diff --git a/libknet/handle.c b/libknet/handle.c index a52fceb4..69e0e3c1 100644 --- a/libknet/handle.c +++ b/libknet/handle.c @@ -1,1469 +1,1501 @@ /* * Copyright (C) 2010-2015 Red Hat, Inc. All rights reserved. * * Authors: Fabio M. Di Nitto * Federico Simoncelli * * This software licensed under GPL-2.0+, LGPL-2.0+ */ #include "config.h" #include #include #include #include #include #include #include #include #include #include #include "internals.h" #include "crypto.h" #include "common.h" #include "threads_common.h" #include "threads_heartbeat.h" #include "threads_pmtud.h" #include "threads_dsthandler.h" #include "threads_send_recv.h" #include "transports.h" #include "logging.h" static pthread_mutex_t handle_config_mutex = PTHREAD_MUTEX_INITIALIZER; static int _init_locks(knet_handle_t knet_h) { int savederrno = 0; savederrno = pthread_rwlock_init(&knet_h->global_rwlock, NULL); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to initialize list rwlock: %s", strerror(savederrno)); goto exit_fail; } knet_h->lock_init_done = 1; savederrno = pthread_rwlock_init(&knet_h->listener_rwlock, NULL); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to initialize listener rwlock: %s", strerror(savederrno)); goto exit_fail; } savederrno = pthread_rwlock_init(&knet_h->host_rwlock, NULL); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to initialize host rwlock: %s", strerror(savederrno)); goto exit_fail; } savederrno = pthread_mutex_init(&knet_h->host_mutex, NULL); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to initialize host mutex: %s", strerror(savederrno)); goto exit_fail; } savederrno = pthread_cond_init(&knet_h->host_cond, NULL); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to initialize host conditional mutex: %s", strerror(savederrno)); goto exit_fail; } savederrno = pthread_mutex_init(&knet_h->pmtud_mutex, NULL); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to initialize pmtud mutex: %s", strerror(savederrno)); goto exit_fail; } savederrno = pthread_cond_init(&knet_h->pmtud_cond, NULL); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to initialize pmtud conditional mutex: %s", strerror(savederrno)); goto exit_fail; } savederrno = pthread_mutex_init(&knet_h->tx_mutex, NULL); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to initialize tx_thread mutex: %s", strerror(savederrno)); goto exit_fail; } return 0; exit_fail: errno = savederrno; return -1; } static void _destroy_locks(knet_handle_t knet_h) { knet_h->lock_init_done = 0; pthread_rwlock_destroy(&knet_h->global_rwlock); pthread_rwlock_destroy(&knet_h->listener_rwlock); pthread_rwlock_destroy(&knet_h->host_rwlock); pthread_mutex_destroy(&knet_h->host_mutex); pthread_cond_destroy(&knet_h->host_cond); pthread_mutex_destroy(&knet_h->pmtud_mutex); pthread_cond_destroy(&knet_h->pmtud_cond); pthread_mutex_destroy(&knet_h->tx_mutex); } static int _init_socketpair(knet_handle_t knet_h, int *sock) { int savederrno = 0; int value; int i; if (socketpair(AF_UNIX, SOCK_SEQPACKET, 0, sock) != 0) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to initialize socketpair: %s", strerror(savederrno)); goto exit_fail; } for (i = 0; i < 2; i++) { if (_fdset_cloexec(sock[i])) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to set CLOEXEC on sock[%d]: %s", i, strerror(savederrno)); goto exit_fail; } if (_fdset_nonblock(sock[i])) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to set NONBLOCK on sock[%d]: %s", i, strerror(savederrno)); goto exit_fail; } value = KNET_RING_RCVBUFF; if (setsockopt(sock[i], SOL_SOCKET, SO_RCVBUFFORCE, &value, sizeof(value)) < 0) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to set receive buffer on sock[%d]: %s", i, strerror(savederrno)); goto exit_fail; } value = KNET_RING_RCVBUFF; if (setsockopt(sock[i], SOL_SOCKET, SO_SNDBUFFORCE, &value, sizeof(value)) < 0) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to set send buffer on sock[%d]: %s", i, strerror(savederrno)); goto exit_fail; } } return 0; exit_fail: errno = savederrno; return -1; } static void _close_socketpair(knet_handle_t knet_h, int *sock) { int i; for (i = 0; i < 2; i++) { if (sock[i]) { close(sock[i]); sock[i] = 0; } } } static int _init_socks(knet_handle_t knet_h) { int savederrno = 0; if (_init_socketpair(knet_h, knet_h->hostsockfd)) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to initialize internal hostsockpair: %s", strerror(savederrno)); goto exit_fail; } if (_init_socketpair(knet_h, knet_h->dstsockfd)) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to initialize internal dstsockpair: %s", strerror(savederrno)); goto exit_fail; } return 0; exit_fail: errno = savederrno; return -1; } static void _close_socks(knet_handle_t knet_h) { _close_socketpair(knet_h, knet_h->dstsockfd); _close_socketpair(knet_h, knet_h->hostsockfd); } static int _init_buffers(knet_handle_t knet_h) { int savederrno = 0; int i; size_t bufsize; for (i = 0; i < PCKT_FRAG_MAX; i++) { bufsize = ceil((float)KNET_MAX_PACKET_SIZE / (i + 1)) + KNET_HEADER_ALL_SIZE; knet_h->send_to_links_buf[i] = malloc(bufsize); if (!knet_h->send_to_links_buf[i]) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to allocate memory datafd to link buffer: %s", strerror(savederrno)); goto exit_fail; } memset(knet_h->send_to_links_buf[i], 0, bufsize); knet_h->recv_from_sock_buf[i] = malloc(KNET_DATABUFSIZE); if (!knet_h->recv_from_sock_buf[i]) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to allocate memory for app to datafd buffer: %s", strerror(savederrno)); goto exit_fail; } memset(knet_h->recv_from_sock_buf[i], 0, KNET_DATABUFSIZE); knet_h->recv_from_links_buf[i] = malloc(KNET_DATABUFSIZE); if (!knet_h->recv_from_links_buf[i]) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to allocate memory for link to datafd buffer: %s", strerror(savederrno)); goto exit_fail; } memset(knet_h->recv_from_links_buf[i], 0, KNET_DATABUFSIZE); } knet_h->pingbuf = malloc(KNET_HEADER_PING_SIZE); if (!knet_h->pingbuf) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to allocate memory for hearbeat buffer: %s", strerror(savederrno)); goto exit_fail; } memset(knet_h->pingbuf, 0, KNET_HEADER_PING_SIZE); knet_h->pmtudbuf = malloc(KNET_PMTUD_SIZE_V6); if (!knet_h->pmtudbuf) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to allocate memory for pmtud buffer: %s", strerror(savederrno)); goto exit_fail; } memset(knet_h->pmtudbuf, 0, KNET_PMTUD_SIZE_V6); for (i = 0; i < PCKT_FRAG_MAX; i++) { bufsize = ceil((float)KNET_MAX_PACKET_SIZE / (i + 1)) + KNET_HEADER_ALL_SIZE + KNET_DATABUFSIZE_CRYPT_PAD; knet_h->send_to_links_buf_crypt[i] = malloc(bufsize); if (!knet_h->send_to_links_buf_crypt[i]) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to allocate memory for crypto datafd to link buffer: %s", strerror(savederrno)); goto exit_fail; } memset(knet_h->send_to_links_buf_crypt[i], 0, bufsize); } knet_h->recv_from_links_buf_decrypt = malloc(KNET_DATABUFSIZE_CRYPT); if (!knet_h->recv_from_links_buf_decrypt) { savederrno = errno; log_err(knet_h, KNET_SUB_CRYPTO, "Unable to allocate memory for crypto link to datafd buffer: %s", strerror(savederrno)); goto exit_fail; } memset(knet_h->recv_from_links_buf_decrypt, 0, KNET_DATABUFSIZE_CRYPT); knet_h->recv_from_links_buf_crypt = malloc(KNET_DATABUFSIZE_CRYPT); if (!knet_h->recv_from_links_buf_crypt) { savederrno = errno; log_err(knet_h, KNET_SUB_CRYPTO, "Unable to allocate memory for crypto link to datafd buffer: %s", strerror(savederrno)); goto exit_fail; } memset(knet_h->recv_from_links_buf_crypt, 0, KNET_DATABUFSIZE_CRYPT); knet_h->pingbuf_crypt = malloc(KNET_DATABUFSIZE_CRYPT); if (!knet_h->pingbuf_crypt) { savederrno = errno; log_err(knet_h, KNET_SUB_CRYPTO, "Unable to allocate memory for crypto hearbeat buffer: %s", strerror(savederrno)); goto exit_fail; } memset(knet_h->pingbuf_crypt, 0, KNET_DATABUFSIZE_CRYPT); knet_h->pmtudbuf_crypt = malloc(KNET_DATABUFSIZE_CRYPT); if (!knet_h->pmtudbuf_crypt) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to allocate memory for crypto pmtud buffer: %s", strerror(savederrno)); goto exit_fail; } memset(knet_h->pmtudbuf_crypt, 0, KNET_DATABUFSIZE_CRYPT); return 0; exit_fail: errno = savederrno; return -1; } static void _destroy_buffers(knet_handle_t knet_h) { int i; for (i = 0; i < PCKT_FRAG_MAX; i++) { free(knet_h->send_to_links_buf[i]); free(knet_h->recv_from_sock_buf[i]); free(knet_h->send_to_links_buf_crypt[i]); free(knet_h->recv_from_links_buf[i]); } free(knet_h->recv_from_links_buf_decrypt); free(knet_h->recv_from_links_buf_crypt); free(knet_h->pingbuf); free(knet_h->pingbuf_crypt); free(knet_h->pmtudbuf); free(knet_h->pmtudbuf_crypt); } static int _init_epolls(knet_handle_t knet_h) { struct epoll_event ev; int savederrno = 0; /* * even if the kernel does dynamic allocation with epoll_ctl * we need to reserve one extra for host to host communication */ knet_h->send_to_links_epollfd = epoll_create(KNET_EPOLL_MAX_EVENTS + 1); if (knet_h->send_to_links_epollfd < 0) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to create epoll datafd to link fd: %s", strerror(savederrno)); goto exit_fail; } knet_h->recv_from_links_epollfd = epoll_create(KNET_EPOLL_MAX_EVENTS); if (knet_h->recv_from_links_epollfd < 0) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to create epoll link to datafd fd: %s", strerror(savederrno)); goto exit_fail; } knet_h->dst_link_handler_epollfd = epoll_create(KNET_EPOLL_MAX_EVENTS); if (knet_h->dst_link_handler_epollfd < 0) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to create epoll dst cache fd: %s", strerror(savederrno)); goto exit_fail; } if (_fdset_cloexec(knet_h->send_to_links_epollfd)) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to set CLOEXEC on datafd to link epoll fd: %s", strerror(savederrno)); goto exit_fail; } if (_fdset_cloexec(knet_h->recv_from_links_epollfd)) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to set CLOEXEC on link to datafd epoll fd: %s", strerror(savederrno)); goto exit_fail; } if (_fdset_cloexec(knet_h->dst_link_handler_epollfd)) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to set CLOEXEC on dst cache epoll fd: %s", strerror(savederrno)); goto exit_fail; } memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = knet_h->hostsockfd[0]; if (epoll_ctl(knet_h->send_to_links_epollfd, EPOLL_CTL_ADD, knet_h->hostsockfd[0], &ev)) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to add hostsockfd[0] to epoll pool: %s", strerror(savederrno)); goto exit_fail; } memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = knet_h->dstsockfd[0]; if (epoll_ctl(knet_h->dst_link_handler_epollfd, EPOLL_CTL_ADD, knet_h->dstsockfd[0], &ev)) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Unable to add dstsockfd[0] to epoll pool: %s", strerror(savederrno)); goto exit_fail; } return 0; exit_fail: errno = savederrno; return -1; } static void _close_epolls(knet_handle_t knet_h) { struct epoll_event ev; int i; memset(&ev, 0, sizeof(struct epoll_event)); for (i = 0; i < KNET_DATAFD_MAX; i++) { if (knet_h->sockfd[i].in_use) { epoll_ctl(knet_h->send_to_links_epollfd, EPOLL_CTL_DEL, knet_h->sockfd[i].sockfd[knet_h->sockfd[i].is_created], &ev); if (knet_h->sockfd[i].sockfd[knet_h->sockfd[i].is_created]) { _close_socketpair(knet_h, knet_h->sockfd[i].sockfd); } } } epoll_ctl(knet_h->send_to_links_epollfd, EPOLL_CTL_DEL, knet_h->hostsockfd[0], &ev); epoll_ctl(knet_h->dst_link_handler_epollfd, EPOLL_CTL_DEL, knet_h->dstsockfd[0], &ev); close(knet_h->send_to_links_epollfd); close(knet_h->recv_from_links_epollfd); close(knet_h->dst_link_handler_epollfd); } +static int _start_transports(knet_handle_t knet_h) +{ + int i, savederrno = 0, err = 0; + + for (i=0; itransport_ops[i] = get_udp_transport(); + break; +#ifdef HAVE_NETINET_SCTP_H + case KNET_TRANSPORT_SCTP: + knet_h->transport_ops[i] = get_sctp_transport(); + break; +#endif + } + if ((knet_h->transport_ops[i]) && + (knet_h->transport_ops[i]->handle_allocate)) { + knet_h->transport_ops[i]->handle_allocate(knet_h, &knet_h->transports[i]); + if (!knet_h->transports[i]) { + savederrno = errno; + log_err(knet_h, KNET_SUB_HANDLE, "Failed to allocate transport handle for %s: %s", + knet_h->transport_ops[i]->transport_name, + strerror(savederrno)); + err = -1; + goto out; + } + } + } + +out: + errno = savederrno; + return err; +} + +static void _stop_transports(knet_handle_t knet_h) +{ + int i; + + for (i=0; itransport_ops[i]) && + (knet_h->transport_ops[i]->handle_free)) { + knet_h->transport_ops[i]->handle_free(knet_h, knet_h->transports[i]); + } + } +} + static int _start_threads(knet_handle_t knet_h) { int savederrno = 0; savederrno = pthread_create(&knet_h->pmtud_link_handler_thread, 0, _handle_pmtud_link_thread, (void *) knet_h); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to start pmtud link thread: %s", strerror(savederrno)); goto exit_fail; } savederrno = pthread_create(&knet_h->dst_link_handler_thread, 0, _handle_dst_link_handler_thread, (void *) knet_h); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to start dst cache thread: %s", strerror(savederrno)); goto exit_fail; } savederrno = pthread_create(&knet_h->send_to_links_thread, 0, _handle_send_to_links_thread, (void *) knet_h); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to start datafd to link thread: %s", strerror(savederrno)); goto exit_fail; } savederrno = pthread_create(&knet_h->recv_from_links_thread, 0, _handle_recv_from_links_thread, (void *) knet_h); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to start link to datafd thread: %s", strerror(savederrno)); goto exit_fail; } savederrno = pthread_create(&knet_h->heartbt_thread, 0, _handle_heartbt_thread, (void *) knet_h); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to start heartbeat thread: %s", strerror(savederrno)); goto exit_fail; } return 0; exit_fail: errno = savederrno; return -1; } - -static void _stop_transports(knet_handle_t knet_h) -{ - int i; - knet_transport_ops_t *ops = NULL; - - for (i=0; ihandle_free(knet_h, knet_h->transports[i]); - } - } -} - static void _stop_threads(knet_handle_t knet_h) { void *retval; /* * allow threads to catch on shutdown request * and release locks before we stop them. * this isn't the most efficent way to handle it * but it works good enough for now */ sleep(1); pthread_mutex_lock(&knet_h->host_mutex); pthread_cond_signal(&knet_h->host_cond); pthread_mutex_unlock(&knet_h->host_mutex); if (knet_h->heartbt_thread) { pthread_cancel(knet_h->heartbt_thread); pthread_join(knet_h->heartbt_thread, &retval); } if (knet_h->send_to_links_thread) { pthread_cancel(knet_h->send_to_links_thread); pthread_join(knet_h->send_to_links_thread, &retval); } if (knet_h->recv_from_links_thread) { pthread_cancel(knet_h->recv_from_links_thread); pthread_join(knet_h->recv_from_links_thread, &retval); } if (knet_h->dst_link_handler_thread) { pthread_cancel(knet_h->dst_link_handler_thread); pthread_join(knet_h->dst_link_handler_thread, &retval); } pthread_mutex_lock(&knet_h->pmtud_mutex); pthread_cond_signal(&knet_h->pmtud_cond); pthread_mutex_unlock(&knet_h->pmtud_mutex); sleep(1); if (knet_h->pmtud_link_handler_thread) { pthread_cancel(knet_h->pmtud_link_handler_thread); pthread_join(knet_h->pmtud_link_handler_thread, &retval); } } knet_handle_t knet_handle_new(uint16_t host_id, int log_fd, uint8_t default_log_level) { knet_handle_t knet_h; int savederrno = 0; struct rlimit cur; if (getrlimit(RLIMIT_NOFILE, &cur) < 0) { return NULL; } if ((log_fd < 0) || (log_fd >= cur.rlim_max)) { errno = EINVAL; return NULL; } /* * validate incoming request */ if ((log_fd) && (default_log_level > KNET_LOG_DEBUG)) { errno = EINVAL; return NULL; } /* * allocate handle */ knet_h = malloc(sizeof(struct knet_handle)); if (!knet_h) { errno = ENOMEM; return NULL; } memset(knet_h, 0, sizeof(struct knet_handle)); savederrno = pthread_mutex_lock(&handle_config_mutex); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get handle mutex lock: %s", strerror(savederrno)); errno = savederrno; goto exit_fail; } /* * copy config in place */ knet_h->host_id = host_id; knet_h->logfd = log_fd; if (knet_h->logfd > 0) { memset(&knet_h->log_levels, default_log_level, KNET_MAX_SUBSYSTEMS); } /* * set pmtud default timers */ knet_h->pmtud_interval = KNET_PMTUD_DEFAULT_INTERVAL; /* * init main locking structures */ if (_init_locks(knet_h)) { savederrno = errno; goto exit_fail; } /* * init sockets */ if (_init_socks(knet_h)) { savederrno = errno; goto exit_fail; } /* * allocate packet buffers */ if (_init_buffers(knet_h)) { savederrno = errno; goto exit_fail; } /* * create epoll fds */ if (_init_epolls(knet_h)) { savederrno = errno; goto exit_fail; } + /* + * start transports + */ + + if (_start_transports(knet_h)) { + savederrno = errno; + goto exit_fail; + } + /* * start internal threads */ if (_start_threads(knet_h)) { savederrno = errno; goto exit_fail; } pthread_mutex_unlock(&handle_config_mutex); return knet_h; exit_fail: pthread_mutex_unlock(&handle_config_mutex); knet_handle_free(knet_h); errno = savederrno; return NULL; } int knet_handle_free(knet_handle_t knet_h) { int savederrno = 0; savederrno = pthread_mutex_lock(&handle_config_mutex); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get handle mutex lock: %s", strerror(savederrno)); errno = savederrno; return -1; } if (!knet_h) { pthread_mutex_unlock(&handle_config_mutex); errno = EINVAL; return -1; } if (!knet_h->lock_init_done) { goto exit_nolock; } savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get write lock: %s", strerror(savederrno)); pthread_mutex_unlock(&handle_config_mutex); errno = savederrno; return -1; } if (knet_h->host_head != NULL) { savederrno = EBUSY; log_err(knet_h, KNET_SUB_HANDLE, "Unable to free handle: host(s) or listener(s) are still active: %s", strerror(savederrno)); pthread_rwlock_unlock(&knet_h->global_rwlock); pthread_mutex_unlock(&handle_config_mutex); errno = savederrno; return -1; } knet_h->fini_in_progress = 1; pthread_rwlock_unlock(&knet_h->global_rwlock); _stop_threads(knet_h); _stop_transports(knet_h); _close_epolls(knet_h); _destroy_buffers(knet_h); _close_socks(knet_h); crypto_fini(knet_h); _destroy_locks(knet_h); exit_nolock: free(knet_h); knet_h = NULL; pthread_mutex_unlock(&handle_config_mutex); return 0; } int knet_handle_enable_sock_notify(knet_handle_t knet_h, void *sock_notify_fn_private_data, void (*sock_notify_fn) ( void *private_data, int datafd, int8_t channel, uint8_t tx_rx, int error, int errorno)) { int savederrno = 0, err = 0; if (!knet_h) { errno = EINVAL; return -1; } if (!sock_notify_fn) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get write lock: %s", strerror(savederrno)); errno = savederrno; return -1; } knet_h->sock_notify_fn_private_data = sock_notify_fn_private_data; knet_h->sock_notify_fn = sock_notify_fn; log_debug(knet_h, KNET_SUB_HANDLE, "sock_notify_fn enabled"); pthread_rwlock_unlock(&knet_h->global_rwlock); return err; } int knet_handle_add_datafd(knet_handle_t knet_h, int *datafd, int8_t *channel) { int err = 0, savederrno = 0; int i; struct epoll_event ev; if (!knet_h) { errno = EINVAL; return -1; } if (datafd == NULL) { errno = EINVAL; return -1; } if (channel == NULL) { errno = EINVAL; return -1; } if (*channel >= KNET_DATAFD_MAX) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get write lock: %s", strerror(savederrno)); errno = savederrno; return -1; } if (!knet_h->sock_notify_fn) { log_err(knet_h, KNET_SUB_HANDLE, "Adding datafd requires sock notify callback enabled!"); savederrno = EINVAL; err = -1; goto out_unlock; } if (*datafd > 0) { for (i = 0; i < KNET_DATAFD_MAX; i++) { if ((knet_h->sockfd[i].in_use) && (knet_h->sockfd[i].sockfd[0] == *datafd)) { log_err(knet_h, KNET_SUB_HANDLE, "requested datafd: %d already exist in index: %d", *datafd, i); savederrno = EEXIST; err = -1; goto out_unlock; } } } /* * auto allocate a channel */ if (*channel < 0) { for (i = 0; i < KNET_DATAFD_MAX; i++) { if (!knet_h->sockfd[i].in_use) { *channel = i; break; } } if (*channel < 0) { savederrno = EBUSY; err = -1; goto out_unlock; } } else { if (knet_h->sockfd[*channel].in_use) { savederrno = EBUSY; err = -1; goto out_unlock; } } knet_h->sockfd[*channel].is_created = 0; knet_h->sockfd[*channel].is_socket = 0; knet_h->sockfd[*channel].has_error = 0; if (*datafd > 0) { int sockopt; socklen_t sockoptlen = sizeof(sockopt); if (_fdset_cloexec(*datafd)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_HANDLE, "Unable to set CLOEXEC on datafd: %s", strerror(savederrno)); goto out_unlock; } if (_fdset_nonblock(*datafd)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_HANDLE, "Unable to set NONBLOCK on datafd: %s", strerror(savederrno)); goto out_unlock; } knet_h->sockfd[*channel].sockfd[0] = *datafd; knet_h->sockfd[*channel].sockfd[1] = 0; if (!getsockopt(knet_h->sockfd[*channel].sockfd[0], SOL_SOCKET, SO_TYPE, &sockopt, &sockoptlen)) { knet_h->sockfd[*channel].is_socket = 1; } } else { if (_init_socketpair(knet_h, knet_h->sockfd[*channel].sockfd)) { savederrno = errno; err = -1; goto out_unlock; } knet_h->sockfd[*channel].is_created = 1; knet_h->sockfd[*channel].is_socket = 1; *datafd = knet_h->sockfd[*channel].sockfd[0]; } memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = knet_h->sockfd[*channel].sockfd[knet_h->sockfd[*channel].is_created]; if (epoll_ctl(knet_h->send_to_links_epollfd, EPOLL_CTL_ADD, knet_h->sockfd[*channel].sockfd[knet_h->sockfd[*channel].is_created], &ev)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_HANDLE, "Unable to add datafd %d to linkfd epoll pool: %s", knet_h->sockfd[*channel].sockfd[knet_h->sockfd[*channel].is_created], strerror(savederrno)); if (knet_h->sockfd[*channel].is_created) { _close_socketpair(knet_h, knet_h->sockfd[*channel].sockfd); } goto out_unlock; } knet_h->sockfd[*channel].in_use = 1; out_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = savederrno; return err; } int knet_handle_remove_datafd(knet_handle_t knet_h, int datafd) { int err = 0, savederrno = 0; int8_t channel = -1; int i; struct epoll_event ev; if (!knet_h) { errno = EINVAL; return -1; } if (datafd <= 0) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get write lock: %s", strerror(savederrno)); errno = savederrno; return -1; } for (i = 0; i < KNET_DATAFD_MAX; i++) { if ((knet_h->sockfd[i].in_use) && (knet_h->sockfd[i].sockfd[0] == datafd)) { channel = i; break; } } if (channel < 0) { savederrno = EINVAL; err = -1; goto out_unlock; } if (!knet_h->sockfd[channel].has_error) { memset(&ev, 0, sizeof(struct epoll_event)); if (epoll_ctl(knet_h->send_to_links_epollfd, EPOLL_CTL_DEL, knet_h->sockfd[channel].sockfd[knet_h->sockfd[channel].is_created], &ev)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_HANDLE, "Unable to del datafd %d from linkfd epoll pool: %s", knet_h->sockfd[channel].sockfd[0], strerror(savederrno)); goto out_unlock; } } if (knet_h->sockfd[channel].is_created) { _close_socketpair(knet_h, knet_h->sockfd[channel].sockfd); } memset(&knet_h->sockfd[channel], 0, sizeof(struct knet_sock)); out_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = savederrno; return err; } int knet_handle_get_datafd(knet_handle_t knet_h, const int8_t channel, int *datafd) { int err = 0, savederrno = 0; if (!knet_h) { errno = EINVAL; return -1; } if ((channel < 0) || (channel >= KNET_DATAFD_MAX)) { errno = EINVAL; return -1; } if (datafd == NULL) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get read lock: %s", strerror(savederrno)); errno = savederrno; return -1; } if (!knet_h->sockfd[channel].in_use) { savederrno = EINVAL; err = -1; goto out_unlock; } *datafd = knet_h->sockfd[channel].sockfd[0]; out_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = savederrno; return err; } int knet_handle_get_channel(knet_handle_t knet_h, const int datafd, int8_t *channel) { int err = 0, savederrno = 0; int i; if (!knet_h) { errno = EINVAL; return -1; } if (datafd <= 0) { errno = EINVAL; return -1; } if (channel == NULL) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get read lock: %s", strerror(savederrno)); errno = savederrno; return -1; } *channel = -1; for (i = 0; i < KNET_DATAFD_MAX; i++) { if ((knet_h->sockfd[i].in_use) && (knet_h->sockfd[i].sockfd[0] == datafd)) { *channel = i; break; } } if (*channel < 0) { savederrno = EINVAL; err = -1; goto out_unlock; } out_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = savederrno; return err; } int knet_handle_enable_filter(knet_handle_t knet_h, void *dst_host_filter_fn_private_data, int (*dst_host_filter_fn) ( void *private_data, const unsigned char *outdata, ssize_t outdata_len, uint8_t tx_rx, uint16_t this_host_id, uint16_t src_node_id, int8_t *channel, uint16_t *dst_host_ids, size_t *dst_host_ids_entries)) { int savederrno = 0; if (!knet_h) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get write lock: %s", strerror(savederrno)); errno = savederrno; return -1; } knet_h->dst_host_filter_fn_private_data = dst_host_filter_fn_private_data; knet_h->dst_host_filter_fn = dst_host_filter_fn; if (knet_h->dst_host_filter_fn) { log_debug(knet_h, KNET_SUB_HANDLE, "dst_host_filter_fn enabled"); } else { log_debug(knet_h, KNET_SUB_HANDLE, "dst_host_filter_fn disabled"); } pthread_rwlock_unlock(&knet_h->global_rwlock); return 0; } int knet_handle_setfwd(knet_handle_t knet_h, unsigned int enabled) { int savederrno = 0; if (!knet_h) { errno = EINVAL; return -1; } if ((enabled < 0) || (enabled > 1)) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get write lock: %s", strerror(savederrno)); errno = savederrno; return -1; } knet_h->enabled = enabled; if (enabled) { log_debug(knet_h, KNET_SUB_HANDLE, "Data forwarding is enabled"); } else { log_debug(knet_h, KNET_SUB_HANDLE, "Data forwarding is disabled"); } pthread_rwlock_unlock(&knet_h->global_rwlock); return 0; } int knet_handle_pmtud_getfreq(knet_handle_t knet_h, unsigned int *interval) { int savederrno = 0; if (!knet_h) { errno = EINVAL; return -1; } if (!interval) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get write lock: %s", strerror(savederrno)); errno = savederrno; return -1; } *interval = knet_h->pmtud_interval; pthread_rwlock_unlock(&knet_h->global_rwlock); return 0; } int knet_handle_pmtud_setfreq(knet_handle_t knet_h, unsigned int interval) { int savederrno = 0; if (!knet_h) { errno = EINVAL; return -1; } if ((!interval) || (interval > 86400)) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get write lock: %s", strerror(savederrno)); errno = savederrno; return -1; } knet_h->pmtud_interval = interval; log_debug(knet_h, KNET_SUB_HANDLE, "PMTUd interval set to: %u seconds", interval); pthread_rwlock_unlock(&knet_h->global_rwlock); return 0; } int