diff --git a/configure.ac b/configure.ac index e1b0c0ea..fb486abf 100644 --- a/configure.ac +++ b/configure.ac @@ -1,389 +1,392 @@ # # Copyright (C) 2010-2015 Red Hat, Inc. All rights reserved. # # Authors: Fabio M. Di Nitto # Federico Simoncelli # # This software licensed under GPL-2.0+, LGPL-2.0+ # # -*- Autoconf -*- # Process this file with autoconf to produce a configure script. # AC_PREREQ([2.63]) AC_INIT([kronosnet], m4_esyscmd([build-aux/git-version-gen .tarball-version]), [devel@lists.kronosnet.org]) AC_USE_SYSTEM_EXTENSIONS AM_INIT_AUTOMAKE([1.11.1 dist-bzip2 dist-xz color-tests -Wno-portability subdir-objects]) LT_PREREQ([2.2.6]) LT_INIT AC_CONFIG_MACRO_DIR([m4]) AC_CONFIG_SRCDIR([kronosnetd/main.c]) AC_CONFIG_HEADERS([config.h]) AC_CANONICAL_HOST AC_PROG_LIBTOOL AC_LANG([C]) systemddir=${prefix}/lib/systemd/system if test "$prefix" = "NONE"; then prefix="/usr" if test "$localstatedir" = "\${prefix}/var"; then localstatedir="/var" fi if test "$sysconfdir" = "\${prefix}/etc"; then sysconfdir="/etc" fi if test "$systemddir" = "NONE/lib/systemd/system"; then systemddir=/lib/systemd/system fi if test "$libdir" = "\${exec_prefix}/lib"; then if test -e /usr/lib64; then libdir="/usr/lib64" else libdir="/usr/lib" fi fi fi # Checks for programs. if ! ${MAKE-make} --version /cannot/make/this >/dev/null 2>&1; then AC_MSG_ERROR(["you don't seem to have GNU make; it is required"]) fi AC_PROG_AWK AC_PROG_GREP AC_PROG_SED AC_PROG_CPP AC_PROG_CC AM_PROG_CC_C_O AC_PROG_LN_S AC_PROG_INSTALL AC_PROG_MAKE_SET AC_PROG_CXX AC_PROG_RANLIB AC_CHECK_PROGS([PUBLICAN], [publican], [:]) AC_CHECK_PROGS([PKGCONFIG], [pkg-config]) AC_ARG_ENABLE([kronosnetd], [ --enable-kronosnetd : Kronosnetd support ],, [ enable_kronosnetd="no" ]) AM_CONDITIONAL([BUILD_KRONOSNETD], test x$enable_kronosnetd = xyes) AC_ARG_ENABLE([libtap], [ --enable-libtap : libtap support ],, [ enable_libtap="no" ]) if test "x$enable_kronosnetd" = xyes; then enable_libtap=yes fi AM_CONDITIONAL([BUILD_LIBTAP], test x$enable_libtap = xyes) AC_ARG_ENABLE([libknet-sctp], [ --enable-libknet-sctp : libknet SCTP support ],, [ enable_libknet_sctp="yes" ]) ## local helper functions # this function checks if CC support options passed as # args. Global CFLAGS are ignored during this test. cc_supports_flag() { saveCPPFLAGS="$CPPFLAGS" CPPFLAGS="$@" if echo $CC | grep -q clang; then CPPFLAGS="-Werror $CPPFLAGS" fi AC_MSG_CHECKING([whether $CC supports "$@"]) AC_PREPROC_IFELSE([AC_LANG_PROGRAM([])], [RC=0; AC_MSG_RESULT([yes])], [RC=1; AC_MSG_RESULT([no])]) CPPFLAGS="$saveCPPFLAGS" return $RC } # helper macro to check libs without adding them to LIBS check_lib_no_libs() { lib_no_libs_arg1=$1 shift lib_no_libs_arg2=$1 shift lib_no_libs_args=$@ AC_CHECK_LIB([$lib_no_libs_arg1], [$lib_no_libs_arg2],,, [$lib_no_libs_args]) LIBS=$ac_check_lib_save_LIBS } # Checks for C features AC_C_INLINE # Checks for libraries. AC_CHECK_LIB([pthread], [pthread_create]) AC_CHECK_LIB([m], [ceil]) AC_CHECK_LIB([rt], [clock_gettime]) PKG_CHECK_MODULES([nss],[nss]) # Checks for header files. AC_CHECK_HEADERS([fcntl.h]) AC_CHECK_HEADERS([stdlib.h]) AC_CHECK_HEADERS([string.h]) AC_CHECK_HEADERS([strings.h]) AC_CHECK_HEADERS([sys/ioctl.h]) AC_CHECK_HEADERS([syslog.h]) AC_CHECK_HEADERS([unistd.h]) AC_CHECK_HEADERS([netinet/in.h]) AC_CHECK_HEADERS([sys/socket.h]) AC_CHECK_HEADERS([arpa/inet.h]) AC_CHECK_HEADERS([netdb.h]) AC_CHECK_HEADERS([limits.h]) AC_CHECK_HEADERS([stdint.h]) if test "x$enable_libknet_sctp" = xyes; then AC_CHECK_HEADERS([netinet/sctp.h],, AC_MSG_ERROR(["missing required SCTP headers"])) fi # Checks for typedefs, structures, and compiler characteristics. AC_C_INLINE AC_TYPE_SIZE_T AC_TYPE_PID_T AC_TYPE_SSIZE_T AC_TYPE_UINT8_T AC_TYPE_UINT16_T AC_TYPE_UINT32_T AC_TYPE_UINT64_T AC_TYPE_INT32_T # Checks for library functions. AC_FUNC_ALLOCA AC_FUNC_FORK AC_FUNC_MALLOC AC_FUNC_REALLOC AC_CHECK_FUNCS([memset]) AC_CHECK_FUNCS([strdup]) AC_CHECK_FUNCS([strerror]) AC_CHECK_FUNCS([dup2]) AC_CHECK_FUNCS([select]) AC_CHECK_FUNCS([socket]) AC_CHECK_FUNCS([inet_ntoa]) AC_CHECK_FUNCS([memmove]) AC_CHECK_FUNCS([strchr]) AC_CHECK_FUNCS([atexit]) AC_CHECK_FUNCS([ftruncate]) AC_CHECK_FUNCS([strrchr]) AC_CHECK_FUNCS([strstr]) AC_CHECK_FUNCS([clock_gettime]) AC_CHECK_FUNCS([strcasecmp]) AC_CHECK_FUNCS([sendmmsg]) AC_CHECK_FUNCS([recvmmsg]) +AC_CHECK_FUNCS([epoll_wait]) +AC_CHECK_FUNCS([kevent]) +# if neither epoll_wait nor kevent are present, we should fail. # Check entries in specific structs AC_CHECK_MEMBER([struct mmsghdr.msg_hdr], [AC_DEFINE_UNQUOTED([HAVE_MMSGHDR], [1], [struct mmsghdr exists])], [], [[#include ]]) # checks (for kronosnetd) if test "x$enable_kronosnetd" = xyes; then AC_CHECK_HEADERS([security/pam_appl.h], [AC_CHECK_LIB([pam], [pam_start])], [AC_MSG_ERROR([Unable to find LinuxPAM devel files])]) AC_CHECK_HEADERS([security/pam_misc.h], [AC_CHECK_LIB([pam_misc], [misc_conv])], [AC_MSG_ERROR([Unable to find LinuxPAM MISC devel files])]) PKG_CHECK_MODULES([libqb], [libqb]) AC_CHECK_LIB([qb], [qb_log_thread_priority_set], [have_qb_log_thread_priority_set="yes"], [have_qb_log_thread_priority_set="no"]) if test "x${have_qb_log_thread_priority_set}" = xyes; then AC_DEFINE_UNQUOTED([HAVE_QB_LOG_THREAD_PRIORITY_SET], 1, [have qb_log_thread_priority_set]) fi fi # local options AC_ARG_ENABLE([debug], [ --enable-debug enable debug build. ], [ default="no" ]) AC_ARG_ENABLE([publicandocs], [ --enable-publicandocs enable docs build. ], [ default="no" ]) AC_ARG_WITH([initdefaultdir], [ --with-initdefaultdir : path to /etc/sysconfig/.. or /etc/default dir. ], [ INITDEFAULTDIR="$withval" ], [ INITDEFAULTDIR="$sysconfdir/default" ]) AC_ARG_WITH([initddir], [ --with-initddir=DIR : path to init script directory. ], [ INITDDIR="$withval" ], [ INITDDIR="$sysconfdir/init.d" ]) AC_ARG_WITH([systemddir], [ --with-systemddir=DIR : path to systemd unit files directory. ], [ SYSTEMDDIR="$withval" ], [ SYSTEMDDIR="$systemddir" ]) AC_ARG_WITH([syslogfacility], [ --with-syslogfacility=FACILITY default syslog facility. ], [ SYSLOGFACILITY="$withval" ], [ SYSLOGFACILITY="LOG_DAEMON" ]) AC_ARG_WITH([sysloglevel], [ --with-sysloglevel=LEVEL default syslog level. ], [ SYSLOGLEVEL="$withval" ], [ SYSLOGLEVEL="LOG_INFO" ]) AC_ARG_WITH([defaultadmgroup], [ --with-defaultadmgroup=GROUP define PAM group. Users part of this group will be allowed to configure kronosnet. Others will only receive read-only rights. ], [ DEFAULTADMGROUP="$withval" ], [ DEFAULTADMGROUP="kronosnetadm" ]) ## random vars LOGDIR=${localstatedir}/log/ RUNDIR=${localstatedir}/run/ DEFAULT_CONFIG_DIR=${sysconfdir}/kronosnet ## do subst AM_CONDITIONAL([BUILD_DOCS], [test "x${enable_publicandocs}" = xyes]) AM_CONDITIONAL([DEBUG], [test "x${enable_debug}" = xyes]) AC_SUBST([DEFAULT_CONFIG_DIR]) AC_SUBST([INITDEFAULTDIR]) AC_SUBST([INITDDIR]) AC_SUBST([SYSTEMDDIR]) AC_SUBST([LOGDIR]) AC_SUBST([DEFAULTADMGROUP]) AC_DEFINE_UNQUOTED([DEFAULT_CONFIG_DIR], ["$(eval echo ${DEFAULT_CONFIG_DIR})"], [Default config directory]) AC_DEFINE_UNQUOTED([DEFAULT_CONFIG_FILE], ["$(eval echo ${DEFAULT_CONFIG_DIR}/kronosnetd.conf)"], [Default config file]) AC_DEFINE_UNQUOTED([LOGDIR], ["$(eval echo ${LOGDIR})"], [Default logging directory]) AC_DEFINE_UNQUOTED([DEFAULT_LOG_FILE], ["$(eval echo ${LOGDIR}/kronosnetd.log)"], [Default log file]) AC_DEFINE_UNQUOTED([RUNDIR], ["$(eval echo ${RUNDIR})"], [Default run directory]) AC_DEFINE_UNQUOTED([SYSLOGFACILITY], [$(eval echo ${SYSLOGFACILITY})], [Default syslog facility]) AC_DEFINE_UNQUOTED([SYSLOGLEVEL], [$(eval echo ${SYSLOGLEVEL})], [Default syslog level]) AC_DEFINE_UNQUOTED([DEFAULTADMGROUP], ["$(eval echo ${DEFAULTADMGROUP})"], [Default admin group]) ## *FLAGS handling ENV_CFLAGS="$CFLAGS" ENV_CPPFLAGS="$CPPFLAGS" ENV_LDFLAGS="$LDFLAGS" # debug build stuff if test "x${enable_debug}" = xyes; then AC_DEFINE_UNQUOTED([DEBUG], [1], [Compiling Debugging code]) OPT_CFLAGS="-O0" else OPT_CFLAGS="-O3" fi # gdb flags if test "x${GCC}" = xyes; then GDB_FLAGS="-ggdb3" else GDB_FLAGS="-g" fi # extra warnings EXTRA_WARNINGS="" WARNLIST=" all shadow missing-prototypes missing-declarations strict-prototypes declaration-after-statement pointer-arith write-strings cast-align bad-function-cast missing-format-attribute format=2 format-security format-nonliteral no-long-long unsigned-char gnu89-inline no-strict-aliasing error address cpp overflow parentheses sequence-point switch uninitialized unused-but-set-variable unused-function unused-result unused-value unused-variable " for j in $WARNLIST; do if cc_supports_flag -W$j; then EXTRA_WARNINGS="$EXTRA_WARNINGS -W$j"; fi done CFLAGS="$ENV_CFLAGS $lt_prog_compiler_pic $OPT_CFLAGS $GDB_FLAGS \ $EXTRA_WARNINGS $WERROR_CFLAGS" CPPFLAGS="$ENV_CPPFLAGS" LDFLAGS="$ENV_LDFLAGS $lt_prog_compiler_pic -Wl,--as-needed" AC_CONFIG_FILES([ Makefile init/Makefile libtap/Makefile libtap/libtap.pc kronosnetd/Makefile kronosnetd/kronosnetd.logrotate libknet/Makefile libknet/libknet.pc libknet/tests/Makefile docs/Makefile poc-code/Makefile poc-code/iov-hash/Makefile poc-code/access-list/Makefile ]) AC_OUTPUT diff --git a/libknet/compat.c b/libknet/compat.c index 95ebe7f3..8f7c637b 100644 --- a/libknet/compat.c +++ b/libknet/compat.c @@ -1,90 +1,194 @@ /* * Copyright (C) 2016 Red Hat, Inc. All rights reserved. * * Author: Jan Friesse * * This software licensed under GPL-2.0+, LGPL-2.0+ */ #include "config.h" #include #include #include "compat.h" #ifndef HAVE_SENDMMSG int sendmmsg(int sockfd, struct mmsghdr *msgvec, unsigned int vlen, unsigned int flags) { #ifdef SYS_sendmmsg /* * For systems where kernel supports sendmmsg but glibc doesn't (RHEL 6) */ return (syscall(SYS_sendmmsg, sockfd, msgvec, vlen, flags)); #else /* * Generic implementation of sendmmsg using sendmsg */ unsigned int i; ssize_t ret; if (vlen == 0) { return (0); } for (i = 0; i < vlen; i++) { ret = sendmsg(sockfd, &msgvec[i].msg_hdr, flags); if (ret >= 0) { msgvec[i].msg_len = ret; } else { break ; } } return ((ret >= 0) ? vlen : ret); #endif } #endif #ifndef HAVE_RECVMMSG extern int recvmmsg(int sockfd, struct mmsghdr *msgvec, unsigned int vlen, unsigned int flags, struct timespec *timeout) { #ifdef SYS_recvmmsg /* * For systems where kernel supports recvmmsg but glibc doesn't (RHEL 6) */ return (syscall(SYS_recvmmsg, sockfd, msgvec, vlen, flags, timeout)); #else /* * Generic implementation of recvmmsg using recvmsg */ unsigned int i; ssize_t ret; if (vlan == 0) { return (0); } if (timeout != NULL || (flags && MSG_WAITFORONE)) { /* * Not implemented */ errno = EINVAL; return (-1); } for (i = 0; i < vlen; i++) { ret = recvmsg(sockfd, &msgvec[i].msg_hdr, flags); if (ret >= 0) { msgvec[i].msg_len = ret; } else { break ; } } return ((ret >= 0) ? vlen : ret); #endif } #endif + + +#ifndef HAVE_EPOLL +#ifdef HAVE_KEVENT + +/* for FreeBSD which has kevent instead of epoll */ + +#include +#include +#include +#include + +static int32_t +_poll_to_filter_(int32_t event) +{ + int32_t out = 0; + if (event & POLLIN) + out |= EVFILT_READ; + if (event & POLLOUT) + out |= EVFILT_WRITE; + return out; +} + +int epoll_create(int size) +{ + return kqueue(); +} + + +int epoll_ctl(int epfd, int op, int fd, struct epoll_event *event) +{ + int ret = 0; + struct kevent ke; + short filters = _poll_to_filter_(event->events); + + switch (op) { + /* The kevent man page says that EV_ADD also does MOD */ + case EPOLL_CTL_ADD: + case EPOLL_CTL_MOD: + EV_SET(&ke, fd, filters, EV_ADD | EV_ENABLE, 0, 0, event->data.ptr); + break; + case EPOLL_CTL_DEL: + EV_SET(&ke, fd, filters, EV_DELETE, 0, 0, event->data.ptr); + break; + default: + errno = EINVAL; + return -1; + } + ret = kevent(epfd, &ke, 1, NULL, 0, NULL); + return ret; +} + +int epoll_wait(int epfd, struct epoll_event *events, int maxevents, int timeout_ms) +{ + struct kevent kevents[maxevents]; + struct timespec timeout = { 0, 0 }; + struct timespec *timeout_ptr = &timeout; + uint32_t revents; + int event_count; + int i; + int returned_events; + + if (timeout_ms != -1) { + timeout.tv_sec = timeout_ms/1000; + timeout.tv_nsec += (timeout_ms % 1000) * 1000000ULL; + } + else { + timeout_ptr = NULL; + } + + event_count = kevent(epfd, NULL, 0, kevents, maxevents, timeout_ptr); + if (event_count == -1) { + return -1; + } + + returned_events = 0; + for (i = 0; i < event_count; i++) { + revents = 0; + + if (kevents[i].flags & EV_ERROR) { + revents |= POLLERR; + } + if (kevents[i].flags & EV_EOF) { + revents |= POLLHUP; + } + if (kevents[i].filter == EVFILT_READ) { + revents |= POLLIN; + } + if (kevents[i].filter == EVFILT_WRITE) { + revents |= POLLOUT; + } + events[returned_events].events = revents; + events[returned_events].data.ptr = kevents[i].udata; + returned_events++; + } + + return returned_events; +} + +/* ifdef HAVE_KEVENT */ +#endif + +/* ifndef EPOLL */ +#endif diff --git a/libknet/compat.h b/libknet/compat.h index 80bd3053..3637cd24 100644 --- a/libknet/compat.h +++ b/libknet/compat.h @@ -1,36 +1,69 @@ /* * Copyright (C) 2016 Red Hat, Inc. All rights reserved. * * Authors: Jan Friesse * * This software licensed under GPL-2.0+, LGPL-2.0+ */ #ifndef __COMPAT_H__ #define __COMPAT_H__ #include "config.h" #include +#include #ifndef HAVE_MMSGHDR struct mmsghdr { struct msghdr msg_hdr; /* Message header */ unsigned int msg_len; /* Number of bytes transmitted */ }; #endif #ifndef MSG_WAITFORONE #define MSG_WAITFORONE 0x10000 #endif #ifndef HAVE_SENDMMSG extern int sendmmsg(int sockfd, struct mmsghdr *msgvec, unsigned int vlen, unsigned int flags); #endif #ifndef HAVE_RECVMMSG extern int recvmmsg(int sockfd, struct mmsghdr *msgvec, unsigned int vlen, unsigned int flags, struct timespec *timeout); #endif +#ifndef ETIME +#define ETIME ETIMEDOUT +#endif + +#ifndef HAVE_EPOLL + +#include +#define EPOLL_CTL_ADD 1 +#define EPOLL_CTL_MOD 2 +#define EPOLL_CTL_DEL 3 + +#define EPOLLIN POLLIN +#define EPOLLOUT POLLOUT + +typedef union epoll_data { + void *ptr; + int fd; + uint32_t u32; + uint64_t u64; +} epoll_data_t; + +struct epoll_event { + uint32_t events; /* Epoll events */ + epoll_data_t data; /* User data variable */ +}; + +int epoll_create(int size); +int epoll_ctl(int epfd, int op, int fd, struct epoll_event *event); +int epoll_wait(int epfd, struct epoll_event *events, int maxevents, int timeout_ms); + +#endif + #endif diff --git a/libknet/handle.c b/libknet/handle.c index 19428881..f655c47b 100644 --- a/libknet/handle.c +++ b/libknet/handle.c @@ -1,1425 +1,1425 @@ /* * Copyright (C) 2010-2015 Red Hat, Inc. All rights reserved. * * Authors: Fabio M. Di Nitto * Federico Simoncelli * * This software licensed under GPL-2.0+, LGPL-2.0+ */ #include "config.h" #include #include #include #include #include -#include #include #include #include #include #include "internals.h" #include "crypto.h" +#include "compat.h" #include "common.h" #include "threads_common.h" #include "threads_heartbeat.h" #include "threads_pmtud.h" #include "threads_dsthandler.h" #include "threads_send_recv.h" #include "transports.h" #include "logging.h" static pthread_mutex_t handle_config_mutex = PTHREAD_MUTEX_INITIALIZER; static int _init_locks(knet_handle_t knet_h) { int savederrno = 0; savederrno = pthread_rwlock_init(&knet_h->global_rwlock, NULL); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to initialize list rwlock: %s", strerror(savederrno)); goto exit_fail; } knet_h->lock_init_done = 1; savederrno = pthread_rwlock_init(&knet_h->host_rwlock, NULL); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to initialize host rwlock: %s", strerror(savederrno)); goto exit_fail; } savederrno = pthread_mutex_init(&knet_h->host_mutex, NULL); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to initialize host mutex: %s", strerror(savederrno)); goto exit_fail; } savederrno = pthread_cond_init(&knet_h->host_cond, NULL); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to initialize host conditional mutex: %s", strerror(savederrno)); goto exit_fail; } savederrno = pthread_mutex_init(&knet_h->pmtud_mutex, NULL); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to initialize pmtud mutex: %s", strerror(savederrno)); goto exit_fail; } savederrno = pthread_cond_init(&knet_h->pmtud_cond, NULL); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to initialize pmtud conditional mutex: %s", strerror(savederrno)); goto exit_fail; } savederrno = pthread_mutex_init(&knet_h->tx_mutex, NULL); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to initialize tx_thread mutex: %s", strerror(savederrno)); goto exit_fail; } return 0; exit_fail: errno = savederrno; return -1; } static void _destroy_locks(knet_handle_t knet_h) { knet_h->lock_init_done = 0; pthread_rwlock_destroy(&knet_h->global_rwlock); pthread_rwlock_destroy(&knet_h->host_rwlock); pthread_mutex_destroy(&knet_h->host_mutex); pthread_cond_destroy(&knet_h->host_cond); pthread_mutex_destroy(&knet_h->pmtud_mutex); pthread_cond_destroy(&knet_h->pmtud_cond); pthread_mutex_destroy(&knet_h->tx_mutex); } static int _init_socks(knet_handle_t knet_h) { int savederrno = 0; if (_init_socketpair(knet_h, knet_h->hostsockfd)) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to initialize internal hostsockpair: %s", strerror(savederrno)); goto exit_fail; } if (_init_socketpair(knet_h, knet_h->dstsockfd)) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to initialize internal dstsockpair: %s", strerror(savederrno)); goto exit_fail; } return 0; exit_fail: errno = savederrno; return -1; } static void _close_socks(knet_handle_t knet_h) { _close_socketpair(knet_h, knet_h->dstsockfd); _close_socketpair(knet_h, knet_h->hostsockfd); } static int _init_buffers(knet_handle_t knet_h) { int savederrno = 0; int i; size_t bufsize; for (i = 0; i < PCKT_FRAG_MAX; i++) { bufsize = ceil((float)KNET_MAX_PACKET_SIZE / (i + 1)) + KNET_HEADER_ALL_SIZE; knet_h->send_to_links_buf[i] = malloc(bufsize); if (!knet_h->send_to_links_buf[i]) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to allocate memory datafd to link buffer: %s", strerror(savederrno)); goto exit_fail; } memset(knet_h->send_to_links_buf[i], 0, bufsize); knet_h->recv_from_sock_buf[i] = malloc(KNET_DATABUFSIZE); if (!knet_h->recv_from_sock_buf[i]) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to allocate memory for app to datafd buffer: %s", strerror(savederrno)); goto exit_fail; } memset(knet_h->recv_from_sock_buf[i], 0, KNET_DATABUFSIZE); knet_h->recv_from_links_buf[i] = malloc(KNET_DATABUFSIZE); if (!knet_h->recv_from_links_buf[i]) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to allocate memory for link to datafd buffer: %s", strerror(savederrno)); goto exit_fail; } memset(knet_h->recv_from_links_buf[i], 0, KNET_DATABUFSIZE); } knet_h->pingbuf = malloc(KNET_HEADER_PING_SIZE); if (!knet_h->pingbuf) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to allocate memory for hearbeat buffer: %s", strerror(savederrno)); goto exit_fail; } memset(knet_h->pingbuf, 0, KNET_HEADER_PING_SIZE); knet_h->pmtudbuf = malloc(KNET_PMTUD_SIZE_V6); if (!knet_h->pmtudbuf) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to allocate memory for pmtud buffer: %s", strerror(savederrno)); goto exit_fail; } memset(knet_h->pmtudbuf, 0, KNET_PMTUD_SIZE_V6); for (i = 0; i < PCKT_FRAG_MAX; i++) { bufsize = ceil((float)KNET_MAX_PACKET_SIZE / (i + 1)) + KNET_HEADER_ALL_SIZE + KNET_DATABUFSIZE_CRYPT_PAD; knet_h->send_to_links_buf_crypt[i] = malloc(bufsize); if (!knet_h->send_to_links_buf_crypt[i]) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to allocate memory for crypto datafd to link buffer: %s", strerror(savederrno)); goto exit_fail; } memset(knet_h->send_to_links_buf_crypt[i], 0, bufsize); } knet_h->recv_from_links_buf_decrypt = malloc(KNET_DATABUFSIZE_CRYPT); if (!knet_h->recv_from_links_buf_decrypt) { savederrno = errno; log_err(knet_h, KNET_SUB_CRYPTO, "Unable to allocate memory for crypto link to datafd buffer: %s", strerror(savederrno)); goto exit_fail; } memset(knet_h->recv_from_links_buf_decrypt, 0, KNET_DATABUFSIZE_CRYPT); knet_h->recv_from_links_buf_crypt = malloc(KNET_DATABUFSIZE_CRYPT); if (!knet_h->recv_from_links_buf_crypt) { savederrno = errno; log_err(knet_h, KNET_SUB_CRYPTO, "Unable to allocate memory for crypto link to datafd buffer: %s", strerror(savederrno)); goto exit_fail; } memset(knet_h->recv_from_links_buf_crypt, 0, KNET_DATABUFSIZE_CRYPT); knet_h->pingbuf_crypt = malloc(KNET_DATABUFSIZE_CRYPT); if (!knet_h->pingbuf_crypt) { savederrno = errno; log_err(knet_h, KNET_SUB_CRYPTO, "Unable to allocate memory for crypto hearbeat buffer: %s", strerror(savederrno)); goto exit_fail; } memset(knet_h->pingbuf_crypt, 0, KNET_DATABUFSIZE_CRYPT); knet_h->pmtudbuf_crypt = malloc(KNET_DATABUFSIZE_CRYPT); if (!knet_h->pmtudbuf_crypt) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to allocate memory for crypto pmtud buffer: %s", strerror(savederrno)); goto exit_fail; } memset(knet_h->pmtudbuf_crypt, 0, KNET_DATABUFSIZE_CRYPT); memset(knet_h->knet_transport_fd_tracker, KNET_MAX_TRANSPORTS, sizeof(knet_h->knet_transport_fd_tracker)); return 0; exit_fail: errno = savederrno; return -1; } static void _destroy_buffers(knet_handle_t knet_h) { int i; for (i = 0; i < PCKT_FRAG_MAX; i++) { free(knet_h->send_to_links_buf[i]); free(knet_h->recv_from_sock_buf[i]); free(knet_h->send_to_links_buf_crypt[i]); free(knet_h->recv_from_links_buf[i]); } free(knet_h->recv_from_links_buf_decrypt); free(knet_h->recv_from_links_buf_crypt); free(knet_h->pingbuf); free(knet_h->pingbuf_crypt); free(knet_h->pmtudbuf); free(knet_h->pmtudbuf_crypt); } static int _init_epolls(knet_handle_t knet_h) { struct epoll_event ev; int savederrno = 0; /* * even if the kernel does dynamic allocation with epoll_ctl * we need to reserve one extra for host to host communication */ knet_h->send_to_links_epollfd = epoll_create(KNET_EPOLL_MAX_EVENTS + 1); if (knet_h->send_to_links_epollfd < 0) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to create epoll datafd to link fd: %s", strerror(savederrno)); goto exit_fail; } knet_h->recv_from_links_epollfd = epoll_create(KNET_EPOLL_MAX_EVENTS); if (knet_h->recv_from_links_epollfd < 0) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to create epoll link to datafd fd: %s", strerror(savederrno)); goto exit_fail; } knet_h->dst_link_handler_epollfd = epoll_create(KNET_EPOLL_MAX_EVENTS); if (knet_h->dst_link_handler_epollfd < 0) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to create epoll dst cache fd: %s", strerror(savederrno)); goto exit_fail; } if (_fdset_cloexec(knet_h->send_to_links_epollfd)) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to set CLOEXEC on datafd to link epoll fd: %s", strerror(savederrno)); goto exit_fail; } if (_fdset_cloexec(knet_h->recv_from_links_epollfd)) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to set CLOEXEC on link to datafd epoll fd: %s", strerror(savederrno)); goto exit_fail; } if (_fdset_cloexec(knet_h->dst_link_handler_epollfd)) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to set CLOEXEC on dst cache epoll fd: %s", strerror(savederrno)); goto exit_fail; } memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = knet_h->hostsockfd[0]; if (epoll_ctl(knet_h->send_to_links_epollfd, EPOLL_CTL_ADD, knet_h->hostsockfd[0], &ev)) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to add hostsockfd[0] to epoll pool: %s", strerror(savederrno)); goto exit_fail; } memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = knet_h->dstsockfd[0]; if (epoll_ctl(knet_h->dst_link_handler_epollfd, EPOLL_CTL_ADD, knet_h->dstsockfd[0], &ev)) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to add dstsockfd[0] to epoll pool: %s", strerror(savederrno)); goto exit_fail; } return 0; exit_fail: errno = savederrno; return -1; } static void _close_epolls(knet_handle_t knet_h) { struct epoll_event ev; int i; memset(&ev, 0, sizeof(struct epoll_event)); for (i = 0; i < KNET_DATAFD_MAX; i++) { if (knet_h->sockfd[i].in_use) { epoll_ctl(knet_h->send_to_links_epollfd, EPOLL_CTL_DEL, knet_h->sockfd[i].sockfd[knet_h->sockfd[i].is_created], &ev); if (knet_h->sockfd[i].sockfd[knet_h->sockfd[i].is_created]) { _close_socketpair(knet_h, knet_h->sockfd[i].sockfd); } } } epoll_ctl(knet_h->send_to_links_epollfd, EPOLL_CTL_DEL, knet_h->hostsockfd[0], &ev); epoll_ctl(knet_h->dst_link_handler_epollfd, EPOLL_CTL_DEL, knet_h->dstsockfd[0], &ev); close(knet_h->send_to_links_epollfd); close(knet_h->recv_from_links_epollfd); close(knet_h->dst_link_handler_epollfd); } static int _start_transports(knet_handle_t knet_h) { int i, savederrno = 0, err = 0; for (i=0; itransport_ops[i] = get_udp_transport(); break; case KNET_TRANSPORT_SCTP: knet_h->transport_ops[i] = get_sctp_transport(); break; } if (knet_h->transport_ops[i]) { if (knet_h->transport_ops[i]->transport_init(knet_h) < 0) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Failed to allocate transport handle for %s: %s", knet_h->transport_ops[i]->transport_name, strerror(savederrno)); err = -1; goto out; } } } out: errno = savederrno; return err; } static void _stop_transports(knet_handle_t knet_h) { int i; for (i=0; itransport_ops[i]) { knet_h->transport_ops[i]->transport_free(knet_h); } } } static int _start_threads(knet_handle_t knet_h) { int savederrno = 0; savederrno = pthread_create(&knet_h->pmtud_link_handler_thread, 0, _handle_pmtud_link_thread, (void *) knet_h); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to start pmtud link thread: %s", strerror(savederrno)); goto exit_fail; } savederrno = pthread_create(&knet_h->dst_link_handler_thread, 0, _handle_dst_link_handler_thread, (void *) knet_h); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to start dst cache thread: %s", strerror(savederrno)); goto exit_fail; } savederrno = pthread_create(&knet_h->send_to_links_thread, 0, _handle_send_to_links_thread, (void *) knet_h); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to start datafd to link thread: %s", strerror(savederrno)); goto exit_fail; } savederrno = pthread_create(&knet_h->recv_from_links_thread, 0, _handle_recv_from_links_thread, (void *) knet_h); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to start link to datafd thread: %s", strerror(savederrno)); goto exit_fail; } savederrno = pthread_create(&knet_h->heartbt_thread, 0, _handle_heartbt_thread, (void *) knet_h); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to start heartbeat thread: %s", strerror(savederrno)); goto exit_fail; } return 0; exit_fail: errno = savederrno; return -1; } static void _stop_threads(knet_handle_t knet_h) { void *retval; /* * allow threads to catch on shutdown request * and release locks before we stop them. * this isn't the most efficent way to handle it * but it works good enough for now */ sleep(1); pthread_mutex_lock(&knet_h->host_mutex); pthread_cond_signal(&knet_h->host_cond); pthread_mutex_unlock(&knet_h->host_mutex); if (knet_h->heartbt_thread) { pthread_cancel(knet_h->heartbt_thread); pthread_join(knet_h->heartbt_thread, &retval); } if (knet_h->send_to_links_thread) { pthread_cancel(knet_h->send_to_links_thread); pthread_join(knet_h->send_to_links_thread, &retval); } if (knet_h->recv_from_links_thread) { pthread_cancel(knet_h->recv_from_links_thread); pthread_join(knet_h->recv_from_links_thread, &retval); } if (knet_h->dst_link_handler_thread) { pthread_cancel(knet_h->dst_link_handler_thread); pthread_join(knet_h->dst_link_handler_thread, &retval); } pthread_mutex_lock(&knet_h->pmtud_mutex); pthread_cond_signal(&knet_h->pmtud_cond); pthread_mutex_unlock(&knet_h->pmtud_mutex); sleep(1); if (knet_h->pmtud_link_handler_thread) { pthread_cancel(knet_h->pmtud_link_handler_thread); pthread_join(knet_h->pmtud_link_handler_thread, &retval); } } knet_handle_t knet_handle_new(uint16_t host_id, int log_fd, uint8_t default_log_level) { knet_handle_t knet_h; int savederrno = 0; struct rlimit cur; if (getrlimit(RLIMIT_NOFILE, &cur) < 0) { return NULL; } if ((log_fd < 0) || (log_fd >= cur.rlim_max)) { errno = EINVAL; return NULL; } /* * validate incoming request */ if ((log_fd) && (default_log_level > KNET_LOG_DEBUG)) { errno = EINVAL; return NULL; } /* * allocate handle */ knet_h = malloc(sizeof(struct knet_handle)); if (!knet_h) { errno = ENOMEM; return NULL; } memset(knet_h, 0, sizeof(struct knet_handle)); savederrno = pthread_mutex_lock(&handle_config_mutex); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get handle mutex lock: %s", strerror(savederrno)); errno = savederrno; goto exit_fail; } /* * copy config in place */ knet_h->host_id = host_id; knet_h->logfd = log_fd; if (knet_h->logfd > 0) { memset(&knet_h->log_levels, default_log_level, KNET_MAX_SUBSYSTEMS); } /* * set pmtud default timers */ knet_h->pmtud_interval = KNET_PMTUD_DEFAULT_INTERVAL; /* * init main locking structures */ if (_init_locks(knet_h)) { savederrno = errno; goto exit_fail; } /* * init sockets */ if (_init_socks(knet_h)) { savederrno = errno; goto exit_fail; } /* * allocate packet buffers */ if (_init_buffers(knet_h)) { savederrno = errno; goto exit_fail; } /* * create epoll fds */ if (_init_epolls(knet_h)) { savederrno = errno; goto exit_fail; } /* * start transports */ if (_start_transports(knet_h)) { savederrno = errno; goto exit_fail; } /* * start internal threads */ if (_start_threads(knet_h)) { savederrno = errno; goto exit_fail; } pthread_mutex_unlock(&handle_config_mutex); return knet_h; exit_fail: pthread_mutex_unlock(&handle_config_mutex); knet_handle_free(knet_h); errno = savederrno; return NULL; } int knet_handle_free(knet_handle_t knet_h) { int savederrno = 0; savederrno = pthread_mutex_lock(&handle_config_mutex); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get handle mutex lock: %s", strerror(savederrno)); errno = savederrno; return -1; } if (!knet_h) { pthread_mutex_unlock(&handle_config_mutex); errno = EINVAL; return -1; } if (!knet_h->lock_init_done) { goto exit_nolock; } savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get write lock: %s", strerror(savederrno)); pthread_mutex_unlock(&handle_config_mutex); errno = savederrno; return -1; } if (knet_h->host_head != NULL) { savederrno = EBUSY; log_err(knet_h, KNET_SUB_HANDLE, "Unable to free handle: host(s) or listener(s) are still active: %s", strerror(savederrno)); pthread_rwlock_unlock(&knet_h->global_rwlock); pthread_mutex_unlock(&handle_config_mutex); errno = savederrno; return -1; } knet_h->fini_in_progress = 1; pthread_rwlock_unlock(&knet_h->global_rwlock); _stop_threads(knet_h); _stop_transports(knet_h); _close_epolls(knet_h); _destroy_buffers(knet_h); _close_socks(knet_h); crypto_fini(knet_h); _destroy_locks(knet_h); exit_nolock: free(knet_h); knet_h = NULL; pthread_mutex_unlock(&handle_config_mutex); return 0; } int knet_handle_enable_sock_notify(knet_handle_t knet_h, void *sock_notify_fn_private_data, void (*sock_notify_fn) ( void *private_data, int datafd, int8_t channel, uint8_t tx_rx, int error, int errorno)) { int savederrno = 0, err = 0; if (!knet_h) { errno = EINVAL; return -1; } if (!sock_notify_fn) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get write lock: %s", strerror(savederrno)); errno = savederrno; return -1; } knet_h->sock_notify_fn_private_data = sock_notify_fn_private_data; knet_h->sock_notify_fn = sock_notify_fn; log_debug(knet_h, KNET_SUB_HANDLE, "sock_notify_fn enabled"); pthread_rwlock_unlock(&knet_h->global_rwlock); return err; } int knet_handle_add_datafd(knet_handle_t knet_h, int *datafd, int8_t *channel) { int err = 0, savederrno = 0; int i; struct epoll_event ev; if (!knet_h) { errno = EINVAL; return -1; } if (datafd == NULL) { errno = EINVAL; return -1; } if (channel == NULL) { errno = EINVAL; return -1; } if (*channel >= KNET_DATAFD_MAX) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get write lock: %s", strerror(savederrno)); errno = savederrno; return -1; } if (!knet_h->sock_notify_fn) { log_err(knet_h, KNET_SUB_HANDLE, "Adding datafd requires sock notify callback enabled!"); savederrno = EINVAL; err = -1; goto out_unlock; } if (*datafd > 0) { for (i = 0; i < KNET_DATAFD_MAX; i++) { if ((knet_h->sockfd[i].in_use) && (knet_h->sockfd[i].sockfd[0] == *datafd)) { log_err(knet_h, KNET_SUB_HANDLE, "requested datafd: %d already exist in index: %d", *datafd, i); savederrno = EEXIST; err = -1; goto out_unlock; } } } /* * auto allocate a channel */ if (*channel < 0) { for (i = 0; i < KNET_DATAFD_MAX; i++) { if (!knet_h->sockfd[i].in_use) { *channel = i; break; } } if (*channel < 0) { savederrno = EBUSY; err = -1; goto out_unlock; } } else { if (knet_h->sockfd[*channel].in_use) { savederrno = EBUSY; err = -1; goto out_unlock; } } knet_h->sockfd[*channel].is_created = 0; knet_h->sockfd[*channel].is_socket = 0; knet_h->sockfd[*channel].has_error = 0; if (*datafd > 0) { int sockopt; socklen_t sockoptlen = sizeof(sockopt); if (_fdset_cloexec(*datafd)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_HANDLE, "Unable to set CLOEXEC on datafd: %s", strerror(savederrno)); goto out_unlock; } if (_fdset_nonblock(*datafd)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_HANDLE, "Unable to set NONBLOCK on datafd: %s", strerror(savederrno)); goto out_unlock; } knet_h->sockfd[*channel].sockfd[0] = *datafd; knet_h->sockfd[*channel].sockfd[1] = 0; if (!getsockopt(knet_h->sockfd[*channel].sockfd[0], SOL_SOCKET, SO_TYPE, &sockopt, &sockoptlen)) { knet_h->sockfd[*channel].is_socket = 1; } } else { if (_init_socketpair(knet_h, knet_h->sockfd[*channel].sockfd)) { savederrno = errno; err = -1; goto out_unlock; } knet_h->sockfd[*channel].is_created = 1; knet_h->sockfd[*channel].is_socket = 1; *datafd = knet_h->sockfd[*channel].sockfd[0]; } memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = knet_h->sockfd[*channel].sockfd[knet_h->sockfd[*channel].is_created]; if (epoll_ctl(knet_h->send_to_links_epollfd, EPOLL_CTL_ADD, knet_h->sockfd[*channel].sockfd[knet_h->sockfd[*channel].is_created], &ev)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_HANDLE, "Unable to add datafd %d to linkfd epoll pool: %s", knet_h->sockfd[*channel].sockfd[knet_h->sockfd[*channel].is_created], strerror(savederrno)); if (knet_h->sockfd[*channel].is_created) { _close_socketpair(knet_h, knet_h->sockfd[*channel].sockfd); } goto out_unlock; } knet_h->sockfd[*channel].in_use = 1; out_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = savederrno; return err; } int knet_handle_remove_datafd(knet_handle_t knet_h, int datafd) { int err = 0, savederrno = 0; int8_t channel = -1; int i; struct epoll_event ev; if (!knet_h) { errno = EINVAL; return -1; } if (datafd <= 0) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get write lock: %s", strerror(savederrno)); errno = savederrno; return -1; } for (i = 0; i < KNET_DATAFD_MAX; i++) { if ((knet_h->sockfd[i].in_use) && (knet_h->sockfd[i].sockfd[0] == datafd)) { channel = i; break; } } if (channel < 0) { savederrno = EINVAL; err = -1; goto out_unlock; } if (!knet_h->sockfd[channel].has_error) { memset(&ev, 0, sizeof(struct epoll_event)); if (epoll_ctl(knet_h->send_to_links_epollfd, EPOLL_CTL_DEL, knet_h->sockfd[channel].sockfd[knet_h->sockfd[channel].is_created], &ev)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_HANDLE, "Unable to del datafd %d from linkfd epoll pool: %s", knet_h->sockfd[channel].sockfd[0], strerror(savederrno)); goto out_unlock; } } if (knet_h->sockfd[channel].is_created) { _close_socketpair(knet_h, knet_h->sockfd[channel].sockfd); } memset(&knet_h->sockfd[channel], 0, sizeof(struct knet_sock)); out_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = savederrno; return err; } int knet_handle_get_datafd(knet_handle_t knet_h, const int8_t channel, int *datafd) { int err = 0, savederrno = 0; if (!knet_h) { errno = EINVAL; return -1; } if ((channel < 0) || (channel >= KNET_DATAFD_MAX)) { errno = EINVAL; return -1; } if (datafd == NULL) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get read lock: %s", strerror(savederrno)); errno = savederrno; return -1; } if (!knet_h->sockfd[channel].in_use) { savederrno = EINVAL; err = -1; goto out_unlock; } *datafd = knet_h->sockfd[channel].sockfd[0]; out_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = savederrno; return err; } int knet_handle_get_channel(knet_handle_t knet_h, const int datafd, int8_t *channel) { int err = 0, savederrno = 0; int i; if (!knet_h) { errno = EINVAL; return -1; } if (datafd <= 0) { errno = EINVAL; return -1; } if (channel == NULL) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get read lock: %s", strerror(savederrno)); errno = savederrno; return -1; } *channel = -1; for (i = 0; i < KNET_DATAFD_MAX; i++) { if ((knet_h->sockfd[i].in_use) && (knet_h->sockfd[i].sockfd[0] == datafd)) { *channel = i; break; } } if (*channel < 0) { savederrno = EINVAL; err = -1; goto out_unlock; } out_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = savederrno; return err; } int knet_handle_enable_filter(knet_handle_t knet_h, void *dst_host_filter_fn_private_data, int (*dst_host_filter_fn) ( void *private_data, const unsigned char *outdata, ssize_t outdata_len, uint8_t tx_rx, uint16_t this_host_id, uint16_t src_node_id, int8_t *channel, uint16_t *dst_host_ids, size_t *dst_host_ids_entries)) { int savederrno = 0; if (!knet_h) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get write lock: %s", strerror(savederrno)); errno = savederrno; return -1; } knet_h->dst_host_filter_fn_private_data = dst_host_filter_fn_private_data; knet_h->dst_host_filter_fn = dst_host_filter_fn; if (knet_h->dst_host_filter_fn) { log_debug(knet_h, KNET_SUB_HANDLE, "dst_host_filter_fn enabled"); } else { log_debug(knet_h, KNET_SUB_HANDLE, "dst_host_filter_fn disabled"); } pthread_rwlock_unlock(&knet_h->global_rwlock); return 0; } int knet_handle_setfwd(knet_handle_t knet_h, unsigned int enabled) { int savederrno = 0; if (!knet_h) { errno = EINVAL; return -1; } if ((enabled < 0) || (enabled > 1)) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get write lock: %s", strerror(savederrno)); errno = savederrno; return -1; } knet_h->enabled = enabled; if (enabled) { log_debug(knet_h, KNET_SUB_HANDLE, "Data forwarding is enabled"); } else { log_debug(knet_h, KNET_SUB_HANDLE, "Data forwarding is disabled"); } pthread_rwlock_unlock(&knet_h->global_rwlock); return 0; } int knet_handle_pmtud_getfreq(knet_handle_t knet_h, unsigned int *interval) { int savederrno = 0; if (!knet_h) { errno = EINVAL; return -1; } if (!interval) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get write lock: %s", strerror(savederrno)); errno = savederrno; return -1; } *interval = knet_h->pmtud_interval; pthread_rwlock_unlock(&knet_h->global_rwlock); return 0; } int knet_handle_pmtud_setfreq(knet_handle_t knet_h, unsigned int interval) { int savederrno = 0; if (!knet_h) { errno = EINVAL; return -1; } if ((!interval) || (interval > 86400)) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get write lock: %s", strerror(savederrno)); errno = savederrno; return -1; } knet_h->pmtud_interval = interval; log_debug(knet_h, KNET_SUB_HANDLE, "PMTUd interval set to: %u seconds", interval); pthread_rwlock_unlock(&knet_h->global_rwlock); return 0; } int knet_handle_enable_pmtud_notify(knet_handle_t knet_h, void *pmtud_notify_fn_private_data, void (*pmtud_notify_fn) ( void *private_data, unsigned int data_mtu)) { int savederrno = 0; if (!knet_h) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get write lock: %s", strerror(savederrno)); errno = savederrno; return -1; } knet_h->pmtud_notify_fn_private_data = pmtud_notify_fn_private_data; knet_h->pmtud_notify_fn = pmtud_notify_fn; if (knet_h->pmtud_notify_fn) { log_debug(knet_h, KNET_SUB_HANDLE, "pmtud_notify_fn enabled"); } else { log_debug(knet_h, KNET_SUB_HANDLE, "pmtud_notify_fn disabled"); } pthread_rwlock_unlock(&knet_h->global_rwlock); return 0; } int knet_handle_pmtud_get(knet_handle_t knet_h, unsigned int *data_mtu) { int savederrno = 0; if (!knet_h) { errno = EINVAL; return -1; } if (!data_mtu) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get write lock: %s", strerror(savederrno)); errno = savederrno; return -1; } *data_mtu = knet_h->data_mtu; pthread_rwlock_unlock(&knet_h->global_rwlock); return 0; } int knet_handle_crypto(knet_handle_t knet_h, struct knet_handle_crypto_cfg *knet_handle_crypto_cfg) { int savederrno = 0; int err = 0; if (!knet_h) { errno = EINVAL; return -1; } if (!knet_handle_crypto_cfg) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get write lock: %s", strerror(savederrno)); errno = savederrno; return -1; } crypto_fini(knet_h); if ((!strncmp("none", knet_handle_crypto_cfg->crypto_model, 4)) || ((!strncmp("none", knet_handle_crypto_cfg->crypto_cipher_type, 4)) && (!strncmp("none", knet_handle_crypto_cfg->crypto_hash_type, 4)))) { log_debug(knet_h, KNET_SUB_CRYPTO, "crypto is not enabled"); err = 0; goto exit_unlock; } if (knet_handle_crypto_cfg->private_key_len < KNET_MIN_KEY_LEN) { log_debug(knet_h, KNET_SUB_CRYPTO, "private key len too short (min %u): %u", KNET_MIN_KEY_LEN, knet_handle_crypto_cfg->private_key_len); savederrno = EINVAL; err = -1; goto exit_unlock; } if (knet_handle_crypto_cfg->private_key_len > KNET_MAX_KEY_LEN) { log_debug(knet_h, KNET_SUB_CRYPTO, "private key len too long (max %u): %u", KNET_MAX_KEY_LEN, knet_handle_crypto_cfg->private_key_len); savederrno = EINVAL; err = -1; goto exit_unlock; } err = crypto_init(knet_h, knet_handle_crypto_cfg); if (err) { err = -2; } exit_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = savederrno; return err; } ssize_t knet_recv(knet_handle_t knet_h, char *buff, const size_t buff_len, const int8_t channel) { int savederrno = 0; ssize_t err = 0; struct iovec iov_in; if (!knet_h) { errno = EINVAL; return -1; } if (buff == NULL) { errno = EINVAL; return -1; } if (buff_len <= 0) { errno = EINVAL; return -1; } if (buff_len > KNET_MAX_PACKET_SIZE) { errno = EINVAL; return -1; } if (channel < 0) { errno = EINVAL; return -1; } if (channel >= KNET_DATAFD_MAX) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get read lock: %s", strerror(savederrno)); errno = savederrno; return -1; } if (!knet_h->sockfd[channel].in_use) { savederrno = EINVAL; err = -1; goto out_unlock; } memset(&iov_in, 0, sizeof(iov_in)); iov_in.iov_base = (void *)buff; iov_in.iov_len = buff_len; err = readv(knet_h->sockfd[channel].sockfd[0], &iov_in, 1); savederrno = errno; out_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = savederrno; return err; } ssize_t knet_send(knet_handle_t knet_h, const char *buff, const size_t buff_len, const int8_t channel) { int savederrno = 0; ssize_t err = 0; struct iovec iov_out[1]; if (!knet_h) { errno = EINVAL; return -1; } if (buff == NULL) { errno = EINVAL; return -1; } if (buff_len <= 0) { errno = EINVAL; return -1; } if (buff_len > KNET_MAX_PACKET_SIZE) { errno = EINVAL; return -1; } if (channel < 0) { errno = EINVAL; return -1; } if (channel >= KNET_DATAFD_MAX) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get read lock: %s", strerror(savederrno)); errno = savederrno; return -1; } if (!knet_h->sockfd[channel].in_use) { savederrno = EINVAL; err = -1; goto out_unlock; } memset(iov_out, 0, sizeof(iov_out)); iov_out[0].iov_base = (void *)buff; iov_out[0].iov_len = buff_len; err = writev(knet_h->sockfd[channel].sockfd[0], iov_out, 1); savederrno = errno; out_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = savederrno; return err; } diff --git a/libknet/internals.h b/libknet/internals.h index 2d114704..f19d577b 100644 --- a/libknet/internals.h +++ b/libknet/internals.h @@ -1,446 +1,449 @@ /* * Copyright (C) 2010-2015 Red Hat, Inc. All rights reserved. * * Authors: Fabio M. Di Nitto * Federico Simoncelli * * This software licensed under GPL-2.0+, LGPL-2.0+ */ #ifndef __INTERNALS_H__ #define __INTERNALS_H__ /* * NOTE: you shouldn't need to include this header normally */ #include "libknet.h" #include "onwire.h" +#include "compat.h" #define KNET_DATABUFSIZE KNET_MAX_PACKET_SIZE + KNET_HEADER_ALL_SIZE #define KNET_DATABUFSIZE_CRYPT_PAD 1024 #define KNET_DATABUFSIZE_CRYPT KNET_DATABUFSIZE + KNET_DATABUFSIZE_CRYPT_PAD #define KNET_RING_RCVBUFF 8388608 #define PCKT_FRAG_MAX UINT8_MAX #define KNET_EPOLL_MAX_EVENTS KNET_DATAFD_MAX typedef void *knet_transport_link_t; /* per link transport handle */ typedef void *knet_transport_t; /* per knet_h transport handle */ struct knet_transport_ops; /* Forward because of circular dependancy */ struct knet_link { /* required */ struct sockaddr_storage src_addr; struct sockaddr_storage dst_addr; /* configurable */ unsigned int dynamic; /* see KNET_LINK_DYN_ define above */ uint8_t priority; /* higher priority == preferred for A/P */ unsigned long long ping_interval; /* interval */ unsigned long long pong_timeout; /* timeout */ unsigned int latency_fix; /* precision */ uint8_t pong_count; /* how many ping/pong to send/receive before link is up */ /* status */ struct knet_link_status status; /* internals */ uint8_t link_id; uint8_t transport_type; /* #defined constant from API */ knet_transport_link_t transport_link; /* link_info_t from transport */ int outsock; unsigned int configured:1; /* set to 1 if src/dst have been configured transport initialized on this link*/ unsigned int transport_connected:1; /* set to 1 if lower level transport is connected */ unsigned int remoteconnected:1; /* link is enabled for data (peer view) */ unsigned int donnotremoteupdate:1; /* define source of the update */ unsigned int host_info_up_sent:1; /* 0 if we need to notify remote that link is up */ unsigned int latency_exp; uint8_t received_pong; struct timespec ping_last; /* used by PMTUD thread as temp per-link variables and should always contain the onwire_len value! */ uint32_t proto_overhead; struct timespec pmtud_last; uint32_t last_ping_size; uint32_t last_good_mtu; uint32_t last_bad_mtu; uint32_t last_sent_mtu; uint32_t last_recv_mtu; uint8_t has_valid_mtu; }; #define KNET_CBUFFER_SIZE 4096 struct knet_host_defrag_buf { char buf[KNET_MAX_PACKET_SIZE]; uint8_t in_use; /* 0 buffer is free, 1 is in use */ seq_num_t pckt_seq; /* identify the pckt we are receiving */ uint8_t frag_recv; /* how many frags did we receive */ uint8_t frag_map[PCKT_FRAG_MAX];/* bitmap of what we received? */ uint8_t last_first; /* special case if we receive the last fragment first */ uint16_t frag_size; /* normal frag size (not the last one) */ uint16_t last_frag_size; /* the last fragment might not be aligned with MTU size */ struct timespec last_update; /* keep time of the last pckt */ }; struct knet_host { /* required */ uint16_t host_id; /* configurable */ uint8_t link_handler_policy; char name[KNET_MAX_HOST_LEN]; /* status */ struct knet_host_status status; /* internals */ char bcast_circular_buffer[KNET_CBUFFER_SIZE]; seq_num_t bcast_seq_num_rx; char ucast_circular_buffer[KNET_CBUFFER_SIZE]; seq_num_t ucast_seq_num_tx; seq_num_t ucast_seq_num_rx; /* defrag/(reassembly buffers */ struct knet_host_defrag_buf defrag_buf[KNET_MAX_LINK]; char bcast_circular_buffer_defrag[KNET_CBUFFER_SIZE]; char ucast_circular_buffer_defrag[KNET_CBUFFER_SIZE]; /* link stuff */ struct knet_link link[KNET_MAX_LINK]; uint8_t active_link_entries; uint8_t active_links[KNET_MAX_LINK]; struct knet_host *next; }; struct knet_sock { int sockfd[2]; /* sockfd[0] will always be application facing * and sockfd[1] internal if sockpair has been created by knet */ int is_socket; /* check if it's a socket for recvmmsg usage */ int is_created; /* knet created this socket and has to clean up on exit/del */ int in_use; /* set to 1 if it's use, 0 if free */ int has_error; /* set to 1 if there were errors reading from the sock * and socket has been removed from epoll */ }; struct knet_fd_trackers { uint8_t transport; /* transport type (UDP/SCTP...) */ uint8_t data_type; /* internal use for transport to define what data are associated * to this fd */ void *data; /* pointer to the data */ }; #define KNET_MAX_FDS KNET_MAX_HOST * KNET_MAX_LINK * 4 struct knet_handle { uint16_t host_id; unsigned int enabled:1; struct knet_sock sockfd[KNET_DATAFD_MAX]; int logfd; uint8_t log_levels[KNET_MAX_SUBSYSTEMS]; int hostsockfd[2]; int dstsockfd[2]; int send_to_links_epollfd; int recv_from_links_epollfd; int dst_link_handler_epollfd; unsigned int pmtud_interval; unsigned int data_mtu; /* contains the max data size that we can send onwire * without frags */ struct knet_host *host_head; struct knet_host *host_index[KNET_MAX_HOST]; knet_transport_t transports[KNET_MAX_TRANSPORTS+1]; struct knet_transport_ops *transport_ops[KNET_MAX_TRANSPORTS+1]; struct knet_fd_trackers knet_transport_fd_tracker[KNET_MAX_FDS]; /* track status for each fd handled by transports */ uint16_t host_ids[KNET_MAX_HOST]; size_t host_ids_entries; struct knet_header *recv_from_sock_buf[PCKT_FRAG_MAX]; struct knet_header *send_to_links_buf[PCKT_FRAG_MAX]; struct knet_header *recv_from_links_buf[PCKT_FRAG_MAX]; struct knet_header *pingbuf; struct knet_header *pmtudbuf; pthread_t send_to_links_thread; pthread_t recv_from_links_thread; pthread_t heartbt_thread; pthread_t dst_link_handler_thread; pthread_t pmtud_link_handler_thread; int lock_init_done; pthread_rwlock_t global_rwlock; /* global config lock */ pthread_rwlock_t host_rwlock; /* send_host_info lock, can switch to mutex? */ pthread_mutex_t host_mutex; /* host mutex for cond wait on pckt send, switch to mutex/sync_send ? */ pthread_cond_t host_cond; /* conditional for above */ pthread_mutex_t pmtud_mutex; /* pmtud mutex to handle conditional send/recv + timeout */ pthread_cond_t pmtud_cond; /* conditional for above */ pthread_mutex_t tx_mutex; struct crypto_instance *crypto_instance; uint16_t sec_header_size; uint16_t sec_block_size; uint16_t sec_hash_size; uint16_t sec_salt_size; unsigned char *send_to_links_buf_crypt[PCKT_FRAG_MAX]; unsigned char *recv_from_links_buf_crypt; unsigned char *recv_from_links_buf_decrypt; unsigned char *pingbuf_crypt; unsigned char *pmtudbuf_crypt; seq_num_t bcast_seq_num_tx; void *dst_host_filter_fn_private_data; int (*dst_host_filter_fn) ( void *private_data, const unsigned char *outdata, ssize_t outdata_len, uint8_t tx_rx, uint16_t this_host_id, uint16_t src_node_id, int8_t *channel, uint16_t *dst_host_ids, size_t *dst_host_ids_entries); void *pmtud_notify_fn_private_data; void (*pmtud_notify_fn) ( void *private_data, unsigned int data_mtu); void *host_status_change_notify_fn_private_data; void (*host_status_change_notify_fn) ( void *private_data, uint16_t host_id, uint8_t reachable, uint8_t remote, uint8_t external); void *sock_notify_fn_private_data; void (*sock_notify_fn) ( void *private_data, int datafd, int8_t channel, uint8_t tx_rx, int error, int errorno); int fini_in_progress; }; /* * NOTE: every single operation must be implementend * for every protocol. */ typedef struct knet_transport_ops { /* * transport generic information */ const char *transport_name; uint32_t transport_mtu_overhead; /* * transport init must allocate the new transport * and perform all internal initializations * (threads, lists, etc). */ int (*transport_init)(knet_handle_t knet_h); /* * transport free must releases _all_ resources * allocated by tranport_init */ int (*transport_free)(knet_handle_t knet_h); /* * link operations should take care of all the * sockets and epoll management for a given link/transport set * transport_link_disable should return err = -1 and errno = EBUSY * if listener is still in use, and any other errno in case * the link cannot be disabled. * * set_config/clear_config are invoked in global write lock context */ int (*transport_link_set_config)(knet_handle_t knet_h, struct knet_link *link); int (*transport_link_clear_config)(knet_handle_t knet_h, struct knet_link *link); /* * per transport error handling of recvmmsg * (see _handle_recv_from_links comments for details) */ /* * transport_rx_sock_error is invoked when recvmmsg returns <= 0 * * transport_rx_sock_error is invoked with both global_rwlock * and fd_tracker read lock (from RX thread) */ int (*transport_rx_sock_error)(knet_handle_t knet_h, int sockfd, int recv_err, int recv_errno); /* * this function is called on _every_ received packet * to verify if the packet is data or internal protocol error handling * * it should return: * -1 on error * 0 packet is not data and we should continue the packet process loop * 1 packet is not data and we should STOP the packet process loop * 2 packet is data and should be parsed as such * * transport_rx_is_data is invoked with both global_rwlock * and fd_tracker read lock (from RX thread) */ int (*transport_rx_is_data)(knet_handle_t knet_h, int sockfd, struct mmsghdr msg); } knet_transport_ops_t; +socklen_t sockaddr_len(const struct sockaddr_storage *ss); + /** * This is a kernel style list implementation. * * @author Steven Dake */ struct knet_list_head { struct knet_list_head *next; struct knet_list_head *prev; }; /** * @def KNET_LIST_DECLARE() * Declare and initialize a list head. */ #define KNET_LIST_DECLARE(name) \ struct knet_list_head name = { &(name), &(name) } #define KNET_INIT_LIST_HEAD(ptr) do { \ (ptr)->next = (ptr); (ptr)->prev = (ptr); \ } while (0) /** * Initialize the list entry. * * Points next and prev pointers to head. * @param head pointer to the list head */ static inline void knet_list_init(struct knet_list_head *head) { head->next = head; head->prev = head; } /** * Add this element to the list. * * @param element the new element to insert. * @param head pointer to the list head */ static inline void knet_list_add(struct knet_list_head *element, struct knet_list_head *head) { head->next->prev = element; element->next = head->next; element->prev = head; head->next = element; } /** * Add to the list (but at the end of the list). * * @param element pointer to the element to add * @param head pointer to the list head * @see knet_list_add() */ static inline void knet_list_add_tail(struct knet_list_head *element, struct knet_list_head *head) { head->prev->next = element; element->next = head; element->prev = head->prev; head->prev = element; } /** * Delete an entry from the list. * * @param _remove the list item to remove */ static inline void knet_list_del(struct knet_list_head *_remove) { _remove->next->prev = _remove->prev; _remove->prev->next = _remove->next; } /** * Replace old entry by new one * @param old: the element to be replaced * @param new: the new element to insert */ static inline void knet_list_replace(struct knet_list_head *old, struct knet_list_head *new) { new->next = old->next; new->next->prev = new; new->prev = old->prev; new->prev->next = new; } /** * Tests whether list is the last entry in list head * @param list: the entry to test * @param head: the head of the list * @return boolean true/false */ static inline int knet_list_is_last(const struct knet_list_head *list, const struct knet_list_head *head) { return list->next == head; } /** * A quick test to see if the list is empty (pointing to it's self). * @param head pointer to the list head * @return boolean true/false */ static inline int32_t knet_list_empty(const struct knet_list_head *head) { return head->next == head; } /** * Get the struct for this entry * @param ptr: the &struct list_head pointer. * @param type: the type of the struct this is embedded in. * @param member: the name of the list_struct within the struct. */ #define knet_list_entry(ptr,type,member)\ ((type *)((char *)(ptr)-(char*)(&((type *)0)->member))) /** * Get the first element from a list * @param ptr: the &struct list_head pointer. * @param type: the type of the struct this is embedded in. * @param member: the name of the list_struct within the struct. */ #define knet_list_first_entry(ptr, type, member) \ knet_list_entry((ptr)->next, type, member) /** * Iterate over a list * @param pos: the &struct list_head to use as a loop counter. * @param head: the head for your list. */ #define knet_list_for_each(pos, head) \ for (pos = (head)->next; pos != (head); pos = pos->next) /** * Iterate over a list backwards * @param pos: the &struct list_head to use as a loop counter. * @param head: the head for your list. */ #define knet_list_for_each_reverse(pos, head) \ for (pos = (head)->prev; pos != (head); pos = pos->prev) /** * Iterate over a list safe against removal of list entry * @param pos: the &struct list_head to use as a loop counter. * @param n: another &struct list_head to use as temporary storage * @param head: the head for your list. */ #define knet_list_for_each_safe(pos, n, head) \ for (pos = (head)->next, n = pos->next; pos != (head); \ pos = n, n = pos->next) /** * Iterate over list of given type * @param pos: the type * to use as a loop counter. * @param head: the head for your list. * @param member: the name of the list_struct within the struct. */ #define knet_list_for_each_entry(pos, head, member) \ for (pos = knet_list_entry((head)->next, typeof(*pos), member); \ &pos->member != (head); \ pos = knet_list_entry(pos->member.next, typeof(*pos), member)) #endif diff --git a/libknet/libknet.h b/libknet/libknet.h index e96bbbcc..0778fe98 100644 --- a/libknet/libknet.h +++ b/libknet/libknet.h @@ -1,1406 +1,1410 @@ /* * Copyright (C) 2010-2015 Red Hat, Inc. All rights reserved. * * Authors: Fabio M. Di Nitto * Federico Simoncelli * * This software licensed under GPL-2.0+, LGPL-2.0+ */ #ifndef __LIBKNET_H__ #define __LIBKNET_H__ +#include +#include +#include + #include #include /* * libknet limits */ /* * Maximum number of hosts */ #define KNET_MAX_HOST 65536 /* * Maximum number of links between 2 hosts */ #define KNET_MAX_LINK 8 /* * Maximum packet size that should be written to datafd * see knet_handle_new for details */ #define KNET_MAX_PACKET_SIZE 65536 /* * Buffers used for pretty logging * host is used to store both ip addresses and hostnames */ #define KNET_MAX_HOST_LEN 256 #define KNET_MAX_PORT_LEN 6 /* * Some notifications can be generated either on TX or RX */ #define KNET_NOTIFY_TX 0 #define KNET_NOTIFY_RX 1 typedef struct knet_handle *knet_handle_t; /* * Handle structs/API calls */ /* * knet_handle_new * * host_id - Each host in a knet is identified with a unique * ID. when creating a new handle local host_id * must be specified (0 to UINT16T_MAX are all valid). * It is the user's responsibility to check that the value * is unique, or bad things might happen. * * log_fd - Write file descriptor. If set to a value > 0, it will be used * to write log packets (see below) from libknet to the application. * Setting to 0 will disable logging from libknet. * It is possible to enable logging at any given time (see logging API * below). * Make sure to either read from this filedescriptor properly and/or * mark it O_NONBLOCK, otherwise if the fd becomes full, libknet could * block. * * default_log_level - * If logfd is specified, it will initialize all subsystems to log * at default_log_level value. (see logging API below) * * on success, a new knet_handle_t is returned. * on failure, NULL is returned and errno is set. */ knet_handle_t knet_handle_new(uint16_t host_id, int log_fd, uint8_t default_log_level); /* * knet_handle_free * * knet_h - pointer to knet_handle_t * * Destroy a knet handle, free all resources * * knet_handle_free returns: * * 0 on success * -1 on error and errno is set. */ int knet_handle_free(knet_handle_t knet_h); /* * knet_handle_enable_sock_notify * * knet_h - pointer to knet_handle_t * * sock_notify_fn_private_data * void pointer to data that can be used to identify * the callback. * * sock_notify_fn * A callback function that is invoked every time * a socket in the datafd pool will report an error (-1) * or an end of read (0) (see socket.7). * This function MUST NEVER block or add substantial delays. * The callback is invoked in an internal unlocked area * to allow calls to knet_handle_add_datafd/knet_handle_remove_datafd * to swap/replace the bad fd. * if both err and errno are 0, it means that the socket * has received a 0 byte packet (EOF?). * The callback function must either remove the fd from knet * (by calling knet_handle_remove_fd()) or dup a new fd in its place. * Failure to do this can cause problems. * * knet_handle_enable_sock_notify returns: * * 0 on success * -1 on error and errno is set. */ int knet_handle_enable_sock_notify(knet_handle_t knet_h, void *sock_notify_fn_private_data, void (*sock_notify_fn) ( void *private_data, int datafd, int8_t channel, uint8_t tx_rx, int error, int errorno)); /* sorry! can't call it errno ;) */ /* * knet_handle_add_datafd * * IMPORTANT: In order to add datafd to knet, knet_handle_enable_sock_notify * _MUST_ be set and be able to handle both errors (-1) and * 0 bytes read / write from the provided datafd. * On read error (< 0) from datafd, the socket is automatically * removed from polling to avoid spinning on dead sockets. * It is safe to call knet_handle_remove_datafd even on sockets * that have been removed. * * knet_h - pointer to knet_handle_t * * *datafd - read/write file descriptor. * knet will read data here to send to the other hosts * and will write data received from the network. * Each data packet can be of max size KNET_MAX_PACKET_SIZE! * Applications using knet_send/knet_recv will receive a * proper error if the packet size is not within boundaries. * Applications using their own functions to write to the * datafd should NOT write more than KNET_MAX_PACKET_SIZE. * * Please refer to handle.c on how to set up a socketpair. * * datafd can be 0, and knet_handle_add_datafd will create a properly * populated socket pair the same way as ping_test, or a value * higher than 0. A negative number will return an error. * On exit knet_handle_free will take care to cleanup the * socketpair only if they have been created by knet_handle_add_datafd. * * It is possible to pass either sockets or normal fds. * User provided datafd will be marked as non-blocking and close-on-exit. * * *channel - This value has the same effect of VLAN tagging. * A negative value will auto-allocate a channel. * Setting a value between 0 and 31 will try to allocate that * specific channel (unless already in use). * * It is possible to add up to 32 datafds but be aware that each * one of them must have a receiving end on the other host. * * Example: * hostA channel 0 will be delivered to datafd on hostB channel 0 * hostA channel 1 to hostB channel 1. * * Each channel must have a unique file descriptor. * * If your application could have 2 channels on one host and one * channel on another host, then you can use dst_host_filter * to manipulate channel values on TX and RX. * * knet_handle_add_datafd returns: * * 0 on success * *datafd will be populated with a socket if the original value was 0 * or if a specific fd was set, the value is untouched. * *channel will be populated with a channel number if the original value * was negative or the value is untouched if a specific channel * was requested. * * -1 on error and errno is set. * *datafd and *channel are untouched or empty. */ #define KNET_DATAFD_MAX 32 int knet_handle_add_datafd(knet_handle_t knet_h, int *datafd, int8_t *channel); /* * knet_handle_remove_datafd * * knet_h - pointer to knet_handle_t * * datafd - file descriptor to remove. * NOTE that if the socket/fd was created by knet_handle_add_datafd, * the socket will be closed by libknet. * * knet_handle_remove_datafd returns: * * 0 on success * * -1 on error and errno is set. */ int knet_handle_remove_datafd(knet_handle_t knet_h, int datafd); /* * knet_handle_enable_sock_notify * * knet_h - pointer to knet_handle_t * * sock_notify_fn_private_data * void pointer to data that can be used to identify * the callback. * * sock_notify_fn * A callback function that is invoked every time * a socket in the datafd pool will report an error (-1) * or an end of read (0) (see socket.7). * This function MUST NEVER block or add substantial delays. * The callback is invoked in an internal unlocked area * to allow calls to knet_handle_add_datafd/knet_handle_remove_datafd * to swap/replace the bad fd. * if both err and errno are 0, it means that the socket * has received a 0 byte packet (EOF?). * The callback function must either remove the fd from knet * (by calling knet_handle_remove_fd()) or dup a new fd in its place. * Failure to do this can cause problems. * * knet_handle_enable_sock_notify returns: * * 0 on success * -1 on error and errno is set. */ int knet_handle_get_channel(knet_handle_t knet_h, const int datafd, int8_t *channel); /* * knet_handle_get_datafd * * knet_h - pointer to knet_handle_t * * channel - get the datafd associated to this channel * * *datafd - will contain the result * * knet_handle_get_datafd returns: * * 0 on success * and *datafd will contain the results * * -1 on error and errno is set. * and *datafd content is meaningless */ int knet_handle_get_datafd(knet_handle_t knet_h, const int8_t channel, int *datafd); /* * knet_recv * * knet_h - pointer to knet_handle_t * * buff - pointer to buffer to store the received data * * buff_len - buffer lenght * * knet_recv is a commodity function to wrap iovec operations * around a socket. It returns a call to readv(2). */ ssize_t knet_recv(knet_handle_t knet_h, char *buff, const size_t buff_len, const int8_t channel); /* * knet_send * * knet_h - pointer to knet_handle_t * * buff - pointer to the buffer of data to send * * buff_len - length of data to send * * knet_send is a commodity function to wrap iovec operations * around a socket. It returns a call to writev(2). */ ssize_t knet_send(knet_handle_t knet_h, const char *buff, const size_t buff_len, const int8_t channel); /* * knet_send_sync * * knet_h - pointer to knet_handle_t * * buff - pointer to the buffer of data to send * * buff_len - length of data to send * * channel - data channel to use (see knet_handle_add_datafd) * * All knet RX/TX operations are async for performance reasons. * There are applications that might need a sync version of data * transmission and receive errors in case of failure to deliver * to another host. * knet_send_sync bypasses the whole TX async layer and delivers * data directly to the link layer, and returns errors accordingly. * knet_send_sync allows to send only one packet to one host at * a time. It does NOT support multiple destinations or multicast * packets. Decision is still based on dst_host_filter_fn. * * knet_send_sync returns 0 on success and -1 on error. * * In addition to normal sendmmsg errors, knet_send_sync can fail * due to: * * ECANCELED - data forward is disabled * EFAULT - dst_host_filter fatal error * EINVAL - dst_host_filter did not provide * dst_host_ids_entries on unicast pckts * E2BIG - dst_host_filter did return more than one * dst_host_ids_entries on unicast pckts * ENOMSG - received unknown message type * EHOSTDOWN - unicast pckt cannot be delivered because * dest host is not connected yet * ECHILD - crypto failed * EAGAIN - sendmmsg was unable to send all messages and * there was no progress during retry */ int knet_send_sync(knet_handle_t knet_h, const char *buff, const size_t buff_len, const int8_t channel); /* * knet_handle_enable_filter * * knet_h - pointer to knet_handle_t * * dst_host_filter_fn_private_data * void pointer to data that can be used to identify * the callback. * * dst_host_filter_fn - * is a callback function that is invoked every time * a packet hits datafd (see knet_handle_new). * the function allows users to tell libknet where the * packet has to be delivered. * * const unsigned char *outdata - is a pointer to the * current packet * ssize_t outdata_len - lenght of the above data * uint8_t tx_rx - filter is called on tx or rx * (see defines below) * uint16_t this_host_id - host_id processing the packet * uint16_t src_host_id - host_id that generated the * packet * uint16_t *dst_host_ids - array of KNET_MAX_HOST uint16_t * where to store the destinations * size_t *dst_host_ids_entries - number of hosts to send the message * * dst_host_filter_fn should return * -1 on error, packet is discarded. * 0 packet is unicast and should be sent to dst_host_ids and there are * dst_host_ids_entries in the buffer. * 1 packet is broadcast/multicast and is sent all hosts. * contents of dst_host_ids and dst_host_ids_entries are ignored. * (see also kronosnetd/etherfilter.* for an example that filters based * on ether protocol) * * knet_handle_enable_filter returns: * * 0 on success * -1 on error and errno is set. */ int knet_handle_enable_filter(knet_handle_t knet_h, void *dst_host_filter_fn_private_data, int (*dst_host_filter_fn) ( void *private_data, const unsigned char *outdata, ssize_t outdata_len, uint8_t tx_rx, uint16_t this_host_id, uint16_t src_host_id, int8_t *channel, uint16_t *dst_host_ids, size_t *dst_host_ids_entries)); /* * knet_handle_setfwd * * knet_h - pointer to knet_handle_t * * enable - set to 1 to allow data forwarding, 0 to disable data forwarding. * * knet_handle_setfwd returns: * * 0 on success * -1 on error and errno is set. * * By default data forwarding is off and no traffic will pass through knet until * it is set on. */ int knet_handle_setfwd(knet_handle_t knet_h, unsigned int enabled); /* * knet_handle_pmtud_setfreq * * knet_h - pointer to knet_handle_t * * interval - define the interval in seconds between PMTUd scans * range from 1 to 86400 (24h) * * knet_handle_pmtud_setfreq returns: * * 0 on success * -1 on error and errno is set. * * default interval is 60. */ #define KNET_PMTUD_DEFAULT_INTERVAL 60 int knet_handle_pmtud_setfreq(knet_handle_t knet_h, unsigned int interval); /* * knet_handle_pmtud_getfreq * * knet_h - pointer to knet_handle_t * * interval - pointer where to store the current interval value * * knet_handle_pmtud_setfreq returns: * * 0 on success * -1 on error and errno is set. */ int knet_handle_pmtud_getfreq(knet_handle_t knet_h, unsigned int *interval); /* * knet_handle_enable_pmtud_notify * * knet_h - pointer to knet_handle_t * * pmtud_notify_fn_private_data * void pointer to data that can be used to identify * the callback. * * pmtud_notify_fn * is a callback function that is invoked every time * a path MTU size change is detected. * The function allows libknet to notify the user * of data MTU, that's the max value that can be send * onwire without fragmentation. The data MTU will always * be lower than real link MTU because it accounts for * protocol overhead, knet packet header and (if configured) * crypto overhead, * This function MUST NEVER block or add substantial delays. * * knet_handle_enable_pmtud_notify returns: * * 0 on success * -1 on error and errno is set. */ int knet_handle_enable_pmtud_notify(knet_handle_t knet_h, void *pmtud_notify_fn_private_data, void (*pmtud_notify_fn) ( void *private_data, unsigned int data_mtu)); /* * knet_handle_pmtud_get * * knet_h - pointer to knet_handle_t * * data_mtu - pointer where to store data_mtu (see above) * * knet_handle_pmtud_get returns: * * 0 on success * -1 on error and errno is set. */ int knet_handle_pmtud_get(knet_handle_t knet_h, unsigned int *data_mtu); /* * knet_handle_crypto * * knet_h - pointer to knet_handle_t * * knet_handle_crypto_cfg - * pointer to a knet_handle_crypto_cfg structure * * crypto_model should contain the model name. * Currently only "nss" is supported. * Setting to "none" will disable crypto. * * crypto_cipher_type * should contain the cipher algo name. * It can be set to "none" to disable * encryption. * Currently supported by "nss" model: * "3des", "aes128", "aes192" and "aes256". * * crypto_hash_type * should contain the hashing algo name. * It can be set to "none" to disable * hashing. * Currently supported by "nss" model: * "md5", "sha1", "sha256", "sha384" and "sha512". * * private_key will contain the private shared key. * It has to be at least KNET_MIN_KEY_LEN long. * * private_key_len * length of the provided private_key. * * Implementation notes/current limitations: * - enabling crypto, will increase latency as packets have * to processed. * - enabling crypto might reduce the overall throughtput * due to crypto data overhead. * - re-keying is not implemented yet. * - private/public key encryption/hashing is not currently * planned. * - crypto key must be the same for all hosts in the same * knet instance. * - it is safe to call knet_handle_crypto multiple times at runtime. * The last config will be used. * IMPORTANT: a call to knet_handle_crypto can fail due to: * 1) failure to obtain locking * 2) errors to initializing the crypto level. * This can happen even in subsequent calls to knet_handle_crypto. * A failure in crypto init, might leave your traffic unencrypted! * It's best to stop data forwarding (see above), change crypto config, * start forward again. * * knet_handle_crypto returns: * * 0 on success * -1 on error and errno is set. * -2 on crypto subsystem initialization error. No errno is provided at the moment (yet). */ #define KNET_MIN_KEY_LEN 1024 #define KNET_MAX_KEY_LEN 4096 struct knet_handle_crypto_cfg { char crypto_model[16]; char crypto_cipher_type[16]; char crypto_hash_type[16]; unsigned char private_key[KNET_MAX_KEY_LEN]; unsigned int private_key_len; }; int knet_handle_crypto(knet_handle_t knet_h, struct knet_handle_crypto_cfg *knet_handle_crypto_cfg); /* * host structs/API calls */ /* * knet_host_add * * knet_h - pointer to knet_handle_t * * host_id - each host in a knet is identified with a unique ID * (see also knet_handle_new documentation above) * * knet_host_add returns: * * 0 on success * -1 on error and errno is set. */ int knet_host_add(knet_handle_t knet_h, uint16_t host_id); /* * knet_host_remove * * knet_h - pointer to knet_handle_t * * host_id - each host in a knet is identified with a unique ID * (see also knet_handle_new documentation above) * * knet_host_remove returns: * * 0 on success * -1 on error and errno is set. */ int knet_host_remove(knet_handle_t knet_h, uint16_t host_id); /* * knet_host_set_name * * knet_h - pointer to knet_handle_t * * host_id - see above * * name - this name will be used for pretty logging and eventually * search for hosts (see also get_name and get_id below). * Only up to KNET_MAX_HOST_LEN - 1 bytes will be accepted and * name has to be unique for each host. * * knet_host_set_name returns: * * 0 on success * -1 on error and errno is set. */ int knet_host_set_name(knet_handle_t knet_h, uint16_t host_id, const char *name); /* * knet_host_get_name_by_host_id * * knet_h - pointer to knet_handle_t * * host_id - see above * * name - pointer to a preallocated buffer of at least size KNET_MAX_HOST_LEN * where the current host name will be stored * (as set by knet_host_set_name or default by knet_host_add) * * knet_host_get_name_by_host_id returns: * * 0 on success * -1 on error and errno is set (name is left untouched) */ int knet_host_get_name_by_host_id(knet_handle_t knet_h, uint16_t host_id, char *name); /* * knet_host_get_id_by_host_name * * knet_h - pointer to knet_handle_t * * name - name to lookup, max len KNET_MAX_HOST_LEN * * host_id - where to store the result * * knet_host_get_id_by_host_name returns: * * 0 on success * -1 on error and errno is set. */ int knet_host_get_id_by_host_name(knet_handle_t knet_h, const char *name, uint16_t *host_id); /* * knet_host_get_host_list * * knet_h - pointer to knet_handle_t * * host_ids - array of at lest KNET_MAX_HOST size * * host_ids_entries - * number of entries writted in host_ids * * knet_host_get_host_list returns: * * 0 on success * -1 on error and errno is set. */ int knet_host_get_host_list(knet_handle_t knet_h, uint16_t *host_ids, size_t *host_ids_entries); /* * define switching policies */ #define KNET_LINK_POLICY_PASSIVE 0 #define KNET_LINK_POLICY_ACTIVE 1 #define KNET_LINK_POLICY_RR 2 /* * knet_host_set_policy * * knet_h - pointer to knet_handle_t * * host_id - see above * * policy - there are currently 3 kind of simple switching policies * as defined above, based on link configuration. * KNET_LINK_POLICY_PASSIVE - the active link with the lowest * priority will be used. * if one or more active links share * the same priority, the one with * lowest link_id will be used. * * KNET_LINK_POLICY_ACTIVE - all active links will be used * simultaneously to send traffic. * link priority is ignored. * * KNET_LINK_POLICY_RR - round-robin policy, every packet * will be send on a different active * link. * * knet_host_set_policy returns: * * 0 on success * -1 on error and errno is set. */ int knet_host_set_policy(knet_handle_t knet_h, uint16_t host_id, uint8_t