2 * Amanda, The Advanced Maryland Automatic Network Disk Archiver
3 * Copyright (c) 1991-1998 University of Maryland at College Park
6 * Permission to use, copy, modify, distribute, and sell this software and its
7 * documentation for any purpose is hereby granted without fee, provided that
8 * the above copyright notice appear in all copies and that both that
9 * copyright notice and this permission notice appear in supporting
10 * documentation, and that the name of U.M. not be used in advertising or
11 * publicity pertaining to distribution of the software without specific,
12 * written prior permission. U.M. makes no representations about the
13 * suitability of this software for any purpose. It is provided "as is"
14 * without express or implied warranty.
16 * U.M. DISCLAIMS ALL WARRANTIES WITH REGARD TO THIS SOFTWARE, INCLUDING ALL
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS, IN NO EVENT SHALL U.M.
18 * BE LIABLE FOR ANY SPECIAL, INDIRECT OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
19 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION
20 * OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN
21 * CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
23 * Authors: the Amanda Development Team. Its members are listed in a
24 * file named AUTHORS, in the root directory of this distribution.
27 * $Id: driver.c 6512 2007-05-24 17:00:24Z ian $
29 * controlling process for the Amanda backup system
33 * XXX possibly modify tape queue to be cognizant of how much room is left on
34 * tape. Probably not effective though, should do this in planner.
48 #include "server_util.h"
49 #include "timestamp.h"
51 #define driver_debug(i, ...) do { \
52 if ((i) <= debug_driver) { \
53 dbprintf(__VA_ARGS__); \
57 #define hold_debug(i, ...) do { \
58 if ((i) <= debug_holding) { \
59 dbprintf(__VA_ARGS__); \
63 static disklist_t waitq; // dle waiting estimate result
64 static disklist_t runq; // dle waiting to be dumped to holding disk
65 static disklist_t directq; // dle waiting to be dumped directly to tape
66 static disklist_t tapeq; // dle on holding disk waiting to be written
68 static disklist_t roomq; // dle waiting for more space on holding disk
69 static int pending_aborts;
70 static disk_t *taper_disk;
71 static int degraded_mode;
72 static off_t reserved_space;
73 static off_t total_disksize;
74 static char *dumper_program;
75 static char *chunker_program;
76 static int inparallel;
77 static int nodump = 0;
78 static off_t tape_length = (off_t)0;
79 static off_t tape_left = (off_t)0;
80 static int current_tape = 0;
81 static int conf_taperalgo;
82 static int conf_runtapes;
83 static time_t sleep_time;
84 static int idle_reason;
85 static char *driver_timestamp;
86 static char *hd_driver_timestamp;
87 static am_host_t *flushhost = NULL;
88 static int need_degraded=0;
89 static holdalloc_t *holdalloc;
90 static int num_holdalloc;
91 static event_handle_t *dumpers_ev_time = NULL;
92 static event_handle_t *schedule_ev_read = NULL;
93 static int conf_flush_threshold_dumped;
94 static int conf_flush_threshold_scheduled;
95 static int conf_taperflush;
96 static off_t flush_threshold_dumped;
97 static off_t flush_threshold_scheduled;
98 static off_t taperflush;
99 static int schedule_done; // 1 if we don't wait for a
100 // schedule from the planner
101 static int force_flush; // All dump are terminated, we
102 // must now respect taper_flush
104 static int wait_children(int count);
105 static void wait_for_children(void);
106 static void allocate_bandwidth(netif_t *ip, unsigned long kps);
107 static int assign_holdingdisk(assignedhd_t **holdp, disk_t *diskp);
108 static void adjust_diskspace(disk_t *diskp, cmd_t cmd);
109 static void delete_diskspace(disk_t *diskp);
110 static assignedhd_t **build_diskspace(char *destname);
111 static int client_constrained(disk_t *dp);
112 static void deallocate_bandwidth(netif_t *ip, unsigned long kps);
113 static void dump_schedule(disklist_t *qp, char *str);
114 static void dump_to_tape(disk_t *dp);
115 static assignedhd_t **find_diskspace(off_t size, int *cur_idle,
116 assignedhd_t *preferred);
117 static unsigned long free_kps(netif_t *ip);
118 static off_t free_space(void);
119 static void dumper_chunker_result(disk_t *dp);
120 static void dumper_taper_result(disk_t *dp);
121 static void file_taper_result(disk_t *dp);
122 static void handle_dumper_result(void *);
123 static void handle_chunker_result(void *);
124 static void handle_dumpers_time(void *);
125 static void handle_taper_result(void *);
127 static void holdingdisk_state(char *time_str);
128 static dumper_t *idle_dumper(void);
129 static void interface_state(char *time_str);
130 static int queue_length(disklist_t q);
131 static disklist_t read_flush(void);
132 static void read_schedule(void *cookie);
133 static void short_dump_state(void);
134 static void startaflush(void);
135 static void start_degraded_mode(disklist_t *queuep);
136 static void start_some_dumps(disklist_t *rq);
137 static void continue_port_dumps(void);
138 static void update_failed_dump_to_tape(disk_t *);
141 TAPE_ACTION_NO_ACTION = 0,
142 TAPE_ACTION_NEW_TAPE = (1 << 0),
143 TAPE_ACTION_NO_NEW_TAPE = (1 << 1),
144 TAPE_ACTION_START_A_FLUSH = (1 << 2)
147 static TapeAction tape_action(void);
149 static const char *idle_strings[] = {
152 #define IDLE_NO_DUMPERS 1
154 #define IDLE_START_WAIT 2
156 #define IDLE_NO_HOLD 3
158 #define IDLE_CLIENT_CONSTRAINED 4
159 T_("client-constrained"),
160 #define IDLE_NO_BANDWIDTH 5
162 #define IDLE_NO_DISKSPACE 6
176 struct fs_usage fsusage;
178 unsigned long reserve = 100;
182 char *result_argv[MAX_ARGS+1];
189 config_overwrites_t *cfg_ovr = NULL;
190 char *cfg_opt = NULL;
191 holdalloc_t *ha, *ha_last;
194 * Configure program for internationalization:
195 * 1) Only set the message locale for now.
196 * 2) Set textdomain for all amanda related programs to "amanda"
197 * We don't want to be forced to support dozens of message catalogs.
199 setlocale(LC_MESSAGES, "C");
200 textdomain("amanda");
204 setvbuf(stdout, (char *)NULL, (int)_IOLBF, 0);
205 setvbuf(stderr, (char *)NULL, (int)_IOLBF, 0);
209 dbopen(DBG_SUBDIR_SERVER);
211 atexit(wait_for_children);
213 /* Don't die when child closes pipe */
214 signal(SIGPIPE, SIG_IGN);
216 erroutput_type = (ERR_AMANDALOG|ERR_INTERACTIVE);
217 set_logerror(logerror);
221 cfg_ovr = extract_commandline_config_overwrites(&argc, &argv);
225 config_init(CONFIG_INIT_EXPLICIT_NAME | CONFIG_INIT_USE_CWD | CONFIG_INIT_FATAL,
227 apply_config_overwrites(cfg_ovr);
229 g_printf(_("%s: pid %ld executable %s version %s\n"),
230 get_pname(), (long) getpid(), argv[0], version());
233 if(strncmp(argv[2], "nodump", 6) == 0) {
238 safe_cd(); /* do this *after* config_init */
240 check_running_as(RUNNING_AS_DUMPUSER);
242 dbrename(config_name, DBG_SUBDIR_SERVER);
244 amfree(driver_timestamp);
245 /* read timestamp from stdin */
246 while ((line = agets(stdin)) != NULL) {
251 if ( line == NULL ) {
252 error(_("Did not get DATE line from planner"));
255 driver_timestamp = alloc(15);
256 strncpy(driver_timestamp, &line[5], 14);
257 driver_timestamp[14] = '\0';
259 log_add(L_START,_("date %s"), driver_timestamp);
261 gethostname(hostname, SIZEOF(hostname));
262 log_add(L_STATS,_("hostname %s"), hostname);
264 /* check that we don't do many dump in a day and usetimestamps is off */
265 if(strlen(driver_timestamp) == 8) {
267 char *conf_logdir = getconf_str(CNF_LOGDIR);
268 char *logfile = vstralloc(conf_logdir, "/log.",
269 driver_timestamp, ".0", NULL);
270 char *oldlogfile = vstralloc(conf_logdir, "/oldlog/log.",
271 driver_timestamp, ".0", NULL);
272 if(access(logfile, F_OK) == 0 || access(oldlogfile, F_OK) == 0) {
273 log_add(L_WARNING, _("WARNING: This is not the first amdump run today. Enable the usetimestamps option in the configuration file if you want to run amdump more than once per calendar day."));
278 hd_driver_timestamp = get_timestamp_from_time(0);
281 hd_driver_timestamp = stralloc(driver_timestamp);
284 taper_program = vstralloc(amlibexecdir, "/", "taper", versionsuffix(), NULL);
285 dumper_program = vstralloc(amlibexecdir, "/", "dumper", versionsuffix(),
287 chunker_program = vstralloc(amlibexecdir, "/", "chunker", versionsuffix(),
290 conf_taperalgo = getconf_taperalgo(CNF_TAPERALGO);
291 conf_tapetype = getconf_str(CNF_TAPETYPE);
292 conf_runtapes = getconf_int(CNF_RUNTAPES);
293 tape = lookup_tapetype(conf_tapetype);
294 tape_length = tapetype_get_length(tape);
295 conf_flush_threshold_dumped = getconf_int(CNF_FLUSH_THRESHOLD_DUMPED);
296 conf_flush_threshold_scheduled = getconf_int(CNF_FLUSH_THRESHOLD_SCHEDULED);
297 conf_taperflush = getconf_int(CNF_TAPERFLUSH);
299 flush_threshold_dumped = (conf_flush_threshold_dumped * tape_length) / 100;
300 flush_threshold_scheduled = (conf_flush_threshold_scheduled * tape_length) / 100;
301 taperflush = (conf_taperflush *tape_length) / 100;
303 driver_debug(1, _("flush_threshold_dumped: %lld\n"), (long long)flush_threshold_dumped);
304 driver_debug(1, _("flush_threshold_scheduled: %lld\n"), (long long)flush_threshold_scheduled);
305 driver_debug(1, _("taperflush: %lld\n"), (long long)taperflush);
307 /* start initializing: read in databases */
309 conf_diskfile = config_dir_relative(getconf_str(CNF_DISKFILE));
310 if (read_diskfile(conf_diskfile, &origq) < 0) {
311 error(_("could not load disklist \"%s\""), conf_diskfile);
314 amfree(conf_diskfile);
316 /* set up any configuration-dependent variables */
318 inparallel = getconf_int(CNF_INPARALLEL);
320 reserve = (unsigned long)getconf_int(CNF_RESERVE);
322 total_disksize = (off_t)0;
325 for(hdp = getconf_holdingdisks(), dsk = 0; hdp != NULL; hdp = holdingdisk_next(hdp), dsk++) {
326 ha = alloc(SIZEOF(holdalloc_t));
329 /* link the list in the same order as getconf_holdingdisks's results */
338 ha->allocated_dumpers = 0;
339 ha->allocated_space = (off_t)0;
340 ha->disksize = holdingdisk_get_disksize(hdp);
343 if(get_fs_usage(holdingdisk_get_diskdir(hdp), NULL, &fsusage) == -1
344 || access(holdingdisk_get_diskdir(hdp), W_OK) == -1) {
345 log_add(L_WARNING, _("WARNING: ignoring holding disk %s: %s\n"),
346 holdingdisk_get_diskdir(hdp), strerror(errno));
351 /* do the division first to avoid potential integer overflow */
352 if (fsusage.fsu_bavail_top_bit_set)
355 kb_avail = fsusage.fsu_bavail / 1024 * fsusage.fsu_blocksize;
357 if(ha->disksize > (off_t)0) {
358 if(ha->disksize > kb_avail) {
360 _("WARNING: %s: %lld KB requested, "
361 "but only %lld KB available."),
362 holdingdisk_get_diskdir(hdp),
363 (long long)ha->disksize,
364 (long long)kb_avail);
365 ha->disksize = kb_avail;
368 /* ha->disksize is negative; use all but that amount */
369 else if(kb_avail < -ha->disksize) {
371 _("WARNING: %s: not %lld KB free."),
372 holdingdisk_get_diskdir(hdp),
373 (long long)-ha->disksize);
374 ha->disksize = (off_t)0;
378 ha->disksize += kb_avail;
380 g_printf(_("driver: adding holding disk %d dir %s size %lld chunksize %lld\n"),
381 dsk, holdingdisk_get_diskdir(hdp),
382 (long long)ha->disksize,
383 (long long)(holdingdisk_get_chunksize(hdp)));
385 newdir = newvstralloc(newdir,
386 holdingdisk_get_diskdir(hdp), "/", hd_driver_timestamp,
388 if(!mkholdingdir(newdir)) {
389 ha->disksize = (off_t)0;
391 total_disksize += ha->disksize;
394 reserved_space = total_disksize * (off_t)(reserve / 100);
396 g_printf(_("reserving %lld out of %lld for degraded-mode dumps\n"),
397 (long long)reserved_space, (long long)free_space());
401 if(inparallel > MAX_DUMPERS) inparallel = MAX_DUMPERS;
403 /* taper takes a while to get going, so start it up right away */
406 if(conf_runtapes > 0) {
407 startup_tape_process(taper_program);
408 taper_cmd(START_TAPER, driver_timestamp, NULL, 0, NULL);
411 /* fire up the dumpers now while we are waiting */
412 if(!nodump) startup_dump_processes(dumper_program, inparallel, driver_timestamp);
415 * Read schedule from stdin. Usually, this is a pipe from planner,
416 * so the effect is that we wait here for the planner to
417 * finish, but meanwhile the taper is rewinding the tape, reading
418 * the label, checking it, writing a new label and all that jazz
419 * in parallel with the planner.
