * + * If successful, stores the address into '*ss' and returns true; otherwise + * zeros '*ss' and returns false. */ +bool +inet_parse_passive(const char *target_, int default_port, + struct sockaddr_storage *ss) +{ + char *target = xstrdup(target_); + const char *port; + const char *host; + char *p; + bool ok; + + p = target; + port = parse_bracketed_token(&p); + host = parse_bracketed_token(&p); + if (!port && default_port < 0) { + VLOG_ERR("%s: port must be specified", target_); + ok = false; + } else { + ok = parse_sockaddr_components(ss, host ? host : "0.0.0.0", + port, default_port, target_); + } + if (!ok) { + memset(ss, 0, sizeof *ss); + } + free(target); + return ok; +} + + +/* Opens a non-blocking IPv4 or IPv6 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). * @@ -427,117 +564,95 @@ exit: * 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'. */ + * If 'ss' is non-null, then on success stores the bound address into '*ss'. + * + * '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. + * + * 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_in *sinp) +inet_open_passive(int style, const char *target, int default_port, + struct sockaddr_storage *ssp, uint8_t dscp, + bool kernel_print_port) { - char *target = xstrdup(target_); - char *string_ptr = target; - struct sockaddr_in sin; - const char *host_name; - const char *port_string; - int fd, error, port; - unsigned int yes = 1; - - /* 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); - - /* Parse optional port number. */ - port_string = strsep(&string_ptr, ":"); - if (port_string && str_to_int(port_string, 10, &port)) { - sin.sin_port = htons(port); - } else if (default_port < 0) { - VLOG_ERR("%s: port number must be specified", target_); - error = EAFNOSUPPORT; - goto exit; - } + bool kernel_chooses_port; + struct sockaddr_storage ss; + int fd = 0, error; + unsigned int yes = 1; - /* 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 (!inet_parse_passive(target, default_port, &ss)) { + return -EAFNOSUPPORT; } + kernel_chooses_port = ss_get_port(&ss) == 0; /* Create non-blocking socket, set SO_REUSEADDR. */ - fd = socket(AF_INET, style, 0); + fd = socket(ss.ss_family, style, 0); if (fd < 0) { - error = errno; - VLOG_ERR("%s: socket: %s", target_, strerror(error)); - goto exit; + error = sock_errno(); + VLOG_ERR("%s: socket: %s", target, sock_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; + error = sock_errno(); + VLOG_ERR("%s: setsockopt(SO_REUSEADDR): %s", + target, sock_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; + if (bind(fd, (struct sockaddr *) &ss, ss_length(&ss)) < 0) { + error = sock_errno(); + VLOG_ERR("%s: bind: %s", target, sock_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, ss.ss_family, dscp); + if (error) { + VLOG_ERR("%s: set_dscp: %s", target, sock_strerror(error)); + goto error; } /* Listen. */ - if (listen(fd, 10) < 0) { - error = errno; - VLOG_ERR("%s: listen: %s", target_, strerror(error)); - goto exit_close; + if (style == SOCK_STREAM && listen(fd, 10) < 0) { + error = sock_errno(); + VLOG_ERR("%s: listen: %s", target, sock_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; + if (ssp || kernel_chooses_port) { + socklen_t ss_len = sizeof ss; + if (getsockname(fd, (struct sockaddr *) &ss, &ss_len) < 0) { + error = sock_errno(); + VLOG_ERR("%s: getsockname: %s", target, sock_strerror(error)); + goto error; + } + if (kernel_chooses_port && kernel_print_port) { + VLOG_INFO("%s: listening on port %"PRIu16, + target, ss_get_port(&ss)); } - if (sin.sin_family != AF_INET || sin_len != sizeof sin) { - VLOG_ERR("%s: getsockname: invalid socket name", target_); - goto exit_close; + if (ssp) { + *ssp = ss; } - *sinp = sin; } - error = 0; - goto exit; - -exit_close: - close(fd); -exit: - free(target); - return error ? -error : fd; -} + return fd; -/* Returns a readable and writable fd for /dev/null, if successful, otherwise - * a negative errno value. The caller must not close the returned fd (because - * the same fd will be handed out to subsequent callers). */ -int -get_null_fd(void) -{ - static int null_fd = -1; - if (null_fd < 0) { - null_fd = open("/dev/null", O_RDWR); - if (null_fd < 0) { - int error = errno; - VLOG_ERR("could not open /dev/null: %s", strerror(error)); - return -error; - } +error: + if (ssp) { + memset(ssp, 0, sizeof *ssp); } - return null_fd; + closesocket(fd); + return -error; } int @@ -589,6 +704,7 @@ int fsync_parent_dir(const char *file_name) { int error = 0; +#ifndef _WIN32 char *dir; int fd; @@ -601,15 +717,16 @@ fsync_parent_dir(const char *file_name) * really an error. */ } else { error = errno; - VLOG_ERR("%s: fsync failed (%s)", dir, strerror(error)); + VLOG_ERR("%s: fsync failed (%s)", dir, ovs_strerror(error)); } } close(fd); } else { error = errno; - VLOG_ERR("%s: open failed (%s)", dir, strerror(error)); + VLOG_ERR("%s: open failed (%s)", dir, ovs_strerror(error)); } free(dir); +#endif return error; } @@ -641,3 +758,248 @@ get_mtime(const char *file_name, struct timespec *mtime) } } +static int +getsockopt_int(int fd, int level, int option, const char *optname, int *valuep) +{ + static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 10); + socklen_t len; + int value; + int error; + + len = sizeof value; + if (getsockopt(fd, level, option, &value, &len)) { + error = sock_errno(); + VLOG_ERR_RL(&rl, "getsockopt(%s): %s", optname, sock_strerror(error)); + } else if (len != sizeof value) { + error = EINVAL; + VLOG_ERR_RL(&rl, "getsockopt(%s): value is %u bytes (expected %"PRIuSIZE")", + optname, (unsigned int) len, sizeof value); + } else { + error = 0; + } + + *valuep = error ? 0 : value; + return error; +} + +static void +describe_sockaddr(struct ds *string, int fd, + int (*getaddr)(int, struct sockaddr *, socklen_t *)) +{ + struct sockaddr_storage ss; + socklen_t len = sizeof ss; + + if (!getaddr(fd, (struct sockaddr *) &ss, &len)) { + if (ss.ss_family == AF_INET || ss.ss_family == AF_INET6) { + char addrbuf[SS_NTOP_BUFSIZE]; + + ds_put_format(string, "%s:%"PRIu16, + ss_format_address(&ss, addrbuf, sizeof addrbuf), + ss_get_port(&ss)); +#ifndef _WIN32 + } else if (ss.ss_family == AF_UNIX) { + struct sockaddr_un sun; + const char *null; + size_t maxlen; + + memcpy(&sun, &ss, sizeof sun); + maxlen = len - offsetof(struct sockaddr_un, sun_path); + null = memchr(sun.sun_path, '\0', maxlen); + ds_put_buffer(string, sun.sun_path, + null ? null - sun.sun_path : maxlen); +#endif + } +#ifdef HAVE_NETLINK + else if (ss.ss_family == AF_NETLINK) { + int protocol; + +/* SO_PROTOCOL was introduced in 2.6.32. Support it regardless of the version + * of the Linux kernel headers in use at build time. */ +#ifndef SO_PROTOCOL +#define SO_PROTOCOL 38 +#endif + + if (!getsockopt_int(fd, SOL_SOCKET, SO_PROTOCOL, "SO_PROTOCOL", + &protocol)) { + switch (protocol) { + case NETLINK_ROUTE: + ds_put_cstr(string, "NETLINK_ROUTE"); + break; + + case NETLINK_GENERIC: + ds_put_cstr(string, "NETLINK_GENERIC"); + break; + + default: + ds_put_format(string, "AF_NETLINK family %d", protocol); + break; + } + } else { + ds_put_cstr(string, "AF_NETLINK"); + } + } +#endif +#if __linux__ + else if (ss.ss_family == AF_PACKET) { + struct sockaddr_ll sll; + + memcpy(&sll, &ss, sizeof sll); + ds_put_cstr(string, "AF_PACKET"); + if (sll.sll_ifindex) { + char name[IFNAMSIZ]; + + if (if_indextoname(sll.sll_ifindex, name)) { + ds_put_format(string, "(%s)", name); + } else { + ds_put_format(string, "(ifindex=%d)", sll.sll_ifindex); + } + } + if (sll.sll_protocol) { + ds_put_format(string, "(protocol=0x%"PRIu16")", + ntohs(sll.sll_protocol)); + } + } +#endif + else if (ss.ss_family == AF_UNSPEC) { + ds_put_cstr(string, "AF_UNSPEC"); + } else { + ds_put_format(string, "AF_%d", (int) ss.ss_family); + } + } +} + + +#ifdef __linux__ +static void +put_fd_filename(struct ds *string, int fd) +{ + char buf[1024]; + char *linkname; + int n; + + linkname = xasprintf("/proc/self/fd/%d", fd); + n = readlink(linkname, buf, sizeof buf); + if (n > 0) { + ds_put_char(string, ' '); + ds_put_buffer(string, buf, n); + if (n > sizeof buf) { + ds_put_cstr(string, "..."); + } + } + free(linkname); +} +#endif + +/* Returns a malloc()'d string describing 'fd', for use in logging. */ +char * +describe_fd(int fd) +{ + struct ds string; + struct stat s; + + ds_init(&string); +#ifndef _WIN32 + if (fstat(fd, &s)) { + ds_put_format(&string, "fstat failed (%s)", ovs_strerror(errno)); + } else if (S_ISSOCK(s.st_mode)) { + describe_sockaddr(&string, fd, getsockname); + ds_put_cstr(&string, "<->"); + describe_sockaddr(&string, fd, getpeername); + } else { + ds_put_cstr(&string, (isatty(fd) ? "tty" + : S_ISDIR(s.st_mode) ? "directory" + : S_ISCHR(s.st_mode) ? "character device" + : S_ISBLK(s.st_mode) ? "block device" + : S_ISREG(s.st_mode) ? "file" + : S_ISFIFO(s.st_mode) ? "FIFO" + : S_ISLNK(s.st_mode) ? "symbolic link" + : "unknown")); +#ifdef __linux__ + put_fd_filename(&string, fd); +#endif + } +#else + ds_put_format(&string,"file descriptor"); +#endif /* _WIN32 */ + return ds_steal_cstr(&string); +} + + +/* sockaddr_storage helpers. */ + +/* Returns the IPv4 or IPv6 port in 'ss'. */ +uint16_t +ss_get_port(const struct sockaddr_storage *ss) +{ + if (ss->ss_family == AF_INET) { + const struct sockaddr_in *sin + = ALIGNED_CAST(const struct sockaddr_in *, ss); + return ntohs(sin->sin_port); + } else if (ss->ss_family == AF_INET6) { + const struct sockaddr_in6 *sin6 + = ALIGNED_CAST(const struct sockaddr_in6 *, ss); + return ntohs(sin6->sin6_port); + } else { + OVS_NOT_REACHED(); + } +} + +/* Formats the IPv4 or IPv6 address in 'ss' into the 'bufsize' bytes in 'buf'. + * If 'ss' is an IPv6 address, puts square brackets around the address. + * 'bufsize' should be at least SS_NTOP_BUFSIZE. + * + * Returns 'buf'. */ +char * +ss_format_address(const struct sockaddr_storage *ss, + char *buf, size_t bufsize) +{ + ovs_assert(bufsize >= SS_NTOP_BUFSIZE); + if (ss->ss_family == AF_INET) { + const struct sockaddr_in *sin + = ALIGNED_CAST(const struct sockaddr_in *, ss); + + snprintf(buf, bufsize, IP_FMT, IP_ARGS(sin->sin_addr.s_addr)); + } else if (ss->ss_family == AF_INET6) { + const struct sockaddr_in6 *sin6 + = ALIGNED_CAST(const struct sockaddr_in6 *, ss); + + buf[0] = '['; + inet_ntop(AF_INET6, sin6->sin6_addr.s6_addr, buf + 1, bufsize - 1); + strcpy(strchr(buf, '\0'), "]"); + } else { + OVS_NOT_REACHED(); + } + + return buf; +} + +size_t +ss_length(const struct sockaddr_storage *ss) +{ + switch (ss->ss_family) { + case AF_INET: + return sizeof(struct sockaddr_in); + + case AF_INET6: + return sizeof(struct sockaddr_in6); + + default: + OVS_NOT_REACHED(); + } +} + +/* For Windows socket calls, 'errno' is not set. One has to call + * WSAGetLastError() to get the error number and then pass it to + * this function to get the correct error string. + * + * ovs_strerror() calls strerror_r() and would not get the correct error + * string for Windows sockets, but is good for POSIX. */ +const char * +sock_strerror(int error) +{ +#ifdef _WIN32 + return ovs_format_message(error); +#else + return ovs_strerror(error); +#endif +}