diff --git a/libknet/host.c b/libknet/host.c index f02ef025..54061fde 100644 --- a/libknet/host.c +++ b/libknet/host.c @@ -1,829 +1,844 @@ /* * Copyright (C) 2010-2021 Red Hat, Inc. All rights reserved. * * Authors: Fabio M. Di Nitto * Federico Simoncelli * * This software licensed under LGPL-2.0+ */ #include "config.h" #include #include #include #include #include #include "host.h" #include "internals.h" #include "logging.h" #include "threads_common.h" static void _host_list_update(knet_handle_t knet_h) { struct knet_host *host; knet_h->host_ids_entries = 0; for (host = knet_h->host_head; host != NULL; host = host->next) { knet_h->host_ids[knet_h->host_ids_entries] = host->host_id; knet_h->host_ids_entries++; } } int knet_host_add(knet_handle_t knet_h, knet_node_id_t host_id) { int savederrno = 0, err = 0; struct knet_host *host = NULL; uint8_t link_idx; if (!_is_valid_handle(knet_h)) { return -1; } savederrno = get_global_wrlock(knet_h); if (savederrno) { log_err(knet_h, KNET_SUB_HOST, "Unable to get write lock: %s", strerror(savederrno)); errno = savederrno; return -1; } if (knet_h->host_index[host_id]) { err = -1; savederrno = EEXIST; log_err(knet_h, KNET_SUB_HOST, "Unable to add host %u: %s", host_id, strerror(savederrno)); goto exit_unlock; } host = malloc(sizeof(struct knet_host)); if (!host) { err = -1; savederrno = errno; log_err(knet_h, KNET_SUB_HOST, "Unable to allocate memory for host %u: %s", host_id, strerror(savederrno)); goto exit_unlock; } memset(host, 0, sizeof(struct knet_host)); host->defrag_bufs = malloc(knet_h->defrag_bufs_min * sizeof(struct knet_host_defrag_buf)); if (!host->defrag_bufs) { err = -1; savederrno = errno; log_err(knet_h, KNET_SUB_HOST, "Unable to allocate memory for host %u defrag buffers: %s", host_id, strerror(savederrno)); goto exit_unlock; } host->allocated_defrag_bufs = knet_h->defrag_bufs_min; memset(host->defrag_bufs, 0, host->allocated_defrag_bufs * sizeof(struct knet_host_defrag_buf)); log_debug(knet_h, KNET_SUB_HOST, "Allocated %u defrag buffers for host %u", host->allocated_defrag_bufs, host_id); /* * set host_id */ host->host_id = host_id; /* * fill up our own data */ if (knet_h->host_id == host->host_id) { host->onwire_ver = knet_h->onwire_ver; host->onwire_max_ver = knet_h->onwire_max_ver; } /* * set default host->name to host_id for logging */ snprintf(host->name, KNET_MAX_HOST_LEN, "%u", host_id); /* * initialize links internal data */ for (link_idx = 0; link_idx < KNET_MAX_LINK; link_idx++) { host->link[link_idx].link_id = link_idx; host->link[link_idx].status.stats.latency_min = UINT32_MAX; } /* * add new host to the index */ knet_h->host_index[host_id] = host; /* * add new host to host list */ if (knet_h->host_head) { host->next = knet_h->host_head; } knet_h->host_head = host; _host_list_update(knet_h); exit_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); if (err < 0) { if (host) { free(host->defrag_bufs); } free(host); } errno = err ? savederrno : 0; return err; } int knet_host_remove(knet_handle_t knet_h, knet_node_id_t host_id) { int savederrno = 0, err = 0; struct knet_host *host, *removed; uint8_t link_idx; if (!_is_valid_handle(knet_h)) { return -1; } savederrno = get_global_wrlock(knet_h); if (savederrno) { log_err(knet_h, KNET_SUB_HOST, "Unable to get write lock: %s", strerror(savederrno)); errno = savederrno; return -1; } host = knet_h->host_index[host_id]; if (!host) { err = -1; savederrno = EINVAL; log_err(knet_h, KNET_SUB_HOST, "Unable to remove host %u: %s", host_id, strerror(savederrno)); goto exit_unlock; } /* * if links are configured we cannot release the host */ for (link_idx = 0; link_idx < KNET_MAX_LINK; link_idx++) { if (host->link[link_idx].configured) { err = -1; savederrno = EBUSY; log_err(knet_h, KNET_SUB_HOST, "Unable to remove host %u, links are still configured: %s", host_id, strerror(savederrno)); goto exit_unlock; } } removed = NULL; /* * removing host from list */ if (knet_h->host_head->host_id == host_id) { removed = knet_h->host_head; knet_h->host_head = removed->next; } else { for (host = knet_h->host_head; host->next != NULL; host = host->next) { if (host->next->host_id == host_id) { removed = host->next; host->next = removed->next; break; } } } knet_h->host_index[host_id] = NULL; if (removed) { free(removed->defrag_bufs); } free(removed); _host_list_update(knet_h); exit_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = err ? savederrno : 0; return err; } int knet_host_set_name(knet_handle_t knet_h, knet_node_id_t host_id, const char *name) { int savederrno = 0, err = 0; struct knet_host *host; if (!_is_valid_handle(knet_h)) { return -1; } savederrno = get_global_wrlock(knet_h); if (savederrno) { log_err(knet_h, KNET_SUB_HOST, "Unable to get write lock: %s", strerror(savederrno)); errno = savederrno; return -1; } if (!knet_h->host_index[host_id]) { err = -1; savederrno = EINVAL; log_err(knet_h, KNET_SUB_HOST, "Unable to find host %u to set name: %s", host_id, strerror(savederrno)); goto exit_unlock; } if (!name) { err = -1; savederrno = EINVAL; log_err(knet_h, KNET_SUB_HOST, "Unable to set name for host %u: %s", host_id, strerror(savederrno)); goto exit_unlock; } if (strlen(name) >= KNET_MAX_HOST_LEN) { err = -1; savederrno = EINVAL; log_err(knet_h, KNET_SUB_HOST, "Requested name for host %u is too long: %s", host_id, strerror(savederrno)); goto exit_unlock; } for (host = knet_h->host_head; host != NULL; host = host->next) { if (!strncmp(host->name, name, KNET_MAX_HOST_LEN)) { err = -1; savederrno = EEXIST; log_err(knet_h, KNET_SUB_HOST, "Duplicated name found on host_id %u", host->host_id); goto exit_unlock; } } snprintf(knet_h->host_index[host_id]->name, KNET_MAX_HOST_LEN, "%s", name); exit_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = err ? savederrno : 0; return err; } int knet_host_get_name_by_host_id(knet_handle_t knet_h, knet_node_id_t host_id, char *name) { int savederrno = 0, err = 0; if (!_is_valid_handle(knet_h)) { return -1; } if (!name) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HOST, "Unable to get read lock: %s", strerror(savederrno)); errno = savederrno; return -1; } if (!knet_h->host_index[host_id]) { savederrno = EINVAL; err = -1; log_debug(knet_h, KNET_SUB_HOST, "Host %u not found", host_id); goto exit_unlock; } snprintf(name, KNET_MAX_HOST_LEN, "%s", knet_h->host_index[host_id]->name); exit_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = err ? savederrno : 0; return err; } int knet_host_get_id_by_host_name(knet_handle_t knet_h, const char *name, knet_node_id_t *host_id) { int savederrno = 0, err = 0, found = 0; struct knet_host *host; if (!_is_valid_handle(knet_h)) { return -1; } if (!name) { errno = EINVAL; return -1; } if (!host_id) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HOST, "Unable to get read lock: %s", strerror(savederrno)); errno = savederrno; return -1; } for (host = knet_h->host_head; host != NULL; host = host->next) { if (!strncmp(name, host->name, KNET_MAX_HOST_LEN)) { found = 1; *host_id = host->host_id; break; } } if (!found) { savederrno = ENOENT; err = -1; } pthread_rwlock_unlock(&knet_h->global_rwlock); errno = err ? savederrno : 0; return err; } int knet_host_get_host_list(knet_handle_t knet_h, knet_node_id_t *host_ids, size_t *host_ids_entries) { int savederrno = 0; if (!_is_valid_handle(knet_h)) { return -1; } if ((!host_ids) || (!host_ids_entries)) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HOST, "Unable to get read lock: %s", strerror(savederrno)); errno = savederrno; return -1; } memmove(host_ids, knet_h->host_ids, sizeof(knet_h->host_ids)); *host_ids_entries = knet_h->host_ids_entries; pthread_rwlock_unlock(&knet_h->global_rwlock); return 0; } int knet_host_set_policy(knet_handle_t knet_h, knet_node_id_t host_id, uint8_t policy) { int savederrno = 0, err = 0; uint8_t old_policy; if (!_is_valid_handle(knet_h)) { return -1; } if (policy > KNET_LINK_POLICY_RR) { errno = EINVAL; return -1; } savederrno = get_global_wrlock(knet_h); if (savederrno) { log_err(knet_h, KNET_SUB_HOST, "Unable to get write lock: %s", strerror(savederrno)); errno = savederrno; return -1; } if (!knet_h->host_index[host_id]) { err = -1; savederrno = EINVAL; log_err(knet_h, KNET_SUB_HOST, "Unable to set name for host %u: %s", host_id, strerror(savederrno)); goto exit_unlock; } old_policy = knet_h->host_index[host_id]->link_handler_policy; knet_h->host_index[host_id]->link_handler_policy = policy; if (_host_dstcache_update_async(knet_h, knet_h->host_index[host_id])) { savederrno = errno; err = -1; knet_h->host_index[host_id]->link_handler_policy = old_policy; log_debug(knet_h, KNET_SUB_HOST, "Unable to update switch cache for host %u: %s", host_id, strerror(savederrno)); } log_debug(knet_h, KNET_SUB_HOST, "Host %u has new switching policy: %u", host_id, policy); exit_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = err ? savederrno : 0; return err; } int knet_host_get_policy(knet_handle_t knet_h, knet_node_id_t host_id, uint8_t *policy) { int savederrno = 0, err = 0; if (!_is_valid_handle(knet_h)) { return -1; } if (!policy) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HOST, "Unable to get read lock: %s", strerror(savederrno)); errno = savederrno; return -1; } if (!knet_h->host_index[host_id]) { err = -1; savederrno = EINVAL; log_err(knet_h, KNET_SUB_HOST, "Unable to get name for host %u: %s", host_id, strerror(savederrno)); goto exit_unlock; } *policy = knet_h->host_index[host_id]->link_handler_policy; exit_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = err ? savederrno : 0; return err; } int knet_host_get_status(knet_handle_t knet_h, knet_node_id_t host_id, struct knet_host_status *status) { int savederrno = 0, err = 0; struct knet_host *host; if (!_is_valid_handle(knet_h)) { return -1; } if (!status) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HOST, "Unable to get read lock: %s", strerror(savederrno)); errno = savederrno; return -1; } host = knet_h->host_index[host_id]; if (!host) { err = -1; savederrno = EINVAL; log_err(knet_h, KNET_SUB_HOST, "Unable to find host %u: %s", host_id, strerror(savederrno)); goto exit_unlock; } memmove(status, &host->status, sizeof(struct knet_host_status)); exit_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = err ? savederrno : 0; return err; } int knet_host_enable_status_change_notify(knet_handle_t knet_h, void *host_status_change_notify_fn_private_data, void (*host_status_change_notify_fn) ( void *private_data, knet_node_id_t host_id, uint8_t reachable, uint8_t remote, uint8_t external)) { int savederrno = 0; if (!_is_valid_handle(knet_h)) { return -1; } savederrno = get_global_wrlock(knet_h); if (savederrno) { log_err(knet_h, KNET_SUB_HOST, "Unable to get write lock: %s", strerror(savederrno)); errno = savederrno; return -1; } knet_h->host_status_change_notify_fn_private_data = host_status_change_notify_fn_private_data; knet_h->host_status_change_notify_fn = host_status_change_notify_fn; if (knet_h->host_status_change_notify_fn) { log_debug(knet_h, KNET_SUB_HOST, "host_status_change_notify_fn enabled"); } else { log_debug(knet_h, KNET_SUB_HOST, "host_status_change_notify_fn disabled"); } pthread_rwlock_unlock(&knet_h->global_rwlock); errno = 0; return 0; } void _clear_defrag_bufs_stats(struct knet_host *host) { memset(&host->in_use_defrag_buffers, 0, sizeof(host->in_use_defrag_buffers)); host->in_use_defrag_buffers_samples = 0; host->in_use_defrag_buffers_index = 0; } static void _clear_cbuffers(struct knet_host *host, seq_num_t rx_seq_num) { int i; memset(host->circular_buffer, 0, KNET_CBUFFER_SIZE); host->rx_seq_num = rx_seq_num; memset(host->circular_buffer_defrag, 0, KNET_CBUFFER_SIZE); for (i = 0; i < host->allocated_defrag_bufs; i++) { memset(&host->defrag_bufs[i], 0, sizeof(struct knet_host_defrag_buf)); } _clear_defrag_bufs_stats(host); } static void _reclaim_old_defrag_bufs(knet_handle_t knet_h, struct knet_host *host, seq_num_t seq_num) { seq_num_t head, tail; /* seq_num boundaries */ int i; head = seq_num + 1; if (knet_h->defrag_bufs_max > host->allocated_defrag_bufs) { tail = seq_num - (knet_h->defrag_bufs_max + 1); } else { tail = seq_num - (host->allocated_defrag_bufs + 1); } /* * expire old defrag buffers */ for (i = 0; i < host->allocated_defrag_bufs; i++) { if (host->defrag_bufs[i].in_use) { /* * head has done a rollover to 0+ */ if (tail > head) { if ((host->defrag_bufs[i].pckt_seq >= head) && (host->defrag_bufs[i].pckt_seq <= tail)) { host->defrag_bufs[i].in_use = 0; } } else { if ((host->defrag_bufs[i].pckt_seq >= head) || (host->defrag_bufs[i].pckt_seq <= tail)){ host->defrag_bufs[i].in_use = 0; } } } } } /* * check if a given packet seq num is in the circular buffers * defrag_buf = 0 -> use normal cbuf 1 -> use the defrag buffer lookup */ int _seq_num_lookup(knet_handle_t knet_h, struct knet_host *host, seq_num_t seq_num, int defrag_buf, int clear_buf) { size_t head, tail; /* circular buffer indexes */ seq_num_t seq_dist; char *dst_cbuf = host->circular_buffer; char *dst_cbuf_defrag = host->circular_buffer_defrag; seq_num_t *dst_seq_num = &host->rx_seq_num; + /* + * There is a potential race condition where the sender + * is overloaded, sending data packets before pings + * can kick in and set the correct dst_seq_num. + * + * if this node is starting up (dst_seq_num = 0), + * it can start rejecing valid packets and get stuck. + * + * Set the dst_seq_num to the first seen packet and + * use that as reference instead. + */ + if (!