X-Git-Url: http://git.tuebingen.mpg.de/?p=dss.git;a=blobdiff_plain;f=dss.c;h=63f5f568fd735a8470f54d8df1663e243de81c06;hp=6101f17742c7ba8ef6e3a57a1878dfccb7227feb;hb=492928ebe8b83c31aeca4866fef37ab07174ed9c;hpb=0ba29646cbd80a8bb7bdf827352bd1332217c398 diff --git a/dss.c b/dss.c index 6101f17..63f5f56 100644 --- a/dss.c +++ b/dss.c @@ -1,5 +1,5 @@ /* - * Copyright (C) 2008-2009 Andre Noll + * Copyright (C) 2008-2010 Andre Noll * * Licensed under the GPL v2. For licencing details see COPYING. */ @@ -33,6 +33,7 @@ #include "df.h" #include "time.h" #include "snap.h" +#include "ipc.h" /** Command line and config file options. */ static struct gengetopt_args_info conf; @@ -47,7 +48,7 @@ static int create_process_stopped; /** Process id of current pre-remove/rm/post-remove process. */ static pid_t remove_pid; /** When the next snapshot is due. */ -static struct timeval next_snapshot_time; +static int64_t next_snapshot_time; /** When to try to remove something. */ static struct timeval next_removal_check; /** Creation time of the snapshot currently being created. */ @@ -64,14 +65,114 @@ enum hook_status snapshot_removal_status; DEFINE_DSS_ERRLIST; +static const char const *hook_status_description[] = {HOOK_STATUS_ARRAY}; +/* may be called with ds == NULL. */ +static int disk_space_low(struct disk_space *ds) +{ + struct disk_space ds_struct; + + if (!ds) { + int ret = get_disk_space(".", &ds_struct); + if (ret < 0) + return ret; + ds = &ds_struct; + } + if (conf.min_free_mb_arg) + if (ds->free_mb < conf.min_free_mb_arg) + return 1; + if (conf.min_free_percent_arg) + if (ds->percent_free < conf.min_free_percent_arg) + return 1; + if (conf.min_free_percent_inodes_arg) + if (ds->percent_free_inodes < conf.min_free_percent_inodes_arg) + return 1; + return 0; +} + +static void dump_dss_config(const char *msg) +{ + const char dash[] = "-----------------------------"; + int ret; + FILE *log = logfile? logfile : stderr; + struct disk_space ds; + int64_t now = get_current_time(); + + if (conf.loglevel_arg > INFO) + return; + + fprintf(log, "%s <%s config> %s\n", dash, msg, dash); + fprintf(log, "\n*** disk space ***\n\n"); + ret = get_disk_space(".", &ds); + if (ret >= 0) { + DSS_INFO_LOG("disk space low: %s\n", disk_space_low(&ds)? + "yes" : "no"); + log_disk_space(&ds); + } else + DSS_ERROR_LOG("can not get free disk space: %s\n", + dss_strerror(-ret)); + + /* we continue on errors from get_disk_space */ + + fprintf(log, "\n*** command line and config file options ***\n\n"); + cmdline_parser_dump(log, &conf); + fprintf(log, "\n*** internal state ***\n\n"); + fprintf(log, + "pid: %d\n" + "logile: %s\n" + "snapshot_currently_being_removed: %s\n" + "path_to_last_complete_snapshot: %s\n" + "reference_snapshot: %s\n" + "snapshot_creation_status: %s\n" + "snapshot_removal_status: %s\n" + , + (int) getpid(), + logfile? conf.logfile_arg : "stderr", + snapshot_currently_being_removed? + snapshot_currently_being_removed->name : "(none)", + path_to_last_complete_snapshot? + path_to_last_complete_snapshot : "(none)", + name_of_reference_snapshot? + name_of_reference_snapshot : "(none)", + hook_status_description[snapshot_creation_status], + hook_status_description[snapshot_removal_status] + ); + if (create_pid != 0) + fprintf(log, + "create_pid: %" PRId32 "\n" + "create process is %sstopped\n" + , + create_pid, + create_process_stopped? "" : "not " + ); + if (remove_pid != 0) + fprintf(log, "remove_pid: %" PRId32 "\n", remove_pid); + if (next_snapshot_time != 