diff --git a/libknet/compress.c b/libknet/compress.c index 48c144fc..d4fbd1de 100644 --- a/libknet/compress.c +++ b/libknet/compress.c @@ -1,179 +1,194 @@ /* * Copyright (C) 2010-2017 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 "internals.h" #include "compress.h" #include "logging.h" #include "compress_zlib.h" #include "compress_lz4.h" #include "compress_lzo2.h" #include "compress_lzma.h" #include "compress_bzip2.h" /* * internal module switch data */ /* - * DO NOT CHANGE ORDER HERE OR ONWIRE COMPATIBILITY + * DO NOT CHANGE MODEL_ID HERE OR ONWIRE COMPATIBILITY * WILL BREAK! * * always add before the last NULL/NULL/NULL. */ compress_model_t compress_modules_cmds[] = { - { "none", NULL, NULL, NULL, NULL, NULL }, - { "zlib", NULL, NULL, zlib_val_level, zlib_compress, zlib_decompress }, - { "lz4", NULL, NULL, lz4_val_level, lz4_compress, lz4_decompress }, - { "lz4hc", NULL, NULL, lz4hc_val_level, lz4hc_compress, lz4_decompress }, - { "lzo2", lzo2_init, lzo2_fini, lzo2_val_level, lzo2_compress, lzo2_decompress }, - { "lzma", NULL, NULL, lzma_val_level, lzma_compress, lzma_decompress }, - { "bzip2", NULL, NULL, bzip2_val_level, bzip2_compress, bzip2_decompress }, - { NULL, NULL, NULL, NULL, NULL, NULL }, + { 0, 1, "none", NULL, NULL, NULL, NULL, NULL }, + { 1, 1, "zlib", NULL, NULL, zlib_val_level, zlib_compress, zlib_decompress }, + { 2, 1, "lz4", NULL, NULL, lz4_val_level, lz4_compress, lz4_decompress }, + { 3, 1, "lz4hc", NULL, NULL, lz4hc_val_level, lz4hc_compress, lz4_decompress }, + { 4, 1, "lzo2", lzo2_init, lzo2_fini, lzo2_val_level, lzo2_compress, lzo2_decompress }, + { 5, 1, "lzma", NULL, NULL, lzma_val_level, lzma_compress, lzma_decompress }, + { 6, 1, "bzip2", NULL, NULL, bzip2_val_level, bzip2_compress, bzip2_decompress }, + { 255, 0, NULL, NULL, NULL, NULL, NULL, NULL }, }; -/* - * used exclusively by the test suite (see api_knet_send_compress) - */ -const char *get_model_by_idx(int idx) -{ - return compress_modules_cmds[idx].model_name; -} - static int get_model(const char *model) { int idx = 0; while (compress_modules_cmds[idx].model_name != NULL) { - if (!strcmp(compress_modules_cmds[idx].model_name, model)) - return idx; + if ((!strcmp(compress_modules_cmds[idx].model_name, model)) && + (compress_modules_cmds[idx].built_in == 1)) { + return compress_modules_cmds[idx].model_id; + } idx++; } return -1; } static int get_max_model(void) { int idx = 0; while (compress_modules_cmds[idx].model_name != NULL) { idx++; } return idx - 1; } +static int is_valid_model(int compress_model) +{ + int idx = 0; + + while (compress_modules_cmds[idx].model_name != NULL) { + if ((compress_model == compress_modules_cmds[idx].model_id) && + (compress_modules_cmds[idx].built_in == 1)) { + return 0; + } + idx++; + } + return -1; +} + static int val_level( knet_handle_t knet_h, int compress_model, int compress_level) { return compress_modules_cmds[compress_model].val_level(knet_h, compress_level); } int compress_init( knet_handle_t knet_h, struct knet_handle_compress_cfg *knet_handle_compress_cfg) { int cmp_model, idx = 0; - knet_h->compress_max_model = get_max_model(); - if (knet_h->compress_max_model > KNET_MAX_COMPRESS_METHODS) { + if (get_max_model() > KNET_MAX_COMPRESS_METHODS) { log_err(knet_h, KNET_SUB_COMPRESS, "Too many compress methods supported by compress.c. Please complain to knet developers to fix internals.h KNET_MAX_COMPRESS_METHODS define!"); errno = EINVAL; return -1; } if (!knet_handle_compress_cfg) { - while (compress_modules_cmds[idx].model_name != NULL) { + while ((compress_modules_cmds[idx].model_name != NULL) && + (compress_modules_cmds[idx].built_in == 1)) { if (compress_modules_cmds[idx].init != NULL) { if (compress_modules_cmds[idx].init(knet_h, idx) < 0) { log_err(knet_h, KNET_SUB_COMPRESS, "Failed to initialize %s