/* * Copyright 2014-2015 Con Kolivas * * This program is free software; you can redistribute it and/or modify it * under the terms of the GNU General Public License as published by the Free * Software Foundation; either version 3 of the License, or (at your option) * any later version. See COPYING for more details. */ #include "config.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "ckpool.h" #include "libckpool.h" #include "generator.h" #include "stratifier.h" #include "connector.h" ckpool_t *global_ckp; static void proclog(ckpool_t *ckp, char *msg) { FILE *LOGFP; int logfd; if (unlikely(!msg)) { fprintf(stderr, "Proclog received null message"); return; } if (unlikely(!strlen(msg))) { fprintf(stderr, "Proclog received zero length message"); free(msg); return; } LOGFP = ckp->logfp; logfd = ckp->logfd; flock(logfd, LOCK_EX); fprintf(LOGFP, "%s", msg); flock(logfd, LOCK_UN); free(msg); } /* Log everything to the logfile, but display warnings on the console as well */ void logmsg(int loglevel, const char *fmt, ...) { if (global_ckp->loglevel >= loglevel && fmt) { int logfd = global_ckp->logfd; char *buf = NULL; struct tm tm; time_t now_t; va_list ap; char stamp[128]; va_start(ap, fmt); VASPRINTF(&buf, fmt, ap); va_end(ap); now_t = time(NULL); localtime_r(&now_t, &tm); sprintf(stamp, "[%d-%02d-%02d %02d:%02d:%02d]", tm.tm_year + 1900, tm.tm_mon + 1, tm.tm_mday, tm.tm_hour, tm.tm_min, tm.tm_sec); if (loglevel <= LOG_WARNING) {\ if (loglevel <= LOG_ERR && errno != 0) fprintf(stderr, "%s %s with errno %d: %s\n", stamp, buf, errno, strerror(errno)); else fprintf(stderr, "%s %s\n", stamp, buf); fflush(stderr); } if (logfd) { char *msg; if (loglevel <= LOG_ERR && errno != 0) ASPRINTF(&msg, "%s %s with errno %d: %s\n", stamp, buf, errno, strerror(errno)); else ASPRINTF(&msg, "%s %s\n", stamp, buf); ckmsgq_add(global_ckp->logger, msg); } free(buf); } } /* Generic function for creating a message queue receiving and parsing thread */ static void *ckmsg_queue(void *arg) { ckmsgq_t *ckmsgq = (ckmsgq_t *)arg; ckpool_t *ckp = ckmsgq->ckp; pthread_detach(pthread_self()); rename_proc(ckmsgq->name); while (42) { ckmsg_t *msg; tv_t now; ts_t abs; mutex_lock(ckmsgq->lock); tv_time(&now); tv_to_ts(&abs, &now); abs.tv_sec++; if (!ckmsgq->msgs) pthread_cond_timedwait(ckmsgq->cond, ckmsgq->lock, &abs); msg = ckmsgq->msgs; if (msg) DL_DELETE(ckmsgq->msgs, msg); mutex_unlock(ckmsgq->lock); if (!msg) continue; ckmsgq->func(ckp, msg->data); free(msg); } return NULL; } ckmsgq_t *create_ckmsgq(ckpool_t *ckp, const char *name, const void *func) { ckmsgq_t *ckmsgq = ckzalloc(sizeof(ckmsgq_t)); strncpy(ckmsgq->name, name, 15); ckmsgq->func = func; ckmsgq->ckp = ckp; ckmsgq->lock = ckalloc(sizeof(pthread_mutex_t)); ckmsgq->cond = ckalloc(sizeof(pthread_cond_t)); mutex_init(ckmsgq->lock); cond_init(ckmsgq->cond); create_pthread(&ckmsgq->pth, ckmsg_queue, ckmsgq); return ckmsgq; } ckmsgq_t *create_ckmsgqs(ckpool_t *ckp, const char *name, const void *func, const int count) { ckmsgq_t *ckmsgq = ckzalloc(sizeof(ckmsgq_t) * count); pthread_mutex_t *lock; pthread_cond_t *cond; int i; lock = ckalloc(sizeof(pthread_mutex_t)); cond = ckalloc(sizeof(pthread_cond_t)); mutex_init(lock); cond_init(cond); for (i = 0; i < count; i++) { snprintf(ckmsgq[i].name, 15, "%.8s%x", name, i); ckmsgq[i].func = func; ckmsgq[i].ckp = ckp; ckmsgq[i].lock = lock; ckmsgq[i].cond = cond; create_pthread(&ckmsgq[i].pth, ckmsg_queue, &ckmsgq[i]); } return ckmsgq; } /* Generic function for adding messages to a ckmsgq linked list and signal the * ckmsgq parsing thread to wake up and process it. */ void ckmsgq_add(ckmsgq_t *ckmsgq, void *data) { ckmsg_t *msg = ckalloc(sizeof(ckmsg_t)); msg->data = data; mutex_lock(ckmsgq->lock); ckmsgq->messages++; DL_APPEND(ckmsgq->msgs, msg); pthread_cond_signal(ckmsgq->cond); mutex_unlock(ckmsgq->lock); } /* Return whether there are any messages queued in the ckmsgq linked list. */ bool ckmsgq_empty(ckmsgq_t *ckmsgq) { bool ret = true; mutex_lock(ckmsgq->lock); if (ckmsgq->msgs) ret = (ckmsgq->msgs->next == ckmsgq->msgs->prev); mutex_unlock(ckmsgq->lock); return ret; } static void broadcast_proc(ckpool_t *ckp, const char *buf) { int i; for (i = 0; i < ckp->proc_instances; i++) { proc_instance_t *pi = ckp->children[i]; send_proc(pi, buf); } } /* Put a sanity check on kill calls to make sure we are not sending them to * pid 0. */ static int kill_pid(const int pid, const int sig) { if (pid < 1) return -1; return kill(pid, sig); } static int pid_wait(const pid_t pid, const int ms) { tv_t start, now; int ret; tv_time(&start); do { ret = kill_pid(pid, 0); if (ret) break; tv_time(&now); } while (ms_tvdiff(&now, &start) < ms); return ret; } static int send_procmsg(const proc_instance_t *pi, const char *buf) { char *path = pi->us.path; int ret = -1; int