*dst_seq_num) { + *dst_seq_num = seq_num; + } + if (clear_buf) { _clear_cbuffers(host, seq_num); } _reclaim_old_defrag_bufs(knet_h, host, *dst_seq_num); if (seq_num < *dst_seq_num) { seq_dist = (SEQ_MAX - seq_num) + *dst_seq_num; } else { seq_dist = *dst_seq_num - seq_num; } head = seq_num % KNET_CBUFFER_SIZE; if (seq_dist < KNET_CBUFFER_SIZE) { /* seq num is in ring buffer */ if (!defrag_buf) { return (dst_cbuf[head] == 0) ? 1 : 0; } else { return (dst_cbuf_defrag[head] == 0) ? 1 : 0; } } else if (seq_dist <= SEQ_MAX - KNET_CBUFFER_SIZE) { memset(dst_cbuf, 0, KNET_CBUFFER_SIZE); memset(dst_cbuf_defrag, 0, KNET_CBUFFER_SIZE); *dst_seq_num = seq_num; } /* cleaning up circular buffer */ tail = (*dst_seq_num + 1) % KNET_CBUFFER_SIZE; if (tail > head) { memset(dst_cbuf + tail, 0, KNET_CBUFFER_SIZE - tail); memset(dst_cbuf, 0, head + 1); memset(dst_cbuf_defrag + tail, 0, KNET_CBUFFER_SIZE - tail); memset(dst_cbuf_defrag, 0, head + 1); } else { memset(dst_cbuf + tail, 0, head - tail + 1); memset(dst_cbuf_defrag + tail, 0, head - tail + 1); } *dst_seq_num = seq_num; return 1; } void _seq_num_set(struct knet_host *host, seq_num_t seq_num, int defrag_buf) { if (!defrag_buf) { host->circular_buffer[seq_num % KNET_CBUFFER_SIZE] = 1; } else { host->circular_buffer_defrag[seq_num % KNET_CBUFFER_SIZE] = 1; } return; } int _host_dstcache_update_async(knet_handle_t knet_h, struct knet_host *host) { int savederrno = 0; knet_node_id_t host_id = host->host_id; if (sendto(knet_h->dstsockfd[1], &host_id, sizeof(host_id), MSG_DONTWAIT | MSG_NOSIGNAL, NULL, 0) != sizeof(host_id)) { savederrno = errno; log_debug(knet_h, KNET_SUB_HOST, "Unable to write to dstpipefd[1]: %s", strerror(savederrno)); errno = savederrno; return -1; } return 0; } int _host_dstcache_update_sync(knet_handle_t knet_h, struct knet_host *host) { int link_idx; int best_priority = -1; int reachable = 0; if (knet_h->host_id == host->host_id && knet_h->has_loop_link) { host->active_link_entries = 1; return 0; } host->active_link_entries = 0; for (link_idx = 0; link_idx < KNET_MAX_LINK; link_idx++) { if (host->link[link_idx].status.enabled != 1) /* link is not enabled */ continue; if (host->link[link_idx].status.connected != 1) /* link is not enabled */ continue; if (host->link[link_idx].has_valid_mtu != 1) /* link does not have valid MTU */ continue; if (host->link_handler_policy == KNET_LINK_POLICY_PASSIVE) { /* for passive we look for the only active link with higher priority */ if (host->link[link_idx].priority > best_priority) { host->active_links[0] = link_idx; best_priority = host->link[link_idx].priority; } host->active_link_entries = 1; } else { /* for RR and ACTIVE we need to copy all available links */ host->active_links[host->active_link_entries] = link_idx; host->active_link_entries++; } } if (host->link_handler_policy == KNET_LINK_POLICY_PASSIVE) { log_info(knet_h, KNET_SUB_HOST, "host: %u (passive) best link: %u (pri: %u)", host->host_id, host->link[host->active_links[0]].link_id, host->link[host->active_links[0]].priority); } else { log_info(knet_h, KNET_SUB_HOST, "host: %u has %u active links", host->host_id, host->active_link_entries); } /* no active links, we can clean the circular buffers and indexes */ if (!host->active_link_entries) { log_warn(knet_h, KNET_SUB_HOST, "host: %u has no active links", host->host_id); _clear_cbuffers(host, 0); } else { reachable = 1; } if (host->status.reachable != reachable) { host->status.reachable = reachable; if (knet_h->host_status_change_notify_fn) { knet_h->host_status_change_notify_fn( knet_h->host_status_change_notify_fn_private_data, host->host_id, host->status.reachable, host->status.remote, host->status.external); } } return 0; } void _handle_onwire_version(knet_handle_t knet_h, struct knet_host *host, struct knet_header *inbuf) { struct knet_host *tmp_host = NULL; uint8_t onwire_ver = knet_h->onwire_max_ver; int docallback = 0; /* * data we process here are onwire independent * we are in a global read only lock context, so it´s safe to parse host lists * and we can change onwire_ver using the dedicated mutex */ /* * update current host onwire info */ host->onwire_ver = inbuf->kh_version; host->onwire_max_ver = inbuf->kh_max_ver; for (tmp_host = knet_h->host_head; tmp_host != NULL; tmp_host = tmp_host->next) { /* * do not attempt to change protocol till * we see all nodes at least once. */ if (!tmp_host->onwire_max_ver) { return; } /* * ignore nodes were max ver is lower than our min ver * logged as error by thread_rx, we need to make sure to skip it * during onwire_ver calculation. */ if (tmp_host->onwire_max_ver < knet_h->onwire_min_ver) { continue; } /* * use the highest max_ver common to all known nodes */ if (tmp_host->onwire_max_ver < onwire_ver) { onwire_ver = tmp_host->onwire_max_ver; } } if (pthread_mutex_lock(&knet_h->onwire_mutex)) { log_debug(knet_h, KNET_SUB_HOST, "Unable to get onwire mutex lock"); return; } if (knet_h->onwire_force_ver) { onwire_ver = knet_h->onwire_force_ver; } if (knet_h->onwire_ver != onwire_ver) { log_debug(knet_h, KNET_SUB_HOST, "node %u updating onwire version to %u", knet_h->host_id, onwire_ver); knet_h->onwire_ver = onwire_ver; docallback = 1; } pthread_mutex_unlock(&knet_h->onwire_mutex); /* * do the callback outside of locked context and use cached value * to avoid blocking on locking */ if ((docallback) && (knet_h->onwire_ver_notify_fn)) { knet_h->onwire_ver_notify_fn(knet_h->onwire_ver_notify_fn_private_data, knet_h->onwire_min_ver, knet_h->onwire_max_ver, onwire_ver); } } diff --git a/libknet/threads_rx.c b/libknet/threads_rx.c index 38e31327..2cd7e5ed 100644 --- a/libknet/threads_rx.c +++ b/libknet/threads_rx.c @@ -1,1206 +1,1206 @@ /* * Copyright (C) 2012-2021 Red Hat, Inc. All rights reserved. * * Authors: Fabio M. Di Nitto * Federico Simoncelli * * This software licensed under 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 "links.h" #include "links_acl.h" #include "logging.h" #include "transports.h" #include "transport_common.h" #include "threads_common.h" #include "threads_heartbeat.h" #include "threads_pmtud.h" #include "threads_rx.h" #include "netutils.h" #include "onwire_v1.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; } } } /* * calculate use % of defrag buffers per host * and if % is <= knet_h->defrag_bufs_shrink_threshold for the last second, then half the size */ static void _shrink_defrag_buffers(knet_handle_t knet_h) { struct knet_host *host; struct knet_host_defrag_buf *new_bufs = NULL; struct timespec now; unsigned long long time_diff; /* nanoseconds */ uint16_t i, x, in_use_bufs; uint32_t sum; /* * first run. */ if ((knet_h->defrag_bufs_last_run.tv_sec == 0) && (knet_h->defrag_bufs_last_run.tv_nsec == 0)) { clock_gettime(CLOCK_MONOTONIC, &knet_h->defrag_bufs_last_run); return; } clock_gettime(CLOCK_MONOTONIC, &now); timespec_diff(knet_h->defrag_bufs_last_run, now, &time_diff); if (time_diff < (((unsigned long long)knet_h->defrag_bufs_usage_samples_timespan * 1000000000) / knet_h->defrag_bufs_usage_samples)) { return; } /* * record the last run */ memmove(&knet_h->defrag_bufs_last_run, &now, sizeof(struct timespec)); /* * do the real work: */ for (host = knet_h->host_head; host != NULL; host = host->next) { /* * Update buffer usage stats. We do this for all nodes. */ in_use_bufs = 0; for (i = 0; i < host->allocated_defrag_bufs; i++) { if (host->defrag_bufs[i].in_use) { in_use_bufs++; } } /* * record only % */ host->in_use_defrag_buffers[host->in_use_defrag_buffers_index] = (in_use_bufs * 100 / host->allocated_defrag_bufs); host->in_use_defrag_buffers_index++; /* * make sure to stay within buffer */ if (host->in_use_defrag_buffers_index == knet_h->defrag_bufs_usage_samples) { host->in_use_defrag_buffers_index = 0; } /* * only allow shrinking if we have enough samples */ if (host->in_use_defrag_buffers_samples < knet_h->defrag_bufs_usage_samples) { host->in_use_defrag_buffers_samples++; continue; } /* * only allow shrinking if in use bufs are <= knet_h->defrag_bufs_shrink_threshold% */ if (knet_h->defrag_bufs_reclaim_policy == RECLAIM_POLICY_AVERAGE) { sum = 0; for (i = 0; i < knet_h->defrag_bufs_usage_samples; i++) { sum += host->in_use_defrag_buffers[i]; } sum = sum / knet_h->defrag_bufs_usage_samples; if (sum > knet_h->defrag_bufs_shrink_threshold) { continue; } } else { sum = 0; for (i = 0; i < knet_h->defrag_bufs_usage_samples; i++) { if (host->in_use_defrag_buffers[i] > knet_h->defrag_bufs_shrink_threshold) { sum = 1; } } if (sum) { continue; } } /* * only allow shrinking if allocated bufs > min_defrag_bufs */ if (host->allocated_defrag_bufs == knet_h->defrag_bufs_min) { continue; } /* * compat all the in_use buffers at the beginning. * we the checks above, we are 100% sure they fit */ x = 0; for (i = 0; i < host->allocated_defrag_bufs; i++) { if (host->defrag_bufs[i].in_use) { memmove(&host->defrag_bufs[x], &host->defrag_bufs[i], sizeof(struct knet_host_defrag_buf)); x++; } } /* * memory allocation is not critical. it just means the system is under * memory pressure and we will need to wait our turn to free memory... how odd :) */ new_bufs = realloc(host->defrag_bufs, sizeof(struct knet_host_defrag_buf) * (host->allocated_defrag_bufs / 2)); if (!new_bufs) { log_err(knet_h, KNET_SUB_RX, "Unable to decrease defrag buffers for host %u: %s", host->host_id, strerror(errno)); continue; } host->defrag_bufs = new_bufs; host->allocated_defrag_bufs = host->allocated_defrag_bufs / 2; /* * clear buffer use stats. Old ones are no good for new one */ _clear_defrag_bufs_stats(host); log_debug(knet_h, KNET_SUB_RX, "Defrag buffers for host %u decreased from %u to: %u", host->host_id, host->allocated_defrag_bufs * 2, host->allocated_defrag_bufs); } } /* * check if we can double the defrag buffers. * * return 0 if we cannot reallocate * return 1 if we have more buffers */ static int _realloc_defrag_buffers(knet_handle_t knet_h, struct knet_host *src_host) { struct knet_host_defrag_buf *new_bufs = NULL; int i; /* * max_defrag_bufs is a power of 2 * allocated_defrag_bufs doubles on each iteration. * Sooner or later (and hopefully never) allocated with be == to max. */ if (src_host->allocated_defrag_bufs < knet_h->defrag_bufs_max) { new_bufs = realloc(src_host->defrag_bufs, src_host->allocated_defrag_bufs * 2 * sizeof(struct knet_host_defrag_buf)); if (!new_bufs) { log_err(knet_h, KNET_SUB_RX, "Unable to increase defrag buffers for host %u: %s", src_host->host_id, strerror(errno)); return 0; } /* * keep the math simple here between arrays, pointers and what not. * Init each buffer individually. */ for (i = src_host->allocated_defrag_bufs; i < src_host->allocated_defrag_bufs * 2; i++) { memset(&new_bufs[i], 0, sizeof(struct knet_host_defrag_buf)); } src_host->allocated_defrag_bufs = src_host->allocated_defrag_bufs * 2; src_host->defrag_bufs = new_bufs; /* * clear buffer use stats. Old ones are no good for new one */ _clear_defrag_bufs_stats(src_host); log_debug(knet_h, KNET_SUB_RX, "Defrag buffers for host %u increased from %u to: %u", src_host->host_id, src_host->allocated_defrag_bufs / 2, src_host->allocated_defrag_bufs); return 1; } return 0; } /* * this functions needs to return an index * to a knet_host_defrag_buf. (-1 on errors) */ static int _find_pckt_defrag_buf(knet_handle_t knet_h, struct knet_host *src_host, seq_num_t seq_num) { int i, oldest; uint16_t cur_allocated_defrag_bufs = src_host->allocated_defrag_bufs; /* * check if there is a buffer already in use handling the same seq_num */ for (i = 0; i < src_host->allocated_defrag_bufs; i++) { if (src_host->defrag_bufs[i].in_use) { if (src_host->defrag_bufs[i].pckt_seq == 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(knet_h, src_host, seq_num, 1, 0)) { errno = ETIME; return -1; } /* * register the pckt as seen */ _seq_num_set(src_host, seq_num, 1); /* * see if there is a free buffer */ for (i = 0; i < src_host->allocated_defrag_bufs; i++) { if (!src_host->defrag_bufs[i].in_use) { return i; } } /* * check