knet_handle_enable_pmtud_notify(knet_handle_t knet_h, void *pmtud_notify_fn_private_data, void (*pmtud_notify_fn) ( void *private_data, unsigned int data_mtu)) { int savederrno = 0; if (!knet_h) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get write lock: %s", strerror(savederrno)); errno = savederrno; return -1; } knet_h->pmtud_notify_fn_private_data = pmtud_notify_fn_private_data; knet_h->pmtud_notify_fn = pmtud_notify_fn; if (knet_h->pmtud_notify_fn) { log_debug(knet_h, KNET_SUB_HANDLE, "pmtud_notify_fn enabled"); } else { log_debug(knet_h, KNET_SUB_HANDLE, "pmtud_notify_fn disabled"); } pthread_rwlock_unlock(&knet_h->global_rwlock); return 0; } int knet_handle_pmtud_get(knet_handle_t knet_h, unsigned int *data_mtu) { int savederrno = 0; if (!knet_h) { errno = EINVAL; return -1; } if (!data_mtu) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get write lock: %s", strerror(savederrno)); errno = savederrno; return -1; } *data_mtu = knet_h->data_mtu; pthread_rwlock_unlock(&knet_h->global_rwlock); return 0; } int knet_handle_crypto(knet_handle_t knet_h, struct knet_handle_crypto_cfg *knet_handle_crypto_cfg) { int savederrno = 0; int err = 0; if (!knet_h) { errno = EINVAL; return -1; } if (!knet_handle_crypto_cfg) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get write lock: %s", strerror(savederrno)); errno = savederrno; return -1; } crypto_fini(knet_h); if ((!strncmp("none", knet_handle_crypto_cfg->crypto_model, 4)) || ((!strncmp("none", knet_handle_crypto_cfg->crypto_cipher_type, 4)) && (!strncmp("none", knet_handle_crypto_cfg->crypto_hash_type, 4)))) { log_debug(knet_h, KNET_SUB_CRYPTO, "crypto is not enabled"); err = 0; goto exit_unlock; } if (knet_handle_crypto_cfg->private_key_len < KNET_MIN_KEY_LEN) { log_debug(knet_h, KNET_SUB_CRYPTO, "private key len too short (min %u): %u", KNET_MIN_KEY_LEN, knet_handle_crypto_cfg->private_key_len); savederrno = EINVAL; err = -1; goto exit_unlock; } if (knet_handle_crypto_cfg->private_key_len > KNET_MAX_KEY_LEN) { log_debug(knet_h, KNET_SUB_CRYPTO, "private key len too long (max %u): %u", KNET_MAX_KEY_LEN, knet_handle_crypto_cfg->private_key_len); savederrno = EINVAL; err = -1; goto exit_unlock; } err = crypto_init(knet_h, knet_handle_crypto_cfg); if (err) { err = -2; } exit_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = savederrno; return err; } ssize_t knet_recv(knet_handle_t knet_h, char *buff, const size_t buff_len, const int8_t channel) { int savederrno = 0; ssize_t err = 0; struct iovec iov_in; if (!knet_h) { errno = EINVAL; return -1; } if (buff == NULL) { errno = EINVAL; return -1; } if (buff_len <= 0) { errno = EINVAL; return -1; } if (buff_len > KNET_MAX_PACKET_SIZE) { errno = EINVAL; return -1; } if (channel < 0) { errno = EINVAL; return -1; } if (channel >= KNET_DATAFD_MAX) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get read lock: %s", strerror(savederrno)); errno = savederrno; return -1; } if (!knet_h->sockfd[channel].in_use) { savederrno = EINVAL; err = -1; goto out_unlock; } memset(&iov_in, 0, sizeof(iov_in)); iov_in.iov_base = (void *)buff; iov_in.iov_len = buff_len; err = readv(knet_h->sockfd[channel].sockfd[0], &iov_in, 1); savederrno = errno; out_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = savederrno; return err; } ssize_t knet_send(knet_handle_t knet_h, const char *buff, const size_t buff_len, const int8_t channel) { int savederrno = 0; ssize_t err = 0; struct iovec iov_out[1]; if (!knet_h) { errno = EINVAL; return -1; } if (buff == NULL) { errno = EINVAL; return -1; } if (buff_len <= 0) { errno = EINVAL; return -1; } if (buff_len > KNET_MAX_PACKET_SIZE) { errno = EINVAL; return -1; } if (channel < 0) { errno = EINVAL; return -1; } if (channel >= KNET_DATAFD_MAX) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get read lock: %s", strerror(savederrno)); errno = savederrno; return -1; } if (!knet_h->sockfd[channel].in_use) { savederrno = EINVAL; err = -1; goto out_unlock; } memset(iov_out, 0, sizeof(iov_out)); iov_out[0].iov_base = (void *)buff; iov_out[0].iov_len = buff_len; err = writev(knet_h->sockfd[channel].sockfd[0], iov_out, 1); savederrno = errno; out_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = savederrno; return err; } diff --git a/libknet/link.c b/libknet/link.c index 1383d31f..8bd8a969 100644 --- a/libknet/link.c +++ b/libknet/link.c @@ -1,983 +1,956 @@ /* * Copyright (C) 2010-2015 Red Hat, Inc. All rights reserved. * * Authors: Fabio M. Di Nitto * Federico Simoncelli * * This software licensed under GPL-2.0+, LGPL-2.0+ */ #include "config.h" #include #include #include #include #include #include "internals.h" #include "logging.h" #include "link.h" #include "listener.h" #include "transports.h" #include "host.h" int _link_updown(knet_handle_t knet_h, uint16_t host_id, uint8_t link_id, unsigned int enabled, unsigned int connected) { struct knet_link *link = &knet_h->host_index[host_id]->link[link_id]; if ((link->status.enabled == enabled) && (link->status.connected == connected)) return 0; link->status.enabled = enabled; link->status.connected = connected; _host_dstcache_update_sync(knet_h, knet_h->host_index[host_id]); if ((link->status.dynconnected) && (!link->status.connected)) link->status.dynconnected = 0; return 0; } int knet_link_set_config(knet_handle_t knet_h, uint16_t host_id, uint8_t link_id, uint8_t transport, struct sockaddr_storage *src_addr, struct sockaddr_storage *dst_addr) { int savederrno = 0, err = 0; struct knet_host *host; struct knet_link *link; if (!knet_h) { errno = EINVAL; return -1; } if (link_id >= KNET_MAX_LINK) { errno = EINVAL; return -1; } if (!src_addr) { errno = EINVAL; return -1; } if (transport >= KNET_MAX_TRANSPORTS) { errno = EINVAL; return -1; } + if (!knet_h->transport_ops[transport]) { + errno = EINVAL; + return -1; + } + savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_LINK, "Unable to get write lock: %s", strerror(savederrno)); errno = savederrno; return -1; } host = knet_h->host_index[host_id]; if (!host) { err = -1; savederrno = EINVAL; log_err(knet_h, KNET_SUB_LINK, "Unable to find host %u: %s", host_id, strerror(savederrno)); goto exit_unlock; } link = &host->link[link_id]; if (link->status.enabled != 0) { err =-1; savederrno = EBUSY; log_err(knet_h, KNET_SUB_LINK, "Host %u link %u is currently in use: %s", host_id, link_id, strerror(savederrno)); goto exit_unlock; } memmove(&link->src_addr, src_addr, sizeof(struct sockaddr_storage)); err = getnameinfo((const struct sockaddr *)src_addr, sizeof(struct sockaddr_storage), link->status.src_ipaddr, KNET_MAX_HOST_LEN, link->status.src_port, KNET_MAX_PORT_LEN, NI_NUMERICHOST | NI_NUMERICSERV); if (err) { if (err == EAI_SYSTEM) { savederrno = errno; log_warn(knet_h, KNET_SUB_LINK, "Unable to resolve host: %u link: %u source addr/port: %s", host_id, link_id, strerror(savederrno)); } else { savederrno = EINVAL; log_warn(knet_h, KNET_SUB_LINK, "Unable to resolve host: %u link: %u source addr/port: %s", host_id, link_id, gai_strerror(err)); } err = -1; goto exit_unlock; } - link->transport_type = transport; - - switch (transport) { - case KNET_TRANSPORT_UDP: - knet_h->transport_ops[link->transport_type] = get_udp_transport(); - break; - case KNET_TRANSPORT_SCTP: -#ifdef HAVE_NETINET_SCTP_H - knet_h->transport_ops[link->transport_type] = get_sctp_transport(); - break; -#else - log_warn(knet_h, KNET_SUB_LINK, - "SCTP protocol not supported in this build"); -#endif - default: - savederrno = EINVAL; - err = -1; - goto exit_unlock; - } - - /* First time we've used this transport for this handle */ - if (!knet_h->transports[transport]) { - knet_h->transport_ops[link->transport_type]->handle_allocate(knet_h, &knet_h->transports[transport]); - } - if (!knet_h->transports[transport]) { - savederrno = errno; - log_err(knet_h, KNET_SUB_LISTENER, "Failed to allocate transport handle for %s: %s", - knet_h->transport_ops[link->transport_type]->transport_name, - strerror(savederrno)); - err = -1; - goto exit_unlock; - } - if (!dst_addr) { link->dynamic = KNET_LINK_DYNIP; err = 0; goto exit_unlock; } link->dynamic = KNET_LINK_STATIC; memmove(&link->dst_addr, dst_addr, sizeof(struct sockaddr_storage)); err = getnameinfo((const struct sockaddr *)dst_addr, sizeof(struct sockaddr_storage), link->status.dst_ipaddr, KNET_MAX_HOST_LEN, link->status.dst_port, KNET_MAX_PORT_LEN, NI_NUMERICHOST | NI_NUMERICSERV); if (err) { if (err == EAI_SYSTEM) { savederrno = errno; log_warn(knet_h, KNET_SUB_LINK, "Unable to resolve host: %u link: %u destination addr/port: %s", host_id, link_id, strerror(savederrno)); } else { savederrno = EINVAL; log_warn(knet_h, KNET_SUB_LINK, "Unable to resolve host: %u link: %u destination addr/port: %s", host_id, link_id, gai_strerror(err)); } err = -1; } exit_unlock: if (!err) { + link->transport_type = transport; link->configured = 1; link->pong_count = KNET_LINK_DEFAULT_PONG_COUNT; link->has_valid_mtu = 0; link->ping_interval = KNET_LINK_DEFAULT_PING_INTERVAL * 1000; /* microseconds */ link->pong_timeout = KNET_LINK_DEFAULT_PING_TIMEOUT * 1000; /* microseconds */ link->latency_fix = KNET_LINK_DEFAULT_PING_PRECISION; link->latency_exp = KNET_LINK_DEFAULT_PING_PRECISION - \ ((link->ping_interval * KNET_LINK_DEFAULT_PING_PRECISION) / 8000000); } pthread_rwlock_unlock(&knet_h->global_rwlock); errno = savederrno; return err; } int knet_link_get_config(knet_handle_t knet_h, uint16_t host_id, uint8_t link_id, uint8_t *transport, struct sockaddr_storage *src_addr, struct sockaddr_storage *dst_addr, uint8_t *dynamic) { int savederrno = 0, err = 0; struct knet_host *host; struct knet_link *link; if (!knet_h) { errno = EINVAL; return -1; } if (link_id >= KNET_MAX_LINK) { errno = EINVAL; return -1; } if (!src_addr) { errno = EINVAL; return -1; } if (!dynamic) { errno = EINVAL; return -1; } if (!transport) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_LINK, "Unable to get read lock: %s", strerror(savederrno)); errno = savederrno; return -1; } host = knet_h->host_index[host_id]; if (!host) { err = -1; savederrno = EINVAL; log_err(knet_h, KNET_SUB_LINK, "Unable to find host %u: %s", host_id, strerror(savederrno)); goto exit_unlock; } link = &host->link[link_id]; if (!link->configured) { err = -1; savederrno = EINVAL; log_err(knet_h, KNET_SUB_LINK, "host %u link %u is not configured: %s", host_id, link_id, strerror(savederrno)); goto exit_unlock; } if ((link->dynamic == KNET_LINK_STATIC) && (!dst_addr)) { savederrno = EINVAL; err = -1; goto exit_unlock; } memmove(src_addr, &link->src_addr, sizeof(struct sockaddr_storage)); *transport = link->transport_type; if (link->dynamic == KNET_LINK_STATIC) { *dynamic = 0; memmove(dst_addr, &link->dst_addr, sizeof(struct sockaddr_storage)); } else { *dynamic = 1; } exit_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = savederrno; return err; } int knet_link_set_enable(knet_handle_t knet_h, uint16_t host_id, uint8_t link_id, unsigned int enabled) { int savederrno = 0, err = 0; struct knet_host *host; struct knet_link *link; if (!knet_h) { errno = EINVAL; return -1; } if (link_id >= KNET_MAX_LINK) { errno = EINVAL; return -1; } if (enabled > 1) { errno = EINVAL; return -1; } /* * this read lock might appear as an API violation, but be * very careful because we cannot use a write lock (yet). * the _send_host_info requires threads to be operational. * a write lock here would deadlock. * a read lock is sufficient as all functions invoked by * this code are already thread safe. */ savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_LINK, "Unable to get read lock: %s", strerror(savederrno)); errno = savederrno; return -1; } host = knet_h->host_index[host_id]; if (!host) { err = -1; savederrno = EINVAL; log_err(knet_h, KNET_SUB_LINK, "Unable to find host %u: %s", host_id, strerror(savederrno)); goto exit_unlock; } link = &host->link[link_id]; if (!link->configured) { err = -1; savederrno = EINVAL; log_err(knet_h, KNET_SUB_LINK, "host %u link %u is not configured: %s", host_id, link_id, strerror(savederrno)); goto exit_unlock; } if (link->status.enabled == enabled) { err = 0; goto exit_unlock; } if (enabled) { if (knet_h->transport_ops[link->transport_type]->link_allocate( knet_h, knet_h->transports[link->transport_type], link, &link->transport, link_id, &link->src_addr, &link->dst_addr, &link->outsock) < 0) { savederrno = errno; err = -1; goto exit_unlock; } if (_listener_add(knet_h, host_id, link_id) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_LINK, "Unable to setup listener for this link"); goto exit_unlock; } log_debug(knet_h, KNET_SUB_LINK, "host: %u link: %u is enabled", host_id, link_id); } if (!enabled) { struct knet_hostinfo knet_hostinfo; knet_hostinfo.khi_type = KNET_HOSTINFO_TYPE_LINK_UP_DOWN; knet_hostinfo.khi_bcast = KNET_HOSTINFO_UCAST; knet_hostinfo.khi_dst_node_id = host_id; knet_hostinfo.khip_link_status_link_id = link_id; knet_hostinfo.khip_link_status_status = KNET_HOSTINFO_LINK_STATUS_DOWN; _send_host_info(knet_h, &knet_hostinfo, KNET_HOSTINFO_LINK_STATUS_SIZE); } err = _link_updown(knet_h, host_id, link_id, enabled, link->status.connected); savederrno = errno; if ((!err) && (enabled)) { err = 0; goto exit_unlock; } if (err) { err = -1; goto exit_unlock; } err = _listener_remove(knet_h, host_id, link_id); savederrno = errno; if ((err) && (savederrno != EBUSY)) { log_err(knet_h, KNET_SUB_LINK, "Unable to remove listener for this link"); if (_link_updown(knet_h, host_id, link_id, 1, link->status.connected)) { /* force link status the hard way */ link->status.enabled = 1; } log_debug(knet_h, KNET_SUB_LINK, "host: %u link: %u is NOT disabled", host_id, link_id); err = -1; goto exit_unlock; } else { err = 0; savederrno = 0; } log_debug(knet_h, KNET_SUB_LINK, "host: %u link: %u is disabled", host_id, link_id); link->host_info_up_sent = 0; exit_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = savederrno; return err; } int knet_link_get_enable(knet_handle_t knet_h, uint16_t host_id, uint8_t link_id, unsigned int *enabled) { int savederrno = 0, err = 0; struct knet_host *host; struct knet_link *link; if (!knet_h) { errno = EINVAL; return -1; } if (link_id >= KNET_MAX_LINK) { errno = EINVAL; return -1; } if (!enabled) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_LINK, "Unable to get read lock: %s", strerror(savederrno)); errno = savederrno; return -1; } host = knet_h->host_index[host_id]; if (!host) { err = -1; savederrno = EINVAL; log_err(knet_h, KNET_SUB_LINK, "Unable to find host %u: %s", host_id, strerror(savederrno)); goto exit_unlock; } link = &host->link[link_id]; if (!link->configured) { err = -1; savederrno = EINVAL; log_err(knet_h, KNET_SUB_LINK, "host %u link %u is not configured: %s", host_id, link_id, strerror(savederrno)); goto exit_unlock; } *enabled = link->status.enabled; exit_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = savederrno; return err; } int knet_link_set_pong_count(knet_handle_t knet_h, uint16_t host_id, uint8_t link_id, uint8_t pong_count) { int savederrno = 0, err = 0; struct knet_host *host; struct knet_link *link; if (!knet_h) { errno = EINVAL; return -1; } if (link_id >= KNET_MAX_LINK) { errno = EINVAL; return -1; } if (pong_count < 1) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_LINK, "Unable to get write lock: %s", strerror(savederrno)); errno = savederrno; return -1; } host = knet_h->host_index[host_id]; if (!host) { err = -1; savederrno = EINVAL; log_err(knet_h, KNET_SUB_LINK, "Unable to find host %u: %s", host_id, strerror(savederrno)); goto exit_unlock; } link = &host->link[link_id]; if (!link->configured) { err = -1; savederrno = EINVAL; log_err(knet_h, KNET_SUB_LINK, "host %u link %u is not configured: %s", host_id, link_id, strerror(savederrno)); goto exit_unlock; } link->pong_count = pong_count; log_debug(knet_h, KNET_SUB_LINK, "host: %u link: %u pong count update: %u", host_id, link_id, link->pong_count); exit_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = savederrno; return err; } int knet_link_get_pong_count(knet_handle_t knet_h, uint16_t host_id, uint8_t link_id, uint8_t *pong_count) { int savederrno = 0, err = 0; struct knet_host *host; struct knet_link *link; if (!knet_h) { errno = EINVAL; return -1; } if (link_id >= KNET_MAX_LINK) { errno = EINVAL; return -1; } if (!pong_count) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_LINK, "Unable to get read lock: %s", strerror(savederrno)); errno = savederrno; return -1; } host = knet_h->host_index[host_id]; if (!host) { err = -1; savederrno = EINVAL; log_err(knet_h, KNET_SUB_LINK, "Unable to find host %u: %s", host_id, strerror(savederrno)); goto exit_unlock; } link = &host->link[link_id]; if (!link->configured) { err = -1; savederrno = EINVAL; log_err(knet_h, KNET_SUB_LINK, "host %u link %u is not configured: %s", host_id, link_id, strerror(savederrno)); goto exit_unlock; } *pong_count = link->pong_count; exit_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = savederrno; return err; } int knet_link_set_ping_timers(knet_handle_t knet_h, uint16_t host_id, uint8_t link_id, time_t interval, time_t timeout, unsigned int precision) { int savederrno = 0, err = 0; struct knet_host *host; struct knet_link *link; if (!knet_h) { errno = EINVAL; return -1; } if (link_id >= KNET_MAX_LINK) { errno = EINVAL; return -1; } if (!interval) { errno = EINVAL; return -1; } if (!timeout) { errno = EINVAL; return -1; } if (!precision) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_LINK, "Unable to get write lock: %s", strerror(savederrno)); errno = savederrno; return -1; } host = knet_h->host_index[host_id]; if (!host) { err = -1; savederrno = EINVAL; log_err(knet_h, KNET_SUB_LINK, "Unable to find host %u: %s", host_id, strerror(savederrno)); goto exit_unlock; } link = &host->link[link_id]; if (!link->configured) { err = -1; savederrno = EINVAL; log_err(knet_h, KNET_SUB_LINK, "host %u link %u is not configured: %s", host_id, link_id, strerror(savederrno)); goto exit_unlock; } link->ping_interval = interval * 1000; /* microseconds */ link->pong_timeout = timeout * 1000; /* microseconds */ link->latency_fix = precision; link->latency_exp = precision - \ ((link->ping_interval * precision) / 8000000); log_debug(knet_h, KNET_SUB_LINK, "host: %u link: %u timeout update - interval: %llu timeout: %llu precision: %d", host_id, link_id, link->ping_interval, link->pong_timeout, precision); exit_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = savederrno; return err; } int knet_link_get_ping_timers(knet_handle_t knet_h, uint16_t host_id, uint8_t link_id, time_t *interval, time_t *timeout, unsigned int *precision) { int savederrno = 0, err = 0; struct knet_host *host; struct knet_link *link; if (!knet_h) { errno = EINVAL; return -1; } if (link_id >= KNET_MAX_LINK) { errno = EINVAL; return -1; } if (!interval) { errno = EINVAL; return -1; } if (!timeout) { errno = EINVAL; return -1; } if (!precision) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_LINK, "Unable to get read lock: %s", strerror(savederrno)); errno = savederrno; return -1; } host = knet_h->host_index[host_id]; if (!host) { err = -1; savederrno = EINVAL; log_err(knet_h, KNET_SUB_LINK, "Unable to find host %u: %s", host_id, strerror(savederrno)); goto exit_unlock; } link = &host->link[link_id]; if (!link->configured) { err = -1; savederrno = EINVAL; log_err(knet_h, KNET_SUB_LINK, "host %u link %u is not configured: %s", host_id, link_id, strerror(savederrno)); goto exit_unlock; } *interval = link->ping_interval / 1000; /* microseconds */ *timeout = link->pong_timeout / 1000; *precision = link->latency_fix; exit_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = savederrno; return err; } int knet_link_set_priority(knet_handle_t knet_h, uint16_t host_id, uint8_t link_id, uint8_t priority) { int savederrno = 0, err = 0; struct knet_host *host; struct knet_link *link; uint8_t old_priority; if (!knet_h) { errno = EINVAL; return -1; } if (link_id >= KNET_MAX_LINK) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_LINK, "Unable to get write lock: %s", strerror(savederrno)); errno = savederrno; return -1; } host = knet_h->host_index[host_id]; if (!host) { err = -1; savederrno = EINVAL; log_err(knet_h, KNET_SUB_LINK, "Unable to find host %u: %s", host_id, strerror(savederrno)); goto exit_unlock; } link = &host->link[link_id]; if (!link->configured) { err = -1; savederrno = EINVAL; log_err(knet_h, KNET_SUB_LINK, "host %u link %u is not configured: %s", host_id, link_id, strerror(savederrno)); goto exit_unlock; } old_priority = link->priority; if (link->priority == priority) { err = 0; goto exit_unlock; } link->priority = priority; if (_host_dstcache_update_async(knet_h, host)) { savederrno = errno; log_debug(knet_h, KNET_SUB_LINK, "Unable to update link priority (host: %u link: %u priority: %u): %s", host_id, link_id, link->priority, strerror(savederrno)); link->priority = old_priority; err = -1; goto exit_unlock; } log_debug(knet_h, KNET_SUB_LINK, "host: %u link: %u priority set to: %u", host_id, link_id, link->priority); exit_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = savederrno; return err; } int knet_link_get_priority(knet_handle_t knet_h, uint16_t host_id, uint8_t link_id, uint8_t *priority) { int savederrno = 0, err = 0; struct knet_host *host; struct knet_link *link; if (!knet_h) { errno = EINVAL; return -1; } if (link_id >= KNET_MAX_LINK) { errno = EINVAL; return -1; } if (!priority) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_LINK, "Unable to get read lock: %s", strerror(savederrno)); errno = savederrno; return -1; } host = knet_h->host_index[host_id]; if (!host) { err = -1; savederrno = EINVAL; log_err(knet_h, KNET_SUB_LINK, "Unable to find host %u: %s", host_id, strerror(savederrno)); goto exit_unlock; } link = &host->link[link_id]; if (!link->configured) { err = -1; savederrno = EINVAL; log_err(knet_h, KNET_SUB_LINK, "host %u link %u is not configured: %s", host_id, link_id, strerror(savederrno)); goto exit_unlock; } *priority = link->priority; exit_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = savederrno; return err; } int knet_link_get_link_list(knet_handle_t knet_h, uint16_t host_id, uint8_t *link_ids, size_t *link_ids_entries) { int savederrno = 0, err = 0, i, count = 0; struct knet_host *host; struct knet_link *link; if (!knet_h) { errno = EINVAL; return -1; } if (!link_ids) { errno = EINVAL; return -1; } if (!link_ids_entries) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_LINK, "Unable to get read lock: %s", strerror(savederrno)); errno = savederrno; return -1; } host = knet_h->host_index[host_id]; if (!host) { err = -1; savederrno = EINVAL; log_err(knet_h, KNET_SUB_LINK, "Unable to find host %u: %s", host_id, strerror(savederrno)); goto exit_unlock; } for (i = 0; i < KNET_MAX_LINK; i++) { link = &host->link[i]; if (!link->configured) { continue; } link_ids[count] = i; count++; } *link_ids_entries = count; exit_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = savederrno; return err; } int knet_link_get_status(knet_handle_t knet_h, uint16_t host_id, uint8_t link_id, struct knet_link_status *status) { int savederrno = 0, err = 0; struct knet_host *host; struct knet_link *link; if (!knet_h) { errno = EINVAL; return -1; } if (link_id >= KNET_MAX_LINK) { errno = EINVAL; return -1; } if (!status) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_LINK, "Unable to get read lock: %s", strerror(savederrno)); errno = savederrno; return -1; } host = knet_h->host_index[host_id]; if (!host) { err = -1; savederrno = EINVAL; log_err(knet_h, KNET_SUB_LINK, "Unable to find host %u: %s", host_id, strerror(savederrno)); goto exit_unlock; } link = &host->link[link_id]; if (!link->configured) { err = -1; savederrno = EINVAL; log_err(knet_h, KNET_SUB_LINK, "host %u link %u is not configured: %s", host_id, link_id, strerror(savederrno)); goto exit_unlock; } memmove(status, &link->status, sizeof(struct knet_link_status)); exit_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = savederrno; return err; } diff --git a/libknet/listener.c b/libknet/listener.c index a58cbc06..92c9c5e1 100644 --- a/libknet/listener.c +++ b/libknet/listener.c @@ -1,164 +1,166 @@ /* * Copyright (C) 2010-2015 Red Hat, Inc. All rights reserved. * * Authors: Fabio M. Di Nitto * Federico Simoncelli * * This software licensed under GPL-2.0+, LGPL-2.0+ */ #include "config.h" #include #include #include #include #include #include #include "internals.h" #include "common.h" #include "logging.h" #include "listener.h" #include "transports.h" int _listener_add(knet_handle_t knet_h, uint16_t host_id, uint8_t link_id) { int count = 0; int savederrno = 0, err = 0; struct knet_link *lnk = &knet_h->host_index[host_id]->link[link_id]; struct knet_listener *listener = NULL; savederrno = pthread_rwlock_wrlock(&knet_h->listener_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_LISTENER, "Unable to get write lock: %s", strerror(savederrno)); errno = savederrno; return -1; } listener = knet_h->listener_head; while (listener) { count++; log_debug(knet_h, KNET_SUB_LISTENER, "checking listener: %d", count); if (!memcmp(&lnk->src_addr, &listener->address, sizeof(struct sockaddr_storage))) { log_debug(knet_h, KNET_SUB_LISTENER, "found active listener"); break; } listener = listener->next; } if (!listener) { listener = malloc(sizeof(struct knet_listener)); if (!listener) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_LISTENER, "out of memory to allocate listener: %s", strerror(savederrno)); goto exit_unlock; } memset(listener, 0, sizeof(struct knet_listener)); memmove(&listener->address, &lnk->src_addr, sizeof(struct sockaddr_storage)); - if (knet_h->transport_ops[lnk->transport_type]->link_listener_start(knet_h, lnk->transport, link_id, - &lnk->src_addr, &lnk->dst_addr) < 0) { + if ((knet_h->transport_ops[lnk->transport_type]) && + (knet_h->transport_ops[lnk->transport_type]->link_listener_start) && + (knet_h->transport_ops[lnk->transport_type]->link_listener_start(knet_h, lnk->transport, link_id, + &lnk->src_addr, &lnk->dst_addr) < 0)) { savederrno = errno; err = -1; free(listener); listener = NULL; log_err(knet_h, KNET_SUB_LISTENER, "Unable to start listener for this link"); goto exit_unlock; } /* pushing new host to the front */ listener->next = knet_h->listener_head; knet_h->listener_head = listener; } lnk->listener_sock = listener->sock; exit_unlock: if ((err) && (listener)) { if (listener->sock >= 0) { close(listener->sock); } free(listener); listener = NULL; } pthread_rwlock_unlock(&knet_h->listener_rwlock); errno = savederrno; return err; } int _listener_remove(knet_handle_t knet_h, uint16_t host_id, uint8_t