427 taper_state = TAPER_STATE_DEFAULT;
428 tapeq = read_flush();
430 roomq.head = roomq.tail = NULL;
432 log_add(L_STATS, _("startup time %s"), walltime_str(curclock()));
434 g_printf(_("driver: start time %s inparallel %d bandwidth %lu diskspace %lld "), walltime_str(curclock()), inparallel,
435 free_kps(NULL), (long long)free_space());
436 g_printf(_(" dir %s datestamp %s driver: drain-ends tapeq %s big-dumpers %s\n"),
437 "OBSOLETE", driver_timestamp, taperalgo2str(conf_taperalgo),
438 getconf_str(CNF_DUMPORDER));
441 /* ok, planner is done, now lets see if the tape is ready */
443 if (conf_runtapes > 0) {
444 cmd = getresult(taper, 1, &result_argc, result_argv, MAX_ARGS+1);
445 if (cmd != TAPER_OK) {
446 /* no tape, go into degraded mode: dump to holding disk */
453 tape_left = tape_length;
455 taper_input_error = NULL;
456 taper_tape_error = NULL;
458 taper_ev_read = NULL;
460 schedule_done = nodump;
463 if(!need_degraded) startaflush();
466 schedule_ev_read = event_register((event_id_t)0, EV_READFD, read_schedule, NULL);
473 /* mv runq to directq */
474 while (!empty(runq)) {
475 diskp = dequeue_disk(&runq);
476 headqueue_disk(&directq, diskp);
479 /* handle any remaining dumps by dumping directly to tape, if possible */
480 while(!empty(directq) && taper > 0) {
481 diskp = dequeue_disk(&directq);
482 if (diskp->to_holdingdisk == HOLD_REQUIRED) {
483 char *qname = quote_string(diskp->name);
484 log_add(L_FAIL, _("%s %s %s %d [%s]"),
485 diskp->host->hostname, qname, sched(diskp)->datestamp,
487 _("can't dump required holdingdisk"));
490 else if (!degraded_mode) {
491 taper_state |= TAPER_STATE_DUMP_TO_TAPE;
494 taper_state &= !TAPER_STATE_DUMP_TO_TAPE;
497 char *qname = quote_string(diskp->name);
498 log_add(L_FAIL, _("%s %s %s %d [%s]"),
499 diskp->host->hostname, qname, sched(diskp)->datestamp,
501 diskp->to_holdingdisk == HOLD_AUTO ?
502 _("no more holding disk space") :
503 _("can't dump no-hold disk in degraded mode"));
508 /* fill up the tape or start new one for taperflush */
512 short_dump_state(); /* for amstatus */
514 g_printf(_("driver: QUITTING time %s telling children to quit\n"),
515 walltime_str(curclock()));
519 for(dumper = dmptable; dumper < dmptable + inparallel; dumper++) {
521 dumper_cmd(dumper, QUIT, NULL);
526 taper_cmd(QUIT, NULL, NULL, 0, NULL);
529 /* wait for all to die */
533 holding_cleanup(NULL, NULL);
537 check_unfree_serial();
538 g_printf(_("driver: FINISHED time %s\n"), walltime_str(curclock()));
540 log_add(L_FINISH,_("date %s time %s"), driver_timestamp, walltime_str(curclock()));
541 amfree(driver_timestamp);
543 amfree(dumper_program);
544 amfree(taper_program);
551 /* sleep up to count seconds, and wait for terminating child process */
552 /* if sleep is negative, this function will not timeout */
553 /* exit once all child process are finished or the timout expired */
554 /* return 0 if no more children to wait */
555 /* return 1 if some children are still alive */
557 wait_children(int count)
569 pid = waitpid((pid_t)-1, &retstat, WNOHANG);
573 if (! WIFEXITED(retstat)) {
575 code = WTERMSIG(retstat);
576 } else if (WEXITSTATUS(retstat) != 0) {
578 code = WEXITSTATUS(retstat);
581 for (dumper = dmptable; dumper < dmptable + inparallel;
583 if (pid == dumper->pid) {
584 who = stralloc(dumper->name);
588 if (dumper->chunker && pid == dumper->chunker->pid) {
589 who = stralloc(dumper->chunker->name);
590 dumper->chunker->pid = -1;
594 if (who == NULL && pid == taper_pid) {
595 who = stralloc("taper");
598 if(what != NULL && who == NULL) {
599 who = stralloc("unknown");
602 log_add(L_WARNING, _("%s pid %u exited with %s %d\n"), who,
603 (unsigned)pid, what, code);
604 g_printf(_("driver: %s pid %u exited with %s %d\n"), who,
605 (unsigned)pid, what, code);
609 } while (pid > 0 || wait_errno == EINTR);
614 } while ((errno != ECHILD) && (count != 0));
615 return (errno != ECHILD);
619 kill_children(int signal)
624 for(dumper = dmptable; dumper < dmptable + inparallel; dumper++) {
625 if (!dumper->down && dumper->pid > 1) {
626 g_printf(_("driver: sending signal %d to %s pid %u\n"), signal,
627 dumper->name, (unsigned)dumper->pid);
628 if (kill(dumper->pid, signal) == -1 && errno == ESRCH) {
630 dumper->chunker->pid = 0;
632 if (dumper->chunker && dumper->chunker->pid > 1) {
633 g_printf(_("driver: sending signal %d to %s pid %u\n"), signal,
634 dumper->chunker->name,
635 (unsigned)dumper->chunker->pid);
636 if (kill(dumper->chunker->pid, signal) == -1 &&
638 dumper->chunker->pid = 0;
645 g_printf(_("driver: sending signal %d to %s pid %u\n"), signal,
646 "taper", (unsigned)taper_pid);
647 if (kill(taper_pid, signal) == -1 && errno == ESRCH)
652 wait_for_children(void)
657 for(dumper = dmptable; dumper < dmptable + inparallel; dumper++) {
658 if (dumper->pid > 1 && dumper->fd >= 0) {
659 dumper_cmd(dumper, QUIT, NULL);
660 if (dumper->chunker && dumper->chunker->pid > 1 &&
661 dumper->chunker->fd >= 0)
662 chunker_cmd(dumper->chunker, QUIT, NULL);
667 if(taper_pid > 1 && taper > 0) {
668 taper_cmd(QUIT, NULL, NULL, 0, NULL);
671 if(wait_children(60) == 0)
674 kill_children(SIGHUP);
675 if(wait_children(60) == 0)
678 kill_children(SIGKILL);
679 if(wait_children(-1) == 0)
692 TapeAction result_tape_action;
694 result_tape_action = tape_action();
696 if (result_tape_action & TAPE_ACTION_NEW_TAPE) {
697 taper_state &= !TAPER_STATE_WAIT_FOR_TAPE;
698 taper_cmd(NEW_TAPE, NULL, NULL, 0, NULL);
699 } else if (result_tape_action & TAPE_ACTION_NO_NEW_TAPE) {
700 taper_state &= !TAPER_STATE_WAIT_FOR_TAPE;
701 taper_cmd(NO_NEW_TAPE, NULL, NULL, 0, NULL);
704 if (!degraded_mode && !taper_busy && !empty(tapeq) &&
705 (result_tape_action & TAPE_ACTION_START_A_FLUSH)) {
707 datestamp = sched(tapeq.head)->datestamp;
708 switch(conf_taperalgo) {
710 dp = dequeue_disk(&tapeq);
714 while (fit != NULL) {
715 extra_tapes = (fit->tape_splitsize > (off_t)0) ?
716 conf_runtapes - current_tape : 0;
717 if(sched(fit)->act_size <= (tape_left +
718 tape_length * (off_t)extra_tapes) &&
719 strcmp(sched(fit)->datestamp, datestamp) <= 0) {
727 if(dp) remove_disk(&tapeq, dp);
730 fit = dp = tapeq.head;
731 while (fit != NULL) {
732 if(sched(fit)->act_size > sched(dp)->act_size &&
733 strcmp(sched(fit)->datestamp, datestamp) <= 0) {
738 if(dp) remove_disk(&tapeq, dp);
740 case ALGO_LARGESTFIT:
742 while (fit != NULL) {
743 extra_tapes = (fit->tape_splitsize > (off_t)0) ?