sockd; if (unlikely(!path || !strlen(path))) { LOGERR("Attempted to send message %s to null path in send_proc", buf ? buf : ""); goto out; } if (unlikely(!buf || !strlen(buf))) { LOGERR("Attempted to send null message to socket %s in send_proc", path); goto out; } if (unlikely(kill_pid(pi->pid, 0))) { LOGALERT("Attempting to send message %s to dead process %s", buf, pi->processname); goto out; } sockd = open_unix_client(path); if (unlikely(sockd < 0)) { LOGWARNING("Failed to open socket %s in send_recv_proc", path); goto out; } if (unlikely(!send_unix_msg(sockd, buf))) LOGWARNING("Failed to send %s to socket %s", buf, path); else ret = sockd; out: if (unlikely(ret == -1)) LOGERR("Failure in send_procmsg"); return ret; } /* Listen for incoming global requests. Always returns a response if possible */ static void *listener(void *arg) { proc_instance_t *pi = (proc_instance_t *)arg; unixsock_t *us = &pi->us; ckpool_t *ckp = pi->ckp; char *buf = NULL, *msg; int sockd; rename_proc(pi->sockname); retry: dealloc(buf); sockd = accept(us->sockd, NULL, NULL); if (sockd < 0) { LOGERR("Failed to accept on socket in listener"); goto out; } buf = recv_unix_msg(sockd); if (!buf) { LOGWARNING("Failed to get message in listener"); send_unix_msg(sockd, "failed"); } else if (cmdmatch(buf, "shutdown")) { LOGWARNING("Listener received shutdown message, terminating ckpool"); send_unix_msg(sockd, "exiting"); goto out; } else if (cmdmatch(buf, "ping")) { LOGDEBUG("Listener received ping request"); send_unix_msg(sockd, "pong"); } else if (cmdmatch(buf, "loglevel")) { int loglevel; if (sscanf(buf, "loglevel=%d", &loglevel) != 1) { LOGWARNING("Failed to parse loglevel message %s", buf); send_unix_msg(sockd, "Failed"); } else if (loglevel < LOG_EMERG || loglevel > LOG_DEBUG) { LOGWARNING("Invalid loglevel %d sent", loglevel); send_unix_msg(sockd, "Invalid"); } else { ckp->loglevel = loglevel; broadcast_proc(ckp, buf); send_unix_msg(sockd, "success"); } } else if (cmdmatch(buf, "getxfd")) { int connfd = send_procmsg(ckp->connector, buf); if (connfd > 0) { int newfd = get_fd(connfd); if (newfd > 0) { LOGDEBUG("Sending new fd %d", newfd); send_fd(newfd, sockd); Close(newfd); } else LOGWARNING("Failed to get_fd"); Close(connfd); } else LOGWARNING("Failed to send_procmsg to connector"); } else if (cmdmatch(buf, "accept")) { LOGWARNING("Listener received accept message, accepting clients"); send_procmsg(ckp->connector, "accept"); send_unix_msg(sockd, "accepting"); } else if (cmdmatch(buf, "reject")) { LOGWARNING("Listener received reject message, rejecting clients"); send_procmsg(ckp->connector, "reject"); send_unix_msg(sockd, "rejecting"); } else if (cmdmatch(buf, "reconnect")) { LOGWARNING("Listener received request to send reconnect to clients"); send_procmsg(ckp->stratifier, buf); send_unix_msg(sockd, "reconnecting"); } else if (cmdmatch(buf, "restart")) { LOGWARNING("Listener received restart message, attempting handover"); send_unix_msg(sockd, "restarting"); if (!fork()) { if (!ckp->handover) { ckp->initial_args[ckp->args++] = strdup("-H"); ckp->initial_args[ckp->args] = NULL; } execv(ckp->initial_args[0], (char *const *)ckp->initial_args); } } else if (cmdmatch(buf, "stratifierstats")) { LOGDEBUG("Listener received stratifierstats request"); msg = send_recv_proc(ckp->stratifier, "stats"); send_unix_msg(sockd, msg); } else if (cmdmatch(buf, "connectorstats")) { LOGDEBUG("Listener received connectorstats request"); msg = send_recv_proc(ckp->connector, "stats"); send_unix_msg(sockd, msg); } else { LOGINFO("Listener received unhandled message: %s", buf); send_unix_msg(sockd, "unknown"); } Close(sockd); goto retry; out: dealloc(buf); close_unix_socket(us->sockd, us->path); return NULL; } bool ping_main(ckpool_t *ckp) { char *buf; if (unlikely(kill_pid(ckp->main.pid, 0))) return false; buf = send_recv_proc(&ckp->main, "ping"); if (unlikely(!buf)) return false; free(buf); return true; } void empty_buffer(connsock_t *cs) { cs->buflen = cs->bufofs = 0; } /* Read from a socket into cs->buf till we get an '\n', converting it to '\0' * and storing how much extra data we've received, to be moved to the beginning * of the buffer for use on the next receive. */ int read_socket_line(connsock_t *cs, const int timeout) { int fd = cs->fd, ret = -1; char *eom = NULL; size_t buflen; if (unlikely(fd < 0)) goto out; if (unlikely(!cs->buf)) cs->buf = ckzalloc(PAGESIZE); else if (cs->buflen) { memmove(cs->buf, cs->buf + cs->bufofs, cs->buflen); memset(cs->buf + cs->buflen, 0, cs->bufofs); cs->bufofs = cs->buflen; cs->buflen = 0; cs->buf[cs->bufofs] = '\0'; eom = strchr(cs->buf, '\n'); } while (42) { char readbuf[PAGESIZE] = {}; int backoff = 1; char *newbuf; ret = wait_read_select(fd, eom ? 