link_id) { int err = 0, savederrno = 0; int link_idx; struct epoll_event ev; /* kernel < 2.6.9 bug (see epoll_ctl man) */ struct knet_host *host; struct knet_link *lnk = &knet_h->host_index[host_id]->link[link_id]; struct knet_listener *tmp_listener; struct knet_listener *listener; int listener_cnt = 0; savederrno = pthread_rwlock_wrlock(&knet_h->listener_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_LISTENER, "Unable to get write lock: %s", strerror(savederrno)); errno = savederrno; return -1; } /* checking if listener is in use */ 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].status.enabled != 1) continue; if (host->link[link_idx].listener_sock == lnk->listener_sock) { listener_cnt++; } } } if (listener_cnt) { lnk->listener_sock = 0; log_debug(knet_h, KNET_SUB_LISTENER, "listener_remove: listener still in use"); savederrno = EBUSY; err = -1; goto exit_unlock; } listener = knet_h->listener_head; while (listener) { if (listener->sock == lnk->listener_sock) break; listener = listener->next; } /* TODO: use a doubly-linked list? */ if (listener == knet_h->listener_head) { knet_h->listener_head = knet_h->listener_head->next; } else { for (tmp_listener = knet_h->listener_head; tmp_listener != NULL; tmp_listener = tmp_listener->next) { if (listener == tmp_listener->next) { tmp_listener->next = tmp_listener->next->next; break; } } } knet_h->transport_ops[lnk->transport_type]->link_free(lnk->transport); lnk->transport = NULL; epoll_ctl(knet_h->recv_from_links_epollfd, EPOLL_CTL_DEL, listener->sock, &ev); close(listener->sock); free(listener); exit_unlock: pthread_rwlock_unlock(&knet_h->listener_rwlock); errno = savederrno; return err; } diff --git a/libknet/threads_pmtud.c b/libknet/threads_pmtud.c index 3011416e..46f53497 100644 --- a/libknet/threads_pmtud.c +++ b/libknet/threads_pmtud.c @@ -1,388 +1,396 @@ /* * Copyright (C) 2010-2015 Red Hat, Inc. All rights reserved. * * Authors: Fabio M. Di Nitto * Federico Simoncelli * * This software licensed under GPL-2.0+, LGPL-2.0+ */ #include "config.h" #include #include #include #include #include "crypto.h" #include "link.h" #include "host.h" #include "logging.h" #include "threads_common.h" #include "threads_pmtud.h" static int _handle_check_link_pmtud(knet_handle_t knet_h, struct knet_host *dst_host, struct knet_link *dst_link) { int ret, savederrno, mutex_retry_limit, failsafe; ssize_t onwire_len; /* current packet onwire size */ ssize_t overhead_len; /* onwire packet overhead (protocol based) */ ssize_t max_mtu_len; /* max mtu for protocol */ ssize_t data_len; /* how much data we can send in the packet * generally would be onwire_len - overhead_len * needs to be adjusted for crypto */ ssize_t pad_len; /* crypto packet pad size, needs to move into crypto.c callbacks */ int len; /* len of what we were able to sendto onwire */ struct timespec ts; unsigned char *outbuf = (unsigned char *)knet_h->pmtudbuf; mutex_retry_limit = 0; failsafe = 0; pad_len = 0; - dst_link->last_bad_mtu = knet_h->transport_ops[dst_link->transport_type]->link_get_mtu_overhead(dst_link->transport); + /* + * FIXME: proto overhead should be included in overhead_len + */ + if ((knet_h->transport_ops[dst_link->transport_type]) && + (knet_h->transport_ops[dst_link->transport_type]->link_get_mtu_overhead)) { + dst_link->last_bad_mtu = knet_h->transport_ops[dst_link->transport_type]->link_get_mtu_overhead(dst_link->transport); + } else { + dst_link->last_bad_mtu = 0; + } knet_h->pmtudbuf->khp_pmtud_link = dst_link->link_id; switch (dst_link->dst_addr.ss_family) { case AF_INET6: max_mtu_len = KNET_PMTUD_SIZE_V6; overhead_len = KNET_PMTUD_OVERHEAD_V6; dst_link->last_good_mtu = dst_link->last_ping_size + KNET_PMTUD_OVERHEAD_V6; break; case AF_INET: max_mtu_len = KNET_PMTUD_SIZE_V4; overhead_len = KNET_PMTUD_OVERHEAD_V4; dst_link->last_good_mtu = dst_link->last_ping_size + KNET_PMTUD_OVERHEAD_V6; break; default: log_debug(knet_h, KNET_SUB_PMTUD_T, "PMTUD aborted, unknown protocol"); return -1; break; } /* * discovery starts from the top because kernel will * refuse to send packets > current iface mtu. * this saves us some time and network bw. */ onwire_len = max_mtu_len; restart: /* * prevent a race when interface mtu is changed _exactly_ during * the discovery process and it's complex to detect. Easier * to wait the next loop. * 30 is not an arbitrary value. To bisect from 576 to 128000 doesn't * take more than 18/19 steps. */ if (failsafe == 30) { log_err(knet_h, KNET_SUB_PMTUD_T, "Aborting PMTUD process: Too many attempts. MTU might have changed during discovery."); return -1; } else { failsafe++; } data_len = onwire_len - overhead_len; if (knet_h->crypto_instance) { if (knet_h->sec_block_size) { pad_len = knet_h->sec_block_size - (data_len % knet_h->sec_block_size); if (pad_len == knet_h->sec_block_size) { pad_len = 0; } data_len = data_len + pad_len; } data_len = data_len + (knet_h->sec_hash_size + knet_h->sec_salt_size + knet_h->sec_block_size); if (knet_h->sec_block_size) { while (data_len + overhead_len >= max_mtu_len) { data_len = data_len - knet_h->sec_block_size; } } if (dst_link->last_bad_mtu) { while (data_len + overhead_len >= dst_link->last_bad_mtu) { data_len = data_len - (knet_h->sec_hash_size + knet_h->sec_salt_size + knet_h->sec_block_size); } } if (data_len < (knet_h->sec_hash_size + knet_h->sec_salt_size + knet_h->sec_block_size) + 1) { log_debug(knet_h, KNET_SUB_PMTUD_T, "Aborting PMTUD process: link mtu smaller than crypto header detected (link might have been disconnected)"); return -1; } onwire_len = data_len + overhead_len; knet_h->pmtudbuf->khp_pmtud_size = onwire_len; if (crypto_encrypt_and_sign(knet_h, (const unsigned char *)knet_h->pmtudbuf, data_len - (knet_h->sec_hash_size + knet_h->sec_salt_size + knet_h->sec_block_size), knet_h->pmtudbuf_crypt, &data_len) < 0) { log_debug(knet_h, KNET_SUB_PMTUD_T, "Unable to crypto pmtud packet"); return -1; } outbuf = knet_h->pmtudbuf_crypt; } else { knet_h->pmtudbuf->khp_pmtud_size = onwire_len; } /* link has gone down, aborting pmtud */ if (dst_link->status.connected != 1) { log_debug(knet_h, KNET_SUB_PMTUD_T, "PMTUD detected host (%u) link (%u) has been disconnected", dst_host->host_id, dst_link->link_id); return -1; } if (pthread_mutex_lock(&knet_h->pmtud_mutex) != 0) { log_debug(knet_h, KNET_SUB_PMTUD_T, "Unable to get mutex lock"); return -1; } len = sendto(dst_link->outsock, outbuf, data_len, MSG_DONTWAIT | MSG_NOSIGNAL, (struct sockaddr *) &dst_link->dst_addr, sizeof(struct sockaddr_storage)); savederrno = errno; if ((len < 0) && (savederrno != EMSGSIZE)) { log_debug(knet_h, KNET_SUB_PMTUD_T, "Unable to send pmtu packet (sendto): %d %s", savederrno, strerror(savederrno)); pthread_mutex_unlock(&knet_h->pmtud_mutex); return -1; } if (len != data_len) { /* * this is coming from "localhost" already. */ if (savederrno == EMSGSIZE) { dst_link->last_bad_mtu = onwire_len; } else { log_debug(knet_h, KNET_SUB_PMTUD_T, "Unable to send pmtu packet len: %zu err: %s", onwire_len, strerror(savederrno)); } } else { dst_link->last_sent_mtu = onwire_len; dst_link->last_recv_mtu = 0; if (clock_gettime(CLOCK_REALTIME, &ts) < 0) { log_debug(knet_h, KNET_SUB_PMTUD_T, "Unable to get current time: %s", strerror(errno)); pthread_mutex_unlock(&knet_h->pmtud_mutex); return -1; } /* * Set an artibrary 2 seconds timeout to receive a PMTUd reply * perhaps this should be configurable but: * 1) too short timeout can cause instability since MTU value * influeces link status * 2) too high timeout slows down the MTU detection process for * small MTU * * Another option is to make the PMTUd process less influent * in link status detection but that could cause data packet loss * without link up/down changes */ ts.tv_sec += 2; ret = pthread_cond_timedwait(&knet_h->pmtud_cond, &knet_h->pmtud_mutex, &ts); if (shutdown_in_progress(knet_h)) { pthread_mutex_unlock(&knet_h->pmtud_mutex); log_debug(knet_h, KNET_SUB_PMTUD_T, "PMTUD aborted. shutdown in progress"); return -1; } if ((ret != 0) && (ret != ETIMEDOUT)) { pthread_mutex_unlock(&knet_h->pmtud_mutex); if (mutex_retry_limit == 3) { log_debug(knet_h, KNET_SUB_PMTUD_T, "PMTUD aborted, unable to get mutex lock"); return -1; } mutex_retry_limit++; goto restart; } if ((dst_link->last_recv_mtu != onwire_len) || (ret)) { dst_link->last_bad_mtu = onwire_len; } else { int found_mtu = 0; if (knet_h->sec_block_size) { if ((onwire_len + knet_h->sec_block_size >= max_mtu_len) || ((dst_link->last_bad_mtu) && (dst_link->last_bad_mtu <= (onwire_len + knet_h->sec_block_size)))) { found_mtu = 1; } } else { if ((onwire_len == max_mtu_len) || ((dst_link->last_bad_mtu) && (dst_link->last_bad_mtu == (onwire_len + 1)))) { found_mtu = 1; } } if (found_mtu) { /* * account for IP overhead, knet headers and crypto in PMTU calculation */ dst_link->status.mtu = onwire_len - dst_link->status.proto_overhead; pthread_mutex_unlock(&knet_h->pmtud_mutex); return 