diff --git a/src/booth.h b/src/booth.h index 904066c..eb1a802 100644 --- a/src/booth.h +++ b/src/booth.h @@ -1,250 +1,251 @@ /* * Copyright (C) 2011 Jiaju Zhang * Copyright (C) 2013-2014 Philipp Marek * * 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 2.1 of the License, or (at your option) any later version. * * This software is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU * General Public License for more details. * * You should have received a copy of the GNU General Public * License along with this library; if not, write to the Free Software * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA */ #ifndef _BOOTH_H #define _BOOTH_H #include #include #include #include #include #define BOOTH_RUN_DIR "/var/run/booth/" #define BOOTH_LOG_DIR "/var/log" #define BOOTH_LOGFILE_NAME "booth.log" #define BOOTH_DEFAULT_CONF_DIR "/etc/booth/" #define BOOTH_DEFAULT_CONF_NAME "booth" #define BOOTH_DEFAULT_CONF_EXT ".conf" #define BOOTH_DEFAULT_CONF \ BOOTH_DEFAULT_CONF_DIR BOOTH_DEFAULT_CONF_NAME BOOTH_DEFAULT_CONF_EXT #define DAEMON_NAME "boothd" #define BOOTH_PATH_LEN 127 #define BOOTH_DEFAULT_PORT 9929 /* TODO: remove */ #define BOOTH_PROTO_FAMILY AF_INET #define BOOTHC_MAGIC 0x5F1BA08C #define BOOTHC_VERSION 0x00010002 /** Timeout value for poll(). * Determines frequency of periodic jobs, eg. when send-retries are done. * See process_tickets(). */ #define POLL_TIMEOUT 1000 /** @{ */ /** The on-network data structures and constants. */ #define BOOTH_NAME_LEN 64 #define CHAR2CONST(a,b,c,d) ((a << 24) | (b << 16) | (c << 8) | d) /* Says that the ticket shouldn't be active anywhere. * NONE wouldn't be specific enough. */ #define NO_ONE (-1) /* Says that another one should recover. */ #define TICKET_LOST CHAR2CONST('L', 'O', 'S', 'T') typedef unsigned char boothc_site [BOOTH_NAME_LEN]; typedef unsigned char boothc_ticket[BOOTH_NAME_LEN]; struct boothc_header { /** Authentication data; not used now. */ uint32_t iv; uint32_t auth1; uint32_t auth2; /** BOOTHC_MAGIC */ uint32_t magic; /** BOOTHC_VERSION */ uint32_t version; /** Packet source; site_id. See add_site(). */ uint32_t from; /** Length including header */ uint32_t length; /** The command respectively protocol state. See cmd_request_t. */ uint32_t cmd; /** Result of operation. 0 == OK */ uint32_t result; char data[0]; } __attribute__((packed)); struct ticket_msg { /** Ticket name. */ boothc_ticket id; /** Current leader. May be NO_ONE. See add_site(). * For a OP_REQ_VOTE this is */ uint32_t leader; /** Current term. */ uint32_t term; uint32_t term_valid_for; /* Perhaps we need to send a status along, too - like * starting, running, stopping, error, ...? */ uint32_t leader_commit; // TODO: NEEDED? } __attribute__((packed)); struct boothc_ticket_msg { struct boothc_header header; struct ticket_msg ticket; } __attribute__((packed)); typedef enum { /* 0x43 = "C"ommands */ CMD_LIST = CHAR2CONST('C', 'L', 's', 't'), CMD_GRANT = CHAR2CONST('C', 'G', 'n', 't'), CMD_REVOKE = CHAR2CONST('C', 'R', 'v', 'k'), /* Replies */ CMR_GENERAL = CHAR2CONST('G', 'n', 'l', 'R'), // Increase distance to CMR_GRANT CMR_LIST = CHAR2CONST('R', 'L', 's', 't'), CMR_GRANT = CHAR2CONST('R', 'G', 'n', 't'), CMR_REVOKE = CHAR2CONST('R', 'R', 'v', 'k'), /* Raft */ OP_REQ_VOTE = CHAR2CONST('R', 'V', 'o', 't'), OP_VOTE_FOR = CHAR2CONST('V', 't', 'F', 'r'), OP_HEARTBEAT= CHAR2CONST('H', 'r', 't', 'B'), /* AppendEntry in Raft */ OP_MY_INDEX = CHAR2CONST('M', 'I', 'd', 'x'), /* Answer to Heartbeat */ OP_REJECTED = CHAR2CONST('R', 'J', 'C', '!'), } cmd_request_t; /* TODO: make readable constants */ typedef enum { /* for compatibility with other functions */ RLT_SUCCESS = 0, RLT_ASYNC = CHAR2CONST('A', 's', 'y', 'n'), RLT_SYNC_SUCC = CHAR2CONST('S', 'c', 'c', 's'), RLT_SYNC_FAIL = CHAR2CONST('F', 'a', 'i', 'l'), RLT_INVALID_ARG = CHAR2CONST('I', 'A', 'r', 'g'), RLT_OVERGRANT = CHAR2CONST('O', 'v', 'e', 'r'), RLT_PROBABLY_SUCCESS = CHAR2CONST('S', 'u', 'c', '?'), RLT_BUSY = CHAR2CONST('B', 'u', 's', 'y'), RLT_TERM_OUTDATED = CHAR2CONST('T', 'O', 'd', 't'), RLT_TERM_STILL_VALID = CHAR2CONST('T', 'V', 'l', 'd'), } cmd_result_t; /** @} */ /** @{ */ struct booth_site { /** Calculated ID. See add_site(). */ int site_id; int type; int local; /** Roles, like ACCEPTOR, PROPOSER, or LEARNER. Not really used ATM. */ int role; char addr_string[BOOTH_NAME_LEN]; int tcp_fd; int udp_fd; /* 0-based, used for indexing into per-ticket weights */ int index; uint64_t bitmask; unsigned short family; union { struct sockaddr_in sa4; struct sockaddr_in6 sa6; }; int saddrlen; int addrlen; } __attribute__((packed)); extern struct booth_site *local; +extern struct booth_site * no_leader; /** @} */ struct booth_transport; struct client { int fd; const struct booth_transport *transport; void (*workfn)(int); void (*deadfn)(int); }; extern struct client *clients; extern struct pollfd *pollfds; int client_add(int fd, const struct booth_transport *tpt, void (*workfn)(int ci), void (*deadfn)(int ci)); int do_read(int fd, void *buf, size_t count); int do_write(int fd, void *buf, size_t count); void process_connection(int ci); void safe_copy(char *dest, char *value, size_t buflen, const char *description); struct