diff --git a/configure.ac b/configure.ac index 8188f1cd..74c00e53 100644 --- a/configure.ac +++ b/configure.ac @@ -1,405 +1,400 @@ # # 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 ! ${MAKE-make} --version /cannot/make/this >/dev/null 2>&1; then AC_MSG_ERROR(["you don't seem to have GNU make; it is required"]) fi AC_PROG_AWK AC_PROG_GREP AC_PROG_SED AC_PROG_CPP AC_PROG_CC AM_PROG_CC_C_O AC_PROG_LN_S AC_PROG_INSTALL AC_PROG_MAKE_SET AC_PROG_CXX AC_PROG_RANLIB AC_CHECK_PROGS([PUBLICAN], [publican], [:]) AC_CHECK_PROGS([PKGCONFIG], [pkg-config]) AC_ARG_ENABLE([poc], [ --enable-poc : build poc code ],, [ enable_poc="yes" ]) AM_CONDITIONAL([BUILD_POC], test x$enable_poc = xyes) 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]) AC_CHECK_HEADERS([sys/epoll.h]) if test "x$enable_libknet_sctp" = xyes; then AC_CHECK_HEADERS([netinet/sctp.h],, AC_MSG_ERROR(["missing required SCTP headers"])) fi # Checks for typedefs, structures, and compiler characteristics. AC_C_INLINE AC_TYPE_SIZE_T AC_TYPE_PID_T AC_TYPE_SSIZE_T AC_TYPE_UINT8_T AC_TYPE_UINT16_T AC_TYPE_UINT32_T AC_TYPE_UINT64_T AC_TYPE_INT32_T # Checks for library functions. AC_FUNC_ALLOCA AC_FUNC_FORK AC_FUNC_MALLOC AC_FUNC_REALLOC AC_CHECK_FUNCS([memset]) AC_CHECK_FUNCS([strdup]) AC_CHECK_FUNCS([strerror]) AC_CHECK_FUNCS([dup2]) AC_CHECK_FUNCS([select]) AC_CHECK_FUNCS([socket]) AC_CHECK_FUNCS([inet_ntoa]) AC_CHECK_FUNCS([memmove]) AC_CHECK_FUNCS([strchr]) AC_CHECK_FUNCS([atexit]) AC_CHECK_FUNCS([ftruncate]) AC_CHECK_FUNCS([strrchr]) AC_CHECK_FUNCS([strstr]) AC_CHECK_FUNCS([clock_gettime]) AC_CHECK_FUNCS([strcasecmp]) AC_CHECK_FUNCS([sendmmsg]) AC_CHECK_FUNCS([recvmmsg]) AC_CHECK_FUNCS([kevent]) # if neither sys/epoll.h nor kevent are present, we should fail. if test "x$ac_cv_header_sys_epoll_h" = xno && test "x$ac_cv_func_kevent" = xno; then AC_MSG_ERROR([Both epoll and kevent unavailable on this OS]) fi if test "x$ac_cv_header_sys_epoll_h" = xyes && test "x$ac_cv_func_kevent" = xyes; then AC_MSG_ERROR([Both epoll and kevent available on this OS, please contact the maintainers to fix the code]) fi -# Check entries in specific structs -AC_CHECK_MEMBER([struct mmsghdr.msg_hdr], - [AC_DEFINE_UNQUOTED([HAVE_MMSGHDR], [1], [struct mmsghdr exists])], - [], [[#include ]]) - # checks (for kronosnetd) if test "x$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.h b/libknet/compat.h index 151580ec..05d6116b 100644 --- a/libknet/compat.h +++ b/libknet/compat.h @@ -1,80 +1,73 @@ /* * 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 /* FreeBSD has recvmmsg but it's a buggy wrapper */ #ifdef __FreeBSD__ #define recvmmsg COMPAT_recvmmsg #define sendmmsg COMPAT_sendmmsg #undef HAVE_RECVMMSG #undef HAVE_SENDMMSG #endif -#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 #ifdef HAVE_SYS_EPOLL_H #include #else #ifdef HAVE_KEVENT #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 /* HAVE_KEVENT */ #endif /* HAVE_SYS_EPOLL_H */ #endif /* __COMPAT_H__ */ diff --git a/libknet/internals.h b/libknet/internals.h index 7c3405e5..ddb06d61 100644 --- a/libknet/internals.h +++ b/libknet/internals.h @@ -1,462 +1,467 @@ /* * 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_mmsghdr { + struct msghdr msg_hdr; /* Message header */ + unsigned int msg_len; /* Number of bytes transmitted */ +}; + 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 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_DATABUFSIZE]; 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 */ knet_node_id_t host_id; /* configurable */ uint8_t link_handler_policy; char name[KNET_MAX_HOST_LEN]; /* status */ struct knet_host_status status; /* internals */ char circular_buffer[KNET_CBUFFER_SIZE]; seq_num_t rx_seq_num; seq_num_t untimed_rx_seq_num; seq_num_t timed_rx_seq_num; uint8_t