diff --git a/libknet/internals.h b/libknet/internals.h index 3e4bbfaf..20f7e105 100644 --- a/libknet/internals.h +++ b/libknet/internals.h @@ -1,514 +1,527 @@ /* * Copyright (C) 2010-2018 Red Hat, Inc. All rights reserved. * * Authors: Fabio M. Di Nitto * Federico Simoncelli * * This software licensed under GPL-2.0+, LGPL-2.0+ */ #ifndef __KNET_INTERNALS_H__ #define __KNET_INTERNALS_H__ /* * NOTE: you shouldn't need to include this header normally */ #include #include "libknet.h" #include "onwire.h" #include "compat.h" #include "threads_common.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_DATABUFSIZE_COMPRESS_PAD 1024 #define KNET_DATABUFSIZE_COMPRESS KNET_DATABUFSIZE + KNET_DATABUFSIZE_COMPRESS_PAD #define KNET_RING_RCVBUFF 8388608 #define PCKT_FRAG_MAX UINT8_MAX #define PCKT_RX_BUFS 512 #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 long long pong_timeout_adj; /* timeout adjusted for latency */ uint8_t pong_timeout_backoff; /* see link.h for definition */ unsigned int latency_fix; /* precision */ uint8_t pong_count; /* how many ping/pong to send/receive before link is up */ uint64_t flags; /* 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 #define KNET_MAX_COMPRESS_METHODS UINT8_MAX struct knet_handle_stats_extra { uint64_t tx_crypt_pmtu_packets; uint64_t tx_crypt_pmtu_reply_packets; uint64_t tx_crypt_ping_packets; uint64_t tx_crypt_pong_packets; }; 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_fd_trackers knet_transport_fd_tracker[KNET_MAX_FDS]; /* track status for each fd handled by transports */ struct knet_handle_stats stats; struct knet_handle_stats_extra stats_extra; uint32_t reconnect_int; knet_node_id_t host_ids[KNET_MAX_HOST]; size_t host_ids_entries; struct knet_header *recv_from_sock_buf; struct knet_header *send_to_links_buf[PCKT_FRAG_MAX]; struct knet_header *recv_from_links_buf[PCKT_RX_BUFS]; struct knet_header *pingbuf; struct knet_header *pmtudbuf; uint8_t threads_status[KNET_THREAD_MAX]; pthread_mutex_t threads_status_mutex; 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; 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 */ pthread_mutex_t backoff_mutex; /* used to protect dst_link->pong_timeout_adj */ pthread_mutex_t kmtu_mutex; /* used to protect kernel_mtu */ uint32_t kernel_mtu; /* contains the MTU detected by the kernel on a given link */ int pmtud_waiting; int pmtud_running; int pmtud_forcerun; int pmtud_abort; struct crypto_instance *crypto_instance; size_t sec_header_size; size_t sec_block_size; size_t sec_hash_size; size_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; int compress_model; int compress_level; size_t compress_threshold; void *compress_int_data[KNET_MAX_COMPRESS_METHODS]; /* for compress method private data */ unsigned char *recv_from_links_buf_decompress; unsigned char *send_to_links_buf_compress; seq_num_t tx_seq_num; pthread_mutex_t tx_seq_num_mutex; uint8_t has_loop_link; uint8_t loop_link; 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; uint64_t flags; }; extern pthread_rwlock_t shlib_rwlock; /* global shared lib load lock */ /* * NOTE: every single operation must be implementend * for every protocol. */ +/* + * make it easier to map values in transports.c + */ +#define TRANSPORT_PROTO_NOT_CONNECTION_ORIENTED 0 +#define TRANSPORT_PROTO_IS_CONNECTION_ORIENTED 1 + typedef struct knet_transport_ops { /* * transport generic information */ const char *transport_name; const uint8_t transport_id; const uint8_t built_in; +/* + * connection oriented protocols like SCTP + * don´t need dst_addr in sendto calls and + * on some OSes are considered EINVAL. + */ + uint8_t transport_is_connection_oriented; + 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 knet_mmsghdr *msg); } knet_transport_ops_t; socklen_t sockaddr_len(const struct sockaddr_storage *ss); struct pretty_names { const char *name; uint8_t val; }; /** * 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/tests/api_knet_send.c b/libknet/tests/api_knet_send.c index e7f84565..3dfb6d0b 100644 --- a/libknet/tests/api_knet_send.c +++ b/libknet/tests/api_knet_send.c @@ -1,320 +1,326 @@ /* * Copyright (C) 2016-2018 Red Hat, Inc. All rights reserved. * * Authors: Fabio M. Di Nitto * * This software licensed under GPL-2.0+, LGPL-2.0+ */ #include "config.h" #include #include #include #include #include #include #include "libknet.h" #include "internals.h" #include "netutils.h" #include "test-common.h" static int private_data; static void sock_notify(void *pvt_data, int datafd, int8_t channel, uint8_t tx_rx, int error, int errorno) { return; } -static void test(void) +static void test(uint8_t transport) { knet_handle_t knet_h; int logfds[2]; int datafd = 0; int8_t channel = 0; struct knet_link_status link_status; char send_buff[KNET_MAX_PACKET_SIZE]; char recv_buff[KNET_MAX_PACKET_SIZE]; ssize_t send_len = 0; int recv_len = 0; int savederrno; struct sockaddr_storage lo; if (make_local_sockaddr(&lo, 0) < 0) { printf("Unable to convert loopback to sockaddr: %s\n", strerror(errno)); exit(FAIL); } memset(send_buff, 0, sizeof(send_buff)); printf("Test knet_send incorrect knet_h\n"); if ((!knet_send(NULL, send_buff, KNET_MAX_PACKET_SIZE, channel)) || (errno != EINVAL)) { printf("knet_send accepted invalid knet_h or returned incorrect error: %s\n", strerror(errno)); exit(FAIL); } setup_logpipes(logfds); knet_h = knet_handle_start(logfds, KNET_LOG_DEBUG); printf("Test knet_send with no send_buff\n"); if ((!knet_send(knet_h, NULL, KNET_MAX_PACKET_SIZE, channel)) || (errno != EINVAL)) { printf("knet_send accepted invalid send_buff or returned incorrect error: %s\n", strerror(errno)); knet_handle_free(knet_h); flush_logs(logfds[0], stdout); close_logpipes(logfds); exit(FAIL); } flush_logs(logfds[0], stdout); printf("Test knet_send with invalid send_buff len (0)\n"); if ((!knet_send(knet_h, send_buff, 0, channel)) || (errno != EINVAL)) { printf("knet_send accepted invalid send_buff len (0) or returned incorrect error: %s\n", strerror(errno)); knet_handle_free(knet_h); flush_logs(logfds[0], stdout); close_logpipes(logfds); exit(FAIL); } flush_logs(logfds[0], stdout); printf("Test knet_send with invalid send_buff len (> KNET_MAX_PACKET_SIZE)\n"); if ((!knet_send(knet_h, send_buff, KNET_MAX_PACKET_SIZE + 1, channel)) || (errno != EINVAL)) { printf("knet_send accepted invalid send_buff len (> KNET_MAX_PACKET_SIZE) or returned incorrect error: %s\n", strerror(errno)); knet_handle_free(knet_h); flush_logs(logfds[0], stdout); close_logpipes(logfds); exit(FAIL); } flush_logs(logfds[0], stdout); printf("Test knet_send with invalid channel (-1)\n"); channel = -1; if ((!knet_send(knet_h, send_buff, KNET_MAX_PACKET_SIZE, channel)) || (errno != EINVAL)) { printf("knet_send accepted invalid channel (-1) or returned incorrect error: %s\n", strerror(errno)); knet_handle_free(knet_h); flush_logs(logfds[0], stdout); close_logpipes(logfds); exit(FAIL); } flush_logs(logfds[0], stdout); printf("Test knet_send with invalid channel (KNET_DATAFD_MAX)\n"); channel = KNET_DATAFD_MAX; if ((!knet_send(knet_h, send_buff, KNET_MAX_PACKET_SIZE, channel)) || (errno != EINVAL)) { printf("knet_send accepted invalid channel (KNET_DATAFD_MAX) or returned incorrect error: %s\n", strerror(errno)); knet_handle_free(knet_h); flush_logs(logfds[0], stdout); close_logpipes(logfds); exit(FAIL); } flush_logs(logfds[0], stdout); printf("Test knet_send with unconfigured channel\n"); channel = 0; if ((!knet_send(knet_h, send_buff, KNET_MAX_PACKET_SIZE, channel)) || (errno != EINVAL)) { printf("knet_send accepted invalid unconfigured channel or returned incorrect error: %s\n", strerror(errno)); knet_handle_free(knet_h); flush_logs(logfds[0], stdout); close_logpipes(logfds); exit(FAIL); } flush_logs(logfds[0], stdout); printf("Test knet_send with valid data\n"); if (knet_handle_enable_sock_notify(knet_h, &private_data, sock_notify) < 0) { printf("knet_handle_enable_sock_notify failed: %s\n", strerror(errno)); knet_handle_free(knet_h); flush_logs(logfds[0], stdout); close_logpipes(logfds); exit(FAIL); } datafd = 0; channel = -1; if (knet_handle_add_datafd(knet_h, &datafd, &channel) < 0) { printf("knet_handle_add_datafd failed: %s\n", strerror(errno)); knet_handle_free(knet_h); flush_logs(logfds[0], stdout); close_logpipes(logfds); exit(FAIL); } if (knet_host_add(knet_h, 1) < 0) { printf("knet_host_add failed: %s\n", strerror(errno)); knet_handle_free(knet_h); flush_logs(logfds[0], stdout); close_logpipes(logfds); exit(FAIL); } - if (knet_link_set_config(knet_h, 1, 0, KNET_TRANSPORT_UDP, &lo, &lo, 0) < 0) { + if (knet_link_set_config(knet_h, 1, 0, transport, &lo, &lo, 0) < 0) { printf("Unable to configure link: %s\n", strerror(errno)); knet_host_remove(knet_h, 1); knet_handle_free(knet_h); flush_logs(logfds[0], stdout); close_logpipes(logfds); exit(FAIL); } if (knet_link_set_enable(knet_h, 1, 0, 1) < 0) { printf("knet_link_set_enable failed: %s\n", strerror(errno)); knet_link_clear_config(knet_h, 1, 0); knet_host_remove(knet_h, 1); knet_handle_free(knet_h); flush_logs(logfds[0], stdout); close_logpipes(logfds); exit(FAIL); } if (knet_handle_setfwd(knet_h, 1) < 0) { printf("knet_handle_setfwd failed: %s\n", strerror(errno)); knet_link_set_enable(knet_h, 1, 0, 0); knet_link_clear_config(knet_h, 1, 0); knet_host_remove(knet_h, 1); knet_handle_free(knet_h); flush_logs(logfds[0], stdout); close_logpipes(logfds); exit(FAIL); } if (wait_for_host(knet_h, 1, 10, logfds[0], stdout) < 0) { printf("timeout waiting for host to be reachable"); knet_link_set_enable(knet_h, 1, 0, 0); knet_link_clear_config(knet_h, 1, 0); knet_host_remove(knet_h, 1); knet_handle_free(knet_h); flush_logs(logfds[0], stdout); close_logpipes(logfds); exit(FAIL); } send_len = knet_send(knet_h, send_buff, KNET_MAX_PACKET_SIZE, channel); if (send_len <= 0) { printf("knet_send failed: %s\n", strerror(errno)); knet_link_set_enable(knet_h, 1, 0, 0); knet_link_clear_config(knet_h, 1, 0); knet_host_remove(knet_h, 1); knet_handle_free(knet_h); flush_logs(logfds[0], stdout); close_logpipes(logfds); exit(FAIL); } if (send_len != sizeof(send_buff)) { printf("knet_send sent only %zd bytes: %s\n", send_len, strerror(errno)); knet_link_set_enable(knet_h, 1, 0, 0); knet_link_clear_config(knet_h, 1, 0); knet_host_remove(knet_h, 1); knet_handle_free(knet_h); flush_logs(logfds[0], stdout); close_logpipes(logfds); exit(FAIL); } flush_logs(logfds[0], stdout); if (wait_for_packet(knet_h, 10, datafd)) { printf("Error waiting for packet: %s\n", strerror(errno)); knet_link_set_enable(knet_h, 1, 0, 0); knet_link_clear_config(knet_h, 1, 0); knet_host_remove(knet_h, 1); knet_handle_free(knet_h); flush_logs(logfds[0], stdout); close_logpipes(logfds); exit(FAIL); } recv_len = knet_recv(knet_h, recv_buff, KNET_MAX_PACKET_SIZE, channel); savederrno = errno; if (recv_len != send_len) { printf("knet_recv received only %d bytes: %s (errno: %d)\n", recv_len, strerror(errno), errno); knet_link_set_enable(knet_h, 1, 0, 0); knet_link_clear_config(knet_h, 1, 0); knet_host_remove(knet_h, 1); knet_handle_free(knet_h); flush_logs(logfds[0], stdout); close_logpipes(logfds); if ((is_helgrind()) && (recv_len == -1) && (savederrno == EAGAIN)) { printf("helgrind exception. this is normal due to possible timeouts\n"); exit(PASS); } exit(FAIL); } if (memcmp(recv_buff, send_buff, KNET_MAX_PACKET_SIZE)) { printf("recv and send buffers are different!