command_line { int type; /* ACT_ */ int op; /* OP_ */ char configfile[BOOTH_PATH_LEN]; char lockfile[BOOTH_PATH_LEN]; char site[BOOTH_NAME_LEN]; struct boothc_ticket_msg msg; }; extern struct command_line cl; /* http://gcc.gnu.org/onlinedocs/gcc/Typeof.html */ #define min(a__,b__) \ ({ typeof (a__) _a = (a__); \ typeof (b__) _b = (b__); \ _a < _b ? _a : _b; }) #define max(a__,b__) \ ({ typeof (a__) _a = (a__); \ typeof (b__) _b = (b__); \ _a > _b ? _a : _b; }) #endif /* _BOOTH_H */ diff --git a/src/main.c b/src/main.c index 3b4a20f..d67dfa3 100644 --- a/src/main.c +++ b/src/main.c @@ -1,1311 +1,1317 @@ /* * Copyright (C) 2011 Jiaju Zhang * Copyright (C) 2013-2014 Philipp Marek * * 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 2.1 of the License, or (at your option) any later version. * * This software is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU * General Public License for more details. * * You should have received a copy of the GNU General Public * License along with this library; if not, write to the Free Software * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "log.h" #include "booth.h" #include "config.h" #include "transport.h" #include "inline-fn.h" #include "pacemaker.h" #include "ticket.h" #define RELEASE_VERSION "1.0" #define CLIENT_NALLOC 32 int daemonize = 0; /** Structure for "clients". * Filehandles with incoming data get registered here (and in pollfds), * along with their callbacks. * Because these can be reallocated with every new fd, addressing * happens _only_ by their numeric index. */ struct client *clients = NULL; struct pollfd *pollfds = NULL; static int client_maxi; static int client_size = 0; +static const struct booth_site _no_leader = { + .addr_string = "none", + .site_id = NO_ONE, +}; +struct booth_site *no_leader = (struct booth_site*)& _no_leader; + typedef enum { BOOTHD_STARTED=0, BOOTHD_STARTING } BOOTH_DAEMON_STATE; int poll_timeout = POLL_TIMEOUT; typedef enum { OP_LIST = 1, OP_GRANT, OP_REVOKE, } operation_t; struct booth_config *booth_conf; struct command_line cl; int do_read(int fd, void *buf, size_t count) { int rv, off = 0; while (off < count) { rv = read(fd, (char *)buf + off, count - off); if (rv == 0) return -1; if (rv == -1 && errno == EINTR) continue; if (rv == -1) return -1; off += rv; } return 0; } int do_write(int fd, void *buf, size_t count) { int rv, off = 0; retry: rv = write(fd, (char *)buf + off, count); if (rv == -1 && errno == EINTR) goto retry; /* If we cannot write _any_ data, we'd be in an (potential) loop. */ if (rv <= 0) { log_error("write failed: %s (%d)", strerror(errno), errno); return rv; } if (rv != count) { count -= rv; off += rv; goto retry; } return 0; } static void client_alloc(void) { int i; if (!clients) { clients = malloc(CLIENT_NALLOC * sizeof(struct client)); pollfds = malloc(CLIENT_NALLOC * sizeof(struct pollfd)); } else { clients = realloc(clients, (client_size + CLIENT_NALLOC) * sizeof(struct client)); pollfds = realloc(pollfds, (client_size + CLIENT_NALLOC) * sizeof(struct pollfd)); } if (!clients || !pollfds) { log_error("can't alloc for client array"); exit(1); } for (i = client_size; i < client_size + CLIENT_NALLOC; i++) { clients[i].workfn = NULL; clients[i].deadfn = NULL; clients[i].fd = -1; pollfds[i].fd = -1; pollfds[i].revents = 0; } client_size += CLIENT_NALLOC; } static void client_dead(int ci) { if (clients[ci].fd != -1) close(clients[ci].fd); clients[ci].fd = -1; clients[ci].workfn = NULL; pollfds[ci].fd = -1; } int client_add(int fd, const struct booth_transport *tpt, void (*workfn)(int ci), void (*deadfn)(int ci)) { int i; struct client *c; if (client_size + 2 >= client_maxi ) { client_alloc(); } for (i = 0; i < client_size; i++) { c = clients + i; if (c->fd != -1) continue; c->workfn = workfn; if (deadfn) c->deadfn = deadfn; else c->deadfn = client_dead; c->transport = tpt; c->fd = fd; pollfds[i].fd = fd; pollfds[i].events = POLLIN; if (i > client_maxi) client_maxi = i; return i; } assert(!"no client"); } /* Only used for client requests, TCP ???*/ void process_connection(int ci) { struct boothc_ticket_msg msg; int rv, len, exp, fd; void (*deadfn) (int ci); fd = clients[ci].fd; rv = do_read(fd, &msg.header, sizeof(msg.header)); if (rv < 0) { if (errno == ECONNRESET) log_debug("client %d connection reset for fd %d", ci, clients[ci].fd); goto kill; } if (check_boothc_header(&msg.header, -1) < 0) goto kill; /* Basic sanity checks already done. */ len = ntohl(msg.header.length); if (len) { if (len != sizeof(msg)) { bad_len: log_error("got wrong length %u", len); return; } exp = len - sizeof(msg.header); rv = do_read(clients[ci].fd, msg.header.data, exp); if (rv < 0) { log_error("connection %d read data error %d, wanted %d", ci, rv, exp); goto kill; } } /* For CMD_GRANT and CMD_REVOKE: * Don't close connection immediately, but send * result a second later? */ switch (ntohl(msg.header.cmd)) { case CMD_LIST: ticket_answer_list(fd, &msg); goto kill; case CMD_GRANT: /* Expect boothc_ticket_site_msg. */ if (len != sizeof(msg)) goto bad_len; ticket_answer_grant(fd, &msg); goto kill; case CMD_REVOKE: /* Expect boothc_ticket_site_msg. */ if (len != sizeof(msg)) goto bad_len; ticket_answer_revoke(fd, &msg); goto