policy); /* * knet_host_get_policy * * knet_h - pointer to knet_handle_t * * host_id - see above * * policy - will contain the current configured switching policy. * Default is passive when creating a new host. * * knet_host_get_policy returns: * * 0 on success * -1 on error and errno is set. */ int knet_host_get_policy(knet_handle_t knet_h, uint16_t host_id, uint8_t *policy); /* * knet_host_enable_status_change_notify * * knet_h - pointer to knet_handle_t * * host_status_change_notify_fn_private_data * void pointer to data that can be used to identify * the callback. * * host_status_change_notify_fn * is a callback function that is invoked every time * there is a change in the host status. * host status is identified by: * - reachable, this host can send/receive data to/from host_id * - remote, 0 if the host_id is connected locally or 1 if * the there is one or more knet host(s) in between. * NOTE: re-switching is NOT currently implemented, * but this is ready for future and can avoid * an API/ABI breakage later on. * - external, 0 if the host_id is configured locally or 1 if * it has been added from remote nodes config. * NOTE: dynamic topology is NOT currently implemented, * but this is ready for future and can avoid * an API/ABI breakage later on. * This function MUST NEVER block or add substantial delays. * * knet_host_status_change_notify returns: * * 0 on success * -1 on error and errno is set. */ int knet_host_enable_status_change_notify(knet_handle_t knet_h, void *host_status_change_notify_fn_private_data, void (*host_status_change_notify_fn) ( void *private_data, uint16_t host_id, uint8_t reachable, uint8_t remote, uint8_t external)); /* * define host status structure for quick lookup * struct is in flux as more stats will be added soon * * reachable host_id can be seen either directly connected * or via another host_id * * remote 0 = node is connected locally, 1 is visible via * via another host_id * * external 0 = node is configured/known locally, * 1 host_id has been received via another host_id */ struct knet_host_status { uint8_t reachable; uint8_t remote; uint8_t external; /* add host statistics */ }; /* * knet_host_status_get * * knet_h - pointer to knet_handle_t * * status - pointer to knet_host_status struct (see above) * * knet_handle_pmtud_get returns: * * 0 on success * -1 on error and errno is set. */ int knet_host_get_status(knet_handle_t knet_h, uint16_t host_id, struct knet_host_status *status); /* * link structs/API calls * * every host allocated/managed by knet_host_* has * KNET_MAX_LINK structures to define the network * paths that connect 2 hosts. * * Each link is identified by a link_id that has a * values between 0 and KNET_MAX_LINK - 1. * * KNOWN LIMITATIONS: * * - let's assume the scenario where two hosts are connected * with any number of links. link_id must match on both sides. * If host_id 0 link_id 0 is configured to connect IP1 to IP2 and * host_id 0 link_id 1 is configured to connect IP3 to IP4, * host_id 1 link_id 0 _must_ connect IP2 to IP1 and likewise * host_id 1 link_id 1 _must_ connect IP4 to IP3. * We might be able to lift this restriction in future, by using * other data to determine src/dst link_id, but for now, deal with it. * * - */ /* * commodity functions to convert strings to sockaddr and viceversa */ /* * knet_strtoaddr * * host - IPaddr/hostname to convert * be aware only the first IP address will be returned * in case a hostname resolves to multiple IP * * port - port to connect to * * ss - sockaddr_storage where to store the converted data * * sslen - len of the sockaddr_storage * * knet_strtoaddr returns same error codes as getaddrinfo * */ int knet_strtoaddr(const char *host, const char *port, struct sockaddr_storage *ss, socklen_t sslen); /* * knet_addrtostr * * ss - sockaddr_storage to convert * * sslen - len of the sockaddr_storage * * host - IPaddr/hostname where to store data * (recommended size: KNET_MAX_HOST_LEN) * * port - port buffer where to store data * (recommended size: KNET_MAX_PORT_LEN) * * knet_strtoaddr returns same error codes as getnameinfo * */ int knet_addrtostr(const struct sockaddr_storage *ss, socklen_t sslen, char *addr_buf, size_t addr_buf_size, char *port_buf, size_t port_buf_size); #define KNET_TRANSPORT_UDP 0 #define KNET_TRANSPORT_SCTP 1 #define KNET_MAX_TRANSPORTS 2 /* * knet_link_set_config * * knet_h - pointer to knet_handle_t * * host_id - see above * * link_id - see above * * transport - one of the above KNET_TRANSPORT_xxx constants * * src_addr - sockaddr_storage that can be either IPv4 or IPv6 * * dst_addr - sockaddr_storage that can be either IPv4 or IPv6 * this can be null if we don't know the incoming * IP address/port and the link will remain quiet * till the node on the other end will initiate a * connection * * knet_link_set_config returns: * * 0 on success * -1 on error and errno is set. */ int knet_link_set_config(knet_handle_t knet_h, uint16_t host_id, uint8_t link_id, uint8_t transport, struct sockaddr_storage *src_addr, struct sockaddr_storage *dst_addr); /* * knet_link_get_config * * knet_h - pointer to knet_handle_t * * host_id - see above * * link_id - see above * * transport - see above * * src_addr - sockaddr_storage that can be either IPv4 or IPv6 * * dst_addr - sockaddr_storage that can be either IPv4 or IPv6 * * dynamic - 0 if dst_addr is static or 1 if dst_addr is dynamic. * In case of 1, dst_addr can be NULL and it will be left * untouched. * * knet_link_get_config returns: * * 0 on success. * -1 on error and errno is set. */ int knet_link_get_config(knet_handle_t knet_h, uint16_t host_id, uint8_t link_id, uint8_t *transport, struct sockaddr_storage *src_addr, struct sockaddr_storage *dst_addr, uint8_t *dynamic); /* * knet_link_clear_config * * knet_h - pointer to knet_handle_t * * host_id - see above * * link_id - see above * * knet_link_clear_config returns: * * 0 on success. * -1 on error and errno is set. */ int knet_link_clear_config(knet_handle_t knet_h, uint16_t host_id, uint8_t link_id); /* * knet_link_set_enable * * knet_h - pointer to knet_handle_t * * host_id - see above * * link_id - see above * * enabled - 0 disable the link, 1 enable the link * * knet_link_set_enable returns: * * 0 on success * -1 on error and errno is set. */ int knet_link_set_enable(knet_handle_t knet_h, uint16_t host_id, uint8_t link_id, unsigned int enabled); /* * knet_link_get_enable * * knet_h - pointer to knet_handle_t * * host_id - see above * * link_id - see above * * enabled - 0 disable the link, 1 enable the link * * knet_link_get_enable returns: * * 0 on success * -1 on error and errno is set. */ int knet_link_get_enable(knet_handle_t knet_h, uint16_t host_id, uint8_t link_id, unsigned int *enabled); /* * knet_link_set_ping_timers * * knet_h - pointer to knet_handle_t * * host_id - see above * * link_id - see above * * interval - specify the ping interval * * timeout - if no pong is received within this time, * the link is declared dead * * precision - how many values of latency are used to calculate * the average link latency (see also get_status below) * * knet_link_set_ping_timers returns: * * 0 on success * -1 on error and errno is set. */ #define KNET_LINK_DEFAULT_PING_INTERVAL 1000 /* 1 second */ #define KNET_LINK_DEFAULT_PING_TIMEOUT 2000 /* 2 seconds */ #define KNET_LINK_DEFAULT_PING_PRECISION 2048 /* samples */ int knet_link_set_ping_timers(knet_handle_t knet_h, uint16_t host_id, uint8_t link_id, time_t interval, time_t timeout, unsigned int precision); /* * knet_link_get_ping_timers * * knet_h - pointer to knet_handle_t * * host_id - see above * * link_id - see above * * interval - ping intervall * * timeout - if no pong is received within this time, * the link is declared dead * * precision - how many values of latency are used to calculate * the average link latency (see also get_status below) * * knet_link_get_ping_timers returns: * * 0 on success * -1 on error and errno is set. */ int knet_link_get_ping_timers(knet_handle_t knet_h, uint16_t host_id, uint8_t link_id, time_t *interval, time_t *timeout, unsigned int *precision); /* * knet_link_set_pong_count * * knet_h - pointer to knet_handle_t * * host_id - see above * * link_id - see above * * pong_count - how many valid ping/pongs before a link is marked UP. * default: 5, value should be > 0 * * knet_link_set_pong_count returns: * * 0 on success * -1 on error and errno is set. */ #define KNET_LINK_DEFAULT_PONG_COUNT 5 int knet_link_set_pong_count(knet_handle_t knet_h, uint16_t host_id, uint8_t link_id, uint8_t pong_count); /* * knet_link_get_pong_count * * knet_h - pointer to knet_handle_t * * host_id - see above * * link_id - see above * * pong_count - see above * * knet_link_get_pong_count returns: * * 0 on success * -1 on error and errno is set. */ int knet_link_get_pong_count(knet_handle_t knet_h, uint16_t host_id, uint8_t link_id, uint8_t *pong_count); /* * knet_link_set_priority * * knet_h - pointer to knet_handle_t * * host_id - see above * * link_id - see above * * priority - specify the switching priority for this link * see also knet_host_set_policy * * knet_link_set_priority returns: * * 0 on success * -1 on error and errno is set. */ int knet_link_set_priority(knet_handle_t knet_h, uint16_t host_id, uint8_t link_id, uint8_t priority); /* * knet_link_get_priority * * knet_h - pointer to knet_handle_t * * host_id - see above * * link_id - see above * * priority - gather the switching priority for this link * see also knet_host_set_policy * * knet_link_get_priority returns: * * 0 on success * -1 on error and errno is set. */ int knet_link_get_priority(knet_handle_t knet_h, uint16_t host_id, uint8_t link_id, uint8_t *priority); /* * knet_link_get_link_list * * knet_h - pointer to knet_handle_t * * link_ids - array of at lest KNET_MAX_LINK size * with the list of configured links for a certain host. * * link_ids_entries - * number of entries contained in link_ids * * knet_link_get_link_list returns: * * 0 on success * -1 on error and errno is set. */ int knet_link_get_link_list(knet_handle_t knet_h, uint16_t host_id, uint8_t *link_ids, size_t *link_ids_entries); /* * define link status structure for quick lookup * struct is in flux as more stats will be added soon * * src/dst_{ipaddr,port} strings are filled by * getnameinfo(3) when configuring the link. * if the link is dynamic (see knet_link_set_config) * dst_ipaddr/port will contain ipaddr/port of the currently * connected peer or "Unknown" if it was not possible * to determine the ipaddr/port at runtime. * * enabled see also knet_link_set/get_enable. * * connected the link is connected to a peer and ping/pong traffic * is flowing. * * dynconnected the link has dynamic ip on the other end, and * we can see the other host is sending pings to us. * * latency average latency of this link * see also knet_link_set/get_timeout. * * pong_last if the link is down, this value tells us how long * ago this link was active. A value of 0 means that the link * has never been active. */ struct knet_link_status { char src_ipaddr[KNET_MAX_HOST_LEN]; char src_port[KNET_MAX_PORT_LEN]; char dst_ipaddr[KNET_MAX_HOST_LEN]; char dst_port[KNET_MAX_PORT_LEN]; unsigned int enabled:1; /* link is configured and admin enabled for traffic */ unsigned int connected:1; /* link is connected for data (local view) */ unsigned int dynconnected:1; /* link has been activated by remote dynip */ unsigned long long latency; /* average latency computed by fix/exp */ struct timespec pong_last; unsigned int mtu; /* current detected MTU on this link */ unsigned int proto_overhead; /* contains the size of the IP protocol, knet headers and * crypto headers (if configured). This value is filled in * ONLY after the first PMTUd run on that given link, * and can change if link configuration or crypto configuration * changes at runtime. * WARNING: in general mtu + proto_overhead might or might * not match the output of ifconfig mtu due to crypto * requirements to pad packets to some specific boundaries. */ /* add link statistics */ }; /* * knet_link_get_status * * knet_h - pointer to knet_handle_t * * host_id - see above * * link_id - see above * * status - pointer to knet_link_status struct (see above) * * knet_link_get_status returns: * * 0 on success * -1 on error and errno is set. */ int knet_link_get_status(knet_handle_t knet_h, uint16_t host_id, uint8_t link_id, struct knet_link_status *status); /* * logging structs/API calls */ /* * libknet is composed of several subsystems. In order * to easily distinguish log messages coming from different * places, each subsystem has its own ID. * * 0-19 config/management * 20-39 internal threads * 40-59 transports * 60-69 crypto implementations */ #define KNET_SUB_COMMON 0 /* common.c */ #define KNET_SUB_HANDLE 1 /* handle.c alloc/dealloc config changes */ #define KNET_SUB_HOST 2 /* host add/del/modify */ #define KNET_SUB_LISTENER 3 /* listeners add/del/modify... */ #define KNET_SUB_LINK 4 /* link add/del/modify */ #define KNET_SUB_TRANSPORT 5 /* Transport common */ #define KNET_SUB_CRYPTO 6 /* crypto.c config generic layer */ #define KNET_SUB_FILTER 19 /* allocated for users to log from dst_filter */ #define KNET_SUB_DSTCACHE 20 /* switching thread (destination cache handling) */ #define KNET_SUB_HEARTBEAT 21 /* heartbeat thread */ #define KNET_SUB_PMTUD 22 /* Path MTU Discovery thread */ #define KNET_SUB_TX 23 /* send to link thread */ #define KNET_SUB_RX 24 /* recv from link thread */ #define KNET_SUB_TRANSP_UDP 40 /* UDP Transport */ #define KNET_SUB_TRANSP_SCTP 41 /* SCTP Transport */ #define KNET_SUB_NSSCRYPTO 60 /* nsscrypto.c */ #define KNET_SUB_UNKNOWN 254 #define KNET_MAX_SUBSYSTEMS KNET_SUB_UNKNOWN + 1 /* * Convert between subsystem IDs and names */ /* * knet_log_get_subsystem_name * * return internal name of the subsystem or "common" */ const char *knet_log_get_subsystem_name(uint8_t subsystem); /* * knet_log_get_subsystem_id * * return internal ID of the subsystem or KNET_SUB_COMMON */ uint8_t knet_log_get_subsystem_id(const char *name); /* * 4 log levels are enough for everybody */ #define KNET_LOG_ERR 0 /* unrecoverable errors/conditions */ #define KNET_LOG_WARN 1 /* recoverable errors/conditions */ #define KNET_LOG_INFO 2 /* info, link up/down, config changes.. */ #define KNET_LOG_DEBUG 3 /* * Convert between log level values and names */ /* * knet_log_get_loglevel_name * * return internal name of the log level or "ERROR" for unknown values */ const char *knet_log_get_loglevel_name(uint8_t level); /* * knet_log_get_loglevel_id * * return internal log level ID or KNET_LOG_ERR for invalid names */ uint8_t knet_log_get_loglevel_id(const char *name); /* * every log message is composed by a text message (including a trailing \n) * and message level/subsystem IDs. * In order to make debugging easier it is possible to send those packets * straight to stdout/stderr (see knet_bench.c stdout option). */ #define KNET_MAX_LOG_MSG_SIZE 256 struct knet_log_msg { char msg[KNET_MAX_LOG_MSG_SIZE - (sizeof(uint8_t)*2)]; uint8_t subsystem; /* KNET_SUB_* */ uint8_t msglevel; /* KNET_LOG_* */ }; /* * knet_log_set_log_level * * knet_h - same as above * * subsystem - same as above * * level - same as above * * knet_log_set_loglevel allows fine control of log levels by subsystem. * See also knet_handle_new for defaults. * * knet_log_set_loglevel returns: * * 0 on success * -1 on error and errno is set. */ int knet_log_set_loglevel(knet_handle_t knet_h, uint8_t subsystem, uint8_t level); /* * knet_log_get_log_level * * knet_h - same as above * * subsystem - same as above * * level - same as above * * knet_log_get_loglevel returns: * * 0 on success * -1 on error and errno is set. */ int knet_log_get_loglevel(knet_handle_t knet_h, uint8_t subsystem, uint8_t *level); #endif diff --git a/libknet/netutils.c b/libknet/netutils.c index 1093ff36..75604a30 100644 --- a/libknet/netutils.c +++ b/libknet/netutils.c @@ -1,138 +1,148 @@ /* * Copyright (C) 2010-2015 Red Hat, Inc. All rights reserved. * * Authors: Fabio M. Di Nitto * Federico Simoncelli * * This software licensed under GPL-2.0+, LGPL-2.0+ */ #include "config.h" #include #include #include #include #include #include #include "internals.h" #include "netutils.h" static int is_v4_mapped(struct sockaddr_storage *ss, socklen_t salen) { char map[] = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0xff, 0xff }; struct sockaddr_in6 *addr6 = (struct sockaddr_in6 *) ss; return memcmp(&addr6->sin6_addr, map, 12); } /* * unused now - move to libknet for ACL */ int cmpaddr(struct sockaddr_storage *ss1, socklen_t sslen1, struct sockaddr_storage *ss2, socklen_t sslen2) { int ss1_offset = 0, ss2_offset = 0; struct sockaddr_in6 *ss1_addr6 = (struct sockaddr_in6 *)ss1; struct sockaddr_in6 *ss2_addr6 = (struct sockaddr_in6 *)ss2; struct sockaddr_in *ss1_addr = (struct sockaddr_in *)ss1; struct sockaddr_in *ss2_addr = (struct sockaddr_in *)ss2; char *addr1, *addr2; if (ss1->ss_family == ss2->ss_family) return memcmp(ss1, ss2, sslen1); if (ss1->ss_family == AF_INET6) { if (is_v4_mapped(ss1, sslen1)) return 1; addr1 = (char *)&ss1_addr6->sin6_addr; ss1_offset = 12; } else addr1 = (char *)&ss1_addr->sin_addr; if (ss2->ss_family == AF_INET6) { if (is_v4_mapped(ss2, sslen2)) return 1; addr2 = (char *)&ss2_addr6->sin6_addr; ss2_offset = 12; } else addr2 = (char *)&ss2_addr->sin_addr; return memcmp(addr1+ss1_offset, addr2+ss2_offset, 4); } int knet_strtoaddr(const char *host, const char *port, struct sockaddr_storage *ss, socklen_t sslen) { int err; struct addrinfo hints; struct addrinfo *result = NULL; if (!host) { errno = EINVAL; return -1; } if (!port) { errno = EINVAL; return -1; } if (!ss) { errno = EINVAL; return -1; } if (!sslen) { errno = EINVAL; return -1; } memset(&hints, 0, sizeof(struct addrinfo)); hints.ai_family = AF_UNSPEC; hints.ai_socktype = SOCK_DGRAM; hints.ai_flags = AI_NUMERICHOST | AI_NUMERICSERV; err = getaddrinfo(host, port, &hints, &result); if (!err) { memmove(ss, result->ai_addr, (sslen < result->ai_addrlen) ? sslen : result->ai_addrlen); freeaddrinfo(result); } return err; } + +socklen_t sockaddr_len(const struct sockaddr_storage *ss) +{ + if (ss->ss_family == AF_INET) { + return sizeof(struct sockaddr_in); + } else { + return sizeof(struct sockaddr_in6); + } +} + int knet_addrtostr(const struct sockaddr_storage *ss, socklen_t sslen, char *addr_buf, size_t addr_buf_size, char *port_buf, size_t port_buf_size) { if (!ss) { errno = EINVAL; return -1; } if (!sslen) { errno = EINVAL; return -1; } if (!addr_buf) { errno = EINVAL; return -1; } if (!port_buf) { errno = EINVAL; return -1; } - return getnameinfo((struct sockaddr *)ss, sslen, addr_buf, addr_buf_size, + return getnameinfo((struct sockaddr *)ss, sockaddr_len(ss), addr_buf, addr_buf_size, port_buf, port_buf_size, NI_NUMERICHOST | NI_NUMERICSERV); } diff --git a/libknet/netutils.h b/libknet/netutils.h index 1bc4b091..4dddcb59 100644 --- a/libknet/netutils.h +++ b/libknet/netutils.h @@ -1,17 +1,17 @@ /* * Copyright (C) 2010-2015 Red Hat, Inc. All rights reserved. * * Authors: Fabio M. Di Nitto * Federico Simoncelli * * This software licensed under GPL-2.0+, LGPL-2.0+ */ #ifndef __NETUTILS_H__ #define __NETUTILS_H__ #include int cmpaddr(struct sockaddr_storage *ss1, socklen_t sslen1, struct sockaddr_storage *ss2, socklen_t sslen2); - +socklen_t knet_sockaddr_len(const struct sockaddr_storage *ss); #endif diff --git a/libknet/tests/knet_bench.c b/libknet/tests/knet_bench.c index 66df4e08..4166e897 100644 --- a/libknet/tests/knet_bench.c +++ b/libknet/tests/knet_bench.c @@ -1,664 +1,663 @@ /* * Copyright (C) 2016 Red Hat, Inc. All rights reserved. * * Authors: Fabio M. Di Nitto * * This software licensed under GPL-2.0+, LGPL-2.0+ */ #include "config.h" #include #include #include #include #include #include #include #include "libknet.h" #include "internals.h" #include "netutils.h" #include "test-common.h" #define MAX_NODES 128 static int senderid = -1; static knet_handle_t knet_h; static int datafd = 0; static int8_t channel = 0; static int globallistener = 0; static int continous = 0; static struct sockaddr_storage allv4; static struct sockaddr_storage allv6; static int broadcast_test = 1; static pthread_t rx_thread = NULL; static char *rx_buf[PCKT_FRAG_MAX]; static int shutdown_in_progress = 0; static pthread_mutex_t shutdown_mutex = PTHREAD_MUTEX_INITIALIZER; #define TEST_PING 0 #define TEST_PING_AND_DATA 1 #define TEST_PERF 2 static int test_type = TEST_PING; struct node { int nodeid; int links; struct sockaddr_storage address[KNET_MAX_LINK]; }; static void print_help(void) { printf("knet_bench usage:\n"); printf(" -h print this help (no really)\n"); printf(" -d enable debug logs (default INFO)\n"); printf(" -c [implementation]:[crypto]:[hashing] crypto configuration. (default disabled)\n"); printf(" Example: -c nss:aes128:sha1\n"); printf(" -p [active|passive|rr] (default: passive)\n"); printf(" -P [udp|sctp] (default: udp) protocol (transport) to use\n"); printf(" -t [nodeid] This nodeid (required)\n"); printf(" -n [nodeid],[link1_ip_addr],[link2_..] Other nodes information (at least one required)\n"); printf(" Example: -t 1,192.168.8.1,3ffe::8:1,..\n"); printf(" can be repeated up to %d and should contain also the localnode info\n", MAX_NODES); printf(" -b [port] baseport (default: 50000)\n"); printf(" -l enable global listener on 0.0.0.0/:: (default: off, incompatible with -o)\n"); printf(" -o enable baseport offset per nodeid\n"); printf(" -w dont wait for all nodes to be up before starting the test (default: wait)\n"); printf(" -T [ping|ping_data|perf] test type (default: ping)\n"); printf(" ping: will wait for all hosts to join the knet network, sleep 5 seconds and quit\n"); printf(" ping_data: will wait for all hosts to join the knet network, sends some data to all nodes and quit\n"); printf(" perf: will wait for all hosts to join the knet network, perform a series of benchmarks and quit\n"); printf(" -s nodeid that will generate traffic for benchmarks\n"); printf(" -C repeat the test continously (default: off)\n"); } static void parse_nodes(char *nodesinfo[MAX_NODES], int onidx, int port, struct node nodes[MAX_NODES], int thisnodeid, int *thisidx) { int i; char *temp = NULL; char port_str[10]; memset(port_str, 0, sizeof(port_str)); sprintf(port_str, "%d", port); for (i = 0; i < onidx; i++) { nodes[i].nodeid = atoi(strtok(nodesinfo[i], ",")); if ((nodes[i].nodeid < 0) || (nodes[i].nodeid > KNET_MAX_HOST)) { printf("Invalid nodeid: %d (0 - %d)\n", nodes[i].nodeid, KNET_MAX_HOST); exit(FAIL); } if (thisnodeid == nodes[i].nodeid) { *thisidx = i; } while((temp = strtok(NULL, ","))) { if (nodes[i].links == KNET_MAX_LINK) { printf("Too many links configured. Max %d\n", KNET_MAX_LINK); exit(FAIL); } if (knet_strtoaddr(temp, port_str, &nodes[i].address[nodes[i].links], sizeof(struct sockaddr_storage)) < 0) { printf("Unable to convert %s to sockaddress\n", temp); exit(FAIL); } nodes[i].links++; } } if (knet_strtoaddr("0.0.0.0", port_str, &allv4, sizeof(struct sockaddr_storage)) < 0) { printf("Unable to convert 0.0.0.0 to sockaddress\n"); exit(FAIL); } if (knet_strtoaddr("::", port_str, &allv6, sizeof(struct sockaddr_storage)) < 0) { printf("Unable to convert :: to sockaddress\n"); exit(FAIL); } for (i = 1; i < onidx; i++) { if (nodes[0].links != nodes[i].links) { printf("knet_bench does not support unbalanced link configuration\n"); exit(FAIL); } } return; } static int private_data; static void sock_notify(void *pvt_data, int local_datafd, int8_t local_channel, uint8_t tx_rx, int error, int errorno) { printf("Error (%d - %d - %s) from socket: %d\n", error, errorno, strerror(errno), local_datafd); return; } static void setup_knet(int argc, char *argv[]) { int logfd; int rv; char *cryptocfg = NULL, *policystr = NULL, *protostr = NULL; char *othernodeinfo[MAX_NODES]; struct node nodes[MAX_NODES]; int thisnodeid = -1; int thisidx = -1; int onidx = 0; int debug = KNET_LOG_INFO; int port = 50000, portoffset = 0; int thisport = 0, otherport = 0; int thisnewport = 0, othernewport = 0; struct sockaddr_in *so_in; struct sockaddr_in6 *so_in6; struct sockaddr_storage *src; int i, link_idx, allnodesup = 0; int policy = KNET_LINK_POLICY_PASSIVE, policyfound = 0; int protocol = KNET_TRANSPORT_UDP, protofound = 0; int wait = 1; struct knet_handle_crypto_cfg knet_handle_crypto_cfg; char *cryptomodel = NULL, *cryptotype = NULL, *cryptohash = NULL; memset(nodes, 0, sizeof(nodes)); - optind = 0; while ((rv = getopt(argc, argv, "CT:s:ldowb:t:n:c:p:P:h")) != EOF) { switch(rv) { case 'h': print_help(); exit(PASS); break; case 'd': debug = KNET_LOG_DEBUG; break; case 'c': if (cryptocfg) { printf("Error: -c can only be specified once\n"); exit(FAIL); } cryptocfg = optarg; break; case 'p': if (policystr) { printf("Error: -p can only be specified once\n"); exit(FAIL); } policystr = optarg; if (!strcmp(policystr, "active")) { policy = KNET_LINK_POLICY_ACTIVE; policyfound = 1; } if (!strcmp(policystr, "rr")) { policy = KNET_LINK_POLICY_RR; policyfound = 1; } if (!strcmp(policystr, "passive")) { policy = KNET_LINK_POLICY_PASSIVE; policyfound = 1; } if (!policyfound) { printf("Error: invalid policy %s specified. -p accepts active|passive|rr\n", policystr); exit(FAIL); } break; case 'P': if (protostr) { printf("Error: -P can only be specified once\n"); exit(FAIL); } protostr = optarg; if (!strcmp(protostr, "udp")) { protocol = KNET_TRANSPORT_UDP; protofound = 1; } if (!strcmp(protostr, "sctp")) { protocol = KNET_TRANSPORT_SCTP; protofound = 1; } if (!protofound) { printf("Error: invalid protocol %s specified. -P accepts udp|sctp\n", policystr); exit(FAIL); } break; case 't': if (thisnodeid >= 0) { printf("Error: -t can only be specified once\n"); exit(FAIL); } thisnodeid = atoi(optarg); if ((thisnodeid < 0) || (thisnodeid > 65536)) { printf("Error: -t nodeid out of range %d (1 - 65536)\n", thisnodeid); exit(FAIL); } break; case 'n': if (onidx == MAX_NODES) { printf("Error: too many other nodes. Max %d\n", MAX_NODES); exit(FAIL); } othernodeinfo[onidx] = optarg; onidx++; break; case 'b': port = atoi(optarg); if ((port < 1) || (port > 65536)) { printf("Error: port %d out of range (1 - 65536)\n", port); exit(FAIL); } case 'o': if (globallistener) { printf("Error: -l cannot be used with -o\n"); exit(FAIL); } portoffset = 1; break; case 'l': if (portoffset) { printf("Error: -o cannot be used with -l\n"); exit(FAIL); } globallistener = 1; break; case 'w': wait = 0; break; case 's': if (senderid >= 0) { printf("Error: -s can only be specified once\n"); exit(FAIL); } senderid = atoi(optarg); if ((senderid < 0) || (senderid > 65536)) { printf("Error: -s nodeid out of range %d (1 - 65536)\n", senderid); exit(FAIL); } break; case 'T': if (!strcmp("ping", optarg)) { test_type = TEST_PING; } if (!strcmp("ping_data", optarg)) { test_type = TEST_PING_AND_DATA; } if (!strcmp("perf", optarg)) { test_type = TEST_PERF; } break; case 'C': continous = 1; break; default: break; } } if (thisnodeid < 0) { printf("Who am I?!? missing -t from command line?