/* * Copyright 2023 Vsevolod Stakhov * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at * * http://www.apache.org/licenses/LICENSE-2.0 * * Unless required by applicable law or agreed to in writing, software * distributed under the License is distributed on an "AS IS" BASIS, * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. * See the License for the specific language governing permissions and * limitations under the License. */ #include "config.h" #include "upstream.h" #include "ottery.h" #include "ref.h" #include "cfg_file.h" #include "rdns.h" #include "cryptobox.h" #include "utlist.h" #include "contrib/libev/ev.h" #include "logger.h" #include "contrib/librdns/rdns.h" #include "contrib/mumhash/mum.h" #include struct upstream_inet_addr_entry { rspamd_inet_addr_t *addr; guint priority; struct upstream_inet_addr_entry *next; }; struct upstream_addr_elt { rspamd_inet_addr_t *addr; guint priority; guint errors; }; struct upstream_list_watcher { rspamd_upstream_watch_func func; GFreeFunc dtor; gpointer ud; enum rspamd_upstreams_watch_event events_mask; struct upstream_list_watcher *next, *prev; }; struct upstream { guint weight; guint cur_weight; guint errors; guint checked; guint dns_requests; gint active_idx; guint ttl; gchar *name; ev_timer ev; gdouble last_fail; gdouble last_resolve; gpointer ud; enum rspamd_upstream_flag flags; struct upstream_list *ls; GList *ctx_pos; struct upstream_ctx *ctx; struct { GPtrArray *addr; /* struct upstream_addr_elt */ guint cur; } addrs; struct upstream_inet_addr_entry *new_addrs; gpointer data; gchar uid[8]; ref_entry_t ref; #ifdef UPSTREAMS_THREAD_SAFE rspamd_mutex_t *lock; #endif }; struct upstream_limits { gdouble revive_time; gdouble revive_jitter; gdouble error_time; gdouble dns_timeout; gdouble lazy_resolve_time; guint max_errors; guint dns_retransmits; }; struct upstream_list { gchar *ups_line; struct upstream_ctx *ctx; GPtrArray *ups; GPtrArray *alive; struct upstream_list_watcher *watchers; guint64 hash_seed; const struct upstream_limits *limits; enum rspamd_upstream_flag flags; guint cur_elt; enum rspamd_upstream_rotation rot_alg; #ifdef UPSTREAMS_THREAD_SAFE rspamd_mutex_t *lock; #endif }; struct upstream_ctx { struct rdns_resolver *res; struct ev_loop *event_loop; struct upstream_limits limits; GQueue *upstreams; gboolean configured; rspamd_mempool_t *pool; ref_entry_t ref; }; #ifndef UPSTREAMS_THREAD_SAFE #define RSPAMD_UPSTREAM_LOCK(x) \ do { \ } while (0) #define RSPAMD_UPSTREAM_UNLOCK(x) \ do { \ } while (0) #else #define RSPAMD_UPSTREAM_LOCK(x) rspamd_mutex_lock(x->lock) #define RSPAMD_UPSTREAM_UNLOCK(x) rspamd_mutex_unlock(x->lock) #endif #define msg_debug_upstream(...) rspamd_conditional_debug_fast(NULL, NULL, \ rspamd_upstream_log_id, "upstream", upstream->uid, \ G_STRFUNC, \ __VA_ARGS__) #define msg_info_upstream(...) rspamd_default_log_function(G_LOG_LEVEL_INFO, \ "upstream", upstream->uid, \ G_STRFUNC, \ __VA_ARGS__) #define msg_err_upstream(...) rspamd_default_log_function(G_LOG_LEVEL_INFO, \ "upstream", upstream->uid, \ G_STRFUNC, \ __VA_ARGS__) INIT_LOG_MODULE(upstream) /* 4 errors in 10 seconds */ #define DEFAULT_MAX_ERRORS 4 static const guint default_max_errors = DEFAULT_MAX_ERRORS; #define DEFAULT_REVIVE_TIME 60 static const gdouble default_revive_time = DEFAULT_REVIVE_TIME; #define DEFAULT_REVIVE_JITTER 0.4 static const gdouble default_revive_jitter = DEFAULT_REVIVE_JITTER; #define DEFAULT_ERROR_TIME 10 static const gdouble default_error_time = DEFAULT_ERROR_TIME; #define DEFAULT_DNS_TIMEOUT 1.0 static const gdouble default_dns_timeout = DEFAULT_DNS_TIMEOUT; #define DEFAULT_DNS_RETRANSMITS 2 static const guint default_dns_retransmits = DEFAULT_DNS_RETRANSMITS; /* TODO: make it configurable */ #define DEFAULT_LAZY_RESOLVE_TIME 3600.0 static const gdouble default_lazy_resolve_time = DEFAULT_LAZY_RESOLVE_TIME; static const struct upstream_limits default_limits = { .revive_time = DEFAULT_REVIVE_TIME, .revive_jitter = DEFAULT_REVIVE_JITTER, .error_time = DEFAULT_ERROR_TIME, .dns_timeout = DEFAULT_DNS_TIMEOUT, .dns_retransmits = DEFAULT_DNS_RETRANSMITS, .max_errors = DEFAULT_MAX_ERRORS, .lazy_resolve_time = DEFAULT_LAZY_RESOLVE_TIME, }; static void rspamd_upstream_lazy_resolve_cb(struct ev_loop *, ev_timer *, int); void rspamd_upstreams_library_config(struct rspamd_config *cfg, struct upstream_ctx *ctx, struct ev_loop *event_loop, struct rdns_resolver *resolver) { g_assert(ctx != NULL); g_assert(cfg != NULL); if (cfg->upstream_error_time) { ctx->limits.error_time = cfg->upstream_error_time; } if (cfg->upstream_max_errors) { ctx->limits.max_errors = cfg->upstream_max_errors; } if (cfg->upstream_revive_time) { ctx->limits.revive_time = cfg->upstream_revive_time; } if (cfg->upstream_lazy_resolve_time) { ctx->limits.lazy_resolve_time = cfg->upstream_lazy_resolve_time; } if (cfg->dns_retransmits) { ctx->limits.dns_retransmits = cfg->dns_retransmits; } if (cfg->dns_timeout) { ctx->limits.dns_timeout = cfg->dns_timeout; } ctx->event_loop = event_loop; ctx->res = resolver; ctx->configured = TRUE; /* Start lazy resolving */ if (event_loop && resolver) { GList *cur; struct upstream *upstream; cur = ctx->upstreams->head; while (cur) { upstream = cur->data; if (!ev_can_stop(&upstream->ev) && upstream->ls && !