X-Git-Url: http://git.cascardo.eti.br/?a=blobdiff_plain;f=lib%2Fsocket-util.c;h=6f959b271a785b2daf98bc188e5c47d6c6e0f916;hb=968eec593cc61690c9e0ed97450c4889258381af;hp=9fea7bdab55455951929e304185578f46c6f4ea6;hpb=b36db1148939c1ca7af3c3973252fea3fbf1c4e6;p=cascardo%2Fovs.git diff --git a/lib/socket-util.c b/lib/socket-util.c index 9fea7bdab..6f959b271 100644 --- a/lib/socket-util.c +++ b/lib/socket-util.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2008, 2009, 2010, 2011, 2012 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. @@ -21,24 +21,24 @@ #include #include #include +#include #include #include #include #include #include -#include #include #include #include #include #include #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 #endif #ifdef HAVE_NETLINK @@ -48,17 +48,6 @@ 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 - static int getsockopt_int(int fd, int level, int option, const char *optname, int *valuep); @@ -67,18 +56,28 @@ static int getsockopt_int(int fd, int level, int option, const char *optname, 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 @@ -89,60 +88,61 @@ xset_nonblocking(int fd) } } +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 errno; - } - return 0; -} + switch (family) { + case AF_INET: + retval = setsockopt(fd, IPPROTO_IP, IP_TOS, &val, sizeof val); + break; -static bool -rlim_is_finite(rlim_t limit) -{ - if (limit == RLIM_INFINITY) { - return false; - } - -#ifdef RLIM_SAVED_CUR /* FreeBSD 8.0 lacks RLIM_SAVED_CUR. */ - if (limit == RLIM_SAVED_CUR) { - return false; - } -#endif + case AF_INET6: + retval = setsockopt(fd, IPPROTO_IPV6, IPV6_TCLASS, &val, sizeof val); + break; -#ifdef RLIM_SAVED_MAX /* FreeBSD 8.0 lacks RLIM_SAVED_MAX. */ - if (limit == RLIM_SAVED_MAX) { - return false; + default: + return ENOPROTOOPT; } -#endif - return true; + return retval ? sock_errno() : 0; } -/* Returns the maximum valid FD value, plus 1. */ -int -get_max_fds(void) +/* 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) { - static int max_fds = -1; - if (max_fds < 0) { - struct rlimit r; - if (!getrlimit(RLIMIT_NOFILE, &r) && rlim_is_finite(r.rlim_cur)) { - max_fds = r.rlim_cur; - } else { - VLOG_WARN("failed to obtain fd limit, defaulting to 1024"); - max_fds = 1024; - } - } - return max_fds; + return strchr(host_name, ':') != NULL; } /* Translates 'host_name', which must be a string representation of an IP @@ -151,7 +151,7 @@ get_max_fds(void) int lookup_ip(const char *host_name, struct in_addr *addr) { - if (!inet_aton(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; @@ -165,7 +165,7 @@ lookup_ip(const char *host_name, struct in_addr *addr) 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; @@ -178,30 +178,64 @@ lookup_ipv6(const char *host_name, struct in6_addr *addr) * 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 + * 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 hostent *h; + struct addrinfo *result; + struct addrinfo hints; - if (inet_aton(host_name, addr)) { + if (ip_parse(host_name, &addr->s_addr)) { return 0; } - h = gethostbyname(host_name); - if (h) { - *addr = *(struct in_addr *) h->h_addr; + 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; - } - return (h_errno == HOST_NOT_FOUND ? ENOENT - : h_errno == TRY_AGAIN ? EAGAIN - : h_errno == NO_RECOVERY ? EIO - : h_errno == NO_ADDRESS ? ENXIO - : EINVAL); +#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; + } } int @@ -213,14 +247,19 @@ check_connection_completion(int fd) 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) { if (pfd.revents & POLLERR) { - ssize_t n = send(fd, "", 1, MSG_DONTWAIT); + ssize_t n = send(fd, "", 1, 0); if (n < 0) { - return errno; + return sock_errno(); } else { VLOG_ERR_RL(&rl, "poll return POLLERR but send succeeded"); return EPROTO; @@ -228,48 +267,13 @@ check_connection_completion(int fd) } return 0; } else if (retval < 0) { - 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. */ -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; -} - /* Returns the size of socket 'sock''s receive buffer (SO_RCVBUF), or a * negative errno value if an error occurs. */ int @@ -300,246 +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; - ovs_strzcpy(un->sun_path, name, sizeof