X-Git-Url: http://git.tuebingen.mpg.de/?p=paraslash.git;a=blobdiff_plain;f=net.c;h=2f720b10477aee8214cac5a5223ca4378b29bd5f;hp=61f50dd71041a2a5e80008dbb2da110b75b70a42;hb=bda95f9508b456dcea89d300f6d4104e30ab9f3e;hpb=29a8e4384c08cb4e295e5b82994e62bf247a0ef7 diff --git a/net.c b/net.c index 61f50dd7..2f720b10 100644 --- a/net.c +++ b/net.c @@ -1,5 +1,5 @@ /* - * Copyright (C) 2005-2009 Andre Noll + * Copyright (C) 2005-2011 Andre Noll * * Licensed under the GPL v2. For licencing details see COPYING. */ @@ -25,15 +25,13 @@ #define AI_ADDRCONFIG 0 #endif -#include #include -#include #include "para.h" #include "error.h" -#include "crypt.h" #include "net.h" #include "string.h" +#include "list.h" #include "fd.h" /** @@ -187,6 +185,31 @@ failed: return NULL; } +/** + * Stringify port number, resolve into service name where defined. + * \param port 2-byte port number, in host-byte-order. + * \param transport Transport protocol name (e.g. "udp", "tcp"), or NULL. + * \return Pointer to static result buffer. + * + * \sa getservent(3), services(5), nsswitch.conf(5) + */ +const char *stringify_port(int port, const char *transport) +{ + static char service[NI_MAXSERV]; + + if (port < 0 || port > 0xFFFF) { + snprintf(service, sizeof(service), "undefined (%d)", port); + } else { + struct servent *se = getservbyport(htons(port), transport); + + if (se == NULL) + snprintf(service, sizeof(service), "%u", port); + else + snprintf(service, sizeof(service), "%s", se->s_name); + } + return service; +} + /** * Determine the socket type for a given layer-4 protocol. * @@ -217,14 +240,119 @@ static const char *layer4_name(const unsigned l4type) return "UNKNOWN PROTOCOL"; } +/** + * Flowopts: Transport-layer independent encapsulation of socket options. + * + * These collect individual socket options into a queue, which is disposed of + * directly after makesock(). The 'pre_conn_opt' structure is for internal use + * only and should not be visible elsewhere. + * + * \sa setsockopt(2), makesock() + */ +struct pre_conn_opt { + int sock_level; /**< Second argument to setsockopt() */ + int sock_option; /**< Third argument to setsockopt() */ + char *opt_name; /**< Stringified \a sock_option */ + void *opt_val; /**< Fourth argument to setsockopt() */ + socklen_t opt_len; /**< Fifth argument to setsockopt() */ + + struct list_head node; /**< FIFO, as sockopt order matters. */ +}; + +/** FIFO list of pre-connection socket options to be set */ +struct flowopts { + struct list_head sockopts; +}; + +struct flowopts *flowopt_new(void) +{ + struct flowopts *new = para_malloc(sizeof(*new)); + + INIT_LIST_HEAD(&new->sockopts); + return new; +} + +/** + * Append new socket option to flowopt queue. + * + * \param fo The flowopt queue to append to. + * \param lev Level at which \a opt resides. + * \param opt New option to add. + * \param name Stringified name of \a opt. + * \param val The value to set \a opt to. + * \param len Length of \a val. + * + * \sa setsockopt(2) + */ +void flowopt_add(struct flowopts *fo, int lev, int opt, + const char *name, const void *val, int len) +{ + struct pre_conn_opt *new = para_malloc(sizeof(*new)); + + new->sock_option = opt; + new->sock_level = lev; + new->opt_name = para_strdup(name); + + if (val == NULL) { + new->opt_val = NULL; + new->opt_len = 0; + } else { + new->opt_val = para_malloc(len); + new->opt_len = len; + memcpy(new->opt_val, val, len); + } + + list_add_tail(&new->node, &fo->sockopts); +} + +void flowopt_add_bool(struct flowopts *fo, int lev, int opt, + const char *optname, bool on_or_off) +{ + int on = on_or_off; /* kernel takes 'int' */ + + flowopt_add(fo, lev, opt, optname, &on, sizeof(on)); +} + +/** Set the entire bunch of pre-connection options at once. */ +static void flowopt_setopts(int sockfd, struct flowopts *fo) +{ + struct pre_conn_opt *pc; + + if (fo == NULL) + return; + + list_for_each_entry(pc, &fo->sockopts, node) + if (setsockopt(sockfd, pc->sock_level, pc->sock_option, + pc->opt_val, pc->opt_len) < 0) { + PARA_EMERG_LOG("Can not set %s socket option: %s", + pc->opt_name, strerror(errno)); + exit(EXIT_FAILURE); + } +} + +static void flowopt_cleanup(struct flowopts *fo) +{ + struct pre_conn_opt *cur, *next; + + if (fo == NULL) + return; + + list_for_each_entry_safe(cur, next, &fo->sockopts, node) { + free(cur->opt_name); + free(cur->opt_val); + free(cur); + } + free(fo); +} + /** * Resolve IPv4/IPv6 address and create a ready-to-use active or passive socket. * - * \param l3type The layer-3 type (\p AF_INET, \p AF_INET6, \p AF_UNSPEC). * \param l4type The layer-4 type (\p IPPROTO_xxx). * \param passive Whether this is a passive (1) or active (0) socket. * \param host Remote or local hostname or IPv/6 address string. * \param port_number Decimal port number. + * \param fo Socket options to be set before making the connection. * * This creates a ready-made IPv4/v6 socket structure after looking up the * necessary parameters. The interpretation of \a host depends on the value of @@ -238,18 +366,20 @@ static const char *layer4_name(const unsigned l4type) * * Furthermore, bind(2) is called on passive sockets, and connect(2) on active * sockets. The algorithm tries all possible address combinations until it - * succeeds. + * succeeds. If \a fo is supplied, options are set and cleanup is performed. * * \return This function returns 1 on success and \a -E_ADDRESS_LOOKUP when no * matching connection could be set up (with details in the error log). * * \sa ipv6(7), getaddrinfo(3), bind(2), connect(2). */ -int makesock(unsigned l3type, unsigned l4type, int passive, - const char *host, unsigned short port_number) +int makesock(unsigned l4type, bool passive, + const char *host, uint16_t port_number, + struct flowopts *fo) { - struct addrinfo *local = NULL, *src, - *remote = NULL, *dst, hints; + struct addrinfo *local = NULL, *src = NULL, *remote = NULL, + *dst = NULL, hints; + unsigned int l3type = AF_UNSPEC; int rc, on = 1, sockfd = -1, socktype = sock_type(l4type); char port[6]; /* port number has at most 5 digits */ @@ -281,7 +411,8 @@ int makesock(unsigned l3type, unsigned l4type, int passive, layer4_name(l4type), host? host : (passive? "[loopback]" : "[localhost]"), port, gai_strerror(rc)); - return -E_ADDRESS_LOOKUP; + rc = -E_ADDRESS_LOOKUP; + goto out; } /* Iterate over all src/dst combination, exhausting dst first */ @@ -296,16 +427,20 @@ int makesock(unsigned l3type, unsigned l4type, int passive, goto get_next_dst; /* - * Set those options that need to be set before establishing - * the connection. Reuse the address on passive (listening) - * sockets to avoid failure on restart. + * Reuse the address on passive sockets to avoid failure on + * restart (protocols using listen()) and when creating + * multiple listener instances (UDP multicast). */ if (passive && setsockopt(sockfd, SOL_SOCKET, SO_REUSEADDR, - &on, sizeof(on)) == -1) { + &on, sizeof(on)) == -1) { + rc = errno; + close(sockfd); PARA_ERROR_LOG("can not set SO_REUSEADDR: %s\n", - strerror(errno)); - return -ERRNO_TO_PARA_ERROR(errno); + strerror(rc)); + rc = -ERRNO_TO_PARA_ERROR(rc); + break; } + flowopt_setopts(sockfd, fo); if (src) { if (bind(sockfd, src->ai_addr, src->ai_addrlen) < 0) { @@ -326,16 +461,20 @@ get_next_src: if (src && (src = src->ai_next)) /* restart inner loop */ dst = remote; } +out: if (local) freeaddrinfo(local); if (remote) freeaddrinfo(remote); + flowopt_cleanup(fo); if (src == NULL && dst == NULL) { + if (rc >= 0) + rc = -E_MAKESOCK; PARA_ERROR_LOG("can not create %s socket %s#%s.