\n"); exit(FAIL); } if (onidx < 1) { printf("no other nodes configured?!? missing -n from command line\n"); exit(FAIL); } parse_nodes(othernodeinfo, onidx, port, nodes, thisnodeid, &thisidx); if (thisidx < 0) { printf("no config for this node found\n"); exit(FAIL); } if (senderid >= 0) { for (i=0; i < onidx; i++) { if (senderid == nodes[i].nodeid) { break; } } if (i == onidx) { printf("Unable to find senderid in nodelist\n"); exit(FAIL); } } if ((test_type == TEST_PERF) && (senderid < 0)) { printf("Error: performance test requires -s to be set (for now)\n"); exit(FAIL); } logfd = start_logging(stdout); knet_h = knet_handle_new(thisnodeid, logfd, debug); if (!knet_h) { printf("Unable to knet_handle_new: %s\n", strerror(errno)); exit(FAIL); } if (cryptocfg) { memset(&knet_handle_crypto_cfg, 0, sizeof(knet_handle_crypto_cfg)); cryptomodel = strtok(cryptocfg, ":"); cryptotype = strtok(NULL, ":"); cryptohash = strtok(NULL, ":"); if (cryptomodel) { strncpy(knet_handle_crypto_cfg.crypto_model, cryptomodel, sizeof(knet_handle_crypto_cfg.crypto_model) - 1); } if (cryptotype) { strncpy(knet_handle_crypto_cfg.crypto_cipher_type, cryptotype, sizeof(knet_handle_crypto_cfg.crypto_cipher_type) - 1); } if (cryptohash) { strncpy(knet_handle_crypto_cfg.crypto_hash_type, cryptohash, sizeof(knet_handle_crypto_cfg.crypto_hash_type) - 1); } knet_handle_crypto_cfg.private_key_len = KNET_MAX_KEY_LEN; if (knet_handle_crypto(knet_h, &knet_handle_crypto_cfg)) { printf("Unable to init crypto\n"); exit(FAIL); } } if (knet_handle_enable_sock_notify(knet_h, &private_data, sock_notify) < 0) { printf("knet_handle_enable_sock_notify failed: %s\n", strerror(errno)); knet_handle_free(knet_h); exit(FAIL); } datafd = 0; channel = -1; if (knet_handle_add_datafd(knet_h, &datafd, &channel) < 0) { printf("knet_handle_add_datafd failed: %s\n", strerror(errno)); knet_handle_free(knet_h); exit(FAIL); } for (i=0; i < onidx; i++) { if (i == thisidx) { continue; } if (knet_host_add(knet_h, nodes[i].nodeid) < 0) { printf("knet_host_add failed: %s\n", strerror(errno)); exit(FAIL); } if (knet_host_set_policy(knet_h, nodes[i].nodeid, policy) < 0) { printf("knet_host_set_policy failed: %s\n", strerror(errno)); exit(FAIL); } for (link_idx = 0; link_idx < nodes[i].links; link_idx++) { if (portoffset) { if (nodes[thisidx].address[link_idx].ss_family == AF_INET) { so_in = (struct sockaddr_in *)&nodes[thisidx].address[link_idx]; thisport = ntohs(so_in->sin_port); thisnewport = thisport + nodes[i].nodeid; so_in->sin_port = (htons(thisnewport)); so_in = (struct sockaddr_in *)&nodes[i].address[link_idx]; otherport = ntohs(so_in->sin_port); othernewport = otherport + nodes[thisidx].nodeid; so_in->sin_port = (htons(othernewport)); } else { so_in6 = (struct sockaddr_in6 *)&nodes[thisidx].address[link_idx]; thisport = ntohs(so_in6->sin6_port); thisnewport = thisport + nodes[i].nodeid; so_in6->sin6_port = (htons(thisnewport)); so_in6 = (struct sockaddr_in6 *)&nodes[i].address[link_idx]; otherport = ntohs(so_in6->sin6_port); othernewport = otherport + nodes[thisidx].nodeid; so_in6->sin6_port = (htons(othernewport)); } } if (!globallistener) { src = &nodes[thisidx].address[link_idx]; } else { if (nodes[thisidx].address[link_idx].ss_family == AF_INET) { src = &allv4; } else { src = &allv6; } } if (knet_link_set_config(knet_h, nodes[i].nodeid, link_idx, protocol, src, &nodes[i].address[link_idx]) < 0) { printf("Unable to configure link: %s\n", strerror(errno)); exit(FAIL); } if (portoffset) { if (nodes[thisidx].address[link_idx].ss_family == AF_INET) { so_in = (struct sockaddr_in *)&nodes[thisidx].address[link_idx]; so_in->sin_port = (htons(thisport)); so_in = (struct sockaddr_in *)&nodes[i].address[link_idx]; so_in->sin_port = (htons(otherport)); } else { so_in6 = (struct sockaddr_in6 *)&nodes[thisidx].address[link_idx]; so_in6->sin6_port = (htons(thisport)); so_in6 = (struct sockaddr_in6 *)&nodes[i].address[link_idx]; so_in6->sin6_port = (htons(otherport)); } } if (knet_link_set_enable(knet_h, nodes[i].nodeid, link_idx, 1) < 0) { printf("knet_link_set_enable failed: %s\n", strerror(errno)); exit(FAIL); } } } if (knet_handle_setfwd(knet_h, 1) < 0) { printf("knet_handle_setfwd failed: %s\n", strerror(errno)); exit(FAIL); } if (wait) { while(!allnodesup) { allnodesup = 1; for (i=0; i < onidx; i++) { if (i == thisidx) { continue; } if(knet_h->host_index[nodes[i].nodeid]->status.reachable != 1) { printf("waiting host %d to be reachable\n", nodes[i].nodeid); allnodesup = 0; } } if (!allnodesup) { sleep(1); } } sleep(1); } } static int ping_dst_host_filter(void *pvt_data, const unsigned char *outdata, ssize_t outdata_len, uint8_t tx_rx, uint16_t this_host_id, uint16_t src_host_id, int8_t *dst_channel, uint16_t *dst_host_ids, size_t *dst_host_ids_entries) { if (broadcast_test) { return 1; } if (tx_rx == KNET_NOTIFY_TX) { memmove(&dst_host_ids[0], outdata, 2); } else { dst_host_ids[0] = this_host_id; } *dst_host_ids_entries = 1; return 0; } static void *_rx_thread(void *args) { fd_set rfds; ssize_t len; struct timeval tv; struct sockaddr_storage address[PCKT_FRAG_MAX]; struct mmsghdr msg[PCKT_FRAG_MAX]; struct iovec iov_in[PCKT_FRAG_MAX]; int i, msg_recv; for (i = 0; i < PCKT_FRAG_MAX; i++) { rx_buf[i] = malloc(KNET_MAX_PACKET_SIZE); if (!rx_buf[i]) { printf("RXT: Unable to malloc!\n"); return NULL; } memset(rx_buf[i], 0, KNET_MAX_PACKET_SIZE); iov_in[i].iov_base = (void *)rx_buf[i]; iov_in[i].iov_len = KNET_MAX_PACKET_SIZE; memset(&msg[i].msg_hdr, 0, sizeof(struct msghdr)); msg[i].msg_hdr.msg_name = &address[i]; msg[i].msg_hdr.msg_namelen = sizeof(struct sockaddr_storage); msg[i].msg_hdr.msg_iov = &iov_in[i]; msg[i].msg_hdr.msg_iovlen = 1; } select_loop: tv.tv_sec = 5; tv.tv_usec = 0; FD_ZERO(&rfds); FD_SET(datafd, &rfds); len = select(FD_SETSIZE, &rfds, NULL, NULL, &tv); if (len < 0) { printf("RXT: Unable select over datafd\nHALTING RX THREAD!\n"); return NULL; } if (!len) { printf("RXT: No data for the past 5 seconds\n"); } if (FD_ISSET(datafd, &rfds)) { msg_recv = recvmmsg(datafd, msg, PCKT_FRAG_MAX, MSG_DONTWAIT | MSG_NOSIGNAL, NULL); if (msg_recv < 0) { printf("RXT: error from recvmmsg: %s\n", strerror(errno)); } for (i = 0; i < msg_recv; i++) { if (msg[i].msg_len == 0) { printf("RXT: received 0 bytes message?\n"); } if (test_type == TEST_PING_AND_DATA) { - printf("received %u bytes message: %s\n", msg[i].msg_len, (char *)msg[i].msg_hdr.msg_iov->iov_base); + printf("received %lu bytes message: %s\n", (long)msg[i].msg_len, (char *)msg[i].msg_hdr.msg_iov->iov_base); } /* * do stats here */ } } goto select_loop; return NULL; } static void setup_data_txrx_common(void) { if (!rx_thread) { if (knet_handle_enable_filter(knet_h, NULL, ping_dst_host_filter)) { printf("Unable to enable dst_host_filter: %s\n", strerror(errno)); exit(FAIL); } printf("Setting up rx thread\n"); if (pthread_create(&rx_thread, 0, _rx_thread, NULL)) { printf("Unable to start rx thread\n"); exit(FAIL); } } } static void stop_rx_thread(void) { void *retval; int i; if (rx_thread) { printf("Shutting down rx thread\n"); pthread_cancel(rx_thread); pthread_join(rx_thread, &retval); for (i = 0; i < PCKT_FRAG_MAX; i ++) { free(rx_buf[i]); } } } static void send_ping_data(void) { const char *buf = "Hello world!\x0"; ssize_t len = strlen(buf); if (knet_send(knet_h, buf, len, channel) != len) { printf("Error sending hello world: %s\n", strerror(errno)); } sleep(1); } static void cleanup_all(void) { if (pthread_mutex_lock(&shutdown_mutex)) { return; } if (shutdown_in_progress) { pthread_mutex_unlock(&shutdown_mutex); return; } shutdown_in_progress = 1; pthread_mutex_unlock(&shutdown_mutex); if (rx_thread) { stop_rx_thread(); } knet_handle_stop(knet_h); } static void sigint_handler(int signum) { printf("Cleaning up... got signal: %d\n", signum); cleanup_all(); exit(PASS); } int main(int argc, char *argv[]) { if (signal(SIGINT, sigint_handler) == SIG_ERR) { printf("Unable to configure SIGINT handler\n"); exit(FAIL); } need_root(); setup_knet(argc, argv); restart: switch(test_type) { default: case TEST_PING: /* basic ping, no data */ sleep(5); break; case TEST_PING_AND_DATA: setup_data_txrx_common(); send_ping_data(); break; case TEST_PERF: setup_data_txrx_common(); break; } if (continous) { goto restart; } cleanup_all(); return PASS; } diff --git a/libknet/tests/test-common.c b/libknet/tests/test-common.c index 9d86cad2..6eb68de4 100644 --- a/libknet/tests/test-common.c +++ b/libknet/tests/test-common.c @@ -1,434 +1,434 @@ /* * Copyright (C) 2016 Red Hat, Inc. All rights reserved. * * Author: Fabio M. Di Nitto * * This software licensed under GPL-2.0+, LGPL-2.0+ */ #include "config.h" #include #include #include #include #include #include #include #include #include #include "libknet.h" #include "test-common.h" static pthread_mutex_t log_mutex = PTHREAD_MUTEX_INITIALIZER; static int log_init = 0; static pthread_mutex_t log_thread_mutex = PTHREAD_MUTEX_INITIALIZER; static pthread_t log_thread; static int log_thread_init = 0; static int log_fds[2]; struct log_thread_data { int logfd; - struct _IO_FILE *std; + FILE *std; }; static struct log_thread_data data; static pthread_mutex_t shutdown_mutex = PTHREAD_MUTEX_INITIALIZER; static int shutdown_in_progress = 0; static int _read_pipe(int fd, char **file, size_t *length) { char buf[4096]; int n; int done = 0; *file = NULL; *length = 0; memset(buf, 0, sizeof(buf)); while (!done) { n = read(fd, buf, sizeof(buf)); if (n < 0) { if (errno == EINTR) continue; if (*file) free(*file); return n; } if (n == 0 && (!*length)) return 0; if (n == 0) done = 1; if (*file) *file = realloc(*file, (*length) + n + done); else *file = malloc(n + done); if (!*file) return -1; memcpy((*file) + (*length), buf, n); *length += (done + n); } /* Null terminator */ (*file)[(*length) - 1] = 0; return 0; } int execute_shell(const char *command, char **error_string) { pid_t pid; int status, err = 0; int fd[2]; size_t size = 0; if ((command == NULL) || (!error_string)) { errno = EINVAL; return FAIL; } *error_string = NULL; err = pipe(fd); if (err) goto out_clean; pid = fork(); if (pid < 0) { err = pid; goto out_clean; } if (pid) { /* parent */ close(fd[1]); err = _read_pipe(fd[0], error_string, &size); if (err) goto out_clean0; waitpid(pid, &status, 0); if (!WIFEXITED(status)) { err = -1; goto out_clean0; } if (WIFEXITED(status) && WEXITSTATUS(status) != 0) { err = WEXITSTATUS(status); goto out_clean0; } goto out_clean0; } else { /* child */ close(0); close(1); close(2); close(fd[0]); dup2(fd[1], 1); dup2(fd[1], 2); close(fd[1]); execlp("/bin/sh", "/bin/sh", "-c", command, NULL); exit(FAIL); } out_clean: close(fd[1]); out_clean0: close(fd[0]); return err; } int is_memcheck(void) { char *val; val = getenv("KNETMEMCHECK"); if (val) { if (!strncmp(val, "yes", 3)) { return 1; } } return 0; } int is_helgrind(void) { char *val; val = getenv("KNETHELGRIND"); if (val) { if (!strncmp(val, "yes", 3)) { return 1; } } return 0; } int need_root(void) { if (geteuid() != 0) { printf("This test requires root privileges\n"); exit(SKIP); } return PASS; } void set_scheduler(int policy) { struct sched_param sched_param; int err; err = sched_get_priority_max(policy); if (err < 0) { printf("Could not get maximum scheduler priority\n"); exit(FAIL); } sched_param.sched_priority = err; err = sched_setscheduler(0, policy, &sched_param); if (err < 0) { printf("Could not set priority\n"); exit(FAIL); } return; } int setup_logpipes(int *logfds) { if (pipe2(logfds, O_CLOEXEC | O_NONBLOCK) < 0) { printf("Unable to setup logging pipe\n"); exit(FAIL); } return PASS; } void close_logpipes(int *logfds) { close(logfds[0]); logfds[0] = 0; close(logfds[1]); logfds[1] = 0; } -void flush_logs(int logfd, struct _IO_FILE *std) +void flush_logs(int logfd, FILE *std) { struct knet_log_msg msg; size_t bytes_read; int len; next: len = 0; bytes_read = 0; memset(&msg, 0, sizeof(struct knet_log_msg)); while (bytes_read < sizeof(struct knet_log_msg)) { len = read(logfd, &msg + bytes_read, sizeof(struct knet_log_msg) - bytes_read); if (len <= 0) { return; } bytes_read += len; } if (len > 0) { fprintf(std, "knet logs: [%s] %s: %s\n", knet_log_get_loglevel_name(msg.msglevel), knet_log_get_subsystem_name(msg.subsystem), msg.msg); goto next; } } static void *_logthread(void *args) { fd_set rfds; ssize_t len; struct timeval tv; select_loop: tv.tv_sec = 5; tv.tv_usec = 0; FD_ZERO(&rfds); FD_SET(data.logfd, &rfds); len = select(FD_SETSIZE, &rfds, NULL, NULL, &tv); if (len < 0) { fprintf(data.std, "Unable select over logfd!\nHALTING LOGTHREAD!\n"); return NULL; } if (!len) { fprintf(data.std, "knet logs: No logs in the last 5 seconds\n"); } if (FD_ISSET(data.logfd, &rfds)) { flush_logs(data.logfd, data.std); } goto select_loop; return NULL; } -int start_logthread(int logfd, struct _IO_FILE *std) +int start_logthread(int logfd, FILE *std) { int savederrno = 0; savederrno = pthread_mutex_lock(&log_thread_mutex); if (savederrno) { printf("Unable to get log_thread mutex lock\n"); return -1; } if (!log_thread_init) { data.logfd = logfd; data.std = std; savederrno = pthread_create(&log_thread, 0, _logthread, NULL); if (savederrno) { printf("Unable to start logging thread: %s\n", strerror(savederrno)); pthread_mutex_unlock(&log_thread_mutex); return -1; } log_thread_init = 1; } pthread_mutex_unlock(&log_thread_mutex); return 0; } int stop_logthread(void) { int savederrno = 0; void *retval; savederrno = pthread_mutex_lock(&log_thread_mutex); if (savederrno) { printf("Unable to get log_thread mutex lock\n"); return -1; } if (log_thread_init) { pthread_cancel(log_thread); pthread_join(log_thread, &retval); log_thread_init = 0; } pthread_mutex_unlock(&log_thread_mutex); return 0; } static void stop_logging(void) { stop_logthread(); flush_logs(log_fds[0], stdout); close_logpipes(log_fds); } -int start_logging(struct _IO_FILE *std) +int start_logging(FILE *std) { int savederrno = 0; savederrno = pthread_mutex_lock(&log_mutex); if (savederrno) { printf("Unable to get log_mutex lock\n"); return -1; } if (!log_init) { setup_logpipes(log_fds); if (atexit(&stop_logging) != 0) { printf("Unable to register atexit handler to stop logging: %s\n", strerror(errno)); exit(FAIL); } if (start_logthread(log_fds[0], std) < 0) { exit(FAIL); } log_init = 1; } pthread_mutex_unlock(&log_mutex); return log_fds[1]; } int knet_handle_stop(knet_handle_t knet_h) { int i, j, savederrno; uint16_t host_ids[KNET_MAX_HOST]; uint8_t link_ids[KNET_MAX_LINK]; size_t host_ids_entries = 0, link_ids_entries = 0; struct knet_link_status status; savederrno = pthread_mutex_lock(&shutdown_mutex); if (savederrno) { printf("Unable to get shutdown mutex lock\n"); return -1; } if (shutdown_in_progress) { pthread_mutex_unlock(&shutdown_mutex); errno = EINVAL; return -1; } shutdown_in_progress = 1; pthread_mutex_unlock(&shutdown_mutex); if (!knet_h) { errno = EINVAL; return -1; } if (knet_host_get_host_list(knet_h, host_ids, &host_ids_entries) < 0) { printf("knet_host_get_host_list failed: %s\n", strerror(errno)); return -1; } for (i = 0; i < host_ids_entries; i++) { if (knet_link_get_link_list(knet_h, host_ids[i], link_ids, &link_ids_entries)) { printf("knet_link_get_link_list failed: %s\n", strerror(errno)); return -1; } for (j = 0; j < link_ids_entries; j++) { if (knet_link_get_status(knet_h, host_ids[i], link_ids[j], &status)) { printf("knet_link_get_status failed: %s\n", strerror(errno)); return -1; } if (status.enabled) { if (knet_link_set_enable(knet_h, host_ids[i], j, 0)) { printf("knet_link_set_enable failed: %s\n", strerror(errno)); return -1; } } knet_link_clear_config(knet_h, host_ids[i], j); } if (knet_host_remove(knet_h, host_ids[i]) < 0) { printf("knet_host_remove failed: %s\n", strerror(errno)); return -1; } } if (knet_handle_free(knet_h)) { printf("knet_handle_free failed: %s\n", strerror(errno)); return -1; } return 0; } diff --git a/libknet/tests/test-common.h b/libknet/tests/test-common.h index 2eca93aa..c45cef55 100644 --- a/libknet/tests/test-common.h +++ b/libknet/tests/test-common.h @@ -1,64 +1,72 @@ /* * Copyright (C) 2016 Red Hat, Inc. All rights reserved. * * Authors: Fabio M. Di Nitto * * This software licensed under GPL-2.0+, LGPL-2.0+ */ #ifndef __TEST_COMMON_H__ #define __TEST_COMMON_H__ #include "internals.h" #include /* * error codes from automake test-driver */ #define PASS 0 #define SKIP 77 #define ERROR 99 #define FAIL -1 +/* For *BSD compatibility */ +#ifndef s6_addr16 +#define s6_addr8 __u6_addr.__u6_addr8 +#define s6_addr16 __u6_addr.__u6_addr16 +#define s6_addr32 __u6_addr.__u6_addr32 +#endif + + /* * common facilities */ int execute_shell(const char *command, char **error_string); int is_memcheck(void); int is_helgrind(void); int need_root(void); void set_scheduler(int policy); /* * consider moving this one as official API */ int knet_handle_stop(knet_handle_t knet_h); /* * high level logging function. * automatically setup logpipes and start/stop logging thread. * * start_logging exit(FAIL) on error or fd to pass to knet_handle_new * and it will install an atexit handle to close logging properly * * WARNING: DO NOT use start_logging for api_ or int_ testing. * while start_logging would work just fine, the output * of the logs is more complex to read because of the way * the thread would interleave the output of printf from api_/int_ testing * with knet logs. Functionally speaking you get the exact same logs, * but a lot harder to read due to the thread latency in printing logs. */ -int start_logging(struct _IO_FILE *std); +int start_logging(FILE *std); int setup_logpipes(int *logfds); void close_logpipes(int *logfds); -void flush_logs(int logfd, struct _IO_FILE *std); -int start_logthread(int logfd, struct _IO_FILE *std); +void flush_logs(int logfd, FILE *std); +int start_logthread(int logfd, FILE *std); int stop_logthread(void); #endif diff --git a/libknet/threads_dsthandler.c b/libknet/threads_dsthandler.c index 63f10734..c01b3847 100644 --- a/libknet/threads_dsthandler.c +++ b/libknet/threads_dsthandler.c @@ -1,61 +1,61 @@ /* * Copyright (C) 2010-2015 Red Hat, Inc. All rights reserved. * * Authors: Fabio M. Di Nitto * Federico Simoncelli * * This software licensed under GPL-2.0+, LGPL-2.0+ */ #include "config.h" #include #include -#include #include "host.h" +#include "compat.h" #include "logging.h" #include "threads_common.h" #include "threads_dsthandler.h" static void _handle_dst_link_updates(knet_handle_t knet_h) { uint16_t host_id; struct knet_host *host; if (recv(knet_h->dstsockfd[0], &host_id, sizeof(host_id), MSG_DONTWAIT | MSG_NOSIGNAL) != sizeof(host_id)) { log_debug(knet_h, KNET_SUB_DSTCACHE, "Short read on dstsockfd"); return; } if (pthread_rwlock_wrlock(&knet_h->global_rwlock) != 0) { log_debug(knet_h, KNET_SUB_DSTCACHE, "Unable to get read lock"); return; } host = knet_h->host_index[host_id]; if (!host) { log_debug(knet_h, KNET_SUB_DSTCACHE, "Unable to find host: %u", host_id); goto out_unlock; } _host_dstcache_update_sync(knet_h, host); out_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); return; } void *_handle_dst_link_handler_thread(void *data) { knet_handle_t knet_h = (knet_handle_t) data; struct epoll_event events[KNET_EPOLL_MAX_EVENTS]; while (!shutdown_in_progress(knet_h)) { if (epoll_wait(knet_h->dst_link_handler_epollfd, events, KNET_EPOLL_MAX_EVENTS, -1) >= 1) _handle_dst_link_updates(knet_h); } return NULL; } diff --git a/libknet/threads_send_recv.c b/libknet/threads_send_recv.c index f7bb6210..15d09723 100644 --- a/libknet/threads_send_recv.c +++ b/libknet/threads_send_recv.c @@ -1,1244 +1,1243 @@ /* * Copyright (C) 2010-2015 Red Hat, Inc. All rights reserved. * * Authors: Fabio M. Di Nitto * Federico Simoncelli * * This software licensed under GPL-2.0+, LGPL-2.0+ */ #include "config.h" #include #include #include #include #include #include -#include #include #include #include #include "crypto.h" #include "compat.h" #include "host.h" #include "link.h" #include "logging.h" #include "transports.h" #include "threads_common.h" #include "threads_send_recv.h" /* * SEND */ static int _dispatch_to_links(knet_handle_t knet_h, struct knet_host *dst_host, struct iovec *iov_out) { int link_idx, msg_idx, sent_msgs, msgs_to_send, prev_sent, progress; struct mmsghdr msg[PCKT_FRAG_MAX]; int err = 0, savederrno = 0; memset(&msg, 0, sizeof(struct mmsghdr)); for (link_idx = 0; link_idx < dst_host->active_link_entries; link_idx++) { msgs_to_send = knet_h->send_to_links_buf[0]->khp_data_frag_num; sent_msgs = 0; prev_sent = 0; progress = 1; retry: msg_idx = 0; while (msg_idx < msgs_to_send) { memset(&msg[msg_idx].msg_hdr, 0, sizeof(struct msghdr)); msg[msg_idx].msg_hdr.msg_name = &dst_host->link[dst_host->active_links[link_idx]].dst_addr; msg[msg_idx].msg_hdr.msg_namelen = sizeof(struct sockaddr_storage); msg[msg_idx].msg_hdr.msg_iov = &iov_out[msg_idx + prev_sent]; msg[msg_idx].msg_hdr.msg_iovlen = 1; msg_idx++; } sent_msgs = sendmmsg(dst_host->link[dst_host->active_links[link_idx]].outsock, msg, msg_idx, MSG_DONTWAIT | MSG_NOSIGNAL); savederrno = errno; if ((sent_msgs >= 0) && (sent_msgs < msg_idx)) { if ((sent_msgs) || (progress)) { msgs_to_send = msg_idx - sent_msgs; prev_sent = prev_sent + sent_msgs; if (sent_msgs) { progress = 1; } else { progress = 0; } log_debug(knet_h, KNET_SUB_TX, "Unable to send all (%d/%d) data packets to host %s (%u) link %s:%s (%u)", sent_msgs, msg_idx, dst_host->name, dst_host->host_id, dst_host->link[dst_host->active_links[link_idx]].status.dst_ipaddr, dst_host->link[dst_host->active_links[link_idx]].status.dst_port, dst_host->link[dst_host->active_links[link_idx]].link_id); goto retry; } if (!progress) { savederrno = EAGAIN; err = -1; goto out_unlock; } } if (sent_msgs < 0) { log_debug(knet_h, KNET_SUB_TX, "Unable to send data packet to host %s (%u) link %s:%s (%u): %s", dst_host->name, dst_host->host_id, dst_host->link[dst_host->active_links[link_idx]].status.dst_ipaddr, dst_host->link[dst_host->active_links[link_idx]].status.dst_port, dst_host->link[dst_host->active_links[link_idx]].link_id, strerror(savederrno)); err = -1; goto out_unlock; } if ((dst_host->link_handler_policy == KNET_LINK_POLICY_RR) && (dst_host->active_link_entries > 1)) { uint8_t cur_link_id = dst_host->active_links[0]; memmove(&dst_host->active_links[0], &dst_host->active_links[1], KNET_MAX_LINK - 1); dst_host->active_links[dst_host->active_link_entries - 1] = cur_link_id; break; } } out_unlock: errno = savederrno; return err; } static int _parse_recv_from_sock(knet_handle_t knet_h, int buf_idx, ssize_t inlen, int8_t channel, int is_sync) { ssize_t outlen, frag_len; struct knet_host *dst_host; uint16_t dst_host_ids_temp[KNET_MAX_HOST]; size_t dst_host_ids_entries_temp = 0; uint16_t dst_host_ids[KNET_MAX_HOST]; size_t dst_host_ids_entries = 0; int bcast = 1; struct knet_hostinfo *knet_hostinfo; struct iovec iov_out[PCKT_FRAG_MAX]; uint8_t frag_idx; unsigned int temp_data_mtu; int host_idx; int send_mcast = 0; struct knet_header *inbuf; int savederrno = 0; int err = 0; inbuf = knet_h->recv_from_sock_buf[buf_idx]; if ((knet_h->enabled != 1) && (inbuf->kh_type != KNET_HEADER_TYPE_HOST_INFO)) { /* data forward is disabled */ log_debug(knet_h, KNET_SUB_TX, "Received data packet but forwarding is disabled"); savederrno = ECANCELED; err = -1; goto out_unlock; } /* * move this into a separate function to expand on * extra switching rules */ switch(inbuf->kh_type) { case KNET_HEADER_TYPE_DATA: if (knet_h->dst_host_filter_fn) { bcast = knet_h->dst_host_filter_fn( knet_h->dst_host_filter_fn_private_data, (const unsigned char *)inbuf->khp_data_userdata, inlen, KNET_NOTIFY_TX, knet_h->host_id, knet_h->host_id, &channel, dst_host_ids_temp, &dst_host_ids_entries_temp); if (bcast < 0) { log_debug(knet_h, KNET_SUB_TX, "Error from dst_host_filter_fn: %d", bcast); savederrno = EFAULT; err = -1; goto out_unlock; } if ((!bcast) && (!dst_host_ids_entries_temp)) { log_debug(knet_h, KNET_SUB_TX, "Message is unicast but no dst_host_ids_entries"); savederrno = EINVAL; err = -1; goto out_unlock; } } break; case KNET_HEADER_TYPE_HOST_INFO: knet_hostinfo = (struct knet_hostinfo *)inbuf->khp_data_userdata; if (knet_hostinfo->khi_bcast == KNET_HOSTINFO_UCAST) { bcast = 0; dst_host_ids_temp[0] = knet_hostinfo->khi_dst_node_id; dst_host_ids_entries_temp = 1; knet_hostinfo->khi_dst_node_id = htons(knet_hostinfo->khi_dst_node_id); } break; default: log_warn(knet_h, KNET_SUB_TX, "Receiving unknown messages from socket"); savederrno = ENOMSG; err = -1; goto out_unlock; break; } if (is_sync) { if ((bcast) || ((!bcast) && (dst_host_ids_entries_temp > 1))) { log_debug(knet_h, KNET_SUB_TX, "knet_send_sync is only supported with unicast packets for one destination"); savederrno = E2BIG; err = -1; goto out_unlock; } } /* * check destinations hosts before spending time * in fragmenting/encrypting packets to save * time processing data for unrechable hosts. * for unicast, also remap the destination data * to skip unreachable hosts. */ if (!bcast) { dst_host_ids_entries = 0; for (host_idx = 0; host_idx < dst_host_ids_entries_temp; host_idx++) { dst_host = knet_h->host_index[dst_host_ids_temp[host_idx]]; if (!dst_host) { continue; } if (dst_host->status.reachable) { dst_host_ids[dst_host_ids_entries] = dst_host_ids_temp[host_idx]; dst_host_ids_entries++; } } if (!dst_host_ids_entries) { savederrno = EHOSTDOWN; err = -1; goto out_unlock; } } else { send_mcast = 0; for (dst_host = knet_h->host_head; dst_host != NULL; dst_host = dst_host->next) { if (dst_host->status.reachable) { send_mcast = 1; break; } } if (!send_mcast) { savederrno = EHOSTDOWN; err = -1; goto out_unlock; } } if (!knet_h->data_mtu) { /* * using MIN_MTU_V4 for data mtu is not completely accurate but safe enough */ log_debug(knet_h, KNET_SUB_TX, "Received data packet but data MTU is still unknown." " Packet might not be delivered." " Assuming mininum IPv4 mtu (%d)", KNET_PMTUD_MIN_MTU_V4); temp_data_mtu = KNET_PMTUD_MIN_MTU_V4; } else { /* * take a copy of the mtu to avoid value changing under * our feet while we are sending a fragmented pckt */ temp_data_mtu = knet_h->data_mtu; } /* * prepare the outgoing buffers */ frag_len = inlen; frag_idx = 0; inbuf->khp_data_bcast = bcast; inbuf->khp_data_frag_num = ceil((float)inlen / temp_data_mtu); inbuf->khp_data_channel = channel; while (frag_idx < inbuf->khp_data_frag_num) { /* * set the iov_base */ iov_out[frag_idx].iov_base = (void *)knet_h->send_to_links_buf[frag_idx]; /* * set the len */ if (frag_len > temp_data_mtu) { iov_out[frag_idx].iov_len = temp_data_mtu + KNET_HEADER_DATA_SIZE; } else { iov_out[frag_idx].iov_len = frag_len + KNET_HEADER_DATA_SIZE; } /* * copy the frag info on all buffers */ knet_h->send_to_links_buf[frag_idx]->kh_type = inbuf->kh_type; knet_h->send_to_links_buf[frag_idx]->khp_data_frag_num = inbuf->khp_data_frag_num; knet_h->send_to_links_buf[frag_idx]->khp_data_bcast = inbuf->khp_data_bcast; knet_h->send_to_links_buf[frag_idx]->khp_data_channel = inbuf->khp_data_channel; memmove(knet_h->send_to_links_buf[frag_idx]->khp_data_userdata, inbuf->khp_data_userdata + (temp_data_mtu * frag_idx), iov_out[frag_idx].iov_len - KNET_HEADER_DATA_SIZE); frag_len = frag_len - temp_data_mtu; frag_idx++; } if (!bcast) { for (host_idx = 0; host_idx < dst_host_ids_entries; host_idx++) { dst_host = knet_h->host_index[dst_host_ids[host_idx]]; knet_h->send_to_links_buf[0]->khp_data_seq_num = htons(++dst_host->ucast_seq_num_tx); frag_idx = 0; while (frag_idx < knet_h->send_to_links_buf[0]->khp_data_frag_num) { knet_h->send_to_links_buf[frag_idx]->khp_data_seq_num = knet_h->send_to_links_buf[0]->khp_data_seq_num; if (knet_h->crypto_instance) { if (crypto_encrypt_and_sign( knet_h, (const unsigned char *)knet_h->send_to_links_buf[frag_idx], iov_out[frag_idx].iov_len, knet_h->send_to_links_buf_crypt[frag_idx], &outlen) < 0) { log_debug(knet_h, KNET_SUB_TX, "Unable to encrypt unicast packet"); savederrno = ECHILD; err = -1; goto out_unlock; } iov_out[frag_idx].iov_base = knet_h->send_to_links_buf_crypt[frag_idx]; iov_out[frag_idx].iov_len = outlen; } frag_idx++; } err = _dispatch_to_links(knet_h, dst_host, iov_out); savederrno = errno; if (err) { goto out_unlock; } } } else { knet_h->send_to_links_buf[0]->khp_data_seq_num = htons(++knet_h->bcast_seq_num_tx); frag_idx = 0; while (frag_idx < knet_h->send_to_links_buf[0]->khp_data_frag_num) { knet_h->send_to_links_buf[frag_idx]->khp_data_seq_num = knet_h->send_to_links_buf[0]->khp_data_seq_num; if (knet_h->crypto_instance) { if (crypto_encrypt_and_sign( knet_h, (const unsigned char *)knet_h->send_to_links_buf[frag_idx], iov_out[frag_idx].iov_len, knet_h->send_to_links_buf_crypt[frag_idx], &outlen) < 0) { log_debug(knet_h, KNET_SUB_TX, "Unable to encrypt unicast packet"); savederrno = ECHILD; err = -1; goto out_unlock; } iov_out[frag_idx].iov_base = knet_h->send_to_links_buf_crypt[frag_idx]; iov_out[frag_idx].iov_len = outlen; } frag_idx++; } for (dst_host = knet_h->host_head; dst_host != NULL; dst_host = dst_host->next) { if (dst_host->status.reachable) { err = _dispatch_to_links(knet_h, dst_host, iov_out); savederrno = errno; if (err) { goto out_unlock; } } } } out_unlock: if ((inlen > 0) && (inbuf->kh_type == KNET_HEADER_TYPE_HOST_INFO)) { if (pthread_mutex_lock(&knet_h->host_mutex) != 0) log_debug(knet_h, KNET_SUB_TX, "Unable to get mutex lock"); pthread_cond_signal(&knet_h->host_cond); pthread_mutex_unlock(&knet_h->host_mutex); } errno = savederrno; return err; } int knet_send_sync(knet_handle_t knet_h, const char *buff, const size_t buff_len, const int8_t channel) { int savederrno = 0, err = 0; if (!knet_h) { errno = EINVAL; return -1; } if (buff == NULL) { errno = EINVAL; return -1; } if (buff_len <= 0) { errno = EINVAL; return -1; } if (buff_len > KNET_MAX_PACKET_SIZE) { errno = EINVAL; return -1; } if (channel < 0) { errno = EINVAL; return -1; } if (channel >= KNET_DATAFD_MAX) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_TX, "Unable to get read lock: %s", strerror(savederrno)); errno = savederrno; return -1; } if (!knet_h->sockfd[channel].in_use) { savederrno = EINVAL; err = -1; goto out; } savederrno = pthread_mutex_lock(&knet_h->tx_mutex); if (savederrno) { log_err(knet_h, KNET_SUB_TX, "Unable to get TX mutex lock: %s", strerror(savederrno)); err = -1; goto out; } knet_h->recv_from_sock_buf[0]->kh_type = KNET_HEADER_TYPE_DATA; memmove(knet_h->recv_from_sock_buf[0]->khp_data_userdata, buff, buff_len); err = _parse_recv_from_sock(knet_h, 0, buff_len, channel, 1); savederrno = errno; pthread_mutex_unlock(&knet_h->tx_mutex); out: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = savederrno; return err; } static void _handle_send_to_links(knet_handle_t knet_h, int sockfd, int8_t channel, struct mmsghdr *msg, int type) { ssize_t inlen = 0; struct iovec iov_in; int msg_recv, i; int savederrno = 0, docallback = 0; if ((channel >= 0) && (channel < KNET_DATAFD_MAX) && (!knet_h->sockfd[channel].is_socket)) { memset(&iov_in, 0, sizeof(iov_in)); iov_in.iov_base = (void *)knet_h->recv_from_sock_buf[0]->khp_data_userdata; iov_in.iov_len = KNET_MAX_PACKET_SIZE; inlen = readv(sockfd, &iov_in, 1); if (inlen <= 0) { savederrno = errno; docallback = 1; goto out; } msg_recv = 1; knet_h->recv_from_sock_buf[0]->kh_type = type; _parse_recv_from_sock(knet_h, 0, inlen, channel, 0); } else { msg_recv = recvmmsg(sockfd, msg, PCKT_FRAG_MAX, MSG_DONTWAIT | MSG_NOSIGNAL, NULL); if (msg_recv < 0) { inlen = msg_recv; savederrno = errno; docallback = 1; goto out; } for (i = 0; i < msg_recv; i++) { inlen = msg[i].msg_len; if (inlen == 0) { savederrno = 0; docallback = 1; goto out; break; } knet_h->recv_from_sock_buf[i]->kh_type = type; _parse_recv_from_sock(knet_h, i, inlen, channel, 0); } } out: if (inlen < 0) { struct epoll_event ev; memset(&ev, 0, sizeof(struct epoll_event)); if (epoll_ctl(knet_h->send_to_links_epollfd, EPOLL_CTL_DEL, knet_h->sockfd[channel].sockfd[knet_h->sockfd[channel].is_created], &ev)) { log_err(knet_h, KNET_SUB_TX, "Unable to del datafd %d from linkfd epoll pool: %s", knet_h->sockfd[channel].sockfd[0], strerror(savederrno)); } else { knet_h->sockfd[channel].has_error = 1; } } if (docallback) { knet_h->sock_notify_fn(knet_h->sock_notify_fn_private_data, knet_h->sockfd[channel].sockfd[0], channel, KNET_NOTIFY_TX, inlen, savederrno); } } void *_handle_send_to_links_thread(void *data) { knet_handle_t knet_h = (knet_handle_t) data; struct epoll_event events[KNET_EPOLL_MAX_EVENTS]; struct sockaddr_storage address[PCKT_FRAG_MAX]; struct mmsghdr msg[PCKT_FRAG_MAX]; struct iovec iov_in[PCKT_FRAG_MAX]; int i, nev, type; int8_t channel; memset(&msg, 0, sizeof(struct mmsghdr)); /* preparing data buffer */ for (i = 0; i < PCKT_FRAG_MAX; i++) { iov_in[i].iov_base = (void *)knet_h->recv_from_sock_buf[i]->khp_data_userdata; iov_in[i].iov_len = KNET_MAX_PACKET_SIZE; memset(&msg[i].msg_hdr, 0, sizeof(struct msghdr)); msg[i].msg_hdr.msg_name = &address[i]; msg[i].msg_hdr.msg_namelen = sizeof(struct sockaddr_storage); msg[i].msg_hdr.msg_iov = &iov_in[i]; msg[i].msg_hdr.msg_iovlen = 1; knet_h->recv_from_sock_buf[i]->kh_version = KNET_HEADER_VERSION; knet_h->recv_from_sock_buf[i]->khp_data_frag_seq = 0; knet_h->recv_from_sock_buf[i]->kh_node = htons(knet_h->host_id); knet_h->send_to_links_buf[i]->kh_version = KNET_HEADER_VERSION; knet_h->send_to_links_buf[i]->khp_data_frag_seq = i + 1; knet_h->send_to_links_buf[i]->kh_node = htons(knet_h->host_id); } while (!shutdown_in_progress(knet_h)) { nev = epoll_wait(knet_h->send_to_links_epollfd, events, KNET_EPOLL_MAX_EVENTS + 1, -1); if (pthread_rwlock_rdlock(&knet_h->global_rwlock) != 0) { log_debug(knet_h, KNET_SUB_TX, "Unable to get read lock"); continue; } for (i = 0; i < nev; i++) { if (events[i].data.fd == knet_h->hostsockfd[0]) { type = KNET_HEADER_TYPE_HOST_INFO; channel = -1; } else { type = KNET_HEADER_TYPE_DATA; for (channel = 0; channel < KNET_DATAFD_MAX; channel++) { if ((knet_h->sockfd[channel].in_use) && (knet_h->sockfd[channel].sockfd[knet_h->sockfd[channel].is_created] == events[i].data.fd)) { break; } } } if (pthread_mutex_lock(&knet_h->tx_mutex) != 0) { log_debug(knet_h, KNET_SUB_TX, "Unable to get mutex lock"); continue; } _handle_send_to_links(knet_h, events[i].data.fd, channel, msg, type); pthread_mutex_unlock(&knet_h->tx_mutex); } pthread_rwlock_unlock(&knet_h->global_rwlock); } return NULL; } /* * RECV */ /* * return 1 if a > b * return -1 if b > a * return 0 if they are equal */ static inline int timecmp(struct timespec a, struct timespec b) { if (a.tv_sec != b.tv_sec) { if (a.tv_sec > b.tv_sec) { return 1; } else { return -1; } } else { if (a.tv_nsec > b.tv_nsec) { return 1; } else if (a.tv_nsec < b.tv_nsec) { return -1; } else { return 0; } } } /* * this functions needs to return an index (0 to 7) * to a knet_host_defrag_buf. (-1 on errors) */ static int find_pckt_defrag_buf(knet_handle_t knet_h, struct knet_header *inbuf) { struct knet_host *src_host = knet_h->host_index[inbuf->kh_node]; int i, oldest; /* * check if there is a buffer already in use handling the same seq_num */ for (i = 0; i < KNET_MAX_LINK; i++) { if (src_host->defrag_buf[i].in_use) { if (src_host->defrag_buf[i].pckt_seq == inbuf->khp_data_seq_num) { return i; } } } /* * If there is no buffer that's handling the current seq_num * either it's new or it's been reclaimed already. * check if it's been reclaimed/seen before using the defrag circular * buffer. If the pckt has been seen before, the buffer expired (ETIME) * and there is no point to try to defrag it again. */ if (!_seq_num_lookup(src_host, inbuf->khp_data_bcast, inbuf->khp_data_seq_num, 1)) { errno = ETIME; return -1; } /* * register the pckt as seen */ _seq_num_set(src_host, inbuf->khp_data_bcast, inbuf->khp_data_seq_num, 1); /* * see if there is a free buffer */ for (i = 0; i < KNET_MAX_LINK; i++) { if (!src_host->defrag_buf[i].in_use) { return i; } } /* * at this point, there are no free buffers, the pckt is new * and we need to reclaim a buffer, and we will take the one * with the oldest timestamp. It's as good as any. */ oldest = 0; for (i = 0; i < KNET_MAX_LINK; i++) { if (timecmp(src_host->defrag_buf[i].last_update, src_host->defrag_buf[oldest].last_update) < 0) { oldest = i; } } src_host->defrag_buf[oldest].in_use = 0; return oldest; } static int pckt_defrag(knet_handle_t knet_h, struct knet_header *inbuf, ssize_t *len) { struct knet_host_defrag_buf *defrag_buf; int defrag_buf_idx; defrag_buf_idx = find_pckt_defrag_buf(knet_h, inbuf); if (defrag_buf_idx < 0) { if (errno == ETIME) { log_debug(knet_h, KNET_SUB_RX, "Defrag buffer expired"); } return 1; } defrag_buf = &knet_h->host_index[inbuf->kh_node]->defrag_buf[defrag_buf_idx]; /* * if the buf is not is use, then make sure it's clean */ if (!defrag_buf->in_use) { memset(defrag_buf, 0, sizeof(struct knet_host_defrag_buf)); defrag_buf->in_use = 1; defrag_buf->pckt_seq = inbuf->khp_data_seq_num; } /* * update timestamp on the buffer */ clock_gettime(CLOCK_MONOTONIC, &defrag_buf->last_update); /* * check if we already received this fragment */ if (defrag_buf->frag_map[inbuf->khp_data_frag_seq]) { /* * if we have received this fragment and we didn't clear the buffer * it means that we don't have all fragments yet */ return 1; } /* * we need to handle the last packet with gloves due to its different size */ if (inbuf->khp_data_frag_seq == inbuf->khp_data_frag_num) { defrag_buf->last_frag_size = *len; /* * in the event when the last packet arrives first, * we still don't know the offset vs the other fragments (based on MTU), * so we store the fragment at the end of the buffer where it's safe * and take a copy of the len so that we can restore its offset later. * remember we can't use the local MTU for this calculation because pMTU * can be asymettric between the same hosts. */ if (!defrag_buf->frag_size) { defrag_buf->last_first = 1; memmove(defrag_buf->buf + (KNET_MAX_PACKET_SIZE - *len), inbuf->khp_data_userdata, *len); } } else { defrag_buf->frag_size = *len; } memmove(defrag_buf->buf + ((inbuf->khp_data_frag_seq - 1) * defrag_buf->frag_size), inbuf->khp_data_userdata, *len); defrag_buf->frag_recv++; defrag_buf->frag_map[inbuf->khp_data_frag_seq] = 1; /* * check if we received all the fragments */ if (defrag_buf->frag_recv == inbuf->khp_data_frag_num) { /* * special case the last pckt */ if (defrag_buf->last_first) { memmove(defrag_buf->buf + ((inbuf->khp_data_frag_num - 1) * defrag_buf->frag_size), defrag_buf->buf + (KNET_MAX_PACKET_SIZE - defrag_buf->last_frag_size), defrag_buf->last_frag_size); } /* * recalculate packet lenght */ *len = ((inbuf->khp_data_frag_num - 1) * defrag_buf->frag_size) + defrag_buf->last_frag_size; /* * copy the pckt back in the user data */ memmove(inbuf->khp_data_userdata, defrag_buf->buf, *len); /* * free this buffer */ defrag_buf->in_use = 0; return 0; } return 1; } static void _parse_recv_from_links(knet_handle_t knet_h, struct sockaddr_storage *address, int ind, ssize_t len) { ssize_t outlen; struct knet_host *src_host; struct knet_link *src_link; unsigned long long latency_last; uint16_t dst_host_ids[KNET_MAX_HOST]; size_t dst_host_ids_entries = 0; int bcast = 1; struct timespec recvtime; struct knet_header *inbuf = knet_h->recv_from_links_buf[ind]; unsigned char *outbuf = (unsigned char *)knet_h->recv_from_links_buf[ind]; struct knet_hostinfo *knet_hostinfo; struct iovec iov_out[1]; int8_t channel; if (knet_h->crypto_instance) { if (crypto_authenticate_and_decrypt(knet_h, (unsigned char *)inbuf, len, knet_h->recv_from_links_buf_decrypt, &outlen) < 0) { log_debug(knet_h, KNET_SUB_RX, "Unable to decrypt/auth packet"); return; } len = outlen; inbuf = (struct knet_header *)knet_h->recv_from_links_buf_decrypt; } if (len < (KNET_HEADER_SIZE + 1)) { log_debug(knet_h, KNET_SUB_RX, "Packet is too short: %ld", len); return; } if (inbuf->kh_version != KNET_HEADER_VERSION) { log_debug(knet_h, KNET_SUB_RX, "Packet version does not match"); return; } inbuf->kh_node = ntohs(inbuf->kh_node); src_host = knet_h->host_index[inbuf->kh_node]; if (src_host == NULL) { /* host not found */ log_debug(knet_h, KNET_SUB_RX, "Unable to find source host for this packet"); return; } src_link = NULL; if ((inbuf->kh_type & KNET_HEADER_TYPE_PMSK) != 0) { src_link = src_host->link + (inbuf->khp_ping_link % KNET_MAX_LINK); if (src_link->dynamic == KNET_LINK_DYNIP) { if (memcmp(&src_link->dst_addr, address, sizeof(struct sockaddr_storage)) != 0) { log_debug(knet_h, KNET_SUB_RX, "host: %u link: %u appears to have changed ip address", src_host->host_id, src_link->link_id); memmove(&src_link->dst_addr, address, sizeof(struct sockaddr_storage)); if (knet_addrtostr(&src_link->dst_addr, sizeof(struct sockaddr_storage), src_link->status.dst_ipaddr, KNET_MAX_HOST_LEN, src_link->status.dst_port, KNET_MAX_PORT_LEN) != 0) { log_debug(knet_h, KNET_SUB_RX, "Unable to resolve ???"); snprintf(src_link->status.dst_ipaddr, KNET_MAX_HOST_LEN - 1, "Unknown!!!"); snprintf(src_link->status.dst_port, KNET_MAX_PORT_LEN - 1, "??"); } } src_link->status.dynconnected = 1; } } switch (inbuf->kh_type) { case KNET_HEADER_TYPE_HOST_INFO: case KNET_HEADER_TYPE_DATA: inbuf->khp_data_seq_num = ntohs(inbuf->khp_data_seq_num); channel = inbuf->khp_data_channel; if (!_seq_num_lookup(src_host, inbuf->khp_data_bcast, inbuf->khp_data_seq_num, 0)) { if (src_host->link_handler_policy != KNET_LINK_POLICY_ACTIVE) { log_debug(knet_h, KNET_SUB_RX, "Packet has already been delivered"); } return; } if (inbuf->khp_data_frag_num > 1) { /* * len as received from the socket also includes extra stuff * that the defrag code doesn't care about. So strip it * here and readd only for repadding once we are done * defragging */ len = len - KNET_HEADER_DATA_SIZE; if (pckt_defrag(knet_h, inbuf, &len)) { return; } len = len + KNET_HEADER_DATA_SIZE; } if (inbuf->kh_type == KNET_HEADER_TYPE_DATA) { if (knet_h->enabled != 1) /* data forward is disabled */ break; if (knet_h->dst_host_filter_fn) { int host_idx; int found = 0; bcast = knet_h->dst_host_filter_fn( knet_h->dst_host_filter_fn_private_data, (const unsigned char *)inbuf->khp_data_userdata, len - KNET_HEADER_DATA_SIZE, KNET_NOTIFY_RX, knet_h->host_id, inbuf->kh_node, &channel, dst_host_ids, &dst_host_ids_entries); if (bcast < 0) { log_debug(knet_h, KNET_SUB_RX, "Error from dst_host_filter_fn: %d", bcast); return; } if ((!bcast) && (!dst_host_ids_entries)) { log_debug(knet_h, KNET_SUB_RX, "Message is unicast but no dst_host_ids_entries"); return; } /* check if we are dst for this packet */ if (!bcast) { for (host_idx = 0; host_idx < dst_host_ids_entries; host_idx++) { if (dst_host_ids[host_idx] == knet_h->host_id) { found = 1; break; } } if (!found) { log_debug(knet_h, KNET_SUB_RX, "Packet is not for us"); return; } } } } if (inbuf->kh_type == KNET_HEADER_TYPE_DATA) { if (!knet_h->sockfd[channel].in_use) { log_debug(knet_h, KNET_SUB_RX, "received packet for channel %d but there is no local sock connected", channel); return; } memset(iov_out, 0, sizeof(iov_out)); iov_out[0].iov_base = (void *) inbuf->khp_data_userdata; iov_out[0].iov_len = len - KNET_HEADER_DATA_SIZE; outlen = writev(knet_h->sockfd[channel].sockfd[knet_h->sockfd[channel].is_created], iov_out, 1); if (outlen <= 0) { knet_h->sock_notify_fn(knet_h->sock_notify_fn_private_data, knet_h->sockfd[channel].sockfd[0], channel, KNET_NOTIFY_RX, outlen, errno); return; } if (outlen == iov_out[0].iov_len) { _seq_num_set(src_host, bcast, inbuf->khp_data_seq_num, 0); } } else { /* HOSTINFO */ knet_hostinfo = (struct knet_hostinfo *)inbuf->khp_data_userdata; if (knet_hostinfo->khi_bcast == KNET_HOSTINFO_UCAST) { bcast = 0; knet_hostinfo->khi_dst_node_id = ntohs(knet_hostinfo->khi_dst_node_id); } if (!_seq_num_lookup(src_host, bcast, inbuf->khp_data_seq_num, 0)) { return; } _seq_num_set(src_host, bcast, inbuf->khp_data_seq_num, 0); switch(knet_hostinfo->khi_type) { case KNET_HOSTINFO_TYPE_LINK_UP_DOWN: src_link = src_host->link + (knet_hostinfo->khip_link_status_link_id % KNET_MAX_LINK); /* * basically if the node is coming back to life from a crash * we should receive a host info where local previous status == remote current status * and so we can detect that node is showing up again * we need to clear cbuffers and notify the node of our status by resending our host info */ if ((src_link->remoteconnected == KNET_HOSTINFO_LINK_STATUS_UP) && (src_link->remoteconnected == knet_hostinfo->khip_link_status_status)) { src_link->host_info_up_sent = 0; } src_link->remoteconnected = knet_hostinfo->khip_link_status_status; if (src_link->remoteconnected == KNET_HOSTINFO_LINK_STATUS_DOWN) { /* * if a host is disconnecting clean, we note that in donnotremoteupdate * so that we don't send host info back immediately but we wait * for the node to send an update when it's alive again */ src_link->host_info_up_sent = 0; src_link->donnotremoteupdate = 1; } else { src_link->donnotremoteupdate = 0; } log_debug(knet_h, KNET_SUB_RX, "host message up/down. from host: %u link: %u remote connected: %u", src_host->host_id, src_link->link_id, src_link->remoteconnected); if (_host_dstcache_update_async(knet_h, src_host)) { log_debug(knet_h, KNET_SUB_RX, "Unable to update switch cache for host: %u link: %u remote connected: %u)", src_host->host_id, src_link->link_id, src_link->remoteconnected); } break; case KNET_HOSTINFO_TYPE_LINK_TABLE: break; default: log_warn(knet_h, KNET_SUB_RX, "Receiving unknown host info message from host %u", src_host->host_id); break; } } break; case KNET_HEADER_TYPE_PING: outlen = KNET_HEADER_PING_SIZE; inbuf->kh_type = KNET_HEADER_TYPE_PONG; inbuf->kh_node = htons(knet_h->host_id); if (knet_h->crypto_instance) { if (crypto_encrypt_and_sign(knet_h, (const unsigned char *)inbuf, len, knet_h->recv_from_links_buf_crypt, &outlen) < 0) { log_debug(knet_h, KNET_SUB_RX, "Unable to encrypt pong packet"); break; } outbuf = knet_h->recv_from_links_buf_crypt; } if (sendto(src_link->outsock, outbuf, outlen, MSG_DONTWAIT | MSG_NOSIGNAL, (struct sockaddr *) &src_link->dst_addr, sizeof(struct sockaddr_storage)) != outlen) { log_debug(knet_h, KNET_SUB_RX, "Unable to send pong reply (sock: %d) packet (sendto): %d %s. recorded src ip: %s src port: %s dst ip: %s dst port: %s", src_link->outsock, errno, strerror(errno), src_link->status.src_ipaddr, src_link->status.src_port, src_link->status.dst_ipaddr, src_link->status.dst_port); } break; case KNET_HEADER_TYPE_PONG: clock_gettime(CLOCK_MONOTONIC, &src_link->status.pong_last); memmove(&recvtime, &inbuf->khp_ping_time[0], sizeof(struct timespec)); timespec_diff(recvtime, src_link->status.pong_last, &latency_last); src_link->status.latency = ((src_link->status.latency * src_link->latency_exp) + ((latency_last / 1000llu) * (src_link->latency_fix - src_link->latency_exp))) / src_link->latency_fix; if (src_link->status.latency < src_link->pong_timeout) { if (!src_link->status.connected) { if (src_link->received_pong >= src_link->pong_count) { log_info(knet_h, KNET_SUB_RX, "host: %u link: %u is up", src_host->host_id, src_link->link_id); _link_updown(knet_h, src_host->host_id, src_link->link_id, src_link->status.enabled, 1); } else { src_link->received_pong++; log_debug(knet_h, KNET_SUB_RX, "host: %u link: %u received pong: %u", src_host->host_id, src_link->link_id, src_link->received_pong); } } } break; case KNET_HEADER_TYPE_PMTUD: outlen = KNET_HEADER_PMTUD_SIZE; inbuf->kh_type = KNET_HEADER_TYPE_PMTUD_REPLY; inbuf->kh_node = htons(knet_h->host_id); if (knet_h->crypto_instance) { if (crypto_encrypt_and_sign(knet_h, (const unsigned char *)inbuf, len, knet_h->recv_from_links_buf_crypt, &outlen) < 0) { log_debug(knet_h, KNET_SUB_RX, "Unable to encrypt PMTUd reply packet"); break; } outbuf = knet_h->recv_from_links_buf_crypt; } if (sendto(src_link->outsock, outbuf, outlen, MSG_DONTWAIT | MSG_NOSIGNAL, (struct sockaddr *) &src_link->dst_addr, sizeof(struct sockaddr_storage)) != outlen) { log_debug(knet_h, KNET_SUB_RX, "Unable to send PMTUd reply (sock: %d) packet (sendto): %d %s. recorded src ip: %s src port: %s dst ip: %s dst port: %s", src_link->outsock, errno, strerror(errno), src_link->status.src_ipaddr, src_link->status.src_port, src_link->status.dst_ipaddr, src_link->status.dst_port); } break; case KNET_HEADER_TYPE_PMTUD_REPLY: if (pthread_mutex_lock(&knet_h->pmtud_mutex) != 0) { log_debug(knet_h, KNET_SUB_RX, "Unable to get mutex lock"); break; } src_link->last_recv_mtu = inbuf->khp_pmtud_size; pthread_cond_signal(&knet_h->pmtud_cond); pthread_mutex_unlock(&knet_h->pmtud_mutex); break; default: return; } } static void _handle_recv_from_links(knet_handle_t knet_h, int sockfd, struct mmsghdr *msg) { int err, savederrno; int i, msg_recv, transport; if (pthread_rwlock_rdlock(&knet_h->global_rwlock) != 0) { log_debug(knet_h, KNET_SUB_RX, "Unable to get global read lock"); return; } if (_is_valid_fd(knet_h, sockfd) < 1) { /* * this is normal if a fd got an event and before we grab the read lock * and the link is removed by another thread */ goto exit_unlock; } transport = knet_h->knet_transport_fd_tracker[sockfd].transport; msg_recv = recvmmsg(sockfd, msg, PCKT_FRAG_MAX, MSG_DONTWAIT | MSG_NOSIGNAL, NULL); savederrno = errno; /* * WARNING: man page for recvmmsg is wrong. Kernel implementation here: * recvmmsg can return: * -1 on error * 0 if the previous run