kill; default: log_error("connection %d cmd %x unknown", ci, ntohl(msg.header.cmd)); init_header(&msg.header,CMR_GENERAL, RLT_INVALID_ARG, sizeof(msg.header)); send_header_only(fd, &msg.header); goto kill; } assert(0); return; kill: deadfn = clients[ci].deadfn; if(deadfn) { deadfn(ci); } return; } /** Callback function for the listening TCP socket. */ static void process_listener(int ci) { int fd, i; fd = accept(clients[ci].fd, NULL, NULL); if (fd < 0) { log_error("process_listener: accept error for fd %d: %s (%d)", clients[ci].fd, strerror(errno), errno); if (clients[ci].deadfn) clients[ci].deadfn(ci); return; } i = client_add(fd, clients[ci].transport, process_connection, NULL); log_debug("add client connection %d fd %d", i, fd); } static int setup_config(int type) { int rv; rv = read_config(cl.configfile); if (rv < 0) goto out; /* Set "local" pointer, ignoring errors. */ if (cl.type == DAEMON && cl.site[0]) { if (!find_site_by_name(cl.site, &local, 1)) { log_error("Cannot find \"%s\" in the configuration.", cl.site); return -EINVAL; } local->local = 1; } else find_myself(NULL, type == CLIENT); rv = check_config(type); if (rv < 0) goto out; /* Per default the PID file name is derived from the * configuration name. */ if (!cl.lockfile[0]) { snprintf(cl.lockfile, sizeof(cl.lockfile)-1, "%s/%s.pid", BOOTH_RUN_DIR, booth_conf->name); } out: return rv; } static int setup_transport(void) { int rv; rv = transport()->init(message_recv); if (rv < 0) { log_error("failed to init booth_transport %s", transport()->name); goto out; } rv = booth_transport[TCP].init(NULL); if (rv < 0) { log_error("failed to init booth_transport[TCP]"); goto out; } out: return rv; } static int write_daemon_state(int fd, int state) { char buffer[1024]; int rv, size; size = sizeof(buffer) - 1; rv = snprintf(buffer, size, "booth_pid=%d " "booth_state=%s " "booth_type=%s " "booth_cfg_name='%s' " "booth_addr_string='%s' " "booth_port=%d\n", getpid(), ( state == BOOTHD_STARTED ? "started" : state == BOOTHD_STARTING ? "starting" : "invalid"), type_to_string(local->type), booth_conf->name, local->addr_string, booth_conf->port); if (rv < 0 || rv == size) { log_error("Buffer filled up in write_daemon_state()."); return -1; } size = rv; rv = ftruncate(fd, 0); if (rv < 0) { log_error("lockfile %s truncate error %d: %s", cl.lockfile, errno, strerror(errno)); return rv; } rv = lseek(fd, 0, SEEK_SET); if (rv < 0) { log_error("lseek set fd(%d) offset to 0 error, return(%d), message(%s)", fd, rv, strerror(errno)); rv = -1; return rv; } rv = write(fd, buffer, size); if (rv != size) { log_error("write to fd(%d, %d) returned %d, errno %d, message(%s)", fd, size, rv, errno, strerror(errno)); return -1; } return 0; } static int loop(int fd) { void (*workfn) (int ci); void (*deadfn) (int ci); int rv, i; rv = setup_transport(); if (rv < 0) goto fail; rv = setup_ticket(); if (rv < 0) goto fail; client_add(local->tcp_fd, booth_transport + TCP, process_listener, NULL); rv = write_daemon_state(fd, BOOTHD_STARTED); if (rv != 0) { log_error("write daemon state %d to lockfile error %s: %s", BOOTHD_STARTED, cl.lockfile, strerror(errno)); goto fail; } if (cl.type == ARBITRATOR) log_info("BOOTH arbitrator daemon started"); else if (cl.type == SITE) log_info("BOOTH cluster site daemon started"); while (1) { rv = poll(pollfds, client_maxi + 1, poll_timeout); if (rv == -1 && errno == EINTR) continue; if (rv < 0) { log_error("poll failed: %s (%d)", strerror(errno), errno); goto fail; } for (i = 0; i <= client_maxi; i++) { if (clients[i].fd < 0) continue; if (pollfds[i].revents & POLLIN) { workfn = clients[i].workfn; if (workfn) workfn(i); } if (pollfds[i].revents & (POLLERR | POLLHUP | POLLNVAL)) { deadfn = clients[i].deadfn; if (deadfn) deadfn(i); } } process_tickets(); } return 0; fail: return -1; } static int query_get_string_answer(cmd_request_t cmd) { struct booth_site *site; struct boothc_header reply; char *data; int data_len; int rv; struct booth_transport const *tpt; data = NULL; init_header(&cl.msg.header, cmd, 0, sizeof(cl.msg)); if (!*cl.site) site = local; else if (!find_site_by_name(cl.site, &site, 1)) { log_error("cannot find site \"%s\"", cl.site); rv = ENOENT; goto out; } tpt = booth_transport + TCP; rv = tpt->open(site); if (rv < 0) goto out_free; rv = tpt->send(site, &cl.msg, sizeof(cl.msg)); if (rv < 0) goto out_free; rv = tpt->recv(site, &reply, sizeof(reply)); if (rv < 0) goto out_free; data_len = ntohl(reply.length) - sizeof(reply); data = malloc(data_len); if (!data) { rv = -ENOMEM; goto out_free; } rv = tpt->recv(site, data, data_len); if (rv < 0) goto out_free; do_write(STDOUT_FILENO, data, data_len); rv = 0; out_free: free(data); tpt->close(site); out: return rv; } static int do_command(cmd_request_t cmd) { struct booth_site *site; struct boothc_header reply; struct booth_transport const *tpt; int rv; rv = 0; site = NULL; if (!