X-Git-Url: http://git.tuebingen.mpg.de/?p=paraslash.git;a=blobdiff_plain;f=net.c;h=2369dc8ce99027a4bb122a9a9d3e71bb5aa10f8b;hp=61f50dd71041a2a5e80008dbb2da110b75b70a42;hb=fdbdd1fd575d7d37d4fb182252107a4d3816853c;hpb=dd462c7e1b61864d6da01a877ced78dab72e2244 diff --git a/net.c b/net.c index 61f50dd7..463033bb 100644 --- a/net.c +++ b/net.c @@ -1,5 +1,5 @@ /* - * Copyright (C) 2005-2009 Andre Noll + * Copyright (C) 2005 Andre Noll * * Licensed under the GPL v2. For licencing details see COPYING. */ @@ -12,6 +12,11 @@ */ #define _GNU_SOURCE +#include +#include +#include +#include +#include #include /* At least NetBSD needs these. */ @@ -25,15 +30,13 @@ #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" /** @@ -45,7 +48,7 @@ * \param netmask Value of the netmask part in \a cidr or the * default of 32 if not specified. * - * \return Pointer to \a addr if succesful, NULL on error. + * \return Pointer to \a addr if successful, NULL on error. * \sa RFC 4632 */ char *parse_cidr(const char *cidr, @@ -144,9 +147,9 @@ static bool host_string_ok(const char *host) * \param hostlen The maximum length of \a host. * \param port To return the port number (if any) of \a url. * - * \return Pointer to \a host, or NULL if failed. - * If NULL is returned, \a host and \a portnum are undefined. If no - * port number was present in \a url, \a portnum is set to -1. + * \return Pointer to \a host, or \p NULL if failed. If \p NULL is returned, + * \a host and \a port are undefined. If no port number was present in \a url, + * \a port is set to -1. * * \sa RFC 3986, 3.2.2/3.2.3 */ @@ -170,16 +173,16 @@ char *parse_url(const char *url, if (*o++ != ']' || (*o != '\0' && *o != ':')) goto failed; } else { - for (; (*c = *o == ':'? '\0' : *o); c++, o++) - if (c == end) + for (; (*c = *o == ':'? '\0' : *o); c++, o++) { + if (c == end && o[1]) goto failed; + } } if (*o == ':') if (para_atoi32(++o, port) < 0 || *port < 0 || *port > 0xffff) goto failed; - if (host_string_ok(host)) return host; failed: @@ -187,6 +190,31 @@ 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), "%d", port); + else + snprintf(service, sizeof(service), "%s", se->s_name); + } + return service; +} + /** * Determine the socket type for a given layer-4 protocol. * @@ -218,149 +246,282 @@ static const char *layer4_name(const unsigned l4type) } /** - * Resolve IPv4/IPv6 address and create a ready-to-use active or passive socket. + * 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; +}; + +/** + * Allocate and initialize a flowopt queue. + * + * \return A new structure to be passed to \ref flowopt_add(). It is + * automatically deallocated in \ref makesock(). + */ +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); +} + +/** 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); + } +} + +/** + * Deallocate all resources of a flowopts structure. + * + * \param fo A pointer as returned from flowopt_new(). + * + * It's OK to pass \p NULL here in which case the function does nothing. + */ +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 an IPv4/IPv6 address. * - * \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 passive Whether \p AI_PASSIVE should be included as hint. * \param host Remote or local hostname or IPv/6 address string. - * \param port_number Decimal port number. + * \param port_number Used to set the port in each returned address structure. + * \param result addrinfo structures are returned here. * - * 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 - * \a passive: - * - on a passive socket host is interpreted as an interface IPv4/6 address - * (can be left NULL); - * - on an active socket, \a host is the peer DNS name or IPv4/6 address - * to connect to; - * - \a port_number is in either case the numeric port number (not service - * string). - * - * Furthermore, bind(2) is called on passive sockets, and connect(2) on active - * sockets. The algorithm tries all possible address combinations until it - * succeeds. - * - * \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) -{ - struct addrinfo *local = NULL, *src, - *remote = NULL, *dst, hints; - int rc, on = 1, sockfd = -1, - socktype = sock_type(l4type); + * The interpretation of \a host depends on the value of \a passive. On a + * passive socket host is interpreted as an interface IPv4/6 address (can be + * left NULL). On an active socket, \a host is the peer DNS name or IPv4/6 + * address to connect to. + * + * \return Standard. + * + * \sa getaddrinfo(3). + */ +int lookup_address(unsigned l4type, bool passive, const char *host, + int port_number, struct addrinfo **result) +{ + int ret; char port[6]; /* port number has at most 5 digits */ + struct addrinfo *addr = NULL, hints; - sprintf(port, "%u", port_number); + *result = NULL; + sprintf(port, "%u", port_number & 0xffff); /* Set up address hint structure */ memset(&hints, 0, sizeof(hints)); - hints.ai_family = l3type; - hints.ai_socktype = socktype; - /* + hints.ai_family = AF_UNSPEC; + hints.ai_socktype = sock_type(l4type); + /* * getaddrinfo does not support SOCK_DCCP, so for the sake of lookup * (and only then) pretend to be UDP. */ if (l4type == IPPROTO_DCCP) hints.ai_socktype = SOCK_DGRAM; - /* only use addresses available on the host */ hints.ai_flags = AI_ADDRCONFIG; - if (l3type == AF_INET6) - /* use v4-mapped-v6 if no v6 addresses found */ - hints.ai_flags |= AI_V4MAPPED | AI_ALL; - if (passive && host == NULL) hints.ai_flags |= AI_PASSIVE; - /* Obtain local/remote address information */ - if ((rc = getaddrinfo(host, port, &hints, passive ? &local : &remote))) { - PARA_ERROR_LOG("can not resolve %s address %s#%s: %s.\n", - layer4_name(l4type), - host? host : (passive? "[loopback]" : "[localhost]"), - port, gai_strerror(rc)); + ret = getaddrinfo(host, port, &hints, &addr); + if (ret != 0) { + PARA_ERROR_LOG("can not resolve %s address %s#%s: %s\n", + layer4_name(l4type), + host? host : (passive? "[loopback]" : "[localhost]"), + port, gai_strerror(ret)); return -E_ADDRESS_LOOKUP; } + *result = addr; + return 1; +} - /* Iterate over all src/dst combination, exhausting dst first */ - for (src = local, dst = remote; src != NULL || dst != NULL; /* no op */ ) { - if (src && dst && src->ai_family == AF_INET - && dst->ai_family == AF_INET6) - goto get_next_dst; /* v4 -> v6 is not possible */ - - sockfd = socket(src ? src->ai_family : dst->ai_family, - socktype, l4type); - if (sockfd < 0) - goto get_next_dst; +/** + * Create an active or passive socket. + * + * \param l4type \p IPPROTO_TCP, \p IPPROTO_UDP, or \p IPPROTO_DCCP. + * \param passive Whether to call bind(2) or connect(2). + * \param ai Address information as obtained from \ref lookup_address(). + * \param fo Socket options to be set before making the connection. + * + * bind(2) is called on passive sockets, and connect(2) on active sockets. The + * algorithm tries all possible address combinations until it succeeds. If \a + * fo is supplied, options are set but cleanup must be performed in the caller. + * + * \return File descriptor on success, \p E_MAKESOCK on errors. + * + * \sa \ref lookup_address(), \ref makesock(), ip(7), ipv6(7), bind(2), + * connect(2). + */ +int makesock_addrinfo(unsigned l4type, bool passive, struct addrinfo *ai, + struct flowopts *fo) +{ + int ret = -E_MAKESOCK, on = 1; + for (; ai; ai = ai->ai_next) { + int fd; + ret = socket(ai->ai_family, sock_type(l4type), l4type); + if (ret < 0) + continue; + fd = ret; + flowopt_setopts(fd, fo); + if (!passive) { + if (connect(fd, ai->ai_addr, ai->ai_addrlen) == 0) + return fd; + close(fd); + continue; + } /* - * 