of recvmmsg recorded an error on the socket * N number of messages (see exception below). * * If there is an error from recvmsg after receiving a frame or more, the recvmmsg * loop is interrupted, error recorded in the socket (getsockopt(SO_ERROR) and * it will be visibile in the next run. * * Need to be careful how we handle errors at this stage. * * error messages need to be handled on a per transport/protocol base * at this point we have different layers of error handling * - msg_recv < 0 -> error from this run * msg_recv = 0 -> error from previous run and error on socket needs to be cleared * - per-transport message data * example: msg[i].msg_hdr.msg_flags & MSG_NOTIFICATION or msg_len for SCTP == EOF, * but for UDP it is perfectly legal to receive a 0 bytes message.. go figure * - NOTE: on SCTP MSG_NOTIFICATION we get msg_recv == PCKT_FRAG_MAX messages and no * errno set. That means the error api needs to be able to abort the loop below. */ if (msg_recv <= 0) { log_err(knet_h, KNET_SUB_RX, "Error message received from recvmmsg on socket %d: %s", sockfd, strerror(savederrno)); knet_h->transport_ops[transport]->transport_rx_sock_error(knet_h, sockfd, msg_recv, savederrno); goto exit_unlock; } for (i = 0; i < msg_recv; i++) { err = knet_h->transport_ops[transport]->transport_rx_is_data(knet_h, sockfd, msg[i]); /* * TODO: make this section silent once we are confident * all protocols packet handlers are good */ switch(err) { case -1: /* on error */ log_debug(knet_h, KNET_SUB_RX, "Transport reported error parsing packet"); goto exit_unlock; break; case 0: /* packet is not data and we should continue the packet process loop */ log_debug(knet_h, KNET_SUB_RX, "Transport reported no data, continue"); break; case 1: /* packet is not data and we should STOP the packet process loop */ log_debug(knet_h, KNET_SUB_RX, "Transport reported no data, stop"); goto exit_unlock; break; case 2: /* packet is data and should be parsed as such */ _parse_recv_from_links(knet_h, (struct sockaddr_storage *)&msg[i].msg_hdr.msg_name, i, msg[i].msg_len); break; } } exit_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); } void *_handle_recv_from_links_thread(void *data) { int i, nev; knet_handle_t knet_h = (knet_handle_t) data; struct epoll_event events[KNET_EPOLL_MAX_EVENTS]; struct sockaddr_storage address[PCKT_FRAG_MAX]; struct mmsghdr msg[PCKT_FRAG_MAX]; struct iovec iov_in[PCKT_FRAG_MAX]; memset(&msg, 0, sizeof(struct mmsghdr)); for (i = 0; i < PCKT_FRAG_MAX; i++) { iov_in[i].iov_base = (void *)knet_h->recv_from_links_buf[i]; iov_in[i].iov_len = KNET_DATABUFSIZE; memset(&msg[i].msg_hdr, 0, sizeof(struct msghdr)); msg[i].msg_hdr.msg_name = &address[i]; msg[i].msg_hdr.msg_namelen = sizeof(struct sockaddr_storage); msg[i].msg_hdr.msg_iov = &iov_in[i]; msg[i].msg_hdr.msg_iovlen = 1; } while (!shutdown_in_progress(knet_h)) { nev = epoll_wait(knet_h->recv_from_links_epollfd, events, KNET_EPOLL_MAX_EVENTS, -1); for (i = 0; i < nev; i++) { _handle_recv_from_links(knet_h, events[i].data.fd, msg); } } return NULL; } diff --git a/libknet/transport_common.c b/libknet/transport_common.c index 3d993ee3..22e7a96f 100644 --- a/libknet/transport_common.c +++ b/libknet/transport_common.c @@ -1,231 +1,286 @@ #include "config.h" #include #include #include #include -#include #include #include #include "libknet.h" +#include "compat.h" #include "host.h" #include "link.h" #include "logging.h" #include "common.h" #include "transports.h" int _configure_common_socket(knet_handle_t knet_h, int sock, const char *type) { int err = 0, savederrno = 0; int value; if (_fdset_cloexec(sock)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set %s CLOEXEC socket opts: %s", type, strerror(savederrno)); goto exit_error; } if (_fdset_nonblock(sock)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set %s NONBLOCK socket opts: %s", type, strerror(savederrno)); goto exit_error; } +#ifdef SO_RCVBUFFORCE value = KNET_RING_RCVBUFF; if (setsockopt(sock, SOL_SOCKET, SO_RCVBUFFORCE, &value, sizeof(value)) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set %s receive buffer: %s", type, strerror(savederrno)); goto exit_error; } +#else + value = KNET_RING_RCVBUFF; + if (setsockopt(sock, SOL_SOCKET, SO_RCVBUF, &value, sizeof(value)) < 0) { + savederrno = errno; + err = -1; + log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set %s SO_RECVBUF: %s", + type, strerror(savederrno)); + goto exit_error; + } +#endif value = KNET_RING_RCVBUFF; +#ifdef SO_SNDBUFFORCE if (setsockopt(sock, SOL_SOCKET, SO_SNDBUFFORCE, &value, sizeof(value)) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set %s send buffer: %s", type, strerror(savederrno)); goto exit_error; } +#else + if (setsockopt(sock, SOL_SOCKET, SO_SNDBUF, &value, sizeof(value)) < 0) { + savederrno = errno; + err = -1; + log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set %s SO_SNDBUF: %s", + type, strerror(savederrno)); + goto exit_error; + } +#endif exit_error: errno = savederrno; return err; } int _configure_transport_socket(knet_handle_t knet_h, int sock, struct sockaddr_storage *address, const char *type) { int err = 0, savederrno = 0; int value; if (_configure_common_socket(knet_h, sock, type) < 0) { savederrno = errno; err = -1; goto exit_error; } +#ifdef IP_FREEBIND value = 1; if (setsockopt(sock, SOL_IP, IP_FREEBIND, &value, sizeof(value)) <0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set FREEBIND on %s socket: %s", type, strerror(savederrno)); goto exit_error; } +#endif +#ifdef IP_BINDANY /* BSD */ + value = 1; + if (setsockopt(sock, IPPROTO_IP, IP_BINDANY, &value, sizeof(value)) <0) { + savederrno = errno; + err = -1; + log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set BINDANY on %s socket: %s", + type, strerror(savederrno)); + goto exit_error; + } +#endif if (address->ss_family == AF_INET6) { value = 1; if (setsockopt(sock, IPPROTO_IPV6, IPV6_V6ONLY, &value, sizeof(value)) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set %s IPv6 only: %s", type, strerror(savederrno)); goto exit_error; } +#ifdef IPV6_MTU_DISCOVER value = IPV6_PMTUDISC_PROBE; if (setsockopt(sock, SOL_IPV6, IPV6_MTU_DISCOVER, &value, sizeof(value)) <0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set PMTUDISC on %s socket: %s", type, strerror(savederrno)); goto exit_error; } +#else + value = 1; + if (setsockopt(sock, IPPROTO_IPV6, IPV6_DONTFRAG, &value, sizeof(value)) <0) { + savederrno = errno; + err = -1; + log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set DONTFRAG on %s socket: %s", + type, strerror(savederrno)); + goto exit_error; + } +#endif } else { +#ifdef IP_MTU_DISCOVER value = IP_PMTUDISC_PROBE; if (setsockopt(sock, SOL_IP, IP_MTU_DISCOVER, &value, sizeof(value)) <0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set PMTUDISC on %s socket: %s", type, strerror(savederrno)); goto exit_error; } +#else + value = 1; + if (setsockopt(sock, IPPROTO_IP, IP_DONTFRAG, &value, sizeof(value)) <0) { + savederrno = errno; + err = -1; + log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set DONTFRAG on %s socket: %s", + type, strerror(savederrno)); + goto exit_error; + } +#endif } value = 1; if (setsockopt(sock, SOL_SOCKET, SO_REUSEADDR, &value, sizeof(value)) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set %s reuseaddr: %s", type, strerror(savederrno)); goto exit_error; } if (_fdset_cloexec(sock)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set %s CLOEXEC socket opts: %s", type, strerror(savederrno)); goto exit_error; } if (_fdset_nonblock(sock)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set %s NONBLOCK socket opts: %s", type, strerror(savederrno)); goto exit_error; } exit_error: errno = savederrno; return err; } int _init_socketpair(knet_handle_t knet_h, int *sock) { int err = 0, savederrno = 0; int i; if (socketpair(AF_UNIX, SOCK_SEQPACKET, 0, sock) != 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_HANDLE, "Unable to initialize socketpair: %s", strerror(savederrno)); goto exit_fail; } for (i = 0; i < 2; i++) { if (_configure_common_socket(knet_h, sock[i], "local socketpair") < 0) { savederrno = errno; err = -1; goto exit_fail; } } exit_fail: errno = savederrno; return err; } void _close_socketpair(knet_handle_t knet_h, int *sock) { int i; for (i = 0; i < 2; i++) { if (sock[i]) { close(sock[i]); sock[i] = 0; } } } /* * must be called with global read lock * * return -1 on error * return 0 if fd is invalid * return 1 if fd is valid */ int _is_valid_fd(knet_handle_t knet_h, int sockfd) { int ret = 0; if (sockfd < 0) { errno = EINVAL; return -1; } if (sockfd > KNET_MAX_FDS) { errno = EINVAL; return -1; } if (knet_h->knet_transport_fd_tracker[sockfd].transport >= KNET_MAX_TRANSPORTS) { ret = 0; } else { ret = 1; } return ret; } /* * must be called with global write lock */ int _set_fd_tracker(knet_handle_t knet_h, int sockfd, uint8_t transport, uint8_t data_type, void *data) { if (sockfd < 0) { errno = EINVAL; return -1; } if (sockfd > KNET_MAX_FDS) { errno = EINVAL; return -1; } knet_h->knet_transport_fd_tracker[sockfd].transport = transport; knet_h->knet_transport_fd_tracker[sockfd].data_type = data_type; knet_h->knet_transport_fd_tracker[sockfd].data = data; return 0; } diff --git a/libknet/transport_sctp.c b/libknet/transport_sctp.c index d0219952..699c401a 100644 --- a/libknet/transport_sctp.c +++ b/libknet/transport_sctp.c @@ -1,1315 +1,1322 @@ #include "config.h" #include #include #include #include -#include #include #include -#include +#include +#include "compat.h" #include "host.h" #include "link.h" #include "logging.h" #include "common.h" #include "transports.h" #include "threads_common.h" #ifdef HAVE_NETINET_SCTP_H #include /* * https://en.wikipedia.org/wiki/SCTP_packet_structure */ #define KNET_PMTUD_SCTP_OVERHEAD_COMMON 12 #define KNET_PMTUD_SCTP_OVERHEAD_DATA_CHUNK 16 #define KNET_PMTUD_SCTP_OVERHEAD KNET_PMTUD_SCTP_OVERHEAD_COMMON + KNET_PMTUD_SCTP_OVERHEAD_DATA_CHUNK /* * Time to sleep before reconnection attempts. in microseconds */ #define KNET_SCTP_SLEEP_TIME 1000000 /* * this value is per listener */ #define MAX_ACCEPTED_SOCKS 256 typedef struct sctp_handle_info { struct knet_list_head listen_links_list; struct knet_list_head connect_links_list; int connect_epollfd; int connectsockfd[2]; int listen_epollfd; int listensockfd[2]; pthread_t connect_thread; pthread_t listen_thread; } sctp_handle_info_t; /* * use by fd_tracker data type */ #define SCTP_NO_LINK_INFO 0 #define SCTP_LISTEN_LINK_INFO 1 #define SCTP_CONNECT_LINK_INFO 2 /* * this value is per listener */ #define MAX_ACCEPTED_SOCKS 256 typedef struct sctp_listen_link_info { struct knet_list_head list; int listen_sock; int accepted_socks[MAX_ACCEPTED_SOCKS]; struct sockaddr_storage src_address; int on_listener_epoll; int on_rx_epoll; } sctp_listen_link_info_t; typedef struct sctp_connect_link_info { struct knet_list_head list; sctp_listen_link_info_t *listener; struct knet_link *link; struct sockaddr_storage dst_address; int connect_sock; int on_connected_epoll; int on_rx_epoll; int close_sock; } sctp_connect_link_info_t; /* * socket handling functions * * those functions do NOT perform locking. locking * should be handled in the right context from callers */ /* * sockets are removed from rx_epoll from callers * see also error handling functions */ static int _close_connect_socket(knet_handle_t knet_h, struct knet_link *link) { int err = 0, savederrno = 0; sctp_connect_link_info_t *info = link->transport_link; sctp_handle_info_t *handle_info = knet_h->transports[KNET_TRANSPORT_SCTP]; struct epoll_event ev; if (info->on_connected_epoll) { memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLOUT; ev.data.fd = info->connect_sock; if (epoll_ctl(handle_info->connect_epollfd, EPOLL_CTL_DEL, info->connect_sock, &ev)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to remove connected socket from the epoll pool: %s", strerror(errno)); goto exit_error; } info->on_connected_epoll = 0; } exit_error: if (info->connect_sock != -1) { if (_set_fd_tracker(knet_h, info->connect_sock, KNET_MAX_TRANSPORTS, SCTP_NO_LINK_INFO, NULL) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to set fd tracker: %s", strerror(savederrno)); goto exit_error; } close(info->connect_sock); info->connect_sock = -1; } errno = savederrno; return err; } static int _enable_sctp_notifications(knet_handle_t knet_h, int sock, const char *type) { int err = 0, savederrno = 0; struct sctp_event_subscribe events; memset(&events, 0, sizeof (events)); events.sctp_data_io_event = 1; events.sctp_association_event = 1; events.sctp_send_failure_event = 1; events.sctp_address_event = 1; events.sctp_peer_error_event = 1; events.sctp_shutdown_event = 1; if (setsockopt(sock, IPPROTO_SCTP, SCTP_EVENTS, &events, sizeof (events)) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to enable %s events: %s", type, strerror(savederrno)); } errno = savederrno; return err; } static int _configure_sctp_socket(knet_handle_t knet_h, int sock, struct sockaddr_storage *address, const char *type) { int err = 0, savederrno = 0; int value; + int level; + +#ifdef SOL_SCTP + level = SOL_SCTP; +#else + level = IPPROTO_SCTP; +#endif if (_configure_transport_socket(knet_h, sock, address, type) < 0) { savederrno = errno; err = -1; goto exit_error; } value = 1; - if (setsockopt(sock, SOL_SCTP, SCTP_NODELAY, &value, sizeof(value)) < 0) { + if (setsockopt(sock, level, SCTP_NODELAY, &value, sizeof(value)) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set sctp nodelay: %s", strerror(savederrno)); goto exit_error; } value = 1; - if (setsockopt(sock, SOL_SCTP, SCTP_DISABLE_FRAGMENTS, &value, sizeof(value)) < 0) { + if (setsockopt(sock, level, SCTP_DISABLE_FRAGMENTS, &value, sizeof(value)) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set sctp disable fragments: %s", strerror(savederrno)); goto exit_error; } if (_enable_sctp_notifications(knet_h, sock, type) < 0) { savederrno = errno; err = -1; } exit_error: errno = savederrno; return err; } static int _reconnect_socket(knet_handle_t knet_h, struct knet_link *link) { int err = 0, savederrno = 0; sctp_connect_link_info_t *info = link->transport_link; sctp_handle_info_t *handle_info = knet_h->transports[KNET_TRANSPORT_SCTP]; struct epoll_event ev; - if (connect(info->connect_sock, (struct sockaddr *)&link->dst_addr, sizeof(struct sockaddr_storage)) < 0) { + if (connect(info->connect_sock, (struct sockaddr *)&link->dst_addr, sockaddr_len(&link->dst_addr)) < 0) { if ((errno != EALREADY) && (errno != EINPROGRESS) && (errno != EISCONN)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to connect SCTP socket %d: %s", info->connect_sock, strerror(savederrno)); goto exit_error; } } if (!info->on_connected_epoll) { memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLOUT; ev.data.fd = info->connect_sock; if (epoll_ctl(handle_info->connect_epollfd, EPOLL_CTL_ADD, info->connect_sock, &ev)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to add send/recv to epoll pool: %s", strerror(savederrno)); goto exit_error; } info->on_connected_epoll = 1; } exit_error: errno = savederrno; return err; } static int _create_connect_socket(knet_handle_t knet_h, struct knet_link *link) { int err = 0, savederrno = 0; sctp_connect_link_info_t *info = link->transport_link; sctp_handle_info_t *handle_info = knet_h->transports[KNET_TRANSPORT_SCTP]; struct epoll_event ev; int connect_sock; connect_sock = socket(link->dst_addr.ss_family, SOCK_STREAM, IPPROTO_SCTP); if (connect_sock < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to create send/recv socket: %s", strerror(savederrno)); goto exit_error; } if (_configure_sctp_socket(knet_h, connect_sock, &link->dst_addr, "SCTP connect") < 0) { savederrno = errno; err = -1; goto exit_error; } if (_set_fd_tracker(knet_h, connect_sock, KNET_TRANSPORT_SCTP, SCTP_CONNECT_LINK_INFO, info) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to set fd tracker: %s", strerror(savederrno)); goto exit_error; } info->connect_sock = connect_sock; info->close_sock = 0; if (_reconnect_socket(knet_h, link) < 0) { savederrno = errno; err = -1; goto exit_error; } exit_error: if (err) { if (info->on_connected_epoll) { epoll_ctl(handle_info->connect_epollfd, EPOLL_CTL_DEL, connect_sock, &ev); } if (connect_sock >= 0) { close(connect_sock); } } errno = savederrno; return err; } /* * socket error management functions * * both called with global read lock. * * NOTE: we need to remove the fd from the epoll as soon as possible * even before we notify the respective thread to take care of it * because scheduling can make it so that this thread will overload * and the threads supposed to take care of the error will never * be able to take action. * we CANNOT handle FDs here diretly (close/reconnect/etc) due * to locking context. We need to delegate that to their respective * management threads within global write lock. * * this function is called from: * - RX thread with recv_err <= 0 directly on recvmmsg error * - transport_rx_is_data when msg_len == 0 (recv_err = 1) * - transport_rx_is_data on notification (recv_err = 2) * * basically this small abouse of recv_err is to detect notifications * generated by sockets created by listen(). */ static int sctp_transport_rx_sock_error(knet_handle_t knet_h, int sockfd, int recv_err, int recv_errno) { struct epoll_event ev; sctp_connect_link_info_t *connect_info = knet_h->knet_transport_fd_tracker[sockfd].data; sctp_listen_link_info_t *listen_info = knet_h->knet_transport_fd_tracker[sockfd].data; sctp_handle_info_t *handle_info = knet_h->transports[KNET_TRANSPORT_SCTP]; switch (knet_h->knet_transport_fd_tracker[sockfd].data_type) { case SCTP_CONNECT_LINK_INFO: /* * all connect link have notifications enabled * and we accept only data from notification and * generic recvmmsg errors. * * Errors generated by msg_len 0 can be ignored because * they follow a notification (double notification) */ if (recv_err != 1) { connect_info->link->transport_connected = 0; if (connect_info->on_rx_epoll) { memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = sockfd; if (epoll_ctl(knet_h->recv_from_links_epollfd, EPOLL_CTL_DEL, sockfd, &ev)) { log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to remove EOFed socket from epoll pool: %s", strerror(errno)); return -1; } connect_info->on_rx_epoll = 0; } log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Notifying connect thread that sockfd %d received an error", sockfd); if (sendto(handle_info->connectsockfd[1], &sockfd, sizeof(int), MSG_DONTWAIT | MSG_NOSIGNAL, NULL, 0) != sizeof(int)) { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to notify connect thread: %s", strerror(errno)); } } break; case SCTP_LISTEN_LINK_INFO: if (listen_info->listen_sock != sockfd) { if (recv_err != 1) { if (listen_info->on_rx_epoll) { memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = sockfd; if (epoll_ctl(knet_h->recv_from_links_epollfd, EPOLL_CTL_DEL, sockfd, &ev)) { log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to remove EOFed socket from epoll pool: %s", strerror(errno)); return -1; } listen_info->on_rx_epoll = 0; } log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Notifying listen thread that sockfd %d received an error", sockfd); if (sendto(handle_info->listensockfd[1], &sockfd, sizeof(int), MSG_DONTWAIT | MSG_NOSIGNAL, NULL, 0) != sizeof(int)) { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to notify listen thread: %s", strerror(errno)); } } } else { /* * this means the listen() socket has generated * a notification. now what? :-) */ log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Received stray notification for listen() socket %d", sockfd); } break; default: break; } return 0; } /* * NOTE: sctp_transport_rx_is_data is called with global rdlock * delegate any FD error management to sctp_transport_rx_sock_error * and keep this code to parsing incoming data only */ static int sctp_transport_rx_is_data(knet_handle_t knet_h, int sockfd, struct mmsghdr msg) { int i; struct iovec *iov = msg.msg_hdr.msg_iov; size_t iovlen = msg.msg_hdr.msg_iovlen; struct sctp_assoc_change *sac; union sctp_notification *snp; if (!(msg.msg_hdr.msg_flags & MSG_NOTIFICATION)) { if (msg.msg_len == 0) { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "received 0 bytes len packet: %d", sockfd); /* * NOTE: with event notification enabled, we receive error twice: * 1) from the event notification * 2) followed by a 0 byte msg_len * * This is generally not a problem if not for causing extra * handling for the same issue. Should we drop notifications * and keep the code generic (handle all errors via msg_len = 0) * or keep the duplication as safety measure, or drop msg_len = 0 * handling (what about sockets without events enabled?) */ sctp_transport_rx_sock_error(knet_h, sockfd, 1, 0); return 1; } return 2; } if (!(msg.msg_hdr.msg_flags & MSG_EOR)) { return 1; } for (i=0; i< iovlen; i++) { snp = iov[i].iov_base; switch (snp->sn_header.sn_type) { case SCTP_ASSOC_CHANGE: log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "[event] sctp assoc change"); sac = &snp->sn_assoc_change; if (sac->sac_state == SCTP_COMM_LOST) { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "[event] sctp assoc change: comm_lost"); sctp_transport_rx_sock_error(knet_h, sockfd, 2, 0); } break; case SCTP_SHUTDOWN_EVENT: log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "[event] sctp shutdown event"); sctp_transport_rx_sock_error(knet_h, sockfd, 2, 0); break; case SCTP_SEND_FAILED: log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "[event] sctp send failed"); break; case SCTP_PEER_ADDR_CHANGE: log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "[event] sctp peer addr change"); break; case SCTP_REMOTE_ERROR: log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "[event] sctp remote error"); break; default: log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "[event] unknown sctp event type: %hu\n", snp->sn_header.sn_type); break; } } return 0; } /* * connect / outgoing socket management thread */ /* * _handle_connected_sctp* are called with a global write lock * from the connect_thread */ static void _handle_connected_sctp(knet_handle_t knet_h, int connect_sock) { int err; struct epoll_event ev; unsigned int status, len = sizeof(status); sctp_handle_info_t *handle_info = knet_h->transports[KNET_TRANSPORT_SCTP]; sctp_connect_link_info_t *info = knet_h->knet_transport_fd_tracker[connect_sock].data; struct knet_link *link = info->link; err = getsockopt(connect_sock, SOL_SOCKET, SO_ERROR, &status, &len); if (err) { log_err(knet_h, KNET_SUB_TRANSP_SCTP, "SCTP getsockopt() on connecting socket %d failed: %s", connect_sock, strerror(errno)); return; } if (info->close_sock) { if (_close_connect_socket(knet_h, link) < 0) { log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to close sock %d from _handle_connected_sctp: %s", connect_sock, strerror(errno)); return; } info->close_sock = 0; if (_create_connect_socket(knet_h, link) < 0) { log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to recreate connecting sock! %s", strerror(errno)); return; } } if (status) { log_info(knet_h, KNET_SUB_TRANSP_SCTP, "SCTP connect on %d to %s port %s failed: %s", connect_sock, link->status.dst_ipaddr, link->status.dst_port, strerror(status)); /* * No need to create a new socket if connect failed, * just retry connect */ _reconnect_socket(knet_h, info->link); return; } /* * Connected - Remove us from the connect epoll */ memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLOUT; ev.data.fd = connect_sock; if (epoll_ctl(handle_info->connect_epollfd, EPOLL_CTL_DEL, connect_sock, &ev)) { log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to remove connected socket %d from epoll pool: %s", connect_sock, strerror(errno)); } info->on_connected_epoll = 0; link->transport_connected = 1; link->outsock = info->connect_sock; memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = connect_sock; if (epoll_ctl(knet_h->recv_from_links_epollfd, EPOLL_CTL_ADD, connect_sock, &ev)) { log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to add connected socket to epoll pool: %s", strerror(errno)); } info->on_rx_epoll = 1; log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "SCTP handler fd %d now connected to %s port %s", connect_sock, link->status.dst_ipaddr, link->status.dst_port); } static void _handle_connected_sctp_errors(knet_handle_t knet_h) { int sockfd = -1; sctp_handle_info_t *handle_info = knet_h->transports[KNET_TRANSPORT_SCTP]; sctp_connect_link_info_t *info; if (recv(handle_info->connectsockfd[0], &sockfd, sizeof(int), MSG_DONTWAIT | MSG_NOSIGNAL) != sizeof(int)) { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Short read on connectsockfd"); return; } if (_is_valid_fd(knet_h, sockfd) < 1) { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Received stray notification for connected socket fd error"); return; } log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Processing connected error on socket: %d", sockfd); info = knet_h->knet_transport_fd_tracker[sockfd].data; info->close_sock = 1; info->link->transport_connected = 0; _reconnect_socket(knet_h, info->link); } static void *_sctp_connect_thread(void *data) { int savederrno; int i, nev; knet_handle_t knet_h = (knet_handle_t) data; sctp_handle_info_t *handle_info = knet_h->transports[KNET_TRANSPORT_SCTP]; struct epoll_event events[KNET_EPOLL_MAX_EVENTS]; while (!shutdown_in_progress(knet_h)) { nev = epoll_wait(handle_info->connect_epollfd, events, KNET_EPOLL_MAX_EVENTS, -1); if (nev < 0) { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "SCTP connect handler EPOLL ERROR: %s", strerror(errno)); continue; } /* * Sort out which FD has a connection */ savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to get write lock: %s", strerror(savederrno)); continue; } /* * minor optimization: deduplicate events * * in some cases we can receive multiple notifcations * of the same FD having issues or need handling. * It's enough to process it once even tho it's safe * to handle them multiple times. */ for (i = 0; i < nev; i++) { if (events[i].data.fd == handle_info->connectsockfd[0]) { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Received notification from rx_error for connected socket"); _handle_connected_sctp_errors(knet_h); } else { if (_is_valid_fd(knet_h, events[i].data.fd) == 1) { _handle_connected_sctp(knet_h, events[i].data.fd); } else { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Received stray notification for dead fd %d\n", events[i].data.fd); } } } pthread_rwlock_unlock(&knet_h->global_rwlock); /* * this thread can generate events for itself. * we need to sleep in between loops to allow other threads * to be scheduled */ usleep(KNET_SCTP_SLEEP_TIME); } return NULL; } /* * listen/incoming connections management thread */ /* * Listener received a new connection * called with a write lock from main thread */ static void _handle_incoming_sctp(knet_handle_t knet_h, int listen_sock) { int err = 0, savederrno = 0; int new_fd; int i = -1; sctp_listen_link_info_t *info = knet_h->knet_transport_fd_tracker[listen_sock].data; struct epoll_event ev; struct sockaddr_storage ss; socklen_t sock_len = sizeof(ss); char addr_str[KNET_MAX_HOST_LEN]; char port_str[KNET_MAX_PORT_LEN]; new_fd = accept(listen_sock, (struct sockaddr *)&ss, &sock_len); if (new_fd < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Incoming: accept error: %s", strerror(errno)); goto exit_error; } if (knet_addrtostr(&ss, sizeof(ss), addr_str, KNET_MAX_HOST_LEN, port_str, KNET_MAX_PORT_LEN) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Incoming: unable to gather socket info"); goto exit_error; } log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Incoming: received connection from: %s port: %s", addr_str, port_str); /* * Keep a track of all accepted FDs */ for (i=0; iaccepted_socks[i] == -1) { info->accepted_socks[i] = new_fd; break; } } if (i == MAX_ACCEPTED_SOCKS) { errno = EBUSY; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Incoming: too many connections!"); goto exit_error; } if (_configure_common_socket(knet_h, new_fd, "SCTP incoming") < 0) { savederrno = errno; err = -1; goto exit_error; } if (_enable_sctp_notifications(knet_h, new_fd, "Incoming connection") < 0) { savederrno = errno; err = -1; goto exit_error; } if (_set_fd_tracker(knet_h, new_fd, KNET_TRANSPORT_SCTP, SCTP_LISTEN_LINK_INFO, info) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to set fd tracker: %s", strerror(errno)); goto exit_error; } memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = new_fd; if (epoll_ctl(knet_h->recv_from_links_epollfd, EPOLL_CTL_ADD, new_fd, &ev)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Incoming: unable to add accepted socket %d to epoll pool: %s", new_fd, strerror(errno)); goto exit_error; } info->on_rx_epoll = 1; log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Incoming: accepted new fd %d for %s/%s (listen fd: %d). index: %d", new_fd, addr_str, port_str, info->listen_sock, i); exit_error: if (err) { if ((i >= 0) || (i < MAX_ACCEPTED_SOCKS)) { info->accepted_socks[i] = -1; } _set_fd_tracker(knet_h, new_fd, KNET_MAX_TRANSPORTS, SCTP_NO_LINK_INFO, NULL); close(new_fd); } errno = savederrno; return; } /* * Listen thread received a notification of a bad socket that needs closing * called with a write lock from main thread */ static void _handle_listen_sctp_errors(knet_handle_t knet_h) { int sockfd = -1; sctp_handle_info_t *handle_info = knet_h->transports[KNET_TRANSPORT_SCTP]; sctp_listen_link_info_t *info; int i; if (recv(handle_info->listensockfd[0], &sockfd, sizeof(int), MSG_DONTWAIT | MSG_NOSIGNAL) != sizeof(int)) { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Short read on listensockfd"); return; } if (_is_valid_fd(knet_h, sockfd) < 1) { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Received stray notification for listen socket fd error"); return; } log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Processing listen error on socket: %d", sockfd); info = knet_h->knet_transport_fd_tracker[sockfd].data; for (i=0; iaccepted_socks[i]) { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Closing accepted socket %d", sockfd); _set_fd_tracker(knet_h, sockfd, KNET_MAX_TRANSPORTS, SCTP_NO_LINK_INFO, NULL); info->accepted_socks[i] = -1; close(sockfd); } } } static void *_sctp_listen_thread(void *data) { int savederrno; int i, nev; knet_handle_t knet_h = (knet_handle_t) data; sctp_handle_info_t *handle_info = knet_h->transports[KNET_TRANSPORT_SCTP]; struct epoll_event events[KNET_EPOLL_MAX_EVENTS]; while (!shutdown_in_progress(knet_h)) { nev = epoll_wait(handle_info->listen_epollfd, events, KNET_EPOLL_MAX_EVENTS, -1); if (nev < 0) { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "SCTP listen handler EPOLL ERROR: %s", strerror(errno)); continue; } savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to get write lock: %s", strerror(savederrno)); continue; } /* * Sort out which FD has an incoming connection */ for (i = 0; i < nev; i++) { if (events[i].data.fd == handle_info->listensockfd[0]) { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Received notification from rx_error for listener/accepted socket"); _handle_listen_sctp_errors(knet_h); } else { if (_is_valid_fd(knet_h, events[i].data.fd) == 1) { _handle_incoming_sctp(knet_h, events[i].data.fd); } else { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Received listen notification from invalid socket"); } } } pthread_rwlock_unlock(&knet_h->global_rwlock); } return NULL; } /* * sctp_link_listener_start/stop are called in global write lock * context from set_config and clear_config. */ static sctp_listen_link_info_t *sctp_link_listener_start(knet_handle_t knet_h, struct knet_link *link) { int err = 0, savederrno = 0; int listen_sock = -1; struct epoll_event ev; sctp_listen_link_info_t *info = NULL; sctp_handle_info_t *handle_info = knet_h->transports[KNET_TRANSPORT_SCTP]; /* * Only allocate a new listener if src address is different */ knet_list_for_each_entry(info, &handle_info->listen_links_list, list) { if (memcmp(&info->src_address, &link->src_addr, sizeof(struct sockaddr_storage)) == 0) { return info; } } info = malloc(sizeof(sctp_listen_link_info_t)); if (!info) { err = -1; goto exit_error; } memset(info, 0, sizeof(sctp_listen_link_info_t)); memset(info->accepted_socks, -1, sizeof(info->accepted_socks)); memcpy(&info->src_address, &link->src_addr, sizeof(struct sockaddr_storage)); listen_sock = socket(link->src_addr.ss_family, SOCK_STREAM, IPPROTO_SCTP); if (listen_sock < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to create listener socket: %s", strerror(savederrno)); goto exit_error; } if (_configure_sctp_socket(knet_h, listen_sock, &link->src_addr, "SCTP listener") < 0) { savederrno = errno; err = -1; goto exit_error; } - if (bind(listen_sock, (struct sockaddr *)&link->src_addr, sizeof(struct sockaddr_storage)) < 0) { + if (bind(listen_sock, (struct sockaddr *)&link->src_addr, sockaddr_len(&link->src_addr)) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to bind listener socket: %s", strerror(savederrno)); goto exit_error; } if (listen(listen_sock, 5) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to listen on listener socket: %s", strerror(savederrno)); goto exit_error; } if (_set_fd_tracker(knet_h, listen_sock, KNET_TRANSPORT_SCTP, SCTP_LISTEN_LINK_INFO, info) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to set fd tracker: %s", strerror(savederrno)); goto exit_error; } memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = listen_sock; if (epoll_ctl(handle_info->listen_epollfd, EPOLL_CTL_ADD, listen_sock, &ev)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to add listener to epoll pool: %s", strerror(savederrno)); goto exit_error; } info->on_listener_epoll = 1; info->listen_sock = listen_sock; knet_list_add(&info->list, &handle_info->listen_links_list); log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Listening on fd %d for %s:%s", listen_sock, link->status.src_ipaddr, link->status.src_port); exit_error: if (err) { if (info->on_listener_epoll) { epoll_ctl(handle_info->listen_epollfd, EPOLL_CTL_DEL, listen_sock, &ev); } if (listen_sock >= 0) { close(listen_sock); } if (info) { free(info); info = NULL; } } errno = savederrno; return info; } static int sctp_link_listener_stop(knet_handle_t knet_h, struct knet_link *link) { int err = 0, savederrno = 0; int found = 0, i; struct knet_host *host; int link_idx; sctp_handle_info_t *handle_info = knet_h->transports[KNET_TRANSPORT_SCTP]; sctp_connect_link_info_t *this_link_info = link->transport_link; sctp_listen_link_info_t *info = this_link_info->listener; sctp_connect_link_info_t *link_info; struct epoll_event ev; for (host = knet_h->host_head; host != NULL; host = host->next) { for (link_idx = 0; link_idx < KNET_MAX_LINK; link_idx++) { if (&host->link[link_idx] == link) continue; link_info = host->link[link_idx].transport_link; if ((link_info) && (link_info->listener == info) && (host->link[link_idx].status.enabled == 1)) { found = 1; break; } } } if (found) { this_link_info->listener = NULL; log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "SCTP listener socket %d still in use", info->listen_sock); savederrno = EBUSY; err = -1; goto exit_error; } if (info->on_listener_epoll) { memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = info->listen_sock; if (epoll_ctl(handle_info->listen_epollfd, EPOLL_CTL_DEL, info->listen_sock, &ev)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to remove listener to epoll pool: %s", strerror(savederrno)); goto exit_error; } info->on_listener_epoll = 0; } if (_set_fd_tracker(knet_h, info->listen_sock, KNET_MAX_TRANSPORTS, SCTP_NO_LINK_INFO, NULL) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to set fd tracker: %s", strerror(savederrno)); goto exit_error; } close(info->listen_sock); for (i=0; i< MAX_ACCEPTED_SOCKS; i++) { if (info->accepted_socks[i] > -1) { memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = info->accepted_socks[i]; if (epoll_ctl(knet_h->recv_from_links_epollfd, EPOLL_CTL_DEL, info->accepted_socks[i], &ev)) { log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to remove EOFed socket from epoll pool: %s", strerror(errno)); } info->on_rx_epoll = 0; close(info->accepted_socks[i]); if (_set_fd_tracker(knet_h, info->accepted_socks[i], KNET_MAX_TRANSPORTS, SCTP_NO_LINK_INFO, NULL) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to set fd tracker: %s", strerror(savederrno)); goto exit_error; } info->accepted_socks[i] = -1; } } knet_list_del(&info->list); free(info); this_link_info->listener = NULL; exit_error: errno = savederrno; return err; } /* * Links config/clear. Both called with global wrlock from link_set_config/clear_config */ static int sctp_transport_link_set_config(knet_handle_t knet_h, struct knet_link *link) { int savederrno = 0, err = 0; sctp_connect_link_info_t *info; sctp_handle_info_t *handle_info = knet_h->transports[KNET_TRANSPORT_SCTP]; info = malloc(sizeof(sctp_connect_link_info_t)); if (!info) { goto exit_error; } memset(info, 0, sizeof(sctp_connect_link_info_t)); link->transport_link = info; info->link = link; memcpy(&info->dst_address, &link->dst_addr, sizeof(struct sockaddr_storage)); info->on_connected_epoll = 0; info->connect_sock = -1; info->listener = sctp_link_listener_start(knet_h, link); if (!info->listener) { savederrno = errno; err = -1; goto exit_error; } if (_create_connect_socket(knet_h, link) < 0) { savederrno = errno; err = -1; goto exit_error; } knet_list_add(&info->list, &handle_info->connect_links_list); link->outsock = info->connect_sock; exit_error: if (err) { if (info) { if (info->connect_sock) { close(info->connect_sock); } if (info->listener) { sctp_link_listener_stop(knet_h, link); } link->transport_link = NULL; free(info); } } errno = savederrno; return err; } /* * called with global wrlock * FIX exit path error handling */ static int sctp_transport_link_clear_config(knet_handle_t knet_h, struct knet_link *link) { int err = 0, savederrno = 0; sctp_connect_link_info_t *info; struct epoll_event ev; if (!link) { errno = EINVAL; return -1; } info = link->transport_link; if (!info) { errno = EINVAL; return -1; } if ((sctp_link_listener_stop(knet_h, link) <0) && (errno != EBUSY)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to remove listener trasport: %s", strerror(savederrno)); goto exit_error; } if (info->on_rx_epoll) { memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = info->connect_sock; if (epoll_ctl(knet_h->recv_from_links_epollfd, EPOLL_CTL_DEL, info->connect_sock, &ev)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to remove connected socket from epoll pool: %s", strerror(savederrno)); goto exit_error; } info->on_rx_epoll = 0; } if (_close_connect_socket(knet_h, link) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to close connected socket: %s", strerror(savederrno)); goto exit_error; } knet_list_del(&info->list); free(info); link->transport_link = NULL; exit_error: errno = savederrno; return err; } /* * transport_free and transport_init are * called only from knet_handle_new and knet_handle_free. * all resources (hosts/links) should have been already freed at this point * and they are called in a write locked context, hence they * don't need their own locking. */ static int sctp_transport_free(knet_handle_t knet_h) { sctp_handle_info_t *handle_info; void *thread_status; struct epoll_event ev; if (!knet_h->transports[KNET_TRANSPORT_SCTP]) { errno = EINVAL; return -1; } handle_info = knet_h->transports[KNET_TRANSPORT_SCTP]; /* * keep it here while we debug list usage and such */ if (!knet_list_empty(&handle_info->listen_links_list)) { log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Internal error. listen links list is not empty"); } if (!knet_list_empty(&handle_info->connect_links_list)) { log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Internal error. connect links list is not empty"); } if (handle_info->listen_thread) { pthread_cancel(handle_info->listen_thread); pthread_join(handle_info->listen_thread, &thread_status); } if (handle_info->connect_thread) { pthread_cancel(handle_info->connect_thread); pthread_join(handle_info->connect_thread, &thread_status); } if (handle_info->listensockfd[0] >= 0) { memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = handle_info->listensockfd[0]; epoll_ctl(handle_info->listen_epollfd, EPOLL_CTL_DEL, handle_info->listensockfd[0], &ev); } if (handle_info->connectsockfd[0] >= 0) { memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = handle_info->connectsockfd[0]; epoll_ctl(handle_info->connect_epollfd, EPOLL_CTL_DEL, handle_info->connectsockfd[0], &ev); } _close_socketpair(knet_h, handle_info->connectsockfd); _close_socketpair(knet_h, handle_info->listensockfd); if (handle_info->listen_epollfd >= 0) { close(handle_info->listen_epollfd); } if (handle_info->connect_epollfd >= 0) { close(handle_info->connect_epollfd); } free(handle_info); knet_h->transports[KNET_TRANSPORT_SCTP] = NULL; return 0; } static int sctp_transport_init(knet_handle_t knet_h) { int err = 0, savederrno = 0; sctp_handle_info_t *handle_info; struct epoll_event ev; if (knet_h->transports[KNET_TRANSPORT_SCTP]) { errno = EEXIST; return -1; } handle_info = malloc(sizeof(sctp_handle_info_t)); if (!handle_info) { return -1; } memset(handle_info, 0,sizeof(sctp_handle_info_t)); knet_h->transports[KNET_TRANSPORT_SCTP] = handle_info; knet_list_init(&handle_info->listen_links_list); knet_list_init(&handle_info->connect_links_list); handle_info->listen_epollfd = epoll_create(KNET_EPOLL_MAX_EVENTS + 1); if (handle_info->listen_epollfd < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to create epoll listen fd: %s", strerror(savederrno)); goto exit_fail; } if (_fdset_cloexec(handle_info->listen_epollfd)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to set CLOEXEC on listen_epollfd: %s", strerror(savederrno)); goto exit_fail; } handle_info->connect_epollfd = epoll_create(KNET_EPOLL_MAX_EVENTS + 1); if (handle_info->connect_epollfd < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to create epoll connect fd: %s", strerror(savederrno)); goto exit_fail; } if (_fdset_cloexec(handle_info->connect_epollfd)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to set CLOEXEC on connect_epollfd: %s", strerror(savederrno)); goto exit_fail; } if (_init_socketpair(knet_h, handle_info->connectsockfd) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to init connect socketpair: %s", strerror(savederrno)); goto exit_fail; } memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = handle_info->connectsockfd[0]; if (epoll_ctl(handle_info->connect_epollfd, EPOLL_CTL_ADD, handle_info->connectsockfd[0], &ev)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to add connectsockfd[0] to connect epoll pool: %s", strerror(savederrno)); goto exit_fail; } if (_init_socketpair(knet_h, handle_info->listensockfd) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to init listen socketpair: %s", strerror(savederrno)); goto exit_fail; } memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = handle_info->listensockfd[0]; if (epoll_ctl(handle_info->listen_epollfd, EPOLL_CTL_ADD, handle_info->listensockfd[0], &ev)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to add listensockfd[0] to listen epoll pool: %s", strerror(savederrno)); goto exit_fail; } /* * Start connect & listener threads */ savederrno = pthread_create(&handle_info->listen_thread, 0, _sctp_listen_thread, (void *) knet_h); if (savederrno) { err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to start sctp listen thread: %s", strerror(savederrno)); goto exit_fail; } savederrno = pthread_create(&handle_info->connect_thread, 0, _sctp_connect_thread, (void *) knet_h); if (savederrno) { err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to start sctp connect thread: %s", strerror(savederrno)); goto exit_fail; } exit_fail: if (err < 0) { sctp_transport_free(knet_h); } errno = savederrno; return err; } static knet_transport_ops_t sctp_transport_ops = { .transport_name = "SCTP", .transport_mtu_overhead = KNET_PMTUD_SCTP_OVERHEAD, .transport_init = sctp_transport_init, .transport_free = sctp_transport_free, .transport_link_set_config = sctp_transport_link_set_config, .transport_link_clear_config = sctp_transport_link_clear_config, .transport_rx_sock_error = sctp_transport_rx_sock_error, .transport_rx_is_data = sctp_transport_rx_is_data, }; knet_transport_ops_t *get_sctp_transport() { return &sctp_transport_ops; } #else // HAVE_NETINET_SCTP_H knet_transport_ops_t *get_sctp_transport() { return NULL; } #endif diff --git a/libknet/transport_udp.c b/libknet/transport_udp.c index 3118b993..d1933518 100644 --- a/libknet/transport_udp.c +++ b/libknet/transport_udp.c @@ -1,267 +1,267 @@ #include "config.h" #include #include #include -#include #include #include -#include +#include #include "libknet.h" +#include "compat.h" #include "host.h" #include "link.h" #include "logging.h" #include "common.h" #include "transports.h" #define KNET_PMTUD_UDP_OVERHEAD 8 typedef struct udp_handle_info { struct knet_list_head links_list; } udp_handle_info_t; typedef struct udp_link_info { struct knet_list_head list; struct sockaddr_storage local_address; int socket_fd; int on_epoll; } udp_link_info_t; static int udp_transport_link_set_config(knet_handle_t knet_h, struct knet_link *link) { int err = 0, savederrno = 0; int sock = -1; struct epoll_event ev; udp_link_info_t *info; udp_handle_info_t *handle_info = knet_h->transports[KNET_TRANSPORT_UDP]; /* * Only allocate a new link if the local address is different */ knet_list_for_each_entry(info, &handle_info->links_list, list) { if (memcmp(&info->local_address, &link->src_addr, sizeof(struct sockaddr_storage)) == 0) { log_debug(knet_h, KNET_SUB_TRANSP_UDP, "Re-using existing UDP socket for new link"); link->outsock = info->socket_fd; link->transport_link = info; link->transport_connected = 1; return 0; } } info = malloc(sizeof(udp_link_info_t)); if (!info) { err = -1; goto exit_error; } sock = socket(link->src_addr.ss_family, SOCK_DGRAM, 0); if (sock < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_LISTENER, "Unable to create listener socket: %s", strerror(savederrno)); goto exit_error; } if (_configure_transport_socket(knet_h, sock, &link->src_addr, "UDP") < 0) { savederrno = errno; err = -1; goto exit_error; } - if (bind(sock, (struct sockaddr *)&link->src_addr, sizeof(struct sockaddr_storage)) < 0) { + if (bind(sock, (struct sockaddr *)&link->src_addr, sockaddr_len(&link->src_addr))) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_UDP, "Unable to bind listener socket: %s", strerror(savederrno)); goto exit_error; } memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = sock; if (epoll_ctl(knet_h->recv_from_links_epollfd, EPOLL_CTL_ADD, sock, &ev)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_UDP, "Unable to add listener to epoll pool: %s", strerror(savederrno)); goto exit_error; } info->on_epoll = 1; if (_set_fd_tracker(knet_h, sock, KNET_TRANSPORT_UDP, 0, info) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_UDP, "Unable to set fd tracker: %s", strerror(savederrno)); goto exit_error; } memcpy(&info->local_address, &link->src_addr, sizeof(struct sockaddr_storage)); info->socket_fd = sock; knet_list_add(&info->list, &handle_info->links_list); link->outsock = sock; link->transport_link = info; link->transport_connected = 1; exit_error: if (err) { if (info) { if (info->on_epoll) { epoll_ctl(knet_h->recv_from_links_epollfd, EPOLL_CTL_DEL, sock, &ev); } free(info); } if (sock >= 0) { close(sock); } } errno = savederrno; return err; } static int udp_transport_link_clear_config(knet_handle_t knet_h, struct knet_link *link) { int err = 0, savederrno = 0; int found = 0; struct knet_host *host; int link_idx; udp_link_info_t *info = link->transport_link; struct epoll_event ev; for (host = knet_h->host_head; host != NULL; host = host->next) { for (link_idx = 0; link_idx < KNET_MAX_LINK; link_idx++) { if (&host->link[link_idx] == link) continue; if ((host->link[link_idx].transport_link == info) && (host->link[link_idx].status.enabled == 1)) { found = 1; break; } } } if (found) { log_debug(knet_h, KNET_SUB_TRANSP_UDP, "UDP socket %d still in use", info->socket_fd); savederrno = EBUSY; err = -1; goto exit_error; } if (info->on_epoll) { memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = info->socket_fd; if (epoll_ctl(knet_h->recv_from_links_epollfd, EPOLL_CTL_DEL, info->socket_fd, &ev) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_UDP, "Unable to remove UDP socket from epoll poll: %s", strerror(errno)); goto exit_error; } info->on_epoll = 0; } if (_set_fd_tracker(knet_h, info->socket_fd, KNET_MAX_TRANSPORTS, 0, NULL) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_UDP, "Unable to set fd tracker: %s", strerror(savederrno)); goto exit_error; } close(info->socket_fd); knet_list_del(&info->list); free(link->transport_link); exit_error: errno = savederrno; return err; } static int udp_transport_free(knet_handle_t knet_h) { udp_handle_info_t *handle_info; if (!knet_h->transports[KNET_TRANSPORT_UDP]) { errno = EINVAL; return -1; } handle_info = knet_h->transports[KNET_TRANSPORT_UDP]; /* * keep it here while we debug list usage and such */ if (!knet_list_empty(&handle_info->links_list)) { log_err(knet_h, KNET_SUB_TRANSP_UDP, "Internal error. handle list is not empty"); return -1; } free(handle_info); knet_h->transports[KNET_TRANSPORT_UDP] = NULL; return 0; } static int udp_transport_init(knet_handle_t knet_h) { udp_handle_info_t *handle_info; if (knet_h->transports[KNET_TRANSPORT_UDP]) { errno = EEXIST; return -1; } handle_info = malloc(sizeof(udp_handle_info_t)); if (!handle_info) { return -1; } knet_h->transports[KNET_TRANSPORT_UDP] = handle_info; knet_list_init(&handle_info->links_list); return 0; } static int udp_transport_rx_sock_error(knet_handle_t knet_h, int sockfd, int recv_err, int recv_errno) { /* * UDP can't do much here afaict, perhaps clear the SO_ERROR? */ log_err(knet_h, KNET_SUB_TRANSP_UDP, "Sock: %d received error: %d (%s)", sockfd, recv_err, strerror(recv_errno)); return 0; } static int udp_transport_rx_is_data(knet_handle_t knet_h, int sockfd, struct mmsghdr msg) { if (msg.msg_len == 0) return 0; return 2; } static knet_transport_ops_t udp_transport_ops = { .transport_name = "UDP", .transport_mtu_overhead = KNET_PMTUD_UDP_OVERHEAD, .transport_init = udp_transport_init, .transport_free = udp_transport_free, .transport_link_set_config = udp_transport_link_set_config, .transport_link_clear_config = udp_transport_link_clear_config, .transport_rx_sock_error = udp_transport_rx_sock_error, .transport_rx_is_data = udp_transport_rx_is_data, }; knet_transport_ops_t *get_udp_transport() { return &udp_transport_ops; }