\n"); knet_link_set_enable(knet_h, 1, 0, 0); knet_link_clear_config(knet_h, 1, 0); knet_host_remove(knet_h, 1); knet_handle_free(knet_h); flush_logs(logfds[0], stdout); close_logpipes(logfds); exit(FAIL); } /* A sanity check on the stats */ if (knet_link_get_status(knet_h, 1, 0, &link_status, sizeof(link_status)) < 0) { printf("knet_link_get_status failed: %s\n", strerror(errno)); knet_link_set_enable(knet_h, 1, 0, 0); knet_link_clear_config(knet_h, 1, 0); knet_host_remove(knet_h, 1); knet_handle_free(knet_h); flush_logs(logfds[0], stdout); close_logpipes(logfds); exit(FAIL); } if (link_status.stats.tx_data_packets != 2 || link_status.stats.rx_data_packets != 2 || link_status.stats.tx_data_bytes < KNET_MAX_PACKET_SIZE || link_status.stats.rx_data_bytes < KNET_MAX_PACKET_SIZE || link_status.stats.tx_data_bytes > KNET_MAX_PACKET_SIZE*2 || link_status.stats.rx_data_bytes > KNET_MAX_PACKET_SIZE*2) { printf("stats look wrong: tx_packets: %" PRIu64 " (%" PRIu64 " bytes), rx_packets: %" PRIu64 " (%" PRIu64 " bytes)\n", link_status.stats.tx_data_packets, link_status.stats.tx_data_bytes, link_status.stats.rx_data_packets, link_status.stats.rx_data_bytes); } flush_logs(logfds[0], stdout); knet_link_set_enable(knet_h, 1, 0, 0); knet_link_clear_config(knet_h, 1, 0); knet_host_remove(knet_h, 1); knet_handle_free(knet_h); flush_logs(logfds[0], stdout); close_logpipes(logfds); } int main(int argc, char *argv[]) { - test(); + printf("Testing with UDP\n"); + test(KNET_TRANSPORT_UDP); + +#ifdef HAVE_NETINET_SCTP_H + printf("Testing with SCTP\n"); + test(KNET_TRANSPORT_SCTP); +#endif return PASS; } diff --git a/libknet/tests/knet_bench.c b/libknet/tests/knet_bench.c index 2871feb8..f60b5b1b 100644 --- a/libknet/tests/knet_bench.c +++ b/libknet/tests/knet_bench.c @@ -1,1296 +1,1296 @@ /* * Copyright (C) 2016-2018 Red Hat, Inc. All rights reserved. * * Authors: Fabio M. Di Nitto * * This software licensed under GPL-2.0+, LGPL-2.0+ */ #include "config.h" #include #include #include #include #include #include #include #include #include #include "libknet.h" #include "compat.h" #include "internals.h" #include "netutils.h" #include "transport_common.h" #include "threads_common.h" #include "test-common.h" #define MAX_NODES 128 static int senderid = -1; static int thisnodeid = -1; static knet_handle_t knet_h; static int datafd = 0; static int8_t channel = 0; static int globallistener = 0; static int continous = 0; static int show_stats = 0; static struct sockaddr_storage allv4; static struct sockaddr_storage allv6; static int broadcast_test = 1; static pthread_t rx_thread = (pthread_t)NULL; static char *rx_buf[PCKT_FRAG_MAX]; static int wait_for_perf_rx = 0; static char *compresscfg = NULL; static char *cryptocfg = NULL; static int machine_output = 0; static int bench_shutdown_in_progress = 0; static pthread_mutex_t shutdown_mutex = PTHREAD_MUTEX_INITIALIZER; #define TEST_PING 0 #define TEST_PING_AND_DATA 1 #define TEST_PERF_BY_SIZE 2 #define TEST_PERF_BY_TIME 3 static int test_type = TEST_PING; #define TEST_START 2 #define TEST_STOP 4 #define TEST_COMPLETE 6 #define ONE_GIGABYTE 1073741824 static uint64_t perf_by_size_size = 1 * ONE_GIGABYTE; static uint64_t perf_by_time_secs = 10; struct node { int nodeid; int links; uint8_t transport[KNET_MAX_LINK]; struct sockaddr_storage address[KNET_MAX_LINK]; }; static void print_help(void) { printf("knet_bench usage:\n"); printf(" -h print this help (no really)\n"); printf(" -d enable debug logs (default INFO)\n"); printf(" -c [implementation]:[crypto]:[hashing] crypto configuration. (default disabled)\n"); printf(" Example: -c nss:aes128:sha1\n"); printf(" -z [implementation]:[level]:[threshold] compress configuration. (default disabled)\n"); printf(" Example: -z zlib:5:100\n"); printf(" -p [active|passive|rr] (default: passive)\n"); printf(" -P [UDP|SCTP] (default: UDP) protocol (transport) to use for all links\n"); printf(" -t [nodeid] This nodeid (required)\n"); printf(" -n [nodeid],[proto]/[link1_ip],[link2_..] Other nodes information (at least one required)\n"); printf(" Example: -n 1,192.168.8.1,SCTP/3ffe::8:1,UDP/172...\n"); printf(" can be repeated up to %d and should contain also the localnode info\n", MAX_NODES); printf(" -b [port] baseport (default: 50000)\n"); printf(" -l enable global listener on 0.0.0.0/:: (default: off, incompatible with -o)\n"); printf(" -o enable baseport offset per nodeid\n"); printf(" -m change PMTUd interval in seconds (default: 60)\n"); printf(" -w dont wait for all nodes to be up before starting the test (default: wait)\n"); printf(" -T [ping|ping_data|perf-by-size|perf-by-time]\n"); printf(" test type (default: ping)\n"); printf(" ping: will wait for all hosts to join the knet network, sleep 5 seconds and quit\n"); printf(" ping_data: will wait for all hosts to join the knet network, sends some data to all nodes and quit\n"); printf(" perf-by-size: will wait for all hosts to join the knet network,\n"); printf(" perform a series of benchmarks by transmitting a known\n"); printf(" size/quantity of packets and measuring the time, then quit\n"); printf(" perf-by-time: will wait for all hosts to join the knet network,\n"); printf(" perform a series of benchmarks by transmitting a known\n"); printf(" size of packets for a given amount of time (10 seconds)\n"); printf(" and measuring the quantity of data transmitted, then quit\n"); printf(" -s nodeid that will generate traffic for benchmarks\n"); printf(" -S [size|seconds] when used in combination with -T perf-by-size it indicates how many GB of traffic to generate for the test. (default: 1GB)\n"); printf(" when used in combination with -T perf-by-time it indicates how many Seconds of traffic to generate for the test. (default: 10 seconds)\n"); printf(" -C repeat the test continously (default: off)\n"); printf(" -X[XX] show stats at the end of the run (default: 1)\n"); printf(" 1: show handle stats, 2: show summary link stats\n"); printf(" 3: show detailed link stats\n"); printf(" -a enable machine parsable output (default: off).\n"); } static void parse_nodes(char *nodesinfo[MAX_NODES], int onidx, int port, struct node nodes[MAX_NODES], int *thisidx) { int i; char *temp = NULL; char port_str[10]; memset(port_str, 0, sizeof(port_str)); sprintf(port_str, "%d", port); for (i = 0; i < onidx; i++) { nodes[i].nodeid = atoi(strtok(nodesinfo[i], ",")); if ((nodes[i].nodeid < 0) || (nodes[i].nodeid > KNET_MAX_HOST)) { printf("Invalid nodeid: %d (0 - %d)\n", nodes[i].nodeid, KNET_MAX_HOST); exit(FAIL); } if (thisnodeid == nodes[i].nodeid) { *thisidx = i; } while((temp = strtok(NULL, ","))) { char *slash = NULL; uint8_t transport; if (nodes[i].links == KNET_MAX_LINK) { printf("Too many links configured. Max %d\n", KNET_MAX_LINK); exit(FAIL); } slash = strstr(temp, "/"); if (slash) { memset(slash, 0, 1); transport = knet_get_transport_id_by_name(temp); if (transport == KNET_MAX_TRANSPORTS) { printf("Unknown transport: %s\n", temp); exit(FAIL); } nodes[i].transport[nodes[i].links] = transport; temp = slash + 1; } else { nodes[i].transport[nodes[i].links] = KNET_TRANSPORT_UDP; } if (knet_strtoaddr(temp, port_str, &nodes[i].address[nodes[i].links], sizeof(struct sockaddr_storage)) < 0) { printf("Unable to convert %s to sockaddress\n", temp); exit(FAIL); } nodes[i].links++; } } if (knet_strtoaddr("0.0.0.0", port_str, &allv4, sizeof(struct sockaddr_storage)) < 0) { printf("Unable to convert 0.0.0.0 to sockaddress\n"); exit(FAIL); } if (knet_strtoaddr("::", port_str, &allv6, sizeof(struct sockaddr_storage)) < 0) { printf("Unable to convert :: to sockaddress\n"); exit(FAIL); } for (i = 1; i < onidx; i++) { if (nodes[0].links != nodes[i].links) { printf("knet_bench does not support unbalanced link configuration\n"); exit(FAIL); } } return; } static int private_data; static void sock_notify(void *pvt_data, int local_datafd, int8_t local_channel, uint8_t tx_rx, int error, int errorno) { printf("[info]: error (%d - %d - %s) from socket: %d\n", error, errorno, strerror(errno), local_datafd); return; } static int ping_dst_host_filter(void *pvt_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_host_id, int8_t *dst_channel, knet_node_id_t *dst_host_ids, size_t *dst_host_ids_entries) { if (broadcast_test) { return 1; } if (tx_rx == KNET_NOTIFY_TX) { memmove(&dst_host_ids[0], outdata, 2); } else { dst_host_ids[0] = this_host_id; } *dst_host_ids_entries = 1; return 0; } static void setup_knet(int argc, char *argv[]) { int logfd = 0; int rv; char *policystr = NULL, *protostr = NULL; char *othernodeinfo[MAX_NODES]; struct node nodes[MAX_NODES]; int thisidx = -1; int onidx = 0; int debug = KNET_LOG_INFO; int port = 50000, portoffset = 0; int thisport = 0, otherport = 0; int thisnewport = 0, othernewport = 0; struct sockaddr_in *so_in; struct sockaddr_in6 *so_in6; struct sockaddr_storage *src; int i, link_idx, allnodesup = 0; int policy = KNET_LINK_POLICY_PASSIVE, policyfound = 0; int protocol = KNET_TRANSPORT_UDP, protofound = 0; int wait = 1; int pmtud_interval = 60; struct knet_handle_crypto_cfg knet_handle_crypto_cfg; char *cryptomodel = NULL, *cryptotype = NULL, *cryptohash = NULL; struct knet_handle_compress_cfg knet_handle_compress_cfg; memset(nodes, 0, sizeof(nodes)); while ((rv = getopt(argc, argv, "aCT:S:s:ldom:wb:t:n:c:p:X::P:z:h")) != EOF) { switch(rv) { case 'h': print_help(); exit(PASS); break; case 'a': machine_output = 1; break; case 'd': debug = KNET_LOG_DEBUG; break; case 'c': if (cryptocfg) { printf("Error: -c can only be specified once\n"); exit(FAIL); } cryptocfg = optarg; break; case 'p': if (policystr) { printf("Error: -p can only be specified once\n"); exit(FAIL); } policystr = optarg; if (!strcmp(policystr, "active")) { policy = KNET_LINK_POLICY_ACTIVE; policyfound = 1; } /* * we can't use rr because clangs can't compile * an array of 3 strings, one of which is 2 bytes long */ if (!strcmp(policystr, "round-robin")) { policy = KNET_LINK_POLICY_RR; policyfound = 1; } if (!strcmp(policystr, "passive")) { policy = KNET_LINK_POLICY_PASSIVE; policyfound = 1; } if (!policyfound) { printf("Error: invalid policy %s specified. -p accepts active|passive|rr\n", policystr); exit(FAIL); } break; case 'P': if (protostr) { printf("Error: -P can only be specified once\n"); exit(FAIL); } protostr = optarg; if (!strcmp(protostr, "UDP")) { protocol = KNET_TRANSPORT_UDP; protofound = 1; } if (!strcmp(protostr, "SCTP")) { protocol = KNET_TRANSPORT_SCTP; protofound = 1; } if (!protofound) { printf("Error: invalid protocol %s specified. -P accepts udp|sctp\n", policystr); exit(FAIL); } break; case 't': if (thisnodeid >= 0) { printf("Error: -t can only be specified once\n"); exit(FAIL); } thisnodeid = atoi(optarg); if ((thisnodeid < 0) || (thisnodeid > 65536)) { printf("Error: -t nodeid out of range %d (1 - 65536)\n", thisnodeid); exit(FAIL); } break; case 'n': if (onidx == MAX_NODES) { printf("Error: too many other nodes. Max %d\n", MAX_NODES); exit(FAIL); } othernodeinfo[onidx] = optarg; onidx++; break; case 'b': port = atoi(optarg); if ((port < 1) || (port > 65536)) { printf("Error: port %d out of range (1 - 65536)\n", port); exit(FAIL); } break; case 'o': if (globallistener) { printf("Error: -l cannot be used with -o\n"); exit(FAIL); } portoffset = 1; break; case 'm': pmtud_interval = atoi(optarg); if (pmtud_interval < 1) { printf("Error: pmtud interval %d out of range (> 0)\n", pmtud_interval); exit(FAIL); } break; case 'l': if (portoffset) { printf("Error: -o cannot be used with -l\n"); exit(FAIL); } globallistener = 1; break; case 'w': wait = 0; break; case 's': if (senderid >= 0) { printf("Error: -s can only be specified once\n"); exit(FAIL); } senderid = atoi(optarg); if ((senderid < 0) || (senderid > 65536)) { printf("Error: -s nodeid out of range %d (1 - 65536)\n", senderid); exit(FAIL); } break; case 'T': if (!strcmp("ping", optarg)) { test_type = TEST_PING; } if (!strcmp("ping_data", optarg)) { test_type = TEST_PING_AND_DATA; } if (!strcmp("perf-by-size", optarg)) { test_type = TEST_PERF_BY_SIZE; } if (!strcmp("perf-by-time", optarg)) { test_type = TEST_PERF_BY_TIME; } break; case 'S': perf_by_size_size = (uint64_t)atoi(optarg) * ONE_GIGABYTE; perf_by_time_secs = (uint64_t)atoi(optarg); break; case 'C': continous = 1; break; case 'X': if (optarg) { show_stats = atoi(optarg); } else { show_stats = 1; } break; case 'z': if (compresscfg) { printf("Error: -c can only be specified once\n"); exit(FAIL); } compresscfg = optarg; break; default: break; } } if (thisnodeid < 0) { printf("Who am I?!? missing -t from command line?\n"); exit(FAIL); } if (onidx < 1) { printf("no other nodes configured?!? missing -n from command line\n"); exit(FAIL); } parse_nodes(othernodeinfo, onidx, port, nodes, &thisidx); if (thisidx < 0) { printf("no config for this node found\n"); exit(FAIL); } if (senderid >= 0) { for (i=0; i < onidx; i++) { if (senderid == nodes[i].nodeid) { break; } } if (i == onidx) { printf("Unable to find senderid in nodelist\n"); exit(FAIL); } } if (((test_type == TEST_PERF_BY_SIZE) || (test_type == TEST_PERF_BY_TIME)) && (senderid < 0)) { printf("Error: performance test requires -s to be set (for now)\n"); exit(FAIL); } logfd = start_logging(stdout); knet_h = knet_handle_new(thisnodeid, logfd, debug); if (!knet_h) { printf("Unable to knet_handle_new: %s\n", strerror(errno)); exit(FAIL); } if (cryptocfg) { memset(&knet_handle_crypto_cfg, 0, sizeof(knet_handle_crypto_cfg)); cryptomodel = strtok(cryptocfg, ":"); cryptotype = strtok(NULL, ":"); cryptohash = strtok(NULL, ":"); if (cryptomodel) { strncpy(knet_handle_crypto_cfg.crypto_model, cryptomodel, sizeof(knet_handle_crypto_cfg.crypto_model) - 1); } if (cryptotype) { strncpy(knet_handle_crypto_cfg.crypto_cipher_type, cryptotype, sizeof(knet_handle_crypto_cfg.crypto_cipher_type) - 1); } if (cryptohash) { strncpy(knet_handle_crypto_cfg.crypto_hash_type, cryptohash, sizeof(knet_handle_crypto_cfg.crypto_hash_type) - 1); } knet_handle_crypto_cfg.private_key_len = KNET_MAX_KEY_LEN; if (knet_handle_crypto(knet_h, &knet_handle_crypto_cfg)) { printf("Unable to init crypto\n"); exit(FAIL); } } if (compresscfg) { memset(&knet_handle_compress_cfg, 0, sizeof(struct knet_handle_compress_cfg)); snprintf(knet_handle_compress_cfg.compress_model, 16, "%s", strtok(compresscfg, ":")); knet_handle_compress_cfg.compress_level = atoi(strtok(NULL, ":")); knet_handle_compress_cfg.compress_threshold = atoi(strtok(NULL, ":")); if (knet_handle_compress(knet_h, &knet_handle_compress_cfg)) { printf("Unable to configure compress\n"); exit(FAIL); } } if (knet_handle_enable_sock_notify(knet_h, &private_data, sock_notify) < 0) { printf("knet_handle_enable_sock_notify failed: %s\n", strerror(errno)); knet_handle_free(knet_h); exit(FAIL); } datafd = 0; channel = -1; if (knet_handle_add_datafd(knet_h, &datafd, &channel) < 0) { printf("knet_handle_add_datafd failed: %s\n", strerror(errno)); knet_handle_free(knet_h); exit(FAIL); } if (knet_handle_pmtud_setfreq(knet_h, pmtud_interval) < 0) { printf("knet_handle_pmtud_setfreq