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) { - PARA_ERROR_LOG("can not set SO_REUSEADDR: %s\n", - strerror(errno)); - return -ERRNO_TO_PARA_ERROR(errno); - } - - if (src) { - if (bind(sockfd, src->ai_addr, src->ai_addrlen) < 0) { - close(sockfd); - goto get_next_src; - } - if (!dst) /* bind-only completed successfully */ - break; + if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &on, + sizeof(on)) == -1) { + close(fd); + continue; } - - if (dst && connect(sockfd, dst->ai_addr, dst->ai_addrlen) == 0) - break; /* connection completed successfully */ - close(sockfd); -get_next_dst: - if (dst && (dst = dst->ai_next)) + if (bind(fd, ai->ai_addr, ai->ai_addrlen) < 0) { + close(fd); continue; -get_next_src: - if (src && (src = src->ai_next)) /* restart inner loop */ - dst = remote; + } + return fd; } - if (local) - freeaddrinfo(local); - if (remote) - freeaddrinfo(remote); - - if (src == NULL && dst == NULL) { - 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 -E_MAKESOCK; +} + +/** + * Resolve IPv4/IPv6 address and create a ready-to-use active or passive socket. + * + * \param l4type The layer-4 type (\p IPPROTO_xxx). + * \param passive Whether this is a passive or active socket. + * \param host Passed to \ref lookup_address(). + * \param port_number Passed to \ref lookup_address(). + * \param fo Passed to \ref makesock_addrinfo(). + * + * This creates a ready-made IPv4/v6 socket structure after looking up the + * necessary parameters. The function first calls \ref lookup_address() and + * passes the address information to makesock_addrinfo() to create and + * initialize the socket. + * + * \return The newly created file descriptor on success, a negative error code + * on failure. + * + * \sa \ref lookup_address(), \ref makesock_addrinfo(). + */ +int makesock(unsigned l4type, bool passive, const char *host, uint16_t port_number, + struct flowopts *fo) +{ + struct addrinfo *ai; + int ret = lookup_address(l4type, passive, host, port_number, &ai); + + if (ret >= 0) + ret = makesock_addrinfo(l4type, passive, ai, fo); + if (ai) + freeaddrinfo(ai); + if (ret < 0) { + PARA_ERROR_LOG("can not create %s socket %s#%d.\n", + layer4_name(l4type), host? host : (passive? + "[loopback]" : "[localhost]"), port_number); } - return sockfd; + return ret; } /** * 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); @@ -369,160 +530,178 @@ 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. - * - * \sa getnameinfo(3). + * Determine IPv4/v6 socket address length. + * \param sa Container of IPv4 or IPv6 address. + * \return Address-family dependent address length. */ -static char *host_and_port(struct sockaddr *sa, socklen_t len) +static socklen_t salen(const struct sockaddr *sa) { - static char output[NI_MAXHOST + NI_MAXSERV + 2]; - char hbuf[NI_MAXHOST], sbuf[NI_MAXSERV]; - int ret; + assert(sa->sa_family == AF_INET || sa->sa_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; + 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); } /** - * Look up the local or remote side of a connected socket structure. - * - * \param fd The socket descriptor of the connected socket. - * \param getname Either \p getsockname() for local, or \p getpeername() for - * remote side. - * - * \return A static character string identifying hostname and port of the - * chosen side. + * 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 getsockname(2), getpeername(2). + * \sa RFC 3493 */ -static char *__get_sock_name(int fd, int (*getname)(int, struct sockaddr*, - socklen_t *)) +static const struct sockaddr * +normalize_ip_address(const struct sockaddr_storage *ss) { - struct sockaddr_storage ss; - socklen_t sslen = sizeof(ss); + assert(ss->ss_family == AF_INET || ss->ss_family == AF_INET6); - 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; + 