0; } dst_link->last_good_mtu = onwire_len; } } onwire_len = (dst_link->last_good_mtu + dst_link->last_bad_mtu) / 2; pthread_mutex_unlock(&knet_h->pmtud_mutex); goto restart; } static int _handle_check_pmtud(knet_handle_t knet_h, struct knet_host *dst_host, struct knet_link *dst_link, unsigned int *min_mtu) { uint8_t saved_valid_pmtud; unsigned int saved_pmtud; struct timespec clock_now; unsigned long long diff_pmtud, interval; interval = knet_h->pmtud_interval * 1000000000llu; /* nanoseconds */ if (clock_gettime(CLOCK_MONOTONIC, &clock_now) != 0) { log_debug(knet_h, KNET_SUB_PMTUD_T, "Unable to get monotonic clock"); return 0; } timespec_diff(dst_link->pmtud_last, clock_now, &diff_pmtud); if (diff_pmtud < interval) { *min_mtu = dst_link->status.mtu; return dst_link->has_valid_mtu; } switch (dst_link->dst_addr.ss_family) { case AF_INET6: dst_link->status.proto_overhead = KNET_PMTUD_OVERHEAD_V6 + KNET_HEADER_ALL_SIZE + knet_h->sec_header_size; break; case AF_INET: dst_link->status.proto_overhead = KNET_PMTUD_OVERHEAD_V4 + KNET_HEADER_ALL_SIZE + knet_h->sec_header_size; break; } saved_pmtud = dst_link->status.mtu; saved_valid_pmtud = dst_link->has_valid_mtu; log_debug(knet_h, KNET_SUB_PMTUD_T, "Starting PMTUD for host: %u link: %u", dst_host->host_id, dst_link->link_id); if (_handle_check_link_pmtud(knet_h, dst_host, dst_link) < 0) { dst_link->has_valid_mtu = 0; } else { dst_link->has_valid_mtu = 1; switch (dst_link->dst_addr.ss_family) { case AF_INET6: if (((dst_link->status.mtu + dst_link->status.proto_overhead) < KNET_PMTUD_MIN_MTU_V6) || ((dst_link->status.mtu + dst_link->status.proto_overhead) > KNET_PMTUD_SIZE_V6)) { log_debug(knet_h, KNET_SUB_PMTUD_T, "PMTUD detected an IPv6 MTU out of bound value (%u) for host: %u link: %u.", dst_link->status.mtu + dst_link->status.proto_overhead, dst_host->host_id, dst_link->link_id); dst_link->has_valid_mtu = 0; } break; case AF_INET: if (((dst_link->status.mtu + dst_link->status.proto_overhead) < KNET_PMTUD_MIN_MTU_V4) || ((dst_link->status.mtu + dst_link->status.proto_overhead) > KNET_PMTUD_SIZE_V4)) { log_debug(knet_h, KNET_SUB_PMTUD_T, "PMTUD detected an IPv4 MTU out of bound value (%u) for host: %u link: %u.", dst_link->status.mtu + dst_link->status.proto_overhead, dst_host->host_id, dst_link->link_id); dst_link->has_valid_mtu = 0; } break; } if (dst_link->has_valid_mtu) { if ((saved_pmtud) && (saved_pmtud != dst_link->status.mtu)) { log_info(knet_h, KNET_SUB_PMTUD_T, "PMTUD link change for host: %u link: %u from %u to %u", dst_host->host_id, dst_link->link_id, saved_pmtud, dst_link->status.mtu); } log_debug(knet_h, KNET_SUB_PMTUD_T, "PMTUD completed for host: %u link: %u current link mtu: %u", dst_host->host_id, dst_link->link_id, dst_link->status.mtu); if (dst_link->status.mtu < *min_mtu) { *min_mtu = dst_link->status.mtu; } dst_link->pmtud_last = clock_now; } } if (saved_valid_pmtud != dst_link->has_valid_mtu) { _host_dstcache_update_sync(knet_h, dst_host); } return dst_link->has_valid_mtu; } void *_handle_pmtud_link_thread(void *data) { knet_handle_t knet_h = (knet_handle_t) data; struct knet_host *dst_host; struct knet_link *dst_link; int link_idx; unsigned int min_mtu, have_mtu; knet_h->data_mtu = KNET_PMTUD_MIN_MTU_V4 - KNET_HEADER_ALL_SIZE - knet_h->sec_header_size; /* preparing pmtu buffer */ knet_h->pmtudbuf->kh_version = KNET_HEADER_VERSION; knet_h->pmtudbuf->kh_type = KNET_HEADER_TYPE_PMTUD; knet_h->pmtudbuf->kh_node = htons(knet_h->host_id); while (!shutdown_in_progress(knet_h)) { usleep(KNET_THREADS_TIMERES); if (pthread_rwlock_rdlock(&knet_h->global_rwlock) != 0) { log_debug(knet_h, KNET_SUB_PMTUD_T, "Unable to get read lock"); continue; } min_mtu = KNET_PMTUD_SIZE_V4 - KNET_HEADER_ALL_SIZE - knet_h->sec_header_size; have_mtu = 0; for (dst_host = knet_h->host_head; dst_host != NULL; dst_host = dst_host->next) { for (link_idx = 0; link_idx < KNET_MAX_LINK; link_idx++) { dst_link = &dst_host->link[link_idx]; if ((dst_link->status.enabled != 1) || (dst_link->status.connected != 1) || (!dst_link->last_ping_size) || ((dst_link->dynamic == KNET_LINK_DYNIP) && (dst_link->status.dynconnected != 1))) continue; if (_handle_check_pmtud(knet_h, dst_host, dst_link, &min_mtu)) { have_mtu = 1; } } } if (have_mtu) { if (knet_h->data_mtu != min_mtu) { knet_h->data_mtu = min_mtu; log_info(knet_h, KNET_SUB_PMTUD_T, "Global data MTU changed to: %u", knet_h->data_mtu); if (knet_h->pmtud_notify_fn) { knet_h->pmtud_notify_fn(knet_h->pmtud_notify_fn_private_data, knet_h->data_mtu); } } } pthread_rwlock_unlock(&knet_h->global_rwlock); } return NULL; } diff --git a/libknet/transport_common.c b/libknet/transport_common.c index b65e93a3..618f0123 100644 --- a/libknet/transport_common.c +++ b/libknet/transport_common.c @@ -1,176 +1,177 @@ #include "config.h" #include #include #include #include #include #include #include #include #include #include #include #include #include "libknet.h" #include "host.h" #include "link.h" #include "logging.h" #include "common.h" #include "transports.h" #include "../common/netutils.h" - int _configure_transport_socket(knet_handle_t knet_h, int sock, struct sockaddr_storage *address, const char *type) { int err = 0; int value; int savederrno; value = KNET_RING_RCVBUFF; if (setsockopt(sock, SOL_SOCKET, SO_RCVBUFFORCE, &value, sizeof(value)) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT_T, "Unable to set %s receive buffer: %s", type, strerror(savederrno)); goto exit_error; } value = KNET_RING_RCVBUFF; if (setsockopt(sock, SOL_SOCKET, SO_SNDBUFFORCE, &value, sizeof(value)) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT_T, "Unable to set %s send buffer: %s", type, strerror(savederrno)); goto exit_error; } value = 1; if (setsockopt(sock, SOL_IP, IP_FREEBIND, &value, sizeof(value)) <0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT_T, "Unable to set FREEBIND on %s socket: %s", type, strerror(savederrno)); goto exit_error; } if (address->ss_family == AF_INET6) { value = 1; if (setsockopt(sock, IPPROTO_IPV6, IPV6_V6ONLY, &value, sizeof(value)) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT_T, "Unable to set %s IPv6 only: %s", type, strerror(savederrno)); goto exit_error; } value = IPV6_PMTUDISC_PROBE; if (setsockopt(sock, SOL_IPV6, IPV6_MTU_DISCOVER, &value, sizeof(value)) <0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT_T, "Unable to set PMTUDISC on %s socket: %s", type, strerror(savederrno)); goto exit_error; } } else { value = IP_PMTUDISC_PROBE; if (setsockopt(sock, SOL_IP, IP_MTU_DISCOVER, &value, sizeof(value)) <0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT_T, "Unable to set PMTUDISC on %s socket: %s", type, strerror(savederrno)); goto exit_error; } } value = 1; if (setsockopt(sock, SOL_SOCKET, SO_REUSEADDR, &value, sizeof(value)) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT_T, "Unable to set %s reuseaddr: %s", type, strerror(savederrno)); goto exit_error; } if (_fdset_cloexec(sock)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT_T, "Unable to set %s CLOEXEC socket opts: %s", type, strerror(savederrno)); goto exit_error; } if (_fdset_nonblock(sock)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT_T, "Unable to set %s NONBLOCK socket opts: %s", type, strerror(savederrno)); goto exit_error; } err = 0; exit_error: return err; } void _close_socket(knet_handle_t knet_h, int sockfd) { struct epoll_event ev; int i; log_err(knet_h, KNET_SUB_LINK_T, "EOF received on socket fd %d", sockfd); 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_LISTENER, "Unable to remove EOFed socket from epoll pool: %s", strerror(errno)); } /* Tell transport that the FD has been closed */ for (i=0; itransports[i] && - !knet_h->transport_ops[i]->handle_fd_eof(knet_h, sockfd)) + if ((knet_h->transport_ops[i]) && + (knet_h->transport_ops[i]->handle_fd_eof) && + (!knet_h->transport_ops[i]->handle_fd_eof(knet_h, sockfd))) break; } } void _handle_socket_notification(knet_handle_t knet_h, int sockfd, struct iovec *iov, size_t iovlen) { int i; /* Find the transport and post the message */ for (i=0; itransports[i] && knet_h->transport_ops[i]->handle_fd_notification && - knet_h->transport_ops[i]->handle_fd_notification(knet_h, sockfd, iov, iovlen)) + if ((knet_h->transport_ops[i]) && + (knet_h->transport_ops[i]->handle_fd_notification) && + (knet_h->transport_ops[i]->handle_fd_notification(knet_h, sockfd, iov, iovlen))) break; } } /* * Wrappers for addrtostr() & addrtostr_free() for use when we only need the IP address * printing in DEBUG mode - it's to heavy for within normal use */ int _transport_addrtostr(const struct sockaddr *sa, socklen_t salen, char *str[2]) { #ifdef DEBUG return addrtostr(sa, salen, str); #else str[0] = (char*)"node"; str[1] = (char*)""; return 0; #endif } void _transport_addrtostr_free(char *str[2]) { #ifdef DEBUG addrtostr_free(str); #else #endif }