diff options
Diffstat (limited to 'erts/emulator/drivers/common/inet_drv.c')
-rw-r--r-- | erts/emulator/drivers/common/inet_drv.c | 1302 |
1 files changed, 1118 insertions, 184 deletions
diff --git a/erts/emulator/drivers/common/inet_drv.c b/erts/emulator/drivers/common/inet_drv.c index f6de343f72..3195ca3874 100644 --- a/erts/emulator/drivers/common/inet_drv.c +++ b/erts/emulator/drivers/common/inet_drv.c @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 1997-2017. All Rights Reserved. + * Copyright Ericsson AB 1997-2018. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -63,6 +63,20 @@ #include <sys/un.h> #endif +#ifdef HAVE_SENDFILE +#if defined(__linux__) || (defined(__sun) && defined(__SVR4)) + #include <sys/sendfile.h> +#elif defined(__FreeBSD__) || defined(__DragonFly__) + /* Need to define __BSD_VISIBLE in order to expose prototype of sendfile */ + #define __BSD_VISIBLE 1 + #include <sys/socket.h> +#endif +#endif + +#if defined(__APPLE__) && defined(__MACH__) && !defined(__DARWIN__) + #define __DARWIN__ 1 +#endif + /* All platforms fail on malloc errors. */ #define FATAL_MALLOC @@ -610,6 +624,26 @@ static size_t my_strnlen(const char *s, size_t maxlen) # define VALGRIND_MAKE_MEM_DEFINED(ptr,size) #endif +#ifndef __WIN32__ +/* Calculate CMSG_NXTHDR without having a struct msghdr*. + * CMSG_LEN only caters for alignment for start of data. + * To get how much to advance we need to use CMSG_SPACE + * on the payload length. To get the payload length we + * take the calculated cmsg->cmsg_len and subtract the + * header length. To get the header length we use + * the pointer difference from the cmsg start pointer + * to the CMSG_DATA(cmsg) pointer. + */ +#define LEN_CMSG_DATA(cmsg) \ + ((cmsg)->cmsg_len - ((char*)CMSG_DATA(cmsg) - (char*)(cmsg))) +#define NXT_CMSG_HDR(cmsg) \ + ((struct cmsghdr*)(((char*)(cmsg)) + CMSG_SPACE(LEN_CMSG_DATA(cmsg)))) +#endif + +#if !defined(IPV6_PKTOPTIONS) && defined(IPV6_2292PKTOPTIONS) +#define IPV6_PKTOPTIONS IPV6_2292PKTOPTIONS +#endif + /* Magic errno value used locally for return of {error, system_limit} - the emulator definition of SYSTEM_LIMIT is not available here. @@ -701,6 +735,7 @@ static size_t my_strnlen(const char *s, size_t maxlen) #define TCP_REQ_RECV 42 #define TCP_REQ_UNRECV 43 #define TCP_REQ_SHUTDOWN 44 +#define TCP_REQ_SENDFILE 45 /* UDP and SCTP requests */ #define PACKET_REQ_RECV 60 /* Common for UDP and SCTP */ /* #define SCTP_REQ_LISTEN 61 MERGED Different from TCP; not for UDP */ @@ -723,6 +758,7 @@ static size_t my_strnlen(const char *s, size_t maxlen) #define TCP_ADDF_DELAYED_ECONNRESET 128 /* An ECONNRESET error occurred on send or shutdown */ #define TCP_ADDF_SHUTDOWN_WR_DONE 256 /* A shutdown(sock, SHUT_WR) or SHUT_RDWR was made */ #define TCP_ADDF_LINGER_ZERO 512 /* Discard driver queue on port close */ +#define TCP_ADDF_SENDFILE 1024 /* Send from an fd instead of the driver queue */ /* *_REQ_* replies */ #define INET_REP_ERROR 0 @@ -771,6 +807,12 @@ static size_t my_strnlen(const char *s, size_t maxlen) #define INET_LOPT_LINE_DELIM 40 /* Line delimiting char */ #define INET_OPT_TCLASS 41 /* IPv6 transport class */ #define INET_OPT_BIND_TO_DEVICE 42 /* get/set network device the socket is bound to */ +#define INET_OPT_RECVTOS 43 /* IP_RECVTOS ancillary data */ +#define INET_OPT_RECVTCLASS 44 /* IPV6_RECVTCLASS ancillary data */ +#define INET_OPT_PKTOPTIONS 45 /* IP(V6)_PKTOPTIONS get ancillary data */ +#define INET_OPT_TTL 46 /* IP_TTL */ +#define INET_OPT_RECVTTL 47 /* IP_RECVTTL ancillary data */ +#define TCP_OPT_NOPUSH 48 /* super-Nagle, aka TCP_CORK */ /* SCTP options: a separate range, from 100: */ #define SCTP_OPT_RTOINFO 100 #define SCTP_OPT_ASSOCINFO 101 @@ -846,6 +888,11 @@ static size_t my_strnlen(const char *s, size_t maxlen) #define SCTP_FLAG_SACDELAY_ENABLE (32 /* am_sackdelay_enable */) #define SCTP_FLAG_SACDELAY_DISABLE (64 /* am_sackdelay_disable */) +/* Flags for recv_cmsgflags */ +#define INET_CMSG_RECVTOS (1 << 0) /* am_recvtos, am_tos */ +#define INET_CMSG_RECVTCLASS (1 << 1) /* am_recvtclass, am_tclass */ +#define INET_CMSG_RECVTTL (1 << 2) /* am_recvttl, am_ttl */ + /* ** End of interface constants. **--------------------------------------------------------------------------*/ @@ -900,9 +947,20 @@ static size_t my_strnlen(const char *s, size_t maxlen) #ifdef HAVE_SCTP #define PACKET_ERL_DRV_TERM_DATA_LEN 512 #else +#ifndef __WIN32__ +/* Assume we have recvmsg() and might need room for ancillary data */ +#define PACKET_ERL_DRV_TERM_DATA_LEN 64 +#else #define PACKET_ERL_DRV_TERM_DATA_LEN 32 #endif +#endif + +#if defined(TCP_CORK) +#define INET_TCP_NOPUSH TCP_CORK +#elif defined(TCP_NOPUSH) && !defined(__DARWIN__) +#define INET_TCP_NOPUSH TCP_NOPUSH +#endif #define BIN_REALLOC_MARGIN(x) ((x)/4) /* 25% */ @@ -1068,6 +1126,7 @@ typedef struct { char *netns; /* Socket network namespace name as full file path */ #endif + int recv_cmsgflags; /* Which ancillary data to expect */ } inet_descriptor; @@ -1137,7 +1196,6 @@ static int packet_inet_init(void); static void packet_inet_stop(ErlDrvData); static void packet_inet_command(ErlDrvData, char*, ErlDrvSizeT); static void packet_inet_drv_input(ErlDrvData data, ErlDrvEvent event); -static void packet_inet_drv_output(ErlDrvData data, ErlDrvEvent event); static ErlDrvData udp_inet_start(ErlDrvPort, char* command); #ifdef HAVE_SCTP static ErlDrvData sctp_inet_start(ErlDrvPort, char* command); @@ -1162,7 +1220,7 @@ static struct erl_drv_entry udp_inet_driver_entry = NULL, #else packet_inet_drv_input, - packet_inet_drv_output, + NULL, #endif "udp_inet", NULL, @@ -1197,7 +1255,7 @@ static struct erl_drv_entry sctp_inet_driver_entry = NULL, #else packet_inet_drv_input, - packet_inet_drv_output, + NULL, #endif "sctp_inet", NULL, @@ -1237,6 +1295,21 @@ typedef struct { inet_async_multi_op *multi_first;/* NULL == no multi-accept-queue, op is in ordinary queue */ inet_async_multi_op *multi_last; MultiTimerData *mtd; /* Timer structures for multiple accept */ +#ifdef HAVE_SENDFILE + struct { + ErlDrvSizeT ioq_skip; /* The number of bytes in the queue at the time + * sendfile was issued, which must be sent + * before issuing the sendfile call itself. */ + int dup_file_fd; /* The file handle to send from; this is + * duplicated when sendfile is issued to + * reduce (but not eliminate) the impact of a + * nasty race, so we have to remember to close + * it. */ + Uint64 bytes_sent; + Uint64 offset; + Uint64 length; + } sendfile; +#endif } tcp_descriptor; /* send function */ @@ -1247,6 +1320,8 @@ static int tcp_deliver(tcp_descriptor* desc, int len); static int tcp_shutdown_error(tcp_descriptor* desc, int err); +static int tcp_inet_sendfile(tcp_descriptor* desc); + static int tcp_inet_output(tcp_descriptor* desc, HANDLE event); static int tcp_inet_input(tcp_descriptor* desc, HANDLE event); @@ -1263,7 +1338,6 @@ typedef struct { static int packet_inet_input(udp_descriptor* udesc, HANDLE event); -static int packet_inet_output(udp_descriptor* udesc, HANDLE event); #endif /* convert descriptor pointer to inet_descriptor pointer */ @@ -1305,6 +1379,11 @@ static ErlDrvTermData am_udp_error; #ifdef HAVE_SYS_UN_H static ErlDrvTermData am_local; #endif +#ifndef __WIN32__ +static ErlDrvTermData am_tos; +static ErlDrvTermData am_tclass; +static ErlDrvTermData am_ttl; +#endif #ifdef HAVE_SCTP