*cl.site) site = local; else { if (!find_site_by_name(cl.site, &site, 1)) { log_error("Site \"%s\" not configured.", cl.site); goto out_close; } if (site->type == ARBITRATOR) { log_error("Site \"%s\" is an arbitrator, cannot grant ticket there.", cl.site); goto out_close; } assert(site->type == SITE); } /* We don't check for existence of ticket, so that asking can be * done without local configuration, too. * Although, that means that the UDP port has to be specified, too. */ if (!cl.msg.ticket.id[0]) { /* If the loaded configuration has only a single ticket defined, use that. */ if (booth_conf->ticket_count == 1) { strcpy(cl.msg.ticket.id, booth_conf->ticket[0].name); } else { log_error("No ticket given."); goto out_close; } } init_header(&cl.msg.header, cmd, 0, sizeof(cl.msg)); /* Always use TCP for client - at least for now. */ tpt = booth_transport + TCP; rv = tpt->open(site); if (rv < 0) goto out_close; rv = tpt->send(site, &cl.msg, sizeof(cl.msg)); if (rv < 0) goto out_close; rv = tpt->recv(site, &reply, sizeof(reply)); if (rv < 0) goto out_close; if (reply.result == htonl(RLT_INVALID_ARG)) { log_info("invalid argument!"); rv = -1; goto out_close; } if (reply.result == htonl(RLT_OVERGRANT)) { log_info("You're granting a granted ticket. " "If you wanted to migrate a ticket, " "use revoke first, then use grant."); rv = -1; goto out_close; } rv = ntohl(reply.result); switch (rv) { case RLT_ASYNC: if (cmd == CMD_GRANT) log_info("grant command sent, result will be returned " "asynchronously, you can get the result from " "the log files"); else if (cmd == CMD_REVOKE) log_info("revoke command sent, result will be returned " "asynchronously, you can get the result from " "the log files."); else log_error("internal error reading reply result!"); rv = 0; break; case RLT_SYNC_SUCC: case RLT_SUCCESS: if (cmd == CMD_GRANT) log_info("grant succeeded!"); else if (cmd == CMD_REVOKE) log_info("revoke succeeded!"); rv = 0; break; case RLT_SYNC_FAIL: if (cmd == CMD_GRANT) log_info("grant failed!"); else if (cmd == CMD_REVOKE) log_info("revoke failed!"); rv = -1; break; case RLT_INVALID_ARG: log_error("\"Invalid argument\", most probably ticket name \"%s\" wrong.", cl.msg.ticket.id); break; default: log_error("got an error code: %x", rv); rv = -1; } out_close: if (site) local_transport->close(site); return rv; } static int do_grant(void) { return do_command(CMD_GRANT); } static int do_revoke(void) { return do_command(CMD_REVOKE); } static int _lockfile(int mode, int *fdp, pid_t *locked_by) { struct flock lock; int fd, rv; /* After reboot the directory may not yet exist. * Try to create it, but ignore errors. */ if (strncmp(cl.lockfile, BOOTH_RUN_DIR, strlen(BOOTH_RUN_DIR)) == 0) mkdir(BOOTH_RUN_DIR, 0775); if (locked_by) *locked_by = 0; *fdp = -1; fd = open(cl.lockfile, mode, 0664); if (fd < 0) return errno; *fdp = fd; lock.l_type = F_WRLCK; lock.l_start = 0; lock.l_whence = SEEK_SET; lock.l_len = 0; lock.l_pid = 0; if (fcntl(fd, F_SETLK, &lock) == 0) return 0; rv = errno; if (locked_by) if (fcntl(fd, F_GETLK, &lock) == 0) *locked_by = lock.l_pid; return rv; } static inline int is_root(void) { /* TODO: getuid()? Better way to check? */ return geteuid() == 0; } static int create_lockfile(void) { int rv, fd; fd = -1; rv = _lockfile(O_CREAT | O_WRONLY, &fd, NULL); if (fd == -1) { log_error("lockfile %s open error %d: %s", cl.lockfile, rv, strerror(rv)); return -1; } if (rv < 0) { log_error("lockfile %s setlk error %d: %s", cl.lockfile, rv, strerror(rv)); goto fail; } rv = write_daemon_state(fd, BOOTHD_STARTING); if (rv != 0) { log_error("write daemon state %d to lockfile error %s: %s", BOOTHD_STARTING, cl.lockfile, strerror(errno)); goto fail; } if (is_root()) { if (fchown(fd, booth_conf->uid, booth_conf->gid) < 0) log_error("fchown() on lockfile said %d: %s", errno, strerror(errno)); } return fd; fail: close(fd); return -1; } static void unlink_lockfile(int fd) { unlink(cl.lockfile); close(fd); } static void print_usage(void) { printf("Usages:\n"); printf(" booth daemon [-c config] [-D]\n"); printf(" booth [client] {list|grant|revoke} [options]\n"); printf(" booth status [-c config] [-D]\n"); printf("\n"); printf("Client operations:\n"); printf(" list: List all the tickets\n"); printf(" grant: Grant ticket to site\n"); printf(" revoke: Revoke ticket from site\n"); printf("\n"); printf("Options:\n"); printf(" -c FILE Specify config file [default " BOOTH_DEFAULT_CONF "]\n"); printf(" Can be a path or a name without \".conf\" suffix\n"); printf(" -D Enable debugging to stderr and don't fork\n"); printf(" -S Systemd mode (no forking)\n"); printf(" -t ticket name\n"); printf(" -s site name\n"); printf(" -l LOCKFILE Specify lock file path (daemon only)\n"); printf(" -h Print this help, then exit\n"); printf("\n"); printf("Please see the man page for details.