got_data; /* defrag/reassembly buffers */ struct knet_host_defrag_buf defrag_buf[KNET_MAX_LINK]; char 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 { knet_node_id_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 */ knet_node_id_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_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; /* used to protect knet_send_sync and TX thread */ pthread_mutex_t hb_mutex; /* used to protect heartbeat thread and seq_num broadcasting */ 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 tx_seq_num; pthread_mutex_t tx_seq_num_mutex; 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, knet_node_id_t this_host_id, knet_node_id_t src_node_id, int8_t *channel, knet_node_id_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, knet_node_id_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; const uint8_t transport_id; 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); /* * transport callback for incoming dynamic connections * this is called in global read lock context */ int (*transport_link_dyn_connect)(knet_handle_t knet_h, int sockfd, 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_rdlock */ int (*transport_rx_sock_error)(knet_handle_t knet_h, int sockfd, int recv_err, int recv_errno); /* * transport_tx_sock_error is invoked with global_rwlock and * it's invoked when sendto or sendmmsg returns =< 0 * * it should return: * -1 on internal error * 0 ignore error and continue * 1 retry * any sleep or wait action should happen inside the transport code */ int (*transport_tx_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); + int (*transport_rx_is_data)(knet_handle_t knet_h, int sockfd, struct knet_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/threads_rx.c b/libknet/threads_rx.c index 93e8ba7f..d85bd429 100644 --- a/libknet/threads_rx.c +++ b/libknet/threads_rx.c @@ -1,727 +1,727 @@ /* * Copyright (C) 2010-2017 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 "compat.h" #include "crypto.h" #include "host.h" #include "link.h" #include "logging.h" #include "transports.h" #include "threads_common.h" #include "threads_heartbeat.h" #include "threads_rx.h" #include "netutils.h" /* * 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_seq_num, 1, 0)) { errno = ETIME; return -1; } /* * register the pckt as seen */ _seq_num_set(src_host, 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, int sockfd, const struct mmsghdr *msg) +static void _parse_recv_from_links(knet_handle_t knet_h, int sockfd, const struct knet_mmsghdr *msg) { int err = 0, savederrno = 0; ssize_t outlen; struct knet_host *src_host; struct knet_link *src_link; unsigned long long latency_last; knet_node_id_t dst_host_ids[KNET_MAX_HOST]; size_t dst_host_ids_entries = 0; int bcast = 1; struct timespec recvtime; struct knet_header *inbuf = msg->msg_hdr.msg_iov->iov_base; unsigned char *outbuf = (unsigned char *)msg->msg_hdr.msg_iov->iov_base; ssize_t len = msg->msg_len; struct knet_hostinfo *knet_hostinfo; struct iovec iov_out[1]; int8_t channel; struct sockaddr_storage pckt_src; seq_num_t recv_seq_num; int wipe_bufs = 0; 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) { /* * cpyaddrport will only copy address and port of the incoming * packet and strip extra bits such as flow and scopeid */ cpyaddrport(&pckt_src, msg->msg_hdr.msg_name); if (cmpaddr(&src_link->dst_addr, sockaddr_len(&src_link->dst_addr), &pckt_src, sockaddr_len(&pckt_src)) != 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, &pckt_src, sizeof(struct sockaddr_storage)); if (knet_addrtostr(&src_link->dst_addr, sockaddr_len(msg->msg_hdr.msg_name), 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, "??"); } else { log_info(knet_h, KNET_SUB_RX, "host: %u link: %u new connection established