0 : timeout); if (eom && !ret) break; if (ret < 1) { if (!ret) LOGDEBUG("Select timed out in read_socket_line"); else LOGERR("Select failed in read_socket_line"); goto out; } ret = recv(fd, readbuf, PAGESIZE - 4, 0); if (ret < 1) { /* Closed socket after valid message */ if (!ret && eom) break; LOGERR("Failed to recv in read_socket_line"); ret = -1; goto out; } buflen = cs->bufofs + ret + 1; while (42) { newbuf = realloc(cs->buf, buflen); if (likely(newbuf)) break; if (backoff == 1) fprintf(stderr, "Failed to realloc %d in read_socket_line, retrying\n", (int)buflen); cksleep_ms(backoff); backoff <<= 1; } cs->buf = newbuf; if (unlikely(!cs->buf)) quit(1, "Failed to alloc buf of %d bytes in read_socket_line", (int)buflen); memcpy(cs->buf + cs->bufofs, readbuf, ret); cs->bufofs += ret; cs->buf[cs->bufofs] = '\0'; eom = strchr(cs->buf, '\n'); } ret = eom - cs->buf; cs->buflen = cs->buf + cs->bufofs - eom - 1; if (cs->buflen) cs->bufofs = eom - cs->buf + 1; else cs->bufofs = 0; *eom = '\0'; out: if (ret < 0) { empty_buffer(cs); dealloc(cs->buf); Close(cs->fd); } return ret; } static void childsighandler(const int sig); static int get_proc_pid(const proc_instance_t *pi) { int ret, pid = 0; char path[256]; FILE *fp; sprintf(path, "%s%s.pid", pi->ckp->socket_dir, pi->processname); fp = fopen(path, "re"); if (!fp) goto out; ret = fscanf(fp, "%d", &pid); if (ret < 1) pid = 0; fclose(fp); out: return pid; } /* Send a single message to a process instance when there will be no response, * closing the socket immediately. */ bool _send_proc(proc_instance_t *pi, const char *msg, const char *file, const char *func, const int line) { char *path = pi->us.path; bool ret = false; int sockd; if (unlikely(!path || !strlen(path))) { LOGERR("Attempted to send message %s to null path in send_proc", msg ? msg : ""); goto out; } if (unlikely(!msg || !strlen(msg))) { LOGERR("Attempted to send null message to socket %s in send_proc", path); goto out; } /* At startup the pid fields are not set up before some processes are * forked so they never inherit them. */ if (unlikely(!pi->pid)) pi->pid = get_proc_pid(pi); if (unlikely(kill_pid(pi->pid, 0))) { LOGALERT("Attempting to send message %s to non existent process %s", msg, pi->processname); goto out; } sockd = open_unix_client(path); if (unlikely(sockd < 0)) { LOGWARNING("Failed to open socket %s", path); goto out; } if (unlikely(!send_unix_msg(sockd, msg))) LOGWARNING("Failed to send %s to socket %s", msg, path); else ret = true; Close(sockd); out: if (unlikely(!ret)) { LOGERR("Failure in send_proc from %s %s:%d", file, func, line); childsighandler(15); } return ret; } /* Send a single message to a process instance and retrieve the response, then * close the socket. */ char *_send_recv_proc(proc_instance_t *pi, const char *msg, const char *file, const char *func, const int line) { char *path = pi->us.path, *buf = NULL; int sockd; if (unlikely(!path || !strlen(path))) { LOGERR("Attempted to send message %s to null path in send_proc", msg ? msg : ""); goto out; } if (unlikely(!msg || !strlen(msg))) { LOGERR("Attempted to send null message to socket %s in send_proc", path); goto out; } if (unlikely(kill_pid(pi->pid, 0))) { LOGALERT("Attempting to send message %s to dead process %s", msg, pi->processname); goto out; } sockd = open_unix_client(path); if (unlikely(sockd < 0)) { LOGWARNING("Failed to open socket %s in send_recv_proc", path); goto out; } if (unlikely(!send_unix_msg(sockd, msg))) LOGWARNING("Failed to send %s to socket %s", msg, path); else buf = recv_unix_msg(sockd); Close(sockd); out: if (unlikely(!buf)) LOGERR("Failure in send_recv_proc from %s %s:%d", file, func, line); return buf; } /* As send_recv_proc but only to ckdb */ char *_send_recv_ckdb(const ckpool_t *ckp, const char *msg, const char *file, const char *func, const int line) { const char *path = ckp->ckdb_sockname; char *buf = NULL; int sockd; if (unlikely(!path || !strlen(path))) { LOGERR("Attempted to send message %s to null path in send_recv_ckdb", msg ? msg : ""); goto out; } if (unlikely(!msg || !strlen(msg))) { LOGERR("Attempted to send null message to ckdb in send_recv_ckdb"); goto out; } sockd = open_unix_client(path); if (unlikely(sockd < 0)) { LOGWARNING("Failed to open socket %s in send_recv_ckdb", path); goto out; } if (unlikely(!send_unix_msg(sockd, msg))) LOGWARNING("Failed to send %s to ckdb", msg); else buf = recv_unix_msg(sockd); Close(sockd); out: if (unlikely(!buf)) LOGERR("Failure in send_recv_ckdb from %s %s:%d", file, func, line); return buf; } /* Send a json msg to ckdb and return the response */ char *_ckdb_msg_call(const ckpool_t *ckp, const char *msg, const char *file, const char *func, const int line) { char *buf = NULL; LOGDEBUG("Sending ckdb: %s", msg); buf = _send_recv_ckdb(ckp, msg, file, func, line); LOGDEBUG("Received from ckdb: %s", buf); return buf; } json_t *json_rpc_call(connsock_t *cs, const char *rpc_req) { char *http_req = NULL; json_error_t err_val; json_t *val = NULL; int len, ret; if (unlikely(cs->fd < 0)) { LOGWARNING("FD %d invalid in json_rpc_call", cs->fd); goto out; } if (unlikely(!cs->url)) { LOGWARNING("No URL in json_rpc_call"); goto out; } if (unlikely(!cs->port)) { LOGWARNING("No port in json_rpc_call"); goto out; } if (unlikely(!cs->auth)) { LOGWARNING("No auth in json_rpc_call"); goto out; } if (unlikely(!rpc_req)) { LOGWARNING("Null rpc_req passed to json_rpc_call"); goto out; } len = strlen(rpc_req); if (unlikely(!len)) { LOGWARNING("Zero length rpc_req passed to json_rpc_call"); goto out; } http_req = ckalloc(len + 256); // Leave room for headers sprintf(http_req, "POST / HTTP/1.1\n" "Authorization: Basic %s\n" "Host: %s:%s\n" "Content-type: application/json\n" "Content-Length: %d\n\n%s", cs->auth, cs->url, cs->port, len, rpc_req); len = strlen(http_req); ret = write_socket(cs->fd, http_req, len); if (ret != len) { LOGWARNING("Failed to write to socket in json_rpc_call"); goto out_empty; } ret = read_socket_line(cs, 5); if (ret < 1) { LOGWARNING("Failed to read socket line in json_rpc_call"); goto out_empty; } if (strncasecmp(cs->buf, "HTTP/1.1 200 OK", 15)) { LOGWARNING("HTTP response not ok: %s", cs->buf); goto out_empty; } do { ret = read_socket_line(cs, 5); if (ret < 1) { LOGWARNING("Failed to read http socket lines in json_rpc_call"); goto out_empty; } } while (strncmp(cs->buf, "{", 1)); val = json_loads(cs->buf, 0, &err_val); if (!val) LOGWARNING("JSON decode failed(%d): %s", err_val.line, err_val.text); out_empty: empty_socket(cs->fd); empty_buffer(cs); if (!val) { /* Assume that a failed request means the socket will be closed * and reopen it */ LOGWARNING("Reopening socket to %s:%s", cs->url, cs->port); Close(cs->fd); cs->fd = connect_socket(cs->url, cs->port); } out: free(http_req); dealloc(cs->buf); return val; } /* Open the file in path, check if there is a pid in there that still exists * and if not, write the pid into that file. */ static bool write_pid(ckpool_t *ckp, const char *path, const pid_t pid) { struct stat statbuf; FILE *fp; int ret; if (!stat(path, &statbuf)) { int oldpid; LOGNOTICE("File %s exists", path); fp = fopen(path, "re"); if (!fp) { LOGEMERG("Failed to open file %s", path); return false; } ret = fscanf(fp, "%d", &oldpid); fclose(fp); if (ret == 1 && !(kill_pid(oldpid, 0))) { if (ckp->handover) { if (pid_wait(oldpid, 500)) goto out; LOGWARNING("Old process pid %d failed to shutdown cleanly, terminating", oldpid); } if (!ckp->killold) { LOGEMERG("Process %s pid %d still exists, start ckpool with -k if you wish to kill it", path, oldpid); return false; } if (kill_pid(oldpid, 15)) { LOGEMERG("Unable to kill old process %s pid %d", path, oldpid); return false; } LOGWARNING("Terminating old process %s pid %d", path, oldpid); if (pid_wait(oldpid, 500)) goto out; if (kill_pid(oldpid, 9)) { LOGEMERG("Unable to kill old process %s pid %d", path, oldpid); return false; } LOGWARNING("Unable to terminate old process %s pid %d, killing", path, oldpid); if (!pid_wait(oldpid, 500)) { LOGEMERG("Unable to kill old process %s pid %d", path, oldpid); return false; } } } out: fp = fopen(path, "we"); if (!fp) { LOGERR("Failed to open file %s", path); return false; } fprintf(fp, "%d", pid); fclose(fp); return true; } static void name_process_sockname(unixsock_t *us, const proc_instance_t *pi) { us->path = strdup(pi->ckp->socket_dir); realloc_strcat(&us->path, pi->sockname); } static void open_process_sock(ckpool_t *ckp, const proc_instance_t *pi, unixsock_t *us) { LOGDEBUG("Opening %s", us->path); us->sockd = open_unix_server(us->path); if (unlikely(us->sockd < 0)) quit(1, "Failed to open %s socket", pi->sockname); if (chown(us->path, -1, ckp->gr_gid)) quit(1, "Failed to set %s to group id %d", us->path, ckp->gr_gid); } static void create_process_unixsock(proc_instance_t *pi) { unixsock_t *us = &pi->us; ckpool_t *ckp = pi->ckp; name_process_sockname(us, pi); open_process_sock(ckp, pi, us); } static void write_namepid(proc_instance_t *pi) { char s[256]; pi->pid = getpid(); sprintf(s, "%s%s.pid", pi->ckp->socket_dir, pi->processname); if (!write_pid(pi->ckp, s, pi->pid)) quit(1, "Failed to write %s pid %d", pi->processname, pi->pid); } static void rm_namepid(const proc_instance_t *pi) { char s[256]; sprintf(s, "%s%s.pid", pi->ckp->socket_dir, pi->processname); unlink(s); } /* Disable signal handlers for child processes, but simply pass them onto the * parent process to shut down cleanly. */ static void childsighandler(const int sig) { signal(sig, SIG_IGN); signal(SIGTERM, SIG_IGN); if (sig != SIGUSR1) { LOGWARNING("Child process received signal %d, forwarding signal to %s main process", sig, global_ckp->name); kill_pid(global_ckp->main.pid, sig); } exit(0); } static