\n", layer4_name(l4type), host? host : (passive? "[loopback]" : "[localhost]"), port); - return -ERRNO_TO_PARA_ERROR(errno); + return rc; } return sockfd; } @@ -343,24 +482,25 @@ get_next_src: /** * Create a passive / listening socket. * - * \param l3type The network-layer type (\p AF_xxx). * \param l4type The transport-layer type (\p IPPROTO_xxx). * \param port The decimal port number to listen on. + * \param fo Flowopts (if any) to set before starting to listen. * * \return Positive integer (socket descriptor) on success, negative value * otherwise. * * \sa makesock(), ip(7), ipv6(7), bind(2), listen(2). */ -int para_listen(unsigned l3type, unsigned l4type, unsigned short port) +int para_listen(unsigned l4type, uint16_t port, struct flowopts *fo) { - int ret, fd = makesock(l3type, l4type, 1, NULL, port); + int ret, fd = makesock(l4type, 1, NULL, port, fo); if (fd > 0) { ret = listen(fd, BACKLOG); if (ret < 0) { + ret = errno; close(fd); - return -ERRNO_TO_PARA_ERROR(errno); + return -ERRNO_TO_PARA_ERROR(ret); } PARA_INFO_LOG("listening on %s port %u, fd %d\n", layer4_name(l4type), port, fd); @@ -369,28 +509,89 @@ int para_listen(unsigned l3type, unsigned l4type, unsigned short port) } /** - * Print numeric host and port number (beware - uses static char). - * - * \param sa The IPv4/IPv6 socket address to use. - * \param len The length of \p sa. + * Determine IPv4/v6 socket address length. + * \param sa Container of IPv4 or IPv6 address. + * \return Address-family dependent address length. + */ +static socklen_t salen(const struct sockaddr *sa) +{ + assert(sa->sa_family == AF_INET || sa->sa_family == AF_INET6); + + return sa->sa_family == AF_INET6 + ? sizeof(struct sockaddr_in6) + : sizeof(struct sockaddr_in); +} + +/** True if @ss holds a v6-mapped-v4 address (RFC 4291, 2.5.5.2) */ +static bool SS_IS_ADDR_V4MAPPED(const struct sockaddr_storage *ss) +{ + const struct sockaddr_in6 *ia6 = (const struct sockaddr_in6 *)ss; + + return ss->ss_family == AF_INET6 && IN6_IS_ADDR_V4MAPPED(&ia6->sin6_addr); +} + +/** + * Process IPv4/v6 address, turn v6-mapped-v4 address into normal IPv4 address. + * \param ss Container of IPv4/6 address. + * \return Pointer to normalized address (may be static storage). * - * \sa getnameinfo(3). + * \sa RFC 3493 */ -static char *host_and_port(struct sockaddr *sa, socklen_t len) +static const struct sockaddr * +normalize_ip_address(const struct sockaddr_storage *ss) { - static char output[NI_MAXHOST + NI_MAXSERV + 2]; - char hbuf[NI_MAXHOST], sbuf[NI_MAXSERV]; - int ret; + assert(ss->ss_family == AF_INET || ss->ss_family == AF_INET6); - ret = getnameinfo(sa, len, hbuf, sizeof(hbuf), sbuf, sizeof(sbuf), - NI_NUMERICHOST | NI_NUMERICSERV); - if (ret) { - PARA_WARNING_LOG("hostname lookup error (%s).\n", - gai_strerror(ret)); - sprintf(output, "(unknown)"); - } else - sprintf(output, "%s#%s", hbuf, sbuf); - return output; + if (SS_IS_ADDR_V4MAPPED(ss)) { + const struct sockaddr_in6 *ia6 = (const struct sockaddr_in6 *)ss; + static struct sockaddr_in ia; + + ia.sin_family = AF_INET; + ia.sin_port = ia6->sin6_port; + memcpy(&ia.sin_addr.s_addr, &(ia6->sin6_addr.s6_addr[12]), 4); + return (const struct sockaddr *)&ia; + } + return (const struct sockaddr *)ss; +} + +/** + * Generic/fallback MTU values + * + * These are taken from RFC 1122, RFC 2460, and RFC 5405. + * - RFC 1122, 3.3.3 defines EMTU_S ("Effective MTU for sending") and recommends + * to use an EMTU_S size of of 576 bytes if the IPv4 path MTU is unknown; + * - RFC 2460, 5. requires a minimum IPv6 MTU of 1280 bytes; + * - RFC 5405, 3.2 recommends that if path MTU discovery is not done, + * UDP senders should use the respective minimum values of EMTU_S. + */ +static inline int generic_mtu(const int af_type) +{ + return af_type == AF_INET6 ? 