library", compress_modules_cmds[idx].model_name); errno = EINVAL; return -1; } } idx++; } return 0; } log_debug(knet_h, KNET_SUB_COMPRESS, "Initizializing compress module [%s/%d/%u]", knet_handle_compress_cfg->compress_model, knet_handle_compress_cfg->compress_level, knet_handle_compress_cfg->compress_threshold); cmp_model = get_model(knet_handle_compress_cfg->compress_model); if (cmp_model < 0) { log_err(knet_h, KNET_SUB_COMPRESS, "compress model %s not supported", knet_handle_compress_cfg->compress_model); errno = EINVAL; return -1; } if (cmp_model > 0) { if (val_level(knet_h, cmp_model, knet_handle_compress_cfg->compress_level) < 0) { log_err(knet_h, KNET_SUB_COMPRESS, "compress level %d not supported for model %s", knet_handle_compress_cfg->compress_level, knet_handle_compress_cfg->compress_model); errno = EINVAL; return -1; } if (knet_handle_compress_cfg->compress_threshold > KNET_MAX_PACKET_SIZE) { log_err(knet_h, KNET_SUB_COMPRESS, "compress threshold cannot be higher than KNET_MAX_PACKET_SIZE (%d).", KNET_MAX_PACKET_SIZE); errno = EINVAL; return -1; } if (knet_handle_compress_cfg->compress_threshold == 0) { knet_h->compress_threshold = KNET_COMPRESS_THRESHOLD; log_debug(knet_h, KNET_SUB_COMPRESS, "resetting compression threshold to default (%d)", KNET_COMPRESS_THRESHOLD); } else { knet_h->compress_threshold = knet_handle_compress_cfg->compress_threshold; } } knet_h->compress_model = cmp_model; knet_h->compress_level = knet_handle_compress_cfg->compress_level; return 0; } void compress_fini( knet_handle_t knet_h) { int idx = 0; - while ((compress_modules_cmds[idx].model_name != NULL) && (idx < KNET_MAX_COMPRESS_METHODS)) { + while ((compress_modules_cmds[idx].model_name != NULL) && + (compress_modules_cmds[idx].built_in == 1) && + (idx < KNET_MAX_COMPRESS_METHODS)) { if (compress_modules_cmds[idx].fini != NULL) { compress_modules_cmds[idx].fini(knet_h, idx); } idx++; } return; } int compress( knet_handle_t knet_h, const unsigned char *buf_in, const ssize_t buf_in_len, unsigned char *buf_out, ssize_t *buf_out_len) { return compress_modules_cmds[knet_h->compress_model].compress(knet_h, buf_in, buf_in_len, buf_out, buf_out_len); } int decompress( knet_handle_t knet_h, int compress_model, const unsigned char *buf_in, const ssize_t buf_in_len, unsigned char *buf_out, ssize_t *buf_out_len) { + if (is_valid_model(compress_model) < 0) { + log_err(knet_h, KNET_SUB_COMPRESS, "This build of libknet does not support %s compression. Please contact your distribution vendor or fix the build", compress_modules_cmds[compress_model].model_name); + errno = EINVAL; + return -1; + } return compress_modules_cmds[compress_model].decompress(knet_h, buf_in, buf_in_len, buf_out, buf_out_len); } diff --git a/libknet/compress.h b/libknet/compress.h index 4fa64bad..ffc6d8d8 100644 --- a/libknet/compress.h +++ b/libknet/compress.h @@ -1,59 +1,56 @@ /* * Copyright (C) 2010-2017 Red Hat, Inc. All rights reserved. * * Author: Fabio M. Di Nitto * * This software licensed under GPL-2.0+, LGPL-2.0+ */ #ifndef __KNET_COMPRESS_H__ #define __KNET_COMPRESS_H__ #include "internals.h" typedef struct { + uint8_t model_id; + uint8_t built_in; const char *model_name; int (*init) (knet_handle_t knet_h, int method_idx); void (*fini) (knet_handle_t knet_h, int method_idx); int (*val_level)(knet_handle_t knet_h, int compress_level); int (*compress) (knet_handle_t knet_h, const unsigned char *buf_in, const ssize_t buf_in_len, unsigned char *buf_out, ssize_t *buf_out_len); int (*decompress)(knet_handle_t knet_h, const unsigned char *buf_in, const ssize_t buf_in_len, unsigned char *buf_out, ssize_t *buf_out_len); } compress_model_t; int compress_init( knet_handle_t knet_h, struct knet_handle_compress_cfg *knet_handle_compress_cfg); void compress_fini( knet_handle_t knet_h); int compress( knet_handle_t knet_h, const unsigned char *buf_in, const ssize_t buf_in_len, unsigned char *buf_out, ssize_t *buf_out_len); int decompress( knet_handle_t knet_h, int compress_model, const unsigned char *buf_in, const ssize_t buf_in_len, unsigned char *buf_out, ssize_t *buf_out_len); -/* - * used exclusively by the test suite (see api_knet_send_compress) - */ -const char *get_model_by_idx(int idx); - #endif diff --git a/libknet/internals.h b/libknet/internals.h index 0f77dd68..0cae38ef 100644 --- a/libknet/internals.h +++ b/libknet/internals.h @@ -1,485 +1,484 @@ /* * Copyright (C) 2010-2015 Red Hat, Inc. All rights reserved. * * Authors: Fabio M. Di Nitto * Federico Simoncelli * * This software licensed under GPL-2.0+, LGPL-2.0+ */ #ifndef __KNET_INTERNALS_H__ #define __KNET_INTERNALS_H__ /* * NOTE: you shouldn't need to include this header normally */ #include "libknet.h" #include "onwire.h" #include "compat.h" #define KNET_DATABUFSIZE KNET_MAX_PACKET_SIZE + KNET_HEADER_ALL_SIZE #define KNET_DATABUFSIZE_CRYPT_PAD 1024 #define KNET_DATABUFSIZE_CRYPT KNET_DATABUFSIZE + KNET_DATABUFSIZE_CRYPT_PAD #define KNET_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 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 32 struct knet_handle { knet_node_id_t host_id; unsigned int enabled:1; struct knet_sock sockfd[KNET_DATAFD_MAX]; int logfd; uint8_t log_levels[KNET_MAX_SUBSYSTEMS]; int hostsockfd[2]; int dstsockfd[2]; int send_to_links_epollfd; int recv_from_links_epollfd; int dst_link_handler_epollfd; unsigned int pmtud_interval; unsigned int data_mtu; /* contains the max data size that we can send onwire * without frags */ struct knet_host *host_head; struct knet_host *host_index[KNET_MAX_HOST]; knet_transport_t transports[KNET_MAX_TRANSPORTS+1]; struct knet_transport_ops *transport_ops[KNET_MAX_TRANSPORTS+1]; struct knet_fd_trackers knet_transport_fd_tracker[KNET_MAX_FDS]; /* track status for each fd handled by transports */ 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; pthread_t send_to_links_thread; pthread_t recv_from_links_thread; pthread_t heartbt_thread; pthread_t dst_link_handler_thread; pthread_t pmtud_link_handler_thread; int lock_init_done; pthread_rwlock_t global_rwlock; /* global config lock */ pthread_mutex_t pmtud_mutex; /* pmtud mutex to handle conditional send/recv + timeout */ pthread_cond_t pmtud_cond; /* conditional for above */ pthread_mutex_t tx_mutex; /* used to protect knet_send_sync and TX thread */ pthread_mutex_t hb_mutex; /* used to protect heartbeat thread and seq_num broadcasting */ struct crypto_instance *crypto_instance; uint16_t sec_header_size; uint16_t sec_block_size; uint16_t sec_hash_size; uint16_t sec_salt_size; unsigned char *send_to_links_buf_crypt[PCKT_FRAG_MAX]; unsigned char *recv_from_links_buf_crypt; unsigned char *recv_from_links_buf_decrypt; unsigned char *pingbuf_crypt; unsigned char *pmtudbuf_crypt; int compress_model; - int compress_max_model; int compress_level; uint32_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; }; /* * NOTE: every single operation must be implementend * for every protocol. */ typedef struct knet_transport_ops { /* * transport generic information */ const char *transport_name; const uint8_t transport_id; uint32_t transport_mtu_overhead; /* * transport init must allocate the new transport * and perform all internal initializations * (threads, lists, etc). */ int (*transport_init)(knet_handle_t knet_h); /* * transport free must releases _all_ resources * allocated by tranport_init */ int (*transport_free)(knet_handle_t knet_h); /* * link operations should take care of all the * sockets and epoll management for a given link/transport set * transport_link_disable should return err = -1 and errno = EBUSY * if listener is still in use, and any other errno in case * the link cannot be disabled. * * set_config/clear_config are invoked