(upstream->flags & RSPAMD_UPSTREAM_FLAG_NORESOLVE)) { gdouble when; if (upstream->flags & RSPAMD_UPSTREAM_FLAG_SRV_RESOLVE) { /* Resolve them immediately ! */ when = 0.0; } else { when = rspamd_time_jitter(upstream->ls->limits->lazy_resolve_time, upstream->ls->limits->lazy_resolve_time * .1); } ev_timer_init(&upstream->ev, rspamd_upstream_lazy_resolve_cb, when, 0); upstream->ev.data = upstream; ev_timer_start(ctx->event_loop, &upstream->ev); } cur = g_list_next(cur); } } } static void rspamd_upstream_ctx_dtor(struct upstream_ctx *ctx) { GList *cur; struct upstream *u; cur = ctx->upstreams->head; while (cur) { u = cur->data; u->ctx = NULL; u->ctx_pos = NULL; cur = g_list_next(cur); } g_queue_free(ctx->upstreams); rspamd_mempool_delete(ctx->pool); g_free(ctx); } void rspamd_upstreams_library_unref(struct upstream_ctx *ctx) { REF_RELEASE(ctx); } struct upstream_ctx * rspamd_upstreams_library_init(void) { struct upstream_ctx *ctx; ctx = g_malloc0(sizeof(*ctx)); memcpy(&ctx->limits, &default_limits, sizeof(ctx->limits)); ctx->pool = rspamd_mempool_new(rspamd_mempool_suggest_size(), "upstreams", 0); ctx->upstreams = g_queue_new(); REF_INIT_RETAIN(ctx, rspamd_upstream_ctx_dtor); return ctx; } static gint rspamd_upstream_af_to_weight(const rspamd_inet_addr_t *addr) { int ret; switch (rspamd_inet_address_get_af(addr)) { case AF_UNIX: ret = 2; break; case AF_INET: ret = 1; break; default: ret = 0; break; } return ret; } /* * Select IPv4 addresses before IPv6 */ static gint rspamd_upstream_addr_sort_func(gconstpointer a, gconstpointer b) { const struct upstream_addr_elt *ip1 = *(const struct upstream_addr_elt **) a, *ip2 = *(const struct upstream_addr_elt **) b; gint w1, w2; if (ip1->priority == 0 && ip2->priority == 0) { w1 = rspamd_upstream_af_to_weight(ip1->addr); w2 = rspamd_upstream_af_to_weight(ip2->addr); } else { w1 = ip1->priority; w2 = ip2->priority; } /* Inverse order */ return w2 - w1; } static void rspamd_upstream_set_active(struct upstream_list *ls, struct upstream *upstream) { RSPAMD_UPSTREAM_LOCK(ls); g_ptr_array_add(ls->alive, upstream); upstream->active_idx = ls->alive->len - 1; if (upstream->ctx && upstream->ctx->configured && !(upstream->flags & RSPAMD_UPSTREAM_FLAG_NORESOLVE)) { if (ev_can_stop(&upstream->ev)) { ev_timer_stop(upstream->ctx->event_loop, &upstream->ev); } /* Start lazy (or not so lazy) names resolution */ gdouble when; if (upstream->flags & RSPAMD_UPSTREAM_FLAG_SRV_RESOLVE) { /* Resolve them immediately ! */ when = 0.0; } else { when = rspamd_time_jitter(upstream->ls->limits->lazy_resolve_time, upstream->ls->limits->lazy_resolve_time * .1); } ev_timer_init(&upstream->ev, rspamd_upstream_lazy_resolve_cb, when, 0); upstream->ev.data = upstream; msg_debug_upstream("start lazy resolving for %s in %.0f seconds", upstream->name, when); ev_timer_start(upstream->ctx->event_loop, &upstream->ev); } RSPAMD_UPSTREAM_UNLOCK(ls); } static void rspamd_upstream_addr_elt_dtor(gpointer a) { struct upstream_addr_elt *elt = a; if (elt) { rspamd_inet_address_free(elt->addr); g_free(elt); } } static void rspamd_upstream_update_addrs(struct upstream *upstream) { guint addr_cnt, i, port; gboolean seen_addr, reset_errors = FALSE; struct upstream_inet_addr_entry *cur, *tmp; GPtrArray *new_addrs; struct upstream_addr_elt *addr_elt, *naddr; /* * We need first of all get the saved port, since DNS gives us no * idea about what port has been used previously */ RSPAMD_UPSTREAM_LOCK(upstream); if (upstream->addrs.addr->len > 0 && upstream->new_addrs) { addr_elt = g_ptr_array_index(upstream->addrs.addr, 0); port = rspamd_inet_address_get_port(addr_elt->addr); /* Now calculate new addrs count */ addr_cnt = 0; LL_FOREACH(upstream->new_addrs, cur) { addr_cnt++; } /* At 10% probability reset errors on addr elements */ if (rspamd_random_double_fast() > 0.9) { reset_errors = TRUE; msg_debug_upstream("reset errors on upstream %s", upstream->name); } new_addrs = g_ptr_array_new_full(addr_cnt, rspamd_upstream_addr_elt_dtor); /* Copy addrs back */ LL_FOREACH(upstream->new_addrs, cur) { seen_addr = FALSE; naddr = NULL; /* Ports are problematic, set to compare in the next block */ rspamd_inet_address_set_port(cur->addr, port); PTR_ARRAY_FOREACH(upstream->addrs.addr, i, addr_elt) { if (rspamd_inet_address_compare(addr_elt->addr, cur->addr, FALSE) == 0) { naddr = g_malloc0(sizeof(*naddr)); naddr->addr = cur->addr; naddr->errors = reset_errors ? 