744 conf_runtapes - current_tape : 0;
745 if(sched(fit)->act_size <=
746 (tape_left + tape_length * (off_t)extra_tapes) &&
747 (!dp || sched(fit)->act_size > sched(dp)->act_size) &&
748 strcmp(sched(fit)->datestamp, datestamp) <= 0) {
753 if(dp) remove_disk(&tapeq, dp);
759 remove_disk(&tapeq, dp);
762 if(!dp) { /* ALGO_SMALLEST, or default if nothing fit. */
763 if(conf_taperalgo != ALGO_SMALLEST) {
765 _("driver: startaflush: Using SMALLEST because nothing fit\n"));
767 fit = dp = tapeq.head;
768 while (fit != NULL) {
769 if(sched(fit)->act_size < sched(dp)->act_size &&
770 strcmp(sched(fit)->datestamp, datestamp) <= 0) {
775 if(dp) remove_disk(&tapeq, dp);
777 if(taper_ev_read == NULL) {
778 taper_ev_read = event_register((event_id_t)taper, EV_READFD,
779 handle_taper_result, NULL);
784 taper_input_error = NULL;
785 taper_tape_error = NULL;
786 taper_result = LAST_TOK;
787 taper_sendresult = 0;
788 taper_first_label = NULL;
790 taper_state &= !TAPER_STATE_DUMP_TO_TAPE;
792 qname = quote_string(dp->name);
793 taper_cmd(FILE_WRITE, dp, sched(dp)->destname, sched(dp)->level,
794 sched(dp)->datestamp);
795 g_fprintf(stderr,_("driver: startaflush: %s %s %s %lld %lld\n"),
796 taperalgo2str(conf_taperalgo), dp->host->hostname, qname,
797 (long long)sched(taper_disk)->act_size,
798 (long long)tape_left);
799 if(sched(dp)->act_size <= tape_left)
800 tape_left -= sched(dp)->act_size;
802 tape_left = (off_t)0;
805 error(_("FATAL: Taper marked busy and no work found."));
808 } else if(!taper_busy && taper_ev_read != NULL) {
809 event_release(taper_ev_read);
810 taper_ev_read = NULL;
820 /* first, check if host is too busy */
822 if(dp->host->inprogress >= dp->host->maxdumps) {
826 /* next, check conflict with other dumps on same spindle */
828 if(dp->spindle == -1) { /* but spindle -1 never conflicts by def. */
832 for(dp2 = dp->host->disks; dp2 != NULL; dp2 = dp2->hostnext)
833 if(dp2->inprogress && dp2->spindle == dp->spindle) {
845 disk_t *diskp, *delayed_diskp, *diskp_accept;
846 assignedhd_t **holdp=NULL, **holdp_accept;
847 const time_t now = time(NULL);
850 char *result_argv[MAX_ARGS+1];
855 int busy_dumpers = 0;
857 idle_reason = IDLE_NO_DUMPERS;
860 if(dumpers_ev_time != NULL) {
861 event_release(dumpers_ev_time);
862 dumpers_ev_time = NULL;
865 for(dumper = dmptable; dumper < (dmptable+inparallel); dumper++) {
871 for (dumper = dmptable; dumper < dmptable+inparallel; dumper++) {
873 if( dumper->busy || dumper->down) {
877 if (dumper->ev_read != NULL) {
878 event_release(dumper->ev_read);
879 dumper->ev_read = NULL;
883 * A potential problem with starting from the bottom of the dump time
884 * distribution is that a slave host will have both one of the shortest
885 * and one of the longest disks, so starting its shortest disk first will
886 * tie up the host and eliminate its longest disk from consideration the
887 * first pass through. This could cause a big delay in starting that long
888 * disk, which could drag out the whole night's dumps.
890 * While starting from the top of the dump time distribution solves the
891 * above problem, this turns out to be a bad idea, because the big dumps
892 * will almost certainly pack the holding disk completely, leaving no
893 * room for even one small dump to start. This ends up shutting out the
894 * small-end dumpers completely (they stay idle).
896 * The introduction of multiple simultaneous dumps to one host alleviates
897 * the biggest&smallest dumps problem: both can be started at the
903 delayed_diskp = NULL;
907 dumporder = getconf_str(CNF_DUMPORDER);
908 if(strlen(dumporder) > (size_t)(dumper-dmptable)) {
909 dumptype = dumporder[dumper-dmptable];
912 if(dumper-dmptable < 3)
918 for(diskp = rq->head; diskp != NULL; diskp = diskp->next) {
919 assert(diskp->host != NULL && sched(diskp) != NULL);
921 if (diskp->host->start_t > now) {
922 cur_idle = max(cur_idle, IDLE_START_WAIT);
923 if (delayed_diskp == NULL || sleep_time > diskp->host->start_t) {
924 delayed_diskp = diskp;
925 sleep_time = diskp->host->start_t;
927 } else if(diskp->start_t > now) {
928 cur_idle = max(cur_idle, IDLE_START_WAIT);
929 if (delayed_diskp == NULL || sleep_time > diskp->start_t) {
930 delayed_diskp = diskp;
931 sleep_time = diskp->start_t;
933 } else if (diskp->host->netif->curusage > 0 &&
934 sched(diskp)->est_kps > free_kps(diskp->host->netif)) {
935 cur_idle = max(cur_idle, IDLE_NO_BANDWIDTH);
936 } else if(sched(diskp)->no_space) {
937 cur_idle = max(cur_idle, IDLE_NO_DISKSPACE);
938 } else if (diskp->to_holdingdisk == HOLD_NEVER) {
939 cur_idle = max(cur_idle, IDLE_NO_HOLD);
941 find_diskspace(sched(diskp)->est_size, &cur_idle, NULL)) == NULL) {
942 cur_idle = max(cur_idle, IDLE_NO_DISKSPACE);
943 if (empty(tapeq) && busy_dumpers == 0) {
944 remove_disk(rq, diskp);
945 enqueue_disk(&directq, diskp);
947 } else if (client_constrained(diskp)) {
948 free_assignedhd(holdp);
949 cur_idle = max(cur_idle, IDLE_CLIENT_CONSTRAINED);
952 /* disk fits, dump it */
953 int accept = !diskp_accept;
956 case 's': accept = (sched(diskp)->est_size < sched(diskp_accept)->est_size);
958 case 'S': accept = (sched(diskp)->est_size > sched(diskp_accept)->est_size);
960 case 't': accept = (sched(diskp)->est_time < sched(diskp_accept)->est_time);
962 case 'T': accept = (sched(diskp)->est_time > sched(diskp_accept)->est_time);
964 case 'b': accept = (sched(diskp)->est_kps < sched(diskp_accept)->est_kps);
966 case 'B': accept = (sched(diskp)->est_kps > sched(diskp_accept)->est_kps);
968 default: log_add(L_WARNING, _("Unknown dumporder character \'%c\', using 's'.\n"),
970 accept = (sched(diskp)->est_size < sched(diskp_accept)->est_size);
975 if( !diskp_accept || !degraded_mode || diskp->priority >= diskp_accept->priority) {
976 if(holdp_accept) free_assignedhd(holdp_accept);
977 diskp_accept = diskp;
978 holdp_accept = holdp;
981 free_assignedhd(holdp);
985 free_assignedhd(holdp);
990 diskp = diskp_accept;
991 holdp = holdp_accept;
993 idle_reason = max(idle_reason, cur_idle);
996 * If we have no disk at this point, and there are disks that
997 * are delayed, then schedule a time event to call this dumper
998 * with the disk with the shortest delay.
1000 if (diskp == NULL && delayed_diskp != NULL) {
1001 assert(sleep_time > now);
1003 dumpers_ev_time = event_register((event_id_t)sleep_time, EV_TIME,
1004 handle_dumpers_time, &runq);
1006 } else if (diskp != NULL) {
1007 sched(diskp)->act_size = (off_t)0;
1008 allocate_bandwidth(diskp->host->netif, sched(diskp)->est_kps);
1009 sched(diskp)->activehd = assign_holdingdisk(holdp, diskp);
1011 sched(diskp)->destname = newstralloc(sched(diskp)->destname,
1012 sched(diskp)->holdp[0]->destname);
1013 diskp->host->inprogress++; /* host is now busy */
1014 diskp->inprogress = 1;
1015 sched(diskp)->dumper = dumper;
1016 sched(diskp)->timestamp = now;
1018 dumper->busy = 1; /* dumper is now busy */
1019 dumper->dp = diskp; /* link disk to dumper */
1020 remove_disk(rq, diskp); /* take it off the run queue */
1022 sched(diskp)->origsize = (off_t)-1;
1023 sched(diskp)->dumpsize = (off_t)-1;
1024 sched(diskp)->dumptime = (time_t)0;
1025 sched(diskp)->tapetime = (time_t)0;
1026 chunker = dumper->chunker;
1027 chunker->result = LAST_TOK;
1028 dumper->result = LAST_TOK;
1029 startup_chunk_process(chunker,chunker_program);
1030 chunker_cmd(chunker, START, (void *)driver_timestamp);
1031 chunker->dumper = dumper;
1032 chunker_cmd(chunker, PORT_WRITE, diskp);
1033 cmd = getresult(chunker->fd, 1, &result_argc, result_argv, MAX_ARGS+1);
1035 assignedhd_t **h=NULL;
1037 char *qname = quote_string(diskp->name);
1039 g_printf(_("driver: did not get PORT from %s for %s:%s\n"),
1040 chunker->name, diskp->host->hostname, qname);
1044 deallocate_bandwidth(diskp->host->netif, sched(diskp)->est_kps);
1045 h = sched(diskp)->holdp;
1046 activehd = sched(diskp)->activehd;
1047 h[activehd]->used = 0;
1048 h[activehd]->disk->allocated_dumpers--;
1049 adjust_diskspace(diskp, DONE);
1050 delete_diskspace(diskp);
1051 diskp->host->inprogress--;
1052 diskp->inprogress = 0;
1053 sched(diskp)->dumper = NULL;
1056 sched(diskp)->dump_attempted++;
1057 free_serial_dp(diskp);
1058 if(sched(diskp)->dump_attempted < 2)
1059 enqueue_disk(rq, diskp);
1062 dumper->ev_read = event_register((event_id_t)dumper->fd, EV_READFD,
1063 handle_dumper_result, dumper);
1064 chunker->ev_read = event_register((event_id_t)chunker->fd, EV_READFD,
1065 handle_chunker_result, chunker);
1066 dumper->output_port = atoi(result_argv[2]);
1068 dumper_cmd(dumper, PORT_DUMP, diskp);
1070 diskp->host->start_t = now + 15;
1076 * This gets called when a dumper is delayed for some reason. It may
1077 * be because a disk has a delayed start, or amanda is constrained
1078 * by network or disk limits.