failed: %s\n", strerror(errno)); knet_handle_free(knet_h); exit(FAIL); } for (i=0; i < onidx; i++) { if (i == thisidx) { continue; } if (knet_host_add(knet_h, nodes[i].nodeid) < 0) { printf("knet_host_add failed: %s\n", strerror(errno)); exit(FAIL); } if (knet_host_set_policy(knet_h, nodes[i].nodeid, policy) < 0) { printf("knet_host_set_policy failed: %s\n", strerror(errno)); exit(FAIL); } for (link_idx = 0; link_idx < nodes[i].links; link_idx++) { if (portoffset) { if (nodes[thisidx].address[link_idx].ss_family == AF_INET) { so_in = (struct sockaddr_in *)&nodes[thisidx].address[link_idx]; thisport = ntohs(so_in->sin_port); thisnewport = thisport + nodes[i].nodeid; so_in->sin_port = (htons(thisnewport)); so_in = (struct sockaddr_in *)&nodes[i].address[link_idx]; otherport = ntohs(so_in->sin_port); othernewport = otherport + nodes[thisidx].nodeid; so_in->sin_port = (htons(othernewport)); } else { so_in6 = (struct sockaddr_in6 *)&nodes[thisidx].address[link_idx]; thisport = ntohs(so_in6->sin6_port); thisnewport = thisport + nodes[i].nodeid; so_in6->sin6_port = (htons(thisnewport)); so_in6 = (struct sockaddr_in6 *)&nodes[i].address[link_idx]; otherport = ntohs(so_in6->sin6_port); othernewport = otherport + nodes[thisidx].nodeid; so_in6->sin6_port = (htons(othernewport)); } } if (!globallistener) { src = &nodes[thisidx].address[link_idx]; } else { if (nodes[thisidx].address[link_idx].ss_family == AF_INET) { src = &allv4; } else { src = &allv6; } } /* * -P overrides per link protocol configuration */ if (protofound) { nodes[i].transport[link_idx] = protocol; } if (knet_link_set_config(knet_h, nodes[i].nodeid, link_idx, nodes[i].transport[link_idx], src, &nodes[i].address[link_idx], 0) < 0) { printf("Unable to configure link: %s\n", strerror(errno)); exit(FAIL); } if (portoffset) { if (nodes[thisidx].address[link_idx].ss_family == AF_INET) { so_in = (struct sockaddr_in *)&nodes[thisidx].address[link_idx]; so_in->sin_port = (htons(thisport)); so_in = (struct sockaddr_in *)&nodes[i].address[link_idx]; so_in->sin_port = (htons(otherport)); } else { so_in6 = (struct sockaddr_in6 *)&nodes[thisidx].address[link_idx]; so_in6->sin6_port = (htons(thisport)); so_in6 = (struct sockaddr_in6 *)&nodes[i].address[link_idx]; so_in6->sin6_port = (htons(otherport)); } } if (knet_link_set_enable(knet_h, nodes[i].nodeid, link_idx, 1) < 0) { printf("knet_link_set_enable failed: %s\n", strerror(errno)); exit(FAIL); } if (knet_link_set_ping_timers(knet_h, nodes[i].nodeid, link_idx, 1000, 10000, 2048) < 0) { printf("knet_link_set_ping_timers failed: %s\n", strerror(errno)); exit(FAIL); } if (knet_link_set_pong_count(knet_h, nodes[i].nodeid, link_idx, 2) < 0) { printf("knet_link_set_pong_count failed: %s\n", strerror(errno)); exit(FAIL); } } } if (knet_handle_enable_filter(knet_h, NULL, ping_dst_host_filter)) { printf("Unable to enable dst_host_filter: %s\n", strerror(errno)); exit(FAIL); } if (knet_handle_setfwd(knet_h, 1) < 0) { printf("knet_handle_setfwd failed: %s\n", strerror(errno)); exit(FAIL); } if (wait) { while(!allnodesup) { allnodesup = 1; for (i=0; i < onidx; i++) { if (i == thisidx) { continue; } if (knet_h->host_index[nodes[i].nodeid]->status.reachable != 1) { printf("[info]: waiting host %d to be reachable\n", nodes[i].nodeid); allnodesup = 0; } } if (!allnodesup) { sleep(1); } } sleep(1); } } static void *_rx_thread(void *args) { int rx_epoll; struct epoll_event ev; struct epoll_event events[KNET_EPOLL_MAX_EVENTS]; struct sockaddr_storage address[PCKT_FRAG_MAX]; struct knet_mmsghdr msg[PCKT_FRAG_MAX]; struct iovec iov_in[PCKT_FRAG_MAX]; int i, msg_recv; struct timespec clock_start, clock_end; unsigned long long time_diff = 0; uint64_t rx_pkts = 0; uint64_t rx_bytes = 0; unsigned int current_pckt_size = 0; for (i = 0; i < PCKT_FRAG_MAX; i++) { rx_buf[i] = malloc(KNET_MAX_PACKET_SIZE); if (!rx_buf[i]) { printf("RXT: Unable to malloc!\nHALTING RX THREAD!\n"); return NULL; } memset(rx_buf[i], 0, KNET_MAX_PACKET_SIZE); iov_in[i].iov_base = (void *)rx_buf[i]; iov_in[i].iov_len = KNET_MAX_PACKET_SIZE; memset(&msg[i].msg_hdr, 0, sizeof(struct msghdr)); msg[i].msg_hdr.msg_name = &address[i]; msg[i].msg_hdr.msg_namelen = sizeof(struct sockaddr_storage); msg[i].msg_hdr.msg_iov = &iov_in[i]; msg[i].msg_hdr.msg_iovlen = 1; } rx_epoll = epoll_create(KNET_EPOLL_MAX_EVENTS + 1); if (rx_epoll < 0) { printf("RXT: Unable to create epoll!\nHALTING RX THREAD!\n"); return NULL; } memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = datafd; if (epoll_ctl(rx_epoll, EPOLL_CTL_ADD, datafd, &ev)) { printf("RXT: Unable to add datafd to epoll\nHALTING RX THREAD!\n"); return NULL; } memset(&clock_start, 0, sizeof(clock_start)); memset(&clock_end, 0, sizeof(clock_start)); while (!bench_shutdown_in_progress) { if (epoll_wait(rx_epoll, events, KNET_EPOLL_MAX_EVENTS, 1) >= 1) { msg_recv = _recvmmsg(datafd, &msg[0], PCKT_FRAG_MAX, MSG_DONTWAIT | MSG_NOSIGNAL); if (msg_recv < 0) { printf("[info]: RXT: error from recvmmsg: %s\n", strerror(errno)); } switch(test_type) { case TEST_PING_AND_DATA: for (i = 0; i < msg_recv; i++) { if (msg[i].msg_len == 0) { printf("[info]: RXT: received 0 bytes message?\n"); } printf("[info]: received %u bytes message: %s\n", msg[i].msg_len, (char *)msg[i].msg_hdr.msg_iov->iov_base); } break; case TEST_PERF_BY_TIME: case TEST_PERF_BY_SIZE: for (i = 0; i < msg_recv; i++) { if (msg[i].msg_len < 64) { if (msg[i].msg_len == 0) { printf("[info]: RXT: received 0 bytes message?\n"); } if (msg[i].msg_len == TEST_START) { if (clock_gettime(CLOCK_MONOTONIC, &clock_start) != 0) { printf("[info]: unable to get start time!\n"); } } if (msg[i].msg_len == TEST_STOP) { double average_rx_mbytes; double average_rx_pkts; double time_diff_sec; if (clock_gettime(CLOCK_MONOTONIC, &clock_end) != 0) { printf("[info]: unable to get end time!\n"); } timespec_diff(clock_start, clock_end, &time_diff); /* * adjust for sleep(2) between sending the last data and TEST_STOP */ time_diff = time_diff - 2000000000llu; /* * convert to seconds */ time_diff_sec = (double)time_diff / 1000000000llu; average_rx_mbytes = (double)((rx_bytes / time_diff_sec) / (1024 * 1024)); average_rx_pkts = (double)(rx_pkts / time_diff_sec); if (!machine_output) { printf("[perf] execution time: %8.4f secs Average speed: %8.4f MB/sec %8.4f pckts/sec (size: %u total: %" PRIu64 ")\n", time_diff_sec, average_rx_mbytes, average_rx_pkts, current_pckt_size, rx_pkts); } else { printf("[perf],%.4f,%u,%" PRIu64 ",%.4f,%.4f\n", time_diff_sec, current_pckt_size, rx_pkts, average_rx_mbytes, average_rx_pkts); } rx_pkts = 0; rx_bytes = 0; current_pckt_size = 0; } if (msg[i].msg_len == TEST_COMPLETE) { wait_for_perf_rx = 1; } continue; } rx_pkts++; rx_bytes = rx_bytes + msg[i].msg_len; current_pckt_size = msg[i].msg_len; } break; } } } epoll_ctl(rx_epoll, EPOLL_CTL_DEL, datafd, &ev); close(rx_epoll); return NULL; } static void setup_data_txrx_common(void) { if (!rx_thread) { if (knet_handle_enable_filter(knet_h, NULL, ping_dst_host_filter)) { printf("Unable to enable dst_host_filter: %s\n", strerror(errno)); exit(FAIL); } printf("[info]: setting up rx thread\n"); if (pthread_create(&rx_thread, 0, _rx_thread, NULL)) { printf("Unable to start rx thread\n"); exit(FAIL); } } } static void stop_rx_thread(void) { void *retval; int i; if (rx_thread) { printf("[info]: shutting down rx thread\n"); sleep(2); pthread_cancel(rx_thread); pthread_join(rx_thread, &retval); for (i = 0; i < PCKT_FRAG_MAX; i ++) { free(rx_buf[i]); } } } static void send_ping_data(void) { char buf[65535]; ssize_t len; memset(&buf, 0, sizeof(buf)); snprintf(buf, sizeof(buf), "Hello world!"); if (compresscfg) { len = sizeof(buf); } else { len = strlen(buf); } if (knet_send(knet_h, buf, len, channel) != len) { printf("[info]: Error sending hello world: %s\n", strerror(errno)); } sleep(1); } static int send_messages(struct knet_mmsghdr *msg, int msgs_to_send) { int sent_msgs, prev_sent, progress, total_sent; total_sent = 0; sent_msgs = 0; prev_sent = 0; progress = 1; retry: errno = 0; - sent_msgs = _sendmmsg(datafd, &msg[0], msgs_to_send, MSG_NOSIGNAL); + sent_msgs = _sendmmsg(datafd, 0, &msg[0], msgs_to_send, MSG_NOSIGNAL); if (sent_msgs < 0) { if ((errno == EAGAIN) || (errno == EWOULDBLOCK)) { usleep(KNET_THREADS_TIMERES / 16); goto retry; } printf("[info]: Unable to send messages: %s\n", strerror(errno)); return -1; } total_sent = total_sent + sent_msgs; if ((sent_msgs >= 0) && (sent_msgs < msgs_to_send)) { if ((sent_msgs) || (progress)) { msgs_to_send = msgs_to_send - sent_msgs; prev_sent = prev_sent + sent_msgs; if (sent_msgs) { progress = 1; } else { progress = 0; } goto retry; } if (!progress) { printf("[info]: Unable to send more messages after retry\n"); } } return total_sent; } static int setup_send_buffers_common(struct knet_mmsghdr *msg, struct iovec *iov_out, char *tx_buf[]) { int i; for (i = 0; i < PCKT_FRAG_MAX; i++) { tx_buf[i] = malloc(KNET_MAX_PACKET_SIZE); if (!tx_buf[i]) { printf("TXT: Unable to malloc!\n"); return -1; } memset(tx_buf[i], 0, KNET_MAX_PACKET_SIZE); iov_out[i].iov_base = (void *)tx_buf[i]; memset(&msg[i].msg_hdr, 0, sizeof(struct msghdr)); msg[i].msg_hdr.msg_iov = &iov_out[i]; msg[i].msg_hdr.msg_iovlen = 1; } return 0; } static void send_perf_data_by_size(void) { char *tx_buf[PCKT_FRAG_MAX]; struct knet_mmsghdr msg[PCKT_FRAG_MAX]; struct iovec iov_out[PCKT_FRAG_MAX]; char ctrl_message[16]; int sent_msgs; int i; uint64_t total_pkts_to_tx; uint64_t packets_to_send; uint32_t packetsize = 64; setup_send_buffers_common(msg, iov_out, tx_buf); while (packetsize <= KNET_MAX_PACKET_SIZE) { for (i = 0; i < PCKT_FRAG_MAX; i++) { iov_out[i].iov_len = packetsize; } total_pkts_to_tx = perf_by_size_size / packetsize; printf("[info]: testing with %u packet size. total bytes to transfer: %" PRIu64 " (%" PRIu64 " packets)\n", packetsize, perf_by_size_size, total_pkts_to_tx); memset(ctrl_message, 0, sizeof(ctrl_message)); knet_send(knet_h, ctrl_message, TEST_START, channel); while (total_pkts_to_tx > 0) { if (total_pkts_to_tx >= PCKT_FRAG_MAX) { packets_to_send = PCKT_FRAG_MAX; } else { packets_to_send = total_pkts_to_tx; } sent_msgs = send_messages(&msg[0], packets_to_send); if (sent_msgs < 0) { printf("Something went wrong, aborting\n"); exit(FAIL); } total_pkts_to_tx = total_pkts_to_tx - sent_msgs; } sleep(2); knet_send(knet_h, ctrl_message, TEST_STOP, channel); if (packetsize == KNET_MAX_PACKET_SIZE) { break; } /* * Use a multiplier that can always divide properly a GB * into smaller chunks without worry about boundaries */ packetsize *= 4; if (packetsize > KNET_MAX_PACKET_SIZE) { packetsize = KNET_MAX_PACKET_SIZE; } } knet_send(knet_h, ctrl_message, TEST_COMPLETE, channel); for (i = 0; i < PCKT_FRAG_MAX; i++) { free(tx_buf[i]); } } /* For sorting the node list into order */ static int node_compare(const void *aptr, const void *bptr) { uint16_t a,b; a = *(uint16_t *)aptr; b = *(uint16_t *)bptr; return a > b; } static void display_stats(int level) { struct knet_handle_stats handle_stats; struct knet_link_status link_status; struct knet_link_stats total_link_stats; knet_node_id_t host_list[KNET_MAX_HOST]; uint8_t link_list[KNET_MAX_LINK]; int res; unsigned int i,j; size_t num_hosts, num_links; res = knet_handle_get_stats(knet_h, &handle_stats, sizeof(handle_stats)); if (res) { perror("[info]: failed to get knet handle stats"); return; } if (compresscfg || cryptocfg) { printf("\n"); printf("[stat]: handle stats\n"); printf("[stat]: ------------\n"); if (compresscfg) { printf("[stat]: tx_uncompressed_packets: %" PRIu64 "\n", handle_stats.tx_uncompressed_packets); printf("[stat]: tx_compressed_packets: %" PRIu64 "\n", handle_stats.tx_compressed_packets); printf("[stat]: tx_compressed_original_bytes: %" PRIu64 "\n", handle_stats.tx_compressed_original_bytes); printf("[stat]: tx_compressed_size_bytes: %" PRIu64 "\n", handle_stats.tx_compressed_size_bytes ); printf("[stat]: tx_compress_time_ave: %" PRIu64 "\n", handle_stats.tx_compress_time_ave); printf("[stat]: tx_compress_time_min: %" PRIu64 "\n", handle_stats.tx_compress_time_min); printf("[stat]: tx_compress_time_max: %" PRIu64 "\n", handle_stats.tx_compress_time_max); printf("[stat]: rx_compressed_packets: %" PRIu64 "\n", handle_stats.rx_compressed_packets); printf("[stat]: rx_compressed_original_bytes: %" PRIu64 "\n", handle_stats.rx_compressed_original_bytes); printf("[stat]: rx_compressed_size_bytes: %" PRIu64 "\n", handle_stats.rx_compressed_size_bytes); printf("[stat]: rx_compress_time_ave: %" PRIu64 "\n", handle_stats.rx_compress_time_ave); printf("[stat]: rx_compress_time_min: %" PRIu64 "\n", handle_stats.rx_compress_time_min); printf("[stat]: rx_compress_time_max: %" PRIu64 "\n", handle_stats.rx_compress_time_max); printf("\n"); } if (cryptocfg) { printf("[stat]: tx_crypt_packets: %" PRIu64 "\n", handle_stats.tx_crypt_packets); printf("[stat]: tx_crypt_byte_overhead: %" PRIu64 "\n", handle_stats.tx_crypt_byte_overhead); printf("[stat]: tx_crypt_time_ave: %" PRIu64 "\n", handle_stats.tx_crypt_time_ave); printf("[stat]: tx_crypt_time_min: %" PRIu64 "\n", handle_stats.tx_crypt_time_min); printf("[stat]: tx_crypt_time_max: %" PRIu64 "\n", handle_stats.tx_crypt_time_max); printf("[stat]: rx_crypt_packets: %" PRIu64 "\n", handle_stats.rx_crypt_packets); printf("[stat]: rx_crypt_time_ave: %" PRIu64 "\n", handle_stats.rx_crypt_time_ave); printf("[stat]: rx_crypt_time_min: %" PRIu64 "\n", handle_stats.rx_crypt_time_min); printf("[stat]: rx_crypt_time_max: %" PRIu64 "\n", handle_stats.rx_crypt_time_max); printf("\n"); } } if (level < 2) { return; } memset(&total_link_stats, 0, sizeof(struct knet_link_stats)); res = knet_host_get_host_list(knet_h, host_list, &num_hosts); if (res) { perror("[info]: cannot get host list for stats"); return; } /* Print in host ID order */ qsort(host_list, num_hosts, sizeof(uint16_t), node_compare); for (j=0; j 2) { printf("\n"); printf("[stat]: Node %d Link %d\n", host_list[j], link_list[i]); printf("[stat]: tx_data_packets: %" PRIu64 "\n", link_status.stats.tx_data_packets); printf("[stat]: rx_data_packets: %" PRIu64 "\n", link_status.stats.rx_data_packets); printf("[stat]: tx_data_bytes: %" PRIu64 "\n", link_status.stats.tx_data_bytes); printf("[stat]: rx_data_bytes: %" PRIu64 "\n", link_status.stats.rx_data_bytes); printf("[stat]: rx_ping_packets: %" PRIu64 "\n", link_status.stats.rx_ping_packets); printf("[stat]: tx_ping_packets: %" PRIu64 "\n", link_status.stats.tx_ping_packets); printf("[stat]: rx_ping_bytes: %" PRIu64 "\n", link_status.stats.rx_ping_bytes); printf("[stat]: tx_ping_bytes: %" PRIu64 "\n", link_status.stats.tx_ping_bytes); printf("[stat]: rx_pong_packets: %" PRIu64 "\n", link_status.stats.rx_pong_packets); printf("[stat]: tx_pong_packets: %" PRIu64 "\n", link_status.stats.tx_pong_packets); printf("[stat]: rx_pong_bytes: %" PRIu64 "\n", link_status.stats.rx_pong_bytes); printf("[stat]: tx_pong_bytes: %" PRIu64 "\n", link_status.stats.tx_pong_bytes); printf("[stat]: rx_pmtu_packets: %" PRIu64 "\n", link_status.stats.rx_pmtu_packets); printf("[stat]: tx_pmtu_packets: %" PRIu64 "\n", link_status.stats.tx_pmtu_packets); printf("[stat]: rx_pmtu_bytes: %" PRIu64 "\n", link_status.stats.rx_pmtu_bytes); printf("[stat]: tx_pmtu_bytes: %" PRIu64 "\n", link_status.stats.tx_pmtu_bytes); printf("[stat]: tx_total_packets: %" PRIu64 "\n", link_status.stats.tx_total_packets); printf("[stat]: rx_total_packets: %" PRIu64 "\n", link_status.stats.rx_total_packets); printf("[stat]: tx_total_bytes: %" PRIu64 "\n", link_status.stats.tx_total_bytes); printf("[stat]: rx_total_bytes: %" PRIu64 "\n", link_status.stats.rx_total_bytes); printf("[stat]: tx_total_errors: %" PRIu64 "\n", link_status.stats.tx_total_errors); printf("[stat]: tx_total_retries: %" PRIu64 "\n", link_status.stats.tx_total_retries); printf("[stat]: tx_pmtu_errors: %" PRIu32 "\n", link_status.stats.tx_pmtu_errors); printf("[stat]: tx_pmtu_retries: %" PRIu32 "\n", link_status.stats.tx_pmtu_retries); printf("[stat]: tx_ping_errors: %" PRIu32 "\n", link_status.stats.tx_ping_errors); printf("[stat]: tx_ping_retries: %" PRIu32 "\n", link_status.stats.tx_ping_retries); printf("[stat]: tx_pong_errors: %" PRIu32 "\n", link_status.stats.tx_pong_errors); printf("[stat]: tx_pong_retries: %" PRIu32 "\n", link_status.stats.tx_pong_retries); printf("[stat]: tx_data_errors: %" PRIu32 "\n", link_status.stats.tx_data_errors); printf("[stat]: tx_data_retries: %" PRIu32 "\n", link_status.stats.tx_data_retries); printf("[stat]: latency_min: %" PRIu32 "\n", link_status.stats.latency_min); printf("[stat]: latency_max: %" PRIu32 "\n", link_status.stats.latency_max); printf("[stat]: latency_ave: %" PRIu32 "\n", link_status.stats.latency_ave); printf("[stat]: latency_samples: %" PRIu32 "\n", link_status.stats.latency_samples); printf("[stat]: down_count: %" PRIu32 "\n", link_status.stats.down_count); printf("[stat]: up_count: %" PRIu32 "\n", link_status.stats.up_count); } } } printf("\n"); printf("[stat]: Total link stats\n"); printf("[stat]: ----------------\n"); printf("[stat]: tx_data_packets: %" PRIu64 "\n", total_link_stats.tx_data_packets); printf("[stat]: rx_data_packets: %" PRIu64 "\n", total_link_stats.rx_data_packets); printf("[stat]: tx_data_bytes: %" PRIu64 "\n", total_link_stats.tx_data_bytes); printf("[stat]: rx_data_bytes: %" PRIu64 "\n", total_link_stats.rx_data_bytes); printf("[stat]: rx_ping_packets: %" PRIu64 "\n", total_link_stats.rx_ping_packets); printf("[stat]: tx_ping_packets: %" PRIu64 "\n", total_link_stats.tx_ping_packets); printf("[stat]: rx_ping_bytes: %" PRIu64 "\n", total_link_stats.rx_ping_bytes); printf("[stat]: tx_ping_bytes: %" PRIu64 "\n", total_link_stats.tx_ping_bytes); printf("[stat]: rx_pong_packets: %" PRIu64 "\n", total_link_stats.rx_pong_packets); printf("[stat]: tx_pong_packets: %" PRIu64 "\n", total_link_stats.tx_pong_packets); printf("[stat]: rx_pong_bytes: %" PRIu64 "\n", total_link_stats.rx_pong_bytes); printf("[stat]: tx_pong_bytes: %" PRIu64 "\n", total_link_stats.tx_pong_bytes); printf("[stat]: rx_pmtu_packets: %" PRIu64 "\n", total_link_stats.rx_pmtu_packets); printf("[stat]: tx_pmtu_packets: %" PRIu64 "\n", total_link_stats.tx_pmtu_packets); printf("[stat]: rx_pmtu_bytes: %" PRIu64 "\n", total_link_stats.rx_pmtu_bytes); printf("[stat]: tx_pmtu_bytes: %" PRIu64 "\n", total_link_stats.tx_pmtu_bytes); printf("[stat]: tx_total_packets: %" PRIu64 "\n", total_link_stats.tx_total_packets); printf("[stat]: rx_total_packets: %" PRIu64 "\n", total_link_stats.rx_total_packets); printf("[stat]: tx_total_bytes: %" PRIu64 "\n", total_link_stats.tx_total_bytes); printf("[stat]: rx_total_bytes: %" PRIu64 "\n", total_link_stats.rx_total_bytes); printf("[stat]: tx_total_errors: %" PRIu64 "\n", total_link_stats.tx_total_errors); printf("[stat]: tx_total_retries: %" PRIu64 "\n", total_link_stats.tx_total_retries); printf("[stat]: tx_pmtu_errors: %" PRIu32 "\n", total_link_stats.tx_pmtu_errors); printf("[stat]: tx_pmtu_retries: %" PRIu32 "\n", total_link_stats.tx_pmtu_retries); printf("[stat]: tx_ping_errors: %" PRIu32 "\n", total_link_stats.tx_ping_errors); printf("[stat]: tx_ping_retries: %" PRIu32 "\n", total_link_stats.tx_ping_retries); printf("[stat]: tx_pong_errors: %" PRIu32 "\n", total_link_stats.tx_pong_errors); printf("[stat]: tx_pong_retries: %" PRIu32 "\n", total_link_stats.tx_pong_retries); printf("[stat]: tx_data_errors: %" PRIu32 "\n", total_link_stats.tx_data_errors); printf("[stat]: tx_data_retries: %" PRIu32 "\n", total_link_stats.tx_data_retries); printf("[stat]: down_count: %" PRIu32 "\n", total_link_stats.down_count); printf("[stat]: up_count: %" PRIu32 "\n", total_link_stats.up_count); } static void send_perf_data_by_time(void) { char *tx_buf[PCKT_FRAG_MAX]; struct knet_mmsghdr msg[PCKT_FRAG_MAX]; struct iovec iov_out[PCKT_FRAG_MAX]; char ctrl_message[16]; int sent_msgs; int i; uint32_t packetsize = 64; struct timespec clock_start, clock_end; unsigned long long time_diff = 0; setup_send_buffers_common(msg, iov_out, tx_buf); memset(&clock_start, 0, sizeof(clock_start)); memset(&clock_end, 0, sizeof(clock_start)); while (packetsize <= KNET_MAX_PACKET_SIZE) { for (i = 0; i < PCKT_FRAG_MAX; i++) { iov_out[i].iov_len = packetsize; } printf("[info]: testing with %u bytes packet size for %" PRIu64 " seconds.\n", packetsize, perf_by_time_secs); memset(ctrl_message, 0, sizeof(ctrl_message)); knet_send(knet_h, ctrl_message, TEST_START, channel); if (clock_gettime(CLOCK_MONOTONIC, &clock_start) != 0) { printf("[info]: unable to get start time!\n"); } time_diff = 0; while (time_diff < (perf_by_time_secs * 1000000000llu)) { sent_msgs = send_messages(&msg[0], PCKT_FRAG_MAX); if (sent_msgs < 0) { printf("Something went wrong, aborting\n"); exit(FAIL); } if (clock_gettime(CLOCK_MONOTONIC, &clock_end) != 0) { printf("[info]: unable to get end time!\n"); } timespec_diff(clock_start, clock_end, &time_diff); } sleep(2); knet_send(knet_h, ctrl_message, TEST_STOP, channel); if (packetsize == KNET_MAX_PACKET_SIZE) { break; } /* * Use a multiplier that can always divide properly a GB * into smaller chunks without worry about boundaries */ packetsize *= 4; if (packetsize > KNET_MAX_PACKET_SIZE) { packetsize = KNET_MAX_PACKET_SIZE; } } knet_send(knet_h, ctrl_message, TEST_COMPLETE, channel); for (i = 0; i < PCKT_FRAG_MAX; i++) { free(tx_buf[i]); } } static void cleanup_all(void) { if (pthread_mutex_lock(&shutdown_mutex)) { return; } if (bench_shutdown_in_progress) { pthread_mutex_unlock(&shutdown_mutex); return; } bench_shutdown_in_progress = 1; pthread_mutex_unlock(&shutdown_mutex); if (rx_thread) { stop_rx_thread(); } knet_handle_stop(knet_h); } static void sigint_handler(int signum) { printf("[info]: cleaning up... got signal: %d\n", signum); cleanup_all(); exit(PASS); } int main(int argc, char *argv[]) { if (signal(SIGINT, sigint_handler) == SIG_ERR) { printf("Unable to configure SIGINT handler\n"); exit(FAIL); } setup_knet(argc, argv); setup_data_txrx_common(); sleep(5); restart: switch(test_type) { default: case TEST_PING: /* basic ping, no data */ sleep(5); break; case TEST_PING_AND_DATA: send_ping_data(); break; case TEST_PERF_BY_SIZE: if (senderid == thisnodeid) { send_perf_data_by_size(); } else { printf("[info]: waiting for perf rx thread to finish\n"); while(!wait_for_perf_rx) { sleep(1); } } break; case TEST_PERF_BY_TIME: if (senderid == thisnodeid) { send_perf_data_by_time(); } else { printf("[info]: waiting for perf rx thread to finish\n"); while(!wait_for_perf_rx) { sleep(1); } } break; } if (continous) { goto restart; } if (show_stats) { display_stats(show_stats); } cleanup_all(); return PASS; } diff --git a/libknet/threads_heartbeat.c b/libknet/threads_heartbeat.c index 12aaa767..4685ee8f 100644 --- a/libknet/threads_heartbeat.c +++ b/libknet/threads_heartbeat.c @@ -1,222 +1,225 @@ /* * Copyright (C) 2015-2018 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 "crypto.h" #include "links.h" #include "logging.h" #include "transports.h" #include "threads_common.h" #include "threads_heartbeat.h" static void _link_down(knet_handle_t knet_h, struct knet_host *dst_host, struct knet_link *dst_link) { memset(&dst_link->pmtud_last, 0, sizeof(struct timespec)); dst_link->received_pong = 0; dst_link->status.pong_last.tv_nsec = 0; dst_link->pong_timeout_backoff = KNET_LINK_PONG_TIMEOUT_BACKOFF; if (dst_link->status.connected == 1) { log_info(knet_h, KNET_SUB_LINK, "host: %u link: %u is down", dst_host->host_id, dst_link->link_id); _link_updown(knet_h, dst_host->host_id, dst_link->link_id, dst_link->status.enabled, 0); } } static void _handle_check_each(knet_handle_t knet_h, struct knet_host *dst_host, struct knet_link *dst_link, int timed) { int err = 0, savederrno = 0; int len; ssize_t outlen = KNET_HEADER_PING_SIZE; struct timespec clock_now, pong_last; unsigned long long diff_ping; unsigned char *outbuf = (unsigned char *)knet_h->pingbuf; if (dst_link->transport_connected == 0) { _link_down(knet_h, dst_host, dst_link); return; } /* caching last pong to avoid race conditions */ pong_last = dst_link->status.pong_last; if (clock_gettime(CLOCK_MONOTONIC, &clock_now) != 0) { log_debug(knet_h, KNET_SUB_HEARTBEAT, "Unable to get monotonic clock"); return; } timespec_diff(dst_link->ping_last, clock_now, &diff_ping); if ((diff_ping >= (dst_link->ping_interval * 1000llu)) || (!timed)) { memmove(&knet_h->pingbuf->khp_ping_time[0], &clock_now, sizeof(struct timespec)); knet_h->pingbuf->khp_ping_link = dst_link->link_id; if (pthread_mutex_lock(&knet_h->tx_seq_num_mutex)) { log_debug(knet_h, KNET_SUB_HEARTBEAT, "Unable to get seq mutex lock"); return; } knet_h->pingbuf->khp_ping_seq_num = htons(knet_h->tx_seq_num); pthread_mutex_unlock(&knet_h->tx_seq_num_mutex); knet_h->pingbuf->khp_ping_timed = timed; if (knet_h->crypto_instance) { if (crypto_encrypt_and_sign(knet_h, (const unsigned char *)knet_h->pingbuf, outlen, knet_h->pingbuf_crypt, &outlen) < 0) { log_debug(knet_h, KNET_SUB_HEARTBEAT, "Unable to crypto ping packet"); return; } outbuf = knet_h->pingbuf_crypt; knet_h->stats_extra.tx_crypt_ping_packets++; } retry: - len = sendto(dst_link->outsock, outbuf, outlen, - MSG_DONTWAIT | MSG_NOSIGNAL, (struct sockaddr *) &dst_link->dst_addr, - sizeof(struct sockaddr_storage)); + if (transport_get_connection_oriented(knet_h, dst_link->transport_type) == TRANSPORT_PROTO_NOT_CONNECTION_ORIENTED) { + len = sendto(dst_link->outsock, outbuf, outlen, MSG_DONTWAIT | MSG_NOSIGNAL, + (struct sockaddr *) &dst_link->dst_addr, sizeof(struct sockaddr_storage)); + } else { + len = sendto(dst_link->outsock, outbuf, outlen, MSG_DONTWAIT | MSG_NOSIGNAL, NULL, 0); + } savederrno = errno; dst_link->ping_last = clock_now; dst_link->status.stats.tx_ping_packets++; dst_link->status.stats.tx_ping_bytes += outlen; if (len != outlen) { err = transport_tx_sock_error(knet_h, dst_link->transport_type, dst_link->outsock, len, savederrno); switch(err) { case -1: /* unrecoverable error */ log_debug(knet_h, KNET_SUB_HEARTBEAT, "Unable to send ping (sock: %d) packet (sendto): %d %s. recorded src ip: %s src port: %s dst ip: %s dst port: %s", dst_link->outsock, savederrno, strerror(savederrno), dst_link->status.src_ipaddr, dst_link->status.src_port, dst_link->status.dst_ipaddr, dst_link->status.dst_port); dst_link->status.stats.tx_ping_errors++; break; case 0: break; case 1: dst_link->status.stats.tx_ping_retries++; goto retry; break; } } else { dst_link->last_ping_size = outlen; } } timespec_diff(pong_last, clock_now, &diff_ping); if ((pong_last.tv_nsec) && (diff_ping >= (dst_link->pong_timeout_adj * 1000llu))) { _link_down(knet_h, dst_host, dst_link); } } void _send_pings(knet_handle_t knet_h, int timed) { struct knet_host *dst_host; int link_idx; if (pthread_mutex_lock(&knet_h->hb_mutex)) { log_debug(knet_h, KNET_SUB_HEARTBEAT, "Unable to get hb mutex lock"); return; } 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++) { if ((dst_host->link[link_idx].status.enabled != 1) || (dst_host->link[link_idx].transport_type == KNET_TRANSPORT_LOOPBACK ) || ((dst_host->link[link_idx].dynamic == KNET_LINK_DYNIP) && (dst_host->link[link_idx].status.dynconnected != 1))) continue; _handle_check_each(knet_h, dst_host, &dst_host->link[link_idx], timed); } } pthread_mutex_unlock(&knet_h->hb_mutex); } static void _adjust_pong_timeouts(knet_handle_t knet_h) { struct knet_host *dst_host; struct knet_link *dst_link; int link_idx; if (pthread_mutex_lock(&knet_h->backoff_mutex)) { log_debug(knet_h, KNET_SUB_HEARTBEAT, "Unable to get backoff_mutex"); return; } 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++) { if ((dst_host->link[link_idx].status.enabled != 1) || (dst_host->link[link_idx].transport_type == KNET_TRANSPORT_LOOPBACK ) || ((dst_host->link[link_idx].dynamic == KNET_LINK_DYNIP) && (dst_host->link[link_idx].status.dynconnected != 1))) continue; dst_link = &dst_host->link[link_idx]; if (dst_link->pong_timeout_backoff > 1) { dst_link->pong_timeout_backoff--; } dst_link->pong_timeout_adj = (dst_link->pong_timeout * dst_link->pong_timeout_backoff) + (dst_link->status.stats.latency_max * KNET_LINK_PONG_TIMEOUT_LAT_MUL); } } pthread_mutex_unlock(&knet_h->backoff_mutex); } void *_handle_heartbt_thread(void *data) { knet_handle_t knet_h = (knet_handle_t) data; int i = 1; set_thread_status(knet_h, KNET_THREAD_HB, KNET_THREAD_STARTED); /* preparing ping buffer */ knet_h->pingbuf->kh_version = KNET_HEADER_VERSION; knet_h->pingbuf->kh_type = KNET_HEADER_TYPE_PING; knet_h->pingbuf->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_HEARTBEAT, "Unable to get read lock"); continue; } /* * _adjust_pong_timeouts should execute approx once a second. */ if ((i % (1000000 / KNET_THREADS_TIMERES)) == 0) { _adjust_pong_timeouts(knet_h); i = 1; } else { i++; } _send_pings(knet_h, 1); pthread_rwlock_unlock(&knet_h->global_rwlock); } set_thread_status(knet_h, KNET_THREAD_HB, KNET_THREAD_STOPPED); return NULL; } diff --git a/libknet/threads_pmtud.c b/libknet/threads_pmtud.c index 442a3465..33bee5ea 100644 --- a/libknet/threads_pmtud.c +++ b/libknet/threads_pmtud.c @@ -1,566 +1,569 @@ /* * Copyright (C) 2015-2018 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 "links.h" #include "host.h" #include "logging.h" #include "transports.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 err, ret, savederrno, mutex_retry_limit, failsafe, use_kernel_mtu, warn_once; uint32_t kernel_mtu; /* record kernel_mtu from EMSGSIZE */ size_t onwire_len; /* current packet onwire size */ size_t overhead_len; /* onwire packet overhead (protocol based) */ size_t max_mtu_len; /* max mtu for protocol */ size_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 */ size_t pad_len; /* crypto packet pad size, needs to move into crypto.c callbacks */ ssize_t len; /* len of what we were able to sendto onwire */ struct timespec ts; unsigned long long pong_timeout_adj_tmp; unsigned char *outbuf = (unsigned char *)knet_h->pmtudbuf; warn_once = 0; mutex_retry_limit = 0; failsafe = 0; pad_len = 0; 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->proto_overhead; dst_link->last_good_mtu = dst_link->last_ping_size + overhead_len; break; case AF_INET: max_mtu_len = KNET_PMTUD_SIZE_V4; overhead_len = KNET_PMTUD_OVERHEAD_V4 + dst_link->proto_overhead; dst_link->last_good_mtu = dst_link->last_ping_size + overhead_len; break; default: log_debug(knet_h, KNET_SUB_PMTUD, "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, "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, "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, (ssize_t *)&data_len) < 0) { log_debug(knet_h, KNET_SUB_PMTUD, "Unable to crypto pmtud packet"); return -1; } outbuf = knet_h->pmtudbuf_crypt; knet_h->stats_extra.tx_crypt_pmtu_packets++; } 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, "PMTUD detected host (%u) link (%u) has been disconnected", dst_host->host_id, dst_link->link_id); return -1; } if (dst_link->transport_connected != 1) { log_debug(knet_h, KNET_SUB_PMTUD, "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, "Unable to get mutex lock"); return -1; } if (knet_h->pmtud_abort) { pthread_mutex_unlock(&knet_h->pmtud_mutex); errno = EDEADLK; return -1; } savederrno = pthread_mutex_lock(&knet_h->tx_mutex); if (savederrno) { log_err(knet_h, KNET_SUB_PMTUD, "Unable to get TX mutex lock: %s", strerror(savederrno)); return -1; } retry: - len = sendto(dst_link->outsock, outbuf, data_len, - MSG_DONTWAIT | MSG_NOSIGNAL, (struct sockaddr *) &dst_link->dst_addr, - sizeof(struct sockaddr_storage)); + if (transport_get_connection_oriented(knet_h, dst_link->transport_type) == TRANSPORT_PROTO_NOT_CONNECTION_ORIENTED) { + len = sendto(dst_link->outsock, outbuf, data_len, MSG_DONTWAIT | MSG_NOSIGNAL, + (struct sockaddr *) &dst_link->dst_addr, sizeof(struct sockaddr_storage)); + } else { + len = sendto(dst_link->outsock, outbuf, data_len, MSG_DONTWAIT | MSG_NOSIGNAL, NULL, 0); + } savederrno = errno; /* * we cannot hold a lock on kmtu_mutex between resetting * knet_h->kernel_mtu here and below where it's used. * use_kernel_mtu tells us if the knet_h->kernel_mtu was * set to 0 and we can trust its value later. */ use_kernel_mtu = 0; if (pthread_mutex_lock(&knet_h->kmtu_mutex) == 0) { use_kernel_mtu = 1; knet_h->kernel_mtu = 0; pthread_mutex_unlock(&knet_h->kmtu_mutex); } kernel_mtu = 0; err = transport_tx_sock_error(knet_h, dst_link->transport_type, dst_link->outsock, len, savederrno); switch(err) { case -1: /* unrecoverable error */ log_debug(knet_h, KNET_SUB_PMTUD, "Unable to send pmtu packet (sendto): %d %s", savederrno, strerror(savederrno)); pthread_mutex_unlock(&knet_h->tx_mutex); pthread_mutex_unlock(&knet_h->pmtud_mutex); dst_link->status.stats.tx_pmtu_errors++; return -1; case 0: /* ignore error and continue */ break; case 1: /* retry to send those same data */ dst_link->status.stats.tx_pmtu_retries++; goto retry; break; } pthread_mutex_unlock(&knet_h->tx_mutex); if (len != (ssize_t )data_len) { if (savederrno == EMSGSIZE) { /* * we cannot hold a lock on kmtu_mutex between resetting * knet_h->kernel_mtu and here. * use_kernel_mtu tells us if the knet_h->kernel_mtu was * set to 0 previously and we can trust its value now. */ if (use_kernel_mtu) { use_kernel_mtu = 0; if (pthread_mutex_lock(&knet_h->kmtu_mutex) == 0) { kernel_mtu = knet_h->kernel_mtu; pthread_mutex_unlock(&knet_h->kmtu_mutex); } } if (kernel_mtu > 0) { dst_link->last_bad_mtu = kernel_mtu + 1; } else { dst_link->last_bad_mtu = onwire_len; } } else { log_debug(knet_h, KNET_SUB_PMTUD, "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; dst_link->status.stats.tx_pmtu_packets++; dst_link->status.stats.tx_pmtu_bytes += data_len; if (clock_gettime(CLOCK_REALTIME, &ts) < 0) { log_debug(knet_h, KNET_SUB_PMTUD, "Unable to get current time: %s", strerror(errno)); pthread_mutex_unlock(&knet_h->pmtud_mutex); return -1; } /* * set PMTUd reply timeout to match pong_timeout on a given link * * math: internally pong_timeout is expressed in microseconds, while * the public API exports milliseconds. So careful with the 0's here. * the loop is necessary because we are grabbing the current time just above * and add values to it that could overflow into seconds. */ if (pthread_mutex_lock(&knet_h->backoff_mutex)) { log_debug(knet_h, KNET_SUB_PMTUD, "Unable to get backoff_mutex"); pthread_mutex_unlock(&knet_h->pmtud_mutex); return -1; } if (knet_h->crypto_instance) { /* * crypto, under pressure, is a royal PITA */ pong_timeout_adj_tmp = dst_link->pong_timeout_adj * 2; } else { pong_timeout_adj_tmp = dst_link->pong_timeout_adj; } ts.tv_sec += pong_timeout_adj_tmp / 1000000; ts.tv_nsec += (((pong_timeout_adj_tmp) % 1000000) * 1000); while (ts.tv_nsec > 1000000000) { ts.tv_sec += 1; ts.tv_nsec -= 1000000000; } pthread_mutex_unlock(&knet_h->backoff_mutex); knet_h->pmtud_waiting = 1; ret = pthread_cond_timedwait(&knet_h->pmtud_cond, &knet_h->pmtud_mutex, &ts); knet_h->pmtud_waiting = 0; if (knet_h->pmtud_abort) { pthread_mutex_unlock(&knet_h->pmtud_mutex); errno = EDEADLK; return -1; } if (shutdown_in_progress(knet_h)) { pthread_mutex_unlock(&knet_h->pmtud_mutex); log_debug(knet_h, KNET_SUB_PMTUD, "PMTUD aborted. shutdown in progress"); return -1; } if (ret) { if (ret == ETIMEDOUT) { if (!warn_once) { log_warn(knet_h, KNET_SUB_PMTUD, "possible MTU misconfiguration detected. " "kernel is reporting MTU: %u bytes for " "host %u link %u but the other node is " "not acknowledging packets of this size. ", dst_link->last_sent_mtu, dst_host->host_id, dst_link->link_id); log_warn(knet_h, KNET_SUB_PMTUD, "This can be caused by this node interface MTU " "too big or a network device that does not " "support or has been misconfigured to manage MTU " "of this size, or packet loss. knet will continue " "to run but performances might be affected."); warn_once = 1; } } else { pthread_mutex_unlock(&knet_h->pmtud_mutex); if (mutex_retry_limit == 3) { log_debug(knet_h, KNET_SUB_PMTUD, "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))) || (dst_link->last_bad_mtu == dst_link->last_good_mtu)) { 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; } } if (kernel_mtu) { onwire_len = kernel_mtu; } else { 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, int force_run) { uint8_t saved_valid_pmtud; unsigned int saved_pmtud; struct timespec clock_now; unsigned long long diff_pmtud, interval; if (!force_run) { interval = knet_h->pmtud_interval * 1000000000llu; /* nanoseconds */ if (clock_gettime(CLOCK_MONOTONIC, &clock_now) != 0) { log_debug(knet_h, KNET_SUB_PMTUD, "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 + dst_link->proto_overhead + KNET_HEADER_ALL_SIZE + knet_h->sec_header_size; break; case AF_INET: dst_link->status.proto_overhead = KNET_PMTUD_OVERHEAD_V4 + dst_link->proto_overhead + 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, "Starting PMTUD for host: %u link: %u", dst_host->host_id, dst_link->link_id); errno = 0; if (_handle_check_link_pmtud(knet_h, dst_host, dst_link) < 0) { if (errno == EDEADLK) { log_debug(knet_h, KNET_SUB_PMTUD, "PMTUD for host: %u link: %u has been rescheduled", dst_host->host_id, dst_link->link_id); dst_link->status.mtu = saved_pmtud; dst_link->has_valid_mtu = saved_valid_pmtud; errno = EDEADLK; return dst_link->has_valid_mtu; } 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, "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, "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, "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, "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; } /* * set pmtud_last, if we can, after we are done with the PMTUd process * because it can take a very long time. */ dst_link->pmtud_last = clock_now; if (!clock_gettime(CLOCK_MONOTONIC, &clock_now)) { 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; unsigned int lower_mtu; int link_has_mtu; int force_run = 0; set_thread_status(knet_h, KNET_THREAD_PMTUD, KNET_THREAD_STARTED); 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_mutex_lock(&knet_h->pmtud_mutex) != 0) { log_debug(knet_h, KNET_SUB_PMTUD, "Unable to get mutex lock"); continue; } knet_h->pmtud_abort = 0; knet_h->pmtud_running = 1; force_run = knet_h->pmtud_forcerun; knet_h->pmtud_forcerun = 0; pthread_mutex_unlock(&knet_h->pmtud_mutex); if (force_run) { log_debug(knet_h, KNET_SUB_PMTUD, "PMTUd request to rerun has been received"); } if (pthread_rwlock_rdlock(&knet_h->global_rwlock) != 0) { log_debug(knet_h, KNET_SUB_PMTUD, "Unable to get read lock"); continue; } lower_mtu = KNET_PMTUD_SIZE_V4; 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_host->link[link_idx].transport_type == KNET_TRANSPORT_LOOPBACK) || (!dst_link->last_ping_size) || ((dst_link->dynamic == KNET_LINK_DYNIP) && (dst_link->status.dynconnected != 1))) continue; link_has_mtu = _handle_check_pmtud(knet_h, dst_host, dst_link, &min_mtu, force_run); if (errno == EDEADLK) { goto out_unlock; } if (link_has_mtu) { have_mtu = 1; if (min_mtu < lower_mtu) { lower_mtu = min_mtu; } } } } if (have_mtu) { if (knet_h->data_mtu != lower_mtu) { knet_h->data_mtu = lower_mtu; log_info(knet_h, KNET_SUB_PMTUD, "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); } } } out_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); if (pthread_mutex_lock(&knet_h->pmtud_mutex) != 0) { log_debug(knet_h, KNET_SUB_PMTUD, "Unable to get mutex lock"); } else { knet_h->pmtud_running = 0; pthread_mutex_unlock(&knet_h->pmtud_mutex); } } set_thread_status(knet_h, KNET_THREAD_PMTUD, KNET_THREAD_STOPPED); return NULL; } diff --git a/libknet/threads_rx.c b/libknet/threads_rx.c index d2767e17..4877b729 100644 --- a/libknet/threads_rx.c +++ b/libknet/threads_rx.c @@ -1,851 +1,857 @@ /* * Copyright (C) 2012-2018 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 "compat.h" #include "compress.h" #include "crypto.h" #include "host.h" #include "links.h" #include "logging.h" #include "transports.h" #include "transport_common.