if we can increase num of buffers */ if (_realloc_defrag_buffers(knet_h, src_host)) { return cur_allocated_defrag_bufs + 1; } /* * 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 < src_host->allocated_defrag_bufs; i++) { if (_timecmp(src_host->defrag_bufs[i].last_update, src_host->defrag_bufs[oldest].last_update) < 0) { oldest = i; } } src_host->defrag_bufs[oldest].in_use = 0; return oldest; } static int _pckt_defrag(knet_handle_t knet_h, struct knet_host *src_host, seq_num_t seq_num, unsigned char *data, ssize_t *len, uint8_t frags, uint8_t frag_seq) { struct knet_host_defrag_buf *defrag_buf; int defrag_buf_idx; defrag_buf_idx = _find_pckt_defrag_buf(knet_h, src_host, seq_num); if (defrag_buf_idx < 0) { return 1; } defrag_buf = &src_host->defrag_bufs[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 = 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[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 (frag_seq == frags) { 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), data, *len); } } else { defrag_buf->frag_size = *len; } if (defrag_buf->frag_size) { memmove(defrag_buf->buf + ((frag_seq - 1) * defrag_buf->frag_size), data, *len); } defrag_buf->frag_recv++; defrag_buf->frag_map[frag_seq] = 1; /* * check if we received all the fragments */ if (defrag_buf->frag_recv == frags) { /* * special case the last pckt */ if (defrag_buf->last_first) { memmove(defrag_buf->buf + ((frags - 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 = ((frags - 1) * defrag_buf->frag_size) + defrag_buf->last_frag_size; /* * copy the pckt back in the user data */ memmove(data, defrag_buf->buf, *len); /* * free this buffer */ defrag_buf->in_use = 0; return 0; } return 1; } static int _handle_data_stats(knet_handle_t knet_h, struct knet_link *src_link, ssize_t len, uint64_t decrypt_time) { int stats_err; /* data stats at the top for consistency with TX */ src_link->status.stats.rx_data_packets++; src_link->status.stats.rx_data_bytes += len; if (decrypt_time) { stats_err = pthread_mutex_lock(&knet_h->handle_stats_mutex); if (stats_err < 0) { log_err(knet_h, KNET_SUB_RX, "Unable to get mutex lock: %s", strerror(stats_err)); return -1; } /* Only update the crypto overhead for data packets. Mainly to be consistent with TX */ if (decrypt_time < knet_h->stats.rx_crypt_time_min) { knet_h->stats.rx_crypt_time_min = decrypt_time; } if (decrypt_time > knet_h->stats.rx_crypt_time_max) { knet_h->stats.rx_crypt_time_max = decrypt_time; } knet_h->stats.rx_crypt_time_ave = (knet_h->stats.rx_crypt_time_ave * knet_h->stats.rx_crypt_packets + decrypt_time) / (knet_h->stats.rx_crypt_packets+1); knet_h->stats.rx_crypt_packets++; pthread_mutex_unlock(&knet_h->handle_stats_mutex); } return 0; } static int _decompress_data(knet_handle_t knet_h, uint8_t decompress_type, unsigned char *data, ssize_t *len, ssize_t header_size) { int err = 0, stats_err = 0; if (decompress_type) { ssize_t decmp_outlen = KNET_DATABUFSIZE_COMPRESS; struct timespec start_time; struct timespec end_time; uint64_t decompress_time; clock_gettime(CLOCK_MONOTONIC, &start_time); err = decompress(knet_h, decompress_type, data, *len - header_size, knet_h->recv_from_links_buf_decompress, &decmp_outlen); clock_gettime(CLOCK_MONOTONIC, &end_time); timespec_diff(start_time, end_time, &decompress_time); stats_err = pthread_mutex_lock(&knet_h->handle_stats_mutex); if (stats_err < 0) { log_err(knet_h, KNET_SUB_RX, "Unable to get mutex lock: %s", strerror(stats_err)); return -1; } if (!err) { /* Collect stats */ if (decompress_time < knet_h->stats.rx_compress_time_min) { knet_h->stats.rx_compress_time_min = decompress_time; } if (decompress_time > knet_h->stats.rx_compress_time_max) { knet_h->stats.rx_compress_time_max = decompress_time; } knet_h->stats.rx_compress_time_ave = (knet_h->stats.rx_compress_time_ave * knet_h->stats.rx_compressed_packets + decompress_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(data, knet_h->recv_from_links_buf_decompress, decmp_outlen); *len = decmp_outlen + header_size; } else { knet_h->stats.rx_failed_to_decompress++; pthread_mutex_unlock(&knet_h->handle_stats_mutex); log_warn(knet_h, KNET_SUB_COMPRESS, "Unable to decompress packet (%d): %s", err, strerror(errno)); return -1; } pthread_mutex_unlock(&knet_h->handle_stats_mutex); } return 0; } static int _check_destination(knet_handle_t knet_h, struct knet_header *inbuf, unsigned char *data, ssize_t len, ssize_t header_size, int8_t *channel) { knet_node_id_t dst_host_ids[KNET_MAX_HOST]; size_t dst_host_ids_entries = 0; int bcast = 1; size_t host_idx; int found = 0; if (knet_h->dst_host_filter_fn) { bcast = knet_h->dst_host_filter_fn( knet_h->dst_host_filter_fn_private_data, data, len - header_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 -1; } if ((!bcast) && (!dst_host_ids_entries)) { log_debug(knet_h, KNET_SUB_RX, "Message is unicast but no dst_host_ids_entries"); return -1; } /* 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 -1; } 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 -1; } } } return 0; } static int _deliver_data(knet_handle_t knet_h, unsigned char *data, ssize_t len, ssize_t header_size, int8_t channel) { struct iovec iov_out[1]; ssize_t outlen = 0; memset(iov_out, 0, sizeof(iov_out)); retry: iov_out[0].iov_base = (void *) data + outlen; iov_out[0].iov_len = len - (outlen + header_size); outlen = writev(knet_h->sockfd[channel].sockfd[knet_h->sockfd[channel].is_created], iov_out, 1); if ((outlen > 0) && (outlen < (ssize_t)iov_out[0].iov_len)) { log_debug(knet_h, KNET_SUB_RX, "Unable to send all data to the application in one go. Expected: %zu Sent: %zd\n", iov_out[0].iov_len, outlen); goto retry; } 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 -1; } if ((size_t)outlen != iov_out[0].iov_len) { return -1; } return 0; } static void _process_data(knet_handle_t knet_h, struct knet_host *src_host, struct knet_link *src_link, struct knet_header *inbuf, ssize_t len, uint64_t decrypt_time) { int8_t channel; uint8_t decompress_type = 0; ssize_t header_size; seq_num_t seq_num; uint8_t frags, frag_seq; unsigned char *data; if (_handle_data_stats(knet_h, src_link, len, decrypt_time) < 0) { return; } /* * register host is sending data. Required to determine if we need * to reset circular buffers. (see onwire_v1.c) */ src_host->got_data = 1; if (knet_h->onwire_ver_remap) { get_data_header_info_v1(knet_h, inbuf, &header_size, &channel, &seq_num, &decompress_type, &frags, &frag_seq); data = get_data_v1(knet_h, inbuf); } else { switch (inbuf->kh_version) { case 1: get_data_header_info_v1(knet_h, inbuf, &header_size, &channel, &seq_num, &decompress_type, &frags, &frag_seq); data = get_data_v1(knet_h, inbuf); break; default: log_warn(knet_h, KNET_SUB_RX, "processing data onwire version %u not supported", inbuf->kh_version); return; break; } } if (!_seq_num_lookup(knet_h, src_host, 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 (frags > 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 * * the defrag code assumes that data packets have all the same size * except the last one that might be smaller. * */ len = len - header_size; if (_pckt_defrag(knet_h, src_host, seq_num, data, &len, frags, frag_seq)) { return; } len = len + header_size; } if (_decompress_data(knet_h, decompress_type, data, &len, header_size) < 0) { return; } if (!src_host->status.reachable) { log_debug(knet_h, KNET_SUB_RX, "Source host %u not reachable yet. Discarding packet.", src_host->host_id); return; } if (knet_h->enabled != 1) /* data forward is disabled */ return; if (_check_destination(knet_h, inbuf, data, len, header_size, &channel) < 0) { return; } 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; } #ifdef ONWIRE_V1_EXTRA_DEBUG if (inbuf->khp_data_v1_checksum != compute_chksum(data, len - header_size)) { log_err(knet_h, KNET_SUB_RX, "Received incorrect data checksum after reassembly from host: %u seq: %u", src_host->host_id, seq_num); /* * give a chance to the log threads to pick up the message */ sleep(1); abort(); } #endif if (_deliver_data(knet_h, data, len, header_size, channel) < 0) { return; } _seq_num_set(src_host, seq_num, 0); } static struct knet_header *_decrypt_packet(knet_handle_t knet_h, struct knet_header *inbuf, ssize_t *len, uint64_t *decrypt_time) { int try_decrypt = 0; int i = 0; struct timespec start_time; struct timespec end_time; ssize_t outlen; for (i = 1; i <= KNET_MAX_CRYPTO_INSTANCES; i++) { if (knet_h->crypto_instance[i]) { try_decrypt = 1; break; } } if ((!try_decrypt) && (knet_h->crypto_only == KNET_CRYPTO_RX_DISALLOW_CLEAR_TRAFFIC)) { log_debug(knet_h, KNET_SUB_RX, "RX thread configured to accept only crypto packets, but no crypto configs are configured!"); return NULL; } if (try_decrypt) { 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"); if (knet_h->crypto_only == KNET_CRYPTO_RX_DISALLOW_CLEAR_TRAFFIC) { return NULL; } log_debug(knet_h, KNET_SUB_RX, "Attempting to process packet as clear data"); } else { clock_gettime(CLOCK_MONOTONIC, &end_time); timespec_diff(start_time, end_time, decrypt_time); *len = outlen; inbuf = (struct knet_header *)knet_h->recv_from_links_buf_decrypt; } } return inbuf; } static int _packet_checks(knet_handle_t knet_h, struct knet_header *inbuf, ssize_t len) { #ifdef ONWIRE_V1_EXTRA_DEBUG uint32_t rx_packet_checksum, expected_packet_checksum; #endif if (len < (ssize_t)(KNET_HEADER_SIZE + 1)) { log_debug(knet_h, KNET_SUB_RX, "Packet is too short: %ld", (long)len); return -1; } #ifdef ONWIRE_V1_EXTRA_DEBUG rx_packet_checksum = inbuf->kh_checksum; inbuf->kh_checksum = 0; expected_packet_checksum = compute_chksum((const unsigned char *)inbuf, len); if (rx_packet_checksum != expected_packet_checksum) { log_err(knet_h, KNET_SUB_RX, "Received packet with incorrect checksum. Received: %u Expected: %u", rx_packet_checksum, expected_packet_checksum); /* * give a chance to the log threads to pick up the message */ sleep(1); abort(); } #endif /* * old versions of knet did not advertise max_ver and max_ver is set to 0. */ if (!inbuf->kh_max_ver) { inbuf->kh_max_ver = 1; } /* * if the node joining max version is lower than the min version * then we reject the node */ if (inbuf->kh_max_ver < knet_h->onwire_min_ver) { log_warn(knet_h, KNET_SUB_RX, "Received packet version %u from node %u, lower than currently minimal supported onwire version. Rejecting.", inbuf->kh_version, inbuf->kh_node); return -1; } /* * if the node joining with version higher than our max version * then we reject the node */ if (inbuf->kh_version > knet_h->onwire_max_ver) { log_warn(knet_h, KNET_SUB_RX, "Received packet version %u from node %u, higher than currently maximum supported onwire version. Rejecting.", inbuf->kh_version, inbuf->kh_node); return -1; } /* * if the node joining with version lower than the current in use version * then we reject the node * * NOTE: should we make this configurable and support downgrades? */ if ((!knet_h->onwire_force_ver) && (inbuf->kh_version < knet_h->onwire_ver) && (inbuf->kh_max_ver > inbuf->kh_version)) { log_warn(knet_h, KNET_SUB_RX, "Received packet version %u from node %u, lower than currently in use onwire version. Rejecting.", inbuf->kh_version, inbuf->kh_node); return -1; } return 0; } static void _handle_dynip(knet_handle_t knet_h, struct knet_host *src_host, struct knet_link *src_link, int sockfd, const struct knet_mmsghdr *msg) { if (src_link->dynamic == KNET_LINK_DYNIP) { if (cmpaddr(&src_link->dst_addr, msg->msg_hdr.msg_name) != 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, msg->msg_hdr.msg_name, sizeof(struct sockaddr_storage)); if (knet_addrtostr(&src_link->dst_addr, sockaddr_len(&src_link->dst_addr), 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); } } /* * processing incoming packets vs access lists */ static int _check_rx_acl(knet_handle_t knet_h, struct knet_link *src_link, const struct knet_mmsghdr *msg) { if (knet_h->use_access_lists) { if (!check_validate(knet_h, src_link, msg->msg_hdr.msg_name)) { char src_ipaddr[KNET_MAX_HOST_LEN]; char src_port[KNET_MAX_PORT_LEN]; memset(src_ipaddr, 0, KNET_MAX_HOST_LEN); memset(src_port, 0, KNET_MAX_PORT_LEN); if (knet_addrtostr(msg->msg_hdr.msg_name, sockaddr_len(msg->msg_hdr.msg_name), src_ipaddr, KNET_MAX_HOST_LEN, src_port, KNET_MAX_PORT_LEN) < 0) { log_debug(knet_h, KNET_SUB_RX, "Packet rejected: unable to resolve host/port"); } else { log_debug(knet_h, KNET_SUB_RX, "Packet rejected from %s/%s", src_ipaddr, src_port); } return 0; } } return 1; } static void _parse_recv_from_links(knet_handle_t knet_h, int sockfd, const struct knet_mmsghdr *msg) { int savederrno = 0, stats_err = 0; struct knet_host *src_host; struct knet_link *src_link; uint64_t decrypt_time = 0; struct knet_header *inbuf = msg->msg_hdr.msg_iov->iov_base; ssize_t len = msg->msg_len; int i, found_link = 0; inbuf = _decrypt_packet(knet_h, inbuf, &len, &decrypt_time); if (!inbuf) { return; } if (_packet_checks(knet_h, inbuf, len) < 0) { return; } inbuf->kh_node = ntohs(inbuf->kh_node); /* * determine source host */ 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; } /* * deteremine source link */ if (inbuf->kh_type == KNET_HEADER_TYPE_PING) { _handle_onwire_version(knet_h, src_host, inbuf); if (knet_h->onwire_ver_remap) { src_link = get_link_from_pong_v1(knet_h, src_host, inbuf); } else { switch (inbuf->kh_version) { case 1: src_link = get_link_from_pong_v1(knet_h, src_host, inbuf); break; default: log_warn(knet_h, KNET_SUB_RX, "Parsing ping onwire version %u not supported", inbuf->kh_version); return; break; } } if (!