1082 handle_dumpers_time(
1085 disklist_t *runq = cookie;
1086 event_release(dumpers_ev_time);
1087 dumpers_ev_time = NULL;
1088 start_some_dumps(runq);
1099 g_printf(_("dump of driver schedule %s:\n--------\n"), str);
1101 for(dp = qp->head; dp != NULL; dp = dp->next) {
1102 qname = quote_string(dp->name);
1103 g_printf(" %-20s %-25s lv %d t %5lu s %lld p %d\n",
1104 dp->host->hostname, qname, sched(dp)->level,
1105 sched(dp)->est_time,
1106 (long long)sched(dp)->est_size, sched(dp)->priority);
1109 g_printf("--------\n");
1113 start_degraded_mode(
1114 /*@keep@*/ disklist_t *queuep)
1118 off_t est_full_size;
1121 newq.head = newq.tail = 0;
1123 dump_schedule(queuep, _("before start degraded mode"));
1125 est_full_size = (off_t)0;
1126 while(!empty(*queuep)) {
1127 dp = dequeue_disk(queuep);
1129 qname = quote_string(dp->name);
1130 if(sched(dp)->level != 0)
1131 /* go ahead and do the disk as-is */
1132 enqueue_disk(&newq, dp);
1134 if (reserved_space + est_full_size + sched(dp)->est_size
1135 <= total_disksize) {
1136 enqueue_disk(&newq, dp);
1137 est_full_size += sched(dp)->est_size;
1139 else if(sched(dp)->degr_level != -1) {
1140 sched(dp)->level = sched(dp)->degr_level;
1141 sched(dp)->dumpdate = sched(dp)->degr_dumpdate;
1142 sched(dp)->est_nsize = sched(dp)->degr_nsize;
1143 sched(dp)->est_csize = sched(dp)->degr_csize;
1144 sched(dp)->est_time = sched(dp)->degr_time;
1145 sched(dp)->est_kps = sched(dp)->degr_kps;
1146 enqueue_disk(&newq, dp);
1149 log_add(L_FAIL,_("%s %s %s %d [can't switch to incremental dump]"),
1150 dp->host->hostname, qname, sched(dp)->datestamp,
1157 /*@i@*/ *queuep = newq;
1160 dump_schedule(queuep, _("after start degraded mode"));
1165 continue_port_dumps(void)
1169 int active_dumpers=0, busy_dumpers=0, i;
1172 /* First we try to grant diskspace to some dumps waiting for it. */
1173 for( dp = roomq.head; dp; dp = ndp ) {
1175 /* find last holdingdisk used by this dump */
1176 for( i = 0, h = sched(dp)->holdp; h[i+1]; i++ ) {
1177 (void)h; /* Quiet lint */
1179 /* find more space */
1180 h = find_diskspace( sched(dp)->est_size - sched(dp)->act_size,
1181 &active_dumpers, h[i] );
1183 for(dumper = dmptable; dumper < dmptable + inparallel &&
1184 dumper->dp != dp; dumper++) {
1185 (void)dp; /* Quiet lint */
1187 assert( dumper < dmptable + inparallel );
1188 sched(dp)->activehd = assign_holdingdisk( h, dp );
1189 chunker_cmd( dumper->chunker, CONTINUE, dp );
1191 remove_disk( &roomq, dp );
1195 /* So for some disks there is less holding diskspace available than
1196 * was asked for. Possible reasons are
1197 * a) diskspace has been allocated for other dumps which are
1198 * still running or already being written to tape
1199 * b) all other dumps have been suspended due to lack of diskspace
1200 * c) this dump doesn't fit on all the holding disks
1201 * Case a) is not a problem. We just wait for the diskspace to
1202 * be freed by moving the current disk to a queue.
1203 * If case b) occurs, we have a deadlock situation. We select
1204 * a dump from the queue to be aborted and abort it. It will
1205 * be retried later dumping to disk.
1206 * If case c) is detected, the dump is aborted. Next time
1207 * it will be dumped directly to tape. Actually, case c is a special
1208 * manifestation of case b) where only one dumper is busy.
1210 for(dp=NULL, dumper = dmptable; dumper < (dmptable+inparallel); dumper++) {
1211 if( dumper->busy ) {
1213 if( !find_disk(&roomq, dumper->dp) ) {
1216 sched(dp)->est_size > sched(dumper->dp)->est_size ) {
1221 if((dp != NULL) && (active_dumpers == 0) && (busy_dumpers > 0) &&
1222 ((!taper_busy && empty(tapeq)) || degraded_mode) &&
1223 pending_aborts == 0 ) { /* not case a */
1224 if( busy_dumpers == 1 ) { /* case c */
1225 sched(dp)->no_space = 1;
1228 /* At this time, dp points to the dump with the smallest est_size.
1229 * We abort that dump, hopefully not wasting too much time retrying it.
1231 remove_disk( &roomq, dp );
1232 chunker_cmd( sched(dp)->dumper->chunker, ABORT, NULL);
1233 dumper_cmd( sched(dp)->dumper, ABORT, NULL );
1240 handle_taper_result(
1246 char *result_argv[MAX_ARGS+1];
1249 (void)cookie; /* Quiet unused parameter warning */
1251 assert(cookie == NULL);
1257 cmd = getresult(taper, 1, &result_argc, result_argv, MAX_ARGS+1);
1261 case FAILED: /* FAILED <handle> INPUT-* TAPE-* <input err mesg> <tape err mesg> */
1262 if(result_argc != 6) {
1263 error(_("error: [taper FAILED result_argc != 6: %d"), result_argc);
1267 dp = serial2disk(result_argv[2]);
1268 assert(dp == taper_disk);
1270 free_serial(result_argv[2]);
1272 qname = quote_string(dp->name);
1273 g_printf(_("driver: finished-cmd time %s taper wrote %s:%s\n"),
1274 walltime_str(curclock()), dp->host->hostname, qname);
1278 if (strcmp(result_argv[3], "INPUT-ERROR") == 0) {
1279 taper_input_error = stralloc(result_argv[5]);
1281 if (strcmp(result_argv[4], "TAPE-ERROR") == 0) {
1282 taper_tape_error = stralloc(result_argv[6]);
1289 case PARTIAL: /* PARTIAL <handle> INPUT-* TAPE-* <stat mess> <input err mesg> <tape err mesg>*/
1290 case DONE: /* DONE <handle> INPUT-GOOD TAPE-GOOD <stat mess> <input err mesg> <tape err mesg> */
1291 if(result_argc != 7) {
1292 error(_("error: [taper PARTIAL result_argc != 7: %d"), result_argc);
1296 dp = serial2disk(result_argv[2]);
1297 assert(dp == taper_disk);
1299 free_serial(result_argv[2]);
1301 qname = quote_string(dp->name);
1302 g_printf(_("driver: finished-cmd time %s taper wrote %s:%s\n"),
1303 walltime_str(curclock()), dp->host->hostname, qname);
1307 if (strcmp(result_argv[3], "INPUT-ERROR") == 0) {
1308 taper_input_error = stralloc(result_argv[5]);
1310 if (strcmp(result_argv[4], "TAPE-ERROR") == 0) {
1311 taper_tape_error = stralloc(result_argv[6]);
1318 case PARTDONE: /* PARTDONE <handle> <label> <fileno> <stat> */
1319 dp = serial2disk(result_argv[2]);
1320 assert(dp == taper_disk);
1321 if (result_argc != 6) {
1322 error(_("error [taper PARTDONE result_argc != 5: %d]"),
1326 if (!taper_first_label) {
1327 taper_first_label = stralloc(result_argv[3]);
1328 taper_first_fileno = OFF_T_ATOI(result_argv[4]);
1330 taper_written = OFF_T_ATOI(result_argv[5]);
1331 if (taper_written > sched(taper_disk)->act_size)
1332 sched(taper_disk)->act_size = taper_written;
1336 case REQUEST_NEW_TAPE: /* REQUEST-NEW-TAPE */
1337 if (result_argc != 2) {
1338 error(_("error [taper REQUEST_NEW_TAPE result_argc != 2: %d]"),
1342 taper_state &= !TAPER_STATE_TAPE_STARTED;
1344 if (current_tape >= conf_runtapes) {
1345 taper_cmd(NO_NEW_TAPE, NULL, NULL, 0, NULL);
1347 _("Out of tapes; going into degraded mode."));
1348 start_degraded_mode(&runq);
1350 TapeAction result_tape_action;
1352 taper_state |= TAPER_STATE_WAIT_FOR_TAPE;
1353 result_tape_action = tape_action();
1354 if (result_tape_action & TAPE_ACTION_NEW_TAPE) {
1355 taper_cmd(NEW_TAPE, NULL, NULL, 0, NULL);
1356 taper_state &= !TAPER_STATE_WAIT_FOR_TAPE;
1357 } else if (result_tape_action & TAPE_ACTION_NO_NEW_TAPE) {
1358 taper_cmd(NO_NEW_TAPE, NULL, NULL, 0, NULL);
1359 taper_state &= !TAPER_STATE_WAIT_FOR_TAPE;
1364 case NEW_TAPE: /* NEW-TAPE <handle> <label> */
1365 if (result_argc != 3) {
1366 error(_("error [taper NEW_TAPE result_argc != 3: %d]"),
1371 /* Update our tape counter and reset tape_left */
1373 tape_left = tape_length;
1374 taper_state |= TAPER_STATE_TAPE_STARTED;
1377 case NO_NEW_TAPE: /* NO-NEW-TAPE <handle> */
1378 if (result_argc != 2) {
1379 error(_("error [taper NO_NEW_TAPE result_argc != 2: %d]"),
1385 case DUMPER_STATUS: /* DUMPER-STATUS <handle> */
1386 if (result_argc != 2) {
1387 error(_("error [taper NO_NEW_TAPE result_argc != 2: %d]"),
1391 if (taper_dumper->result == LAST_TOK) {
1392 taper_sendresult = 1;
1394 if( taper_dumper->result == DONE) {
1395 taper_cmd(DONE, NULL, NULL, 0, NULL);
1397 taper_cmd(FAILED, NULL, NULL, 0, NULL);
1402 case TAPE_ERROR: /* TAPE-ERROR <handle> <err mess> */
1403 dp = serial2disk(result_argv[2]);
1405 free_serial(result_argv[2]);
1406 qname = quote_string(dp->name);
1407 g_printf(_("driver: finished-cmd time %s taper wrote %s:%s\n"),
1408 walltime_str(curclock()), dp->host->hostname, qname);
1411 log_add(L_WARNING, _("Taper error: %s"), result_argv[3]);
1412 taper_tape_error = stralloc(result_argv[3]);
1417 log_add(L_WARNING, _("Taper protocol error"));
1418 taper_tape_error = stralloc("BOGUS");
1421 * Since we received a taper error, we can't send anything more
1422 * to the taper. Go into degraded mode to try to get everthing
1423 * onto disk. Later, these dumps can be flushed to a new tape.