in global write lock context */ int (*transport_link_set_config)(knet_handle_t knet_h, struct knet_link *link); int (*transport_link_clear_config)(knet_handle_t knet_h, struct knet_link *link); /* * transport callback for incoming dynamic connections * this is called in global read lock context */ int (*transport_link_dyn_connect)(knet_handle_t knet_h, int sockfd, struct knet_link *link); /* * per transport error handling of recvmmsg * (see _handle_recv_from_links comments for details) */ /* * transport_rx_sock_error is invoked when recvmmsg returns <= 0 * * transport_rx_sock_error is invoked with both global_rdlock */ int (*transport_rx_sock_error)(knet_handle_t knet_h, int sockfd, int recv_err, int recv_errno); /* * transport_tx_sock_error is invoked with global_rwlock and * it's invoked when sendto or sendmmsg returns =< 0 * * it should return: * -1 on internal error * 0 ignore error and continue * 1 retry * any sleep or wait action should happen inside the transport code */ int (*transport_tx_sock_error)(knet_handle_t knet_h, int sockfd, int recv_err, int recv_errno); /* * this function is called on _every_ received packet * to verify if the packet is data or internal protocol error handling * * it should return: * -1 on error * 0 packet is not data and we should continue the packet process loop * 1 packet is not data and we should STOP the packet process loop * 2 packet is data and should be parsed as such * * transport_rx_is_data is invoked with both global_rwlock * and fd_tracker read lock (from RX thread) */ int (*transport_rx_is_data)(knet_handle_t knet_h, int sockfd, struct knet_mmsghdr *msg); } knet_transport_ops_t; socklen_t sockaddr_len(const struct sockaddr_storage *ss); /** * This is a kernel style list implementation. * * @author Steven Dake */ struct knet_list_head { struct knet_list_head *next; struct knet_list_head *prev; }; /** * @def KNET_LIST_DECLARE() * Declare and initialize a list head. */ #define KNET_LIST_DECLARE(name) \ struct knet_list_head name = { &(name), &(name) } #define KNET_INIT_LIST_HEAD(ptr) do { \ (ptr)->next = (ptr); (ptr)->prev = (ptr); \ } while (0) /** * Initialize the list entry. * * Points next and prev pointers to head. * @param head pointer to the list head */ static inline void knet_list_init(struct knet_list_head *head) { head->next = head; head->prev = head; } /** * Add this element to the list. * * @param element the new element to insert. * @param head pointer to the list head */ static inline void knet_list_add(struct knet_list_head *element, struct knet_list_head *head) { head->next->prev = element; element->next = head->next; element->prev = head; head->next = element; } /** * Add to the list (but at the end of the list). * * @param element pointer to the element to add * @param head pointer to the list head * @see knet_list_add() */ static inline void knet_list_add_tail(struct knet_list_head *element, struct knet_list_head *head) { head->prev->next = element; element->next = head; element->prev = head->prev; head->prev = element; } /** * Delete an entry from the list. * * @param _remove the list item to remove */ static inline void knet_list_del(struct knet_list_head *_remove) { _remove->next->prev = _remove->prev; _remove->prev->next = _remove->next; } /** * Replace old entry by new one * @param old: the element to be replaced * @param new: the new element to insert */ static inline void knet_list_replace(struct knet_list_head *old, struct knet_list_head *new) { new->next = old->next; new->next->prev = new; new->prev = old->prev; new->prev->next = new; } /** * Tests whether list is the last entry in list head * @param list: the entry to test * @param head: the head of the list * @return boolean true/false */ static inline int knet_list_is_last(const struct knet_list_head *list, const struct knet_list_head *head) { return list->next == head; } /** * A quick test to see if the list is empty (pointing to it's self). * @param head pointer to the list head * @return boolean true/false */ static inline int32_t knet_list_empty(const struct knet_list_head *head) { return head->next == head; } /** * Get the struct for this entry * @param ptr: the &struct list_head pointer. * @param type: the type of the struct this is embedded in. * @param member: the name of the list_struct within the struct. */ #define knet_list_entry(ptr,type,member)\ ((type *)((char *)(ptr)-(char*)(&((type *)0)->member))) /** * Get the first element from a list * @param ptr: the &struct list_head pointer. * @param type: the type of the struct this is embedded in. * @param member: the name of the list_struct within the struct. */ #define knet_list_first_entry(ptr, type, member) \ knet_list_entry((ptr)->next, type, member) /** * Iterate over a list * @param pos: the &struct list_head to use as a loop counter. * @param head: the head for your list. */ #define knet_list_for_each(pos, head) \ for (pos = (head)->next; pos != (head); pos = pos->next) /** * Iterate over a list backwards * @param pos: the &struct list_head to use as a loop counter. * @param head: the head for your list. */ #define knet_list_for_each_reverse(pos, head) \ for (pos = (head)->prev; pos != (head); pos = pos->prev) /** * Iterate over a list safe against removal of list entry * @param pos: the &struct list_head to use as a loop counter. * @param n: another &struct list_head to use as temporary storage * @param head: the head for your list. */ #define knet_list_for_each_safe(pos, n, head) \ for (pos = (head)->next, n = pos->next; pos != (head); \ pos = n, n = pos->next) /** * Iterate over list of given type * @param pos: the type * to use as a loop counter. * @param head: the head for your list. * @param member: the name of the list_struct within the struct. */ #define knet_list_for_each_entry(pos, head, member) \ for (pos = knet_list_entry((head)->next, typeof(*pos), member); \ &pos->member != (head); \ pos = knet_list_entry(pos->member.next, typeof(*pos), member)) #endif diff --git a/libknet/tests/api_knet_send_compress.c b/libknet/tests/api_knet_send_compress.c index 12400dd5..0f3d6ffe 100644 --- a/libknet/tests/api_knet_send_compress.c +++ b/libknet/tests/api_knet_send_compress.c @@ -1,255 +1,254 @@ /* * Copyright (C) 2016 Red Hat, Inc. All rights reserved. * * Authors: Fabio M. Di Nitto * * This software licensed under GPL-2.0+, LGPL-2.0+ */ #include "config.h" #include #include #include #include #include #include "libknet.h" #include "compress.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(const char *model) { 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; struct knet_handle_compress_cfg knet_handle_compress_cfg; memset(&lo, 0, sizeof(struct sockaddr_storage)); if (knet_strtoaddr("127.0.0.1", "50000", &lo, sizeof(struct sockaddr_storage)) < 0) { printf("Unable to convert loopback to sockaddr: %s\n", strerror(errno)); exit(FAIL); } memset(send_buff, 0, sizeof(send_buff)); setup_logpipes(logfds); knet_h = knet_handle_new(1, logfds[1], KNET_LOG_DEBUG); if (!knet_h) { printf("knet_handle_new failed: %s\n", strerror(errno)); flush_logs(logfds[0], stdout); close_logpipes(logfds); exit(FAIL); } flush_logs(logfds[0], stdout); printf("Test knet_send with %s and valid data\n", model); memset(&knet_handle_compress_cfg, 0, sizeof(struct knet_handle_compress_cfg)); strncpy(knet_handle_compress_cfg.compress_model, model, sizeof(knet_handle_compress_cfg.compress_model) - 1); knet_handle_compress_cfg.compress_level = 4; knet_handle_compress_cfg.compress_threshold = 0; if (knet_handle_compress(knet_h, &knet_handle_compress_cfg) < 0) { printf("knet_handle_compress did not accept zlib compress mode with compress level 1 