X-Git-Url: https://git.exim.org/exim.git/blobdiff_plain/3e60dd410992e3b732d5633e25d3cbf76d8b1e2d..cfeaa009b9b407d8b404438afbf4c5d85d61e8c1:/src/src/ip.c diff --git a/src/src/ip.c b/src/src/ip.c index e5945090a..b50130be3 100644 --- a/src/src/ip.c +++ b/src/src/ip.c @@ -2,18 +2,26 @@ * Exim - an Internet mail transport agent * *************************************************/ -/* Copyright (c) University of Cambridge 1995 - 2014 */ +/* Copyright (c) University of Cambridge 1995 - 2018 */ +/* Copyright (c) The Exim Maintainers 2020 - 2021 */ /* See the file NOTICE for conditions of use and distribution. */ +/* SPDX-License-Identifier: GPL-2.0-or-later */ /* Functions for doing things with sockets. With the advent of IPv6 this has got messier, so that it's worth pulling out the code into separate functions -that other parts of Exim can call, expecially as there are now several +that other parts of Exim can call, especially as there are now several different places in the code where sockets are used. */ #include "exim.h" +#if defined(TCP_FASTOPEN) +# if defined(MSG_FASTOPEN) || defined(EXIM_TFO_CONNECTX) || defined(EXIM_TFO_FREEBSD) +# define EXIM_SUPPORT_TFO +# endif +#endif + /************************************************* * Create a socket * *************************************************/ @@ -97,24 +105,11 @@ ip_addrinfo(const uschar *address, struct sockaddr_in6 *saddr) * Bind socket to interface and port * *************************************************/ -/* This function binds a socket to a local interface address and port. For a -wildcard IPv6 bind, the address is ":". - -Arguments: - sock the socket - af AF_INET or AF_INET6 - the socket type - address the IP address, in text form - port the IP port (host order) - -Returns: the result of bind() -*/ - int -ip_bind(int sock, int af, uschar *address, int port) +ip_addr(void * sin_, int af, const uschar * address, int port) { -int s_len; -union sockaddr_46 sin; -memset(&sin, 0, sizeof(sin)); +union sockaddr_46 * sin = sin_; +memset(sin, 0, sizeof(*sin)); /* Setup code when using an IPv6 socket. The wildcard address is ":", to ensure an IPv6 socket is used. */ @@ -124,35 +119,48 @@ if (af == AF_INET6) { if (address[0] == ':' && address[1] == 0) { - sin.v6.sin6_family = AF_INET6; - sin.v6.sin6_addr = in6addr_any; + sin->v6.sin6_family = AF_INET6; + sin->v6.sin6_addr = in6addr_any; } else - { - ip_addrinfo(address, &sin.v6); /* Panic-dies on error */ - } - sin.v6.sin6_port = htons(port); - s_len = sizeof(sin.v6); + ip_addrinfo(address, &sin->v6); /* Panic-dies on error */ + sin->v6.sin6_port = htons(port); + return sizeof(sin->v6); } else -#else /* HAVE_IPv6 */ -af = af; /* Avoid compiler warning */ #endif /* HAVE_IPV6 */ /* Setup code when using IPv4 socket. The wildcard address is "". */ { - sin.v4.sin_family = AF_INET; - sin.v4.sin_port = htons(port); - s_len = sizeof(sin.v4); - if (address[0] == 0) - sin.v4.sin_addr.s_addr = (S_ADDR_TYPE)INADDR_ANY; - else - sin.v4.sin_addr.s_addr = (S_ADDR_TYPE)inet_addr(CS address); + sin->v4.sin_family = AF_INET; + sin->v4.sin_port = htons(port); + sin->v4.sin_addr.s_addr = address[0] == 0 + ? (S_ADDR_TYPE)INADDR_ANY + : (S_ADDR_TYPE)inet_addr(CS address); + return sizeof(sin->v4); } +} + + + +/* This function binds a socket to a local interface address and port. For a +wildcard IPv6 bind, the address is ":". + +Arguments: + sock the socket + af AF_INET or AF_INET6 - the socket type + address the IP address, in text form + port the IP port (host order) -/* Now we can call the bind() function */ +Returns: the result of bind() +*/ +int +ip_bind(int sock, int af, uschar *address, int port) +{ +union sockaddr_46 sin; +int s_len = ip_addr(&sin, af, address, port); return bind(sock, (struct sockaddr *)&sin, s_len); } @@ -173,12 +181,15 @@ Arguments: address the remote address, in text form port the remote port timeout a timeout (zero for indefinite timeout) + fastopen_blob non-null iff TCP_FASTOPEN can be used; may indicate early-data to + be sent in SYN segment. Any such data must be idempotent. Returns: 0 on success; -1 on failure, with errno set */ int -ip_connect(int sock, int af, const uschar *address, int port, int timeout) +ip_connect(int sock, int af, const uschar *address, int port, int timeout, + const blob * fastopen_blob) { struct sockaddr_in s_in4; struct sockaddr *s_ptr; @@ -197,8 +208,6 @@ if (af == AF_INET6) s_len = sizeof(s_in6); } else -#else /* HAVE_IPV6 */ -af = af; /* Avoid compiler warning */ #endif /* HAVE_IPV6 */ /* For an IPv4 address, use an IPv4 sockaddr structure, even on a system with @@ -216,34 +225,155 @@ IPv6 support. */ /* If no connection timeout is set, just call connect() without setting a timer, thereby allowing the inbuilt OS timeout to operate. */ +callout_address = string_sprintf("[%s]:%d", address, port); sigalrm_seen = FALSE; -if (timeout > 0) alarm(timeout); -rc = connect(sock, s_ptr, s_len); +if (timeout > 0) ALARM(timeout); + +#ifdef EXIM_SUPPORT_TFO +/* TCP Fast Open, if the system has a cookie from a previous call to +this peer, can send data in the SYN packet. The peer can send data +before it gets our ACK of its SYN,ACK - the latter is useful for +the SMTP banner. Other (than SMTP) cases of TCP connections can +possibly use the data-on-syn, so support that too. */ + +if (fastopen_blob && f.tcp_fastopen_ok) + { +# ifdef MSG_FASTOPEN + /* This is a Linux implementation. */ + + if ((rc = sendto(sock, fastopen_blob->data, fastopen_blob->len, + MSG_FASTOPEN | MSG_DONTWAIT, s_ptr, s_len)) >= 0) + /* seen for with-data, experimental TFO option, with-cookie case */ + /* seen for with-data, proper TFO opt, with-cookie case */ + { + DEBUG(D_transport|D_v) + debug_printf(" TFO mode connection attempt to %s, %lu data\n", + address, (unsigned long)fastopen_blob->len); + /*XXX also seen on successful TFO, sigh */ + tcp_out_fastopen = fastopen_blob->len > 0 ? TFO_ATTEMPTED_DATA : TFO_ATTEMPTED_NODATA; + } + else switch (errno) + { + case EINPROGRESS: /* expected if we had no cookie for peer */ + /* seen for no-data, proper TFO option, both cookie-request and with-cookie cases */ + /* apparently no visibility of the diffference at this point */ + /* seen for with-data, proper TFO opt, cookie-req */ + /* with netwk delay, post-conn tcp_info sees unacked 1 for R, 2 for C; code in smtp_out.c */ + /* ? older Experimental TFO option behaviour ? */ + DEBUG(D_transport|D_v) debug_printf(" TFO mode sendto, %s data: EINPROGRESS\n", + fastopen_blob->len > 0 ? "with" : "no"); + if (!fastopen_blob->data) + { + tcp_out_fastopen = TFO_ATTEMPTED_NODATA; /* we tried; unknown if useful yet */ + rc = 0; + } + else /* queue unsent data */ + rc = send(sock, fastopen_blob->data, fastopen_blob->len, 0); + break; + + case EOPNOTSUPP: + DEBUG(D_transport) + debug_printf("Tried TCP Fast Open but apparently not enabled by sysctl\n"); + goto