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 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; int was_decrypted = 0; uint64_t crypt_time = 0; 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) { struct timespec start_time; struct timespec end_time; clock_gettime(CLOCK_MONOTONIC, &start_time); 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; } clock_gettime(CLOCK_MONOTONIC, &end_time); timespec_diff(start_time, end_time, &crypt_time); if (crypt_time < knet_h->stats.rx_crypt_time_min) { knet_h->stats.rx_crypt_time_min = crypt_time; } if (crypt_time > knet_h->stats.rx_crypt_time_max) { knet_h->stats.rx_crypt_time_max = crypt_time; } len = outlen; inbuf = (struct knet_header *)knet_h->recv_from_links_buf_decrypt; was_decrypted++; } if (len < (ssize_t)(KNET_HEADER_SIZE + 1)) { log_debug(knet_h, KNET_SUB_RX, "Packet is too short: %ld", (long)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; src_link = src_host->link + (inbuf->khp_ping_link % KNET_MAX_LINK); if ((inbuf->kh_type & KNET_HEADER_TYPE_PMSK) != 0) { 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 */ 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 (src_link) { src_link->status.stats.rx_data_packets++; src_link->status.stats.rx_data_bytes += len; } 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->khp_data_compress) { ssize_t decmp_outlen = KNET_DATABUFSIZE_COMPRESS; struct timespec start_time; struct timespec end_time; uint64_t compress_time; clock_gettime(CLOCK_MONOTONIC, &start_time); err = decompress(knet_h, inbuf->khp_data_compress, (const unsigned char *)inbuf->khp_data_userdata, len - KNET_HEADER_DATA_SIZE, knet_h->recv_from_links_buf_decompress, &decmp_outlen); if (!err) { /* Collect stats */ clock_gettime(CLOCK_MONOTONIC, &end_time); timespec_diff(start_time, end_time, &compress_time); if (compress_time < knet_h->stats.rx_compress_time_min) { knet_h->stats.rx_compress_time_min = compress_time; } if (compress_time > knet_h->stats.rx_compress_time_max) { knet_h->stats.rx_compress_time_max = compress_time; } knet_h->stats.rx_compress_time_ave = (knet_h->stats.rx_compress_time_ave * knet_h->stats.rx_compressed_packets + compress_time) / (knet_h->stats.rx_compressed_packets+1); knet_h->stats.rx_compressed_packets++; knet_h->stats.rx_compressed_original_bytes += decmp_outlen; knet_h->stats.rx_compressed_size_bytes += len - KNET_HEADER_SIZE; memmove(inbuf->khp_data_userdata, knet_h->recv_from_links_buf_decompress, decmp_outlen); len = decmp_outlen + KNET_HEADER_DATA_SIZE; } else { log_warn(knet_h, KNET_SUB_COMPRESS, "Unable to decompress packet (%d): %s", err, strerror(errno)); return; } } if (inbuf->kh_type == KNET_HEADER_TYPE_DATA) { if (knet_h->enabled != 1) /* data forward is disabled */ break; /* Only update the crypto overhead for data packets. Mainly to be consistent with TX */ knet_h->stats.rx_crypt_time_ave = (knet_h->stats.rx_crypt_time_ave * knet_h->stats.rx_crypt_packets + crypt_time) / (knet_h->stats.rx_crypt_packets+1); knet_h->stats.rx_crypt_packets++; if (knet_h->dst_host_filter_fn) { size_t 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) { if (dst_host_ids_entries > KNET_MAX_HOST) { log_debug(knet_h, KNET_SUB_RX, "dst_host_filter_fn returned too many destinations"); return; } 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 ((size_t)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); src_link->status.stats.rx_ping_packets++; src_link->status.stats.rx_ping_bytes += len; 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, outlen, 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; knet_h->stats_extra.tx_crypt_pong_packets++; } retry_pong: - len = sendto(src_link->outsock, outbuf, outlen, MSG_DONTWAIT | MSG_NOSIGNAL, - (struct sockaddr *) &src_link->dst_addr, - sizeof(struct sockaddr_storage)); + if (transport_get_connection_oriented(knet_h, src_link->transport_type) == TRANSPORT_PROTO_NOT_CONNECTION_ORIENTED) { + len = sendto(src_link->outsock, outbuf, outlen, MSG_DONTWAIT | MSG_NOSIGNAL, + (struct sockaddr *) &src_link->dst_addr, sizeof(struct sockaddr_storage)); + } else { + len = sendto(src_link->outsock, outbuf, outlen, MSG_DONTWAIT | MSG_NOSIGNAL, NULL, 0); + } savederrno = errno; if (len != outlen) { err = transport_tx_sock_error(knet_h, src_link->transport_type, 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); src_link->status.stats.tx_pong_errors++; break; case 0: /* ignore error and continue */ break; case 1: /* retry to send those same data */ src_link->status.stats.tx_pong_retries++; goto retry_pong; break; } } src_link->status.stats.tx_pong_packets++; src_link->status.stats.tx_pong_bytes += outlen; break; case KNET_HEADER_TYPE_PONG: src_link->status.stats.rx_pong_packets++; src_link->status.stats.rx_pong_bytes += len; 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_adj) { 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); } } } /* Calculate latency stats */ if (src_link->status.latency > src_link->status.stats.latency_max) { src_link->status.stats.latency_max = src_link->status.latency; } if (src_link->status.latency < src_link->status.stats.latency_min) { src_link->status.stats.latency_min = src_link->status.latency; } src_link->status.stats.latency_ave = (src_link->status.stats.latency_ave * src_link->status.stats.latency_samples + src_link->status.latency) / (src_link->status.stats.latency_samples+1); src_link->status.stats.latency_samples++; break; case KNET_HEADER_TYPE_PMTUD: src_link->status.stats.rx_pmtu_packets++; src_link->status.stats.rx_pmtu_bytes += len; 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, outlen, 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; knet_h->stats_extra.tx_crypt_pmtu_reply_packets++; } savederrno = pthread_mutex_lock(&knet_h->tx_mutex); if (savederrno) { log_err(knet_h, KNET_SUB_RX, "Unable to get TX mutex lock: %s", strerror(savederrno)); goto out_pmtud; } retry_pmtud: - len = sendto(src_link->outsock, outbuf, outlen, MSG_DONTWAIT | MSG_NOSIGNAL, - (struct sockaddr *) &src_link->dst_addr, - sizeof(struct sockaddr_storage)); + if (transport_get_connection_oriented(knet_h, src_link->transport_type) == TRANSPORT_PROTO_NOT_CONNECTION_ORIENTED) { + len = sendto(src_link->outsock, outbuf, outlen, MSG_DONTWAIT | MSG_NOSIGNAL, + (struct sockaddr *) &src_link->dst_addr, sizeof(struct sockaddr_storage)); + } else { + len = sendto(src_link->outsock, outbuf, outlen, MSG_DONTWAIT | MSG_NOSIGNAL, NULL, 0); + } savederrno = errno; if (len != outlen) { err = transport_tx_sock_error(knet_h, src_link->transport_type, 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); src_link->status.stats.tx_pmtu_errors++; break; case 0: /* ignore error and continue */ src_link->status.stats.tx_pmtu_errors++; break; case 1: /* retry to send those same data */ src_link->status.stats.tx_pmtu_retries++; goto retry_pmtud; break; } } pthread_mutex_unlock(&knet_h->tx_mutex); out_pmtud: break; case KNET_HEADER_TYPE_PMTUD_REPLY: src_link->status.stats.rx_pmtu_packets++; src_link->status.stats.rx_pmtu_bytes += len; 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 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_RX_BUFS; i++) { msg[i].msg_hdr.msg_namelen = sizeof(struct sockaddr_storage); } msg_recv = _recvmmsg(sockfd, &msg[0], PCKT_RX_BUFS, MSG_DONTWAIT | MSG_NOSIGNAL); 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) { transport_rx_sock_error(knet_h, transport, sockfd, msg_recv, savederrno); goto exit_unlock; } for (i = 0; i < msg_recv; i++) { err = transport_rx_is_data(knet_h, transport, 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_RX_BUFS]; struct knet_mmsghdr msg[PCKT_RX_BUFS]; struct iovec iov_in[PCKT_RX_BUFS]; set_thread_status(knet_h, KNET_THREAD_RX, KNET_THREAD_STARTED); memset(&msg, 0, sizeof(msg)); for (i = 0; i < PCKT_RX_BUFS; 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, KNET_THREADS_TIMERES / 1000); /* * we use timeout to detect if thread is shutting down */ if (nev == 0) { continue; } for (i = 0; i < nev; i++) { _handle_recv_from_links(knet_h, events[i].data.fd, msg); } } set_thread_status(knet_h, KNET_THREAD_RX, KNET_THREAD_STOPPED); return NULL; } diff --git a/libknet/threads_tx.c b/libknet/threads_tx.c index 1048bf60..da986140 100644 --- a/libknet/threads_tx.c +++ b/libknet/threads_tx.c @@ -1,748 +1,749 @@ /* * Copyright (C) 2012-2018 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 "compat.h" #include "compress.h" #include "crypto.h" #include "host.h" #include "link.h" #include "logging.h" #include "transports.h" #include "transport_common.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 knet_mmsghdr *msg, int msgs_to_send) { int link_idx, msg_idx, sent_msgs, prev_sent, progress; int err = 0, savederrno = 0; unsigned int i; struct knet_mmsghdr *cur; struct knet_link *cur_link; for (link_idx = 0; link_idx < dst_host->active_link_entries; link_idx++) { sent_msgs = 0; prev_sent = 0; progress = 1; cur_link = &dst_host->link[dst_host->active_links[link_idx]]; if (cur_link->transport_type == KNET_TRANSPORT_LOOPBACK) { continue; } msg_idx = 0; while (msg_idx < msgs_to_send) { msg[msg_idx].msg_hdr.msg_name = &cur_link->dst_addr; /* Cast for Linux/BSD compatibility */ for (i=0; i<(unsigned int)msg[msg_idx].msg_hdr.msg_iovlen; i++) { cur_link->status.stats.tx_data_bytes += msg[msg_idx].msg_hdr.msg_iov[i].iov_len; } cur_link->status.stats.tx_data_packets++; msg_idx++; } retry: cur = &msg[prev_sent]; sent_msgs = _sendmmsg(dst_host->link[dst_host->active_links[link_idx]].outsock, + transport_get_connection_oriented(knet_h, dst_host->link[dst_host->active_links[link_idx]].transport_type), &cur[0], msgs_to_send - prev_sent, MSG_DONTWAIT | MSG_NOSIGNAL); savederrno = errno; err = transport_tx_sock_error(knet_h, dst_host->link[dst_host->active_links[link_idx]].transport_type, dst_host->link[dst_host->active_links[link_idx]].outsock, sent_msgs, savederrno); switch(err) { case -1: /* unrecoverable error */ cur_link->status.stats.tx_data_errors++; goto out_unlock; break; case 0: /* ignore error and continue */ break; case 1: /* retry to send those same data */ cur_link->status.stats.tx_data_retries++; 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, size_t inlen, int8_t channel, int is_sync) { size_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][2]; int iovcnt_out = 2; uint8_t frag_idx; unsigned int temp_data_mtu; size_t host_idx; int send_mcast = 0; struct knet_header *inbuf; int savederrno = 0; int err = 0; seq_num_t tx_seq_num; struct knet_mmsghdr msg[PCKT_FRAG_MAX]; int msgs_to_send, msg_idx; unsigned int i; int j; int send_local = 0; int data_compressed = 0; size_t uncrypted_frag_size; inbuf = knet_h->recv_from_sock_buf; 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; } if ((!bcast) && (dst_host_ids_entries_temp > KNET_MAX_HOST)) { log_debug(knet_h, KNET_SUB_TX, "dst_host_filter_fn returned too many destinations"); savederrno = EINVAL; err = -1; goto out_unlock; } } /* Send to localhost if appropriate and enabled */ if (knet_h->has_loop_link) { send_local = 0; if (bcast) { send_local = 1; } else { for (i=0; i< dst_host_ids_entries_temp; i++) { if (dst_host_ids_temp[i] == knet_h->host_id) { send_local = 1; } } } if (send_local) { const unsigned char *buf = inbuf->khp_data_userdata; ssize_t buflen = inlen; struct knet_link *local_link; local_link = knet_h->host_index[knet_h->host_id]->link; local_retry: err = write(knet_h->sockfd[channel].sockfd[knet_h->sockfd[channel].is_created], buf, buflen); if (err < 0) { log_err(knet_h, KNET_SUB_TRANSP_LOOPBACK, "send local failed. error=%s\n", strerror(errno)); local_link->status.stats.tx_data_errors++; } if (err > 0 && err < buflen) { log_debug(knet_h, KNET_SUB_TRANSP_LOOPBACK, "send local incomplete=%d bytes of %zu\n", err, inlen); local_link->status.stats.tx_data_retries++; buf += err; buflen -= err; usleep(KNET_THREADS_TIMERES / 16); goto local_retry; } if (err == buflen) { local_link->status.stats.tx_data_packets++; local_link->status.stats.tx_data_bytes += inlen; } } } 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 unreachable 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->host_id == knet_h->host_id && knet_h->has_loop_link) && 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->host_id == knet_h->host_id && knet_h->has_loop_link) && 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 minimum 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; } /* * compress data */ if ((knet_h->compress_model > 0) && (inlen > knet_h->compress_threshold)) { size_t cmp_outlen = KNET_DATABUFSIZE_COMPRESS; struct timespec start_time; struct timespec end_time; uint64_t compress_time; clock_gettime(CLOCK_MONOTONIC, &start_time); err = compress(knet_h, (const unsigned char *)inbuf->khp_data_userdata, inlen, knet_h->send_to_links_buf_compress, (ssize_t *)&cmp_outlen); if (err < 0) { log_warn(knet_h, KNET_SUB_COMPRESS, "Compression failed (%d): %s", err, strerror(errno)); } else { /* Collect stats */ clock_gettime(CLOCK_MONOTONIC, &end_time); timespec_diff(start_time, end_time, &compress_time); if (compress_time < knet_h->stats.tx_compress_time_min) { knet_h->stats.tx_compress_time_min = compress_time; } if (compress_time > knet_h->stats.tx_compress_time_max) { knet_h->stats.tx_compress_time_max = compress_time; } knet_h->stats.tx_compress_time_ave = (unsigned long long)(knet_h->stats.tx_compress_time_ave * knet_h->stats.tx_compressed_packets + compress_time) / (knet_h->stats.tx_compressed_packets+1); knet_h->stats.tx_compressed_packets++; knet_h->stats.tx_compressed_original_bytes += inlen; knet_h->stats.tx_compressed_size_bytes += cmp_outlen; if (cmp_outlen < inlen) { memmove(inbuf->khp_data_userdata, knet_h->send_to_links_buf_compress, cmp_outlen); inlen = cmp_outlen; data_compressed = 1; } } } if (knet_h->compress_model > 0 && !data_compressed) { knet_h->stats.tx_uncompressed_packets++; } /* * 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 (data_compressed) { inbuf->khp_data_compress = knet_h->compress_model; } else { inbuf->khp_data_compress = 0; } 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][0].iov_base = (void *)knet_h->send_to_links_buf[frag_idx]; iov_out[frag_idx][0].iov_len = KNET_HEADER_DATA_SIZE; iov_out[frag_idx][1].iov_base = inbuf->khp_data_userdata + (temp_data_mtu * frag_idx); /* * set the len */ if (frag_len > temp_data_mtu) { iov_out[frag_idx][1].iov_len = temp_data_mtu; } else { iov_out[frag_idx][1].iov_len = frag_len; } /* * 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; knet_h->send_to_links_buf[frag_idx]->khp_data_compress = inbuf->khp_data_compress; frag_len = frag_len - temp_data_mtu; frag_idx++; } iovcnt_out = 2; } else { iov_out[frag_idx][0].iov_base = (void *)inbuf; iov_out[frag_idx][0].iov_len = frag_len + KNET_HEADER_DATA_SIZE; iovcnt_out = 1; } if (knet_h->crypto_instance) { struct timespec start_time; struct timespec end_time; uint64_t crypt_time; frag_idx = 0; while (frag_idx < inbuf->khp_data_frag_num) { clock_gettime(CLOCK_MONOTONIC, &start_time); if (crypto_encrypt_and_signv( knet_h, iov_out[frag_idx], iovcnt_out, knet_h->send_to_links_buf_crypt[frag_idx], (ssize_t *)&outlen) < 0) { log_debug(knet_h, KNET_SUB_TX, "Unable to encrypt packet"); savederrno = ECHILD; err = -1; goto out_unlock; } clock_gettime(CLOCK_MONOTONIC, &end_time); timespec_diff(start_time, end_time, &crypt_time); if (crypt_time < knet_h->stats.tx_crypt_time_min) { knet_h->stats.tx_crypt_time_min = crypt_time; } if (crypt_time > knet_h->stats.tx_crypt_time_max) { knet_h->stats.tx_crypt_time_max = crypt_time; } knet_h->stats.tx_crypt_time_ave = (knet_h->stats.tx_crypt_time_ave * knet_h->stats.tx_crypt_packets + crypt_time) / (knet_h->stats.tx_crypt_packets+1); uncrypted_frag_size = 0; for (j=0; j < iovcnt_out; j++) { uncrypted_frag_size += iov_out[frag_idx][j].iov_len; } knet_h->stats.tx_crypt_byte_overhead += (outlen - uncrypted_frag_size); knet_h->stats.tx_crypt_packets++; iov_out[frag_idx][0].iov_base = knet_h->send_to_links_buf_crypt[frag_idx]; iov_out[frag_idx][0].iov_len = outlen; frag_idx++; } iovcnt_out = 1; } 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][0]; msg[msg_idx].msg_hdr.msg_iovlen = iovcnt_out; 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->kh_type = KNET_HEADER_TYPE_DATA; memmove(knet_h->recv_from_sock_buf->khp_data_userdata, buff, buff_len); err = _parse_recv_from_sock(knet_h, buff_len, channel, 1); savederrno = errno; pthread_mutex_unlock(&knet_h->tx_mutex); out: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = err ? savederrno : 0; return err; } static void _handle_send_to_links(knet_handle_t knet_h, struct msghdr *msg, int sockfd, int8_t channel, int type) { ssize_t inlen = 0; int savederrno = 0, docallback = 0; if ((channel >= 0) && (channel < KNET_DATAFD_MAX) && (!knet_h->sockfd[channel].is_socket)) { inlen = readv(sockfd, msg->msg_iov, 1); } else { inlen = recvmsg(sockfd, msg, MSG_DONTWAIT | MSG_NOSIGNAL); } if (inlen == 0) { savederrno = 0; docallback = 1; } else if (inlen < 0) { struct epoll_event ev; savederrno = errno; docallback = 1; 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; } } else { knet_h->recv_from_sock_buf->kh_type = type; _parse_recv_from_sock(knet_h, inlen, channel, 0); } 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]; int i, nev, type; int8_t channel; struct iovec iov_in; struct msghdr msg; struct sockaddr_storage address; set_thread_status(knet_h, KNET_THREAD_TX, KNET_THREAD_STARTED); memset(&iov_in, 0, sizeof(iov_in)); iov_in.iov_base = (void *)knet_h->recv_from_sock_buf->khp_data_userdata; iov_in.iov_len = KNET_MAX_PACKET_SIZE; memset(&msg, 0, sizeof(struct msghdr)); msg.msg_name = &address; msg.msg_namelen = sizeof(struct sockaddr_storage); msg.msg_iov = &iov_in; msg.msg_iovlen = 1; knet_h->recv_from_sock_buf->kh_version = KNET_HEADER_VERSION; knet_h->recv_from_sock_buf->khp_data_frag_seq = 0; knet_h->recv_from_sock_buf->kh_node = htons(knet_h->host_id); for (i = 0; i < PCKT_FRAG_MAX; i++) { 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, KNET_THREADS_TIMERES / 1000); /* * we use timeout to detect if thread is shutting down */ if (nev == 0) { continue; } 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 (channel >= KNET_DATAFD_MAX) { log_debug(knet_h, KNET_SUB_TX, "No available channels"); continue; /* channel not found */ } } 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, &msg, events[i].data.fd, channel, type); pthread_mutex_unlock(&knet_h->tx_mutex); } pthread_rwlock_unlock(&knet_h->global_rwlock); } set_thread_status(knet_h, KNET_THREAD_TX, KNET_THREAD_STOPPED); return NULL; } diff --git a/libknet/transport_common.c b/libknet/transport_common.c index 6df50027..cd344932 100644 --- a/libknet/transport_common.c +++ b/libknet/transport_common.c @@ -1,439 +1,443 @@ /* * Copyright (C) 2016-2018 Red Hat, Inc. All rights reserved. * * Author: Fabio M. Di Nitto * * This software licensed under GPL-2.0+, LGPL-2.0+ */ #include "config.h" #include #include #include #include #include #include #include #include #include "libknet.h" #include "compat.h" #include "host.h" #include "link.h" #include "logging.h" #include "common.h" #include "transport_common.h" /* * reuse Jan Friesse's compat layer as wrapper to drop usage of sendmmsg * * TODO: kill those wrappers once we work on packet delivery guarantees */ int _recvmmsg(int sockfd, struct knet_mmsghdr *msgvec, unsigned int vlen, unsigned int flags) { int savederrno = 0, err = 0; unsigned int i; for (i = 0; i < vlen; i++) { err = recvmsg(sockfd, &msgvec[i].msg_hdr, flags); savederrno = errno; if (err >= 0) { msgvec[i].msg_len = err; } else { if ((i > 0) && ((errno == EAGAIN) || (errno == EWOULDBLOCK))) { savederrno = 0; } break; } } errno = savederrno; return ((i > 0) ? (int)i : err); } -int _sendmmsg(int sockfd, struct knet_mmsghdr *msgvec, unsigned int vlen, unsigned int flags) +int _sendmmsg(int sockfd, int connection_oriented, struct knet_mmsghdr *msgvec, unsigned int vlen, unsigned int flags) { int savederrno = 0, err = 0; unsigned int i; for (i = 0; i < vlen; i++) { + if (connection_oriented == TRANSPORT_PROTO_IS_CONNECTION_ORIENTED) { + msgvec[i].msg_hdr.msg_name = NULL; + msgvec[i].msg_hdr.msg_namelen = 0; + } err = sendmsg(sockfd, &msgvec[i].msg_hdr, flags); savederrno = errno; if (err < 0) { break; } } errno = savederrno; return ((i > 0) ? (int)i : err); } /* Assume neither of these constants can ever be zero */ #ifndef SO_RCVBUFFORCE #define SO_RCVBUFFORCE 0 #endif #ifndef SO_SNDBUFFORCE #define SO_SNDBUFFORCE 0 #endif static int _configure_sockbuf(knet_handle_t knet_h, int sock, int option, int force, int target) { int savederrno = 0; int new_value; socklen_t value_len = sizeof new_value; if (setsockopt(sock, SOL_SOCKET, option, &target, sizeof target) != 0) { savederrno = errno; log_err(knet_h, KNET_SUB_TRANSPORT, "Error setting socket buffer via option %d to value %d: %s\n", option, target, strerror(savederrno)); errno = savederrno; return -1; } if (getsockopt(sock, SOL_SOCKET, option, &new_value, &value_len) != 0) { savederrno = errno; log_err(knet_h, KNET_SUB_TRANSPORT, "Error getting socket buffer via option %d: %s\n", option, strerror(savederrno)); errno = savederrno; return -1; } if (value_len != sizeof new_value) { log_err(knet_h, KNET_SUB_TRANSPORT, "Socket option %d returned unexpected size %u\n", option, value_len); errno = ERANGE; return -1; } if (target <= new_value) { return 0; } if (!force || !(knet_h->flags & KNET_HANDLE_FLAG_PRIVILEGED)) { log_err(knet_h, KNET_SUB_TRANSPORT, "Failed to set socket buffer via option %d to value %d: capped at %d", option, target, new_value); if (!(knet_h->flags & KNET_HANDLE_FLAG_PRIVILEGED)) { log_err(knet_h, KNET_SUB_TRANSPORT, "Continuing regardless, as the handle is not privileged." " Expect poor performance!"); return 0; } else { errno = ENAMETOOLONG; return -1; } } if (setsockopt(sock, SOL_SOCKET, force, &target, sizeof target) < 0) { savederrno = errno; log_err(knet_h, KNET_SUB_TRANSPORT, "Failed to set socket buffer via force option %d: %s", force, strerror(savederrno)); if (savederrno == EPERM) { errno = ENAMETOOLONG; } else { errno = savederrno; } return -1; } return 0; } int _configure_common_socket(knet_handle_t knet_h, int sock, uint64_t flags, const char *type) { int err = 0, savederrno = 0; int value; if (_fdset_cloexec(sock)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set %s CLOEXEC socket opts: %s", type, strerror(savederrno)); goto exit_error; } if (_fdset_nonblock(sock)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set %s NONBLOCK socket opts: %s", type, strerror(savederrno)); goto exit_error; } if (_configure_sockbuf(knet_h, sock, SO_RCVBUF, SO_RCVBUFFORCE, KNET_RING_RCVBUFF)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set %s receive buffer: %s", type, strerror(savederrno)); goto exit_error; } if (_configure_sockbuf(knet_h, sock, SO_SNDBUF, SO_SNDBUFFORCE, KNET_RING_RCVBUFF)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set %s send buffer: %s", type, strerror(savederrno)); goto exit_error; } if (flags & KNET_LINK_FLAG_TRAFFICHIPRIO) { #ifdef KNET_LINUX #ifdef SO_PRIORITY value = 6; /* TC_PRIO_INTERACTIVE */ if (setsockopt(sock, SOL_SOCKET, SO_PRIORITY, &value, sizeof(value)) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set %s priority: %s", type, strerror(savederrno)); goto exit_error; } log_debug(knet_h, KNET_SUB_TRANSPORT, "TC_PRIO_INTERACTIVE enabled on socket: %i", sock); #else log_debug(knet_h, KNET_SUB_TRANSPORT, "TC_PRIO_INTERACTIVE not available in this build/platform"); #endif #endif #if defined(IP_TOS) && defined(IPTOS_LOWDELAY) value = IPTOS_LOWDELAY; if (setsockopt(sock, IPPROTO_IP, IP_TOS, &value, sizeof(value)) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set %s priority: %s", type, strerror(savederrno)); goto exit_error; } log_debug(knet_h, KNET_SUB_TRANSPORT, "IPTOS_LOWDELAY enabled on socket: %i", sock); #else log_debug(knet_h, KNET_SUB_TRANSPORT, "IPTOS_LOWDELAY not available in this build/platform"); #endif } exit_error: errno = savederrno; return err; } int _configure_transport_socket(knet_handle_t knet_h, int sock, struct sockaddr_storage *address, uint64_t flags, const char *type) { int err = 0, savederrno = 0; int value; if (_configure_common_socket(knet_h, sock, flags, type) < 0) { savederrno = errno; err = -1; goto exit_error; } #ifdef KNET_LINUX #ifdef IP_FREEBIND value = 1; if (setsockopt(sock, SOL_IP, IP_FREEBIND, &value, sizeof(value)) <0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set FREEBIND on %s socket: %s", type, strerror(savederrno)); goto exit_error; } log_debug(knet_h, KNET_SUB_TRANSPORT, "FREEBIND enabled on socket: %i", sock); #else log_debug(knet_h, KNET_SUB_TRANSPORT, "FREEBIND not available in this build/platform"); #endif #endif #ifdef KNET_BSD #ifdef IP_BINDANY /* BSD */ value = 1; if (setsockopt(sock, IPPROTO_IP, IP_BINDANY, &value, sizeof(value)) <0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set BINDANY on %s socket: %s", type, strerror(savederrno)); goto exit_error; } log_debug(knet_h, KNET_SUB_TRANSPORT, "BINDANY enabled on socket: %i", sock); #else log_debug(knet_h, KNET_SUB_TRANSPORT, "BINDANY not available in this build/platform"); #endif #endif if (address->ss_family == AF_INET6) { value = 1; if (setsockopt(sock, IPPROTO_IPV6, IPV6_V6ONLY, &value, sizeof(value)) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set %s IPv6 only: %s", type, strerror(savederrno)); goto exit_error; } #ifdef KNET_LINUX #ifdef IPV6_MTU_DISCOVER value = IPV6_PMTUDISC_PROBE; if (setsockopt(sock, SOL_IPV6, IPV6_MTU_DISCOVER, &value, sizeof(value)) <0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set PMTUDISC on %s socket: %s", type, strerror(savederrno)); goto exit_error; } log_debug(knet_h, KNET_SUB_TRANSPORT, "IPV6_MTU_DISCOVER enabled on socket: %i", sock); #else log_debug(knet_h, KNET_SUB_TRANSPORT, "IPV6_MTU_DISCOVER not available in this build/platform"); #endif #endif #ifdef IPV6_DONTFRAG value = 1; if (setsockopt(sock, IPPROTO_IPV6, IPV6_DONTFRAG, &value, sizeof(value)) <0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set