\n"); } #define OPTION_STRING "c:Dl:t:s:hS" void safe_copy(char *dest, char *value, size_t buflen, const char *description) { int content_len = buflen - 1; if (strlen(value) >= content_len) { fprintf(stderr, "'%s' exceeds maximum %s length of %d\n", value, description, content_len); exit(EXIT_FAILURE); } strncpy(dest, value, content_len); dest[content_len] = 0; } static int host_convert(char *hostname, char *ip_str, size_t ip_size) { struct addrinfo *result = NULL, hints = {0}; int re = -1; memset(&hints, 0, sizeof(hints)); hints.ai_family = BOOTH_PROTO_FAMILY; hints.ai_socktype = SOCK_DGRAM; re = getaddrinfo(hostname, NULL, &hints, &result); if (re == 0) { struct in_addr addr = ((struct sockaddr_in *)result->ai_addr)->sin_addr; const char *re_ntop = inet_ntop(BOOTH_PROTO_FAMILY, &addr, ip_str, ip_size); if (re_ntop == NULL) { re = -1; } } freeaddrinfo(result); return re; } static int read_arguments(int argc, char **argv) { int optchar; char *arg1 = argv[1]; char *op = NULL; char *cp; char site_arg[INET_ADDRSTRLEN] = {0}; int left; if (argc < 2 || !strcmp(arg1, "help") || !strcmp(arg1, "--help") || !strcmp(arg1, "-h")) { print_usage(); exit(EXIT_SUCCESS); } if (!strcmp(arg1, "version") || !strcmp(arg1, "--version") || !strcmp(arg1, "-V")) { printf("%s %s (built %s %s)\n", argv[0], RELEASE_VERSION, __DATE__, __TIME__); exit(EXIT_SUCCESS); } if (strcmp(arg1, "arbitrator") == 0 || strcmp(arg1, "site") == 0 || strcmp(arg1, "start") == 0 || strcmp(arg1, "daemon") == 0) { cl.type = DAEMON; optind = 2; } else if (strcmp(arg1, "status") == 0) { cl.type = STATUS; optind = 2; } else if (strcmp(arg1, "client") == 0) { cl.type = CLIENT; if (argc < 3) { print_usage(); exit(EXIT_FAILURE); } op = argv[2]; optind = 3; } else { cl.type = CLIENT; op = argv[1]; optind = 2; } if (cl.type == CLIENT) { if (!strcmp(op, "list")) cl.op = OP_LIST; else if (!strcmp(op, "grant")) cl.op = OP_GRANT; else if (!strcmp(op, "revoke")) cl.op = OP_REVOKE; else { fprintf(stderr, "client operation \"%s\" is unknown\n", op); exit(EXIT_FAILURE); } } while (optind < argc) { optchar = getopt(argc, argv, OPTION_STRING); switch (optchar) { case 'c': if (strchr(optarg, '/')) { safe_copy(cl.configfile, optarg, sizeof(cl.configfile), "config file"); } else { /* If no "/" in there, use with default directory. */ strcpy(cl.configfile, BOOTH_DEFAULT_CONF_DIR); cp = cl.configfile + strlen(BOOTH_DEFAULT_CONF_DIR); assert(cp > cl.configfile); assert(*(cp-1) == '/'); /* Write at the \0, ie. after the "/" */ safe_copy(cp, optarg, (sizeof(cl.configfile) - (cp - cl.configfile) - strlen(BOOTH_DEFAULT_CONF_EXT)), "config name"); /* If no extension, append ".conf". * Space is available, see -strlen() above. */ if (!strchr(cp, '.')) strcat(cp, BOOTH_DEFAULT_CONF_EXT); } break; case 'D': debug_level++; /* Fall through */ case 'S': daemonize = 1; break; case 'l': safe_copy(cl.lockfile, optarg, sizeof(cl.lockfile), "lock file"); break; case 't': if (cl.op == OP_GRANT || cl.op == OP_REVOKE) { safe_copy(cl.msg.ticket.id, optarg, sizeof(cl.msg.ticket.id), "ticket name"); } else { print_usage(); exit(EXIT_FAILURE); } break; case 's': /* For testing and debugging: allow "-s site" also for * daemon start, so that the address that should be used * can be set manually. * This makes it easier to start multiple processes * on one machine. */ if (cl.type == CLIENT || (cl.type == DAEMON && debug_level)) { int re = host_convert(optarg, site_arg, INET_ADDRSTRLEN); if (re == 0) { safe_copy(cl.site, site_arg, sizeof(cl.site), "site name"); } else { safe_copy(cl.site, optarg, sizeof(cl.site), "site name"); } } else { log_error("\"-s\" not allowed in daemon mode."); exit(EXIT_FAILURE); } break; case 'h': print_usage(); exit(EXIT_SUCCESS); break; case ':': case '?': fprintf(stderr, "Please use '-h' for usage.\n"); exit(EXIT_FAILURE); break; case -1: /* No more parameters on cmdline, only arguments. */ goto extra_args; default: goto unknown; }; } return 0; extra_args: if (cl.type == CLIENT && !cl.msg.ticket.id[0]) { /* Use additional argument as ticket name. */ safe_copy(cl.msg.ticket.id, argv[optind], sizeof(cl.msg.ticket.id), "ticket name"); optind++; } if (optind == argc) return 0; left = argc - optind; fprintf(stderr, "Superfluous argument%s: %s%s\n", left == 1 ? "" : "s", argv[optind], left == 1 ? "" : "..."); exit(EXIT_FAILURE); unknown: fprintf(stderr, "unknown option: %s\n", argv[optind]); exit(EXIT_FAILURE); } static void set_scheduler(void) { struct sched_param sched_param; struct rlimit rlimit; int rv; rlimit.rlim_cur = RLIM_INFINITY; rlimit.rlim_max = RLIM_INFINITY; setrlimit(RLIMIT_MEMLOCK, &rlimit); rv = mlockall(MCL_CURRENT | MCL_FUTURE); if (rv < 0) { log_error("mlockall failed"); } rv = sched_get_priority_max(SCHED_RR); if (rv != -1) { sched_param.sched_priority = rv; rv = sched_setscheduler(0, SCHED_RR, &sched_param); if (rv == -1) log_error("could not set SCHED_RR priority %d: %s (%d)", sched_param.sched_priority, strerror(errno), errno); } else { log_error("could not get maximum scheduler priority err %d", errno); } } static void set_oom_adj(int val) { FILE *fp; fp = fopen("/proc/self/oom_adj", "w"); if (!fp) return; fprintf(fp, "%i", val); fclose(fp); } static int do_status(int type) { pid_t pid; int rv, lock_fd, ret; const char *reason = NULL; char lockfile_data[1024], *cp; ret = PCMK_OCF_NOT_RUNNING; /* TODO: query all, and return quit only if it's _cleanly_ not * running, ie. _neither_ of port/lockfile/process is available? * * Currently a single failure says "not running", even if "only" the * lockfile has been removed. */ rv = setup_config(type); if (rv) { reason = "Error reading configuration."; ret = PCMK_OCF_UNKNOWN_ERROR; goto quit; } if (!local) { reason = "No Service IP active here."; goto quit; } rv = _lockfile(O_RDWR, &lock_fd, &pid); if (rv == 0) { reason = "PID file not locked."; goto quit; } if (lock_fd == -1) { reason = "No PID file."; goto quit; } if (pid) { fprintf(stdout, "booth_lockpid=%d ", pid); fflush(stdout); } rv = read(lock_fd, lockfile_data, sizeof(lockfile_data) - 1); if (rv < 4) { reason = "Cannot read lockfile data."; ret = PCMK_LSB_UNKNOWN_ERROR; goto quit; } lockfile_data[rv] = 0; if (lock_fd != -1) close(lock_fd); /* Make sure it's only a single line */ cp = strchr(lockfile_data, '\r'); if (cp) *cp = 0; cp = strchr(lockfile_data, '\n'); if (cp) *cp = 0; rv = setup_udp_server(1); if (rv == 0) { reason = "UDP port not in use."; goto quit; } fprintf(stdout, "booth_lockfile='%s' %s\n", cl.lockfile, lockfile_data); if (daemonize) fprintf(stderr, "Booth at %s port %d seems to be running.