from: %s %s", src_host->host_id, src_link->link_id, src_link->status.dst_ipaddr, src_link->status.dst_port); } } /* * transport has already accepted the connection here * otherwise we would not be receiving packets */ knet_h->transport_ops[src_link->transport_type]->transport_link_dyn_connect(knet_h, sockfd, src_link); } } switch (inbuf->kh_type) { case KNET_HEADER_TYPE_HOST_INFO: case KNET_HEADER_TYPE_DATA: /* * TODO: should we accept data even if we can't reply to the other node? * how would that work with SCTP and guaranteed delivery? */ if (!src_host->status.reachable) { log_debug(knet_h, KNET_SUB_RX, "Source host %u not reachable yet", src_host->host_id); //return; } inbuf->khp_data_seq_num = ntohs(inbuf->khp_data_seq_num); channel = inbuf->khp_data_channel; src_host->got_data = 1; if (!_seq_num_lookup(src_host, inbuf->khp_data_seq_num, 0, 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, 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, inbuf->khp_data_seq_num, 0, 0)) { return; } _seq_num_set(src_host, inbuf->khp_data_seq_num, 0); switch(knet_hostinfo->khi_type) { case KNET_HOSTINFO_TYPE_LINK_UP_DOWN: 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); recv_seq_num = ntohs(inbuf->khp_ping_seq_num); wipe_bufs = 0; if (!inbuf->khp_ping_timed) { /* * we might be receiving this message from all links, but we want * to process it only the first time */ if (recv_seq_num != src_host->untimed_rx_seq_num) { /* * cache the untimed seq num */ src_host->untimed_rx_seq_num = recv_seq_num; /* * if the host has received data in between * untimed ping, then we don't need to wipe the bufs */ if (src_host->got_data) { src_host->got_data = 0; wipe_bufs = 0; } else { wipe_bufs = 1; } } _seq_num_lookup(src_host, recv_seq_num, 0, wipe_bufs); } else { /* * pings always arrives in bursts over all the link * catch the first of them to cache the seq num and * avoid duplicate processing */ if (recv_seq_num != src_host->timed_rx_seq_num) { src_host->timed_rx_seq_num = recv_seq_num; if (recv_seq_num == 0) { _seq_num_lookup(src_host, recv_seq_num, 0, 1); } } } 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; } retry_pong: len = sendto(src_link->outsock, outbuf, outlen, MSG_DONTWAIT | MSG_NOSIGNAL, (struct sockaddr *) &src_link->dst_addr, sizeof(struct sockaddr_storage)); savederrno = errno; if (len != outlen) { err = knet_h->transport_ops[src_link->transport_type]->transport_tx_sock_error(knet_h, src_link->outsock, len, savederrno); switch(err) { case -1: /* unrecoverable error */ 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 0: /* ignore error and continue */ break; case 1: /* retry to send those same data */ goto retry_pong; break; } } 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; } retry_pmtud: len = sendto(src_link->outsock, outbuf, outlen, MSG_DONTWAIT | MSG_NOSIGNAL, (struct sockaddr *) &src_link->dst_addr, sizeof(struct sockaddr_storage)); if (len != outlen) { err = knet_h->transport_ops[src_link->transport_type]->transport_tx_sock_error(knet_h, src_link->outsock, len, savederrno); switch(err) { case -1: /* unrecoverable error */ 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 0: /* ignore error and continue */ break; case 1: /* retry to send those same data */ goto retry_pmtud; break; } } 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) +static void _handle_recv_from_links(knet_handle_t knet_h, int sockfd, struct knet_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; /* * reset msg_namelen to buffer size because after recvmmsg * each msg_namelen will contain sizeof sockaddr_in or sockaddr_in6 */ for (i = 0; i < PCKT_FRAG_MAX; i++) { msg[i].msg_hdr.msg_namelen = sizeof(struct sockaddr_storage); } - msg_recv = recvmmsg(sockfd, msg, PCKT_FRAG_MAX, MSG_DONTWAIT | MSG_NOSIGNAL, NULL); + msg_recv = recvmmsg(sockfd, (struct mmsghdr *)&msg[0], 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) { 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, sockfd, &msg[i]); 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 knet_mmsghdr msg[PCKT_FRAG_MAX]; struct iovec iov_in[PCKT_FRAG_MAX]; memset(&msg, 0, sizeof(msg)); 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/threads_tx.c b/libknet/threads_tx.c index 147618d5..4cad4374 100644 --- a/libknet/threads_tx.c +++ b/libknet/threads_tx.c @@ -1,609 +1,609 @@ /* * Copyright (C) 2010-2017 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 "compat.h" #include "crypto.h" #include "host.h" #include "link.h" #include "logging.h" #include "transports.h" #include "threads_common.h" #include "threads_heartbeat.h" #include "threads_tx.h" #include "netutils.h" /* * SEND */ -static int _dispatch_to_links(knet_handle_t knet_h, struct knet_host *dst_host, struct mmsghdr *msg, int msgs_to_send) +static int _dispatch_to_links(knet_handle_t knet_h, struct knet_host *dst_host, struct knet_mmsghdr *msg, int msgs_to_send) { int link_idx, msg_idx, sent_msgs, prev_sent, progress; int err = 0, savederrno = 0; - struct mmsghdr *cur; + struct knet_mmsghdr *cur; for (link_idx = 0; link_idx < dst_host->active_link_entries; link_idx++) { sent_msgs = 0; prev_sent = 0; progress = 1; msg_idx = 0; while (msg_idx < msgs_to_send) { msg[msg_idx].msg_hdr.msg_name = &dst_host->link[dst_host->active_links[link_idx]].dst_addr; msg_idx++; } retry: cur = &msg[prev_sent]; sent_msgs = sendmmsg(dst_host->link[dst_host->active_links[link_idx]].outsock, - cur, msgs_to_send - prev_sent, MSG_DONTWAIT | MSG_NOSIGNAL); + (struct mmsghdr *)&cur[0], msgs_to_send - prev_sent, MSG_DONTWAIT | MSG_NOSIGNAL); savederrno = errno; err = knet_h->transport_ops[dst_host->link[dst_host->active_links[link_idx]].transport_type]->transport_tx_sock_error(knet_h, dst_host->link[dst_host->active_links[link_idx]].outsock, sent_msgs, savederrno); switch(err) { case -1: /* unrecoverable error */ goto out_unlock; break; case 0: /* ignore error and continue */ break; case 1: /* retry to send those same data */ goto retry; break; } prev_sent = prev_sent + sent_msgs; if ((sent_msgs >= 0) && (prev_sent < msgs_to_send)) { if ((sent_msgs) || (progress)) { if (sent_msgs) { progress = 1; } else { progress = 0; } #ifdef DEBUG 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); #endif goto retry; } if (!progress) { savederrno = EAGAIN; 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; knet_node_id_t dst_host_ids_temp[KNET_MAX_HOST]; size_t dst_host_ids_entries_temp = 0; knet_node_id_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; seq_num_t tx_seq_num; - struct mmsghdr msg[PCKT_FRAG_MAX]; + struct knet_mmsghdr msg[PCKT_FRAG_MAX]; int msgs_to_send, msg_idx; 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; if (pthread_mutex_lock(&knet_h->tx_seq_num_mutex)) { log_debug(knet_h, KNET_SUB_TX, "Unable to get seq mutex lock"); goto out_unlock; } knet_h->tx_seq_num++; /* * force seq_num 0 to detect a node that has crashed and rejoining * the knet instance. seq_num 0 will clear the buffers in the RX * thread */ if (knet_h->tx_seq_num == 0) { knet_h->tx_seq_num++; } /* * cache the value in locked context */ tx_seq_num = knet_h->tx_seq_num; inbuf->khp_data_seq_num = htons(knet_h->tx_seq_num); pthread_mutex_unlock(&knet_h->tx_seq_num_mutex); /* * forcefully broadcast a ping to all nodes every SEQ_MAX / 8 * pckts. * this solves 2 problems: * 1) on TX socket overloads we generate extra pings to keep links alive * 2) in 3+ nodes setup, where all the traffic is flowing between node 1 and 2, * node 3+ will be able to keep in sync on the TX seq_num even without * receiving traffic or pings in