1424 * The tape queue is zapped so that it appears empty in future
1425 * checks. If there are dumps waiting for diskspace to be freed,
1430 _("going into degraded mode because of taper component error."));
1431 start_degraded_mode(&runq);
1433 tapeq.head = tapeq.tail = NULL;
1435 if(taper_ev_read != NULL) {
1436 event_release(taper_ev_read);
1437 taper_ev_read = NULL;
1439 if(cmd != TAPE_ERROR) aclose(taper);
1445 error(_("driver received unexpected token (%s) from taper"),
1450 if (taper_result != LAST_TOK) {
1452 if (taper_dumper->result != LAST_TOK) {
1453 // Dumper already returned it's result
1454 dumper_taper_result(taper_disk);
1457 file_taper_result(taper_disk);
1461 } while(areads_dataready(taper));
1469 char *qname = quote_string(dp->name);
1471 if (taper_result == DONE) {
1472 update_info_taper(dp, taper_first_label, taper_first_fileno,
1476 sched(dp)->taper_attempted += 1;
1478 if (taper_input_error) {
1479 g_printf("driver: taper failed %s %s: %s\n",
1480 dp->host->hostname, qname, taper_input_error);
1481 if (strcmp(sched(dp)->datestamp, driver_timestamp) == 0) {
1482 if(sched(dp)->taper_attempted >= 2) {
1483 log_add(L_FAIL, _("%s %s %s %d [too many taper retries after holding disk error: %s]"),
1484 dp->host->hostname, qname, sched(dp)->datestamp,
1485 sched(dp)->level, taper_input_error);
1486 g_printf("driver: taper failed %s %s, too many taper retry after holding disk error\n",
1487 dp->host->hostname, qname);
1488 amfree(sched(dp)->destname);
1489 amfree(sched(dp)->dumpdate);
1490 amfree(sched(dp)->degr_dumpdate);
1491 amfree(sched(dp)->datestamp);
1494 log_add(L_INFO, _("%s %s %s %d [Will retry dump because of holding disk error: %s]"),
1495 dp->host->hostname, qname, sched(dp)->datestamp,
1496 sched(dp)->level, taper_input_error);
1497 g_printf("driver: taper will retry %s %s because of holding disk error\n",
1498 dp->host->hostname, qname);
1499 if (dp->to_holdingdisk != HOLD_REQUIRED) {
1500 dp->to_holdingdisk = HOLD_NEVER;
1501 sched(dp)->dump_attempted -= 1;
1502 headqueue_disk(&directq, dp);
1504 amfree(sched(dp)->destname);
1505 amfree(sched(dp)->dumpdate);
1506 amfree(sched(dp)->degr_dumpdate);
1507 amfree(sched(dp)->datestamp);
1512 amfree(sched(dp)->destname);
1513 amfree(sched(dp)->dumpdate);
1514 amfree(sched(dp)->degr_dumpdate);
1515 amfree(sched(dp)->datestamp);
1518 } else if (taper_tape_error) {
1519 if(sched(dp)->taper_attempted >= 2) {
1520 log_add(L_FAIL, _("%s %s %s %d [too many taper retries]"),
1521 dp->host->hostname, qname, sched(dp)->datestamp,
1523 g_printf("driver: taper failed %s %s, too many taper retry\n",
1524 dp->host->hostname, qname);
1525 amfree(sched(dp)->destname);
1526 amfree(sched(dp)->dumpdate);
1527 amfree(sched(dp)->degr_dumpdate);
1528 amfree(sched(dp)->datestamp);
1531 g_printf("driver: taper will retry %s %s\n",
1532 dp->host->hostname, qname);
1533 /* Re-insert into taper queue. */
1534 headqueue_disk(&tapeq, dp);
1537 delete_diskspace(dp);
1538 amfree(sched(dp)->destname);
1539 amfree(sched(dp)->dumpdate);
1540 amfree(sched(dp)->degr_dumpdate);
1541 amfree(sched(dp)->datestamp);
1548 taper_input_error = NULL;
1549 taper_tape_error = NULL;
1552 /* continue with those dumps waiting for diskspace */
1553 continue_port_dumps();
1554 start_some_dumps(&runq);
1559 dumper_taper_result(
1566 dumper = sched(dp)->dumper;
1569 if(dumper->result == DONE && taper_result == DONE) {
1570 update_info_dumper(dp, sched(dp)->origsize,
1571 sched(dp)->dumpsize, sched(dp)->dumptime);
1572 update_info_taper(dp, taper_first_label, taper_first_fileno,
1574 qname = quote_string(dp->name); /*quote to take care of spaces*/
1576 log_add(L_STATS, _("estimate %s %s %s %d [sec %ld nkb %lld ckb %lld kps %lu]"),
1577 dp->host->hostname, qname, sched(dp)->datestamp,
1579 sched(dp)->est_time, (long long)sched(dp)->est_nsize,
1580 (long long)sched(dp)->est_csize,
1581 sched(dp)->est_kps);
1584 update_failed_dump_to_tape(dp);
1587 is_partial = dumper->result != DONE || taper_result != DONE;
1589 sched(dp)->dump_attempted += 1;
1590 sched(dp)->taper_attempted += 1;
1592 if((dumper->result != DONE || taper_result != DONE) &&
1593 sched(dp)->dump_attempted <= 1 &&
1594 sched(dp)->taper_attempted <= 1) {
1595 enqueue_disk(&directq, dp);
1598 if(dumper->ev_read != NULL) {
1599 event_release(dumper->ev_read);
1600 dumper->ev_read = NULL;
1602 if(taper_ev_read != NULL) {
1603 event_release(taper_ev_read);
1604 taper_ev_read = NULL;
1607 taper_input_error = NULL;
1608 taper_tape_error = NULL;
1610 dp->host->inprogress -= 1;
1612 deallocate_bandwidth(dp->host->netif, sched(dp)->est_kps);
1621 for(dumper = dmptable; dumper < dmptable+inparallel; dumper++)
1622 if(!dumper->busy && !dumper->down) return dumper;
1628 dumper_chunker_result(
1633 assignedhd_t **h=NULL;
1640 dumper = sched(dp)->dumper;
1641 chunker = dumper->chunker;
1645 h = sched(dp)->holdp;
1646 activehd = sched(dp)->activehd;
1648 if(dumper->result == DONE && chunker->result == DONE) {
1649 update_info_dumper(dp, sched(dp)->origsize,
1650 sched(dp)->dumpsize, sched(dp)->dumptime);
1651 qname = quote_string(dp->name);/*quote to take care of spaces*/
1653 log_add(L_STATS, _("estimate %s %s %s %d [sec %ld nkb %lld ckb %lld kps %lu]"),
1654 dp->host->hostname, qname, sched(dp)->datestamp,
1656 sched(dp)->est_time, (long long)sched(dp)->est_nsize,
1657 (long long)sched(dp)->est_csize,
1658 sched(dp)->est_kps);
1662 deallocate_bandwidth(dp->host->netif, sched(dp)->est_kps);
1664 is_partial = dumper->result != DONE || chunker->result != DONE;
1665 rename_tmp_holding(sched(dp)->destname, !is_partial);
1668 for( i = 0, h = sched(dp)->holdp; i < activehd; i++ ) {
1669 dummy += h[i]->used;
1672 size = holding_file_size(sched(dp)->destname, 0);
1673 h[activehd]->used = size - dummy;
1674 h[activehd]->disk->allocated_dumpers--;
1675 adjust_diskspace(dp, DONE);
1677 sched(dp)->dump_attempted += 1;
1679 if((dumper->result != DONE || chunker->result != DONE) &&
1680 sched(dp)->dump_attempted <= 1) {
1681 delete_diskspace(dp);
1682 if (sched(dp)->no_space) {
1683 enqueue_disk(&directq, dp);
1685 enqueue_disk(&runq, dp);
1688 else if(size > (off_t)DISK_BLOCK_KB) {
1689 enqueue_disk(&tapeq, dp);
1692 delete_diskspace(dp);
1696 dp->host->inprogress -= 1;
1699 waitpid(chunker->pid, NULL, 0 );
1700 aclose(chunker->fd);
1705 if (chunker->result == ABORT_FINISHED)
1707 continue_port_dumps();
1709 * Wakeup any dumpers that are sleeping because of network
1710 * or disk constraints.
1712 start_some_dumps(&runq);
1718 handle_dumper_result(
1721 /*static int pending_aborts = 0;*/
1722 dumper_t *dumper = cookie;
1727 char *result_argv[MAX_ARGS+1];
1729 assert(dumper != NULL);
1732 assert(sched(dp) != NULL);
1737 cmd = getresult(dumper->fd, 1, &result_argc, result_argv, MAX_ARGS+1);
1740 /* result_argv[2] always contains the serial number */
1741 sdp = serial2disk(result_argv[2]);
1743 error(_("Invalid serial number %s"), result_argv[2]);
1744 g_assert_not_reached();
1748 qname = quote_string(dp->name);
1751 case DONE: /* DONE <handle> <origsize> <dumpsize> <dumptime> <errstr> */
1752 if(result_argc != 6) {
1753 error(_("error [dumper DONE result_argc != 6: %d]"), result_argc);
1757 sched(dp)->origsize = OFF_T_ATOI(result_argv[3]);
1758 sched(dp)->dumptime = TIME_T_ATOI(result_argv[5]);
1760 g_printf(_("driver: finished-cmd time %s %s dumped %s:%s\n"),
1761 walltime_str(curclock()), dumper->name,
1762 dp->host->hostname, qname);
1765 dumper->result = cmd;
1769 case TRYAGAIN: /* TRY-AGAIN <handle> <errstr> */
1771 * Requeue this disk, and fall through to the FAILED
1774 if(sched(dp)->dump_attempted) {
1775 char *qname = quote_string(dp->name);
1776 log_add(L_FAIL, _("%s %s %s %d [too many dumper retry: %s]"),
1777 dp->host->hostname, qname, sched(dp)->datestamp,
1778 sched(dp)->level, result_argv[3]);
1779 g_printf(_("driver: dump failed %s %s %s, too many dumper retry: %s\n"),
1780 result_argv[2], dp->host->hostname, qname,
1785 case FAILED: /* FAILED <handle> <errstr> */
1786 /*free_serial(result_argv[2]);*/
1787 dumper->result = cmd;
1790 case ABORT_FINISHED: /* ABORT-FINISHED <handle> */
1792 * We sent an ABORT from the NO-ROOM case because this dump
1793 * wasn't going to fit onto the holding disk. We now need to
1794 * clean up the remains of this image, and try to finish
1795 * other dumps that are waiting on disk space.
1797 assert(pending_aborts);
1798 /*free_serial(result_argv[2]);*/
1799 dumper->result = cmd;
1803 /* either EOF or garbage from dumper. Turn it off */
1804 log_add(L_WARNING, _("%s pid %ld is messed up, ignoring it.\n"),
1805 dumper->name, (long)dumper->pid);
1806 if (dumper->ev_read) {
1807 event_release(dumper->ev_read);
1808 dumper->ev_read = NULL;
1812 dumper->down = 1; /* mark it down so it isn't used again */
1814 /* if it was dumping something, zap it and try again */
1815 if(sched(dp)->dump_attempted) {
1816 log_add(L_FAIL, _("%s %s %s %d [%s died]"),
1817 dp->host->hostname, qname, sched(dp)->datestamp,
1818 sched(dp)->level, dumper->name);
1821 log_add(L_WARNING, _("%s died while dumping %s:%s lev %d."),
1822 dumper->name, dp->host->hostname, qname,
1826 dumper->result = cmd;
1834 /* send the dumper result to the chunker */
1835 if (dumper->chunker) {
1836 if (dumper->chunker->down == 0 && dumper->chunker->fd != -1 &&
1837 dumper->chunker->result == LAST_TOK) {
1839 chunker_cmd(dumper->chunker, DONE, dp);
1842 chunker_cmd(dumper->chunker, FAILED, dp);
1845 if( dumper->result != LAST_TOK &&
1846 dumper->chunker->result != LAST_TOK)
1847 dumper_chunker_result(dp);
1848 } else { /* send the dumper result to the taper */
1849 if (taper_sendresult) {
1851 taper_cmd(DONE, driver_timestamp, NULL, 0, NULL);
1853 taper_cmd(FAILED, driver_timestamp, NULL, 0, NULL);
1855 taper_sendresult = 0;
1858 if (taper_dumper && taper_result != LAST_TOK) {
1859 dumper_taper_result(dp);
1861 } while(areads_dataready(dumper->fd));
1866 handle_chunker_result(
1869 /*static int pending_aborts = 0;*/
1870 chunker_t *chunker = cookie;
1871 assignedhd_t **h=NULL;
1876 char *result_argv[MAX_ARGS+1];
1881 assert(chunker != NULL);
1882 dumper = chunker->dumper;
1883 assert(dumper != NULL);
1886 assert(sched(dp) != NULL);
1887 assert(sched(dp)->destname != NULL);
1888 assert(dp != NULL && sched(dp) != NULL && sched(dp)->destname);
1890 if(dp && sched(dp) && sched(dp)->holdp) {
1891 h = sched(dp)->holdp;
1892 activehd = sched(dp)->activehd;
1899 cmd = getresult(chunker->fd, 1, &result_argc, result_argv, MAX_ARGS+1);
1902 /* result_argv[2] always contains the serial number */
1903 sdp = serial2disk(result_argv[2]);
1905 error(_("Invalid serial number %s"), result_argv[2]);
1906 g_assert_not_reached();
1912 case PARTIAL: /* PARTIAL <handle> <dumpsize> <errstr> */
1913 case DONE: /* DONE <handle> <dumpsize> <errstr> */
1914 if(result_argc != 4) {
1915 error(_("error [chunker %s result_argc != 4: %d]"), cmdstr[cmd],
1919 /*free_serial(result_argv[2]);*/
1921 sched(dp)->dumpsize = (off_t)atof(result_argv[3]);
1923 qname = quote_string(dp->name);
1924 g_printf(_("driver: finished-cmd time %s %s chunked %s:%s\n"),
1925 walltime_str(curclock()), chunker->name,
1926 dp->host->hostname, qname);
1930 event_release(chunker->ev_read);
1932 chunker->result = cmd;
1936 case TRYAGAIN: /* TRY-AGAIN <handle> <errstr> */
1937 event_release(chunker->ev_read);
1939 chunker->result = cmd;
1942 case FAILED: /* FAILED <handle> <errstr> */
1943 /*free_serial(result_argv[2]);*/
1945 event_release(chunker->ev_read);
1947 chunker->result = cmd;
1951 case NO_ROOM: /* NO-ROOM <handle> <missing_size> */
1952 if (!h || activehd < 0) { /* should never happen */
1953 error(_("!h || activehd < 0"));
1956 h[activehd]->used -= OFF_T_ATOI(result_argv[3]);
1957 h[activehd]->reserved -= OFF_T_ATOI(result_argv[3]);
1958 h[activehd]->disk->allocated_space -= OFF_T_ATOI(result_argv[3]);
1959 h[activehd]->disk->disksize -= OFF_T_ATOI(result_argv[3]);
1962 case RQ_MORE_DISK: /* RQ-MORE-DISK <handle> */
1963 if (!h || activehd < 0) { /* should never happen */
1964 error(_("!h || activehd < 0"));
1967 h[activehd]->disk->allocated_dumpers--;
1968 h[activehd]->used = h[activehd]->reserved;
1969 if( h[++activehd] ) { /* There's still some allocated space left.