_check_rx_acl(knet_h, src_link, msg)) { return; } _handle_dynip(knet_h, src_host, src_link, sockfd, msg); } else { /* all other packets */ for (i = 0; i < KNET_MAX_LINK; i++) { src_link = &src_host->link[i]; if (cmpaddr(&src_link->dst_addr, msg->msg_hdr.msg_name) == 0) { found_link = 1; break; } } if (found_link) { /* * this check is currently redundant.. Keep it here for now */ if (!_check_rx_acl(knet_h, src_link, msg)) { return; } } else { log_debug(knet_h, KNET_SUB_RX, "Unable to determine source link for data packet. Discarding packet."); return; } } stats_err = pthread_mutex_lock(&src_link->link_stats_mutex); if (stats_err) { log_err(knet_h, KNET_SUB_RX, "Unable to get stats mutex lock for host %u link %u: %s", src_host->host_id, src_link->link_id, strerror(savederrno)); return; } switch (inbuf->kh_type) { case KNET_HEADER_TYPE_DATA: _process_data(knet_h, src_host, src_link, inbuf, len, decrypt_time); break; case KNET_HEADER_TYPE_PING: process_ping(knet_h, src_host, src_link, inbuf, len); break; case KNET_HEADER_TYPE_PONG: process_pong(knet_h, src_host, src_link, inbuf, len); break; case KNET_HEADER_TYPE_PMTUD: src_link->status.stats.rx_pmtu_packets++; src_link->status.stats.rx_pmtu_bytes += len; /* Unlock so we don't deadlock with tx_mutex */ pthread_mutex_unlock(&src_link->link_stats_mutex); process_pmtud(knet_h, src_link, inbuf); return; /* Don't need to unlock link_stats_mutex */ break; case KNET_HEADER_TYPE_PMTUD_REPLY: src_link->status.stats.rx_pmtu_packets++; src_link->status.stats.rx_pmtu_bytes += len; /* pmtud_mutex can't be acquired while we hold a link_stats_mutex (ordering) */ pthread_mutex_unlock(&src_link->link_stats_mutex); process_pmtud_reply(knet_h, src_link, inbuf); return; break; default: pthread_mutex_unlock(&src_link->link_stats_mutex); return; break; } pthread_mutex_unlock(&src_link->link_stats_mutex); } 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 = knet_h->knet_transport_fd_tracker[sockfd].sockaddr_len; } 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 KNET_TRANSPORT_RX_ERROR: /* on error */ + case KNET_TRANSPORT_RX_ISDATA_ERROR: /* on error */ log_debug(knet_h, KNET_SUB_RX, "Transport reported error parsing packet"); goto exit_unlock; break; case KNET_TRANSPORT_RX_NOT_DATA_CONTINUE: /* 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 KNET_TRANSPORT_RX_NOT_DATA_STOP: /* 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 KNET_TRANSPORT_RX_IS_DATA: /* packet is data and should be parsed as such */ _parse_recv_from_links(knet_h, sockfd, &msg[i]); break; case KNET_TRANSPORT_RX_OOB_DATA_CONTINUE: log_debug(knet_h, KNET_SUB_RX, "Transport is processing sock OOB data, continue"); break; case KNET_TRANSPORT_RX_OOB_DATA_STOP: log_debug(knet_h, KNET_SUB_RX, "Transport has completed processing sock OOB data, stop"); goto exit_unlock; break; } } exit_unlock: _shrink_defrag_buffers(knet_h); 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)); memset(&events, 0, sizeof(events)); 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); /* Real value filled in before actual use */ 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_h->threads_timer_res / 1000); /* * the RX threads only need to notify that there has been at least * one successful run after queue flush has been requested. * See setfwd in handle.c */ if (get_thread_flush_queue(knet_h, KNET_THREAD_RX) == KNET_THREAD_QUEUE_FLUSH) { set_thread_flush_queue(knet_h, KNET_THREAD_RX, KNET_THREAD_QUEUE_FLUSHED); } /* * 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; } ssize_t knet_recv(knet_handle_t knet_h, char *buff, const size_t buff_len, const int8_t channel) { int savederrno = 0; ssize_t err = 0; struct iovec iov_in; if (!_is_valid_handle(knet_h)) { return -1; } if (buff == NULL) { errno = EINVAL; return -1; } if (buff_len <= 0) { errno = EINVAL; return -1; } if (buff_len > KNET_MAX_PACKET_SIZE) { errno = EINVAL; return -1; } if (channel < 0) { errno = EINVAL; return -1; } if (channel >= KNET_DATAFD_MAX) { errno = EINVAL; return -1; } savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock); if (savederrno) { log_err(knet_h, KNET_SUB_HANDLE, "Unable to get read lock: %s", strerror(savederrno)); errno = savederrno; return -1; } if (!knet_h->sockfd[channel].in_use) { savederrno = EINVAL; err = -1; goto out_unlock; } memset(&iov_in, 0, sizeof(iov_in)); iov_in.iov_base = (void *)buff; iov_in.iov_len = buff_len; err = readv(knet_h->sockfd[channel].sockfd[0], &iov_in, 1); savederrno = errno; out_unlock: pthread_rwlock_unlock(&knet_h->global_rwlock); errno = err ? savederrno : 0; return err; } diff --git a/libknet/transport_sctp.c b/libknet/transport_sctp.c index ff3dd589..d2d13ba6 100644 --- a/libknet/transport_sctp.c +++ b/libknet/transport_sctp.c @@ -1,1639 +1,1639 @@ /* * Copyright (C) 2016-2021 Red Hat, Inc. All rights reserved. * * Author: Christine Caulfield * * This software licensed under LGPL-2.0+ */ #include "config.h" #include #include #include #include #include #include #include #include #include "compat.h" #include "host.h" #include "links.h" #include "links_acl.h" #include "links_acl_ip.h" #include "logging.h" #include "netutils.h" #include "common.h" #include "transport_common.h" #include "transports.h" #include "threads_common.h" #ifdef HAVE_NETINET_SCTP_H #include #include "transport_sctp.h" typedef struct sctp_handle_info { struct qb_list_head listen_links_list; struct qb_list_head connect_links_list; int connect_epollfd; int connectsockfd[2]; int listen_epollfd; int listensockfd[2]; pthread_t connect_thread; pthread_t listen_thread; socklen_t event_subscribe_kernel_size; char *event_subscribe_buffer; } sctp_handle_info_t; /* * use by fd_tracker data type */ #define SCTP_NO_LINK_INFO 0 #define SCTP_LISTENER_LINK_INFO 1 #define SCTP_ACCEPTED_LINK_INFO 2 #define SCTP_CONNECT_LINK_INFO 3 /* * this value is per listener */ #define MAX_ACCEPTED_SOCKS 256 typedef struct sctp_listen_link_info { struct qb_list_head list; int listen_sock; int accepted_socks[MAX_ACCEPTED_SOCKS]; struct sockaddr_storage src_address; int on_listener_epoll; int on_rx_epoll; int sock_shutdown; } sctp_listen_link_info_t; typedef struct sctp_accepted_link_info { char mread_buf[KNET_DATABUFSIZE]; ssize_t mread_len; sctp_listen_link_info_t *link_info; } sctp_accepted_link_info_t ; typedef struct sctp_connect_link_info { struct qb_list_head list; sctp_listen_link_info_t *listener; struct knet_link *link; struct sockaddr_storage dst_address; int connect_sock; int on_rx_epoll; int close_sock; int sock_shutdown; } sctp_connect_link_info_t; /* * socket handling functions * * those functions do NOT perform locking. locking * should be handled in the right context from callers */ /* * sockets are removed from rx_epoll from callers * see also error handling functions */ static int _close_connect_socket(knet_handle_t knet_h, struct knet_link *kn_link) { int err = 0, savederrno = 0; struct epoll_event ev; sctp_connect_link_info_t *info = kn_link->transport_link; if (info->connect_sock != -1) { if (info->on_rx_epoll) { memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = info->connect_sock; if (epoll_ctl(knet_h->recv_from_links_epollfd, EPOLL_CTL_DEL, info->connect_sock, &ev)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to remove connected socket from epoll pool: %s", strerror(savederrno)); goto exit_error; } info->on_rx_epoll = 0; } if (_set_fd_tracker(knet_h, info->connect_sock, KNET_MAX_TRANSPORTS, SCTP_NO_LINK_INFO, 0, NULL) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to set fd tracker: %s", strerror(savederrno)); } else { close(info->connect_sock); info->connect_sock = -1; } } exit_error: errno = savederrno; return err; } static int _enable_sctp_notifications(knet_handle_t knet_h, int sock, const char *type) { int err = 0, savederrno = 0; sctp_handle_info_t *handle_info = knet_h->transports[KNET_TRANSPORT_SCTP]; if (setsockopt(sock, IPPROTO_SCTP, SCTP_EVENTS, handle_info->event_subscribe_buffer, handle_info->event_subscribe_kernel_size) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to enable %s events: %s", type, strerror(savederrno)); } errno = savederrno; return err; } static int _configure_sctp_socket(knet_handle_t knet_h, int sock, struct sockaddr_storage *address, uint64_t flags, const char *type) { int err = 0, savederrno = 0; int value; int level; #ifdef SOL_SCTP level = SOL_SCTP; #else level = IPPROTO_SCTP; #endif if (_configure_transport_socket(knet_h, sock, address, flags, type) < 0) { savederrno = errno; err = -1; goto exit_error; } value = 1; if (setsockopt(sock, SOL_SOCKET, SO_REUSEADDR, &value, sizeof(value)) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set reuseaddr on socket %d: %s", sock, strerror(savederrno)); goto exit_error; } value = 1; if (setsockopt(sock, level, SCTP_NODELAY, &value, sizeof(value)) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSPORT, "Unable to set sctp nodelay: %s", strerror(savederrno)); goto exit_error; } if (_enable_sctp_notifications(knet_h, sock, type) < 0) { savederrno = errno; err = -1; } exit_error: errno = savederrno; return err; } static int _reconnect_socket(knet_handle_t knet_h, struct knet_link *kn_link) { int err = 0, savederrno = 0; sctp_connect_link_info_t *info = kn_link->transport_link; if (connect(info->connect_sock, (struct sockaddr *)&kn_link->dst_addr, sockaddr_len(&kn_link->dst_addr)) < 0) { savederrno = errno; log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "SCTP socket %d received error: %s", info->connect_sock, strerror(savederrno)); if ((savederrno != EALREADY) && (savederrno != EINPROGRESS) && (savederrno != EISCONN)) { err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to connect SCTP socket %d: %s", info->connect_sock, strerror(savederrno)); } } errno = savederrno; return err; } static int _create_connect_socket(knet_handle_t knet_h, struct knet_link *kn_link) { int err = 0, savederrno = 0; struct epoll_event ev; sctp_connect_link_info_t *info = kn_link->transport_link; int connect_sock; struct sockaddr_storage connect_addr; connect_sock = socket(kn_link->dst_addr.ss_family, SOCK_STREAM, IPPROTO_SCTP); if (connect_sock < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to create send/recv socket: %s", strerror(savederrno)); goto exit_error; } if (_configure_sctp_socket(knet_h, connect_sock, &kn_link->dst_addr, kn_link->flags, "SCTP connect") < 0) { savederrno = errno; err = -1; goto exit_error; } memset(&connect_addr, 0, sizeof(struct sockaddr_storage)); if (knet_strtoaddr(kn_link->status.src_ipaddr, "0", &connect_addr, sockaddr_len(&connect_addr)) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to resolve connecting socket: %s", strerror(savederrno)); goto exit_error; } if (bind(connect_sock, (struct sockaddr *)&connect_addr, sockaddr_len(&connect_addr)) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to bind connecting socket: %s", strerror(savederrno)); goto exit_error; } if (_set_fd_tracker(knet_h, connect_sock, KNET_TRANSPORT_SCTP, SCTP_CONNECT_LINK_INFO, sockaddr_len(&kn_link->src_addr), info) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to set fd tracker: %s", strerror(savederrno)); goto exit_error; } memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = connect_sock; if (epoll_ctl(knet_h->recv_from_links_epollfd, EPOLL_CTL_ADD, connect_sock, &ev)) { log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to add connected socket to epoll pool: %s", strerror(errno)); } info->on_rx_epoll = 1; info->connect_sock = connect_sock; info->close_sock = 0; kn_link->outsock = info->connect_sock; if (_reconnect_socket(knet_h, kn_link) < 0) { savederrno = errno; err = -1; goto exit_error; } exit_error: if (err) { if (connect_sock >= 0) { close(connect_sock); } } errno = savederrno; return err; } static void _lock_sleep_relock(knet_handle_t knet_h) { int i = 0; /* Don't hold onto the lock while sleeping */ pthread_rwlock_unlock(&knet_h->global_rwlock); while (i < 5) { usleep(knet_h->threads_timer_res / 16); if (!pthread_rwlock_rdlock(&knet_h->global_rwlock)) { /* * lock acquired, we can go out */ return; } else { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to get read lock!"); i++; } } /* * time to crash! if we cannot re-acquire the lock * there is no easy way out of this one */ assert(0); } int sctp_transport_tx_sock_error(knet_handle_t knet_h, int sockfd, int recv_err, int recv_errno) { sctp_connect_link_info_t *connect_info = knet_h->knet_transport_fd_tracker[sockfd].data; sctp_accepted_link_info_t *accepted_info = knet_h->knet_transport_fd_tracker[sockfd].data; sctp_listen_link_info_t *listen_info; if (recv_err < 0) { switch (knet_h->knet_transport_fd_tracker[sockfd].data_type) { case SCTP_CONNECT_LINK_INFO: if (connect_info->link->transport_connected == 0) { return KNET_TRANSPORT_SOCK_ERROR_INTERNAL; } break; case SCTP_ACCEPTED_LINK_INFO: listen_info = accepted_info->link_info; if (listen_info->listen_sock != sockfd) { if (listen_info->on_rx_epoll == 0) { - return -KNET_TRANSPORT_SOCK_ERROR_INTERNAL; + return KNET_TRANSPORT_SOCK_ERROR_INTERNAL; } } break; } if (recv_errno == EAGAIN) { #ifdef DEBUG log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Sock: %d is overloaded. Slowing TX down", sockfd); #endif _lock_sleep_relock(knet_h); return KNET_TRANSPORT_SOCK_ERROR_RETRY; } return KNET_TRANSPORT_SOCK_ERROR_INTERNAL; } return KNET_TRANSPORT_SOCK_ERROR_IGNORE; } /* * socket error management functions * * both called with global read lock. * * NOTE: we need to remove the fd from the epoll as soon as possible * even before we notify the respective thread to take care of it * because scheduling can make it so that this thread will overload * and the threads supposed to take care of the error will never * be able to take action. * we CANNOT handle FDs here directly (close/reconnect/etc) due * to locking context. We need to delegate that to their respective * management threads within the global write lock. * * this function is called from: * - RX thread with recv_err <= 0 directly on recvmmsg error * - transport_rx_is_data when msg_len == 0 (recv_err = 1) * - transport_rx_is_data on notification (recv_err = 2) * * basically this small abuse of recv_err is to detect notifications * generated by sockets created by listen(). */ transport_sock_error_t sctp_transport_rx_sock_error(knet_handle_t knet_h, int sockfd, int recv_err, int recv_errno) { struct epoll_event ev; sctp_accepted_link_info_t *accepted_info = knet_h->knet_transport_fd_tracker[sockfd].data; sctp_listen_link_info_t *listen_info; sctp_handle_info_t *handle_info = knet_h->transports[KNET_TRANSPORT_SCTP]; switch (knet_h->knet_transport_fd_tracker[sockfd].data_type) { case SCTP_CONNECT_LINK_INFO: /* * all connect link have notifications enabled * and we accept only data from notification and * generic recvmmsg errors. * * Errors generated by msg_len 0 can be ignored because * they follow a notification (double notification) */ if (recv_err != 1) { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Notifying connect thread that sockfd %d received an error", sockfd); if (sendto(handle_info->connectsockfd[1], &sockfd, sizeof(int), MSG_DONTWAIT | MSG_NOSIGNAL, NULL, 0) != sizeof(int)) { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to notify connect thread: %s", strerror(errno)); } } break; case SCTP_ACCEPTED_LINK_INFO: listen_info = accepted_info->link_info; if (listen_info->listen_sock != sockfd) { if (recv_err != 1) { if (listen_info->on_rx_epoll) { memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = sockfd; if (epoll_ctl(knet_h->recv_from_links_epollfd, EPOLL_CTL_DEL, sockfd, &ev)) { log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to remove EOFed socket from epoll pool: %s", strerror(errno)); return -1; } listen_info->on_rx_epoll = 0; } log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Notifying listen thread that sockfd %d received an error", sockfd); if (sendto(handle_info->listensockfd[1], &sockfd, sizeof(int), MSG_DONTWAIT | MSG_NOSIGNAL, NULL, 0) != sizeof(int)) { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to notify listen thread: %s", strerror(errno)); } } } else { /* * this means the listen() socket has generated * a notification. now what? :-) */ log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Received stray notification for listen() socket %d", sockfd); } break; default: log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Received unknown notification? %d", sockfd); break; } /* * Under RX pressure we need to give time to IPC to pick up the message */ _lock_sleep_relock(knet_h); return 0; } /* * NOTE: sctp_transport_rx_is_data is called with global rdlock * delegate any FD error management to sctp_transport_rx_sock_error * and keep this code to parsing incoming data only */ transport_rx_isdata_t sctp_transport_rx_is_data(knet_handle_t knet_h, int sockfd, struct knet_mmsghdr *msg) { size_t i; struct iovec *iov = msg->msg_hdr.msg_iov; size_t iovlen = msg->msg_hdr.msg_iovlen; struct sctp_assoc_change *sac; union sctp_notification *snp; sctp_accepted_link_info_t *listen_info = knet_h->knet_transport_fd_tracker[sockfd].data; sctp_connect_link_info_t *connect_info = knet_h->knet_transport_fd_tracker[sockfd].data; if (!(msg->msg_hdr.msg_flags & MSG_NOTIFICATION)) { if (msg->msg_len == 0) { /* * NOTE: with event notification enabled, we receive error twice: * 1) from the event notification * 2) followed by a 0 byte msg_len * * the event handler should take care to avoid #2 by stopping * the rx thread from processing more packets than necessary. */ if (knet_h->knet_transport_fd_tracker[sockfd].data_type == SCTP_CONNECT_LINK_INFO) { if (connect_info->sock_shutdown) { return KNET_TRANSPORT_RX_OOB_DATA_CONTINUE; } } else { if (listen_info->link_info->sock_shutdown) { return KNET_TRANSPORT_RX_OOB_DATA_CONTINUE; } } /* * this is pretty much dead code and we should never hit it. * keep it for safety and avoid the rx thread to process * bad info / data. */ return KNET_TRANSPORT_RX_NOT_DATA_STOP; } /* * missing MSG_EOR has to be treated as a short read * from the socket and we need to fill in the mread buf * while we wait for MSG_EOR */ if (!(msg->msg_hdr.msg_flags & MSG_EOR)) { /* * copy the incoming data into mread_buf + mread_len (incremental) * and increase mread_len */ memmove(listen_info->mread_buf + listen_info->mread_len, iov->iov_base, msg->msg_len); listen_info->mread_len = listen_info->mread_len + msg->msg_len; return KNET_TRANSPORT_RX_NOT_DATA_CONTINUE; } /* * got EOR. * if mread_len is > 0 we are completing a packet from short reads * complete reassembling the packet in mread_buf, copy it back in the iov * and set the iov/msg len numbers (size) correctly */ if (listen_info->mread_len) { /* * add last fragment to mread_buf */ memmove(listen_info->mread_buf + listen_info->mread_len, iov->iov_base, msg->msg_len); listen_info->mread_len = listen_info->mread_len + msg->msg_len; /* * move all back into the iovec */ memmove(iov->iov_base, listen_info->mread_buf, listen_info->mread_len); msg->msg_len = listen_info->mread_len; listen_info->mread_len = 0; } return KNET_TRANSPORT_RX_IS_DATA; } if (!(msg->msg_hdr.msg_flags & MSG_EOR)) { return KNET_TRANSPORT_RX_NOT_DATA_STOP; } for (i = 0; i < iovlen; i++) { snp = iov[i].iov_base; switch (snp->sn_header.sn_type) { case SCTP_ASSOC_CHANGE: sac = &snp->sn_assoc_change; switch (sac->sac_state) { case SCTP_COMM_LOST: log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "[event] sctp assoc change socket %d: comm_lost", sockfd); if (knet_h->knet_transport_fd_tracker[sockfd].data_type == SCTP_CONNECT_LINK_INFO) { connect_info->close_sock = 1; connect_info->link->transport_connected = 0; } sctp_transport_rx_sock_error(knet_h, sockfd, 2, 0); return KNET_TRANSPORT_RX_OOB_DATA_STOP; break; case SCTP_COMM_UP: log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "[event] sctp assoc change socket %d: comm_up", sockfd); if (knet_h->knet_transport_fd_tracker[sockfd].data_type == SCTP_CONNECT_LINK_INFO) { connect_info->link->transport_connected = 1; } break; case SCTP_RESTART: log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "[event] sctp assoc change socket %d: restart", sockfd); break; case SCTP_SHUTDOWN_COMP: log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "[event] sctp assoc change socket %d: shutdown comp", sockfd); if (knet_h->knet_transport_fd_tracker[sockfd].data_type == SCTP_CONNECT_LINK_INFO) { connect_info->close_sock = 1; } sctp_transport_rx_sock_error(knet_h, sockfd, 2, 0); return KNET_TRANSPORT_RX_OOB_DATA_STOP; break; case SCTP_CANT_STR_ASSOC: log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "[event] sctp assoc change socket %d: cant str assoc", sockfd); sctp_transport_rx_sock_error(knet_h, sockfd, 2, 0); break; default: log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "[event] sctp assoc change socket %d: unknown %d", sockfd, sac->sac_state); break; } break; case SCTP_SHUTDOWN_EVENT: log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "[event] sctp shutdown event socket %d", sockfd); if (knet_h->knet_transport_fd_tracker[sockfd].data_type == SCTP_CONNECT_LINK_INFO) { connect_info->link->transport_connected = 0; connect_info->sock_shutdown = 1; } else { listen_info->link_info->sock_shutdown = 1; } break; case SCTP_SEND_FAILED: log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "[event] sctp send failed socket: %d", sockfd); break; case SCTP_PEER_ADDR_CHANGE: log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "[event] sctp peer addr change socket %d", sockfd); break; case SCTP_REMOTE_ERROR: log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "[event] sctp remote error socket %d", sockfd); break; default: log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "[event] unknown sctp event socket: %d type: %hu", sockfd, snp->sn_header.sn_type); break; } } return KNET_TRANSPORT_RX_OOB_DATA_CONTINUE; } int sctp_transport_link_is_down(knet_handle_t knet_h, struct knet_link *kn_link) { sctp_handle_info_t *handle_info = knet_h->transports[KNET_TRANSPORT_SCTP]; sctp_connect_link_info_t *info = kn_link->transport_link; kn_link->transport_connected = 0; info->close_sock = 1; log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Notifying connect thread that sockfd %d received a link down event", info->connect_sock); if (sendto(handle_info->connectsockfd[1], &info->connect_sock, sizeof(int), MSG_DONTWAIT | MSG_NOSIGNAL, NULL, 0) != sizeof(int)) { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to notify connect thread: %s", strerror(errno)); } return 0; } /* * connect / outgoing socket management thread */ /* * _handle_connected_sctp* are called with a global write lock * from the connect_thread */ static void _handle_connected_sctp_socket(knet_handle_t knet_h, int connect_sock) { int err; unsigned int status, len = sizeof(status); sctp_connect_link_info_t *info = knet_h->knet_transport_fd_tracker[connect_sock].data; struct knet_link *kn_link = info->link; if (info->close_sock) { if (_close_connect_socket(knet_h, kn_link) < 0) { log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to close sock %d from _handle_connected_sctp_socket: %s", connect_sock, strerror(errno)); return; } info->close_sock = 0; if (_create_connect_socket(knet_h, kn_link) < 0) { log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to recreate connecting sock! %s", strerror(errno)); return; } } _reconnect_socket(knet_h, info->link); err = getsockopt(connect_sock, SOL_SOCKET, SO_ERROR, &status, &len); if (err) { log_err(knet_h, KNET_SUB_TRANSP_SCTP, "SCTP getsockopt() on connecting socket %d failed: %s", connect_sock, strerror(errno)); return; } if (status) { log_info(knet_h, KNET_SUB_TRANSP_SCTP, "SCTP connect on %d to %s port %s failed: %s", connect_sock, kn_link->status.dst_ipaddr, kn_link->status.dst_port, strerror(status)); /* * No need to create a new socket if connect failed, * just retry connect */ return; } log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "SCTP handler fd %d now connected to %s port %s", connect_sock, kn_link->status.dst_ipaddr, kn_link->status.dst_port); } static void _handle_connected_sctp_notifications(knet_handle_t knet_h) { int sockfd = -1; sctp_handle_info_t *handle_info = knet_h->transports[KNET_TRANSPORT_SCTP]; if (recv(handle_info->connectsockfd[0], &sockfd, sizeof(int), MSG_DONTWAIT | MSG_NOSIGNAL) != sizeof(int)) { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Short read on connectsockfd"); return; } if (_is_valid_fd(knet_h, sockfd) < 1) { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Received stray notification for connected socket fd error"); return; } /* * revalidate