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 host_and_port((struct sockaddr *)&ss, sslen); + return (const struct sockaddr *)ss; } /** - * Look up the local side of a connected socket structure. - * - * \param sockfd The file descriptor of the socket. - * - * \return A pointer to a static buffer containing hostname an port. This - * buffer must not be freed by the caller. - * - * \sa remote_name(). + * 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. */ -char *local_name(int sockfd) +static inline int generic_mtu(const int af_type) { - return __get_sock_name(sockfd, getsockname); + return af_type == AF_INET6 ? 1280 : 576; } -/** - * Look up the remote side of a connected socket structure. - * - * \param sockfd The file descriptor of the socket. - * - * \return Analogous to the return value of \ref local_name() but for the - * remote side. - * - * \sa local_name(). - */ -char *remote_name(int sockfd) +/** Crude approximation of IP header overhead - neglecting options. */ +static inline int estimated_header_overhead(const int af_type) { - return __get_sock_name(sockfd, getpeername); + return af_type == AF_INET6 ? 40 : 20; } /** - * Extract IPv4 or IPv6-mapped-IPv4 address from sockaddr_storage. - * \param ss Container of IPv4/6 address - * \return Extracted IPv4 address (different from 0) or 0 if unsuccessful. + * Get the maximum transport-layer message size (MMS_S). * - * \sa RFC 3493 + * \param sockfd The socket file descriptor. + * + * The socket must be connected. See RFC 1122, 3.3.3. If the protocol family + * could not be determined, \p AF_INET is assumed. + * + * \return The maximum message size of the address family type. */ -struct in_addr extract_v4_addr(const struct sockaddr_storage *ss) +int generic_max_transport_msg_size(int sockfd) { - struct in_addr ia = {.s_addr = 0}; - - 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; + struct sockaddr_storage ss; + socklen_t sslen = sizeof(ss); + int af_type = AF_INET; - if (IN6_IS_ADDR_V4MAPPED(&v6_addr)) - memcpy(&ia.s_addr, &(v6_addr.s6_addr[12]), 4); + 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 ia; + return generic_mtu(af_type) - estimated_header_overhead(af_type); } /** - * Send a binary buffer. + * Look up the remote side of a connected socket structure. * - * \param fd The file descriptor. - * \param buf The buffer to be sent. - * \param len The length of \a buf. + * \param fd The socket descriptor of the connected socket. * - * Send out the buffer and try to resend the remaining part in case of short - * writes. + * \return A static character string identifying hostname and port of the + * chosen side in numeric host:port format. * - * \return Standard. + * \sa getsockname(2), getpeername(2), parse_url(), getnameinfo(3), + * services(5), nsswitch.conf(5). */ -int send_bin_buffer(int fd, const char *buf, size_t len) +char *remote_name(int fd) { - if (!len) - PARA_CRIT_LOG("len == 0\n"); - return write_all(fd, buf, &len); + 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 (getpeername(fd, (struct sockaddr *)&ss, &sslen) < 0) { + PARA_ERROR_LOG("can not determine address from fd %d: %s\n", + fd, strerror(errno)); + snprintf(output, sizeof(output), "(unknown)"); + return output; + } + 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; } /** - * Send a \p NULL-terminated buffer. - * - * \param fd The file descriptor. - * \param buf The null-terminated buffer to be send. + * Extract IPv4 or IPv6-mapped-IPv4 address from sockaddr_storage. * - * This is equivalent to send_bin_buffer(fd, buf, strlen(buf)). + * \param ss Container of IPv4/6 address. + * \param ia Extracted IPv4 address (different from 0) or 0 if unsuccessful. * - * \return Standard. + * \sa RFC 3493. */ -int send_buffer(int fd, const