1970 * Tell the dumper about it. */
1971 sched(dp)->activehd++;
1972 chunker_cmd( chunker, CONTINUE, dp );
1973 } else { /* !h[++activehd] - must allocate more space */
1974 sched(dp)->act_size = sched(dp)->est_size; /* not quite true */
1975 sched(dp)->est_size = (sched(dp)->act_size/(off_t)20) * (off_t)21; /* +5% */
1976 sched(dp)->est_size = am_round(sched(dp)->est_size, (off_t)DISK_BLOCK_KB);
1977 if (sched(dp)->est_size < sched(dp)->act_size + 2*DISK_BLOCK_KB)
1978 sched(dp)->est_size += 2 * DISK_BLOCK_KB;
1979 h = find_diskspace( sched(dp)->est_size - sched(dp)->act_size,
1983 /* No diskspace available. The reason for this will be
1984 * determined in continue_port_dumps(). */
1985 enqueue_disk( &roomq, dp );
1986 continue_port_dumps();
1988 /* OK, allocate space for disk and have chunker continue */
1989 sched(dp)->activehd = assign_holdingdisk( h, dp );
1990 chunker_cmd( chunker, CONTINUE, dp );
1996 case ABORT_FINISHED: /* ABORT-FINISHED <handle> */
1998 * We sent an ABORT from the NO-ROOM case because this dump
1999 * wasn't going to fit onto the holding disk. We now need to
2000 * clean up the remains of this image, and try to finish
2001 * other dumps that are waiting on disk space.
2003 /*assert(pending_aborts);*/
2005 /*free_serial(result_argv[2]);*/
2007 event_release(chunker->ev_read);
2009 chunker->result = cmd;
2014 /* either EOF or garbage from chunker. Turn it off */
2015 log_add(L_WARNING, _("%s pid %ld is messed up, ignoring it.\n"),
2016 chunker->name, (long)chunker->pid);
2019 /* if it was dumping something, zap it and try again */
2020 if (!h || activehd < 0) { /* should never happen */
2021 error(_("!h || activehd < 0"));
2024 qname = quote_string(dp->name);
2025 if(sched(dp)->dump_attempted) {
2026 log_add(L_FAIL, _("%s %s %s %d [%s died]"),
2027 dp->host->hostname, qname, sched(dp)->datestamp,
2028 sched(dp)->level, chunker->name);
2031 log_add(L_WARNING, _("%s died while dumping %s:%s lev %d."),
2032 chunker->name, dp->host->hostname, qname,
2039 event_release(chunker->ev_read);
2041 chunker->result = cmd;
2049 if(chunker->result != LAST_TOK && chunker->dumper->result != LAST_TOK)
2050 dumper_chunker_result(dp);
2052 } while(areads_dataready(chunker->fd));
2063 char *hostname, *diskname, *datestamp;
2067 char *inpline = NULL;
2073 char *qdestname = NULL;
2075 tq.head = tq.tail = NULL;
2077 for(line = 0; (inpline = agets(stdin)) != NULL; free(inpline)) {
2079 if (inpline[0] == '\0')
2085 skip_whitespace(s, ch); /* find the command */
2087 error(_("flush line %d: syntax error (no command)"), line);
2091 skip_non_whitespace(s, ch);
2094 if(strcmp(command,"ENDFLUSH") == 0) {
2098 if(strcmp(command,"FLUSH") != 0) {
2099 error(_("flush line %d: syntax error (%s != FLUSH)"), line, command);
2103 skip_whitespace(s, ch); /* find the hostname */
2105 error(_("flush line %d: syntax error (no hostname)"), line);
2109 skip_non_whitespace(s, ch);
2112 skip_whitespace(s, ch); /* find the diskname */
2114 error(_("flush line %d: syntax error (no diskname)"), line);
2118 skip_quoted_string(s, ch);
2119 s[-1] = '\0'; /* terminate the disk name */
2120 diskname = unquote_string(qname);
2122 skip_whitespace(s, ch); /* find the datestamp */
2124 error(_("flush line %d: syntax error (no datestamp)"), line);
2128 skip_non_whitespace(s, ch);
2131 skip_whitespace(s, ch); /* find the level number */
2132 if(ch == '\0' || sscanf(s - 1, "%d", &level) != 1) {
2133 error(_("flush line %d: syntax error (bad level)"), line);
2136 skip_integer(s, ch);
2138 skip_whitespace(s, ch); /* find the filename */
2140 error(_("flush line %d: syntax error (no filename)"), line);
2144 skip_quoted_string(s, ch);
2146 destname = unquote_string(qdestname);
2148 holding_file_get_dumpfile(destname, &file);
2149 if( file.type != F_DUMPFILE) {
2150 if( file.type != F_CONT_DUMPFILE )
2151 log_add(L_INFO, _("%s: ignoring cruft file."), destname);
2157 if(strcmp(hostname, file.name) != 0 ||
2158 strcmp(diskname, file.disk) != 0 ||
2159 strcmp(datestamp, file.datestamp) != 0) {
2160 log_add(L_INFO, _("disk %s:%s not consistent with file %s"),
2161 hostname, diskname, destname);
2168 dp = lookup_disk(file.name, file.disk);
2171 log_add(L_INFO, _("%s: disk %s:%s not in database, skipping it."),
2172 destname, file.name, file.disk);
2177 if(file.dumplevel < 0 || file.dumplevel > 9) {
2178 log_add(L_INFO, _("%s: ignoring file with bogus dump level %d."),
2179 destname, file.dumplevel);
2184 if (holding_file_size(destname,1) <= 0) {
2185 log_add(L_INFO, "%s: removing file with no data.", destname);
2186 holding_file_unlink(destname);
2191 dp1 = (disk_t *)alloc(SIZEOF(disk_t));
2193 dp1->next = dp1->prev = NULL;
2195 /* add it to the flushhost list */
2197 flushhost = alloc(SIZEOF(am_host_t));
2198 flushhost->next = NULL;
2199 flushhost->hostname = stralloc("FLUSHHOST");
2200 flushhost->up = NULL;
2201 flushhost->features = NULL;
2203 dp1->hostnext = flushhost->disks;
2204 flushhost->disks = dp1;
2206 sp = (sched_t *) alloc(SIZEOF(sched_t));
2207 sp->destname = destname;
2208 sp->level = file.dumplevel;
2209 sp->dumpdate = NULL;
2210 sp->degr_dumpdate = NULL;
2211 sp->datestamp = stralloc(file.datestamp);
2212 sp->est_nsize = (off_t)0;
2213 sp->est_csize = (off_t)0;
2217 sp->degr_level = -1;
2218 sp->dump_attempted = 0;
2219 sp->taper_attempted = 0;
2220 sp->act_size = holding_file_size(destname, 0);
2221 sp->holdp = build_diskspace(destname);
2222 if(sp->holdp == NULL) continue;
2224 sp->timestamp = (time_t)0;
2226 dp1->up = (char *)sp;
2228 enqueue_disk(&tq, dp1);
2241 int level, line, priority;
2242 char *dumpdate, *degr_dumpdate;
2244 time_t time, degr_time;
2245 time_t *time_p = &time;
2246 time_t *degr_time_p = °r_time;
2247 off_t nsize, csize, degr_nsize, degr_csize;
2248 unsigned long kps, degr_kps;
2249 char *hostname, *features, *diskname, *datestamp, *inpline = NULL;
2253 off_t flush_size = (off_t)0;
2258 long long degr_nsize_;
2259 long long degr_csize_;
2261 (void)cookie; /* Quiet unused parameter warning */
2263 event_release(schedule_ev_read);
2265 /* read schedule from stdin */
2267 for(line = 0; (inpline = agets(stdin)) != NULL; free(inpline)) {
2268 if (inpline[0] == '\0')
2275 skip_whitespace(s, ch); /* find the command */
2277 error(_("schedule line %d: syntax error (no command)"), line);
2281 skip_non_whitespace(s, ch);
2284 if(strcmp(command,"DUMP") != 0) {
2285 error(_("schedule line %d: syntax error (%s != DUMP)"), line, command);
2289 skip_whitespace(s, ch); /* find the host name */
2291 error(_("schedule line %d: syntax error (no host name)"), line);
2295 skip_non_whitespace(s, ch);
2298 skip_whitespace(s, ch); /* find the feature list */
2300 error(_("schedule line %d: syntax error (no feature list)"), line);
2304 skip_non_whitespace(s, ch);
2307 skip_whitespace(s, ch); /* find the disk name */
2309 error(_("schedule line %d: syntax error (no disk name)"), line);
2313 skip_quoted_string(s, ch);
2314 s[-1] = '\0'; /* terminate the disk name */
2315 diskname = unquote_string(qname);
2317 skip_whitespace(s, ch); /* find the datestamp */
2319 error(_("schedule line %d: syntax error (no datestamp)"), line);
2323 skip_non_whitespace(s, ch);
2326 skip_whitespace(s, ch); /* find the priority number */
2327 if(ch == '\0' || sscanf(s - 1, "%d", &priority) != 1) {
2328 error(_("schedule line %d: syntax error (bad priority)"), line);
2331 skip_integer(s, ch);
2333 skip_whitespace(s, ch); /* find the level number */
2334 if(ch == '\0' || sscanf(s - 1, "%d", &level) != 1) {
2335 error(_("schedule line %d: syntax error (bad level)"), line);
2338 skip_integer(s, ch);
2340 skip_whitespace(s, ch); /* find the dump date */
2342 error(_("schedule line %d: syntax error (bad dump date)"), line);
2346 skip_non_whitespace(s, ch);
2349 skip_whitespace(s, ch); /* find the native size */
2351 if(ch == '\0' || sscanf(s - 1, "%lld", &nsize_) != 1) {
2352 error(_("schedule line %d: syntax error (bad nsize)"), line);
2355 nsize = (off_t)nsize_;
2356 skip_integer(s, ch);
2358 skip_whitespace(s, ch); /* find the compressed size */
2360 if(ch == '\0' || sscanf(s - 1, "%lld", &csize_) != 1) {
2361 error(_("schedule line %d: syntax error (bad csize)"), line);
2364 csize = (off_t)csize_;
2365 skip_integer(s, ch);
2367 skip_whitespace(s, ch); /* find the time number */
2368 if(ch == '\0' || sscanf(s - 1, "%lld", &time_) != 1) {
2369 error(_("schedule line %d: syntax error (bad estimated time)"), line);
2372 *time_p = (time_t)time_;
2373 skip_integer(s, ch);