\n", local->addr_string, booth_conf->port); return 0; quit: log_debug("not running: %s", reason); /* Ie. "DEBUG" */ if (daemonize) fprintf(stderr, "not running: %s\n", reason); return ret; } static int limit_this_process(void) { int rv; if (!is_root()) return 0; if (setregid(booth_conf->gid, booth_conf->gid) < 0) { rv = errno; log_error("setregid() didn't work: %s", strerror(rv)); return rv; } if (setreuid(booth_conf->uid, booth_conf->uid) < 0) { rv = errno; log_error("setreuid() didn't work: %s", strerror(rv)); return rv; } /* TODO: ulimits? But that would restrict crm_ticket and handler * scripts, too! */ return 0; } static int do_server(int type) { int lock_fd = -1; int rv = -1; static char log_ent[128] = DAEMON_NAME "-"; rv = setup_config(type); if (rv < 0) goto out; if (!local) { log_error("Cannot find myself in the configuration."); exit(EXIT_FAILURE); } if (!daemonize) { if (daemon(0, 0) < 0) { perror("daemon error"); exit(EXIT_FAILURE); } } /* The lockfile must be written to _after_ the call to daemon(), so * that the lockfile contains the pid of the daemon, not the parent. */ lock_fd = create_lockfile(); if (lock_fd < 0) return lock_fd; strcat(log_ent, type_to_string(local->type)); cl_log_set_entity(log_ent); cl_log_enable_stderr(debug_level ? TRUE : FALSE); cl_log_set_facility(HA_LOG_FACILITY); cl_inherit_logging_environment(0); log_info("BOOTH %s daemon is starting, node id is 0x%08X (%d).", type_to_string(local->type), local->site_id, local->site_id); signal(SIGUSR1, (__sighandler_t)tickets_log_info); set_scheduler(); set_oom_adj(-16); set_proc_title("%s %s for [%s]:%d", DAEMON_NAME, type_to_string(local->type), local->addr_string, booth_conf->port); rv = limit_this_process(); if (rv) return rv; rv = loop(lock_fd); out: if (lock_fd >= 0) { /* We might not be able to delete it, but at least * make it empty. */ rv = ftruncate(lock_fd, 0); (void)rv; unlink_lockfile(lock_fd); } return rv; } static int do_client(void) { int rv = -1; rv = setup_config(CLIENT); if (rv < 0) { log_error("cannot read config"); goto out; } switch (cl.op) { case OP_LIST: rv = query_get_string_answer(CMD_LIST); break; case OP_GRANT: rv = do_grant(); break; case OP_REVOKE: rv = do_revoke(); break; } out: return rv; } int main(int argc, char *argv[], char *envp[]) { int rv; init_set_proc_title(argc, argv, envp); memset(&cl, 0, sizeof(cl)); strncpy(cl.configfile, BOOTH_DEFAULT_CONF, BOOTH_PATH_LEN - 1); cl.lockfile[0] = 0; debug_level = 0; cl_log_set_entity("booth"); cl_log_enable_stderr(TRUE); cl_log_set_facility(0); rv = read_arguments(argc, argv); if (rv < 0) goto out; switch (cl.type) { case STATUS: rv = do_status(cl.type); break; case ARBITRATOR: case DAEMON: case SITE: rv = do_server(cl.type); break; case CLIENT: rv = do_client(); break; } out: /* Normalize values. 0x100 would be seen as "OK" by waitpid(). */ return (rv >= 0 && rv < 0x70) ? rv : 1; } diff --git a/src/ticket.c b/src/ticket.c index 5a50419..426aa4b 100644 --- a/src/ticket.c +++ b/src/ticket.c @@ -1,648 +1,649 @@ /* * Copyright (C) 2011 Jiaju Zhang * Copyright (C) 2013-2014 Philipp Marek * * 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 2.1 of the License, or (at your option) any later version. * * This software is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU * General Public License for more details. * * You should have received a copy of the GNU General Public * License along with this library; if not, write to the Free Software * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA */ #include #include #include #include #include #include #include #include #include "ticket.h" #include "config.h" #include "pacemaker.h" #include "inline-fn.h" #include "log.h" #include "booth.h" #include "raft.h" #include "handler.h" #define TK_LINE 256 /* Untrusted input, must fit (incl. \0) in a buffer of max chars. */ int check_max_len_valid(const char *s, int max) { int i; for(i=0; iticket_count; i++) { if (!strcmp(booth_conf->ticket[i].name, ticket)) { if (found) *found = booth_conf->ticket + i; return 1; } } return 0; } int check_ticket(char *ticket, struct ticket_config **found) { if (found) *found = NULL; if (!booth_conf) return 0; if (!check_max_len_valid(ticket, sizeof(booth_conf->ticket[0].name))) return 0; return find_ticket_by_name(ticket, found); } int check_site(char *site, int *is_local) { struct booth_site *node; if (!check_max_len_valid(site, sizeof(node->addr_string))) return 0; if (find_site_by_name(site, &node, 0)) { *is_local = node->local; return 1; } return 