cfg\n"); knet_handle_free(knet_h); flush_logs(logfds[0], stdout); close_logpipes(logfds); 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); 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) { 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); } while(knet_h->host_index[1]->status.reachable != 1) { printf("waiting host to be reachable\n"); sleep(1); } 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); sleep(1); 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); } #if 0 /* 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: %lu (%lu bytes), rx_packets: %lu (%lu 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); #endif 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[]) { - int idx = 0; - const char *model = NULL; - need_root(); - while ((model = get_model_by_idx(idx)) != NULL) { - test(model); - idx++; - } + test("zlib"); + test("lz4"); + test("lz4hc"); + test("lzo2"); + test("lzma"); + test("bzip2"); return PASS; } diff --git a/libknet/threads_rx.c b/libknet/threads_rx.c index 35300245..f465ca32 100644 --- a/libknet/threads_rx.c +++ b/libknet/threads_rx.c @@ -1,784 +1,779 @@ /* * Copyright (C) 2010-2017 Red Hat, Inc. All rights reserved. * * Authors: Fabio M. Di Nitto * Federico Simoncelli * * This software licensed under GPL-2.0+, LGPL-2.0+ */ #include "config.h" #include #include #include #include #include #include "compat.h" #include "compress.h" #include "crypto.h" #include "host.h" #include "link.h" #include "logging.h" #include "transports.h" #include "threads_common.h" #include "threads_heartbeat.h" #include "threads_rx.h" #include "netutils.h" /* * RECV */ /* * return 1 if a > b * return -1 if b > a * return 0 if they are equal */ static inline int timecmp(struct timespec a, struct timespec b) { if (a.tv_sec != b.tv_sec) { if (a.tv_sec > b.tv_sec) { return 1; } else { return -1; } } else { if (a.tv_nsec > b.tv_nsec) { return 1; } else if (a.tv_nsec < b.tv_nsec) { return -1; } else { return 0; } } } /* * this functions needs to return an index (0 to 7) * to a knet_host_defrag_buf. (-1 on errors) */ static int find_pckt_defrag_buf(knet_handle_t knet_h, struct knet_header *inbuf) { struct knet_host *src_host = knet_h->host_index[inbuf->kh_node]; int i, oldest; /* * check if there is a buffer already in use handling the same seq_num */ for (i = 0; i < KNET_MAX_LINK; i++) { if (src_host->defrag_buf[i].in_use) { if (src_host->defrag_buf[i].pckt_seq == inbuf->khp_data_seq_num) { return i; } } } /* * If there is no buffer that's handling the current seq_num * either it's new or it's been reclaimed already. * check if it's been reclaimed/seen before using the defrag circular * buffer. If the pckt has been seen before, the buffer expired (ETIME) * and there is no point to try to defrag it again. */ if (!_seq_num_lookup(src_host, inbuf->khp_data_seq_num, 1, 0)) { errno = ETIME; return -1; } /* * register the pckt as seen */ _seq_num_set(src_host, inbuf->khp_data_seq_num, 1); /* * see if there is a free buffer */ for (i = 0; i < KNET_MAX_LINK; i++) { if (!src_host->defrag_buf[i].in_use) { return i; } } /* * at this point, there are no free buffers, the pckt is new * and we need to reclaim a buffer, and we will take the one * with the oldest timestamp. It's as good as any. */ oldest = 0; for (i = 0; i < KNET_MAX_LINK; i++) { if (timecmp(src_host->defrag_buf[i].last_update, src_host->defrag_buf[oldest].last_update) < 0) { oldest = i; } } src_host->defrag_buf[oldest].in_use = 0; return oldest; } static int pckt_defrag(knet_handle_t knet_h, struct knet_header *inbuf, ssize_t *len) { struct knet_host_defrag_buf *defrag_buf; int defrag_buf_idx; defrag_buf_idx = find_pckt_defrag_buf(knet_h, inbuf); if (defrag_buf_idx < 0) { if (errno == ETIME) { log_debug(knet_h, KNET_SUB_RX, "Defrag buffer expired"); } return 1; } defrag_buf = &knet_h->host_index[inbuf->kh_node]->defrag_buf[defrag_buf_idx]; /* * if the buf is not is use, then make sure it's clean */ if (!defrag_buf->in_use) { memset(defrag_buf, 0, sizeof(struct knet_host_defrag_buf)); defrag_buf->in_use = 1; defrag_buf->pckt_seq = inbuf->khp_data_seq_num; } /* * update timestamp on the buffer */ clock_gettime(CLOCK_MONOTONIC, &defrag_buf->last_update); /* * check if we already received this fragment */ if (defrag_buf->frag_map[inbuf->khp_data_frag_seq]) { /* * if we have received this fragment and we didn't clear the buffer * it means that we don't have all fragments yet */ return 1; } /* * we need to handle the last packet with gloves due to its different size */ if (inbuf->khp_data_frag_seq == inbuf->khp_data_frag_num) { defrag_buf->last_frag_size = *len; /* * in the event when the last packet arrives first, * we still don't know the offset vs the other fragments (based on MTU), * so we store the fragment at the end of the buffer where it's safe * and take a copy of the len so that we can restore its offset later. * remember we can't use the local MTU for this calculation because pMTU * can be asymettric between the same hosts. */ if (!defrag_buf->frag_size) { defrag_buf->last_first = 1; memmove(defrag_buf->buf + (KNET_MAX_PACKET_SIZE - *len), inbuf->khp_data_userdata, *len); } } else { defrag_buf->frag_size = *len; } memmove(defrag_buf->buf + ((inbuf->khp_data_frag_seq - 1) * defrag_buf->frag_size), inbuf->khp_data_userdata, *len); defrag_buf->frag_recv++; defrag_buf->frag_map[inbuf->khp_data_frag_seq] = 1; /* * check if we received all the fragments */ if (defrag_buf->frag_recv == inbuf->khp_data_frag_num) { /* * special case the last pckt */ if (defrag_buf->last_first) { memmove(defrag_buf->buf + ((inbuf->khp_data_frag_num - 1) * defrag_buf->frag_size), defrag_buf->buf + (KNET_MAX_PACKET_SIZE - defrag_buf->last_frag_size), defrag_buf->last_frag_size); } /* * recalculate packet lenght */ *len = ((inbuf->khp_data_frag_num - 1) * defrag_buf->frag_size) + defrag_buf->last_frag_size; /* * copy the pckt back in the user data */ memmove(inbuf->khp_data_userdata, defrag_buf->buf, *len); /* * free this buffer */ defrag_buf->in_use = 0; return 0; } return 1; } static void _parse_recv_from_links(knet_handle_t knet_h, int sockfd, const struct knet_mmsghdr *msg) { int err = 0, savederrno = 0; ssize_t outlen; struct knet_host *src_host; struct knet_link *src_link; unsigned long long latency_last; knet_node_id_t dst_host_ids[KNET_MAX_HOST]; size_t dst_host_ids_entries = 0; int bcast = 1; struct timespec recvtime; struct knet_header *inbuf = msg->msg_hdr.msg_iov->iov_base; unsigned char *outbuf = (unsigned char *)msg->msg_hdr.msg_iov->iov_base; ssize_t len = msg->msg_len; struct knet_hostinfo *knet_hostinfo; struct iovec iov_out[1]; int8_t channel; struct sockaddr_storage pckt_src; seq_num_t recv_seq_num; int wipe_bufs = 0; if (knet_h->crypto_instance) { if (crypto_authenticate_and_decrypt(knet_h, (unsigned char *)inbuf, len, knet_h->recv_from_links_buf_decrypt, &outlen) < 0) { log_debug(knet_h, KNET_SUB_RX, "Unable to decrypt/auth packet"); return; } len = outlen; inbuf = (struct knet_header *)knet_h->recv_from_links_buf_decrypt; } if (len < (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 */ knet_h->transport_ops[src_link->transport_type]->transport_link_dyn_connect(knet_h, sockfd, src_link); } } switch (inbuf->kh_type) { case KNET_HEADER_TYPE_HOST_INFO: case KNET_HEADER_TYPE_DATA: /* * TODO: should we accept data even if we can't reply to the other node? * how would that work with SCTP and guaranteed delivery? */ if (!src_host->status.reachable) { log_debug(knet_h, KNET_SUB_RX, "Source host %u not reachable yet", src_host->host_id); //return; } inbuf->khp_data_seq_num = ntohs(inbuf->khp_data_seq_num); channel = inbuf->khp_data_channel; src_host->got_data = 1; if (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; - if (inbuf->khp_data_compress > knet_h->compress_max_model) { - log_err(knet_h, KNET_SUB_COMPRESS, "Received packet with unsupported compression method. Dropping"); - return; - } - 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) { 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; 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) { 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, len, knet_h->recv_from_links_buf_crypt, &outlen) < 0) { log_debug(knet_h, KNET_SUB_RX, "Unable to encrypt pong packet"); break; } outbuf = knet_h->recv_from_links_buf_crypt; } retry_pong: len = sendto(src_link->outsock, outbuf, outlen, MSG_DONTWAIT | MSG_NOSIGNAL, (struct sockaddr *) &src_link->dst_addr, sizeof(struct sockaddr_storage)); savederrno = errno; if (len != outlen) { err = knet_h->transport_ops[src_link->transport_type]->transport_tx_sock_error(knet_h, src_link->outsock, len, savederrno); switch(err) { case -1: /* unrecoverable error */ log_debug(knet_h, KNET_SUB_RX, "Unable to send pong reply (sock: %d) packet (sendto): %d %s. recorded src ip: %s src port: %s dst ip: %s dst port: %s", src_link->outsock, errno, strerror(errno), src_link->status.src_ipaddr, src_link->status.src_port, src_link->status.dst_ipaddr, src_link->status.dst_port); 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) { 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, len, knet_h->recv_from_links_buf_crypt, &outlen) < 0) { log_debug(knet_h, KNET_SUB_RX, "Unable to encrypt PMTUd reply packet"); break; } outbuf = knet_h->recv_from_links_buf_crypt; } retry_pmtud: len = sendto(src_link->outsock, outbuf, outlen, MSG_DONTWAIT | MSG_NOSIGNAL, (struct sockaddr *) &src_link->dst_addr, sizeof(struct sockaddr_storage)); if (len != outlen) { err = knet_h->transport_ops[src_link->transport_type]->transport_tx_sock_error(knet_h, src_link->outsock, len, savederrno); switch(err) { case -1: /* unrecoverable error */ log_debug(knet_h, KNET_SUB_RX, "Unable to send PMTUd reply (sock: %d) packet (sendto): %d %s. recorded src ip: %s src port: %s dst ip: %s dst port: %s", src_link->outsock, errno, strerror(errno), src_link->status.src_ipaddr, src_link->status.src_port, src_link->status.dst_ipaddr, src_link->status.dst_port); 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; } } 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) { knet_h->transport_ops[transport]->transport_rx_sock_error(knet_h, sockfd, msg_recv, savederrno); goto exit_unlock; } for (i = 0; i < msg_recv; i++) { err = knet_h->transport_ops[transport]->transport_rx_is_data(knet_h, sockfd, &msg[i]); /* * TODO: make this section silent once we are confident * all protocols packet handlers are good */ switch(err) { case -1: /* on error */ log_debug(knet_h, KNET_SUB_RX, "Transport reported error parsing packet"); goto exit_unlock; break; case 0: /* packet is not data and we should continue the packet process loop */ log_debug(knet_h, KNET_SUB_RX, "Transport reported no data, continue"); break; case 1: /* packet is not data and we should STOP the packet process loop */ log_debug(knet_h, KNET_SUB_RX, "Transport reported no data, stop"); goto exit_unlock; break; case 2: /* packet is data and should be parsed as such */ _parse_recv_from_links(knet_h, sockfd, &msg[i]); break; } } exit_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); } void *_handle_recv_from_links_thread(void *data) { int i, nev; knet_handle_t knet_h = (knet_handle_t) data; struct epoll_event events[KNET_EPOLL_MAX_EVENTS]; struct sockaddr_storage address[PCKT_RX_BUFS]; struct knet_mmsghdr msg[PCKT_RX_BUFS]; struct iovec iov_in[PCKT_RX_BUFS]; 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, -1); for (i = 0; i < nev; i++) { _handle_recv_from_links(knet_h, events[i].data.fd, msg); } } return NULL; }