0) + fprintf(log, "next snapshot due in %" PRId64 " seconds\n", + next_snapshot_time - now); + if (current_snapshot_creation_time != 0) + fprintf(log, "current_snapshot_creation_time: %" + PRId64 " (%" PRId64 " seconds ago)\n", + current_snapshot_creation_time, + now - current_snapshot_creation_time + ); + if (next_removal_check.tv_sec != 0) { + fprintf(log, "next removal check: %llu (%llu seconds ago)\n", + (long long unsigned)next_removal_check.tv_sec, + now - (long long unsigned)next_removal_check.tv_sec + ); + + } + fprintf(log, "%s %s\n", dash, msg, dash); +} /* a litte cpp magic helps to DRY */ #define COMMANDS \ COMMAND(ls) \ COMMAND(create) \ COMMAND(prune) \ - COMMAND(run) + COMMAND(run) \ + COMMAND(kill) #define COMMAND(x) static int com_ ##x(void); COMMANDS #undef COMMAND @@ -127,23 +228,35 @@ static __printf_1_2 void dss_msg(const char* fmt,...) va_end(argp); } -static int disk_space_low(void) +static char *get_config_file_name(void) { - struct disk_space ds; - int ret = get_disk_space(".", &ds); + char *home, *config_file; + + if (conf.config_file_given) + return dss_strdup(conf.config_file_arg); + home = get_homedir(); + config_file = make_message("%s/.dssrc", home); + free(home); + return config_file; +} + +static int com_kill(void) +{ + pid_t pid; + char *config_file = get_config_file_name(); + int ret = get_dss_pid(config_file, &pid); + free(config_file); if (ret < 0) return ret; - if (conf.min_free_mb_arg) - if (ds.free_mb < conf.min_free_mb_arg) - return 1; - if (conf.min_free_percent_arg) - if (ds.percent_free < conf.min_free_percent_arg) - return 1; - if (conf.min_free_percent_inodes_arg) - if (ds.percent_free_inodes < conf.min_free_percent_inodes_arg) - return 1; - return 0; + if (conf.dry_run_given) { + dss_msg("%d\n", (int)pid); + return 0; + } + ret = kill(pid, SIGTERM); + if (ret < 0) + return -ERRNO_TO_DSS_ERROR(errno); + return 1; } static void dss_get_snapshot_list(struct snapshot_list *sl) @@ -151,22 +264,16 @@ static void dss_get_snapshot_list(struct snapshot_list *sl) get_snapshot_list(sl, conf.unit_interval_arg, conf.num_intervals_arg); } -static int next_snapshot_is_due(void) +static int64_t compute_next_snapshot_time(void) { - struct timeval now, unit_interval = {.tv_sec = 24 * 3600 * conf.unit_interval_arg}, - tmp, diff; - int64_t x = 0; + int64_t x = 0, now = get_current_time(), unit_interval + = 24 * 3600 * conf.unit_interval_arg, ret; unsigned wanted = desired_number_of_snapshots(0, conf.num_intervals_arg), num_complete_snapshots = 0; - int i, ret; + int i; struct snapshot *s = NULL; struct snapshot_list sl; - gettimeofday(&now, NULL); - if (tv_diff(&next_snapshot_time, &now, NULL) > 0) - return 0; - assert(snapshot_creation_status == HS_READY); - current_snapshot_creation_time = 0; dss_get_snapshot_list(&sl); FOR_EACH_SNAPSHOT(s, i, &sl) { if (!(s->flags & SS_COMPLETE)) @@ -175,41 +282,51 @@ static int next_snapshot_is_due(void) x += s->completion_time - s->creation_time; } assert(x >= 0); - if (num_complete_snapshots) - x /= num_complete_snapshots; /* avg time to create one snapshot */ - x *= wanted; /* time to create all snapshots in interval 0 */ - tmp.tv_sec = x; - tmp.tv_usec = 0; - ret = tv_diff(&unit_interval, &tmp, &diff); /* total sleep time per unit interval */ - if (ret < 0 || !s) { /* unit_interval < tmp or no snapshot */ - ret = 1; - goto out; - } - tv_divide(wanted, &diff, &tmp); /* sleep time between two snapshots */ - diff.tv_sec = s->completion_time; /* completion time of the latest snapshot */ - diff.tv_usec = 0; - tv_add(&diff, &tmp, &next_snapshot_time); - ret = (tv_diff(&now, &next_snapshot_time, &diff) < 0)? 