static ErlDrvTermData am_sctp; static ErlDrvTermData am_sctp_passive; @@ -1325,12 +1404,16 @@ static ErlDrvTermData am_sndbuf; static ErlDrvTermData am_reuseaddr; static ErlDrvTermData am_dontroute; static ErlDrvTermData am_priority; -static ErlDrvTermData am_tos; -static ErlDrvTermData am_tclass; +static ErlDrvTermData am_recvtos; +static ErlDrvTermData am_recvtclass; +static ErlDrvTermData am_recvttl; static ErlDrvTermData am_ipv6_v6only; static ErlDrvTermData am_netns; static ErlDrvTermData am_bind_to_device; #endif +#ifdef HAVE_SENDFILE +static ErlDrvTermData am_sendfile; +#endif static char str_eafnosupport[] = "eafnosupport"; static char str_einval[] = "einval"; @@ -1514,8 +1597,10 @@ static void *realloc_wrapper(void *current, ErlDrvSizeT size){ # define ASSOC_ID_LEN 4 # define LOAD_ASSOC_ID LOAD_UINT # define LOAD_ASSOC_ID_CNT LOAD_UINT_CNT -# define SCTP_ANC_BUFF_SIZE INET_DEF_BUFFER/2 /* XXX: not very good... */ +#else +# define IS_SCTP(desc) 0 #endif +# define ANC_BUFF_SIZE INET_DEF_BUFFER/2 /* XXX: not very good... */ #ifdef HAVE_UDP static int load_address(ErlDrvTermData* spec, int i, char* buf) @@ -1751,6 +1836,7 @@ static void release_buffer(ErlDrvBinary* buf) #ifdef HAVE_UDP static ErlDrvBinary* realloc_buffer(ErlDrvBinary* buf, ErlDrvSizeT newsz) { + DEBUGF(("realloc_buffer: %ld -> %ld\r\n", (buf==NULL) ? 0 : buf->orig_size, newsz)); return driver_realloc_binary(buf, newsz); } #endif @@ -2715,6 +2801,66 @@ static int inet_async_data(inet_descriptor* desc, const char* buf, int len) } } +#ifndef __WIN32__ +static int load_cmsg_int(ErlDrvTermData *spec, int i, + struct cmsghdr *cmsg) { + union u { + byte uint8; + Uint16 uint16; + Uint32 uint32; + Uint64 uint64; + } *p; + p = (union u*) CMSG_DATA(cmsg); + switch (LEN_CMSG_DATA(cmsg) * CHAR_BIT) { + case 8: + return LOAD_INT(spec, i, p->uint8); + case 16: + return LOAD_INT(spec, i, p->uint16); + + case 32: + return LOAD_INT(spec, i, p->uint32); + + case 64: + return LOAD_INT(spec, i, p->uint64); + } + return LOAD_INT(spec, i, 0); +} + +static int parse_ancillary_data_item(ErlDrvTermData *spec, int i, + struct cmsghdr *cmsg, int *n) { +#define LOAD_CMSG_INT(proto, type, am) \ + if (cmsg->cmsg_level == (proto) && \ + cmsg->cmsg_type == (type)) { \ + i = LOAD_ATOM(spec, i, (am)); \ + i = load_cmsg_int(spec, i, cmsg); \ + i = LOAD_TUPLE(spec, i, 2); \ + (*n)++; \ + return i; \ + } +#if defined(IPPROTO_IP) && defined(IP_TOS) + LOAD_CMSG_INT(IPPROTO_IP, IP_TOS, am_tos); +#endif +#if defined(IPPROTO_IPV6) && defined(IPV6_TCLASS) + LOAD_CMSG_INT(IPPROTO_IPV6, IPV6_TCLASS, am_tclass); +#endif +#if defined(IPPROTO_IP) && defined(IP_TTL) + LOAD_CMSG_INT(IPPROTO_IP, IP_TTL, am_ttl); +#endif + /* BSD uses the RECV* names in CMSG fields */ +#if defined(IPPROTO_IP) && defined(IP_RECVTOS) + LOAD_CMSG_INT(IPPROTO_IP, IP_RECVTOS, am_tos); +#endif +#if defined(IPPROTO_IPV6) && defined(IPV6_RECVTCLASS) + LOAD_CMSG_INT(IPPROTO_IPV6, IPV6_RECVTCLASS, am_tclass); +#endif +#if defined(IPPROTO_IP) && defined(IP_RECVTTL) + LOAD_CMSG_INT(IPPROTO_IP, IP_RECVTTL, am_ttl); +#endif +#undef LOAD_CMSG_INT + return i; +} +#endif /* #ifndef __WIN32__ */ + #ifdef HAVE_SCTP /* ** SCTP-related atoms: @@ -2846,11 +2992,18 @@ static int sctp_parse_ancillary_data for (cmsg = frst_msg; cmsg != NULL; cmsg = CMSG_NXTHDR(mptr,cmsg)) { struct sctp_sndrcvinfo * sri; - +#ifndef __WIN32 + int old_s; + + /* Parse ancillary data common to UDP */ + old_s = s; + i = parse_ancillary_data_item(spec, i, cmsg, &s); + if (s > old_s) continue; /* Skip other possible ancillary data, e.g. from IPv6: */ if (cmsg->cmsg_level != IPPROTO_SCTP || cmsg->cmsg_type != SCTP_SNDRCV) continue; +#endif if (((char*)cmsg + cmsg->cmsg_len) - (char*)frst_msg > mptr->msg_controllen) @@ -3190,6 +3343,23 @@ static int sctp_parse_async_event } #endif /* HAVE_SCTP */ +#ifndef __WIN32__ +static int udp_parse_ancillary_data(ErlDrvTermData *spec, int i, + struct msghdr *mptr) { + struct cmsghdr *cmsg; + int n; + + n = 0; + for (cmsg = CMSG_FIRSTHDR(mptr); + cmsg != NULL; + cmsg = CMSG_NXTHDR(mptr, cmsg)) { + i = parse_ancillary_data_item(spec, i, cmsg, &n); + } + i = LOAD_NIL(spec, i); + return LOAD_LIST(spec, i, n+1); +} +#endif /* ifndef __WIN32__ */ + /* ** passive mode reply: ** for UDP: @@ -3212,7 +3382,7 @@ static int sctp_parse_async_event static int inet_async_binary_data (inet_descriptor* desc, unsigned int phsz, - ErlDrvBinary * bin, int offs, int len, void * extra) + ErlDrvBinary * bin, int offs, int len, void *mp) { unsigned int hsz = desc->hsz + phsz; ErlDrvTermData spec [PACKET_ERL_DRV_TERM_DATA_LEN]; @@ -3245,9 +3415,10 @@ inet_async_binary_data if (IS_SCTP(desc)) { /* For SCTP we always have desc->hsz==0 (i.e., no application-level headers are used), so hsz==phsz (see above): */ - struct msghdr* mptr; int sz; - + struct msghdr *mptr; + + mptr = mp; ASSERT (hsz == phsz && hsz != 0); sz = len - hsz; /* Size of the msg data proper, w/o the addr */ @@ -3255,7 +3426,6 @@ inet_async_binary_data i = LOAD_STRING(spec, i, bin->orig_bytes+offs, hsz); /* Put in the list (possibly empty) of Ancillary Data: */ - mptr = (struct msghdr *) extra; i = sctp_parse_ancillary_data (spec, i, mptr); /* Then: Data or Event (Notification)? */ @@ -3284,20 +3454,32 @@ inet_async_binary_data } else #endif /* HAVE_SCTP */ - /* Generic case. Both Addr and Data (or a single list of them together) are - returned: */ + { + /* Generic case. Both Addr and Data + * (or a single list of them together) are returned: */ - if ((desc->mode == INET_MODE_LIST) || (hsz > len)) { - /* INET_MODE_LIST => [H1,H2,...Hn] */ - i = LOAD_STRING(spec, i, bin->orig_bytes+offs, len); - } - else { - /* INET_MODE_BINARY => [H1,H2,...HSz | Binary] or [Binary]: */ - int sz = len - hsz; - i = LOAD_BINARY(spec, i, bin, offs+hsz, sz); - if (hsz > 0) - i = LOAD_STRING_CONS(spec, i, bin->orig_bytes+offs, hsz); + if ((desc->mode == INET_MODE_LIST) || (hsz > len)) { + /* INET_MODE_LIST => [H1,H2,...Hn] */ + i = LOAD_STRING(spec, i, bin->orig_bytes+offs, len); + } + else { + /* INET_MODE_BINARY => [H1,H2,...HSz | Binary] or [Binary]: */ + int sz = len - hsz; + i = LOAD_BINARY(spec, i, bin, offs+hsz, sz); + if (hsz > 0) + i = LOAD_STRING_CONS(spec, i, bin->orig_bytes+offs, hsz); + } + +#ifndef __WIN32__ + if (mp) { + /* We got ancillary data from an UDP recvmsg. + * Insert an additional tuple level {[F|AddrData],AncData} */ + i = udp_parse_ancillary_data(spec, i, (struct msghdr*)mp); + i = LOAD_TUPLE(spec, i, 2); + } +#endif } + /* Close up the {ok, ...} or {error, ...