un->sun_path); - *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)); /* ??? */ } -/* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in - * '*un_len' the size of the sockaddr_un. +/* This is like strsep() except: * - * Returns 0 on success, otherwise a positive errno value. On success, - * '*dirfdp' is either -1 or a nonnegative file descriptor that the caller - * should close after using '*un' to bind or connect. On failure, '*dirfdp' is - * -1. */ -static int -make_sockaddr_un(const char *name, struct sockaddr_un *un, socklen_t *un_len, - int *dirfdp) -{ - enum { MAX_UN_LEN = sizeof un->sun_path - 1 }; - - *dirfdp = -1; - if (strlen(name) > MAX_UN_LEN) { - static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1); - - if (LINUX_DATAPATH) { - /* 'name' is too long to fit in a sockaddr_un, but we have a - * workaround for that on Linux: shorten it by opening a file - * descriptor for the directory part of the name and indirecting - * through /proc/self/fd//. */ - char *dir, *base; - char *short_name; - int dirfd; - - dir = dir_name(name); - base = base_name(name); - - dirfd = open(dir, O_DIRECTORY | O_RDONLY); - if (dirfd < 0) { - free(base); - free(dir); - return errno; - } - - short_name = xasprintf("/proc/self/fd/%d/%s", dirfd, base); - free(dir); - free(base); - - if (strlen(short_name) <= MAX_UN_LEN) { - make_sockaddr_un__(short_name, un, un_len); - free(short_name); - *dirfdp = dirfd; - return 0; - } - free(short_name); - close(dirfd); - - VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum " - "%d bytes (even shortened)", name, MAX_UN_LEN); - } else { - /* 'name' is too long and we have no workaround. */ - VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum " - "%d bytes", name, MAX_UN_LEN); - } - - return ENAMETOOLONG; + * - The separator string is ":". + * + * - Square brackets [] quote ":" separators and are removed from the + * tokens. */ +static char * +parse_bracketed_token(char **pp) +{ + 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 { - make_sockaddr_un__(name, un, un_len); - return 0; + char *start = p; + char *end = start + strcspn(start, ":"); + *pp = *end == '\0' ? NULL : end + 1; + *end = '\0'; + return start; } } -/* 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) +static bool +parse_sockaddr_components(struct sockaddr_storage *ss, + const char *host_s, + const char *port_s, uint16_t default_port, + const char *s) { - 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) { - struct sockaddr_un un; - socklen_t un_len; - int dirfd; - - if (unlink(bind_path) && errno != ENOENT) { - VLOG_WARN("unlinking \"%s\": %s\n", bind_path, strerror(errno)); - } - fatal_signal_add_file_to_unlink(bind_path); + struct sockaddr_in *sin = ALIGNED_CAST(struct sockaddr_in *, ss); + int port; - error = make_sockaddr_un(bind_path, &un, &un_len, &dirfd); - if (!error) { - error = bind_unix_socket(fd, (struct sockaddr *) &un, un_len); - } - if (dirfd >= 0) { - close(dirfd); - } - if (error) { - 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; - int dirfd; + memset(ss, 0, sizeof *ss); + if (strchr(host_s, ':')) { + struct sockaddr_in6 *sin6 + = ALIGNED_CAST(struct sockaddr_in6 *, ss); - error = make_sockaddr_un(connect_path, &un, &un_len, &dirfd); - if (!error - && connect(fd, (struct sockaddr*) &un, un_len) - && errno != EINPROGRESS) { - error = errno; + 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; } - if (dirfd >= 0) { - close(dirfd); - } - if (error) { - goto error; + } 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; } } - 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); -} + return true; -ovs_be32 -guess_netmask(ovs_be32 ip_) -{ - 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)); /* ??? */ +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). * @@ -548,31 +441,30 @@ 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, uint8_t dscp) + 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); @@ -580,35 +472,46 @@ inet_open_active(int style, const char *target, uint16_t default_port, 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 + /* 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, strerror(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; } 