0 : addr_elt->errors; seen_addr = TRUE; break; } } if (!seen_addr) { naddr = g_malloc0(sizeof(*naddr)); naddr->addr = cur->addr; naddr->errors = 0; msg_debug_upstream("new address for %s: %s", upstream->name, rspamd_inet_address_to_string_pretty(naddr->addr)); } else { msg_debug_upstream("existing address for %s: %s", upstream->name, rspamd_inet_address_to_string_pretty(cur->addr)); } g_ptr_array_add(new_addrs, naddr); } /* Free old addresses */ g_ptr_array_free(upstream->addrs.addr, TRUE); upstream->addrs.cur = 0; upstream->addrs.addr = new_addrs; g_ptr_array_sort(upstream->addrs.addr, rspamd_upstream_addr_sort_func); } LL_FOREACH_SAFE(upstream->new_addrs, cur, tmp) { /* Do not free inet address pointer since it has been transferred to up */ g_free(cur); } upstream->new_addrs = NULL; RSPAMD_UPSTREAM_UNLOCK(upstream); } static void rspamd_upstream_dns_cb(struct rdns_reply *reply, void *arg) { struct upstream *up = (struct upstream *) arg; struct rdns_reply_entry *entry; struct upstream_inet_addr_entry *up_ent; if (reply->code == RDNS_RC_NOERROR) { entry = reply->entries; RSPAMD_UPSTREAM_LOCK(up); while (entry) { if (entry->type == RDNS_REQUEST_A) { up_ent = g_malloc0(sizeof(*up_ent)); up_ent->addr = rspamd_inet_address_new(AF_INET, &entry->content.a.addr); LL_PREPEND(up->new_addrs, up_ent); } else if (entry->type == RDNS_REQUEST_AAAA) { up_ent = g_malloc0(sizeof(*up_ent)); up_ent->addr = rspamd_inet_address_new(AF_INET6, &entry->content.aaa.addr); LL_PREPEND(up->new_addrs, up_ent); } entry = entry->next; } RSPAMD_UPSTREAM_UNLOCK(up); } up->dns_requests--; if (up->dns_requests == 0) { rspamd_upstream_update_addrs(up); } REF_RELEASE(up); } struct rspamd_upstream_srv_dns_cb { struct upstream *up; guint priority; guint port; guint requests_inflight; }; /* Used when we have resolved SRV record and resolved addrs */ static void rspamd_upstream_dns_srv_phase2_cb(struct rdns_reply *reply, void *arg) { struct rspamd_upstream_srv_dns_cb *cbdata = (struct rspamd_upstream_srv_dns_cb *) arg; struct upstream *up; struct rdns_reply_entry *entry; struct upstream_inet_addr_entry *up_ent; up = cbdata->up; if (reply->code == RDNS_RC_NOERROR) { entry = reply->entries; RSPAMD_UPSTREAM_LOCK(up); while (entry) { if (entry->type == RDNS_REQUEST_A) { up_ent = g_malloc0(sizeof(*up_ent)); up_ent->addr = rspamd_inet_address_new(AF_INET, &entry->content.a.addr); up_ent->priority = cbdata->priority; rspamd_inet_address_set_port(up_ent->addr, cbdata->port); LL_PREPEND(up->new_addrs, up_ent); } else if (entry->type == RDNS_REQUEST_AAAA) { up_ent = g_malloc0(sizeof(*up_ent)); up_ent->addr = rspamd_inet_address_new(AF_INET6, &entry->content.aaa.addr); up_ent->priority = cbdata->priority; rspamd_inet_address_set_port(up_ent->addr, cbdata->port); LL_PREPEND(up->new_addrs, up_ent); } entry = entry->next; } RSPAMD_UPSTREAM_UNLOCK(up); } up->dns_requests--; cbdata->requests_inflight--; if (cbdata->requests_inflight == 0) { g_free(cbdata); } if (up->dns_requests == 0) { rspamd_upstream_update_addrs(up); } REF_RELEASE(up); } static void rspamd_upstream_dns_srv_cb(struct rdns_reply *reply, void *arg) { struct upstream *upstream = (struct upstream *) arg; struct rdns_reply_entry *entry; struct rspamd_upstream_srv_dns_cb *ncbdata; if (reply->code == RDNS_RC_NOERROR) { entry = reply->entries; RSPAMD_UPSTREAM_LOCK(upstream); while (entry) { /* XXX: we ignore weight as it contradicts with upstreams logic */ if (entry->type == RDNS_REQUEST_SRV) { msg_debug_upstream("got srv reply for %s: %s " "(weight=%d, priority=%d, port=%d)", upstream->name, entry->content.srv.target, entry->content.srv.weight, entry->content.srv.priority, entry->content.srv.port); ncbdata = g_malloc0(sizeof(*ncbdata)); ncbdata->priority = entry->content.srv.weight; ncbdata->port = entry->content.srv.port; /* XXX: for all entries? */ upstream->ttl = entry->ttl; if (rdns_make_request_full(upstream->ctx->res, rspamd_upstream_dns_srv_phase2_cb, ncbdata, upstream->ls->limits->dns_timeout, upstream->ls->limits->dns_retransmits, 1, entry->content.srv.target, RDNS_REQUEST_A) != NULL) { upstream->dns_requests++; REF_RETAIN(upstream); ncbdata->requests_inflight++; } if (rdns_make_request_full(upstream->ctx->res, rspamd_upstream_dns_srv_phase2_cb, ncbdata, upstream->ls->limits->dns_timeout, upstream->ls->limits->dns_retransmits, 1, entry->content.srv.target, RDNS_REQUEST_AAAA) != NULL) { upstream->dns_requests++; REF_RETAIN(upstream); ncbdata->requests_inflight++; } if (ncbdata->requests_inflight == 0) { g_free(ncbdata); } } entry = entry->next; } RSPAMD_UPSTREAM_UNLOCK(upstream); } upstream->dns_requests--; REF_RELEASE(upstream); } static void rspamd_upstream_revive_cb(struct ev_loop *loop, ev_timer *w, int revents) { struct upstream *upstream = (struct upstream *) w->data; RSPAMD_UPSTREAM_LOCK(upstream); ev_timer_stop(loop, w); msg_debug_upstream("revive upstream %s", upstream->name); if (upstream->ls) { rspamd_upstream_set_active(upstream->ls, upstream); } RSPAMD_UPSTREAM_UNLOCK(upstream); g_assert(upstream->ref.refcount > 1); REF_RELEASE(upstream); } static void rspamd_upstream_resolve_addrs(const struct upstream_list *ls, struct upstream *upstream) { /* XXX: maybe make it configurable */ static const gdouble min_resolve_interval = 60.0; if (upstream->ctx->res != NULL && upstream->ctx->configured && upstream->dns_requests == 0 && !