0; } #if 0 /** Find out what others think about this ticket. * * If we're a SITE, we can ask (and have to tell) Pacemaker. * An ARBITRATOR can only ask others. */ static int ticket_send_catchup(struct ticket_config *tk) { int i, rv = 0; struct booth_site *site; struct boothc_ticket_msg msg; foreach_node(i, site) { if (!site->local) { init_ticket_msg(&msg, CMD_CATCHUP, RLT_SUCCESS, tk); log_debug("attempting catchup from %s", site->addr_string); rv = booth_udp_send(site, &msg, sizeof(msg)); } } ticket_activate_timeout(tk); return rv; } #endif int ticket_write(struct ticket_config *tk) { if (local->type != SITE) return -EINVAL; disown_if_expired(tk); if (tk->leader == local) { pcmk_handler.grant_ticket(tk); } else { pcmk_handler.revoke_ticket(tk); } return 0; } /* Ask an external program whether getting the ticket * makes sense. * Eg. if the services have a failcount of INFINITY, * we can't serve here anyway. */ int get_ticket_locally_if_allowed(struct ticket_config *tk) { int rv; if (!tk->ext_verifier) goto get_it; rv = run_handler(tk, tk->ext_verifier, 1); if (rv) { log_error("May not acquire ticket."); /* Give it to somebody else. * Just send a commit message, as the * others couldn't help anyway. */ if (leader_and_valid(tk)) { disown_ticket(tk); #if 0 tk->proposed_owner = NULL; /* Just go one further - others may easily override. */ tk->new_ballot++; ticket_broadcast_proposed_state(tk, OP_COMMITTED); tk->state = ST_STABLE; #endif ticket_broadcast(tk, OP_VOTE_FOR, RLT_SUCCESS); } return rv; } else { log_info("May keep ticket."); } get_it: if (leader_and_valid(tk)) { return send_heartbeat(tk); } else { /* Have to take away?? */ /* shouldn't happen*/ assert(0); new_election(tk, local); return ticket_broadcast(tk, OP_REQ_VOTE, RLT_SUCCESS); } #if 0 return paxos_start_round(tk, local); #endif } /** Try to get the ticket for the local site. * */ int do_grant_ticket(struct ticket_config *tk) { int rv; if (tk->leader == local) return RLT_SUCCESS; if (tk->leader) return RLT_OVERGRANT; rv = get_ticket_locally_if_allowed(tk); return rv; } /** Start a PAXOS round for revoking. * That can be started from any site. */ int do_revoke_ticket(struct ticket_config *tk) { int rv; if (!tk->leader) return RLT_SUCCESS; disown_ticket(tk); + tk->voted_for = no_leader; ticket_write(tk); return ticket_broadcast(tk, OP_REQ_VOTE, RLT_SUCCESS); #if 0 rv = paxos_start_round(tk, NULL); #endif return rv; } int list_ticket(char **pdata, unsigned int *len) { struct ticket_config *tk; char timeout_str[64]; char *data, *cp; int i, alloc; *pdata = NULL; *len = 0; alloc = 256 + booth_conf->ticket_count * (BOOTH_NAME_LEN * 2 + 128); data = malloc(alloc); if (!data) return -ENOMEM; cp = data; foreach_ticket(i, tk) { if (tk->term_expires != 0) strftime(timeout_str, sizeof(timeout_str), "%F %T", localtime(&tk->term_expires)); else strcpy(timeout_str, "INF"); cp += sprintf(cp, "ticket: %s, leader: %s, expires: %s, commit: %d\n", tk->name, ticket_leader_string(tk), timeout_str, tk->commit_index); *len = cp - data; assert(*len < alloc); } *pdata = data; return 0; } int setup_ticket(void) { struct ticket_config *tk; int i; /* TODO */ foreach_ticket(i, tk) { tk->leader = NULL; tk->term_expires = 0; // abort_proposal(tk); if (local->type == SITE) { pcmk_handler.load_ticket(tk); } /* There might be a leader; wait for its notification. */ tk->term_expires = time(NULL) + tk->term_duration; tk->state = ST_FOLLOWER; /* TODO: send query packet to see sooner who's online. */ } return 0; } int ticket_answer_list(int fd, struct boothc_ticket_msg *msg) { char *data; int olen, rv; struct boothc_header hdr; rv = list_ticket(&data, &olen); if (rv < 0) return rv; init_header(&hdr, CMR_LIST, RLT_SUCCESS, sizeof(hdr) + olen); return send_header_plus(fd, &hdr, data, olen); } int ticket_answer_grant(int fd, struct boothc_ticket_msg *msg) { int rv; struct ticket_config *tk; if (!check_ticket(msg->ticket.id, &tk)) { log_error("Client asked to grant unknown ticket"); rv = RLT_INVALID_ARG; goto reply; } if (tk->leader) { log_error("client wants to get an (already granted!) ticket \"%s\"", msg->ticket.id); rv = RLT_OVERGRANT; goto reply; } rv = do_grant_ticket(tk); reply: init_header(&msg->header, CMR_GRANT, rv ?: RLT_ASYNC, sizeof(*msg)); return send_ticket_msg(fd, msg); } int ticket_answer_revoke(int fd, struct boothc_ticket_msg *msg) { int rv; struct ticket_config *tk; if (!check_ticket(msg->ticket.id, &tk)) { log_error("Client asked to grant unknown ticket"); rv = RLT_INVALID_ARG; goto reply; } if (!tk->leader) { log_info("client wants to revoke a free ticket \"%s\"", msg->ticket.id); /* Return a different result code? */ rv = RLT_SUCCESS; goto reply; } rv = do_revoke_ticket(tk); if (rv == 0) rv = RLT_ASYNC; reply: init_ticket_msg(msg, CMR_REVOKE, rv, tk); return send_ticket_msg(fd, msg); } int ticket_broadcast(struct ticket_config *tk, cmd_request_t cmd, cmd_result_t res) { struct boothc_ticket_msg msg; init_ticket_msg(&msg, cmd, res, tk); log_debug("broadcasting '%s' for ticket \"%s\"", state_to_string(cmd), tk->name); return transport()->broadcast(&msg, sizeof(msg)); } static void ticket_cron(struct ticket_config *tk) { time_t now; int rv; now = time(NULL); R(tk); /* Has