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; + } + } else { + if (ssp) { + *ssp = ss; } - } else if (fd >= 0) { - close(fd); - fd = -1; } *fdp = fd; return error; } -/* Parses 'target', which 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. @@ -616,54 +519,41 @@ 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 '*sinp' and returns true; otherwise - * zeros '*sinp' and returns false. */ + * 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_in *sinp) + struct sockaddr_storage *ss) { char *target = xstrdup(target_); - char *string_ptr = target; - const char *host_name; - const char *port_string; - bool ok = false; - int port; - - /* Address defaults. */ - 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)) { - sinp->sin_port = htons(port); - } else if (default_port < 0) { - VLOG_ERR("%s: port number must be specified", target_); - goto exit; - } - - /* Parse optional bind IP. */ - host_name = strsep(&string_ptr, ":"); - if (host_name && host_name[0] && lookup_ip(host_name, &sinp->sin_addr)) { - goto exit; + 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_); } - - 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', binds to +/* 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(). * @@ -674,29 +564,34 @@ 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. */ + * 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, uint8_t dscp) + struct sockaddr_storage *ssp, uint8_t dscp, + bool kernel_print_port) { - struct sockaddr_in sin; + bool kernel_chooses_port; + struct sockaddr_storage ss; int fd = 0, error; unsigned int yes = 1; - if (!inet_parse_passive(target, default_port, &sin)) { + 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)); + error = sock_errno(); + VLOG_ERR("%s: socket: %s", target, sock_strerror(error)); return -error; } error = set_nonblocking(fd); @@ -705,72 +600,59 @@ inet_open_passive(int style, const char *target, int default_port, } 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)); + 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)); + 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, dscp); + error = set_dscp(fd, ss.ss_family, dscp); if (error) { - VLOG_ERR("%s: socket: %s", target, strerror(error)); + VLOG_ERR("%s: set_dscp: %s", target, sock_strerror(error)); goto error; } /* Listen. */ if (style == SOCK_STREAM && listen(fd, 10) < 0) { - error = errno; - VLOG_ERR("%s: listen: %s", target, strerror(error)); + 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)); + 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 (sin.sin_family != AF_INET || sin_len != sizeof sin) { - error = EAFNOSUPPORT; - VLOG_ERR("%s: getsockname: invalid socket name", target); - goto error; + if (kernel_chooses_port && kernel_print_port) { + VLOG_INFO("%s: listening on port %"PRIu16, + target, ss_get_port(&ss)); + } + if (ssp) { + *ssp = ss; } - *sinp = sin; } return fd; error: - close(fd); - return -error; -} - -/* 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; - } + if (ssp) { + memset(ssp, 0, sizeof *ssp); } - return null_fd; + closesocket(fd); + return -error; } int @@ -822,6 +704,7 @@ int fsync_parent_dir(const char *file_name) { int error = 0; +#ifndef _WIN32 char *dir; int fd; @@ -834,15 +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; } @@ -874,30 +758,6 @@ get_mtime(const char *file_name, struct timespec *mtime) } } -void -xpipe(int fds[2]) -{ - if (pipe(fds)) { - VLOG_FATAL("failed to create pipe (%s)", strerror(errno)); - } -} - -void -xpipe_nonblocking(int fds[2]) -{ - xpipe(fds); - xset_nonblocking(fds[0]); - xset_nonblocking(fds[1]); -} - -void -xsocketpair(int domain, int type, int protocol, int fds[2]) -{ - if (socketpair(domain, type, protocol, fds)) { - VLOG_FATAL("failed to create socketpair (%s)", strerror(errno)); - } -} - static int getsockopt_int(int fd, int level, int option, const char *optname, int *valuep) { @@ -908,11 +768,11 @@ getsockopt_int(int fd, int level, int option, const char *optname, int *valuep) len = sizeof value; if (getsockopt(fd, level, option, &value, &len)) { - error = errno; - VLOG_ERR_RL(&rl, "getsockopt(%s): %s", optname, strerror(error)); + 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 %zu)", + VLOG_ERR_RL(&rl, "getsockopt(%s): value is %u bytes (expected %"PRIuSIZE")", optname, (unsigned int) len, sizeof value); } else { error = 0; @@ -930,12 +790,13 @@ describe_sockaddr(struct ds *string, int fd, socklen_t len = sizeof ss; if (!getaddr(fd, (struct sockaddr *) &ss, &len)) { - if (ss.ss_family == AF_INET) { - struct sockaddr_in sin; + if (ss.ss_family == AF_INET || ss.ss_family == AF_INET6) { + char addrbuf[SS_NTOP_BUFSIZE]; - memcpy(&sin, &ss, sizeof sin); - ds_put_format(string, IP_FMT":%"PRIu16, - IP_ARGS(sin.sin_addr.s_addr), ntohs(sin.sin_port)); + 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; @@ -946,6 +807,7 @@ describe_sockaddr(struct ds *string, int fd, 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) { @@ -977,7 +839,7 @@ describe_sockaddr(struct ds *string, int fd, } } #endif -#if AF_PACKET && LINUX_DATAPATH +#if __linux__ else if (ss.ss_family == AF_PACKET) { struct sockaddr_ll sll; @@ -1007,7 +869,7 @@ describe_sockaddr(struct ds *string, int fd, } -#ifdef LINUX_DATAPATH +#ifdef __linux__ static void put_fd_filename(struct ds *string, int fd) { @@ -1036,8 +898,9 @@ describe_fd(int fd) struct stat s; ds_init(&string); +#ifndef _WIN32 if (fstat(fd, &s)) { - ds_put_format(&string, "fstat failed (%s)", strerror(errno)); + 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, "<->"); @@ -1051,255 +914,92 @@ describe_fd(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 } +#else + ds_put_format(&string,"file descriptor"); +#endif /* _WIN32 */ return ds_steal_cstr(&string); } -/* Returns the total of the 'iov_len' members of the 'n_iovs' in 'iovs'. - * The caller must ensure that the total does not exceed SIZE_MAX. */ -size_t -iovec_len(const struct iovec iovs[], size_t n_iovs) -{ - size_t len = 0; - size_t i; + +/* sockaddr_storage helpers. */ - for (i = 0; i < n_iovs; i++) { - len += iovs[i].iov_len; - } - return len; -} - -/* Returns true if all of the 'n_iovs' iovecs in 'iovs' have length zero. */ -bool -iovec_is_empty(const struct iovec iovs[], size_t n_iovs) +/* Returns the IPv4 or IPv6 port in 'ss'. */ +uint16_t +ss_get_port(const struct sockaddr_storage *ss) { - size_t i; - - for (i = 0; i < n_iovs; i++) { - if (iovs[i].iov_len) { - return false; - } - } - return true; -} - -/* Sends the 'n_iovs' iovecs of data in 'iovs' and the 'n_fds' file descriptors - * in 'fds' on Unix domain socket 'sock'. Returns the number of bytes - * successfully sent or -1 if an error occurred. On error, sets errno - * appropriately. */ -int -send_iovec_and_fds(int sock, - const struct iovec *iovs, size_t n_iovs, - const int fds[], size_t n_fds) -{ - ovs_assert(sock >= 0); - if (n_fds > 0) { - union { - struct cmsghdr cm; - char control[CMSG_SPACE(SOUTIL_MAX_FDS * sizeof *fds)]; - } cmsg; - struct msghdr msg; - - ovs_assert(!iovec_is_empty(iovs, n_iovs)); - ovs_assert(n_fds <= SOUTIL_MAX_FDS); - - memset(&cmsg, 0, sizeof cmsg); - cmsg.cm.cmsg_len = CMSG_LEN(n_fds * sizeof *fds); - cmsg.cm.cmsg_level = SOL_SOCKET; - cmsg.cm.cmsg_type = SCM_RIGHTS; - memcpy(CMSG_DATA(&cmsg.cm), fds, n_fds * sizeof *fds); - - msg.msg_name = NULL; - msg.msg_namelen = 0; - msg.msg_iov = CONST_CAST(struct iovec *, iovs); - msg.msg_iovlen = n_iovs; - msg.msg_control = &cmsg.cm; - msg.msg_controllen = CMSG_SPACE(n_fds * sizeof *fds); - msg.msg_flags = 0; - - return sendmsg(sock, &msg, 0); + 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 { - return writev(sock, iovs, n_iovs); + OVS_NOT_REACHED(); } } -/* Sends the 'n_iovs' iovecs of data in 'iovs' and the 'n_fds' file descriptors - * in 'fds' on Unix domain socket 'sock'. If 'skip_bytes' is nonzero, then the - * first 'skip_bytes' of data in the iovecs are not sent, and none of the file - * descriptors are sent. The function continues to retry sending until an - * error (other than EINTR) occurs or all the data and fds are sent. - * - * Returns 0 if all the data and fds were successfully sent, otherwise a - * positive errno value. Regardless of success, stores the number of bytes - * sent (always at least 'skip_bytes') in '*bytes_sent'. (If at least one byte - * is sent, then all the fds have been sent.) +/* 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. * - * 'skip_bytes' must be less than or equal to iovec_len(iovs, n_iovs). */ -int -send_iovec_and_fds_fully(int sock, - const struct iovec iovs[], size_t n_iovs, - const int fds[], size_t n_fds, - size_t skip_bytes, size_t *bytes_sent) -{ - *bytes_sent = 0; - while (n_iovs > 0) { - int retval; - - if (skip_bytes) { - retval = skip_bytes; - skip_bytes = 0; - } else if (!