legacy_connect; + + case EPIPE: + DEBUG(D_transport) + debug_printf("Tried TCP Fast Open but kernel too old to support it\n"); + goto legacy_connect; + } + +# elif defined(EXIM_TFO_FREEBSD) + /* Re: https://people.freebsd.org/~pkelsey/tfo-tools/tfo-client.c */ + + if (setsockopt(sock, IPPROTO_TCP, TCP_FASTOPEN, &on, sizeof(on)) < 0) + { + DEBUG(D_transport) + debug_printf("Tried TCP Fast Open but apparently not enabled by sysctl\n"); + goto legacy_connect; + } + if ((rc = sendto(sock, fastopen_blob->data, fastopen_blob->len, 0, + s_ptr, s_len)) >= 0) + { + DEBUG(D_transport|D_v) + debug_printf(" TFO mode connection attempt to %s, %lu data\n", + address, (unsigned long)fastopen_blob->len); + tcp_out_fastopen = fastopen_blob->len > 0 ? TFO_ATTEMPTED_DATA : TFO_ATTEMPTED_NODATA; + } + +# elif defined(EXIM_TFO_CONNECTX) + /* MacOS */ + sa_endpoints_t ends = { + .sae_srcif = 0, .sae_srcaddr = NULL, .sae_srcaddrlen = 0, + .sae_dstaddr = s_ptr, .sae_dstaddrlen = s_len }; + struct iovec iov = { + .iov_base = fastopen_blob->data, .iov_len = fastopen_blob->len }; + size_t len; + + if ((rc = connectx(sock, &ends, SAE_ASSOCID_ANY, + CONNECT_DATA_IDEMPOTENT, &iov, 1, &len, NULL)) == 0) + { + DEBUG(D_transport|D_v) + debug_printf(" TFO mode connection attempt to %s, %lu data\n", + address, (unsigned long)fastopen_blob->len); + tcp_out_fastopen = fastopen_blob->len > 0 ? TFO_ATTEMPTED_DATA : TFO_ATTEMPTED_NODATA; + + if (len != fastopen_blob->len) + DEBUG(D_transport|D_v) + debug_printf(" only queued %lu data!\n", (unsigned long)len); + } + else if (errno == EINPROGRESS) + { + DEBUG(D_transport|D_v) debug_printf(" TFO mode connectx, %s data: EINPROGRESS\n", + fastopen_blob->len > 0 ? "with" : "no"); + if (!fastopen_blob->data) + { + tcp_out_fastopen = TFO_ATTEMPTED_NODATA; /* we tried; unknown if useful yet */ + rc = 0; + } + else /* assume that no data was queued; block in send */ + rc = send(sock, fastopen_blob->data, fastopen_blob->len, 0); + } +# endif + } +else +#endif /*EXIM_SUPPORT_TFO*/ + { +#if defined(EXIM_SUPPORT_TFO) && !defined(EXIM_TFO_CONNECTX) +legacy_connect: +#endif + + DEBUG(D_transport|D_v) if (fastopen_blob) + debug_printf(" non-TFO mode connection attempt to %s, %lu data\n", + address, (unsigned long)fastopen_blob->len); + if ((rc = connect(sock, s_ptr, s_len)) >= 0) + if ( fastopen_blob && fastopen_blob->data && fastopen_blob->len + && send(sock, fastopen_blob->data, fastopen_blob->len, 0) < 0) + rc = -1; + } + save_errno = errno; -alarm(0); +ALARM_CLR(0); /* There is a testing facility for simulating a connection timeout, as I can't think of any other way of doing this. It converts a connection refused into a timeout if the timeout is set to 999999. */ -if (running_in_test_harness) +if (f.running_in_test_harness && save_errno == ECONNREFUSED && timeout == 999999) { - if (save_errno == ECONNREFUSED && timeout == 999999) - { - rc = -1; - save_errno = EINTR; - sigalrm_seen = TRUE; - } + rc = -1; + save_errno = EINTR; + sigalrm_seen = TRUE; } /* Success */ -if (rc >= 0) return 0; +if (rc >= 0) + return 0; /* A failure whose error code is "Interrupted system call" is in fact an externally applied timeout if the signal handler has been run. */ -errno = (save_errno == EINTR && sigalrm_seen)? ETIMEDOUT : save_errno; +errno = save_errno == EINTR && sigalrm_seen ? ETIMEDOUT : save_errno; return -1; } @@ -259,22 +389,23 @@ return -1; Arguments: type SOCK_DGRAM or SOCK_STREAM af AF_INET6 