DONTFRAG on %s socket: %s", type, strerror(savederrno)); goto exit_error; } log_debug(knet_h, KNET_SUB_TRANSPORT, "IPV6_DONTFRAG enabled on socket: %i", sock); #else log_debug(knet_h, KNET_SUB_TRANSPORT, "IPV6_DONTFRAG not available in this build/platform"); #endif } else { #ifdef KNET_LINUX #ifdef IP_MTU_DISCOVER value = IP_PMTUDISC_PROBE; if (setsockopt(sock, SOL_IP, IP_MTU_DISCOVER, &value, sizeof(value)) <0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set PMTUDISC on %s socket: %s", type, strerror(savederrno)); goto exit_error; } log_debug(knet_h, KNET_SUB_TRANSPORT, "PMTUDISC enabled on socket: %i", sock); #else log_debug(knet_h, KNET_SUB_TRANSPORT, "PMTUDISC not available in this build/platform"); #endif #endif #ifdef KNET_BSD #ifdef IP_DONTFRAG value = 1; if (setsockopt(sock, IPPROTO_IP, IP_DONTFRAG, &value, sizeof(value)) <0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set DONTFRAG on %s socket: %s", type, strerror(savederrno)); goto exit_error; } log_debug(knet_h, KNET_SUB_TRANSPORT, "DONTFRAG enabled on socket: %i", sock); #else log_debug(knet_h, KNET_SUB_TRANSPORT, "DONTFRAG not available in this build/platform"); #endif #endif } value = 1; if (setsockopt(sock, SOL_SOCKET, SO_REUSEADDR, &value, sizeof(value)) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set %s reuseaddr: %s", type, strerror(savederrno)); goto exit_error; } exit_error: errno = savederrno; return err; } int _init_socketpair(knet_handle_t knet_h, int *sock) { int err = 0, savederrno = 0; int i; if (socketpair(AF_UNIX, SOCK_SEQPACKET, 0, sock) != 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_HANDLE, "Unable to initialize socketpair: %s", strerror(savederrno)); goto exit_fail; } for (i = 0; i < 2; i++) { if (_configure_common_socket(knet_h, sock[i], 0, "local socketpair") < 0) { savederrno = errno; err = -1; goto exit_fail; } } exit_fail: errno = savederrno; return err; } void _close_socketpair(knet_handle_t knet_h, int *sock) { int i; for (i = 0; i < 2; i++) { if (sock[i]) { close(sock[i]); sock[i] = 0; } } } /* * must be called with global read lock * * return -1 on error * return 0 if fd is invalid * return 1 if fd is valid */ int _is_valid_fd(knet_handle_t knet_h, int sockfd) { int ret = 0; if (sockfd < 0) { errno = EINVAL; return -1; } if (sockfd > KNET_MAX_FDS) { errno = EINVAL; return -1; } if (knet_h->knet_transport_fd_tracker[sockfd].transport >= KNET_MAX_TRANSPORTS) { ret = 0; } else { ret = 1; } return ret; } /* * must be called with global write lock */ int _set_fd_tracker(knet_handle_t knet_h, int sockfd, uint8_t transport, uint8_t data_type, void *data) { if (sockfd < 0) { errno = EINVAL; return -1; } if (sockfd > KNET_MAX_FDS) { errno = EINVAL; return -1; } knet_h->knet_transport_fd_tracker[sockfd].transport = transport; knet_h->knet_transport_fd_tracker[sockfd].data_type = data_type; knet_h->knet_transport_fd_tracker[sockfd].data = data; return 0; } diff --git a/libknet/transport_common.h b/libknet/transport_common.h index 61af9bb4..295d2b26 100644 --- a/libknet/transport_common.h +++ b/libknet/transport_common.h @@ -1,24 +1,24 @@ /* * Copyright (C) 2016-2018 Red Hat, Inc. All rights reserved. * * Authors: Fabio M. Di Nitto * * This software licensed under GPL-2.0+, LGPL-2.0+ */ #ifndef __KNET_TRANSPORT_COMMON_H__ #define __KNET_TRANSPORT_COMMON_H__ int _configure_common_socket(knet_handle_t knet_h, int sock, uint64_t flags, const char *type); int _configure_transport_socket(knet_handle_t knet_h, int sock, struct sockaddr_storage *address, uint64_t flags, const char *type); int _init_socketpair(knet_handle_t knet_h, int *sock); void _close_socketpair(knet_handle_t knet_h, int *sock); int _set_fd_tracker(knet_handle_t knet_h, int sockfd, uint8_t transport, uint8_t data_type, void *data); int _is_valid_fd(knet_handle_t knet_h, int sockfd); -int _sendmmsg(int sockfd, struct knet_mmsghdr *msgvec, unsigned int vlen, unsigned int flags); +int _sendmmsg(int sockfd, int connection_oriented, struct knet_mmsghdr *msgvec, unsigned int vlen, unsigned int flags); int _recvmmsg(int sockfd, struct knet_mmsghdr *msgvec, unsigned int vlen, unsigned int flags); #endif diff --git a/libknet/transports.c b/libknet/transports.c index b685212f..b8713700 100644 --- a/libknet/transports.c +++ b/libknet/transports.c @@ -1,272 +1,277 @@ /* * Copyright (C) 2017-2018 Red Hat, Inc. All rights reserved. * * Author: Fabio M. Di Nitto * * This software licensed under GPL-2.0+, LGPL-2.0+ */ #include "config.h" #include #include #include #include #include #include #include "libknet.h" #include "compat.h" #include "host.h" #include "link.h" #include "logging.h" #include "common.h" #include "transports.h" #include "transport_loopback.h" #include "transport_udp.h" #include "transport_sctp.h" #include "threads_common.h" -#define empty_module 0, 0, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL }, +#define empty_module 0, 0, 0, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL }, static knet_transport_ops_t transport_modules_cmd[KNET_MAX_TRANSPORTS] = { - { "LOOPBACK", KNET_TRANSPORT_LOOPBACK, 1, KNET_PMTUD_LOOPBACK_OVERHEAD, loopback_transport_init, loopback_transport_free, loopback_transport_link_set_config, loopback_transport_link_clear_config, loopback_transport_link_dyn_connect, loopback_transport_rx_sock_error, loopback_transport_tx_sock_error, loopback_transport_rx_is_data }, - { "UDP", KNET_TRANSPORT_UDP, 1, KNET_PMTUD_UDP_OVERHEAD, udp_transport_init, udp_transport_free, udp_transport_link_set_config, udp_transport_link_clear_config, udp_transport_link_dyn_connect, udp_transport_rx_sock_error, udp_transport_tx_sock_error, udp_transport_rx_is_data }, + { "LOOPBACK", KNET_TRANSPORT_LOOPBACK, 1, TRANSPORT_PROTO_NOT_CONNECTION_ORIENTED, KNET_PMTUD_LOOPBACK_OVERHEAD, loopback_transport_init, loopback_transport_free, loopback_transport_link_set_config, loopback_transport_link_clear_config, loopback_transport_link_dyn_connect, loopback_transport_rx_sock_error, loopback_transport_tx_sock_error, loopback_transport_rx_is_data }, + { "UDP", KNET_TRANSPORT_UDP, 1, TRANSPORT_PROTO_NOT_CONNECTION_ORIENTED, KNET_PMTUD_UDP_OVERHEAD, udp_transport_init, udp_transport_free, udp_transport_link_set_config, udp_transport_link_clear_config, udp_transport_link_dyn_connect, udp_transport_rx_sock_error, udp_transport_tx_sock_error, udp_transport_rx_is_data }, { "SCTP", KNET_TRANSPORT_SCTP, #ifdef HAVE_NETINET_SCTP_H - 1, KNET_PMTUD_SCTP_OVERHEAD, sctp_transport_init, sctp_transport_free, sctp_transport_link_set_config, sctp_transport_link_clear_config, sctp_transport_link_dyn_connect, sctp_transport_rx_sock_error, sctp_transport_tx_sock_error, sctp_transport_rx_is_data }, + 1, TRANSPORT_PROTO_IS_CONNECTION_ORIENTED, KNET_PMTUD_SCTP_OVERHEAD, sctp_transport_init, sctp_transport_free, sctp_transport_link_set_config, sctp_transport_link_clear_config, sctp_transport_link_dyn_connect, sctp_transport_rx_sock_error, sctp_transport_tx_sock_error, sctp_transport_rx_is_data }, #else empty_module #endif { NULL, KNET_MAX_TRANSPORTS, empty_module }; /* * transport wrappers */ int start_all_transports(knet_handle_t knet_h) { int idx = 0, savederrno = 0, err = 0; while (transport_modules_cmd[idx].transport_name != NULL) { if (transport_modules_cmd[idx].built_in) { if (transport_modules_cmd[idx].transport_init(knet_h) < 0) { savederrno = errno; log_err(knet_h, KNET_SUB_HANDLE, "Failed to allocate transport handle for %s: %s", transport_modules_cmd[idx].transport_name, strerror(savederrno)); err = -1; goto out; } } idx++; } out: errno = savederrno; return err; } void stop_all_transports(knet_handle_t knet_h) { int idx = 0; while (transport_modules_cmd[idx].transport_name != NULL) { if (transport_modules_cmd[idx].built_in) { transport_modules_cmd[idx].transport_free(knet_h); } idx++; } } int transport_link_set_config(knet_handle_t knet_h, struct knet_link *kn_link, uint8_t transport) { if (!transport_modules_cmd[transport].built_in) { errno = EINVAL; return -1; } kn_link->transport_connected = 0; kn_link->transport_type = transport; kn_link->proto_overhead = transport_modules_cmd[transport].transport_mtu_overhead; return transport_modules_cmd[transport].transport_link_set_config(knet_h, kn_link); } int transport_link_clear_config(knet_handle_t knet_h, struct knet_link *kn_link) { return transport_modules_cmd[kn_link->transport_type].transport_link_clear_config(knet_h, kn_link); } int transport_link_dyn_connect(knet_handle_t knet_h, int sockfd, struct knet_link *kn_link) { return transport_modules_cmd[kn_link->transport_type].transport_link_dyn_connect(knet_h, sockfd, kn_link); } int transport_rx_sock_error(knet_handle_t knet_h, uint8_t transport, int sockfd, int recv_err, int recv_errno) { return transport_modules_cmd[transport].transport_rx_sock_error(knet_h, sockfd, recv_err, recv_errno); } int transport_tx_sock_error(knet_handle_t knet_h, uint8_t transport, int sockfd, int recv_err, int recv_errno) { return transport_modules_cmd[transport].transport_tx_sock_error(knet_h, sockfd, recv_err, recv_errno); } int transport_rx_is_data(knet_handle_t knet_h, uint8_t transport, int sockfd, struct knet_mmsghdr *msg) { return transport_modules_cmd[transport].transport_rx_is_data(knet_h, sockfd, msg); } +int transport_get_connection_oriented(knet_handle_t knet_h, uint8_t transport) +{ + return transport_modules_cmd[transport].transport_is_connection_oriented; +} + /* * public api */ int knet_get_transport_list(struct knet_transport_info *transport_list, size_t *transport_list_entries) { int err = 0; int idx = 0; int outidx = 0; if (!transport_list_entries) { errno = EINVAL; return -1; } while (transport_modules_cmd[idx].transport_name != NULL) { if (transport_modules_cmd[idx].built_in) { if (transport_list) { transport_list[outidx].name = transport_modules_cmd[idx].transport_name; transport_list[outidx].id = transport_modules_cmd[idx].transport_id; } outidx++; } idx++; } *transport_list_entries = outidx; if (!err) errno = 0; return err; } const char *knet_get_transport_name_by_id(uint8_t transport) { int savederrno = 0; const char *name = NULL; if (transport == KNET_MAX_TRANSPORTS) { errno = EINVAL; return name; } if ((transport_modules_cmd[transport].transport_name) && (transport_modules_cmd[transport].built_in)) { name = transport_modules_cmd[transport].transport_name; } else { savederrno = ENOENT; } errno = name ? 0 : savederrno; return name; } uint8_t knet_get_transport_id_by_name(const char *name) { int savederrno = 0; uint8_t err = KNET_MAX_TRANSPORTS; int i, found; if (!name) { errno = EINVAL; return err; } i = 0; found = 0; while (transport_modules_cmd[i].transport_name != NULL) { if (transport_modules_cmd[i].built_in) { if (!strcmp(transport_modules_cmd[i].transport_name, name)) { err = transport_modules_cmd[i].transport_id; found = 1; break; } } i++; } if (!found) { savederrno = EINVAL; } errno = err == KNET_MAX_TRANSPORTS ? savederrno : 0; return err; } int knet_handle_set_transport_reconnect_interval(knet_handle_t knet_h, uint32_t msecs) { int savederrno = 0; if (!knet_h) { errno = EINVAL; return -1; } if (!msecs) { errno = EINVAL; return -1; } if (msecs < 1000) { log_warn(knet_h, KNET_SUB_HANDLE, "reconnect internval below 1 sec (%u msecs) might be too aggressive", msecs); } if (msecs > 60000) { log_warn(knet_h, KNET_SUB_HANDLE, "reconnect internval above 1 minute (%u msecs) could cause long delays in network convergiance", msecs); } savederrno = get_global_wrlock(knet_h); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get read lock: %s", strerror(savederrno)); errno = savederrno; return -1; } knet_h->reconnect_int = msecs; pthread_rwlock_unlock(&knet_h->global_rwlock); errno = 0; return 0; } int knet_handle_get_transport_reconnect_interval(knet_handle_t knet_h, uint32_t *msecs) { int savederrno = 0; if (!knet_h) { errno = EINVAL; return -1; } if (!msecs) { 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; } *msecs = knet_h->reconnect_int; pthread_rwlock_unlock(&knet_h->global_rwlock); errno = 0; return 0; } diff --git a/libknet/transports.h b/libknet/transports.h index 0419451d..ec1158b5 100644 --- a/libknet/transports.h +++ b/libknet/transports.h @@ -1,22 +1,23 @@ /* * Copyright (C) 2016-2018 Red Hat, Inc. All rights reserved. * * Authors: Fabio M. Di Nitto * * This software licensed under GPL-2.0+, LGPL-2.0+ */ #ifndef __KNET_TRANSPORTS_H__ #define __KNET_TRANSPORTS_H__ int start_all_transports(knet_handle_t knet_h); void stop_all_transports(knet_handle_t knet_h); int transport_link_set_config(knet_handle_t knet_h, struct knet_link *kn_link, uint8_t transport); int transport_link_clear_config(knet_handle_t knet_h, struct knet_link *kn_link); int transport_link_dyn_connect(knet_handle_t knet_h, int sockfd, struct knet_link *kn_link); int transport_rx_sock_error(knet_handle_t knet_h, uint8_t transport, int sockfd, int recv_err, int recv_errno); int transport_tx_sock_error(knet_handle_t knet_h, uint8_t transport, int sockfd, int recv_err, int recv_errno); int transport_rx_is_data(knet_handle_t knet_h, uint8_t transport, int sockfd, struct knet_mmsghdr *msg); +int transport_get_connection_oriented(knet_handle_t knet_h, uint8_t transport); #endif