0 : 1; + ret = now; + if (num_complete_snapshots == 0) + goto out; + x /= num_complete_snapshots; /* avg time to create one snapshot */ + if (unit_interval < x * wanted) /* oops, no sleep at all */ + goto out; + ret = s->completion_time + unit_interval / wanted - x; out: free_snapshot_list(&sl); - if (ret == 0) - DSS_DEBUG_LOG("next snapshot due in %lu seconds\n", - tv2ms(&diff) / 1000); - else - DSS_DEBUG_LOG("next snapshot: now\n"); return ret; } +static inline void invalidate_next_snapshot_time(void) +{ + next_snapshot_time = 0; +} + +static inline int next_snapshot_time_is_valid(void) +{ + return next_snapshot_time != 0; +} + +static int next_snapshot_is_due(void) +{ + int64_t now = get_current_time(); + + if (!next_snapshot_time_is_valid()) + next_snapshot_time = compute_next_snapshot_time(); + if (next_snapshot_time <= now) { + DSS_DEBUG_LOG("next snapshot: now\n"); + return 1; + } + DSS_DEBUG_LOG("next snapshot due in %" PRId64 " seconds\n", + next_snapshot_time - now); + return 0; +} + static int pre_create_hook(void) { int ret, fds[3] = {0, 0, 0}; assert(snapshot_creation_status == HS_READY); - if (!conf.pre_create_hook_given) { - snapshot_creation_status = HS_PRE_SUCCESS; - return 0; - } + /* make sure that the next snapshot time will be recomputed */ + invalidate_next_snapshot_time(); DSS_DEBUG_LOG("executing %s\n", conf.pre_create_hook_arg); ret = dss_exec_cmdline_pid(&create_pid, conf.pre_create_hook_arg, fds); @@ -235,15 +352,10 @@ static int pre_remove_hook(struct snapshot *s, const char *why) *snapshot_currently_being_removed = *s; snapshot_currently_being_removed->name = dss_strdup(s->name); - if (!conf.pre_remove_hook_given) { - snapshot_removal_status = HS_PRE_SUCCESS; - return 0; - } cmd = make_message("%s %s/%s", conf.pre_remove_hook_arg, conf.dest_dir_arg, s->name); DSS_DEBUG_LOG("executing %s\n", cmd); - ret = dss_exec_cmdline_pid(&remove_pid, - conf.pre_remove_hook_arg, fds); + ret = dss_exec_cmdline_pid(&remove_pid, cmd, fds); free(cmd); if (ret < 0) return ret; @@ -385,14 +497,14 @@ static struct snapshot *find_outdated_snapshot(struct snapshot_list *sl) int i; struct snapshot *s; - DSS_DEBUG_LOG("looking for snapshots belonging to intervals greater than %d\n", + DSS_DEBUG_LOG("looking for snapshots belonging to intervals >= %d\n", conf.num_intervals_arg); FOR_EACH_SNAPSHOT(s, i, sl) { if (snapshot_is_being_created(s)) continue; if (is_reference_snapshot(s)) continue; - if (s->interval <= conf.num_intervals_arg) + if (s->interval < conf.num_intervals_arg) continue; return s; } @@ -442,7 +554,7 @@ static int try_to_free_disk_space(void) const char *why; int low_disk_space; - ret = disk_space_low(); + ret = disk_space_low(NULL); if (ret < 0) return ret; low_disk_space = ret; @@ -495,10 +607,6 @@ static int post_create_hook(void) int ret, fds[3] = {0, 0, 0}; char *cmd; - if (!conf.post_create_hook_given) { - snapshot_creation_status = HS_READY; - return 0; - } cmd = make_message("%s %s/%s", conf.post_create_hook_arg, conf.dest_dir_arg, path_to_last_complete_snapshot); DSS_NOTICE_LOG("executing %s\n", cmd); @@ -518,10 +626,6 @@ static int post_remove_hook(void) assert(s); - if (!conf.post_remove_hook_given) { - snapshot_removal_status = HS_READY; - return 0; - } cmd = make_message("%s %s/%s", conf.post_remove_hook_arg, conf.dest_dir_arg, s->name); DSS_NOTICE_LOG("executing %s\n", cmd); @@ -533,27 +637,48 @@ static int post_remove_hook(void) return ret; } -static void kill_process(pid_t pid) +static void dss_kill(pid_t pid, int sig, const char *msg) { - if (!pid) + const char *signame, *process_name; + + if (pid == 0) + return; + switch (sig) { + case SIGTERM: signame = "TERM"; break; + case SIGSTOP: signame = "STOP"; break; + case SIGCONT: signame = "CONT"; break; + default: signame = "????"; + } + + if (pid == create_pid) + process_name = "create"; + else if (pid == remove_pid) + process_name = "remove"; + else process_name = "??????"; + + if (msg) + DSS_INFO_LOG("%s\n", msg); + DSS_DEBUG_LOG("sending signal %d (%s) to pid %d (%s process)\n", + sig, signame, (int)pid, process_name); + if (kill(pid, sig) >= 0) return; - DSS_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid); - kill(pid, SIGTERM); + DSS_INFO_LOG("failed to send signal %d (%s) to pid %d (%s process)\n", + sig, signame, (int)pid, process_name); } static void stop_create_process(void) { - if (!create_pid || create_process_stopped) + if (create_process_stopped) return; - kill(SIGSTOP, create_pid); + dss_kill(create_pid, SIGSTOP, "suspending create process"); create_process_stopped = 1; } static void restart_create_process(void) { - if (!create_pid || !create_process_stopped) + if (!create_process_stopped) return; - kill (SIGCONT, create_pid); + dss_kill(create_pid, SIGCONT, "resuming create process"); create_process_stopped = 0; } @@ -598,8 +723,7 @@ static int wait_for_process(pid_t pid, int *status) } } /* SIGINT or SIGTERM */ - DSS_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid); - kill(pid, SIGTERM); + dss_kill(pid, SIGTERM, "killing child process"); } if (ret < 0) DSS_ERROR_LOG("failed to wait for process %d\n", (int)pid); @@ -629,10 +753,7 @@ static int handle_rm_exit(int status) snapshot_removal_status = HS_READY; return -E_BAD_EXIT_CODE; } - if (conf.post_remove_hook_given) - snapshot_removal_status = HS_SUCCESS; - else - snapshot_removal_status = HS_READY; + snapshot_removal_status = HS_SUCCESS; return 1; } @@ -707,8 +828,7 @@ static int handle_rsync_exit(int status) DSS_WARNING_LOG("rsync process %d returned %d -- restarting\n", (int)create_pid, es); snapshot_creation_status = HS_NEEDS_RESTART; - gettimeofday(&next_snapshot_time, NULL); - next_snapshot_time.tv_sec += 60; + next_snapshot_time = get_current_time() + 60; ret = 1; goto out; } @@ -725,7 +845,6 @@ static int handle_rsync_exit(int status) free(name_of_reference_snapshot); name_of_reference_snapshot = NULL; out: - create_pid = 0; create_process_stopped = 0; return ret; } @@ -748,8 +867,7 @@ static int handle_pre_create_hook_exit(int status) DSS_NOTICE_LOG("deferring snapshot creation...\n"); warn_count = 60; /* warn only once per hour */ } - gettimeofday(&next_snapshot_time, NULL); - next_snapshot_time.tv_sec += 60; + next_snapshot_time = get_current_time() + 60; snapshot_creation_status = HS_READY; ret = 0; goto out; @@ -758,7 +876,6 @@ static int handle_pre_create_hook_exit(int status) snapshot_creation_status = HS_PRE_SUCCESS; ret = 1; out: - create_pid = 0; return ret; } @@ -773,17 +890,22 @@ static int handle_sigchld(void) if (pid == create_pid) { switch (snapshot_creation_status) { case HS_PRE_RUNNING: - return handle_pre_create_hook_exit(status); + ret = handle_pre_create_hook_exit(status); + break; case HS_RUNNING: - return handle_rsync_exit(status); + ret = handle_rsync_exit(status); + break; case HS_POST_RUNNING: snapshot_creation_status = HS_READY; - return 1; + ret = 1; + break; default: DSS_EMERG_LOG("BUG: create can't die in status %d\n", snapshot_creation_status); return -E_BUG; } + create_pid = 0; + return ret; } if (pid == remove_pid) { ret = handle_remove_exit(status); @@ -810,34 +932,31 @@ static int check_config(void) return 1; } +/* + * Returns < 0 on errors, 0 if no config file is given and > 0 if the config + * file was read successfully. + */ static int parse_config_file(int override) { - int ret; - char *config_file; + int ret, config_file_exists; + char *config_file = get_config_file_name(); struct stat statbuf; char *old_logfile_arg = NULL; int old_daemon_given = 0; - if (conf.config_file_given) - config_file = dss_strdup(conf.config_file_arg); - else { - char *home = get_homedir(); - config_file = make_message("%s/.dssrc", home); - free(home); - } if (override) { /* SIGHUP */ if (conf.logfile_given) old_logfile_arg = dss_strdup(conf.logfile_arg); old_daemon_given = conf.daemon_given; } - ret = stat(config_file, &statbuf); - if (ret && conf.config_file_given) { + config_file_exists = !stat(config_file, &statbuf); + if (!config_file_exists && conf.config_file_given) { ret = -ERRNO_TO_DSS_ERROR(errno); DSS_ERROR_LOG("failed to stat config file %s\n", config_file); goto out; } - if (!ret) { + if (config_file_exists) { struct cmdline_parser_params params = { .override = override, .initialize = 0, @@ -871,12 +990,12 @@ static int parse_config_file(int override) conf.logfile_given = 1; } } - if (conf.logfile_given) { + if (conf.logfile_given && conf.run_given && conf.daemon_given) { logfile = open_log(conf.logfile_arg); log_welcome(conf.loglevel_arg); } DSS_DEBUG_LOG("loglevel: %d\n", conf.loglevel_arg); -// cmdline_parser_dump(logfile? logfile : stdout, &conf); + ret = config_file_exists; out: free(config_file); if (ret < 0) @@ -894,10 +1013,13 @@ static int handle_sighup(void) { int ret; - DSS_NOTICE_LOG("SIGHUP\n"); + DSS_NOTICE_LOG("SIGHUP, re-reading config\n"); + dump_dss_config("old"); ret = parse_config_file(1); if (ret < 0) return ret; + dump_dss_config("reloaded"); + invalidate_next_snapshot_time(); return change_to_dest_dir(); } @@ -912,8 +1034,8 @@ static int handle_signal(void) case SIGINT: case SIGTERM: restart_create_process(); - kill_process(create_pid); - kill_process(remove_pid); + dss_kill(create_pid, SIGTERM, NULL); + dss_kill(remove_pid, SIGTERM, NULL); ret = -E_SIGNAL; break; case SIGHUP: @@ -944,6 +1066,58 @@ static int use_rsync_locally(char *logname) return 1; } +static int rename_resume_snap(int64_t creation_time) +{ + struct snapshot_list sl = {.num_snapshots = 0}; + struct snapshot *s = NULL; + char *new_name = incomplete_name(creation_time); + int ret; + const char *why; + + ret = 0; + if (conf.no_resume_given) + goto out; + dss_get_snapshot_list(&sl); + /* + * Snapshot recycling: We first look at the newest snapshot. If this + * snapshot happens to be incomplete, the last rsync process was + * aborted and we reuse this one. Otherwise we look at snapshots which + * could be removed (outdated and redundant snapshots) as candidates + * for recycling. If no outdated/redundant snapshot exists, we check if + * there is an orphaned snapshot, which likely is useless anyway. + * + * Only if no existing snapshot is suitable for recycling, we bite the + * bullet and create a new one. + */ + s = get_newest_snapshot(&sl); + if (!s) /* no snapshots at all */ + goto out; + /* re-use last snapshot if it is incomplete */ + why = "aborted"; + if ((s->flags & SS_COMPLETE) == 0) + goto out; + why = "outdated"; + s = find_outdated_snapshot(&sl); + if (s) + goto out; + why = "redundant"; + s = find_redundant_snapshot(&sl); + if (s) + goto out; + why = "orphaned"; + s = find_orphaned_snapshot(&sl); +out: + if (s) { + DSS_INFO_LOG("reusing %s snapshot %s\n", why, s->name); + ret = dss_rename(s->name, new_name); + } + if (ret >= 0) + DSS_NOTICE_LOG("creating new snapshot %s\n", new_name); + free(new_name); + free_snapshot_list(&sl); + return ret; +} + static void create_rsync_argv(char ***argv, int64_t *num) { char *logname; @@ -996,11 +1170,10 @@ static void free_rsync_argv(char **argv) static int create_snapshot(char **argv) { int ret, fds[3] = {0, 0, 0}; - char *name; - name = incomplete_name(current_snapshot_creation_time); - DSS_NOTICE_LOG("creating new snapshot %s\n", name); - free(name); + ret = rename_resume_snap(current_snapshot_creation_time); + if (ret < 0) + return ret; ret = dss_exec(&create_pid, argv[0], argv, fds); if (ret < 0) return ret; @@ -1028,7 +1201,6 @@ static int select_loop(void) } FD_ZERO(&rfds); FD_SET(signal_pipe, &rfds); - DSS_DEBUG_LOG("tvp: %p, tv_sec : %lu\n", tvp, (long unsigned) tv.tv_sec); ret = dss_select(signal_pipe + 1, &rfds, NULL, tvp); if (ret < 0) goto out; @@ -1072,10 +1244,17 @@ static int select_loop(void) case HS_POST_RUNNING: continue; case HS_PRE_SUCCESS: - free_rsync_argv(rsync_argv); - create_rsync_argv(&rsync_argv, ¤t_snapshot_creation_time); - /* fall through */ + if (!name_of_reference_snapshot) { + free_rsync_argv(rsync_argv); + create_rsync_argv(&rsync_argv, ¤t_snapshot_creation_time); + } + ret = create_snapshot(rsync_argv); + if (ret < 0) + goto out; + continue; case HS_NEEDS_RESTART: + if (!next_snapshot_is_due()) + continue; ret = create_snapshot(rsync_argv); if (ret < 0) goto out; @@ -1097,16 +1276,27 @@ static void exit_hook(int exit_code) char *argv[] = {conf.exit_hook_arg, dss_strerror(-exit_code), NULL}; pid_t pid; - if (!conf.exit_hook_given) - return; DSS_NOTICE_LOG("executing %s %s\n", argv[0], argv[1]); dss_exec(&pid, conf.exit_hook_arg, argv, fds); } +static void lock_dss_or_die(void) +{ + char *config_file = get_config_file_name(); + int ret = lock_dss(config_file); + + free(config_file); + if (ret < 0) { + DSS_EMERG_LOG("failed to lock: %s\n", dss_strerror(-ret)); + exit(EXIT_FAILURE); + } +} + static int com_run(void) { int ret; + lock_dss_or_die(); if (conf.dry_run_given) { DSS_ERROR_LOG("dry_run not supported by this command\n"); return -E_SYNTAX; @@ -1129,6 +1319,7 @@ static int com_prune(void) struct disk_space ds; const char *why; + lock_dss_or_die(); ret = get_disk_space(".", &ds); if (ret < 0) return ret; @@ -1188,6 +1379,7 @@ static int com_create(void) int ret, status; char **rsync_argv; + lock_dss_or_die(); if (conf.dry_run_given) { int i; char *msg = NULL; @@ -1285,11 +1477,26 @@ int main(int argc, char **argv) ret = parse_config_file(0); if (ret < 0) goto out; + if (ret == 0) { /* no config file given */ + /* + * Parse the command line options again, but this time check + * that all required options are given. + */ + params = (struct cmdline_parser_params) { + .override = 1, + .initialize = 1, + .check_required = 1, + .check_ambiguity = 1, + .print_errors = 1 + }; + cmdline_parser_ext(argc, argv, &conf, ¶ms); /* aborts on errors */ + } if (conf.daemon_given) daemon_init(); ret = change_to_dest_dir(); if (ret < 0) goto out; + dump_dss_config("startup"); ret = setup_signal_handling(); if (ret < 0) goto out;