2375 skip_whitespace(s, ch); /* find the kps number */
2376 if(ch == '\0' || sscanf(s - 1, "%lu", &kps) != 1) {
2377 error(_("schedule line %d: syntax error (bad kps)"), line);
2380 skip_integer(s, ch);
2382 degr_dumpdate = NULL; /* flag if degr fields found */
2383 skip_whitespace(s, ch); /* find the degr level number */
2385 if(sscanf(s - 1, "%d", °r_level) != 1) {
2386 error(_("schedule line %d: syntax error (bad degr level)"), line);
2389 skip_integer(s, ch);
2391 skip_whitespace(s, ch); /* find the degr dump date */
2393 error(_("schedule line %d: syntax error (bad degr dump date)"), line);
2396 degr_dumpdate = s - 1;
2397 skip_non_whitespace(s, ch);
2400 skip_whitespace(s, ch); /* find the degr native size */
2401 degr_nsize_ = (off_t)0;
2402 if(ch == '\0' || sscanf(s - 1, "%lld", °r_nsize_) != 1) {
2403 error(_("schedule line %d: syntax error (bad degr nsize)"), line);
2406 degr_nsize = (off_t)degr_nsize_;
2407 skip_integer(s, ch);
2409 skip_whitespace(s, ch); /* find the degr compressed size */
2410 degr_csize_ = (off_t)0;
2411 if(ch == '\0' || sscanf(s - 1, "%lld", °r_csize_) != 1) {
2412 error(_("schedule line %d: syntax error (bad degr csize)"), line);
2415 degr_csize = (off_t)degr_csize_;
2416 skip_integer(s, ch);
2418 skip_whitespace(s, ch); /* find the degr time number */
2419 if(ch == '\0' || sscanf(s - 1, "%lld", &time_) != 1) {
2420 error(_("schedule line %d: syntax error (bad degr estimated time)"), line);
2423 *degr_time_p = (time_t)time_;
2424 skip_integer(s, ch);
2426 skip_whitespace(s, ch); /* find the degr kps number */
2427 if(ch == '\0' || sscanf(s - 1, "%lu", °r_kps) != 1) {
2428 error(_("schedule line %d: syntax error (bad degr kps)"), line);
2431 skip_integer(s, ch);
2434 degr_nsize = (off_t)0;
2435 degr_csize = (off_t)0;
2436 degr_time = (time_t)0;
2440 dp = lookup_disk(hostname, diskname);
2443 _("schedule line %d: %s:'%s' not in disklist, ignored"),
2444 line, hostname, qname);
2449 sp = (sched_t *) alloc(SIZEOF(sched_t));
2452 sp->dumpdate = stralloc(dumpdate);
2453 sp->est_nsize = DISK_BLOCK_KB + nsize; /* include header */
2454 sp->est_csize = DISK_BLOCK_KB + csize; /* include header */
2455 /* round estimate to next multiple of DISK_BLOCK_KB */
2456 sp->est_csize = am_round(sp->est_csize, DISK_BLOCK_KB);
2457 sp->est_size = sp->est_csize;
2458 sp->est_time = time;
2460 sp->priority = priority;
2461 sp->datestamp = stralloc(datestamp);
2464 sp->degr_level = degr_level;
2465 sp->degr_dumpdate = stralloc(degr_dumpdate);
2466 sp->degr_nsize = DISK_BLOCK_KB + degr_nsize;
2467 sp->degr_csize = DISK_BLOCK_KB + degr_csize;
2468 /* round estimate to next multiple of DISK_BLOCK_KB */
2469 sp->degr_csize = am_round(sp->degr_csize, DISK_BLOCK_KB);
2470 sp->degr_time = degr_time;
2471 sp->degr_kps = degr_kps;
2473 sp->degr_level = -1;
2474 sp->degr_dumpdate = NULL;
2478 sp->dump_attempted = 0;
2479 sp->taper_attempted = 0;
2484 sp->timestamp = (time_t)0;
2485 sp->destname = NULL;
2488 dp->up = (char *) sp;
2489 if(dp->host->features == NULL) {
2490 dp->host->features = am_string_to_feature(features);
2492 remove_disk(&waitq, dp);
2493 if (dp->to_holdingdisk == HOLD_NEVER) {
2494 enqueue_disk(&directq, dp);
2496 enqueue_disk(&runq, dp);
2498 flush_size += sp->act_size;
2501 g_printf(_("driver: flush size %lld\n"), (long long)flush_size);
2504 log_add(L_WARNING, _("WARNING: got empty schedule from planner"));
2505 if(need_degraded==1) start_degraded_mode(&runq);
2507 start_some_dumps(&runq);
2510 static unsigned long
2518 unsigned long maxusage=0;
2519 unsigned long curusage=0;
2520 for(p = disklist_netifs(); p != NULL; p = p->next) {
2521 maxusage += interface_get_maxusage(p->config);
2522 curusage += p->curusage;
2524 if (maxusage >= curusage)
2525 res = maxusage - curusage;
2530 if ((unsigned long)interface_get_maxusage(ip->config) >= ip->curusage)
2531 res = interface_get_maxusage(ip->config) - ip->curusage;
2546 g_printf(_("driver: interface-state time %s"), time_str);
2548 for(ip = disklist_netifs(); ip != NULL; ip = ip->next) {
2549 g_printf(_(" if %s: free %lu"), interface_name(ip->config), free_kps(ip));
2559 ip->curusage += kps;
2563 deallocate_bandwidth(
2567 assert(kps <= ip->curusage);
2568 ip->curusage -= kps;
2579 total_free = (off_t)0;
2580 for(ha = holdalloc; ha != NULL; ha = ha->next) {
2581 diff = ha->disksize - ha->allocated_space;
2589 * We return an array of pointers to assignedhd_t. The array contains at
2590 * most one entry per holding disk. The list of pointers is terminated by
2591 * a NULL pointer. Each entry contains a pointer to a holdingdisk and
2592 * how much diskspace to use on that disk. Later on, assign_holdingdisk
2593 * will allocate the given amount of space.
2594 * If there is not enough room on the holdingdisks, NULL is returned.
2597 static assignedhd_t **
2601 assignedhd_t * pref)
2603 assignedhd_t **result = NULL;
2604 holdalloc_t *ha, *minp;
2608 off_t halloc, dalloc, hfree, dfree;
2610 (void)cur_idle; /* Quiet unused parameter warning */
2612 if (size < 2*DISK_BLOCK_KB)
2613 size = 2*DISK_BLOCK_KB;
2614 size = am_round(size, (off_t)DISK_BLOCK_KB);
2616 hold_debug(1, _("find_diskspace: want %lld K\n"),
2619 used = alloc(SIZEOF(*used) * num_holdalloc);/*disks used during this run*/
2620 memset( used, 0, (size_t)num_holdalloc );
2621 result = alloc(SIZEOF(assignedhd_t *) * (num_holdalloc + 1));
2624 while( i < num_holdalloc && size > (off_t)0 ) {
2625 /* find the holdingdisk with the fewest active dumpers and among
2626 * those the one with the biggest free space
2628 minp = NULL; minj = -1;
2629 for(j = 0, ha = holdalloc; ha != NULL; ha = ha->next, j++ ) {
2630 if( pref && pref->disk == ha && !used[j] &&
2631 ha->allocated_space <= ha->disksize - (off_t)DISK_BLOCK_KB) {
2636 else if( ha->allocated_space <= ha->disksize - (off_t)(2*DISK_BLOCK_KB) &&
2639 ha->allocated_dumpers < minp->allocated_dumpers ||
2640 (ha->allocated_dumpers == minp->allocated_dumpers &&
2641 ha->disksize-ha->allocated_space > minp->disksize-minp->allocated_space)) ) {
2648 if( !minp ) { break; } /* all holding disks are full */
2651 /* hfree = free space on the disk */
2652 hfree = minp->disksize - minp->allocated_space;
2654 /* dfree = free space for data, remove 1 header for each chunksize */
2655 dfree = hfree - (((hfree-(off_t)1)/holdingdisk_get_chunksize(minp->hdisk))+(off_t)1) * (off_t)DISK_BLOCK_KB;
2657 /* dalloc = space I can allocate for data */
2658 dalloc = ( dfree < size ) ? dfree : size;
2660 /* halloc = space to allocate, including 1 header for each chunksize */
2661 halloc = dalloc + (((dalloc-(off_t)1)/holdingdisk_get_chunksize(minp->hdisk))+(off_t)1) * (off_t)DISK_BLOCK_KB;
2663 hold_debug(1, _("find_diskspace: find diskspace: size %lld hf %lld df %lld da %lld ha %lld\n"),
2670 result[i] = alloc(SIZEOF(assignedhd_t));
2671 result[i]->disk = minp;
2672 result[i]->reserved = halloc;
2673 result[i]->used = (off_t)0;
2674 result[i]->destname = NULL;
2680 if(size != (off_t)0) { /* not enough space available */
2681 g_printf(_("find diskspace: not enough diskspace. Left with %lld K\n"), (long long)size);
2683 free_assignedhd(result);
2687 if (debug_holding > 1) {
2688 for( i = 0; result && result[i]; i++ ) {
2689 hold_debug(1, _("find_diskspace: find diskspace: selected %s free %lld reserved %lld dumpers %d\n"),
2690 holdingdisk_get_diskdir(result[i]->disk->hdisk),
2691 (long long)(result[i]->disk->disksize -
2692 result[i]->disk->allocated_space),
2693 (long long)result[i]->reserved,
2694 result[i]->disk->allocated_dumpers);
2703 assignedhd_t ** holdp,
2708 char *sfn = sanitise_filename(diskp->name);
2710 assignedhd_t **new_holdp;
2713 g_snprintf( lvl, SIZEOF(lvl), "%d", sched(diskp)->level );
2715 size = am_round(sched(diskp)->est_size - sched(diskp)->act_size,
2716 (off_t)DISK_BLOCK_KB);
2718 for( c = 0; holdp[c]; c++ )
2719 (void)c; /* count number of disks */
2721 /* allocate memory for sched(diskp)->holdp */
2722 for(j = 0; sched(diskp)->holdp && sched(diskp)->holdp[j]; j++)
2723 (void)j; /* Quiet lint */
2724 new_holdp = (assignedhd_t **)alloc(SIZEOF(assignedhd_t*)*(j+c+1));
2725 if (sched(diskp)->holdp) {
2726 memcpy(new_holdp, sched(diskp)->holdp, j * SIZEOF(*new_holdp));
2727 amfree(sched(diskp)->holdp);
2729 sched(diskp)->holdp = new_holdp;
2733 if( j > 0 ) { /* This is a request for additional diskspace. See if we can
2734 * merge assignedhd_t's */
2736 if( sched(diskp)->holdp[j-1]->disk == holdp[0]->disk ) { /* Yes! */