} tuple: */ i = LOAD_TUPLE(spec, i, 2); @@ -3429,8 +3611,9 @@ static int tcp_error_message(tcp_descriptor* desc, int err) ** [AddrLen, H2,...,HSz] are msg headers for UDP AF_UNIX only ** Data : List() | Binary() */ -static int packet_binary_message - (inet_descriptor* desc, ErlDrvBinary* bin, int offs, int len, void* extra) +static int packet_binary_message(inet_descriptor* desc, + ErlDrvBinary* bin, int offs, int len, + void *mp) { unsigned int hsz = desc->hsz; ErlDrvTermData spec [PACKET_ERL_DRV_TERM_DATA_LEN]; @@ -3455,8 +3638,14 @@ static int packet_binary_message # ifdef HAVE_SCTP if (!IS_SCTP(desc)) - { # endif + { +#ifndef __WIN32__ + if (mp) i = udp_parse_ancillary_data(spec, i, (struct msghdr*)mp); +#endif + /* We got ancillary data from an UDP recvmsg. + * Insert an additional tuple level {AncData,[F|AddrData]} + */ if ((desc->mode == INET_MODE_LIST) || (hsz > len)) /* INET_MODE_LIST, or only headers => [H1,H2,...Hn] */ i = LOAD_STRING(spec, i, bin->orig_bytes+offs, len); @@ -3468,16 +3657,24 @@ static int packet_binary_message if (hsz > 0) i = LOAD_STRING_CONS(spec, i, bin->orig_bytes+offs, hsz); } -# ifdef HAVE_SCTP + /* Close up the outer 5-or-6-tuple */ +#ifndef __WIN32__ + if (mp) i = LOAD_TUPLE(spec, i, 6); + else +#endif + i = LOAD_TUPLE(spec, i, 5); } +# ifdef HAVE_SCTP else - { /* For SCTP we always have desc->hsz==0 (i.e., no application-level + { + struct msghdr *mptr; + + mptr = mp; + /* For SCTP we always have desc->hsz==0 (i.e., no application-level headers are used): */ - struct msghdr* mptr; ASSERT(hsz == 0); /* Put in the list (possibly empty) of Ancillary Data: */ - mptr = (struct msghdr *) extra; i = sctp_parse_ancillary_data (spec, i, mptr); /* Then: Data or Event (Notification)? */ @@ -3503,11 +3700,11 @@ static int packet_binary_message /* Close up the {[AncilData], Event_OR_Data} tuple: */ i = LOAD_TUPLE (spec, i, 2); + /* Close up the outer 5-tuple: */ + i = LOAD_TUPLE(spec, i, 5); } # endif /* HAVE_SCTP */ - /* Close up the outer 5-tuple: */ - i = LOAD_TUPLE(spec, i, 5); ASSERT(i <= PACKET_ERL_DRV_TERM_DATA_LEN); return erl_drv_output_term(desc->dport, spec, i); } @@ -3641,19 +3838,19 @@ tcp_reply_binary_data(tcp_descriptor* desc, ErlDrvBinary* bin, int offs, int len static int packet_reply_binary_data(inet_descriptor* desc, unsigned int hsz, ErlDrvBinary * bin, int offs, int len, - void * extra) + void *mp) { int code; if (desc->active == INET_PASSIVE) /* "inet" is actually for both UDP and SCTP, as well as TCP! */ - return inet_async_binary_data(desc, hsz, bin, offs, len, extra); + return inet_async_binary_data(desc, hsz, bin, offs, len, mp); else { /* INET_ACTIVE or INET_ONCE: */ if (desc->deliver == INET_DELIVER_PORT) code = inet_port_binary_data(desc, bin, offs, len); else - code = packet_binary_message(desc, bin, offs, len, extra); + code = packet_binary_message(desc, bin, offs, len, mp); if (code < 0) return code; INET_CHECK_ACTIVE_TO_PASSIVE(desc); @@ -3720,8 +3917,9 @@ static void inet_init_sctp(void) { INIT_ATOM(reuseaddr); INIT_ATOM(dontroute); INIT_ATOM(priority); - INIT_ATOM(tos); - INIT_ATOM(tclass); + INIT_ATOM(recvtos); + INIT_ATOM(recvtclass); + INIT_ATOM(recvttl); INIT_ATOM(ipv6_v6only); INIT_ATOM(netns); INIT_ATOM(bind_to_device); @@ -3864,6 +4062,11 @@ static int inet_init() #endif INIT_ATOM(empty_out_q); INIT_ATOM(ssl_tls); +#ifndef __WIN32__ + INIT_ATOM(tos); + INIT_ATOM(tclass); + INIT_ATOM(ttl); +#endif INIT_ATOM(http_eoh); INIT_ATOM(http_header); @@ -3877,6 +4080,10 @@ static int inet_init() INIT_ATOM(https); INIT_ATOM(scheme); +#ifdef HAVE_SENDFILE + INIT_ATOM(sendfile); +#endif + /* add TCP, UDP and SCTP drivers */ add_driver_entry(&tcp_inet_driver_entry); #ifdef HAVE_UDP @@ -4978,6 +5185,71 @@ static int hwaddr_libdlpi_lookup(const char *ifnm, } #endif +#ifdef HAVE_GETIFADDRS +/* Returns 0 for success and errno() for failure */ +static int call_getifaddrs(inet_descriptor* desc_p, struct ifaddrs **ifa_pp) +{ + int result, save_errno; +#ifdef HAVE_SETNS + int current_ns; + + current_ns = 0; + if (desc_p->netns != NULL) { + int new_ns; + /* Temporarily change network namespace for this thread + * over the getifaddrs() call + */ + current_ns = open("/proc/self/ns/net", O_RDONLY); + if (current_ns == INVALID_SOCKET) + return sock_errno(); + new_ns = open(desc_p->netns, O_RDONLY); + if (new_ns == INVALID_SOCKET) { + save_errno = sock_errno(); + while (close(current_ns) == INVALID_SOCKET && + sock_errno() == EINTR); + return save_errno; + } + if (setns(new_ns, CLONE_NEWNET) != 0) { + save_errno = sock_errno(); + while (close(new_ns) == INVALID_SOCKET && + sock_errno() == EINTR); + while (close(current_ns) == INVALID_SOCKET && + sock_errno() == EINTR); + return save_errno; + } + else { + while (close(new_ns) == INVALID_SOCKET && + sock_errno() == EINTR); + } + } +#endif + save_errno = 0; + result = getifaddrs(ifa_pp); + if (result < 0) + save_errno = sock_errno(); +#ifdef HAVE_SETNS + if (desc_p->netns != NULL) { + /* Restore network namespace */ + if (setns(current_ns, CLONE_NEWNET) != 0) { + /* XXX Failed to restore network namespace. + * What to do? Tidy up and return an error... + * Note that the thread now might still be in the set namespace. + * Can this even happen? Should the emulator be aborted? + */ + if (result >= 0) { + /* We got a result but have to waste it */ + save_errno = sock_errno(); + freeifaddrs(*ifa_pp); + } + } + while (close(current_ns) == INVALID_SOCKET && + sock_errno() == EINTR); + } +#endif + return save_errno; +} +#endif /* #ifdef HAVE_GETIFADDRS */ + /* FIXME: temporary hack */ #ifndef IFHWADDRLEN #define IFHWADDRLEN 6 @@ -5055,8 +5327,8 @@ static ErlDrvSSizeT inet_ctl_ifget(inet_descriptor* desc, struct sockaddr_dl *sdlp; int found = 0; - if (getifaddrs(&ifa) == -1) - goto error; + if (call_getifaddrs(desc, &ifa) != 0) + goto error; for (ifp = ifa; ifp; ifp = ifp->ifa_next) { if ((ifp->ifa_addr->sa_family == AF_LINK) && @@ -5078,8 +5350,8 @@ static ErlDrvSSizeT inet_ctl_ifget(inet_descriptor* desc, sys_memcpy(sptr, sdlp->sdl_data + sdlp->sdl_nlen, sdlp->sdl_alen); - freeifaddrs(ifa); sptr += sdlp->sdl_alen; + freeifaddrs(ifa); #endif break; } @@ -5774,6 +6046,7 @@ static ErlDrvSSizeT inet_ctl_getifaddrs(inet_descriptor* desc_p, ErlDrvSizeT buf_size; char *buf_p; char *buf_alloc_p; + int save_errno; buf_size = GETIFADDRS_BUFSZ; buf_alloc_p = ALLOC(GETIFADDRS_BUFSZ); @@ -5808,9 +6081,9 @@ static ErlDrvSSizeT inet_ctl_getifaddrs(inet_descriptor* desc_p, } \ } while (0) - if (getifaddrs(&ifa_p) < 0) { - return ctl_error(sock_errno(), rbuf_pp, rsize); - } + if ((save_errno = call_getifaddrs(desc_p, &ifa_p)) != 0) + return ctl_error(save_errno, rbuf_pp, rsize); + ifa_free_p = ifa_p; *buf_p++ = INET_REP_OK; for (; ifa_p; ifa_p = ifa_p->ifa_next) { @@ -5892,7 +6165,8 @@ static ErlDrvSSizeT inet_ctl_getifaddrs(inet_descriptor* desc_p, but ditto for the other worked and that was actually the requested option, failure was still reported to erlang. */ -#if defined(IP_TOS) && defined(SOL_IP) && defined(SO_PRIORITY) +#if defined(IP_TOS) && defined(IPPROTO_IP) \ + && defined(SO_PRIORITY) && !defined(__WIN32__) static int setopt_prio_tos_trick (int fd, int proto, int type, char* arg_ptr, int arg_sz, int propagate) { @@ -5914,14 +6188,14 @@ static int setopt_prio_tos_trick res_prio = sock_getopt(fd, SOL_SOCKET, SO_PRIORITY, (char *) &tmp_ival_prio, &tmp_arg_sz_prio); - res_tos = sock_getopt(fd, SOL_IP, IP_TOS, + res_tos = sock_getopt(fd, IPPROTO_IP, IP_TOS, (char *) &tmp_ival_tos, &tmp_arg_sz_tos); res = sock_setopt(fd, proto, type, arg_ptr, arg_sz); if (res == 0) { if (type != SO_PRIORITY) { if (type != IP_TOS && res_tos == 0) { res_tos = sock_setopt(fd, - SOL_IP, + IPPROTO_IP, IP_TOS, (char *) &tmp_ival_tos, tmp_arg_sz_tos); @@ -5965,7 +6239,7 @@ static int inet_set_opts(inet_descriptor* desc, char* ptr, int len) int proto; int opt; struct linger li_val; -#ifdef HAVE_MULTICAST_SUPPORT +#if defined(HAVE_MULTICAST_SUPPORT) && defined(IPPROTO_IP) struct ip_mreq mreq_val; #endif int ival; @@ -5988,6 +6262,8 @@ static int inet_set_opts(inet_descriptor* desc, char* ptr, int len) /* XXX { int