void launch_logger(const proc_instance_t *pi) { ckpool_t *ckp = pi->ckp; char loggername[16]; /* Note that the logger is unique per process so it is the only value * in ckp that differs between processes */ snprintf(loggername, 15, "%clogger", pi->processname[0]); ckp->logger = create_ckmsgq(ckp, loggername, &proclog); } static void launch_process(proc_instance_t *pi) { pid_t pid; pid = fork(); if (pid < 0) quit(1, "Failed to fork %s in launch_process", pi->processname); if (!pid) { struct sigaction handler; int ret; json_set_alloc_funcs(json_ckalloc, free); launch_logger(pi); handler.sa_handler = &childsighandler; handler.sa_flags = 0; sigemptyset(&handler.sa_mask); sigaction(SIGUSR1, &handler, NULL); sigaction(SIGTERM, &handler, NULL); signal(SIGINT, SIG_IGN); signal(SIGQUIT, SIG_IGN); rename_proc(pi->processname); write_namepid(pi); ret = pi->process(pi); close_unix_socket(pi->us.sockd, pi->us.path); rm_namepid(pi); exit(ret); } pi->pid = pid; } static void launch_processes(const ckpool_t *ckp) { int i; for (i = 0; i < ckp->proc_instances; i++) launch_process(ckp->children[i]); } int process_exit(ckpool_t *ckp, const proc_instance_t *pi, int ret) { if (ret) { /* Abnormal termination, kill entire process */ LOGWARNING("%s %s exiting with return code %d, shutting down!", ckp->name, pi->processname, ret); send_proc(&ckp->main, "shutdown"); cksleep_ms(100); ret = 1; } else /* Should be part of a normal shutdown */ LOGNOTICE("%s %s exited normally", ckp->name, pi->processname); return ret; } static void clean_up(ckpool_t *ckp) { int i, children = ckp->proc_instances; rm_namepid(&ckp->main); dealloc(ckp->socket_dir); ckp->proc_instances = 0; for (i = 0; i < children; i++) dealloc(ckp->children[i]); dealloc(ckp->children); } static void cancel_join_pthread(pthread_t *pth) { if (!pth || !*pth) return; pthread_cancel(*pth); join_pthread(*pth); pth = NULL; } static void cancel_pthread(pthread_t *pth) { if (!pth || !*pth) return; pthread_cancel(*pth); pth = NULL; } static void wait_child(const pid_t *pid) { int ret; do { ret = waitpid(*pid, NULL, 0); } while (ret != *pid); } static void __shutdown_children(ckpool_t *ckp) { int i; cancel_join_pthread(&ckp->pth_watchdog); /* They never got set up in the first place */ if (!ckp->children) return; /* Send the children a SIGUSR1 for them to shutdown gracefully, then * wait for them to exit and kill them if they don't for 500ms. */ for (i = 0; i < ckp->proc_instances; i++) { pid_t pid = ckp->children[i]->pid; kill_pid(pid, SIGUSR1); if (!ck_completion_timeout(&wait_child, (void *)&pid, 500)) kill_pid(pid, SIGKILL); } } static void shutdown_children(ckpool_t *ckp) { cancel_join_pthread(&ckp->pth_watchdog); __shutdown_children(ckp); } static void sighandler(const int sig) { ckpool_t *ckp = global_ckp; signal(sig, SIG_IGN); signal(SIGTERM, SIG_IGN); LOGWARNING("Parent process %s received signal %d, shutting down", ckp->name, sig); cancel_join_pthread(&ckp->pth_watchdog); __shutdown_children(ckp); cancel_pthread(&ckp->pth_listener); exit(0); } static bool _json_get_string(char **store, const json_t *entry, const char *res) { bool ret = false; const char *buf; *store = NULL; if (!entry || json_is_null(entry)) { LOGDEBUG("Json did not find entry %s", res); goto out; } if (!json_is_string(entry)) { LOGWARNING("Json entry %s is not a string", res); goto out; } buf = json_string_value(entry); LOGDEBUG("Json found entry %s: %s", res, buf); *store = strdup(buf); ret = true; out: return ret; } bool json_get_string(char **store, const json_t *val, const char *res) { return _json_get_string(store, json_object_get(val, res), res); } static void json_get_int64(int64_t *store, const json_t *val, const char *res) { json_t *entry = json_object_get(val, res); if (!entry) { LOGDEBUG("Json did not find entry %s", res); return; } if (!json_is_integer(entry)) { LOGWARNING("Json entry %s is not an integer", res); return; } *store = json_integer_value(entry); LOGDEBUG("Json found entry %s: %ld", res, *store); } bool json_get_int(int *store, const json_t *val, const char *res) { json_t *entry = json_object_get(val, res); bool ret = false; if (!entry) { LOGDEBUG("Json did not find entry %s", res); goto out; } if (!json_is_integer(entry)) { LOGWARNING("Json entry %s is not an integer", res); goto out; } *store = json_integer_value(entry); LOGDEBUG("Json found entry %s: %d", res, *store); ret = true; out: return ret; } bool json_get_double(double *store, const json_t *val, const char *res) { json_t *entry = json_object_get(val, res); bool ret = false; if (!entry) { LOGDEBUG("Json did not find entry %s", res); goto out; } if (!json_is_real(entry)) { LOGWARNING("Json entry %s is not a double", res); goto out; } *store = json_real_value(entry); LOGDEBUG("Json found entry %s: %f", res, *store); ret = true; out: return ret; } static bool json_get_bool(bool *store, const json_t *val, const char *res) { json_t *entry = json_object_get(val, res); bool ret = false; if (!entry) { LOGDEBUG("Json did not find entry %s", res); goto out; } if (!json_is_boolean(entry)) { LOGWARNING("Json entry %s is not a boolean", res); goto out; } *store = json_is_true(entry); LOGDEBUG("Json found entry %s: %s", res, *store ? "true" : "false"); ret = true; out: return ret; } static void parse_btcds(ckpool_t *ckp, const json_t *arr_val, const int arr_size) { json_t *val; int i; ckp->btcds = arr_size; ckp->btcdurl = ckzalloc(sizeof(char *) * arr_size); ckp->btcdauth = ckzalloc(sizeof(char *) * arr_size); ckp->btcdpass = ckzalloc(sizeof(char *) * arr_size); ckp->btcdnotify = ckzalloc(sizeof(bool *) * arr_size); for (i = 0; i < arr_size; i++) { val = json_array_get(arr_val, i); json_get_string(&ckp->btcdurl[i], val, "url"); json_get_string(&ckp->btcdauth[i], val, "auth"); json_get_string(&ckp->btcdpass[i], val, "pass"); json_get_bool(&ckp->btcdnotify[i], val, "notify"); } } static void parse_proxies(ckpool_t *ckp, const json_t *arr_val, const int arr_size) { json_t *val; int i; ckp->proxies = arr_size; ckp->proxyurl = ckzalloc(sizeof(char *) * arr_size); ckp->proxyauth = ckzalloc(sizeof(char *) * arr_size); ckp->proxypass = ckzalloc(sizeof(char *) * arr_size); for (i = 0; i < arr_size; i++) { val = json_array_get(arr_val, i); json_get_string(&ckp->proxyurl[i], val, "url"); json_get_string(&ckp->proxyauth[i], val, "auth"); json_get_string(&ckp->proxypass[i], val, "pass"); } } static bool parse_serverurls(ckpool_t *ckp, const json_t *arr_val) { bool ret = false; int arr_size, i; if (!arr_val) goto out; if (!json_is_array(arr_val)) { LOGWARNING("Unable to parse serverurl entries as an array"); goto out; } arr_size = json_array_size(arr_val); if (!arr_size) { LOGWARNING("Serverurl array empty"); goto out; } ckp->serverurls = arr_size; ckp->serverurl = ckalloc(sizeof(char *) * arr_size); for (i = 0; i < arr_size; i++) { json_t *val = json_array_get(arr_val, i); if (!_json_get_string(&ckp->serverurl[i], val, "serverurl")) LOGWARNING("Invalid serverurl entry number %d", i); } ret = true; out: return ret; } static void parse_config(ckpool_t *ckp) { json_t *json_conf, *arr_val; json_error_t err_val; int arr_size; char *url; json_conf = json_load_file(ckp->config, JSON_DISABLE_EOF_CHECK, &err_val); if (!json_conf) { LOGWARNING("Json decode error for config file %s: (%d): %s", ckp->config, err_val.line, err_val.text); return; } arr_val = json_object_get(json_conf, "btcd"); if (arr_val && json_is_array(arr_val)) { arr_size = json_array_size(arr_val); if (arr_size) parse_btcds(ckp, arr_val, arr_size); } json_get_string(&ckp->btcaddress, json_conf, "btcaddress"); json_get_string(&ckp->btcsig, json_conf, "btcsig"); if (ckp->btcsig && strlen(ckp->btcsig) > 38) { LOGWARNING("Signature %s too long, truncating to 38 bytes", ckp->btcsig); ckp->btcsig[38] = '\0'; } json_get_int(&ckp->blockpoll, json_conf, "blockpoll"); json_get_int(&ckp->nonce1length, json_conf, "nonce1length"); json_get_int(&ckp->nonce2length, json_conf, "nonce2length"); json_get_int(&ckp->update_interval, json_conf, "update_interval"); /* Look for an array first and then a single entry */ arr_val = json_object_get(json_conf, "serverurl"); if (!parse_serverurls(ckp, arr_val)) { if (json_get_string(&url, json_conf, "serverurl")) { ckp->serverurl = ckalloc(sizeof(char *)); ckp->serverurl[0] = url; ckp->serverurls = 1; } } json_get_int64(&ckp->mindiff, json_conf, "mindiff"); json_get_int64(&ckp->startdiff, json_conf, "startdiff"); json_get_int64(&ckp->maxdiff, json_conf, "maxdiff"); json_get_string(&ckp->logdir, json_conf, "logdir"); json_get_int(&ckp->maxclients, json_conf, "maxclients"); arr_val = json_object_get(json_conf, "proxy"); if (arr_val && json_is_array(arr_val)) { arr_size = json_array_size(arr_val); if (arr_size) parse_proxies(ckp, arr_val, arr_size); } json_decref(json_conf); } static proc_instance_t *prepare_child(ckpool_t *ckp, int (*process)(), char *name) { proc_instance_t *pi = ckzalloc(sizeof(proc_instance_t)); ckp->children = realloc(ckp->children, sizeof(proc_instance_t *) * (ckp->proc_instances + 1)); ckp->children[ckp->proc_instances++] = pi; pi->ckp = ckp; pi->processname = name; pi->sockname = pi->processname; pi->process = process; create_process_unixsock(pi); return pi; } static proc_instance_t *child_by_pid(const ckpool_t *ckp, const pid_t pid) { proc_instance_t *pi = NULL; int i; for (i = 0; i < ckp->proc_instances; i++) { if (ckp->children[i]->pid == pid) { pi = ckp->children[i]; break; } } return pi; } static void *watchdog(void *arg) { #if 0 time_t last_relaunch_t = time(NULL); #endif ckpool_t *ckp = (ckpool_t *)arg; rename_proc("watchdog"); sleep(1); while (42) { proc_instance_t *pi; #if 0 time_t relaunch_t; #endif int pid, status; pid = waitpid(0, &status, 0); pi = child_by_pid(ckp, pid); if (pi && WIFEXITED(status)) { LOGWARNING("Child process %s exited, terminating!", pi->processname); break; } #if 0 /* Don't bother trying to respawn for now since communication * breakdown between the processes will make them exit. */ relaunch_t = time(NULL); if (relaunch_t == last_relaunch_t) { LOGEMERG("Respawning processes too fast, exiting!"); break; } last_relaunch_t = relaunch_t; if (pi) { LOGERR("%s process dead! Relaunching", pi->processname); launch_process(pi); } else { LOGEMERG("Unknown child process %d dead, exiting!", pid); break; } #else if (pi) LOGEMERG("%s process dead, terminating!", pi->processname); else LOGEMERG("Unknown child process %d dead, exiting!", pid); break; #endif } send_proc(&ckp->main, "shutdown"); return NULL; } #ifdef USE_CKDB static struct option long_options[] = { {"standalone", no_argument, 0, 'A'}, {"config", required_argument, 0, 'c'}, {"daemonise", no_argument, 0, 'D'}, {"ckdb-name", required_argument, 0, 'd'}, {"group", required_argument, 0, 'g'}, {"handover", no_argument, 0, 'H'}, {"help", no_argument, 0, 'h'}, {"killold", no_argument, 0, 'k'}, {"log-shares", no_argument, 0, 'L'}, {"loglevel", required_argument, 0, 'l'}, {"name", required_argument, 0, 'n'}, {"passthrough", no_argument, 0, 'P'}, {"proxy", no_argument, 0, 'p'}, {"ckdb-sockdir",required_argument, 0, 'S'}, {"sockdir", required_argument, 0, 's'}, {0, 0, 0, 0} }; #else static struct option long_options[] = { {"config", required_argument, 0, 'c'}, {"daemonise", no_argument, 0, 'D'}, {"group", required_argument, 0, 'g'}, {"handover", no_argument, 0, 'H'}, {"help", no_argument, 0, 'h'}, {"killold", no_argument, 0, 'k'}, {"log-shares", no_argument, 0, 'L'}, {"loglevel", required_argument, 0, 'l'}, {"name", required_argument, 0, 'n'}, {"passthrough", no_argument, 0, 'P'}, {"proxy", no_argument, 0, 'p'}, {"sockdir", required_argument, 0, 's'}, {0, 0, 0, 0} }; #endif static bool send_recv_path(const char *path, const char *msg) { int sockd = open_unix_client(path); bool ret = false; char *response; send_unix_msg(sockd, msg); response = recv_unix_msg(sockd); if (response) { ret = true; LOGWARNING("Received: %s in response to %s request", response, msg); dealloc(response); } Close(sockd); return ret; } int main(int argc, char **argv) { struct sigaction handler; int c, ret, i = 0, j; char buf[512] = {}; ckpool_t ckp; /* Make significant floating point errors fatal to avoid subtle bugs being missed */ feenableexcept(FE_DIVBYZERO | FE_INVALID | FE_OVERFLOW ); json_set_alloc_funcs(json_ckalloc, free); global_ckp = &ckp; memset(&ckp, 0, sizeof(ckp)); ckp.loglevel = LOG_NOTICE; ckp.initial_args = ckalloc(sizeof(char *) * (argc + 2)); /* Leave room for extra -H */ for (ckp.args = 0; ckp.args < argc; ckp.args++) ckp.initial_args[ckp.args] = strdup(argv[ckp.args]); ckp.initial_args[ckp.args] = NULL; while ((c = getopt_long(argc, argv, "Ac:Dd:g:HhkLl:n:PpS:s:", long_options, &i)) != -1) { switch (c) { case 'A': ckp.standalone = true; break; case 'c': ckp.config = optarg; break; case 'D': ckp.daemon = true; break; case 'd': ckp.ckdb_name = optarg; break; case 'g': ckp.grpnam = optarg; break; case 'H': ckp.handover = true; ckp.killold = true; break; case 'h': for (j = 0; long_options[j].val; j++) { struct option *jopt = &long_options[j]; if (jopt->has_arg) { char *upper = alloca(strlen(jopt->name) + 1); int offset = 0; do { upper[offset] = toupper(jopt->name[offset]); } while (upper[offset++] != '\0'); printf("-%c %s | --%s %s\n", jopt->val, upper, jopt->name, upper); } else printf("-%c | --%s\n", jopt->val, jopt->name); } exit(0); case 'k': ckp.killold = true; break; case 'L': ckp.logshares = true; break; case 'l': ckp.loglevel = atoi(optarg); if (ckp.loglevel < LOG_EMERG || ckp.loglevel > LOG_DEBUG) { quit(1, "Invalid loglevel (range %d - %d): %d", LOG_EMERG, LOG_DEBUG, ckp.loglevel); } break; case 'n': ckp.name = optarg; break; case 'P': if (ckp.proxy) quit(1, "Cannot set both proxy and passthrough mode"); ckp.standalone = ckp.proxy = ckp.passthrough = true; break; case 'p': if (ckp.passthrough) quit(1, "Cannot set both passthrough and proxy mode"); ckp.proxy = true; break; case 'S': ckp.ckdb_sockdir = strdup(optarg); break; case 's': ckp.socket_dir = strdup(optarg); break; } } if (!ckp.name) { if (ckp.proxy) ckp.name = "ckproxy"; else ckp.name = "ckpool"; } snprintf(buf, 15, "%s", ckp.name); prctl(PR_SET_NAME, buf, 0, 0, 0); memset(buf, 0, 15); if (ckp.grpnam) { struct group *group = getgrnam(ckp.grpnam); if (!group) quit(1, "Failed to find group %s", ckp.grpnam); ckp.gr_gid = group->gr_gid; } else ckp.gr_gid = getegid(); if (!ckp.config) { ckp.config = strdup(ckp.name); realloc_strcat(&ckp.config, ".conf"); } if (!ckp.socket_dir) { ckp.socket_dir = strdup("/tmp/"); realloc_strcat(&ckp.socket_dir, ckp.name); } trail_slash(&ckp.socket_dir); if (!CKP_STANDALONE(&ckp)) { if (!ckp.ckdb_name) ckp.ckdb_name = "ckdb"; if (!ckp.ckdb_sockdir) { ckp.ckdb_sockdir = strdup("/opt/"); realloc_strcat(&ckp.ckdb_sockdir, ckp.ckdb_name); } trail_slash(&ckp.ckdb_sockdir); ret = mkdir(ckp.ckdb_sockdir, 0750); if (ret && errno != EEXIST) quit(1, "Failed to make directory %s", ckp.ckdb_sockdir); ckp.ckdb_sockname = ckp.ckdb_sockdir; realloc_strcat(&ckp.ckdb_sockname, "listener"); } /* Ignore sigpipe */ signal(SIGPIPE, SIG_IGN); ret = mkdir(ckp.socket_dir, 0750); if (ret && errno != EEXIST) quit(1, "Failed to make directory %s", ckp.socket_dir); parse_config(&ckp); /* Set defaults if not found in config file */ if (!ckp.btcds && !ckp.proxy) { ckp.btcds = 1; ckp.btcdurl = ckzalloc(sizeof(char *)); ckp.btcdauth = ckzalloc(sizeof(char *)); ckp.btcdpass = ckzalloc(sizeof(char *)); ckp.btcdnotify = ckzalloc(sizeof(bool)); } if (ckp.btcds) { for (i = 0; i < ckp.btcds; i++) { if (!ckp.btcdurl[i]) ckp.btcdurl[i] = strdup("localhost:8332"); if (!ckp.btcdauth[i]) ckp.btcdauth[i] = strdup("user"); if (!ckp.btcdpass[i]) ckp.btcdpass[i] = strdup("pass"); } } ckp.donaddress = "14BMjogz69qe8hk9thyzbmR5pg34mVKB1e"; if (!ckp.btcaddress) ckp.btcaddress = ckp.donaddress; if (!ckp.blockpoll) ckp.blockpoll = 100; if (!ckp.nonce1length) ckp.nonce1length = 4; else if (ckp.nonce1length < 2 || ckp.nonce1length > 8) quit(0, "Invalid nonce1length %d specified, must be 2~8", ckp.nonce1length); if (!ckp.nonce2length) ckp.nonce2length = 8; else if (ckp.nonce2length < 2 || ckp.nonce2length > 8) quit(0, "Invalid nonce2length %d specified, must be 2~8", ckp.nonce2length); if (!ckp.update_interval) ckp.update_interval = 30; if (!ckp.mindiff) ckp.mindiff = 1; if (!ckp.startdiff) ckp.startdiff = 42; if (!ckp.logdir) ckp.logdir = strdup("logs"); if (!ckp.serverurls) ckp.serverurl = ckzalloc(sizeof(char *)); if (ckp.proxy && !ckp.proxies) quit(0, "No proxy entries found in config file %s", ckp.config); /* Create the log directory */ trail_slash(&ckp.logdir); ret = mkdir(ckp.logdir, 0750); if (ret && errno != EEXIST) quit(1, "Failed to make log directory %s", ckp.logdir); /* Create the workers logdir */ sprintf(buf, "%s/workers", ckp.logdir); ret = mkdir(buf, 0750); if (ret && errno != EEXIST) quit(1, "Failed to make workers log directory %s", buf); /* Create the user logdir */ sprintf(buf, "%s/users", ckp.logdir); ret = mkdir(buf, 0750); if (ret && errno != EEXIST) quit(1, "Failed to make user log directory %s", buf); /* Create the pool logdir */ sprintf(buf, "%s/pool", ckp.logdir); ret = mkdir(buf, 0750); if (ret && errno != EEXIST) quit(1, "Failed to make pool log directory %s", buf); /* Create the logfile */ sprintf(buf, "%s%s.log", ckp.logdir, ckp.name); ckp.logfp = fopen(buf, "ae"); if (!ckp.logfp) quit(1, "Failed to make open log file %s", buf); /* Make logging line buffered */ setvbuf(ckp.logfp, NULL, _IOLBF, 0); ckp.main.ckp = &ckp; ckp.main.processname = strdup("main"); ckp.main.sockname = strdup("listener"); name_process_sockname(&ckp.main.us, &ckp.main); ckp.oldconnfd = ckzalloc(sizeof(int *) * ckp.serverurls); if (ckp.handover) { const char *path = ckp.main.us.path; if (send_recv_path(path, "ping")) { for (i = 0; i < ckp.serverurls; i++) { char getfd[16]; int sockd; snprintf(getfd, 15, "getxfd%d", i); sockd = open_unix_client(path); if (sockd < 1) break; if (!send_unix_msg(sockd, getfd)) break; ckp.oldconnfd[i] = get_fd(sockd); Close(sockd); if (!ckp.oldconnfd[i]) break; LOGWARNING("Inherited old server socket %d with new file descriptor %d!", i, ckp.oldconnfd[i]); } send_recv_path(path, "reject"); send_recv_path(path, "reconnect"); send_recv_path(path, "shutdown"); } } if (ckp.daemon) { int fd; if (fork()) exit(0); setsid(); fd = open("/dev/null",O_RDWR, 0); if (fd != -1) { dup2(fd, STDIN_FILENO); dup2(fd, STDOUT_FILENO); dup2(fd, STDERR_FILENO); } } write_namepid(&ckp.main); open_process_sock(&ckp, &ckp.main, &ckp.main.us); launch_logger(&ckp.main); ckp.logfd = fileno(ckp.logfp); create_pthread(&ckp.pth_listener, listener, &ckp.main); /* Launch separate processes from here */ ckp.generator = prepare_child(&ckp, &generator, "generator"); ckp.stratifier = prepare_child(&ckp, &stratifier, "stratifier"); ckp.connector = prepare_child(&ckp, &connector, "connector"); launch_processes(&ckp); create_pthread(&ckp.pth_watchdog, watchdog, &ckp); handler.sa_handler = &sighandler; handler.sa_flags = 0; sigemptyset(&handler.sa_mask); sigaction(SIGTERM, &handler, NULL); sigaction(SIGINT, &handler, NULL); /* Shutdown from here if the listener is sent a shutdown message */ if (ckp.pth_listener) join_pthread(ckp.pth_listener); shutdown_children(&ckp); clean_up(&ckp); return 0; }