*bytes_sent) { - retval = send_iovec_and_fds(sock, iovs, n_iovs, fds, n_fds); - } else { - retval = writev(sock, iovs, n_iovs); - } - - if (retval > 0) { - *bytes_sent += retval; - while (retval > 0) { - const uint8_t *base = iovs->iov_base; - size_t len = iovs->iov_len; - - if (retval < len) { - size_t sent; - int error; - - error = write_fully(sock, base + retval, len - retval, - &sent); - *bytes_sent += sent; - retval += sent; - if (error) { - return error; - } - } - retval -= len; - iovs++; - n_iovs--; - } - } else if (retval == 0) { - if (iovec_is_empty(iovs, n_iovs)) { - break; - } - VLOG_WARN("send returned 0"); - return EPROTO; - } else if (errno != EINTR) { - return errno; - } + * 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 0; + return buf; } -/* Sends the 'n_iovs' iovecs of data in 'iovs' and the 'n_fds' file descriptors - * in 'fds' on Unix domain socket 'sock'. The function continues to retry - * sending until an error (other than EAGAIN or EINTR) occurs or all the data - * and fds are sent. Upon EAGAIN, the function blocks until the socket is - * ready for more data. - * - * Returns 0 if all the data and fds were successfully sent, otherwise a - * positive errno value. */ -int -send_iovec_and_fds_fully_block(int sock, - const struct iovec iovs[], size_t n_iovs, - const int fds[], size_t n_fds) +size_t +ss_length(const struct sockaddr_storage *ss) { - size_t sent = 0; + switch (ss->ss_family) { + case AF_INET: + return sizeof(struct sockaddr_in); - for (;;) { - int error; + case AF_INET6: + return sizeof(struct sockaddr_in6); - error = send_iovec_and_fds_fully(sock, iovs, n_iovs, - fds, n_fds, sent, &sent); - if (error != EAGAIN) { - return error; - } - poll_fd_wait(sock, POLLOUT); - poll_block(); + default: + OVS_NOT_REACHED(); } } -/* Attempts to receive from Unix domain socket 'sock' up to 'size' bytes of - * data into 'data' and up to SOUTIL_MAX_FDS file descriptors into 'fds'. - * - * - Upon success, returns the number of bytes of data copied into 'data' - * and stores the number of received file descriptors into '*n_fdsp'. +/* 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. * - * - On failure, returns a negative errno value and stores 0 in - * '*n_fdsp'. - * - * - On EOF, returns 0 and stores 0 in '*n_fdsp'. */ -int -recv_data_and_fds(int sock, - void *data, size_t size, - int fds[SOUTIL_MAX_FDS], size_t *n_fdsp) + * 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) { - union { - struct cmsghdr cm; - char control[CMSG_SPACE(SOUTIL_MAX_FDS * sizeof *fds)]; - } cmsg; - struct msghdr msg; - int retval; - struct cmsghdr *p; - size_t i; - - *n_fdsp = 0; - - do { - struct iovec iov; - - iov.iov_base = data; - iov.iov_len = size; - - msg.msg_name = NULL; - msg.msg_namelen = 0; - msg.msg_iov = &iov; - msg.msg_iovlen = 1; - msg.msg_control = &cmsg.cm; - msg.msg_controllen = sizeof cmsg.control; - msg.msg_flags = 0; - - retval = recvmsg(sock, &msg, 0); - } while (retval < 0 && errno == EINTR); - if (retval <= 0) { - return retval < 0 ? -errno : 0; - } - - for (p = CMSG_FIRSTHDR(&msg); p; p = CMSG_NXTHDR(&msg, p)) { - if (p->cmsg_level != SOL_SOCKET || p->cmsg_type != SCM_RIGHTS) { - VLOG_ERR("unexpected control message %d:%d", - p->cmsg_level, p->cmsg_type); - goto error; - } else if (*n_fdsp) { - VLOG_ERR("multiple SCM_RIGHTS received"); - goto error; - } else { - size_t n_fds = (p->cmsg_len - CMSG_LEN(0)) / sizeof *fds; - const int *fds_data = (const int *) CMSG_DATA(p); - - ovs_assert(n_fds > 0); - if (n_fds > SOUTIL_MAX_FDS) { - VLOG_ERR("%zu fds received but only %d supported", - n_fds, SOUTIL_MAX_FDS); - for (i = 0; i < n_fds; i++) { - close(fds_data[i]); - } - goto error; - } - - *n_fdsp = n_fds; - memcpy(fds, fds_data, n_fds * sizeof *fds); - } - } - - return retval; - -error: - for (i = 0; i < *n_fdsp; i++) { - close(fds[i]); - } - *n_fdsp = 0; - return EPROTO; +#ifdef _WIN32 + return ovs_format_message(error); +#else + return ovs_strerror(error); +#endif }