i; for(i=0;i<len;++i) fprintf(stderr,"0x%02X, ", (unsigned) ptr[i]); fprintf(stderr,"\r\n");} */ while(len >= 5) { + int recv_cmsgflags; + opt = *ptr++; ival = get_int32(ptr); ptr += 4; @@ -5996,6 +6272,7 @@ static int inet_set_opts(inet_descriptor* desc, char* ptr, int len) arg_sz = sizeof(ival); proto = SOL_SOCKET; propagate = 0; + recv_cmsgflags = desc->recv_cmsgflags; switch(opt) { case INET_LOPT_HEADER: @@ -6246,28 +6523,80 @@ static int inet_set_opts(inet_descriptor* desc, char* ptr, int len) (long)desc->port, desc->s, ival)); break; #else + /* inet_fill_opts always returns a value for this option, + * so we need to ignore it if not implemented */ continue; #endif case INET_OPT_TOS: -#if defined(IP_TOS) && defined(SOL_IP) - proto = SOL_IP; +#if defined(IP_TOS) && defined(IPPROTO_IP) + proto = IPPROTO_IP; type = IP_TOS; propagate = 1; DEBUGF(("inet_set_opts(%ld): s=%d, IP_TOS=%d\r\n", (long)desc->port, desc->s, ival)); break; #else + /* inet_fill_opts always returns a value for this option, + * so we need to ignore it if not implemented. */ continue; #endif -#if defined(IPV6_TCLASS) && defined(SOL_IPV6) +#if defined(IPV6_TCLASS) && defined(IPPROTO_IPV6) case INET_OPT_TCLASS: - proto = SOL_IPV6; + proto = IPPROTO_IPV6; type = IPV6_TCLASS; propagate = 1; DEBUGF(("inet_set_opts(%ld): s=%d, IPV6_TCLASS=%d\r\n", (long)desc->port, desc->s, ival)); break; #endif +#if defined(IP_TTL) && defined(IPPROTO_IP) + case INET_OPT_TTL: + proto = IPPROTO_IP; + type = IP_TTL; + propagate = 1; + DEBUGF(("inet_set_opts(%ld): s=%d, IP_TTL=%d\r\n", + (long)desc->port, desc->s, ival)); + break; +#endif +#if defined(IP_RECVTOS) && defined(IPPROTO_IP) + case INET_OPT_RECVTOS: + proto = IPPROTO_IP; + type = IP_RECVTOS; + propagate = 1; + recv_cmsgflags = + ival ? + (desc->recv_cmsgflags | INET_CMSG_RECVTOS) : + (desc->recv_cmsgflags & ~INET_CMSG_RECVTOS); + DEBUGF(("inet_set_opts(%ld): s=%d, IP_RECVTOS=%d\r\n", + (long)desc->port, desc->s, ival)); + break; +#endif +#if defined(IPV6_RECVTCLASS) && defined(IPPROTO_IPV6) + case INET_OPT_RECVTCLASS: + proto = IPPROTO_IPV6; + type = IPV6_RECVTCLASS; + propagate = 1; + recv_cmsgflags = + ival ? + (desc->recv_cmsgflags | INET_CMSG_RECVTCLASS) : + (desc->recv_cmsgflags & ~INET_CMSG_RECVTCLASS); + DEBUGF(("inet_set_opts(%ld): s=%d, IPV6_RECVTCLASS=%d\r\n", + (long)desc->port, desc->s, ival)); + break; +#endif +#if defined(IP_RECVTTL) && defined(IPPROTO_IP) + case INET_OPT_RECVTTL: + proto = IPPROTO_IP; + type = IP_RECVTTL; + propagate = 1; + recv_cmsgflags = + ival ? + (desc->recv_cmsgflags | INET_CMSG_RECVTTL) : + (desc->recv_cmsgflags & ~INET_CMSG_RECVTTL); + DEBUGF(("inet_set_opts(%ld): s=%d, IP_RECVTTL=%d\r\n", + (long)desc->port, desc->s, ival)); + break; +#endif case TCP_OPT_NODELAY: proto = IPPROTO_TCP; @@ -6276,7 +6605,20 @@ static int inet_set_opts(inet_descriptor* desc, char* ptr, int len) (long)desc->port, desc->s, ival)); break; -#ifdef HAVE_MULTICAST_SUPPORT + case TCP_OPT_NOPUSH: +#if defined(INET_TCP_NOPUSH) + proto = IPPROTO_TCP; + type = INET_TCP_NOPUSH; + DEBUGF(("inet_set_opts(%ld): s=%d, t=%d TCP_NOPUSH=%d\r\n", + (long)desc->port, desc->s, type, ival)); + break; +#else + /* inet_fill_opts always returns a value for this option, + * so we need to ignore it if not implemented, just in case */ + continue; +#endif + +#if defined(HAVE_MULTICAST_SUPPORT) && defined(IPPROTO_IP) case UDP_OPT_MULTICAST_TTL: proto = IPPROTO_IP; @@ -6322,10 +6664,10 @@ static int inet_set_opts(inet_descriptor* desc, char* ptr, int len) arg_sz = sizeof(mreq_val); break; -#endif /* HAVE_MULTICAST_SUPPORT */ +#endif /* defined(HAVE_MULTICAST_SUPPORT) && defined(IPPROTO_IP) */ case INET_OPT_IPV6_V6ONLY: -#if HAVE_DECL_IPV6_V6ONLY +#if HAVE_DECL_IPV6_V6ONLY && defined(IPPROTO_IPV6) proto = IPPROTO_IPV6; type = IPV6_V6ONLY; propagate = 1; @@ -6385,11 +6727,13 @@ static int inet_set_opts(inet_descriptor* desc, char* ptr, int len) default: return -1; } -#if defined(IP_TOS) && defined(SOL_IP) && defined(SO_PRIORITY) +#if defined(IP_TOS) && defined(IPPROTO_IP) \ + && defined(SO_PRIORITY) && !defined(__WIN32__) res = setopt_prio_tos_trick (desc->s, proto, type, arg_ptr, arg_sz, propagate); #else res = sock_setopt (desc->s, proto, type, arg_ptr, arg_sz); #endif + if (res == 0) desc->recv_cmsgflags = recv_cmsgflags; if (propagate && res != 0) { return -1; } @@ -6397,7 +6741,12 @@ static int inet_set_opts(inet_descriptor* desc, char* ptr, int len) (long)desc->port, desc->s, res)); if (type == SO_RCVBUF) { /* make sure we have desc->bufsz >= SO_RCVBUF */ - if (ival > desc->bufsz) + if (ival > (1 << 16) && desc->stype == SOCK_DGRAM && !IS_SCTP(desc)) + /* For UDP we don't want to automatically + set the buffer size to be larger than + the theoretical max MTU */ + desc->bufsz = 1 << 16; + else if (ival > desc->bufsz) desc->bufsz = ival; } } @@ -6526,10 +6875,14 @@ static int sctp_set_opts(inet_descriptor* desc, char* ptr, int len) while (curr < ptr + len) { + int recv_cmsgflags; /* Get the Erlang-encoded option type -- always 1 byte: */ - int eopt = *curr; + int eopt; + + eopt = *curr; curr++; + recv_cmsgflags = desc->recv_cmsgflags; /* Get the option value. XXX: The condition (curr < ptr + len) does not preclude us from reading from beyond the buffer end, if the Erlang part of the driver specifies its input wrongly! @@ -6710,28 +7063,32 @@ static int sctp_set_opts(inet_descriptor* desc, char* ptr, int len) break; } # else - continue; /* Option not supported -- ignore it */ + /* inet_fill_opts always returns a value for this option, + * so we need to ignore it if not implemented, just in case */ + continue; # endif case INET_OPT_TOS: -# if defined(IP_TOS) && defined(SOL_IP) +# if defined(IP_TOS) && defined(IPPROTO_IP) { arg.ival= get_int32 (curr); curr += 4; - proto = SOL_IP; + proto = IPPROTO_IP; type = IP_TOS; arg_ptr = (char*) (&arg.ival); arg_sz = sizeof ( arg.ival); break; } # else - continue; /* Option not supported -- ignore it */ + /* inet_fill_opts always returns a value for this option, + * so we need to ignore it if not implemented, just in case */ + continue; # endif -# if defined(IPV6_TCLASS) && defined(SOL_IPV6) +# if defined(IPV6_TCLASS) && defined(IPPROTO_IPV6) case INET_OPT_TCLASS: { arg.ival= get_int32 (curr); curr += 4; - proto = SOL_IPV6; + proto = IPPROTO_IPV6; type = IPV6_TCLASS; arg_ptr = (char*) (&arg.ival); arg_sz = sizeof ( arg.ival); @@ -6739,9 +7096,69 @@ static int sctp_set_opts(inet_descriptor* desc, char* ptr, int len) } # endif +# if defined(IP_TTL) && defined(IPPROTO_IP) + case INET_OPT_TTL: + { + arg.ival= get_int32 (curr); curr += 4; + proto = IPPROTO_IP; + type = IP_TTL; + arg_ptr = (char*) (&arg.ival); + arg_sz = sizeof ( arg.ival); + break; + } +# endif + +# if defined(IP_RECVTOS) && defined(IPPROTO_IP) + case INET_OPT_RECVTOS: + { + arg.ival= get_int32 (curr); curr += 4; + proto = IPPROTO_IP; + type = IP_RECVTOS; + arg_ptr = (char*) (&arg.ival); + arg_sz = sizeof ( arg.ival); + recv_cmsgflags = + arg.ival ? + (desc->recv_cmsgflags | INET_CMSG_RECVTOS) : + (desc->recv_cmsgflags & ~INET_CMSG_RECVTOS); + break; + } +# endif + +# if defined(IPV6_RECVTCLASS) && defined(IPPROTO_IPV6) + case INET_OPT_RECVTCLASS: + { + arg.ival= get_int32 (curr); curr += 4; + proto = IPPROTO_IPV6; + type = IPV6_RECVTCLASS; + arg_ptr = (char*) (&arg.ival); + arg_sz = sizeof ( arg.ival); + recv_cmsgflags = + arg.ival ? + (desc->recv_cmsgflags | INET_CMSG_RECVTCLASS) : + (desc->recv_cmsgflags & ~INET_CMSG_RECVTCLASS); + break; + } +# endif + +# if defined(IP_RECVTTL) && defined(IPPROTO_IP) + case INET_OPT_RECVTTL: + { + arg.ival= get_int32 (curr); curr += 4; + proto = IPPROTO_IP; + type = IP_RECVTTL; + arg_ptr = (char*) (&arg.ival); + arg_sz = sizeof ( arg.ival); + recv_cmsgflags = + arg.ival ? + (desc->recv_cmsgflags | INET_CMSG_RECVTTL) : + (desc->recv_cmsgflags & ~INET_CMSG_RECVTTL); + break; + } +# endif + case INET_OPT_IPV6_V6ONLY: -# if HAVE_DECL_IPV6_V6ONLY +# if HAVE_DECL_IPV6_V6ONLY && defined(IPPROTO_IPV6) { arg.ival= get_int32 (curr); curr += 4; proto = IPPROTO_IPV6; @@ -6991,13 +7408,15 @@ static int sctp_set_opts(inet_descriptor* desc, char* ptr, int len) */ return -1; } -#if defined(IP_TOS) && defined(SOL_IP) && defined(SO_PRIORITY) +#if defined(IP_TOS) && defined(IPPROTO_IP) \ + && defined(SO_PRIORITY) && !defined(__WIN32__) res = setopt_prio_tos_trick (desc->s, proto, type, arg_ptr, arg_sz, 1); #else res = sock_setopt (desc->s, proto, type, arg_ptr, arg_sz); #endif /* The return values of "sock_setopt" can only be 0 or -1: */ ASSERT(res == 0 || res == -1); + if (res == 0) desc->recv_cmsgflags = recv_cmsgflags; if (res == -1) { /* Got an error, DO NOT continue with other options. However, on Solaris 10, we DO allow SO_SNDBUF and SO_RCVBUF to fail, assu- @@ -7018,6 +7437,35 @@ static int sctp_set_opts(inet_descriptor* desc, char* ptr, int len) } #endif /* HAVE_SCTP */ +#ifndef __WIN32__ +static void put_cmsg_int32(struct cmsghdr *cmsg, char *ptr) { + union u { + byte uint8; + Uint16 uint16; + Uint32 uint32; + Uint64 uint64; + } *p; + p = (union u*) CMSG_DATA(cmsg); + switch (LEN_CMSG_DATA(cmsg) * CHAR_BIT) { + case 8: + put_int32((Uint32) p->uint8, ptr); + break; + case 16: + put_int32((Uint32) p->uint16, ptr); + break; + case 32: + put_int32(p->uint32, ptr); + break; + case 64: + put_int32((Uint32) p->uint64, ptr); + break; + default: + put_int32(0, ptr); + } + return; +} +#endif + /* load all option values into the buf and reply ** return total length of reply filled into ptr ** ptr should point to a buffer with 9*len +1 to be safe!! @@ -7252,8 +7700,8 @@ static ErlDrvSSizeT inet_fill_opts(inet_descriptor* desc, continue; #endif case INET_OPT_TOS: -#if defined(IP_TOS) && defined(SOL_IP) - proto = SOL_IP; +#if defined(IP_TOS) && defined(IPPROTO_IP) + proto = IPPROTO_IP; type = IP_TOS; break; #else @@ -7262,14 +7710,50 @@ static ErlDrvSSizeT inet_fill_opts(inet_descriptor* desc, continue; #endif case INET_OPT_TCLASS: -#if defined(IPV6_TCLASS) && defined(SOL_IPV6) - proto = SOL_IPV6; +#if defined(IPV6_TCLASS) && defined(IPPROTO_IPV6) + proto = IPPROTO_IPV6; type = IPV6_TCLASS; break; #else TRUNCATE_TO(0,ptr); continue; #endif + case INET_OPT_TTL: +#if defined(IP_TTL) && defined(IPPROTO_IP) + proto = IPPROTO_IP; + type = IP_TTL; + break; +#else + TRUNCATE_TO(0,ptr); + continue; +#endif + case INET_OPT_RECVTOS: +#if defined(IP_RECVTOS) && defined(IPPROTO_IP) + proto = IPPROTO_IP; + type = IP_RECVTOS; + break; +#else + TRUNCATE_TO(0,ptr); + continue; +#endif + case INET_OPT_RECVTCLASS: +#if defined(IPV6_RECVTCLASS) && defined(IPPROTO_IPV6) + proto = IPPROTO_IPV6; + type = IPV6_RECVTCLASS; + break; +#else + TRUNCATE_TO(0,ptr); + continue; +#endif + case INET_OPT_RECVTTL: +#if defined(IP_RECVTTL) && defined(IPPROTO_IP) + proto = IPPROTO_IP; + type = IP_RECVTTL; + break; +#else + TRUNCATE_TO(0,ptr); + continue; +#endif case INET_OPT_REUSEADDR: type = SO_REUSEADDR; break; @@ -7295,8 +7779,18 @@ static ErlDrvSSizeT inet_fill_opts(inet_descriptor* desc, proto = IPPROTO_TCP; type = TCP_NODELAY; break; + case TCP_OPT_NOPUSH: +#if defined(INET_TCP_NOPUSH) + proto = IPPROTO_TCP; + type = INET_TCP_NOPUSH; + break; +#else + *ptr++ = opt; + put_int32(0, ptr); + continue; +#endif -#ifdef HAVE_MULTICAST_SUPPORT +#if defined(HAVE_MULTICAST_SUPPORT) && defined(IPPROTO_IP) case UDP_OPT_MULTICAST_TTL: proto = IPPROTO_IP; type = IP_MULTICAST_TTL; @@ -7315,10 +7809,10 @@ static ErlDrvSSizeT inet_fill_opts(inet_descriptor* desc, arg_ptr = (char*) &li_val; type = SO_LINGER; break; -#endif /* HAVE_MULTICAST_SUPPORT */ +#endif /* defined(HAVE_MULTICAST_SUPPORT) && defined(IPPROTO_IP) */ case INET_OPT_IPV6_V6ONLY: -#if HAVE_DECL_IPV6_V6ONLY +#if HAVE_DECL_IPV6_V6ONLY && defined(IPPROTO_IPV6) proto = IPPROTO_IPV6; type = IPV6_V6ONLY; break; @@ -7396,6 +7890,93 @@ static ErlDrvSSizeT inet_fill_opts(inet_descriptor* desc, continue; #endif +#ifndef __WIN32__ + /* Winsock does not have struct cmsghdr */ + case INET_OPT_PKTOPTIONS: { + struct cmsghdr *cmsg, *cmsg_top; + SOCKLEN_T cmsg_sz; + union { + /* Ensure alignment */ + struct cmsghdr hdr; + /* Room for (IP_TOS | IPV6_TCLASS) + IP_TTL */ + char buf[2*CMSG_SPACE(sizeof(int))]; + } cmsgbuf; + /* Select between IPv4 or IPv6 PKTOPTIONS + * depending on the socket protocol family + */ + switch (desc->sfamily) { +#if defined(IPPROTO_IP) && defined(IP_PKTOPTIONS) + case AF_INET: { + proto = IPPROTO_IP; + type = IP_PKTOPTIONS; + } + break; +#endif +#if defined(IPPROTO_IPV6) && defined(IPV6_PKTOPTIONS) && defined(AF_INET6) + case AF_INET6: { + proto = IPPROTO_IPV6; + type = IPV6_PKTOPTIONS; + } + break; +#endif + default: { + RETURN_ERROR(); + } + } /* switch */ + TRUNCATE_TO(0, ptr); + /* Fetch a cmsg buffer from the socket */ + cmsg_sz = sizeof(cmsgbuf.buf); + if (IS_SOCKET_ERROR(sock_getopt(desc->s, proto, type, + cmsgbuf.buf, &cmsg_sz))) { + continue; + } + /* Reply with Opt/8, Length/32, [COpt/8, Value/32]* + * i.e opt, total length and then all returned + * cmsg options and values + */ + PLACE_FOR(1+4, ptr); + *ptr++ = opt; + arg_ptr = ptr; /* Where to put total length */ + arg_sz = 0; /* Total length */ + for (cmsg_top = (struct cmsghdr*)(cmsgbuf.buf + cmsg_sz), + cmsg = (struct cmsghdr*)cmsgbuf.buf; + cmsg < cmsg_top; + cmsg = NXT_CMSG_HDR(cmsg)) { +#define PUT_CMSG_INT32(CMSG_LEVEL, CMSG_TYPE, OPT) \ + if ((cmsg->cmsg_level == CMSG_LEVEL) && \ + (cmsg->cmsg_type == CMSG_TYPE)) { \ + PLACE_FOR(1+4, ptr); \ + *ptr++ = OPT; \ + put_cmsg_int32(cmsg, ptr); \ + arg_sz += 1+4; \ + continue; \ + } +#if defined(IPPROTO_IP) && defined(IP_TOS) + PUT_CMSG_INT32(IPPROTO_IP, IP_TOS, INET_OPT_TOS); +#endif +#if defined(IPPROTO_IPV6) && defined(IPV6_TCLASS) + PUT_CMSG_INT32(IPPROTO_IPV6, IPV6_TCLASS, INET_OPT_TCLASS); +#endif +#if defined(IPPROTO_IP) && defined(IP_TTL) + PUT_CMSG_INT32(IPPROTO_IP, IP_TTL, INET_OPT_TTL); +#endif + /* BSD uses the RECV* names in CMSG fields */ +#if defined(IPPROTO_IP) && defined(IP_RECVTOS) + PUT_CMSG_INT32(IPPROTO_IP, IP_RECVTOS, INET_OPT_TOS); +#endif +#if defined(IPPROTO_IPV6) && defined(IPV6_RECVTCLASS) + PUT_CMSG_INT32(IPPROTO_IPV6, IPV6_RECVTCLASS, INET_OPT_TCLASS); +#endif +#if defined(IPPROTO_IP) && defined(IP_RECVTTL) + PUT_CMSG_INT32(IPPROTO_IP, IP_RECVTTL, INET_OPT_TTL); +#endif +#undef PUT_CMSG_INT32 + } + put_int32(arg_sz, arg_ptr); /* Put total length */ + continue; + } +#endif /* #ifdef __WIN32__ */ + default: RETURN_ERROR(); } @@ -7455,6 +8036,7 @@ static int load_paddrinfo (ErlDrvTermData * spec, int i, return i; } + /* ** "sctp_fill_opts": Returns {ok, Results}, or an error: */ @@ -7704,6 +8286,7 @@ static ErlDrvSSizeT sctp_fill_opts(inet_descriptor* desc, case INET_OPT_PRIORITY : case INET_OPT_TOS : case INET_OPT_TCLASS : + case