(upstream->flags & RSPAMD_UPSTREAM_FLAG_NORESOLVE)) { gdouble now = ev_now(upstream->ctx->event_loop); if (now - upstream->last_resolve < min_resolve_interval) { msg_info_upstream("do not resolve upstream %s as it was checked %.0f " "seconds ago (%.0f is minimum)", upstream->name, now - upstream->last_resolve, min_resolve_interval); return; } /* Resolve name of the upstream one more time */ if (upstream->name[0] != '/') { upstream->last_resolve = now; /* * If upstream name has a port, then we definitely need to resolve * merely host part! */ char dns_name[253 + 1]; /* 253 == max dns name + \0 */ const char *semicolon_pos = strchr(upstream->name, ':'); if (semicolon_pos != NULL && semicolon_pos > upstream->name) { if (sizeof(dns_name) > semicolon_pos - upstream->name) { rspamd_strlcpy(dns_name, upstream->name, semicolon_pos - upstream->name + 1); } else { /* XXX: truncated */ msg_err_upstream("internal error: upstream name is larger than" "max DNS name: %s", upstream->name); rspamd_strlcpy(dns_name, upstream->name, sizeof(dns_name)); } } else { rspamd_strlcpy(dns_name, upstream->name, sizeof(dns_name)); } if (upstream->flags & RSPAMD_UPSTREAM_FLAG_SRV_RESOLVE) { if (rdns_make_request_full(upstream->ctx->res, rspamd_upstream_dns_srv_cb, upstream, ls->limits->dns_timeout, ls->limits->dns_retransmits, 1, dns_name, RDNS_REQUEST_SRV) != NULL) { upstream->dns_requests++; REF_RETAIN(upstream); } } else { if (rdns_make_request_full(upstream->ctx->res, rspamd_upstream_dns_cb, upstream, ls->limits->dns_timeout, ls->limits->dns_retransmits, 1, dns_name, RDNS_REQUEST_A) != NULL) { upstream->dns_requests++; REF_RETAIN(upstream); } if (rdns_make_request_full(upstream->ctx->res, rspamd_upstream_dns_cb, upstream, ls->limits->dns_timeout, ls->limits->dns_retransmits, 1, dns_name, RDNS_REQUEST_AAAA) != NULL) { upstream->dns_requests++; REF_RETAIN(upstream); } } } } else if (upstream->dns_requests != 0) { msg_info_upstream("do not resolve upstream %s as another request for " "resolving has been already issued", upstream->name); } } static void rspamd_upstream_lazy_resolve_cb(struct ev_loop *loop, ev_timer *w, int revents) { struct upstream *up = (struct upstream *) w->data; RSPAMD_UPSTREAM_LOCK(up); ev_timer_stop(loop, w); if (up->ls) { rspamd_upstream_resolve_addrs(up->ls, up); if (up->ttl == 0 || up->ttl > up->ls->limits->lazy_resolve_time) { w->repeat = rspamd_time_jitter(up->ls->limits->lazy_resolve_time, up->ls->limits->lazy_resolve_time * .1); } else { w->repeat = up->ttl; } ev_timer_again(loop, w); } RSPAMD_UPSTREAM_UNLOCK(up); } static void rspamd_upstream_set_inactive(struct upstream_list *ls, struct upstream *upstream) { gdouble ntim; guint i; struct upstream *cur; struct upstream_list_watcher *w; RSPAMD_UPSTREAM_LOCK(ls); g_ptr_array_remove_index(ls->alive, upstream->active_idx); upstream->active_idx = -1; /* We need to update all indices */ for (i = 0; i < ls->alive->len; i++) { cur = g_ptr_array_index(ls->alive, i); cur->active_idx = i; } if (upstream->ctx) { rspamd_upstream_resolve_addrs(ls, upstream); REF_RETAIN(upstream); ntim = rspamd_time_jitter(ls->limits->revive_time, ls->limits->revive_time * ls->limits->revive_jitter); if (ev_can_stop(&upstream->ev)) { ev_timer_stop(upstream->ctx->event_loop, &upstream->ev); } msg_debug_upstream("mark upstream %s inactive; revive in %.0f seconds", upstream->name, ntim); ev_timer_init(&upstream->ev, rspamd_upstream_revive_cb, ntim, 0); upstream->ev.data = upstream; if (upstream->ctx->event_loop != NULL && upstream->ctx->configured) { ev_timer_start(upstream->ctx->event_loop, &upstream->ev); } } DL_FOREACH(upstream->ls->watchers, w) { if (w->events_mask & RSPAMD_UPSTREAM_WATCH_OFFLINE) { w->func(upstream, RSPAMD_UPSTREAM_WATCH_OFFLINE, upstream->errors, w->ud); } } RSPAMD_UPSTREAM_UNLOCK(ls); } void rspamd_upstream_fail(struct upstream *upstream, gboolean addr_failure, const gchar *reason) { gdouble error_rate = 0, max_error_rate = 0; gdouble sec_last, sec_cur; struct upstream_addr_elt *addr_elt; struct upstream_list_watcher *w; msg_debug_upstream("upstream %s failed; reason: %s", upstream->name, reason); if (upstream->ctx && upstream->active_idx != -1 && upstream->ls) { sec_cur = rspamd_get_ticks(FALSE); RSPAMD_UPSTREAM_LOCK(upstream); if (upstream->errors == 0) { /* We have the first error */ upstream->last_fail = sec_cur; upstream->errors = 1; if (upstream->ls && upstream->dns_requests == 0) { /* Try to re-resolve address immediately */ rspamd_upstream_resolve_addrs(upstream->ls, upstream); } DL_FOREACH(upstream->ls->watchers, w) { if (w->events_mask & RSPAMD_UPSTREAM_WATCH_FAILURE) { w->func(upstream, RSPAMD_UPSTREAM_WATCH_FAILURE, 1, w->ud); } } } else { sec_last = upstream->last_fail; if (sec_cur >= sec_last) { upstream->errors++; DL_FOREACH(upstream->ls->watchers, w) { if (w->events_mask & RSPAMD_UPSTREAM_WATCH_FAILURE) { w->func(upstream, RSPAMD_UPSTREAM_WATCH_FAILURE, upstream->errors, w->ud); } } if (sec_cur - sec_last >= upstream->ls->limits->error_time) { error_rate = ((gdouble) upstream->errors) / (sec_cur - sec_last); max_error_rate = ((gdouble) upstream->ls->limits->max_errors) / upstream->ls->limits->error_time; } if (error_rate > max_error_rate) { /* Remove upstream from the active list */ if (upstream->ls->ups->len > 1) { msg_debug_upstream("mark upstream %s inactive; " "reason: %s; %.2f " "error rate (%d errors), " "%.2f max error rate, " "%.1f first error time, " "%.1f current ts, " "%d upstreams left", upstream->name, reason, error_rate, upstream->errors, max_error_rate, sec_last, sec_cur, upstream->ls->alive->len - 1); rspamd_upstream_set_inactive(upstream->ls, upstream); upstream->errors = 0; } else { msg_debug_upstream("cannot mark last alive upstream %s " "inactive; reason: %s; %.2f " "error rate (%d errors), " "%.2f max error rate, " "%.1f first error time, " "%.1f current ts", upstream->name, reason, error_rate, upstream->errors, max_error_rate, sec_last, sec_cur); /* Just re-resolve addresses */ if (sec_cur - sec_last > upstream->ls->limits->revive_time) { upstream->errors = 0; rspamd_upstream_resolve_addrs(upstream->ls, upstream); } } } else if (sec_cur - sec_last >= upstream->ls->limits->error_time) { /* Forget the whole interval */ upstream->last_fail = sec_cur; upstream->errors = 1; } } } if (addr_failure) { /* Also increase count of errors for this specific address */ if (upstream->addrs.addr) { addr_elt = g_ptr_array_index(upstream->addrs.addr, upstream->addrs.cur); addr_elt->errors++; } } RSPAMD_UPSTREAM_UNLOCK(upstream); } } void rspamd_upstream_ok(struct upstream *upstream) { struct upstream_addr_elt *addr_elt; struct upstream_list_watcher *w; RSPAMD_UPSTREAM_LOCK(upstream); if (upstream->errors > 0 && upstream->active_idx != -1 && upstream->ls) { /* We touch upstream if and only if it is active */ msg_debug_upstream("reset errors on upstream %s (was %ud)", upstream->name, upstream->errors); upstream->errors = 0; if (upstream->addrs.addr) { addr_elt = g_ptr_array_index(upstream->addrs.addr, upstream->addrs.cur); addr_elt->errors = 0; } DL_FOREACH(upstream->ls->watchers, w) { if (w->events_mask & RSPAMD_UPSTREAM_WATCH_SUCCESS) { w->func(upstream, RSPAMD_UPSTREAM_WATCH_SUCCESS, 0, w->ud); } } } RSPAMD_UPSTREAM_UNLOCK(upstream); } void rspamd_upstream_set_weight(struct upstream *up, guint weight) { RSPAMD_UPSTREAM_LOCK(up); up->weight = weight; RSPAMD_UPSTREAM_UNLOCK(up); } #define SEED_CONSTANT 0xa574de7df64e9b9dULL struct upstream_list * rspamd_upstreams_create(struct upstream_ctx *ctx) { struct upstream_list *ls; ls = g_malloc0(sizeof(*ls)); ls->hash_seed = SEED_CONSTANT; ls->ups = g_ptr_array_new(); ls->alive = g_ptr_array_new(); #ifdef UPSTREAMS_THREAD_SAFE ls->lock = rspamd_mutex_new(); #endif ls->cur_elt = 0; ls->ctx = ctx; ls->rot_alg = RSPAMD_UPSTREAM_UNDEF; if (ctx) { ls->limits = &ctx->limits; } else { ls->limits = &default_limits; } return ls; } gsize rspamd_upstreams_count(struct upstream_list *ups) { return ups != NULL ? ups->ups->len : 0; } gsize rspamd_upstreams_alive(struct upstream_list *ups) { return ups != NULL ? ups->alive->len : 0; } static void rspamd_upstream_dtor(struct upstream *up) { struct upstream_inet_addr_entry *cur, *tmp; if (up->new_addrs) { LL_FOREACH_SAFE(up->new_addrs, cur, tmp) { /* Here we need to free pointer as well */ rspamd_inet_address_free(cur->addr); g_free(cur); } } if (up->addrs.addr) { g_ptr_array_free(up->addrs.addr, TRUE); } #ifdef UPSTREAMS_THREAD_SAFE rspamd_mutex_free(up->lock); #endif if (up->ctx) { if (ev_can_stop(&up->ev)) { ev_timer_stop(up->ctx->event_loop, &up->ev); } g_queue_delete_link(up->ctx->upstreams, up->ctx_pos); REF_RELEASE(up->ctx); } g_free(up); } rspamd_inet_addr_t * rspamd_upstream_addr_next(struct upstream *up) { guint idx, next_idx; struct upstream_addr_elt *e1, *e2; do { idx = up->addrs.cur; next_idx = (idx + 1) % up->addrs.addr->len; e1 = g_ptr_array_index(up->addrs.addr, idx); e2 = g_ptr_array_index(up->addrs.addr, next_idx); up->addrs.cur = next_idx; } while (e2->errors > e1->errors); return e2->addr; } rspamd_inet_addr_t * rspamd_upstream_addr_cur(const struct upstream *up) { struct upstream_addr_elt *elt; elt = g_ptr_array_index(up->addrs.addr, up->addrs.cur); return elt->addr; } const gchar * rspamd_upstream_name(struct upstream *up) { return up->name; } gint rspamd_upstream_port(struct upstream *up) { struct upstream_addr_elt *elt; elt = g_ptr_array_index(up->addrs.addr, up->addrs.cur); return rspamd_inet_address_get_port(elt->addr); } gboolean rspamd_upstreams_add_upstream(struct upstream_list *ups, const gchar *str, guint16 def_port, enum rspamd_upstream_parse_type parse_type, void *data) { struct upstream *upstream; GPtrArray *addrs = NULL; guint i, slen; rspamd_inet_addr_t *addr; enum rspamd_parse_host_port_result ret = RSPAMD_PARSE_ADDR_FAIL; upstream = g_malloc0(sizeof(*upstream)); slen = strlen(str); switch (parse_type) { case