sockfd */ if ((sockfd < 0) || (sockfd >= KNET_MAX_FDS)) { return; } log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Processing connected error on socket: %d", sockfd); _handle_connected_sctp_socket(knet_h, sockfd); } static void *_sctp_connect_thread(void *data) { int savederrno; int i, nev; knet_handle_t knet_h = (knet_handle_t) data; sctp_handle_info_t *handle_info = knet_h->transports[KNET_TRANSPORT_SCTP]; struct epoll_event events[KNET_EPOLL_MAX_EVENTS]; set_thread_status(knet_h, KNET_THREAD_SCTP_CONN, KNET_THREAD_STARTED); memset(&events, 0, sizeof(events)); while (!shutdown_in_progress(knet_h)) { nev = epoll_wait(handle_info->connect_epollfd, events, KNET_EPOLL_MAX_EVENTS, knet_h->threads_timer_res / 1000); /* * we use timeout to detect if thread is shutting down */ if (nev == 0) { continue; } if (nev < 0) { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "SCTP connect handler EPOLL ERROR: %s", strerror(errno)); continue; } /* * Sort out which FD has a connection */ savederrno = get_global_wrlock(knet_h); if (savederrno) { log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to get write lock: %s", strerror(savederrno)); continue; } /* * minor optimization: deduplicate events * * in some cases we can receive multiple notifcations * of the same FD having issues or need handling. * It's enough to process it once even tho it's safe * to handle them multiple times. */ for (i = 0; i < nev; i++) { if (events[i].data.fd == handle_info->connectsockfd[0]) { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Received notification from rx_error for connected socket"); _handle_connected_sctp_notifications(knet_h); } else { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Received stray notification on connected sockfd %d\n", events[i].data.fd); } } pthread_rwlock_unlock(&knet_h->global_rwlock); /* * this thread can generate events for itself. * we need to sleep in between loops to allow other threads * to be scheduled */ usleep(knet_h->reconnect_int * 1000); } set_thread_status(knet_h, KNET_THREAD_SCTP_CONN, KNET_THREAD_STOPPED); return NULL; } /* * listen/incoming connections management thread */ /* * Listener received a new connection * called with a write lock from main thread */ static void _handle_incoming_sctp(knet_handle_t knet_h, int listen_sock) { int err = 0, savederrno = 0; int new_fd; int i = -1; sctp_listen_link_info_t *info = knet_h->knet_transport_fd_tracker[listen_sock].data; struct epoll_event ev; struct sockaddr_storage ss; socklen_t sock_len = sizeof(ss); char addr_str[KNET_MAX_HOST_LEN]; char port_str[KNET_MAX_PORT_LEN]; sctp_accepted_link_info_t *accept_info = NULL; struct knet_host *host; struct knet_link *kn_link; int link_idx; sctp_connect_link_info_t *this_link_connect_info; sctp_listen_link_info_t *this_link_listen_info; int pass_acl = 0; memset(&ss, 0, sizeof(ss)); new_fd = accept(listen_sock, (struct sockaddr *)&ss, &sock_len); if (new_fd < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Incoming: accept error: %s", strerror(errno)); goto exit_error; } if (knet_addrtostr(&ss, sizeof(ss), addr_str, KNET_MAX_HOST_LEN, port_str, KNET_MAX_PORT_LEN) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Incoming: unable to gather socket info"); goto exit_error; } log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Incoming: received connection from: %s port: %s", addr_str, port_str); if (knet_h->use_access_lists) { for (host = knet_h->host_head; host != NULL; host = host->next) { for (link_idx = 0; link_idx < KNET_MAX_LINK; link_idx++) { kn_link = &host->link[link_idx]; if ((kn_link->configured) && (kn_link->transport == KNET_TRANSPORT_SCTP)) { this_link_connect_info = kn_link->transport_link; this_link_listen_info = this_link_connect_info->listener; if ((this_link_listen_info->listen_sock == listen_sock) && (check_validate(knet_h, kn_link, &ss))) { pass_acl = 1; break; } } } if (pass_acl) { break; } } if (!pass_acl) { savederrno = EINVAL; log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Connection rejected from %s/%s", addr_str, port_str); close(new_fd); errno = savederrno; return; } } /* * Keep a track of all accepted FDs */ for (i=0; iaccepted_socks[i] == -1) { info->accepted_socks[i] = new_fd; break; } } if (i == MAX_ACCEPTED_SOCKS) { errno = EBUSY; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Incoming: too many connections!"); goto exit_error; } if (_configure_common_socket(knet_h, new_fd, 0, "SCTP incoming") < 0) { /* Inherit flags from listener? */ savederrno = errno; err = -1; goto exit_error; } if (_enable_sctp_notifications(knet_h, new_fd, "Incoming connection") < 0) { savederrno = errno; err = -1; goto exit_error; } accept_info = malloc(sizeof(sctp_accepted_link_info_t)); if (!accept_info) { savederrno = errno; err = -1; goto exit_error; } memset(accept_info, 0, sizeof(sctp_accepted_link_info_t)); accept_info->link_info = info; if (_set_fd_tracker(knet_h, new_fd, KNET_TRANSPORT_SCTP, SCTP_ACCEPTED_LINK_INFO, knet_h->knet_transport_fd_tracker[listen_sock].sockaddr_len, accept_info) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to set fd tracker: %s", strerror(errno)); goto exit_error; } memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = new_fd; if (epoll_ctl(knet_h->recv_from_links_epollfd, EPOLL_CTL_ADD, new_fd, &ev)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Incoming: unable to add accepted socket %d to epoll pool: %s", new_fd, strerror(errno)); goto exit_error; } info->on_rx_epoll = 1; log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Incoming: accepted new fd %d for %s/%s (listen fd: %d). index: %d", new_fd, addr_str, port_str, info->listen_sock, i); exit_error: if (err) { if ((i >= 0) && (i < MAX_ACCEPTED_SOCKS)) { info->accepted_socks[i] = -1; } /* * check the error to make coverity scan happy. * _set_fd_tracker cannot fail at this stage */ if (_set_fd_tracker(knet_h, new_fd, KNET_MAX_TRANSPORTS, SCTP_NO_LINK_INFO, 0, NULL) < 0){ log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to update fdtracker for socket %d", new_fd); } free(accept_info); if (new_fd >= 0) { close(new_fd); } } errno = savederrno; return; } /* * Listen thread received a notification of a bad socket that needs closing * called with a write lock from main thread */ static void _handle_listen_sctp_errors(knet_handle_t knet_h) { int sockfd = -1; sctp_handle_info_t *handle_info = knet_h->transports[KNET_TRANSPORT_SCTP]; sctp_accepted_link_info_t *accept_info; sctp_listen_link_info_t *info; struct knet_host *host; int link_idx; int i; if (recv(handle_info->listensockfd[0], &sockfd, sizeof(int), MSG_DONTWAIT | MSG_NOSIGNAL) != sizeof(int)) { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Short read on listensockfd"); return; } if (_is_valid_fd(knet_h, sockfd) < 1) { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Received stray notification for listen socket fd error"); return; } /* * revalidate sockfd */ if ((sockfd < 0) || (sockfd >= KNET_MAX_FDS)) { return; } log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Processing listen error on socket: %d", sockfd); accept_info = knet_h->knet_transport_fd_tracker[sockfd].data; info = accept_info->link_info; /* * clear all links using this accepted socket as * outbound dynamically connected socket */ for (host = knet_h->host_head; host != NULL; host = host->next) { for (link_idx = 0; link_idx < KNET_MAX_LINK; link_idx++) { if ((host->link[link_idx].dynamic == KNET_LINK_DYNIP) && (host->link[link_idx].outsock == sockfd)) { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Found dynamic connection on host %d link %d (%d)", host->host_id, link_idx, sockfd); host->link[link_idx].status.dynconnected = 0; host->link[link_idx].transport_connected = 0; host->link[link_idx].outsock = 0; memset(&host->link[link_idx].dst_addr, 0, sizeof(struct sockaddr_storage)); } } } for (i=0; iaccepted_socks[i]) { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Closing accepted socket %d", sockfd); /* * check the error to make coverity scan happy. * _set_fd_tracker cannot fail at this stage */ if (_set_fd_tracker(knet_h, sockfd, KNET_MAX_TRANSPORTS, SCTP_NO_LINK_INFO, 0, NULL) < 0) { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to update fdtracker for socket %d", sockfd); } info->accepted_socks[i] = -1; free(accept_info); close(sockfd); break; /* Keeps covscan happy */ } } } static void *_sctp_listen_thread(void *data) { int savederrno; int i, nev; knet_handle_t knet_h = (knet_handle_t) data; sctp_handle_info_t *handle_info = knet_h->transports[KNET_TRANSPORT_SCTP]; struct epoll_event events[KNET_EPOLL_MAX_EVENTS]; set_thread_status(knet_h, KNET_THREAD_SCTP_LISTEN, KNET_THREAD_STARTED); memset(&events, 0, sizeof(events)); while (!shutdown_in_progress(knet_h)) { nev = epoll_wait(handle_info->listen_epollfd, events, KNET_EPOLL_MAX_EVENTS, knet_h->threads_timer_res / 1000); /* * we use timeout to detect if thread is shutting down */ if (nev == 0) { continue; } if (nev < 0) { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "SCTP listen handler EPOLL ERROR: %s", strerror(errno)); continue; } savederrno = get_global_wrlock(knet_h); if (savederrno) { log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to get write lock: %s", strerror(savederrno)); continue; } /* * Sort out which FD has an incoming connection */ for (i = 0; i < nev; i++) { if (events[i].data.fd == handle_info->listensockfd[0]) { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Received notification from rx_error for listener/accepted socket"); _handle_listen_sctp_errors(knet_h); } else { if (_is_valid_fd(knet_h, events[i].data.fd) == 1) { _handle_incoming_sctp(knet_h, events[i].data.fd); } else { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Received listen notification from invalid socket"); } } } pthread_rwlock_unlock(&knet_h->global_rwlock); } set_thread_status(knet_h, KNET_THREAD_SCTP_LISTEN, KNET_THREAD_STOPPED); return NULL; } /* * sctp_link_listener_start/stop are called in global write lock * context from set_config and clear_config. */ static sctp_listen_link_info_t *sctp_link_listener_start(knet_handle_t knet_h, struct knet_link *kn_link) { int err = 0, savederrno = 0; int listen_sock = -1; struct epoll_event ev; sctp_listen_link_info_t *info = NULL; sctp_handle_info_t *handle_info = knet_h->transports[KNET_TRANSPORT_SCTP]; /* * Only allocate a new listener if src address is different */ qb_list_for_each_entry(info, &handle_info->listen_links_list, list) { if (memcmp(&info->src_address, &kn_link->src_addr, sizeof(struct sockaddr_storage)) == 0) { if ((check_add(knet_h, kn_link, -1, &kn_link->dst_addr, &kn_link->dst_addr, CHECK_TYPE_ADDRESS, CHECK_ACCEPT) < 0) && (errno != EEXIST)) { return NULL; } return info; } } info = malloc(sizeof(sctp_listen_link_info_t)); if (!info) { err = -1; goto exit_error; } memset(info, 0, sizeof(sctp_listen_link_info_t)); memset(info->accepted_socks, -1, sizeof(info->accepted_socks)); memmove(&info->src_address, &kn_link->src_addr, sizeof(struct sockaddr_storage)); listen_sock = socket(kn_link->src_addr.ss_family, SOCK_STREAM, IPPROTO_SCTP); if (listen_sock < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to create listener socket: %s", strerror(savederrno)); goto exit_error; } if (_configure_sctp_socket(knet_h, listen_sock, &kn_link->src_addr, kn_link->flags, "SCTP listener") < 0) { savederrno = errno; err = -1; goto exit_error; } if (bind(listen_sock, (struct sockaddr *)&kn_link->src_addr, sockaddr_len(&kn_link->src_addr)) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to bind listener socket: %s", strerror(savederrno)); goto exit_error; } if (listen(listen_sock, 5) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to listen on listener socket: %s", strerror(savederrno)); goto exit_error; } if (_set_fd_tracker(knet_h, listen_sock, KNET_TRANSPORT_SCTP, SCTP_LISTENER_LINK_INFO, sockaddr_len(&kn_link->src_addr), info) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to set fd tracker: %s", strerror(savederrno)); goto exit_error; } if ((check_add(knet_h, kn_link, -1, &kn_link->dst_addr, &kn_link->dst_addr, CHECK_TYPE_ADDRESS, CHECK_ACCEPT) < 0) && (errno != EEXIST)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to configure default access lists: %s", strerror(savederrno)); goto exit_error; } memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = listen_sock; if (epoll_ctl(handle_info->listen_epollfd, EPOLL_CTL_ADD, listen_sock, &ev)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to add listener to epoll pool: %s", strerror(savederrno)); goto exit_error; } info->on_listener_epoll = 1; info->listen_sock = listen_sock; qb_list_add(&info->list, &handle_info->listen_links_list); log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Listening