1280 : 576; +} + +/** Crude approximation of IP header overhead - neglecting options. */ +static inline int estimated_header_overhead(const int af_type) +{ + return af_type == AF_INET6 ? 40 : 20; +} + +/** + * Maximum transport-layer message size (MMS_S) as per RFC 1122, 3.3.3 + * Socket must be connected. + */ +int generic_max_transport_msg_size(int sockfd) +{ + struct sockaddr_storage ss; + socklen_t sslen = sizeof(ss); + int af_type = AF_INET; + + if (getpeername(sockfd, (struct sockaddr *)&ss, &sslen) < 0) { + PARA_ERROR_LOG("can not determine remote address type: %s\n", + strerror(errno)); + } else if (!SS_IS_ADDR_V4MAPPED(&ss)) { + af_type = ss.ss_family; + } + return generic_mtu(af_type) - estimated_header_overhead(af_type); } /** @@ -401,23 +602,39 @@ static char *host_and_port(struct sockaddr *sa, socklen_t len) * remote side. * * \return A static character string identifying hostname and port of the - * chosen side. + * chosen side in numeric host:port format. * - * \sa getsockname(2), getpeername(2). + * \sa getsockname(2), getpeername(2), parse_url(), getnameinfo(3), + * services(5), nsswitch.conf(5). */ static char *__get_sock_name(int fd, int (*getname)(int, struct sockaddr*, socklen_t *)) { struct sockaddr_storage ss; + const struct sockaddr *sa; socklen_t sslen = sizeof(ss); + char hbuf[NI_MAXHOST], sbuf[NI_MAXSERV]; + static char output[sizeof(hbuf) + sizeof(sbuf) + 4]; + int ret; if (getname(fd, (struct sockaddr *)&ss, &sslen) < 0) { - static char *dont_know = "(don't know)"; PARA_ERROR_LOG("can not determine address from fd %d: %s\n", fd, strerror(errno)); - return dont_know; + snprintf(output, sizeof(output), "(unknown)"); + return output; } - return host_and_port((struct sockaddr *)&ss, sslen); + sa = normalize_ip_address(&ss); + ret = getnameinfo(sa, salen(sa), hbuf, sizeof(hbuf), sbuf, + sizeof(sbuf), NI_NUMERICHOST | NI_NUMERICSERV); + if (ret) { + PARA_WARNING_LOG("hostname lookup error (%s).\n", + gai_strerror(ret)); + snprintf(output, sizeof(output), "(lookup error)"); + } else if (sa->sa_family == AF_INET6) + snprintf(output, sizeof(output), "[%s]:%s", hbuf, sbuf); + else + snprintf(output, sizeof(output), "%s:%s", hbuf, sbuf); + return output; } /** @@ -460,15 +677,10 @@ char *remote_name(int sockfd) struct in_addr extract_v4_addr(const struct sockaddr_storage *ss) { struct in_addr ia = {.s_addr = 0}; + const struct sockaddr *sa = normalize_ip_address(ss); - if (ss->ss_family == AF_INET) - ia.s_addr = ((struct sockaddr_in *)ss)->sin_addr.s_addr; - if (ss->ss_family == AF_INET6) { - const struct in6_addr v6_addr = ((struct sockaddr_in6 *)ss)->sin6_addr; - - if (IN6_IS_ADDR_V4MAPPED(&v6_addr)) - memcpy(&ia.s_addr, &(v6_addr.s6_addr[12]), 4); - } + if (sa->sa_family == AF_INET) + ia = ((struct sockaddr_in *)sa)->sin_addr; return ia; } @@ -580,23 +792,68 @@ int recv_buffer(int fd, char *buf, size_t size) * Wrapper around the accept system call. * * \param fd The listening socket. + * \param rfds An optional fd_set pointer. * \param addr Structure which is filled in with the address of the peer socket. * \param size Should contain the size of the structure pointed to by \a addr. + * \param new_fd Result pointer. * - * Accept incoming