char *buf) +void extract_v4_addr(const struct sockaddr_storage *ss, struct in_addr *ia) { - return send_bin_buffer(fd, buf, strlen(buf)); + const struct sockaddr *sa = normalize_ip_address(ss); + + memset(ia, 0, sizeof(*ia)); + if (sa->sa_family == AF_INET) + *ia = ((struct sockaddr_in *)sa)->sin_addr; } /** - * Send a buffer given by a format string. + * Compare the address part of IPv4/6 addresses. * - * \param fd The file descriptor. - * \param fmt A format string. + * \param sa1 First address. + * \param sa2 Second address. * - * \return Standard. + * \return True iff the IP address of \a sa1 and \a sa2 match. */ -__printf_2_3 int send_va_buffer(int fd, const char *fmt, ...) +bool sockaddr_equal(const struct sockaddr *sa1, const struct sockaddr *sa2) { - char *msg; - int ret; - - PARA_VSPRINTF(fmt, msg); - ret = send_buffer(fd, msg); - free(msg); - return ret; + if (!sa1 || !sa2) + return false; + if (sa1->sa_family != sa2->sa_family) + return false; + if (sa1->sa_family == AF_INET) { + struct sockaddr_in *a1 = (typeof(a1))sa1, + *a2 = (typeof (a2))sa2; + return a1->sin_addr.s_addr == a2->sin_addr.s_addr; + } else if (sa1->sa_family == AF_INET6) { + struct sockaddr_in6 *a1 = (typeof(a1))sa1, + *a2 = (typeof (a2))sa2; + return !memcmp(a1, a2, sizeof(*a1)); + } else + return false; } /** @@ -580,23 +759,68 @@ 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; } /** @@ -610,52 +834,66 @@ int para_accept(int fd, void *addr, socklen_t size) * \return Positive on success, \p -E_NAME_TOO_LONG if \a name is longer * than \p UNIX_PATH_MAX. */ -static int init_unix_addr(struct sockaddr_un *u, const char *name) +static int init_unix_addr(struct sockaddr_un *u, const char *name, + bool abstract) { - if (strlen(name) >= UNIX_PATH_MAX) + if (strlen(name) + abstract >= UNIX_PATH_MAX) return -E_NAME_TOO_LONG; memset(u->sun_path, 0, UNIX_PATH_MAX); u->sun_family = PF_UNIX; - strcpy(u->sun_path, name); + strcpy(u->sun_path + abstract, name); return 1; } /** - * Prepare, create, and bind a socket for local communication. + * Create a socket for local communication and listen on it. * * \param name The socket pathname. - * \param unix_addr Pointer to the \p AF_UNIX socket structure. - * \param mode The desired mode of the socket. + * \param mode The desired permissions of the socket. * - * This function creates a local socket for sequenced, reliable, - * two-way, connection-based byte streams. + * This function creates a passive local socket for sequenced, reliable, + * two-way, connection-based byte streams. The socket file descriptor is set to + * nonblocking mode and listen(2) is called to prepare the socket for + * accepting incoming connection requests. * - * \return The file descriptor, on success, negative on errors. + * If mode is zero, an abstract socket (a non-portable Linux extension) is + * created. In this case the socket name has no connection with filesystem + * pathnames. * - * \sa socket(2) - * \sa bind(2) - * \sa chmod(2) + * \return The file descriptor on success, negative error code on failure. + * + * \sa socket(2), \sa bind(2), \sa chmod(2), listen(2), unix(7). */ -int create_local_socket(const char *name, struct sockaddr_un *unix_addr, - mode_t mode) +int create_local_socket(const char *name, mode_t mode) { + struct sockaddr_un unix_addr; int fd, ret; + bool abstract = mode == 0; - ret = init_unix_addr(unix_addr, name); + ret = init_unix_addr(&unix_addr, name, abstract); if (ret < 0) return ret; ret = socket(PF_UNIX, SOCK_STREAM, 0); if (ret < 0) return -ERRNO_TO_PARA_ERROR(errno); fd = ret; - ret = bind(fd, (struct sockaddr *) unix_addr, UNIX_PATH_MAX); + ret = mark_fd_nonblocking(fd); + if (ret < 0) + goto err; + ret = bind(fd, (struct sockaddr *)&unix_addr, sizeof(unix_addr)); if (ret < 0) { ret = -ERRNO_TO_PARA_ERROR(errno); goto err; } - ret = -E_CHMOD; - if (chmod(name, mode) < 0) + if (!abstract) { + ret = -E_CHMOD; + if (chmod(name, mode) < 0) + goto err; + } + if (listen(fd , 5) < 0) { + ret = -ERRNO_TO_PARA_ERROR(errno); goto err; + } return fd; err: close(fd); @@ -670,26 +908,33 @@ 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; - ret = init_unix_addr(&unix_addr, name); - if (ret < 0) - return ret; + PARA_DEBUG_LOG("connecting to %s\n", name); fd = socket(PF_UNIX, SOCK_STREAM, 0); if (fd < 0) return -ERRNO_TO_PARA_ERROR(errno); - if (connect(fd, (struct sockaddr *)&unix_addr, sizeof(unix_addr)) == -1) { - ret = -ERRNO_TO_PARA_ERROR(errno); + /* first try (linux-only) abstract socket */ + ret = init_unix_addr(&unix_addr, name, true); + if (ret < 0) goto err; - } - return fd; + if (connect(fd, (struct sockaddr *)&unix_addr, sizeof(unix_addr)) != -1) + return fd; + /* next try pathname socket */ + ret = init_unix_addr(&unix_addr, name, false); + if (ret < 0) + goto err; + if (connect(fd, (struct sockaddr *)&unix_addr, sizeof(unix_addr)) != -1) + return fd; + ret = -ERRNO_TO_PARA_ERROR(errno); err: close(fd); return ret; @@ -698,23 +943,25 @@ err: #ifndef HAVE_UCRED ssize_t send_cred_buffer(int sock, char *buf) { - return send_buffer(sock, buf); + return write_buffer(sock, buf); } int recv_cred_buffer(int fd, char *buf, size_t size) { return recv_buffer(fd, buf, size) > 0? 1 : -E_RECVMSG; } #else /* HAVE_UCRED */ + /** - * Send \p NULL-terminated buffer and Unix credentials of the current process. + * Send a buffer and the credentials of the current process to a socket. * - * \param sock The socket file descriptor. - * \param buf The buffer to be sent. + * \param sock The file descriptor of the sending socket. + * \param buf The zero-terminated buffer to send. * - * \return On success, this call returns the number of characters sent. On - * error, \p -E_SENDMSG is returned. + * \return On success, this call returns the number of bytes sent. On errors, + * \p -E_SENDMSG is returned. * - * \sa sendmsg(2), okir's Black Hats Manual. + * \sa \ref recv_cred_buffer, sendmsg(2), socket(7), unix(7), okir's Black Hats + * Manual. */ ssize_t send_cred_buffer(int sock, char *buf) { @@ -727,7 +974,7 @@ ssize_t send_cred_buffer(int sock, char *buf) /* Response data */ iov.iov_base = buf; - iov.iov_len = strlen(buf); + iov.iov_len = strlen(buf); c.pid = getpid(); c.uid = getuid(); c.gid = getgid(); @@ -745,7 +992,7 @@ ssize_t send_cred_buffer(int sock, char *buf) *(struct ucred *)CMSG_DATA(cmsg) = c; msg.msg_controllen = cmsg->cmsg_len; ret = sendmsg(sock, &msg, 0); - if (ret < 0) + if (ret < 0) ret = -E_SENDMSG; return ret; } @@ -761,13 +1008,13 @@ static void dispose_fds(int *fds, unsigned num) /** * Receive a buffer and the Unix credentials of the sending process. * - * \param fd the socket file descriptor. - * \param buf the buffer to store the message. - * \param size the size of \a buffer. + * \param fd The file descriptor of the receiving socket. + * \param buf The buffer to store the received message. + * \param size The length of \a buf in bytes. * - * \return negative on errors, the user id on success. + * \return Negative on errors, the user id of the sending process on success. * - * \sa recvmsg(2), okir's Black Hats Manual. + * \sa \ref send_cred_buffer and the references given there. */ int recv_cred_buffer(int fd, char *buf, size_t size) { @@ -800,7 +1047,7 @@ int recv_cred_buffer(int fd, char *buf, size_t size) } else if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) { - dispose_fds((int *) CMSG_DATA(cmsg), + dispose_fds((int *)CMSG_DATA(cmsg), (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int)); } @@ -809,40 +1056,3 @@ 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; -}