on fd %d for %s:%s", listen_sock, kn_link->status.src_ipaddr, kn_link->status.src_port); exit_error: if (err) { if ((info) && (info->on_listener_epoll)) { epoll_ctl(handle_info->listen_epollfd, EPOLL_CTL_DEL, listen_sock, &ev); } if (listen_sock >= 0) { check_rmall(knet_h, kn_link); close(listen_sock); } if (info) { free(info); info = NULL; } } errno = savederrno; return info; } static int sctp_link_listener_stop(knet_handle_t knet_h, struct knet_link *kn_link) { int err = 0, savederrno = 0; int found = 0, i; struct knet_host *host; int link_idx; sctp_handle_info_t *handle_info = knet_h->transports[KNET_TRANSPORT_SCTP]; sctp_connect_link_info_t *this_link_info = kn_link->transport_link; sctp_listen_link_info_t *info = this_link_info->listener; sctp_connect_link_info_t *link_info; struct epoll_event ev; for (host = knet_h->host_head; host != NULL; host = host->next) { for (link_idx = 0; link_idx < KNET_MAX_LINK; link_idx++) { if (&host->link[link_idx] == kn_link) continue; link_info = host->link[link_idx].transport_link; if ((link_info) && (link_info->listener == info)) { found = 1; break; } } } if ((check_rm(knet_h, kn_link, &kn_link->dst_addr, &kn_link->dst_addr, CHECK_TYPE_ADDRESS, CHECK_ACCEPT) < 0) && (errno != ENOENT)) { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to remove default access lists for %d", info->listen_sock); } if (found) { this_link_info->listener = NULL; log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "SCTP listener socket %d still in use", info->listen_sock); savederrno = EBUSY; err = -1; goto exit_error; } if (info->on_listener_epoll) { memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = info->listen_sock; if (epoll_ctl(handle_info->listen_epollfd, EPOLL_CTL_DEL, info->listen_sock, &ev)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to remove listener to epoll pool: %s", strerror(savederrno)); goto exit_error; } info->on_listener_epoll = 0; } if (_set_fd_tracker(knet_h, info->listen_sock, KNET_MAX_TRANSPORTS, SCTP_NO_LINK_INFO, 0, NULL) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to set fd tracker: %s", strerror(savederrno)); goto exit_error; } check_rmall(knet_h, kn_link); close(info->listen_sock); for (i=0; i< MAX_ACCEPTED_SOCKS; i++) { if (info->accepted_socks[i] > -1) { memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = info->accepted_socks[i]; if (epoll_ctl(knet_h->recv_from_links_epollfd, EPOLL_CTL_DEL, info->accepted_socks[i], &ev)) { log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to remove EOFed socket from epoll pool: %s", strerror(errno)); } info->on_rx_epoll = 0; free(knet_h->knet_transport_fd_tracker[info->accepted_socks[i]].data); close(info->accepted_socks[i]); if (_set_fd_tracker(knet_h, info->accepted_socks[i], KNET_MAX_TRANSPORTS, SCTP_NO_LINK_INFO, 0, NULL) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to set fd tracker: %s", strerror(savederrno)); goto exit_error; } info->accepted_socks[i] = -1; } } qb_list_del(&info->list); free(info); this_link_info->listener = NULL; exit_error: errno = savederrno; return err; } /* * Links config/clear. Both called with global wrlock from link_set_config/clear_config */ int sctp_transport_link_set_config(knet_handle_t knet_h, struct knet_link *kn_link) { int savederrno = 0, err = 0; sctp_connect_link_info_t *info; sctp_handle_info_t *handle_info = knet_h->transports[KNET_TRANSPORT_SCTP]; info = malloc(sizeof(sctp_connect_link_info_t)); if (!info) { goto exit_error; } memset(info, 0, sizeof(sctp_connect_link_info_t)); kn_link->transport_link = info; info->link = kn_link; memmove(&info->dst_address, &kn_link->dst_addr, sizeof(struct sockaddr_storage)); info->connect_sock = -1; info->listener = sctp_link_listener_start(knet_h, kn_link); if (!info->listener) { savederrno = errno; err = -1; goto exit_error; } if (kn_link->dynamic == KNET_LINK_STATIC) { if (_create_connect_socket(knet_h, kn_link) < 0) { savederrno = errno; err = -1; goto exit_error; } kn_link->outsock = info->connect_sock; } qb_list_add(&info->list, &handle_info->connect_links_list); exit_error: if (err) { if (info) { if (info->connect_sock >= 0) { close(info->connect_sock); } if (info->listener) { sctp_link_listener_stop(knet_h, kn_link); } kn_link->transport_link = NULL; free(info); } } errno = savederrno; return err; } /* * called with global wrlock */ int sctp_transport_link_clear_config(knet_handle_t knet_h, struct knet_link *kn_link) { int err = 0, savederrno = 0; sctp_connect_link_info_t *info; if (!kn_link) { errno = EINVAL; return -1; } info = kn_link->transport_link; if (!info) { errno = EINVAL; return -1; } if ((sctp_link_listener_stop(knet_h, kn_link) <0) && (errno != EBUSY)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to remove listener transport: %s", strerror(savederrno)); goto exit_error; } if (_close_connect_socket(knet_h, kn_link) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to close connected socket: %s", strerror(savederrno)); goto exit_error; } qb_list_del(&info->list); free(info); kn_link->transport_link = NULL; exit_error: errno = savederrno; return err; } /* * transport_free and transport_init are * called only from knet_handle_new and knet_handle_free. * all resources (hosts/links) should have been already freed at this point * and they are called in a write locked context, hence they * don't need their own locking. */ int sctp_transport_free(knet_handle_t knet_h) { sctp_handle_info_t *handle_info; void *thread_status; struct epoll_event ev; if (!knet_h->transports[KNET_TRANSPORT_SCTP]) { errno = EINVAL; return -1; } handle_info = knet_h->transports[KNET_TRANSPORT_SCTP]; /* * keep it here while we debug list usage and such */ if (!qb_list_empty(&handle_info->listen_links_list)) { log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Internal error. listen links list is not empty"); } if (!qb_list_empty(&handle_info->connect_links_list)) { log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Internal error. connect links list is not empty"); } if (handle_info->listen_thread) { pthread_cancel(handle_info->listen_thread); pthread_join(handle_info->listen_thread, &thread_status); } if (handle_info->connect_thread) { pthread_cancel(handle_info->connect_thread); pthread_join(handle_info->connect_thread, &thread_status); } if (handle_info->listensockfd[0] >= 0) { memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = handle_info->listensockfd[0]; epoll_ctl(handle_info->listen_epollfd, EPOLL_CTL_DEL, handle_info->listensockfd[0], &ev); } if (handle_info->connectsockfd[0] >= 0) { memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = handle_info->connectsockfd[0]; epoll_ctl(handle_info->connect_epollfd, EPOLL_CTL_DEL, handle_info->connectsockfd[0], &ev); } _close_socketpair(knet_h, handle_info->connectsockfd); _close_socketpair(knet_h, handle_info->listensockfd); if (handle_info->listen_epollfd >= 0) { close(handle_info->listen_epollfd); } if (handle_info->connect_epollfd >= 0) { close(handle_info->connect_epollfd); } free(handle_info->event_subscribe_buffer); free(handle_info); knet_h->transports[KNET_TRANSPORT_SCTP] = NULL; return 0; } static int _sctp_subscribe_init(knet_handle_t knet_h) { int test_socket, savederrno; sctp_handle_info_t *handle_info = knet_h->transports[KNET_TRANSPORT_SCTP]; char dummy_events[100]; struct sctp_event_subscribe *events; /* Below we set the first 6 fields of this expanding struct. * SCTP_EVENTS is deprecated, but SCTP_EVENT is not available * on Linux; on the other hand, FreeBSD and old Linux does not * accept small transfers, so we can't simply use this minimum * everywhere. Thus we query and store the native size. */ const unsigned int subscribe_min = 6; test_socket = socket(PF_INET, SOCK_STREAM, IPPROTO_SCTP); if (test_socket < 0) { if (errno == EPROTONOSUPPORT) { log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "SCTP not supported, skipping initialization"); return 0; } savederrno = errno; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to create test socket: %s", strerror(savederrno)); return savederrno; } handle_info->event_subscribe_kernel_size = sizeof dummy_events; if (getsockopt(test_socket, IPPROTO_SCTP, SCTP_EVENTS, &dummy_events, &handle_info->event_subscribe_kernel_size)) { close(test_socket); savederrno = errno; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to query kernel size of struct sctp_event_subscribe: %s", strerror(savederrno)); return savederrno; } close(test_socket); if (handle_info->event_subscribe_kernel_size < subscribe_min) { savederrno = ERANGE; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "No kernel support for the necessary notifications: struct sctp_event_subscribe is %u bytes, %u needed", handle_info->event_subscribe_kernel_size, subscribe_min); return savederrno; } events = malloc(handle_info->event_subscribe_kernel_size); if (!events) { savederrno = errno; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Failed to allocate event subscribe buffer: %s", strerror(savederrno)); return savederrno; } memset(events, 0, handle_info->event_subscribe_kernel_size); events->sctp_data_io_event = 1; events->sctp_association_event = 1; events->sctp_address_event = 1; events->sctp_send_failure_event = 1; events->sctp_peer_error_event = 1; events->sctp_shutdown_event = 1; handle_info->event_subscribe_buffer = (char *)events; log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Size of struct sctp_event_subscribe is %u in kernel, %zu in user space", handle_info->event_subscribe_kernel_size, sizeof(struct sctp_event_subscribe)); return 0; } int sctp_transport_init(knet_handle_t knet_h) { int err = 0, savederrno = 0; sctp_handle_info_t *handle_info; struct epoll_event ev; if (knet_h->transports[KNET_TRANSPORT_SCTP]) { errno = EEXIST; return -1; } handle_info = malloc(sizeof(sctp_handle_info_t)); if (!handle_info) { return -1; } memset(handle_info, 0,sizeof(sctp_handle_info_t)); knet_h->transports[KNET_TRANSPORT_SCTP] = handle_info; savederrno = _sctp_subscribe_init(knet_h); if (savederrno) { err = -1; goto exit_fail; } qb_list_init(&handle_info->listen_links_list); qb_list_init(&handle_info->connect_links_list); handle_info->listen_epollfd = epoll_create(KNET_EPOLL_MAX_EVENTS + 1); if (handle_info->listen_epollfd < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to create epoll listen fd: %s", strerror(savederrno)); goto exit_fail; } if (_fdset_cloexec(handle_info->listen_epollfd)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to set CLOEXEC on listen_epollfd: %s", strerror(savederrno)); goto exit_fail; } handle_info->connect_epollfd = epoll_create(KNET_EPOLL_MAX_EVENTS + 1); if (handle_info->connect_epollfd < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to create epoll connect fd: %s", strerror(savederrno)); goto exit_fail; } if (_fdset_cloexec(handle_info->connect_epollfd)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to set CLOEXEC on connect_epollfd: %s", strerror(savederrno)); goto exit_fail; } if (_init_socketpair(knet_h, handle_info->connectsockfd) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to init connect socketpair: %s", strerror(savederrno)); goto exit_fail; } memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = handle_info->connectsockfd[0]; if (epoll_ctl(handle_info->connect_epollfd, EPOLL_CTL_ADD, handle_info->connectsockfd[0], &ev)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to add connectsockfd[0] to connect epoll pool: %s", strerror(savederrno)); goto exit_fail; } if (_init_socketpair(knet_h, handle_info->listensockfd) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to init listen socketpair: %s", strerror(savederrno)); goto exit_fail; } memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = handle_info->listensockfd[0]; if (epoll_ctl(handle_info->listen_epollfd, EPOLL_CTL_ADD, handle_info->listensockfd[0], &ev)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to add listensockfd[0] to listen epoll pool: %s", strerror(savederrno)); goto exit_fail; } /* * Start connect & listener threads */ set_thread_status(knet_h, KNET_THREAD_SCTP_LISTEN, KNET_THREAD_REGISTERED); savederrno = pthread_create(&handle_info->listen_thread, 0, _sctp_listen_thread, (void *) knet_h); if (savederrno) { err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to start sctp listen thread: %s", strerror(savederrno)); goto exit_fail; } set_thread_status(knet_h, KNET_THREAD_SCTP_CONN, KNET_THREAD_REGISTERED); savederrno = pthread_create(&handle_info->connect_thread, 0, _sctp_connect_thread, (void *) knet_h); if (savederrno) { err = -1; log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to start sctp connect thread: %s", strerror(savederrno)); goto