INET_OPT_TTL : case INET_OPT_IPV6_V6ONLY: case SCTP_OPT_AUTOCLOSE: case SCTP_OPT_MAXSEG : @@ -7711,6 +8294,9 @@ static ErlDrvSSizeT sctp_fill_opts(inet_descriptor* desc, case SCTP_OPT_NODELAY : case SCTP_OPT_DISABLE_FRAGMENTS: case SCTP_OPT_I_WANT_MAPPED_V4_ADDR: + case INET_OPT_RECVTOS : + case INET_OPT_RECVTCLASS : + case INET_OPT_RECVTTL : { int res = 0; unsigned int sz = sizeof(res); @@ -7766,8 +8352,8 @@ static ErlDrvSSizeT sctp_fill_opts(inet_descriptor* desc, } case INET_OPT_TOS: { -# if defined(IP_TOS) && defined(SOL_IP) - proto = SOL_IP; +# if defined(IP_TOS) && defined(IPPROTO_IP) + proto = IPPROTO_IP; type = IP_TOS; is_int = 1; tag = am_tos; @@ -7779,8 +8365,8 @@ static ErlDrvSSizeT sctp_fill_opts(inet_descriptor* desc, } case INET_OPT_TCLASS: { -# if defined(IPV6_TCLASS) && defined(SOL_IPV6) - proto = SOL_IPV6; +# if defined(IPV6_TCLASS) && defined(IPPROTO_IPV6) + proto = IPPROTO_IPV6; type = IPV6_TCLASS; is_int = 1; tag = am_tclass; @@ -7790,8 +8376,60 @@ static ErlDrvSSizeT sctp_fill_opts(inet_descriptor* desc, continue; # endif } + case INET_OPT_TTL: + { +# if defined(IP_TTL) && defined(IPPROTO_IP) + proto = IPPROTO_IP; + type = IP_TTL; + is_int = 1; + tag = am_ttl; + break; +# else + /* Not supported -- ignore */ + continue; +# endif + } + case INET_OPT_RECVTOS: + { +# if defined(IP_RECVTOS) && defined(IPPROTO_IP) + proto = IPPROTO_IP; + type = IP_RECVTOS; + is_int = 0; + tag = am_recvtos; + break; +# else + /* Not supported -- ignore */ + continue; +# endif + } + case INET_OPT_RECVTCLASS: + { +# if defined(IPV6_RECVTCLASS) && defined(IPPROTO_IPV6) + proto = IPPROTO_IPV6; + type = IPV6_RECVTCLASS; + is_int = 0; + tag = am_recvtclass; + break; +# else + /* Not supported -- ignore */ + continue; +# endif + } + case INET_OPT_RECVTTL: + { +# if defined(IP_RECVTTL) && defined(IPPROTO_IP) + proto = IPPROTO_IP; + type = IP_RECVTTL; + is_int = 0; + tag = am_recvttl; + break; +# else + /* Not supported -- ignore */ + continue; +# endif + } case INET_OPT_IPV6_V6ONLY: -# if HAVE_DECL_IPV6_V6ONLY +# if HAVE_DECL_IPV6_V6ONLY && defined(IPPROTO_IPV6) { proto = IPPROTO_IPV6; type = IPV6_V6ONLY; @@ -8456,6 +9094,8 @@ static ErlDrvData inet_start(ErlDrvPort port, int size, int protocol) desc->netns = NULL; #endif + desc->recv_cmsgflags = 0; + return (ErlDrvData)desc; } @@ -9273,6 +9913,13 @@ static void tcp_inet_stop(ErlDrvData e) * will be freed through tcp_inet_stop later on. */ static void tcp_desc_close(tcp_descriptor* desc) { +#ifdef HAVE_SENDFILE + if(desc->tcp_add_flags & TCP_ADDF_SENDFILE) { + desc->tcp_add_flags &= ~TCP_ADDF_SENDFILE; + close(desc->sendfile.dup_file_fd); + } +#endif + tcp_clear_input(desc); tcp_clear_output(desc); @@ -9611,6 +10258,60 @@ static ErlDrvSSizeT tcp_inet_ctl(ErlDrvData e, unsigned int cmd, return ctl_reply(INET_REP_OK, NULL, 0, rbuf, rsize); } } + + case TCP_REQ_SENDFILE: { +#ifdef HAVE_SENDFILE + const ErlDrvSizeT required_len = + sizeof(desc->sendfile.dup_file_fd) + + sizeof(Uint64) * 2; + + int raw_file_fd; + + DEBUGF(("tcp_inet_ctl(%ld): SENDFILE\r\n", (long)desc->inet.port)); + + if (len != required_len) { + return ctl_error(EINVAL, rbuf, rsize); + } else if (!IS_CONNECTED(INETP(desc))) { + return ctl_error(ENOTCONN, rbuf, rsize); + } + + sys_memcpy(&raw_file_fd, buf, sizeof(raw_file_fd)); + buf += sizeof(raw_file_fd); + + desc->sendfile.dup_file_fd = dup(raw_file_fd); + + if(desc->sendfile.dup_file_fd == -1) { + return ctl_error(errno, rbuf, rsize); + } + + desc->sendfile.offset = get_int64(buf); + buf += sizeof(Uint64); + + desc->sendfile.length = get_int64(buf); + buf += sizeof(Uint64); + + ASSERT(desc->sendfile.offset >= 0); + ASSERT(desc->sendfile.length >= 0); + + desc->sendfile.ioq_skip = driver_sizeq(desc->inet.port); + desc->sendfile.bytes_sent = 0; + + desc->inet.caller = driver_caller(desc->inet.port); + desc->tcp_add_flags |= TCP_ADDF_SENDFILE; + + /* See if we can finish sending without selecting & rescheduling. */ + tcp_inet_sendfile(desc); + + if(desc->sendfile.length > 0) { + sock_select(INETP(desc), FD_WRITE, 1); + } + + return ctl_reply(INET_REP_OK, NULL, 0, rbuf, rsize); +#else + return ctl_error(ENOTSUP, rbuf, rsize); +#endif + } + default: DEBUGF(("tcp_inet_ctl(%ld): %u\r\n", (long)desc->inet.port, cmd)); return inet_ctl(INETP(desc), cmd, buf, len, rbuf, rsize); @@ -9750,12 +10451,27 @@ static void tcp_inet_commandv(ErlDrvData e, ErlIOVec* ev) static void tcp_inet_flush(ErlDrvData e) { tcp_descriptor* desc = (tcp_descriptor*)e; - if (!(desc->inet.event_mask & FD_WRITE)) { - /* Discard send queue to avoid hanging port (OTP-7615) */ - tcp_clear_output(desc); + int discard_output; + + /* Discard send queue to avoid hanging port (OTP-7615) */ + discard_output = !(desc->inet.event_mask & FD_WRITE); + + discard_output |= desc->tcp_add_flags & TCP_ADDF_LINGER_ZERO; + +#ifdef HAVE_SENDFILE + /* The old file driver aborted when it was stopped during sendfile, so + * we'll clear the flag and discard all output. */ + if(desc->tcp_add_flags & TCP_ADDF_SENDFILE) { + desc->tcp_add_flags &= ~TCP_ADDF_SENDFILE; + close(desc->sendfile.dup_file_fd); + + discard_output = 1; + } +#endif + + if (discard_output) { + tcp_clear_output(desc); } - if (desc->tcp_add_flags & TCP_ADDF_LINGER_ZERO) - tcp_clear_output(desc); } static void tcp_inet_process_exit(ErlDrvData e, ErlDrvMonitor *monitorp) @@ -9817,6 +10533,12 @@ static int tcp_recv_closed(tcp_descriptor* desc) set_busy_port(desc->inet.port, 0); inet_reply_error_am(INETP(desc), am_closed); DEBUGF(("tcp_recv_closed(%ld): busy reply 'closed'\r\n", port)); + } else { + /* No blocking send op to reply to right now. + * If next op is a send, make sure it returns {error,closed} + * rather than {error,enotconn}. + */ + desc->tcp_add_flags |= TCP_ADDF_DELAYED_CLOSE_SEND; } if (!desc->inet.active) { /* We must cancel any timer here ! */ @@ -10650,7 +11372,9 @@ static int tcp_sendv(tcp_descriptor* desc, ErlIOVec* ev) ev->size += h_len; } - if ((sz = driver_sizeq(ix)) > 0) { + sz = driver_sizeq(ix); + + if ((desc->tcp_add_flags & TCP_ADDF_SENDFILE) || sz > 0) { driver_enqv(ix, ev, 0); if (sz+ev->size >= desc->high) { DEBUGF(("tcp_sendv(%ld): s=%d, sender forced busy\r\n", @@ -10744,8 +11468,9 @@ static int tcp_send(tcp_descriptor* desc, char* ptr, ErlDrvSizeT len) inet_output_count(INETP(desc), len+h_len); + sz = driver_sizeq(ix); - if ((sz = driver_sizeq(ix)) > 0) { + if ((desc->tcp_add_flags & TCP_ADDF_SENDFILE) || sz > 0) { if (h_len > 0) driver_enq(ix, buf, h_len); driver_enq(ix, ptr, len); @@ -10835,6 +11560,246 @@ static void tcp_inet_drv_input(ErlDrvData data, ErlDrvEvent event) (void)tcp_inet_input((tcp_descriptor*)data, (HANDLE)event); } +#ifdef HAVE_SENDFILE +static int tcp_sendfile_completed(tcp_descriptor* desc) { + ErlDrvTermData spec[LOAD_PORT_CNT + LOAD_TUPLE_CNT * 2 + + LOAD_ATOM_CNT * 2 + LOAD_UINT_CNT * 2]; + Uint32 sent_low, sent_high; + int i; + + desc->tcp_add_flags &= ~TCP_ADDF_SENDFILE; + close(desc->sendfile.dup_file_fd); + + /* While we flushed the output queue prior to sending the file, we've + * deferred clearing busy status until now as there's no point in doing so + * while we still have a file to send. + * + * The watermark is checked since more data may have been added while we + * were sending the file. */ + + if (driver_sizeq(desc->inet.port) <= desc->low) { + if (IS_BUSY(INETP(desc))) { + desc->inet.caller = desc->inet.busy_caller; + desc->inet.state &= ~INET_F_BUSY; + + set_busy_port(desc->inet.port, 0); + + /* if we have a timer then cancel and send ok to client */ + if (desc->busy_on_send) { + driver_cancel_timer(desc->inet.port); + desc->busy_on_send = 0; + } + + inet_reply_ok(INETP(desc)); + } + } + + if (driver_sizeq(desc->inet.port) == 0) { + sock_select(INETP(desc), FD_WRITE, 0); + send_empty_out_q_msgs(INETP(desc)); + + if (desc->tcp_add_flags & TCP_ADDF_PENDING_SHUTDOWN) { + tcp_shutdown_async(desc); + } + } + + sent_low = ((Uint64)desc->sendfile.bytes_sent >> 0) & 0xFFFFFFFF; + sent_high = ((Uint64)desc->sendfile.bytes_sent >> 32) & 0xFFFFFFFF; + + i = LOAD_ATOM(spec, 0, am_sendfile); + i = LOAD_PORT(spec, i, desc->inet.dport); + i = LOAD_ATOM(spec, i, am_ok); + i = LOAD_UINT(spec, i, sent_low); + i = LOAD_UINT(spec, i, sent_high); + i = LOAD_TUPLE(spec, i, 3); + i = LOAD_TUPLE(spec, i, 3); + + ASSERT(i == sizeof(spec)/sizeof(*spec)); + + return erl_drv_output_term(desc->inet.dport, spec, i); +} + +static int tcp_sendfile_aborted(tcp_descriptor* desc, int socket_error) { + ErlDrvTermData spec[LOAD_PORT_CNT + LOAD_TUPLE_CNT * 2 + LOAD_ATOM_CNT * 3]; + int i; + + /* We don't clean up sendfile state here, as that's done in tcp_desc_close + * following normal error handling. All we do here is report the failure. */ + + i = LOAD_ATOM(spec, 0, am_sendfile); + i = LOAD_PORT(spec, i, desc->inet.dport); + i = LOAD_ATOM(spec, i, am_error); + + switch (socket_error) { + case ECONNRESET: + case ENOTCONN: + case EPIPE: + i = LOAD_ATOM(spec, i, am_closed); + break; + default: + i = LOAD_ATOM(spec, i, error_atom(socket_error)); + } + + i = LOAD_TUPLE(spec, i, 2); + i = LOAD_TUPLE(spec, i, 3); + + ASSERT(i == sizeof(spec)/sizeof(*spec)); + + return erl_drv_output_term(desc->inet.dport, spec, i); +} + +static int tcp_inet_sendfile(tcp_descriptor* desc) { + ErlDrvPort ix = desc->inet.port; + int result = 0; + ssize_t n; + + DEBUGF(("tcp_inet_sendfile(%ld) {s=%d\r\n", (long)ix, desc->inet.s)); + + /* If there was any data in the queue by the time sendfile was issued, + * we'll need to skip it first. Note that we don't clear busy status until + * we're finished sending the file. */ + while (desc->sendfile.ioq_skip > 0) { + ssize_t bytes_to_send; + SysIOVec* iov; + int vsize; + + ASSERT(driver_sizeq(ix) >= desc->sendfile.ioq_skip); + + if ((iov = driver_peekq(ix, &vsize)) == NULL) { + ERTS_INTERNAL_ERROR("ioq empty when sendfile.ioq_skip > 0"); + } + + bytes_to_send = MIN(desc->sendfile.ioq_skip, iov[0].iov_len); + n = sock_send(desc->inet.s, iov[0].iov_base, bytes_to_send, 0); + + if (!IS_SOCKET_ERROR(n)) { + desc->sendfile.ioq_skip -= n; + driver_deq(ix, n); + } else if (sock_errno() == ERRNO_BLOCK) { +#ifdef __WIN32__ + desc->inet.send_would_block = 1; +#endif + goto done; + } else if (sock_errno() != EINTR) { + goto socket_error; + } + } + + while (desc->sendfile.length > 0) { + /* For some reason the maximum ssize_t cannot be used as the max size. + * 1GB seems to work on all platforms */ + const Sint64 SENDFILE_CHUNK_SIZE = ((1UL << 30) - 1); + + ssize_t bytes_to_send = MIN(SENDFILE_CHUNK_SIZE, desc->sendfile.length); + off_t offset = desc->sendfile.offset; + +#if defined(__linux__) + n = sendfile(desc->inet.s, desc->sendfile.dup_file_fd, &offset, + bytes_to_send); +#elif defined(__FreeBSD__) || defined(__DragonFly__) || defined(__DARWIN__) + { + off_t bytes_sent; + int error; + + #if defined(__DARWIN__) + bytes_sent = bytes_to_send; + + error = sendfile(desc->sendfile.dup_file_fd, desc->inet.s, offset, + &bytes_sent, NULL, 0); + n = bytes_sent; + #else + error = sendfile(desc->sendfile.dup_file_fd, desc->inet.s, offset, + bytes_to_send, NULL, &bytes_sent, 0); + n = bytes_sent; + #endif + + if(error < 0) { + /* EAGAIN/EINTR report partial success by setting bytes_sent, + * so we have to skip error handling if nonzero, and skip EOF + * handling if zero, as it's possible that we didn't manage to + * send anything at all before being interrupted by a + * signal. */ + if((errno != EAGAIN && errno != EINTR) || bytes_sent == 0) { + n = -1; + } + } + } +#elif defined(__sun) && defined(__SVR4) && defined(HAVE_SENDFILEV) + { + sendfilevec_t sfvec[1]; + size_t bytes_sent; + ssize_t error; + + sfvec[0].sfv_fd = desc->sendfile.dup_file_fd; + sfvec[0].sfv_len = bytes_to_send; + sfvec[0].sfv_off = offset; + sfvec[0].sfv_flag = 0; + + error = sendfilev(desc->inet.s, sfvec, 1, &bytes_sent); + n = bytes_sent; + + if(error < 0) { + if(errno == EINVAL) { + /* On some solaris versions (I've seen it on SunOS 5.10), + * using a sfv_len larger than the filesize will result in + * a (-1 && errno == EINVAL). We translate this to a + * successful send of the data.*/ + } else { + /* EAGAIN/EINTR behavior is identical to *BSD. */ + if((errno != EAGAIN && errno != EINTR) || bytes_sent == 0) { + n = -1; + } + } + } + } +#else + #error "Unsupported sendfile syscall; update configure test." +#endif + + if (n > 0) { + desc->sendfile.bytes_sent += n; + desc->sendfile.offset += n; + desc->sendfile.length -= n; + } else if (n == 0) { + /* EOF. */ + desc->sendfile.length = 0; + break; + } else if (IS_SOCKET_ERROR(n) && sock_errno() != EINTR) { + if (sock_errno() != ERRNO_BLOCK) { + goto socket_error; + } + +#ifdef __WIN32__ + desc->inet.send_would_block = 1; +#endif + break; + } + } + + if (desc->sendfile.length == 0) { + tcp_sendfile_completed(desc); + } + + goto done; + +socket_error: { + int socket_errno = sock_errno(); + + DEBUGF(("tcp_inet_sendfile(%ld): send errno = %d (errno %d)\r\n", + (long)desc->inet.port, socket_errno, errno)); + + result = tcp_send_error(desc, socket_errno); + tcp_sendfile_aborted(desc, socket_errno); + + goto done; + } + +done: + DEBUGF(("tcp_inet_sendfile(%ld) }\r\n", (long)desc->inet.port)); + return result; +} +#endif /* HAVE_SENDFILE */ + /* socket ready for ouput: ** 1. INET_STATE_CONNECTING => non block connect ? ** 2. INET_STATE_CONNECTED => write output @@ -10895,7 +11860,14 @@ static int tcp_inet_output(tcp_descriptor* desc, HANDLE event) async_ok(INETP(desc)); } else if (IS_CONNECTED(INETP(desc))) { - for (;;) { + +#ifdef HAVE_SENDFILE + if(desc->tcp_add_flags & TCP_ADDF_SENDFILE) { + return tcp_inet_sendfile(desc); + } +#endif + + for (;;) { int vsize; ssize_t n; SysIOVec* iov; @@ -11301,24 +12273,20 @@ static ErlDrvSSizeT packet_inet_ctl(ErlDrvData e, unsigned int cmd, char* buf, (desc->sfamily, &remote, &buf, &len)) != NULL) return ctl_xerror(xerror, rbuf, rsize); - sock_select(desc, FD_CONNECT, 1); code = sock_connect(desc->s, &remote.sa, len); if (IS_SOCKET_ERROR(code) && (sock_errno() == EINPROGRESS)) { /* XXX: Unix only -- WinSock would have a different cond! */ - desc->state = INET_STATE_CONNECTING; if (timeout != INET_INFINITY) driver_set_timer(desc->port, timeout); enq_async(desc, tbuf, INET_REQ_CONNECT); + async_ok(desc); } else if (code == 0) { /* OK we are connected */ - sock_select(desc, FD_CONNECT, 0); - desc->state = INET_STATE_CONNECTED; enq_async(desc, tbuf, INET_REQ_CONNECT); async_ok(desc); } else { - sock_select(desc, FD_CONNECT, 0); return ctl_error(sock_errno(), rbuf, rsize); } return ctl_reply(INET_REP_OK, tbuf, 2, rbuf, rsize); @@ -11546,10 +12514,10 @@ static void packet_inet_command(ErlDrvData e, char* buf, ErlDrvSizeT len) if (IS_SCTP(desc)) { ErlDrvSizeT data_len; - struct iovec iov[1]; /* For real data */ - struct msghdr mhdr; /* Message wrapper */ - struct sctp_sndrcvinfo *sri; /* The actual ancilary data */ - union { /* For ancilary data */ + struct iovec iov[1]; /* For real data */ + struct msghdr mhdr; /* Message wrapper */ + struct sctp_sndrcvinfo *sri; /* The actual ancillary data */ + union { /* For ancillary data */ struct cmsghdr hdr; char ancd[CMSG_SPACE(sizeof(*sri))]; } cmsg; @@ -11559,12 +12527,12 @@ static void packet_inet_command(ErlDrvData e, char* buf, ErlDrvSizeT len) return; } - /* The ancilary data */ + /* The ancillary data */ sri = (struct sctp_sndrcvinfo *) (CMSG_DATA(&cmsg.hdr)); /* Get the "sndrcvinfo" from the buffer, advancing the "ptr": */ ptr = sctp_get_sendparams(sri, ptr); - /* The ancilary data wrapper */ + /* The ancillary data wrapper */ cmsg.hdr.cmsg_level = IPPROTO_SCTP; cmsg.hdr.cmsg_type = SCTP_SNDRCV; cmsg.hdr.cmsg_len = CMSG_LEN(sizeof(*sri)); @@ -11579,7 +12547,7 @@ static void packet_inet_command(ErlDrvData e, char* buf, ErlDrvSizeT len) iov[0].iov_base = ptr; /* The real data */ mhdr.msg_iov = iov; mhdr.msg_iovlen = 1; - mhdr.msg_control = cmsg.ancd; /* For ancilary data */ + mhdr.msg_control = cmsg.ancd; /* For ancillary data */ mhdr.msg_controllen = cmsg.hdr.cmsg_len; VALGRIND_MAKE_MEM_DEFINED(mhdr.msg_control, mhdr.msg_controllen); /*suppress "uninitialised bytes"*/ mhdr.msg_flags = 0; /* Not used with "sendmsg" */ @@ -11663,10 +12631,12 @@ static int packet_inet_input(udp_descriptor* udesc, HANDLE event) char abuf[sizeof(inet_address)]; /* buffer address; enough??? */ int packet_count = udesc->read_packets; int count = 0; /* number of packets delivered to owner */ -#ifdef HAVE_SCTP +#ifndef __WIN32__ struct msghdr mhdr; /* Top-level msg structure */ struct iovec iov[1]; /* Data or Notification Event */ - char ancd[SCTP_ANC_BUFF_SIZE]; /* Ancillary Data */ + char ancd[ANC_BUFF_SIZE]; /* Ancillary Data */ +#endif +#ifdef HAVE_SCTP int short_recv = 0; #endif @@ -11676,15 +12646,11 @@ static int packet_inet_input(udp_descriptor* udesc, HANDLE event) sys_memzero((char *) &other, sizeof(other)); /* udesc->i_buf is only kept between SCTP fragments */ - if (udesc->i_buf == NULL) { - udesc->i_bufsz = desc->bufsz + len; - if ((udesc->i_buf = alloc_buffer(udesc->i_bufsz)) == NULL) - return packet_error(udesc, ENOMEM); - /* pointer to message start */ - udesc->i_ptr = udesc->i_buf->orig_bytes + len; - } else { - ErlDrvBinary* tmp; +#ifdef HAVE_SCTP + if (udesc->i_buf != NULL) { + ErlDrvBinary* tmp; int bufsz; + ASSERT(IS_SCTP(desc)); bufsz = desc->bufsz + (udesc->i_ptr - udesc->i_buf->orig_bytes); if ((tmp = realloc_buffer(udesc->i_buf, bufsz)) == NULL) { release_buffer(udesc->i_buf); @@ -11696,6 +12662,15 @@ static int packet_inet_input(udp_descriptor* udesc, HANDLE event) udesc->i_buf = tmp; udesc->i_bufsz = bufsz; } + } else +#endif + { + ASSERT(udesc->i_buf == NULL); + udesc->i_bufsz = desc->bufsz + len; + if ((udesc->i_buf = alloc_buffer(udesc->i_bufsz)) == NULL) + return packet_error(udesc, ENOMEM); + /* pointer to message start */ + udesc->i_ptr = udesc->i_buf->orig_bytes + len; } /* Note: On Windows NT, recvfrom() fails if the socket is connected. */ @@ -11710,7 +12685,7 @@ static int packet_inet_input(udp_descriptor* udesc, HANDLE event) mhdr.msg_iov = iov; mhdr.msg_iovlen = 1; mhdr.msg_control = ancd; - mhdr.msg_controllen = SCTP_ANC_BUFF_SIZE; + mhdr.msg_controllen = ANC_BUFF_SIZE; mhdr.msg_flags = 0; /* To be filled by "recvmsg" */ /* Do the actual SCTP receive: */ @@ -11725,6 +12700,24 @@ static int packet_inet_input(udp_descriptor* udesc, HANDLE event) other = desc->remote; goto check_result; } +#ifndef __WIN32__ + /* recvmsg() does not exist in the Winsock API */ + if (desc->recv_cmsgflags) { + /* Use recvmsg() */ + iov->iov_base = udesc->i_ptr; + iov->iov_len = desc->bufsz; + mhdr.msg_name = &other; + mhdr.msg_namelen = len; + mhdr.msg_iov = iov; + mhdr.msg_iovlen = 1; + mhdr.msg_control = ancd; + mhdr.msg_controllen = ANC_BUFF_SIZE; + mhdr.msg_flags = 0; + n = sock_recvmsg(desc->s, &mhdr, 0); + len = mhdr.msg_namelen; + goto check_result; + } +#endif n = sock_recvfrom(desc->s, udesc->i_ptr, desc->bufsz, 0, &other.sa, &len); check_result: @@ -11754,6 +12747,14 @@ static int packet_inet_input(udp_descriptor* udesc, HANDLE event) ) { sock_select(desc,FD_READ,1); } +#ifdef HAVE_SCTP + if (!short_recv) { +#endif + release_buffer(udesc->i_buf); + udesc->i_buf = NULL; +#ifdef HAVE_SCTP + } +#endif return count; /* strange, not ready */ } @@ -11769,7 +12770,7 @@ static int packet_inet_input(udp_descriptor* udesc, HANDLE event) { /* message received */ int code; - void * extra = NULL; + void *mp; char * ptr; int nsz; @@ -11800,14 +12801,18 @@ static int packet_inet_input(udp_descriptor* udesc, HANDLE event) udesc->i_ptr = NULL; /* not used from here */ } } + mp = NULL; #ifdef HAVE_SCTP - if (IS_SCTP(desc)) extra = &mhdr; + if (IS_SCTP(desc)) mp = &mhdr; +#endif +#ifndef __WIN32__ + if (desc->recv_cmsgflags) mp = &mhdr; #endif /* Actual parsing and return of the data received, occur here: */ code = packet_reply_binary_data(desc, len, udesc->i_buf, (sizeof(other) - len), nsz, - extra); + mp); free_buffer(udesc->i_buf); udesc->i_buf = NULL; if (code < 0) @@ -11833,77 +12838,6 @@ static int packet_inet_input(udp_descriptor* udesc, HANDLE event) return count; } -static void packet_inet_drv_output(ErlDrvData e, ErlDrvEvent event) -{ - (void) packet_inet_output((udp_descriptor*)e, (HANDLE)event); -} - -/* UDP/SCTP socket ready for output: -** This is a Back-End for Non-Block SCTP Connect (INET_STATE_CONNECTING) -*/ -static int packet_inet_output(udp_descriptor* udesc, HANDLE event) -{ - inet_descriptor* desc = INETP(udesc); - int ret = 0; - ErlDrvPort ix = desc->port; - - DEBUGF(("packet_inet_output(%ld) {s=%d\r\n", - (long)desc->port, desc->s)); - - if (desc->state == INET_STATE_CONNECTING) { - sock_select(desc, FD_CONNECT, 0); - - driver_cancel_timer(ix); /* posssibly cancel a timer */ -#ifndef __WIN32__ - /* - * XXX This is strange. This *should* work on Windows NT too, - * but doesn't. An bug in Winsock 2.0 for Windows NT? - * - * See "Unix Netwok Programming", W.R.Stevens, p 412 for a - * discussion about Unix portability and non blocking connect. - */ - -#ifndef SO_ERROR - { - int sz = sizeof(desc->remote); - int code = sock_peer(desc->s, - (struct sockaddr*) &desc->remote, &sz); - - if (IS_SOCKET_ERROR(code)) { - desc->state = INET_STATE_OPEN; /* restore state */ - ret = async_error(desc, sock_errno()); - goto done; - } - } -#else - { - int error = 0; /* Has to be initiated, we check it */ - unsigned int sz = sizeof(error); /* even if we get -1 */ - int code = sock_getopt(desc->s, SOL_SOCKET, SO_ERROR, - (void *)&error, &sz); - - if ((code < 0) || error) { - desc->state = INET_STATE_OPEN; /* restore state */ - ret = async_error(desc, error); - goto done; - } - } -#endif /* SO_ERROR */ -#endif /* !__WIN32__ */ - - desc->state = INET_STATE_CONNECTED; - async_ok(desc); - } - else { - sock_select(desc,FD_CONNECT,0); - - DEBUGF(("packet_inet_output(%ld): bad state: %04x\r\n", - (long)desc->port, desc->state)); - } - done: - DEBUGF(("packet_inet_output(%ld) }\r\n", (long)desc->port)); - return ret; -} #endif /*---------------------------------------------------------------------------*/ |