RSPAMD_UPSTREAM_PARSE_DEFAULT: if (slen > sizeof("service=") && RSPAMD_LEN_CHECK_STARTS_WITH(str, slen, "service=")) { const gchar *plus_pos, *service_pos, *semicolon_pos; /* Accept service=srv_name+hostname[:priority] */ service_pos = str + sizeof("service=") - 1; plus_pos = strchr(service_pos, '+'); if (plus_pos != NULL) { semicolon_pos = strchr(plus_pos + 1, ':'); if (semicolon_pos) { upstream->weight = strtoul(semicolon_pos + 1, NULL, 10); } else { semicolon_pos = plus_pos + strlen(plus_pos); } /* * Now our name is _service._tcp. * where is string between semicolon_pos and plus_pos +1 * while service is a string between service_pos and plus_pos */ guint namelen = (semicolon_pos - (plus_pos + 1)) + (plus_pos - service_pos) + (sizeof("tcp") - 1) + 4; addrs = g_ptr_array_sized_new(1); upstream->name = ups->ctx ? rspamd_mempool_alloc(ups->ctx->pool, namelen + 1) : g_malloc(namelen + 1); rspamd_snprintf(upstream->name, namelen + 1, "_%*s._tcp.%*s", (gint) (plus_pos - service_pos), service_pos, (gint) (semicolon_pos - (plus_pos + 1)), plus_pos + 1); upstream->flags |= RSPAMD_UPSTREAM_FLAG_SRV_RESOLVE; ret = RSPAMD_PARSE_ADDR_RESOLVED; } } else { ret = rspamd_parse_host_port_priority(str, &addrs, &upstream->weight, &upstream->name, def_port, FALSE, ups->ctx ? ups->ctx->pool : NULL); } break; case RSPAMD_UPSTREAM_PARSE_NAMESERVER: addrs = g_ptr_array_sized_new(1); if (rspamd_parse_inet_address(&addr, str, strlen(str), RSPAMD_INET_ADDRESS_PARSE_DEFAULT)) { if (ups->ctx) { upstream->name = rspamd_mempool_strdup(ups->ctx->pool, str); } else { upstream->name = g_strdup(str); } if (rspamd_inet_address_get_port(addr) == 0) { rspamd_inet_address_set_port(addr, def_port); } g_ptr_array_add(addrs, addr); ret = RSPAMD_PARSE_ADDR_NUMERIC; if (ups->ctx) { rspamd_mempool_add_destructor(ups->ctx->pool, (rspamd_mempool_destruct_t) rspamd_inet_address_free, addr); rspamd_mempool_add_destructor(ups->ctx->pool, (rspamd_mempool_destruct_t) rspamd_ptr_array_free_hard, addrs); } } else { g_ptr_array_free(addrs, TRUE); } break; } if (ret == RSPAMD_PARSE_ADDR_FAIL) { g_free(upstream); return FALSE; } else { upstream->flags |= ups->flags; if (ret == RSPAMD_PARSE_ADDR_NUMERIC) { /* Add noresolve flag */ upstream->flags |= RSPAMD_UPSTREAM_FLAG_NORESOLVE; } for (i = 0; i < addrs->len; i++) { addr = g_ptr_array_index(addrs, i); rspamd_upstream_add_addr(upstream, rspamd_inet_address_copy(addr, NULL)); } } if (upstream->weight == 0 && ups->rot_alg == RSPAMD_UPSTREAM_MASTER_SLAVE) { /* Special heuristic for master-slave rotation */ if (ups->ups->len == 0) { /* Prioritize the first */ upstream->weight = 1; } } g_ptr_array_add(ups->ups, upstream); upstream->ud = data; upstream->cur_weight = upstream->weight; upstream->ls = ups; REF_INIT_RETAIN(upstream, rspamd_upstream_dtor); #ifdef UPSTREAMS_THREAD_SAFE upstream->lock = rspamd_mutex_new(); #endif upstream->ctx = ups->ctx; if (upstream->ctx) { REF_RETAIN(ups->ctx); g_queue_push_tail(ups->ctx->upstreams, upstream); upstream->ctx_pos = g_queue_peek_tail_link(ups->ctx->upstreams); } guint h = rspamd_cryptobox_fast_hash(upstream->name, strlen(upstream->name), 0); memset(upstream->uid, 0, sizeof(upstream->uid)); rspamd_encode_base32_buf((const guchar *) &h, sizeof(h), upstream->uid, sizeof(upstream->uid) - 1, RSPAMD_BASE32_DEFAULT); msg_debug_upstream("added upstream %s (%s)", upstream->name, upstream->flags & RSPAMD_UPSTREAM_FLAG_NORESOLVE ? "numeric ip" : "DNS name"); g_ptr_array_sort(upstream->addrs.addr, rspamd_upstream_addr_sort_func); rspamd_upstream_set_active(ups, upstream); return TRUE; } void rspamd_upstreams_set_flags(struct upstream_list *ups, enum rspamd_upstream_flag flags) { ups->flags = flags; } void rspamd_upstreams_set_rotation(struct upstream_list *ups, enum rspamd_upstream_rotation rot) { ups->rot_alg = rot; } gboolean rspamd_upstream_add_addr(struct upstream *up, rspamd_inet_addr_t *addr) { struct upstream_addr_elt *elt; /* * XXX: slow and inefficient */ if (up->addrs.addr == NULL) { up->addrs.addr = g_ptr_array_new_full(8, rspamd_upstream_addr_elt_dtor); } elt = g_malloc0(sizeof(*elt)); elt->addr = addr; g_ptr_array_add(up->addrs.addr, elt); g_ptr_array_sort(up->addrs.addr, rspamd_upstream_addr_sort_func); return TRUE; } gboolean rspamd_upstreams_parse_line_len(struct upstream_list *ups, const gchar *str, gsize len, guint16 def_port, void *data) { const gchar *end = str + len, *p = str; const gchar *separators = ";, \n\r\t"; gchar *tmp; guint span_len; gboolean ret = FALSE; if (RSPAMD_LEN_CHECK_STARTS_WITH(p, len, "random:")) { ups->rot_alg = RSPAMD_UPSTREAM_RANDOM; p += sizeof("random:") - 1; } else if (RSPAMD_LEN_CHECK_STARTS_WITH(p, len, "master-slave:")) { ups->rot_alg = RSPAMD_UPSTREAM_MASTER_SLAVE; p += sizeof("master-slave:") - 1; } else if (RSPAMD_LEN_CHECK_STARTS_WITH(p, len, "round-robin:")) { ups->rot_alg = RSPAMD_UPSTREAM_ROUND_ROBIN; p += sizeof("round-robin:") - 1; } else if (RSPAMD_LEN_CHECK_STARTS_WITH(p, len, "hash:")) { ups->rot_alg = RSPAMD_UPSTREAM_HASHED; p += sizeof("hash:") - 1; } while (p < end) { span_len = rspamd_memcspn(p, separators, end - p); if (span_len > 0) { tmp = g_malloc(span_len + 1); rspamd_strlcpy(tmp, p, span_len + 1); if (rspamd_upstreams_add_upstream(ups, tmp, def_port, RSPAMD_UPSTREAM_PARSE_DEFAULT, data)) { ret = TRUE; } g_free(tmp); } p += span_len; /* Skip separators */ if (p < end) { p += rspamd_memspn(p, separators, end - p); } } if (!ups->ups_line) { ups->ups_line = g_malloc(len + 1); rspamd_strlcpy(ups->ups_line, str, len + 1); } return ret; } gboolean rspamd_upstreams_parse_line(struct upstream_list *ups, const gchar *str, guint16 def_port, void *data) { return rspamd_upstreams_parse_line_len(ups, str, strlen(str), def_port, data); } gboolean rspamd_upstreams_from_ucl(struct upstream_list *ups, const ucl_object_t *in, guint16 def_port, void *data) { gboolean ret = FALSE; const ucl_object_t *cur; ucl_object_iter_t it = NULL; it = ucl_object_iterate_new(in); while ((cur = ucl_object_iterate_safe(it, true)) != NULL) { if (ucl_object_type(cur) == UCL_STRING) { ret = rspamd_upstreams_parse_line(ups, ucl_object_tostring(cur), def_port, data); } } ucl_object_iterate_free(it); return ret; } void rspamd_upstreams_destroy(struct upstream_list *ups) { guint i; struct upstream *up; struct upstream_list_watcher *w, *tmp; if (ups != NULL) { g_ptr_array_free(ups->alive, TRUE); for (i = 0; i < ups->ups->len; i++) { up = g_ptr_array_index(ups->ups, i); up->ls = NULL; REF_RELEASE(up); } DL_FOREACH_SAFE(ups->watchers, w, tmp) { if (w->dtor) { w->dtor(w->ud); } g_free(w); } g_free(ups->ups_line); g_ptr_array_free(ups->ups, TRUE); #ifdef UPSTREAMS_THREAD_SAFE rspamd_mutex_free(ups->lock); #endif g_free(ups); } } static void rspamd_upstream_restore_cb(gpointer elt, gpointer ls) { struct upstream *up = (struct upstream *) elt; struct upstream_list *ups = (struct upstream_list *) ls; struct upstream_list_watcher *w; /* Here the upstreams list is already locked */ RSPAMD_UPSTREAM_LOCK(up); if (ev_can_stop(&up->ev)) { ev_timer_stop(up->ctx->event_loop, &up->ev); } g_ptr_array_add(ups->alive, up); up->active_idx = ups->alive->len - 1; RSPAMD_UPSTREAM_UNLOCK(up); DL_FOREACH(up->ls->watchers, w) { if (w->events_mask & RSPAMD_UPSTREAM_WATCH_ONLINE) { w->func(up, RSPAMD_UPSTREAM_WATCH_ONLINE, up->errors, w->ud); } } /* For revive event */ g_assert(up->ref.refcount > 1); REF_RELEASE(up); } static struct upstream * rspamd_upstream_get_random(struct upstream_list *ups, struct upstream *except) { for (;;) { guint idx = ottery_rand_range(ups->alive->len - 1); struct upstream *up; up = g_ptr_array_index(ups->alive, idx); if (except && up == except) { continue; } return up; } } static struct upstream * rspamd_upstream_get_round_robin(struct upstream_list *ups, struct upstream *except, gboolean use_cur) { guint max_weight = 0, min_checked = G_MAXUINT; struct upstream *up = NULL, *selected = NULL, *min_checked_sel = NULL; guint i; /* Select upstream with the maximum cur_weight */ RSPAMD_UPSTREAM_LOCK(ups); for (i = 0; i < ups->alive->len; i++) { up = g_ptr_array_index(ups->alive, i); if (except != NULL && up == except) { continue; } if (use_cur) { if (up->cur_weight > max_weight) { selected = up; max_weight = up->cur_weight; } } else { if (up->weight > max_weight) { selected = up; max_weight = up->weight; } } /* * This code is used when all upstreams have zero weight * The logic is to select least currently used upstream and penalise * upstream with errors. The error penalty should no be too high * to avoid sudden traffic drop in this case. */ if (up->checked + up->errors * 2 < min_checked) { min_checked_sel = up; min_checked = up->checked; } } if (max_weight == 0) { /* All upstreams have zero weight */ if (min_checked > G_MAXUINT / 2) { /* Reset all checked counters to avoid overflow */ for (i = 0; i < ups->alive->len; i++) { up = g_ptr_array_index(ups->alive, i); up->checked = 0; } } selected = min_checked_sel; } if (use_cur && selected) { if (selected->cur_weight > 0) { selected->cur_weight--; } else { selected->cur_weight = selected->weight; } } RSPAMD_UPSTREAM_UNLOCK(ups); return selected; } /* * The key idea of this function is obtained from the following paper: * A Fast, Minimal Memory, Consistent Hash Algorithm * John Lamping, Eric Veach * * http://arxiv.org/abs/1406.2294 */ static guint32 rspamd_consistent_hash(guint64 key, guint32 nbuckets) { gint64 b = -1, j = 0; while (j < nbuckets) { b = j; key *= 2862933555777941757ULL + 1; j = (b + 1) * (double) (1ULL << 31) / (double) ((key >> 33) + 1ULL); } return b; } static struct upstream * rspamd_upstream_get_hashed(struct upstream_list *ups, struct upstream *except, const guint8 *key, guint keylen) { guint64 k; guint32 idx; static const guint max_tries = 20; struct upstream *up = NULL; /* Generate 64 bits input key */ k = rspamd_cryptobox_fast_hash_specific(RSPAMD_CRYPTOBOX_XXHASH64, key, keylen, ups->hash_seed); RSPAMD_UPSTREAM_LOCK(ups); /* * Select new