betweens. This avoids issues with * rollover of the circular buffer */ if (tx_seq_num % (SEQ_MAX / 8) == 0) { _send_pings(knet_h, 0); } if (inbuf->khp_data_frag_num > 1) { 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_seq_num = inbuf->khp_data_seq_num; 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++; } } else { iov_out[frag_idx].iov_base = (void *)inbuf; iov_out[frag_idx].iov_len = frag_len + KNET_HEADER_DATA_SIZE; } if (knet_h->crypto_instance) { frag_idx = 0; while (frag_idx < inbuf->khp_data_frag_num) { if (crypto_encrypt_and_sign( knet_h, (const unsigned char *)iov_out[frag_idx].iov_base, 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 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++; } } memset(&msg, 0, sizeof(msg)); msgs_to_send = inbuf->khp_data_frag_num; msg_idx = 0; while (msg_idx < msgs_to_send) { msg[msg_idx].msg_hdr.msg_namelen = sizeof(struct sockaddr_storage); msg[msg_idx].msg_hdr.msg_iov = &iov_out[msg_idx]; msg[msg_idx].msg_hdr.msg_iovlen = 1; msg_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]]; err = _dispatch_to_links(knet_h, dst_host, &msg[0], msgs_to_send); savederrno = errno; if (err) { goto out_unlock; } } } else { 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, &msg[0], msgs_to_send); savederrno = errno; if (err) { goto out_unlock; } } } } out_unlock: 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) +static void _handle_send_to_links(knet_handle_t knet_h, int sockfd, int8_t channel, struct knet_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); + msg_recv = recvmmsg(sockfd, (struct mmsghdr *)&msg[0], 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 knet_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)); + memset(&msg, 0, sizeof(msg)); /* 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[0], type); pthread_mutex_unlock(&knet_h->tx_mutex); } pthread_rwlock_unlock(&knet_h->global_rwlock); } return NULL; } diff --git a/libknet/transport_sctp.c b/libknet/transport_sctp.c index c8ec42b8..f30976e5 100644 --- a/libknet/transport_sctp.c +++ b/libknet/transport_sctp.c @@ -1,1438 +1,1438 @@ #include "config.h" #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 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_LISTENER_LINK_INFO 1 #define SCTP_ACCEPTED_LINK_INFO 2 #define SCTP_CONNECT_LINK_INFO 3 /* * 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_accepted_link_info { char mread_buf[KNET_DATABUFSIZE]; ssize_t mread_len; sctp_listen_link_info_t *link_info; } sctp_accepted_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, 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; } 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 *kn_link) { int err = 0, savederrno = 0; sctp_connect_link_info_t *info = kn_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 *)&kn_link->dst_addr, sockaddr_len(&kn_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 *kn_link) { int err = 0, savederrno = 0; sctp_connect_link_info_t *info = kn_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(kn_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, &kn_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, kn_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; } static int sctp_transport_tx_sock_error(knet_handle_t knet_h, int sockfd, int recv_err, int recv_errno) { sctp_connect_link_info_t *connect_info = knet_h->knet_transport_fd_tracker[sockfd].data; sctp_accepted_link_info_t *accepted_info = knet_h->knet_transport_fd_tracker[sockfd].data; sctp_listen_link_info_t *listen_info; if (recv_err < 0) { switch (knet_h->knet_transport_fd_tracker[sockfd].data_type) { case SCTP_CONNECT_LINK_INFO: if (connect_info->link->transport_connected == 0) { return -1; } break; case SCTP_ACCEPTED_LINK_INFO: listen_info = accepted_info->link_info; if (listen_info->listen_sock != sockfd) { if (listen_info->on_rx_epoll == 0) { return -1; } } break; } if (recv_errno == EAGAIN) { #ifdef DEBUG log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Sock: %d is overloaded. Slowing TX down", sockfd); #endif usleep(KNET_THREADS_TIMERES / 16); return 1; } return -1; } return 0; } /* * 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_accepted_link_info_t *accepted_info = knet_h->knet_transport_fd_tracker[sockfd].data; sctp_listen_link_info_t *listen_info; 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_ACCEPTED_LINK_INFO: listen_info = accepted_info->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: log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Received unknown notification? %d", sockfd); break; } /* * Under RX pressure we need to give time to IPC to pick up the message */ usleep(KNET_THREADS_TIMERES / 2); 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) +static int sctp_transport_rx_is_data(knet_handle_t knet_h, int sockfd, struct knet_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; sctp_accepted_link_info_t *info = knet_h->knet_transport_fd_tracker[sockfd].data; 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; } /* * missing MSG_EOR has to be treated as a short read * from the socket and we need to fill in the mread buf * while we wait for MSG_EOR */ if (!(msg->msg_hdr.msg_flags & MSG_EOR)) { /* * copy the incoming data into mread_buf + mread_len (incremental) * and increase mread_len */ memmove(info->mread_buf + info->mread_len, iov->iov_base, msg->msg_len); info->mread_len = info->mread_len + msg->msg_len; return 0; } /* * got EOR. * if mread_len is > 0 we are completing a packet from short reads * complete reassembling the packet in mread_buf, copy it back in the iov * and set the iov/msg len numbers (size) correctly */ if (info->mread_len) { /* * add last fragment to mread_buf */ memmove(info->mread_buf + info->mread_len, iov->iov_base, msg->msg_len); info->mread_len = info->mread_len + msg->msg_len; /* * move all back into the iovec */ memmove(iov->iov_base, info->mread_buf, info->mread_len); msg->msg_len = info->mread_len; info->mread_len = 0; } 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 *kn_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, kn_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, kn_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, kn_link->status.dst_ipaddr, kn_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; kn_link->transport_connected = 1; kn_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, kn_link->status.dst_ipaddr, kn_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]; sctp_accepted_link_info_t *accept_info = NULL; 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; } accept_info = malloc(sizeof(sctp_accepted_link_info_t)); if (!accept_info) { savederrno = errno; err = -1; goto exit_error; } memset(accept_info, 0, sizeof(sctp_accepted_link_info_t)); accept_info->link_info = info; if (_set_fd_tracker(knet_h, new_fd, KNET_TRANSPORT_SCTP, SCTP_ACCEPTED_LINK_INFO, accept_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); free(accept_info); 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_accepted_link_info_t *accept_info; sctp_listen_link_info_t *info; struct knet_host *host; int link_idx; 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); accept_info = knet_h->knet_transport_fd_tracker[sockfd].data; info = accept_info->link_info; /* * clear all links using this accepted socket as * outbound dynamically connected socket */ 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].dynamic == KNET_LINK_DYNIP) && (host->link[link_idx].outsock == sockfd)) { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Found dynamic