or AF_INET for the socket type - address the remote address, in text form + hostname host name, or ip address (as text) portlo,porthi the remote port range timeout a timeout - connhost if not NULL, host_item filled in with connection details + connhost if not NULL, host_item to be filled in with connection details errstr pointer for allocated string on error + fastopen_blob with SOCK_STREAM, if non-null, request TCP Fast Open. + Additionally, optional idempotent early-data to send Return: socket fd, or -1 on failure (having allocated an error string) */ int ip_connectedsocket(int type, const uschar * hostname, int portlo, int porthi, - int timeout, host_item * connhost, uschar ** errstr) + int timeout, host_item * connhost, uschar ** errstr, const blob * fastopen_blob) { -int namelen, port; +int namelen; host_item shost; -host_item *h; int af = 0, fd, fd4 = -1, fd6 = -1; shost.next = NULL; @@ -289,9 +420,7 @@ namelen = Ustrlen(hostname); if (hostname[0] == '[' && hostname[namelen - 1] == ']') { - uschar * host = string_copy(hostname); - host[namelen - 1] = 0; - host++; + uschar * host = string_copyn(hostname+1, namelen-2); if (string_is_ip_address(host, NULL) == 0) { *errstr = string_sprintf("malformed IP address \"%s\"", hostname); @@ -303,15 +432,15 @@ if (hostname[0] == '[' && /* Otherwise check for an unadorned IP address */ else if (string_is_ip_address(hostname, NULL) != 0) - shost.name = shost.address = string_copy(hostname); + shost.name = shost.address = string_copyn(hostname, namelen); /* Otherwise lookup IP address(es) from the name */ else { - shost.name = string_copy(hostname); - if (host_find_byname(&shost, NULL, HOST_FIND_QUALIFY_SINGLE, NULL, - FALSE) != HOST_FOUND) + shost.name = string_copyn(hostname, namelen); + if (host_find_byname(&shost, NULL, HOST_FIND_QUALIFY_SINGLE, + NULL, FALSE) != HOST_FOUND) { *errstr = string_sprintf("no IP address found for host %s", shost.name); return -1; @@ -320,11 +449,11 @@ else /* Try to connect to the server - test each IP till one works */ -for (h = &shost; h != NULL; h = h->next) +for (host_item * h = &shost; h; h = h->next) { - fd = (Ustrchr(h->address, ':') != 0) - ? (fd6 < 0) ? (fd6 = ip_socket(type, af = AF_INET6)) : fd6 - : (fd4 < 0) ? (fd4 = ip_socket(type, af = AF_INET )) : fd4; + fd = Ustrchr(h->address, ':') != 0 + ? fd6 < 0 ? (fd6 = ip_socket(type, af = AF_INET6)) : fd6 + : fd4 < 0 ? (fd4 = ip_socket(type, af = AF_INET )) : fd4; if (fd < 0) { @@ -332,12 +461,13 @@ for (h = &shost; h != NULL; h = h->next) goto bad; } - for(port = portlo; port <= porthi; port++) - if (ip_connect(fd, af, h->address, port, timeout) == 0) + for (int port = portlo; port <= porthi; port++) + if (ip_connect(fd, af, h->address, port, timeout, fastopen_blob) == 0) { - if (fd != fd6) close(fd6); - if (fd != fd4) close(fd4); - if (connhost) { + if (fd6 >= 0 && fd != fd6) close(fd6); + if (fd4 >= 0 && fd != fd4) close(fd4); + if (connhost) + { h->port = port; *connhost = *h; connhost->next = NULL; @@ -354,56 +484,70 @@ bad: } +/*XXX TFO? */ int -ip_tcpsocket(const uschar * hostport, uschar ** errstr, int tmo) +ip_tcpsocket(const uschar * hostport, uschar ** errstr, int tmo, + host_item * connhost) { - int scan; - uschar hostname[256]; - unsigned int portlow, porthigh; - - /* extract host and port part */ - scan = sscanf(CS hostport, "%255s %u-%u", hostname, &portlow, &porthigh); - if ( scan != 3 ) { - if ( scan != 2 ) { - *errstr = string_sprintf("invalid socket '%s'", hostport); - return -1; +int scan; +uschar hostname[256]; +unsigned int portlow, porthigh; + +/* extract host and port part */ +scan = sscanf(CS hostport, "%255s %u-%u", hostname, &portlow, &porthigh); +if (scan != 3) + { + if (scan != 2) + { + *errstr = string_sprintf("invalid socket '%s'", hostport); + return -1; } - porthigh = portlow; + porthigh = portlow; } - return ip_connectedsocket(SOCK_STREAM, hostname, portlow, porthigh, - tmo, NULL, errstr); +return ip_connectedsocket(SOCK_STREAM, hostname, portlow, porthigh, + tmo, connhost, errstr, NULL); } int ip_unixsocket(const uschar * path, uschar ** errstr) { - int sock; - struct sockaddr_un server; +int sock; +struct sockaddr_un server; - if ((sock = socket(AF_UNIX, SOCK_STREAM, 0)) < 0) { - *errstr = US"can't open UNIX socket."; - return -1; +if ((sock = socket(AF_UNIX, SOCK_STREAM, 0)) < 0) + { + *errstr = US"can't open UNIX socket."; + return -1; } - server.sun_family = AF_UNIX; - Ustrncpy(server.sun_path, path, sizeof(server.sun_path)-1); - server.sun_path[sizeof(server.sun_path)-1] = '\0'; - if (connect(sock, (struct sockaddr *) &server, sizeof(server)) < 0) { - int err = errno; - (void)close(sock); - *errstr = string_sprintf("unable to connect to UNIX socket (%s): %s", - path, strerror(err)); - return -1; - } - return sock; +callout_address = string_copy(path); +server.sun_family = AF_UNIX; +Ustrncpy(US server.sun_path, path, sizeof(server.sun_path)-1); +server.sun_path[sizeof(server.sun_path)-1] = '\0'; +if (connect(sock, (struct sockaddr *) &server, sizeof(server)) < 0) + { + int err = errno; + (void)close(sock); + *errstr = string_sprintf("unable to connect to UNIX socket (%s): %s", + path, strerror(err)); + return -1; + } +return sock; } +/* spec is either an absolute path (with a leading /), or +a host (name or IP) and port (whitespace-separated). +The port can be a range, dash-separated, or a single number. + +For a TCP socket, optionally fill in a host_item. +*/ int -ip_streamsocket(const uschar * spec, uschar ** errstr, int tmo) +ip_streamsocket(const uschar * spec, uschar ** errstr, int tmo, + host_item * connhost) { - return *spec == '/' - ? ip_unixsocket(spec, errstr) : ip_tcpsocket(spec, errstr, tmo); +return *spec == '/' + ? ip_unixsocket(spec, errstr) : ip_tcpsocket(spec, errstr, tmo, connhost); } /************************************************* @@ -421,11 +565,11 @@ Returns: nothing */ void -ip_keepalive(int sock, uschar *address, BOOL torf) +ip_keepalive(int sock, const uschar *address, BOOL torf) { int fodder = 1; if (setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE, - (uschar *)(&fodder), sizeof(fodder)) != 0) + US (&fodder), sizeof(fodder)) != 0) log_write(0, LOG_MAIN, "setsockopt(SO_KEEPALIVE) on connection %s %s " "failed: %s", torf? "to":"from", address, strerror(errno)); } @@ -439,34 +583,26 @@ if (setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE, /* Arguments: fd the file descriptor - timeout the timeout, seconds + timelimit the timeout endpoint, seconds-since-epoch Returns: TRUE => ready for i/o FALSE => timed out, or other error */ BOOL -fd_ready(int fd, int timeout) +fd_ready(int fd, time_t timelimit) { -fd_set select_inset; -struct timeval tv; -time_t start_recv = time(NULL); -int rc; +int rc, time_left = timelimit - time(NULL); -if (timeout <= 0) +if (time_left <= 0) { errno = ETIMEDOUT; return FALSE; } /* Wait until the socket is ready */ -for (;;) +do { - FD_ZERO (&select_inset); - FD_SET (fd, &select_inset); - tv.tv_sec = timeout; - tv.tv_usec = 0; - /*DEBUG(D_transport) debug_printf("waiting for data on fd\n");*/ - rc = select(fd + 1, (SELECT_ARG2_TYPE *)&select_inset, NULL, NULL, &tv); + rc = poll_one_fd(fd, POLLIN, time_left * 1000); /* If some interrupt arrived, just retry. We presume this to be rare, but it can happen (e.g. the SIGUSR1 signal sent by exiwhat causes @@ -475,17 +611,16 @@ for (;;) Aug 2004: Somebody set up a cron job that ran exiwhat every 2 minutes, making the interrupt not at all rare. Since the timeout is typically more than 2 minutes, the effect was to block the timeout completely. To prevent this - happening again, we do an explicit time test. */ + happening again, we do an explicit time test and adjust the timeout + accordingly */ if (rc < 0 && errno == EINTR) { DEBUG(D_transport) debug_printf("EINTR while waiting for socket data\n"); - if (time(NULL) - start_recv < timeout) continue; - DEBUG(D_transport) debug_printf("total wait time exceeds timeout\n"); - } - /* Handle a timeout, and treat any other select error as a timeout, including - an EINTR when we have been in this loop for longer than timeout. */ + /* Watch out, 'continue' jumps to the condition, not to the loops top */ + if ((time_left = timelimit - time(NULL)) > 0) continue; + } if (rc <= 0) { @@ -493,10 +628,10 @@ for (;;) return FALSE; } - /* If the socket is ready, break out of the loop. */ - - if (FD_ISSET(fd, &select_inset)) break; + /* Checking the FD_ISSET is not enough, if we're interrupted, the + select_inset may still contain the 'input'. */ } +while (rc < 0); return TRUE; } @@ -505,34 +640,34 @@ getting interrupted, and the possibility of select() returning with a positive result but no ready descriptor. Is this in fact possible? Arguments: - sock the socket + cctx the connection context (socket fd, possibly TLS context) buffer to read into bufsize the buffer size - timeout the timeout + timelimit the timeout endpoint, seconds-since-epoch Returns: > 0 => that much data read <= 0 on error or EOF; errno set - zero for EOF */ int -ip_recv(int sock, uschar *buffer, int buffsize, int timeout) +ip_recv(client_conn_ctx * cctx, uschar * buffer, int buffsize, time_t timelimit) { int rc; -if (!fd_ready(sock, timeout)) +if (!fd_ready(cctx->sock, timelimit)) return -1; /* The socket is ready, read from it (via TLS if it's active). On EOF (i.e. close down of the connection), set errno to zero; otherwise leave it alone. */ -#ifdef SUPPORT_TLS -if (tls_out.active == sock) - rc = tls_read(FALSE, buffer, buffsize); -else if (tls_in.active == sock) - rc = tls_read(TRUE, buffer, buffsize); +#ifndef DISABLE_TLS +if (cctx->tls_ctx) /* client TLS */ + rc = tls_read(cctx->tls_ctx, buffer, buffsize); +else if (tls_in.active.sock == cctx->sock) /* server TLS */ + rc = tls_read(NULL, buffer, buffsize); else #endif - rc = recv(sock, buffer, buffsize, 0); + rc = recv(cctx->sock, buffer, buffsize, 0); if (rc > 0) return rc; if (rc == 0) errno = 0; @@ -699,13 +834,9 @@ while (last > first) return TRUE; } else if (c > 0) - { first = middle + 1; - } else - { last = middle; - } } return FALSE; } @@ -713,10 +844,11 @@ return FALSE; void dscp_list_to_stream(FILE *stream) { -int i; -for (i=0; i < dscp_table_size; ++i) +for (int i = 0; i < dscp_table_size; ++i) fprintf(stream, "%s\n", dscp_table[i].name); } /* End of ip.c */ +/* vi: aw ai sw=2 +*/