an owner, has an expiry date, and expiry date in the past? * Losing the ticket must happen in _every_ state. */ if (tk->term_expires && tk->leader && now > tk->term_expires) { log_info("LOST ticket: \"%s\" no longer at %s", tk->name, ticket_leader_string(tk)); /* Couldn't renew in time - ticket lost. */ disown_ticket(tk); /* New vote round; §5.2 */ if (local->type == SITE) new_election(tk, NULL); /* should be "always" that way else tk->state = ST_FOLLOWER; */ // abort_proposal(tk); TODO ticket_write(tk); ticket_activate_timeout(tk); /* May not try to re-acquire now, need to find out * what others think. */ return; } R(tk); switch(tk->state) { case ST_INIT: /* Unknown state, ask others. */ // ticket_send_catchup(tk); break; case ST_FOLLOWER: if (tk->term_expires && now > tk->term_expires) { new_election(tk, NULL); } break; case ST_CANDIDATE: /* §5.2 */ if (now > tk->election_end) new_election(tk, NULL); break; case ST_LEADER: if (tk->hb_sent_at + tk->timeout > now) { /* Heartbeat timeout reached. Oops ... */ tk->retry_number ++; log_error("Not enough answers to heartbeat on try #%d: " "only got %d answers (mask 0x%" PRIx64 ")!", tk->retry_number, count_bits(tk->hb_received), tk->hb_received); /* Don't give up, though - there's still some time until leadership is lost. */ } rv = run_handler(tk, tk->ext_verifier, 1); if (rv) { tk->state = ST_FOLLOWER; tk->leader= NULL; // resp. no owner anymore, new takers? ticket_broadcast(tk, OP_REQ_VOTE, RLT_SUCCESS); ticket_write(tk); } else { tk->term_expires = now + tk->term_duration; send_heartbeat(tk); // ticket_write(tk); // not correct here -- no acks received yet } ticket_activate_timeout(tk); break; default: break; } R(tk); } void process_tickets(void) { struct ticket_config *tk; int i; struct timeval now; float sec_until; gettimeofday(&now, NULL); foreach_ticket(i, tk) { sec_until = timeval_to_float(tk->next_cron) - timeval_to_float(now); if (0) log_debug("ticket %s next cron %" PRIx64 ".%03d, " "now %" PRIx64 "%03d, in %f", tk->name, (uint64_t)tk->next_cron.tv_sec, timeval_msec(tk->next_cron), (uint64_t)now.tv_sec, timeval_msec(now), sec_until); if (sec_until > 0.0) continue; log_debug("ticket cron: doing %s", tk->name); /* Set next value, handler may override. * This should already be handled via the state logic; * but to be on the safe side the renew repetition is * duplicated here, too. */ set_ticket_wakeup(tk); ticket_cron(tk); } } void tickets_log_info(void) { struct ticket_config *tk; int i; foreach_ticket(i, tk) { log_info("Ticket %s: state '%s' " "commit index %d " "leader \"%s\" " "expires %-24.24s", tk->name, state_to_string(tk->state), tk->commit_index, ticket_leader_string(tk), ctime(&tk->term_expires)); } } /* UDP message receiver. */ int message_recv(struct boothc_ticket_msg *msg, int msglen) { uint32_t from; struct booth_site *source; struct ticket_config *tk; struct booth_site *leader; uint32_t leader_u; if (check_boothc_header(&msg->header, sizeof(*msg)) < 0 || msglen != sizeof(*msg)) { log_error("message receive error"); return -1; } from = ntohl(msg->header.from); if (!find_site_by_id(from, &source) || !source) { log_error("unknown sender: %08x", from); return -1; } if (!check_ticket(msg->ticket.id, &tk)) { log_error("got invalid ticket name \"%s\" from %s", msg->ticket.id, source->addr_string); return -EINVAL; } leader_u = ntohl(msg->ticket.leader); if (!find_site_by_id(leader_u, &leader)) { log_error("Message with unknown owner %x received", leader_u); return -EINVAL; } return raft_answer(tk, source, leader, msg); } void set_ticket_wakeup(struct ticket_config *tk) { struct timeval tv, now; /* At least every hour, perhaps sooner. */ ticket_next_cron_in(tk, 3600); switch (tk->state) { case ST_LEADER: assert(tk->leader == local); gettimeofday(&now, NULL); tv = now; tv.tv_sec = next_vote_starts_at(tk); /* If timestamp is in the past, look again in one second. */ if (timeval_compare(tv, now) <= 0) tv.tv_sec = now.tv_sec + 1; ticket_next_cron_at(tk, tv); break; case ST_CANDIDATE: assert(tk->election_end); ticket_next_cron_at_coarse(tk, tk->election_end); break; case ST_INIT: case ST_FOLLOWER: /* If there is (or should be) some owner, check on her later on. * If no one is interested - don't care. */ if ((tk->leader || tk->acquire_after) && (local->type == SITE)) ticket_next_cron_at_coarse(tk, tk->term_expires + tk->acquire_after); break; default: log_error("why here?"); } } /* Given a state (in host byte order), return a human-readable (char*). * An array is used so that multiple states can be printed in a single printf(). */ char *state_to_string(uint32_t state_ho) { union mu { cmd_request_t s; char c[5]; }; static union mu cache[6] = { { 0 } }, *cur; static int current = 0; current ++; if (current >= sizeof(cache)/sizeof(cache[0])) current = 0; cur = cache + current; cur->s = htonl(state_ho); /* Shouldn't be necessary, union array is initialized with zeroes, and * these bytes never get written. */ cur->c[4] = 0; return cur->c; } int send_reject(struct booth_site *dest, struct ticket_config *tk, cmd_result_t code) { struct boothc_ticket_msg msg; init_ticket_msg(&msg, OP_REJECTED, code, tk); return booth_udp_send(dest, &msg, sizeof(msg)); }