connection on host %d link %d (%d)", host->host_id, link_idx, sockfd); host->link[link_idx].status.dynconnected = 0; host->link[link_idx].transport_connected = 0; host->link[link_idx].outsock = 0; memset(&host->link[link_idx].dst_addr, 0, sizeof(struct sockaddr_storage)); } } } 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; free(accept_info); 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, 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_LISTENER_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; free(knet_h->knet_transport_fd_tracker[info->accepted_socks[i]].data); 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 (link->dynamic == KNET_LINK_STATIC) { if (_create_connect_socket(knet_h, link) < 0) { savederrno = errno; err = -1; goto exit_error; } link->outsock = info->connect_sock; } knet_list_add(&info->list, &handle_info->connect_links_list); 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 */ 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 int sctp_transport_link_dyn_connect(knet_handle_t knet_h, int sockfd, struct knet_link *kn_link) { kn_link->outsock = sockfd; kn_link->status.dynconnected = 1; kn_link->transport_connected = 1; return 0; } static knet_transport_ops_t sctp_transport_ops = { .transport_name = "SCTP", .transport_id = KNET_TRANSPORT_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_link_dyn_connect = sctp_transport_link_dyn_connect, .transport_rx_sock_error = sctp_transport_rx_sock_error, .transport_tx_sock_error = sctp_transport_tx_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 5ef7e766..946879fe 100644 --- a/libknet/transport_udp.c +++ b/libknet/transport_udp.c @@ -1,408 +1,408 @@ #include "config.h" #include #include #include #include #include #include #include #include #include #if defined (IP_RECVERR) || defined (IPV6_RECVERR) #include #endif #include "libknet.h" #include "compat.h" #include "host.h" #include "link.h" #include "logging.h" #include "common.h" #include "transports.h" #include "threads_common.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 *kn_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]; #if defined (IP_RECVERR) || defined (IPV6_RECVERR) int value; #endif /* * 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, &kn_link->src_addr, sizeof(struct sockaddr_storage)) == 0) { log_debug(knet_h, KNET_SUB_TRANSP_UDP, "Re-using existing UDP socket for new link"); kn_link->outsock = info->socket_fd; kn_link->transport_link = info; kn_link->transport_connected = 1; return 0; } } info = malloc(sizeof(udp_link_info_t)); if (!info) { err = -1; goto exit_error; } sock = socket(kn_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, &kn_link->src_addr, "UDP") < 0) { savederrno = errno; err = -1; goto exit_error; } #ifdef IP_RECVERR if (kn_link->src_addr.ss_family == AF_INET) { value = 1; if (setsockopt(sock, SOL_IP, IP_RECVERR, &value, sizeof(value)) <0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_UDP, "Unable to set RECVERR on socket: %s", strerror(savederrno)); goto exit_error; } } #endif #ifdef IPV6_RECVERR if (kn_link->src_addr.ss_family == AF_INET6) { value = 1; if (setsockopt(sock, SOL_IPV6, IPV6_RECVERR, &value, sizeof(value)) <0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_UDP, "Unable to set RECVERR on socket: %s", strerror(savederrno)); goto exit_error; } } #endif if (bind(sock, (struct sockaddr *)&kn_link->src_addr, sockaddr_len(&kn_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, &kn_link->src_addr, sizeof(struct sockaddr_storage)); info->socket_fd = sock; knet_list_add(&info->list, &handle_info->links_list); kn_link->outsock = sock; kn_link->transport_link = info; kn_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 *kn_link) { int err = 0, savederrno = 0; int found = 