upstream from all upstreams */ for (guint i = 0; i < max_tries; i++) { idx = rspamd_consistent_hash(k, ups->ups->len); up = g_ptr_array_index(ups->ups, idx); if (up->active_idx < 0 || (except != NULL && up == except)) { /* Found inactive or excluded upstream */ k = mum_hash_step(k, ups->hash_seed); } else { break; } } RSPAMD_UPSTREAM_UNLOCK(ups); if (up->active_idx >= 0) { return up; } /* We failed to find any active upstream */ up = rspamd_upstream_get_random(ups, except); msg_info("failed to find hashed upstream for %s, fallback to random: %s", ups->ups_line, up->name); return up; } static struct upstream * rspamd_upstream_get_common(struct upstream_list *ups, struct upstream *except, enum rspamd_upstream_rotation default_type, const guchar *key, gsize keylen, gboolean forced) { enum rspamd_upstream_rotation type; struct upstream *up = NULL; RSPAMD_UPSTREAM_LOCK(ups); if (ups->alive->len == 0) { /* We have no upstreams alive */ msg_warn("there are no alive upstreams left for %s, revive all of them", ups->ups_line); g_ptr_array_foreach(ups->ups, rspamd_upstream_restore_cb, ups); } RSPAMD_UPSTREAM_UNLOCK(ups); if (ups->alive->len == 1 && default_type != RSPAMD_UPSTREAM_SEQUENTIAL) { /* Fast path */ up = g_ptr_array_index(ups->alive, 0); goto end; } if (!forced) { type = ups->rot_alg != RSPAMD_UPSTREAM_UNDEF ? ups->rot_alg : default_type; } else { type = default_type != RSPAMD_UPSTREAM_UNDEF ? default_type : ups->rot_alg; } if (type == RSPAMD_UPSTREAM_HASHED && (keylen == 0 || key == NULL)) { /* Cannot use hashed rotation when no key is specified, switch to random */ type = RSPAMD_UPSTREAM_RANDOM; } switch (type) { default: case RSPAMD_UPSTREAM_RANDOM: up = rspamd_upstream_get_random(ups, except); break; case RSPAMD_UPSTREAM_HASHED: up = rspamd_upstream_get_hashed(ups, except, key, keylen); break; case RSPAMD_UPSTREAM_ROUND_ROBIN: up = rspamd_upstream_get_round_robin(ups, except, TRUE); break; case RSPAMD_UPSTREAM_MASTER_SLAVE: up = rspamd_upstream_get_round_robin(ups, except, FALSE); break; case RSPAMD_UPSTREAM_SEQUENTIAL: if (ups->cur_elt >= ups->alive->len) { ups->cur_elt = 0; return NULL; } up = g_ptr_array_index(ups->alive, ups->cur_elt++); break; } end: if (up) { up->checked++; } return up; } struct upstream * rspamd_upstream_get(struct upstream_list *ups, enum rspamd_upstream_rotation default_type, const guchar *key, gsize keylen) { return rspamd_upstream_get_common(ups, NULL, default_type, key, keylen, FALSE); } struct upstream * rspamd_upstream_get_forced(struct upstream_list *ups, enum rspamd_upstream_rotation forced_type, const guchar *key, gsize keylen) { return rspamd_upstream_get_common(ups, NULL, forced_type, key, keylen, TRUE); } struct upstream *rspamd_upstream_get_except(struct upstream_list *ups, struct upstream *except, enum rspamd_upstream_rotation default_type, const guchar *key, gsize keylen) { return rspamd_upstream_get_common(ups, except, default_type, key, keylen, FALSE); } void rspamd_upstream_reresolve(struct upstream_ctx *ctx) { GList *cur; struct upstream *up; cur = ctx->upstreams->head; while (cur) { up = cur->data; REF_RETAIN(up); rspamd_upstream_resolve_addrs(up->ls, up); REF_RELEASE(up); cur = g_list_next(cur); } } gpointer rspamd_upstream_set_data(struct upstream *up, gpointer data) { gpointer prev_data = up->data; up->data = data; return prev_data; } gpointer rspamd_upstream_get_data(struct upstream *up) { return up->data; } void rspamd_upstreams_foreach(struct upstream_list *ups, rspamd_upstream_traverse_func cb, void *ud) { struct upstream *up; guint i; for (i = 0; i < ups->ups->len; i++) { up = g_ptr_array_index(ups->ups, i); cb(up, i, ud); } } void rspamd_upstreams_set_limits(struct upstream_list *ups, gdouble revive_time, gdouble revive_jitter, gdouble error_time, gdouble dns_timeout, guint max_errors, guint dns_retransmits) { struct upstream_limits *nlimits; g_assert(ups != NULL); nlimits = rspamd_mempool_alloc(ups->ctx->pool, sizeof(*nlimits)); memcpy(nlimits, ups->limits, sizeof(*nlimits)); if (!isnan(revive_time)) { nlimits->revive_time = revive_time; } if (!isnan(revive_jitter)) { nlimits->revive_jitter = revive_jitter; } if (!isnan(error_time)) { nlimits->error_time = error_time; } if (!isnan(dns_timeout)) { nlimits->dns_timeout = dns_timeout; } if (max_errors > 0) { nlimits->max_errors = max_errors; } if (dns_retransmits > 0) { nlimits->dns_retransmits = dns_retransmits; } ups->limits = nlimits; } void rspamd_upstreams_add_watch_callback(struct upstream_list *ups, enum rspamd_upstreams_watch_event events, rspamd_upstream_watch_func func, GFreeFunc dtor, gpointer ud) { struct upstream_list_watcher *nw; g_assert((events & RSPAMD_UPSTREAM_WATCH_ALL) != 0); nw = g_malloc(sizeof(*nw)); nw->func = func; nw->events_mask = events; nw->ud = ud; nw->dtor = dtor; DL_APPEND(ups->watchers, nw); } struct upstream * rspamd_upstream_ref(struct upstream *up) { REF_RETAIN(up); return up; } void rspamd_upstream_unref(struct upstream *up) { REF_RELEASE(up); }