exit_fail; } exit_fail: if (err < 0) { sctp_transport_free(knet_h); } errno = savederrno; return err; } int sctp_transport_link_dyn_connect(knet_handle_t knet_h, int sockfd, struct knet_link *kn_link) { kn_link->outsock = sockfd; kn_link->status.dynconnected = 1; kn_link->transport_connected = 1; return 0; } #endif diff --git a/libknet/transport_udp.c b/libknet/transport_udp.c index 963340d4..32dd0323 100644 --- a/libknet/transport_udp.c +++ b/libknet/transport_udp.c @@ -1,440 +1,484 @@ /* * Copyright (C) 2016-2021 Red Hat, Inc. All rights reserved. * * Author: Christine Caulfield * * This software licensed under LGPL-2.0+ */ #include "config.h" #include #include #include #include #include #include #include #include #include #if defined (IP_RECVERR) || defined (IPV6_RECVERR) #include #endif #include "libknet.h" #include "compat.h" #include "host.h" #include "link.h" #include "logging.h" #include "common.h" #include "netutils.h" #include "transport_common.h" #include "transport_udp.h" #include "transports.h" #include "threads_common.h" typedef struct udp_handle_info { struct qb_list_head links_list; } udp_handle_info_t; typedef struct udp_link_info { struct qb_list_head list; struct sockaddr_storage local_address; int socket_fd; int on_epoll; } udp_link_info_t; int udp_transport_link_set_config(knet_handle_t knet_h, struct knet_link *kn_link) { int err = 0, savederrno = 0; int sock = -1; struct epoll_event ev; udp_link_info_t *info; udp_handle_info_t *handle_info = knet_h->transports[KNET_TRANSPORT_UDP]; #if defined (IP_RECVERR) || defined (IPV6_RECVERR) int value; #endif /* * Only allocate a new link if the local address is different */ qb_list_for_each_entry(info, &handle_info->links_list, list) { if (memcmp(&info->local_address, &kn_link->src_addr, sizeof(struct sockaddr_storage)) == 0) { log_debug(knet_h, KNET_SUB_TRANSP_UDP, "Re-using existing UDP socket for new link"); kn_link->outsock = info->socket_fd; kn_link->transport_link = info; kn_link->transport_connected = 1; return 0; } } info = malloc(sizeof(udp_link_info_t)); if (!info) { err = -1; goto exit_error; } memset(info, 0, sizeof(udp_link_info_t)); sock = socket(kn_link->src_addr.ss_family, SOCK_DGRAM, 0); if (sock < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_UDP, "Unable to create listener socket: %s", strerror(savederrno)); goto exit_error; } if (_configure_transport_socket(knet_h, sock, &kn_link->src_addr, kn_link->flags, "UDP") < 0) { savederrno = errno; err = -1; goto exit_error; } #ifdef IP_RECVERR if (kn_link->src_addr.ss_family == AF_INET) { value = 1; if (setsockopt(sock, SOL_IP, IP_RECVERR, &value, sizeof(value)) <0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_UDP, "Unable to set RECVERR on socket: %s", strerror(savederrno)); goto exit_error; } log_debug(knet_h, KNET_SUB_TRANSP_UDP, "IP_RECVERR enabled on socket: %i", sock); } #else log_debug(knet_h, KNET_SUB_TRANSP_UDP, "IP_RECVERR not available in this build/platform"); #endif #ifdef IPV6_RECVERR if (kn_link->src_addr.ss_family == AF_INET6) { value = 1; if (setsockopt(sock, SOL_IPV6, IPV6_RECVERR, &value, sizeof(value)) <0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_UDP, "Unable to set RECVERR on socket: %s", strerror(savederrno)); goto exit_error; } log_debug(knet_h, KNET_SUB_TRANSP_UDP, "IPV6_RECVERR enabled on socket: %i", sock); } #else log_debug(knet_h, KNET_SUB_TRANSP_UDP, "IPV6_RECVERR not available in this build/platform"); #endif if (bind(sock, (struct sockaddr *)&kn_link->src_addr, sockaddr_len(&kn_link->src_addr))) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_UDP, "Unable to bind listener socket: %s", strerror(savederrno)); goto exit_error; } memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = sock; if (epoll_ctl(knet_h->recv_from_links_epollfd, EPOLL_CTL_ADD, sock, &ev)) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_UDP, "Unable to add listener to epoll pool: %s", strerror(savederrno)); goto exit_error; } info->on_epoll = 1; if (_set_fd_tracker(knet_h, sock, KNET_TRANSPORT_UDP, 0, sockaddr_len(&kn_link->src_addr), info) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_UDP, "Unable to set fd tracker: %s", strerror(savederrno)); goto exit_error; } memmove(&info->local_address, &kn_link->src_addr, sizeof(struct sockaddr_storage)); info->socket_fd = sock; qb_list_add(&info->list, &handle_info->links_list); kn_link->outsock = sock; kn_link->transport_link = info; kn_link->transport_connected = 1; exit_error: if (err) { if (info) { if (info->on_epoll) { epoll_ctl(knet_h->recv_from_links_epollfd, EPOLL_CTL_DEL, sock, &ev); } free(info); } if (sock >= 0) { close(sock); } } errno = savederrno; return err; } int udp_transport_link_clear_config(knet_handle_t knet_h, struct knet_link *kn_link) { int err = 0, savederrno = 0; int found = 0; struct knet_host *host; int link_idx; udp_link_info_t *info = kn_link->transport_link; struct epoll_event ev; for (host = knet_h->host_head; host != NULL; host = host->next) { for (link_idx = 0; link_idx < KNET_MAX_LINK; link_idx++) { if (&host->link[link_idx] == kn_link) continue; if (host->link[link_idx].transport_link == info) { found = 1; break; } } } if (found) { log_debug(knet_h, KNET_SUB_TRANSP_UDP, "UDP socket %d still in use", info->socket_fd); savederrno = EBUSY; err = -1; goto exit_error; } if (info->on_epoll) { memset(&ev, 0, sizeof(struct epoll_event)); ev.events = EPOLLIN; ev.data.fd = info->socket_fd; if (epoll_ctl(knet_h->recv_from_links_epollfd, EPOLL_CTL_DEL, info->socket_fd, &ev) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_UDP, "Unable to remove UDP socket from epoll poll: %s", strerror(errno)); goto exit_error; } info->on_epoll = 0; } if (_set_fd_tracker(knet_h, info->socket_fd, KNET_MAX_TRANSPORTS, 0, sockaddr_len(&kn_link->src_addr), NULL) < 0) { savederrno = errno; err = -1; log_err(knet_h, KNET_SUB_TRANSP_UDP, "Unable to set fd tracker: %s", strerror(savederrno)); goto exit_error; } close(info->socket_fd); qb_list_del(&info->list); free(kn_link->transport_link); exit_error: errno = savederrno; return err; } int udp_transport_free(knet_handle_t knet_h) { udp_handle_info_t *handle_info; if (!knet_h->transports[KNET_TRANSPORT_UDP]) { errno = EINVAL; return -1; } handle_info = knet_h->transports[KNET_TRANSPORT_UDP]; /* * keep it here while we debug list usage and such */ if (!qb_list_empty(&handle_info->links_list)) { log_err(knet_h, KNET_SUB_TRANSP_UDP, "Internal error. handle list is not empty"); return -1; } free(handle_info); knet_h->transports[KNET_TRANSPORT_UDP] = NULL; return 0; } int udp_transport_init(knet_handle_t knet_h) { udp_handle_info_t *handle_info; if (knet_h->transports[KNET_TRANSPORT_UDP]) { errno = EEXIST; return -1; } handle_info = malloc(sizeof(udp_handle_info_t)); if (!handle_info) { return -1; } memset(handle_info, 0, sizeof(udp_handle_info_t)); knet_h->transports[KNET_TRANSPORT_UDP] = handle_info; qb_list_init(&handle_info->links_list); return 0; } #if defined (IP_RECVERR) || defined (IPV6_RECVERR) static int read_errs_from_sock(knet_handle_t knet_h, int sockfd) { int err = 0, savederrno = 0; int got_err = 0; char buffer[1024]; struct iovec iov; struct msghdr msg; struct cmsghdr *cmsg; struct sock_extended_err *sock_err; struct icmphdr icmph; struct sockaddr_storage remote; struct sockaddr_storage *origin; char addr_str[KNET_MAX_HOST_LEN]; char port_str[KNET_MAX_PORT_LEN]; char addr_remote_str[KNET_MAX_HOST_LEN]; char port_remote_str[KNET_MAX_PORT_LEN]; iov.iov_base = &icmph; iov.iov_len = sizeof(icmph); msg.msg_name = (void*)&remote; msg.msg_namelen = sizeof(remote); msg.msg_iov = &iov; msg.msg_iovlen = 1; msg.msg_flags = 0; msg.msg_control = buffer; msg.msg_controllen = sizeof(buffer); for (;;) { err = recvmsg(sockfd, &msg, MSG_ERRQUEUE); savederrno = errno; if (err < 0) { if (!got_err) { errno = savederrno; return -1; } else { return 0; } } got_err = 1; for (cmsg = CMSG_FIRSTHDR(&msg);cmsg; cmsg = CMSG_NXTHDR(&msg, cmsg)) { if (((cmsg->cmsg_level == SOL_IP) && (cmsg->cmsg_type == IP_RECVERR)) || ((cmsg->cmsg_level == SOL_IPV6 && (cmsg->cmsg_type == IPV6_RECVERR)))) { sock_err = (struct sock_extended_err*)(void *)CMSG_DATA(cmsg); if (sock_err) { switch (sock_err->ee_origin) { case SO_EE_ORIGIN_NONE: /* no origin */ case SO_EE_ORIGIN_LOCAL: /* local source (EMSGSIZE) */ if (sock_err->ee_errno == EMSGSIZE) { if (pthread_mutex_lock(&knet_h->kmtu_mutex) != 0) { log_debug(knet_h, KNET_SUB_TRANSP_UDP, "Unable to get mutex lock"); knet_h->kernel_mtu = 0; break; } else { knet_h->kernel_mtu = sock_err->ee_info; log_debug(knet_h, KNET_SUB_TRANSP_UDP, "detected kernel MTU: %u", knet_h->kernel_mtu); pthread_mutex_unlock(&knet_h->kmtu_mutex); } force_pmtud_run(knet_h, KNET_SUB_TRANSP_UDP, 0); } /* * those errors are way too noisy */ break; case SO_EE_ORIGIN_ICMP: /* ICMP */ case SO_EE_ORIGIN_ICMP6: /* ICMP6 */ origin = (struct sockaddr_storage *)(void *)SO_EE_OFFENDER(sock_err); if (knet_addrtostr(origin, sizeof(*origin), addr_str, KNET_MAX_HOST_LEN, port_str, KNET_MAX_PORT_LEN) < 0) { log_debug(knet_h, KNET_SUB_TRANSP_UDP, "Received ICMP error from unknown source: %s", strerror(sock_err->ee_errno)); } else { if (knet_addrtostr(&remote, sizeof(remote), addr_remote_str, KNET_MAX_HOST_LEN, port_remote_str, KNET_MAX_PORT_LEN) < 0) { log_debug(knet_h, KNET_SUB_TRANSP_UDP, "Received ICMP error from %s: %s destination unknown", addr_str, strerror(sock_err->ee_errno)); } else { log_debug(knet_h, KNET_SUB_TRANSP_UDP, "Received ICMP error from %s: %s %s", addr_str, strerror(sock_err->ee_errno), addr_remote_str); + if ((sock_err->ee_errno == ECONNREFUSED) || /* knet is not running on the other node */ + (sock_err->ee_errno == ECONNABORTED) || /* local kernel closed the socket */ + (sock_err->ee_errno == ENONET) || /* network does not exist */ + (sock_err->ee_errno == ENETUNREACH) || /* network unreachable */ + (sock_err->ee_errno == EHOSTUNREACH) || /* host unreachable */ + (sock_err->ee_errno == EHOSTDOWN) || /* host down (from kernel/net/ipv4/icmp.c */ + (sock_err->ee_errno == ENETDOWN)) { /* network down */ + struct knet_host *host = NULL; + struct knet_link *kn_link = NULL; + int link_idx, found = 0; + + for (host = knet_h->host_head; host != NULL; host = host->next) { + for (link_idx = 0; link_idx < KNET_MAX_LINK; link_idx++) { + kn_link = &host->link[link_idx]; + if (kn_link->outsock == sockfd) { + if (!cmpaddr(&remote, &kn_link->dst_addr)) { + found = 1; + break; + } + } + } + if (found) { + break; + } + } + + if ((host) && (kn_link) && + (kn_link->status.connected)) { + log_debug(knet_h, KNET_SUB_TRANSP_UDP, "Setting down host %u link %i", host->host_id, kn_link->link_id); + /* + * setting transport_connected = 0 will trigger + * thread_heartbeat link_down process. + * + * the process terminates calling into transport_link_down + * below that will set transport_connected = 1 + */ + kn_link->transport_connected = 0; + } + + } } } break; } } else { log_debug(knet_h, KNET_SUB_TRANSP_UDP, "No data in MSG_ERRQUEUE"); } } } } } #else static int read_errs_from_sock(knet_handle_t knet_h, int sockfd) { return 0; } #endif int udp_transport_rx_sock_error(knet_handle_t knet_h, int sockfd, int recv_err, int recv_errno) { if (recv_errno == EAGAIN) { read_errs_from_sock(knet_h, sockfd); } return 0; } transport_sock_error_t udp_transport_tx_sock_error(knet_handle_t knet_h, int sockfd, int recv_err, int recv_errno) { if (recv_err < 0) { if (recv_errno == EMSGSIZE) { read_errs_from_sock(knet_h, sockfd); return KNET_TRANSPORT_SOCK_ERROR_IGNORE; } if ((recv_errno == EINVAL) || (recv_errno == EPERM) || (recv_errno == ENETUNREACH) || (recv_errno == ENETDOWN)) { #ifdef DEBUG if ((recv_errno == ENETUNREACH) || (recv_errno == ENETDOWN)) { log_debug(knet_h, KNET_SUB_TRANSP_UDP, "Sock: %d is unreachable.", sockfd); } #endif return KNET_TRANSPORT_SOCK_ERROR_INTERNAL; } if ((recv_errno == ENOBUFS) || (recv_errno == EAGAIN)) { #ifdef DEBUG log_debug(knet_h, KNET_SUB_TRANSP_UDP, "Sock: %d is overloaded. Slowing TX down", sockfd); #endif usleep(knet_h->threads_timer_res / 16); } else { read_errs_from_sock(knet_h, sockfd); } return KNET_TRANSPORT_SOCK_ERROR_RETRY; } return KNET_TRANSPORT_SOCK_ERROR_IGNORE; } transport_rx_isdata_t udp_transport_rx_is_data(knet_handle_t knet_h, int sockfd, struct knet_mmsghdr *msg) { if (msg->msg_len == 0) return KNET_TRANSPORT_RX_NOT_DATA_CONTINUE; return KNET_TRANSPORT_RX_IS_DATA; } int udp_transport_link_dyn_connect(knet_handle_t knet_h, int sockfd, struct knet_link *kn_link) { kn_link->status.dynconnected = 1; return 0; } int udp_transport_link_is_down(knet_handle_t knet_h, struct knet_link *kn_link) { + /* + * see comments about handling ICMP error messages + */ + kn_link->transport_connected = 1; return 0; }