0; struct knet_host *host; int link_idx; udp_link_info_t *info = kn_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] == kn_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(kn_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; } memset(handle_info, 0, sizeof(udp_handle_info_t)); knet_h->transports[KNET_TRANSPORT_UDP] = handle_info; knet_list_init(&handle_info->links_list); return 0; } #if defined (IP_RECVERR) || defined (IPV6_RECVERR) static int read_errs_from_sock(knet_handle_t knet_h, int sockfd) { int err = 0, savederrno = 0; int got_err = 0; char buffer[1024]; struct iovec iov; struct msghdr msg; struct cmsghdr *cmsg; struct sock_extended_err *sock_err; struct icmphdr icmph; struct sockaddr_storage remote; struct sockaddr_storage *origin; char addr_str[KNET_MAX_HOST_LEN]; char port_str[KNET_MAX_PORT_LEN]; iov.iov_base = &icmph; iov.iov_len = sizeof(icmph); msg.msg_name = (void*)&remote; msg.msg_namelen = sizeof(remote); msg.msg_iov = &iov; msg.msg_iovlen = 1; msg.msg_flags = 0; msg.msg_control = buffer; msg.msg_controllen = sizeof(buffer); for (;;) { err = recvmsg(sockfd, &msg, MSG_ERRQUEUE); savederrno = errno; if (err < 0) { if (!got_err) { errno = savederrno; return -1; } else { return 0; } } got_err = 1; for (cmsg = CMSG_FIRSTHDR(&msg);cmsg; cmsg = CMSG_NXTHDR(&msg, cmsg)) { if (((cmsg->cmsg_level == SOL_IP) && (cmsg->cmsg_type == IP_RECVERR)) || ((cmsg->cmsg_level == SOL_IPV6 && (cmsg->cmsg_type == IPV6_RECVERR)))) { sock_err = (struct sock_extended_err*)CMSG_DATA(cmsg); if (sock_err) { switch (sock_err->ee_origin) { case 0: /* no origin */ case 1: /* local source (EMSGSIZE) */ /* * those errors are way too noisy */ break; case 2: /* ICMP */ case 3: /* ICMP6 */ origin = (struct sockaddr_storage *)SO_EE_OFFENDER(sock_err); if (knet_addrtostr(origin, sizeof(origin), addr_str, KNET_MAX_HOST_LEN, port_str, KNET_MAX_PORT_LEN) < 0) { log_debug(knet_h, KNET_SUB_TRANSP_UDP, "Received ICMP error from unknown source: %s", strerror(sock_err->ee_errno)); } else { log_debug(knet_h, KNET_SUB_TRANSP_UDP, "Received ICMP error from %s: %s", addr_str, strerror(sock_err->ee_errno)); } break; } } else { log_debug(knet_h, KNET_SUB_TRANSP_UDP, "No data in MSG_ERRQUEUE"); } } } } } #else static int read_errs_from_sock(knet_handle_t knet_h, int sockfd) { return 0; } #endif static int udp_transport_rx_sock_error(knet_handle_t knet_h, int sockfd, int recv_err, int recv_errno) { if (recv_errno == EAGAIN) { read_errs_from_sock(knet_h, sockfd); } return 0; } static int udp_transport_tx_sock_error(knet_handle_t knet_h, int sockfd, int recv_err, int recv_errno) { if (recv_err < 0) { if ((recv_errno == ENOBUFS) || (recv_errno == EAGAIN)) { #ifdef DEBUG log_debug(knet_h, KNET_SUB_TRANSP_UDP, "Sock: %d is overloaded. Slowing TX down", sockfd); #endif usleep(KNET_THREADS_TIMERES / 16); return 1; } read_errs_from_sock(knet_h, sockfd); if (recv_errno == EMSGSIZE) { return 0; } return -1; } return 0; } -static int udp_transport_rx_is_data(knet_handle_t knet_h, int sockfd, struct mmsghdr *msg) +static int udp_transport_rx_is_data(knet_handle_t knet_h, int sockfd, struct knet_mmsghdr *msg) { if (msg->msg_len == 0) return 0; return 2; } static int udp_transport_link_dyn_connect(knet_handle_t knet_h, int sockfd, struct knet_link *kn_link) { kn_link->status.dynconnected = 1; return 0; } static knet_transport_ops_t udp_transport_ops = { .transport_name = "UDP", .transport_id = KNET_TRANSPORT_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_link_dyn_connect = udp_transport_link_dyn_connect, .transport_rx_sock_error = udp_transport_rx_sock_error, .transport_tx_sock_error = udp_transport_tx_sock_error, .transport_rx_is_data = udp_transport_rx_is_data, }; knet_transport_ops_t *get_udp_transport() { return &udp_transport_ops; }