connections on \a addr. Retry if interrupted. + * Accept incoming connections on \a addr, retry if interrupted. If \a rfds is + * not \p NULL, return 0 if \a fd is not set in \a rfds without calling accept(). * - * \return The new file descriptor on success, negative on errors. + * \return Negative on errors, zero if no connections are present to be accepted, + * one otherwise. * * \sa accept(2). */ -int para_accept(int fd, void *addr, socklen_t size) +int para_accept(int fd, fd_set *rfds, void *addr, socklen_t size, int *new_fd) { - int new_fd; + int ret; + if (rfds && !FD_ISSET(fd, rfds)) + return 0; do - new_fd = accept(fd, (struct sockaddr *) addr, &size); - while (new_fd < 0 && errno == EINTR); - return new_fd < 0? -ERRNO_TO_PARA_ERROR(errno) : new_fd; + ret = accept(fd, (struct sockaddr *) addr, &size); + while (ret < 0 && errno == EINTR); + + if (ret >= 0) { + *new_fd = ret; + return 1; + } + if (errno == EAGAIN || errno == EWOULDBLOCK) + return 0; + return -ERRNO_TO_PARA_ERROR(errno); +} + +/** + * Probe the list of DCCP CCIDs configured on this host. + * \param ccid_array Pointer to return statically allocated array in. + * \return Number of elements returned in \a ccid_array or error. + * + * NB: This feature is only available on Linux > 2.6.30; on older kernels + * ENOPROTOOPT ("Protocol not available") will be returned. + */ +int dccp_available_ccids(uint8_t **ccid_array) +{ + static uint8_t ccids[DCCP_MAX_HOST_CCIDS]; + socklen_t nccids = sizeof(ccids); + int ret, fd; + + ret = fd = makesock(IPPROTO_DCCP, 1, NULL, 0, NULL); + if (ret < 0) + return ret; + + if (getsockopt(fd, SOL_DCCP, DCCP_SOCKOPT_AVAILABLE_CCIDS, + ccids, &nccids) < 0) { + ret = errno; + close(fd); + PARA_ERROR_LOG("No DCCP_SOCKOPT_AVAILABLE_CCIDS: %s\n", + strerror(ret)); + return -ERRNO_TO_PARA_ERROR(ret); + } + + close(fd); + *ccid_array = ccids; + return nccids; } /** @@ -670,15 +927,17 @@ err: * This function creates a local socket for sequenced, reliable, two-way, * connection-based byte streams. * - * \return The file descriptor, on success, negative on errors. + * \return The file descriptor of the connected socket on success, negative on + * errors. * * \sa create_local_socket(), unix(7), connect(2). */ -int create_remote_socket(const char *name) +int connect_local_socket(const char *name) { struct sockaddr_un unix_addr; int fd, ret; + PARA_DEBUG_LOG("connecting to %s\n", name); ret = init_unix_addr(&unix_addr, name); if (ret < 0) return ret; @@ -809,40 +1068,3 @@ int recv_cred_buffer(int fd, char *buf, size_t size) return result; } #endif /* HAVE_UCRED */ - -/** - * Receive a buffer and check for a pattern. - * - * \param fd The file descriptor to receive from. - * \param pattern The expected pattern. - * \param bufsize The size of the internal buffer. - * - * \return Positive if \a pattern was received, negative otherwise. - * - * This function tries to receive at most \a bufsize bytes from file descriptor - * \a fd. If at least \p strlen(\a pattern) bytes were received, the beginning - * of the received buffer is compared with \a pattern, ignoring case. - * - * \sa recv_buffer(), \sa strncasecmp(3). - */ -int recv_pattern(int fd, const char *pattern, size_t bufsize) -{ - size_t len = strlen(pattern); - char *buf = para_malloc(bufsize + 1); - int ret = -E_RECV_PATTERN, n = recv_buffer(fd, buf, bufsize + 1); - - if (n < len) - goto out; - if (strncasecmp(buf, pattern, len)) - goto out; - ret = 1; -out: - if (ret < 0) { - PARA_NOTICE_LOG("n = %d, did not receive pattern '%s'\n", n, - pattern); - if (n > 0) - PARA_NOTICE_LOG("recvd: %s\n", buf); - } - free(buf); - return ret; -}