2737 sched(diskp)->holdp[j-1]->reserved += holdp[0]->reserved;
2738 holdp[0]->disk->allocated_space += holdp[0]->reserved;
2739 size = (holdp[0]->reserved>size) ? (off_t)0 : size-holdp[0]->reserved;
2740 qname = quote_string(diskp->name);
2741 hold_debug(1, _("assign_holdingdisk: merging holding disk %s to disk %s:%s, add %lld for reserved %lld, left %lld\n"),
2742 holdingdisk_get_diskdir(
2743 sched(diskp)->holdp[j-1]->disk->hdisk),
2744 diskp->host->hostname, qname,
2745 (long long)holdp[0]->reserved,
2746 (long long)sched(diskp)->holdp[j-1]->reserved,
2755 /* copy assignedhd_s to sched(diskp), adjust allocated_space */
2756 for( ; holdp[i]; i++ ) {
2757 holdp[i]->destname = newvstralloc( holdp[i]->destname,
2758 holdingdisk_get_diskdir(holdp[i]->disk->hdisk), "/",
2759 hd_driver_timestamp, "/",
2760 diskp->host->hostname, ".",
2763 sched(diskp)->holdp[j++] = holdp[i];
2764 holdp[i]->disk->allocated_space += holdp[i]->reserved;
2765 size = (holdp[i]->reserved > size) ? (off_t)0 :
2766 (size - holdp[i]->reserved);
2767 qname = quote_string(diskp->name);
2769 _("assign_holdingdisk: %d assigning holding disk %s to disk %s:%s, reserved %lld, left %lld\n"),
2770 i, holdingdisk_get_diskdir(holdp[i]->disk->hdisk),
2771 diskp->host->hostname, qname,
2772 (long long)holdp[i]->reserved,
2775 holdp[i] = NULL; /* so it doesn't get free()d... */
2777 sched(diskp)->holdp[j] = NULL;
2788 assignedhd_t **holdp;
2789 off_t total = (off_t)0;
2792 char *qname, *hqname, *qdest;
2794 (void)cmd; /* Quiet unused parameter warning */
2796 qname = quote_string(diskp->name);
2797 qdest = quote_string(sched(diskp)->destname);
2798 hold_debug(1, _("adjust_diskspace: %s:%s %s\n"),
2799 diskp->host->hostname, qname, qdest);
2801 holdp = sched(diskp)->holdp;
2803 assert(holdp != NULL);
2805 for( i = 0; holdp[i]; i++ ) { /* for each allocated disk */
2806 diff = holdp[i]->used - holdp[i]->reserved;
2807 total += holdp[i]->used;
2808 holdp[i]->disk->allocated_space += diff;
2809 hqname = quote_string(holdingdisk_name(holdp[i]->disk->hdisk));
2810 hold_debug(1, _("adjust_diskspace: hdisk %s done, reserved %lld used %lld diff %lld alloc %lld dumpers %d\n"),
2811 holdingdisk_name(holdp[i]->disk->hdisk),
2812 (long long)holdp[i]->reserved,
2813 (long long)holdp[i]->used,
2815 (long long)holdp[i]->disk->allocated_space,
2816 holdp[i]->disk->allocated_dumpers );
2817 holdp[i]->reserved += diff;
2821 sched(diskp)->act_size = total;
2823 hold_debug(1, _("adjust_diskspace: after: disk %s:%s used %lld\n"),
2824 diskp->host->hostname, qname,
2825 (long long)sched(diskp)->act_size);
2834 assignedhd_t **holdp;
2837 holdp = sched(diskp)->holdp;
2839 assert(holdp != NULL);
2841 for( i = 0; holdp[i]; i++ ) { /* for each disk */
2842 /* find all files of this dump on that disk, and subtract their
2843 * reserved sizes from the disk's allocated space
2845 holdp[i]->disk->allocated_space -= holdp[i]->used;
2848 holding_file_unlink(holdp[0]->destname); /* no need for the entire list,
2849 * because holding_file_unlink
2850 * will walk through all files
2851 * using cont_filename */
2852 free_assignedhd(sched(diskp)->holdp);
2853 sched(diskp)->holdp = NULL;
2854 sched(diskp)->act_size = (off_t)0;
2857 static assignedhd_t **
2864 char buffer[DISK_BLOCK_BYTES];
2866 assignedhd_t **result;
2869 char dirname[1000], *ch;
2871 char *filename = destname;
2873 memset(buffer, 0, sizeof(buffer));
2874 used = alloc(SIZEOF(off_t) * num_holdalloc);
2875 for(i=0;i<num_holdalloc;i++)
2877 result = alloc(SIZEOF(assignedhd_t *) * (num_holdalloc + 1));
2879 while(filename != NULL && filename[0] != '\0') {
2880 strncpy(dirname, filename, 999);
2882 ch = strrchr(dirname,'/');
2884 ch = strrchr(dirname,'/');
2887 for(j = 0, ha = holdalloc; ha != NULL; ha = ha->next, j++ ) {
2888 if(strcmp(dirname, holdingdisk_get_diskdir(ha->hdisk))==0) {
2893 if(stat(filename, &finfo) == -1) {
2894 g_fprintf(stderr, _("stat %s: %s\n"), filename, strerror(errno));
2895 finfo.st_size = (off_t)0;
2897 used[j] += ((off_t)finfo.st_size+(off_t)1023)/(off_t)1024;
2898 if((fd = open(filename,O_RDONLY)) == -1) {
2899 g_fprintf(stderr,_("build_diskspace: open of %s failed: %s\n"),
2900 filename, strerror(errno));
2903 if ((buflen = fullread(fd, buffer, SIZEOF(buffer))) > 0) {;
2904 parse_file_header(buffer, &file, (size_t)buflen);
2907 filename = file.cont_filename;
2910 for(j = 0, i=0, ha = holdalloc; ha != NULL; ha = ha->next, j++ ) {
2911 if(used[j] != (off_t)0) {
2912 result[i] = alloc(SIZEOF(assignedhd_t));
2913 result[i]->disk = ha;
2914 result[i]->reserved = used[j];
2915 result[i]->used = used[j];
2916 result[i]->destname = stralloc(destname);
2934 g_printf(_("driver: hdisk-state time %s"), time_str);
2936 for(ha = holdalloc, dsk = 0; ha != NULL; ha = ha->next, dsk++) {
2937 diff = ha->disksize - ha->allocated_space;
2938 g_printf(_(" hdisk %d: free %lld dumpers %d"), dsk,
2939 (long long)diff, ha->allocated_dumpers);
2945 update_failed_dump_to_tape(
2949 * should simply set no_bump
2952 time_t save_timestamp = sched(dp)->timestamp;
2953 /* setting timestamp to 0 removes the current level from the
2954 * database, so that we ensure that it will not be bumped to the
2955 * next level on the next run. If we didn't do this, dumpdates or
2956 * gnutar-lists might have been updated already, and a bumped
2957 * incremental might be created. */
2958 sched(dp)->timestamp = 0;
2959 update_info_dumper(dp, (off_t)-1, (off_t)-1, (time_t)-1);
2960 sched(dp)->timestamp = save_timestamp;
2963 /* ------------------- */
2971 char *result_argv[MAX_ARGS+1];
2974 qname = quote_string(dp->name);
2975 g_printf(_("driver: dumping %s:%s directly to tape\n"),
2976 dp->host->hostname, qname);
2979 /* pick a dumper and fail if there are no idle dumpers */
2981 dumper = idle_dumper();
2983 g_printf(_("driver: no idle dumpers for %s:%s.\n"),
2984 dp->host->hostname, qname);
2986 log_add(L_WARNING, _("no idle dumpers for %s:%s.\n"),
2987 dp->host->hostname, qname);
2989 return; /* fatal problem */
2992 /* tell the taper to read from a port number of its choice */
2994 taper_cmd(PORT_WRITE, dp, NULL, sched(dp)->level, sched(dp)->datestamp);
2995 cmd = getresult(taper, 1, &result_argc, result_argv, MAX_ARGS+1);
2997 g_printf(_("driver: did not get PORT from taper for %s:%s\n"),
2998 dp->host->hostname, qname);
3000 log_add(L_WARNING, _("driver: did not get PORT from taper for %s:%s.\n"),
3001 dp->host->hostname, qname);
3003 return; /* fatal problem */
3007 /* copy port number */
3008 dumper->output_port = atoi(result_argv[2]);
3011 dumper->chunker = NULL;
3012 dumper->result = LAST_TOK;
3013 taper_result = LAST_TOK;
3014 sched(dp)->dumper = dumper;
3016 /* tell the dumper to dump to a port */
3017 dumper_cmd(dumper, PORT_DUMP, dp);
3018 dp->host->start_t = time(NULL) + 15;
3020 /* update statistics & print state */
3022 taper_busy = dumper->busy = 1;
3023 taper_input_error = NULL;
3024 taper_tape_error = NULL;
3025 taper_dumper = dumper;
3027 taper_input_error = NULL;
3028 taper_tape_error = NULL;
3029 taper_first_label = NULL;
3031 taper_state |= TAPER_STATE_DUMP_TO_TAPE;
3032 sched(dp)->act_size = sched(dp)->est_size;
3033 dp->host->inprogress += 1;
3035 sched(dp)->timestamp = time((time_t *)0);
3036 allocate_bandwidth(dp->host->netif, sched(dp)->est_kps);
3037 idle_reason = NOT_IDLE;
3041 dumper->ev_read = event_register(dumper->fd, EV_READFD,
3042 handle_dumper_result, dumper);
3043 taper_ev_read = event_register(taper, EV_READFD,
3044 handle_taper_result, NULL);
3054 for(len = 0, p = q.head; p != NULL; len++, p = p->next)
3055 (void)len; /* Quiet lint */
3060 short_dump_state(void)
3065 wall_time = walltime_str(curclock());
3067 g_printf(_("driver: state time %s "), wall_time);
3068 g_printf(_("free kps: %lu space: %lld taper: "),
3070 (long long)free_space());
3071 if(degraded_mode) g_printf(_("DOWN"));
3072 else if(!taper_busy) g_printf(_("idle"));
3073 else g_printf(_("writing"));
3075 for(i = 0; i < inparallel; i++) if(!dmptable[i].busy) nidle++;
3076 g_printf(_(" idle-dumpers: %d"), nidle);
3077 g_printf(_(" qlen tapeq: %d"), queue_length(tapeq));
3078 g_printf(_(" runq: %d"), queue_length(runq));
3079 g_printf(_(" roomq: %d"), queue_length(roomq));
3080 g_printf(_(" wakeup: %d"), (int)sleep_time);
3081 g_printf(_(" driver-idle: %s\n"), _(idle_strings[idle_reason]));
3082 interface_state(wall_time);
3083 holdingdisk_state(wall_time);
3087 static TapeAction tape_action(void)
3089 TapeAction result = TAPE_ACTION_NO_ACTION;
3097 off_t dump_to_disk_size;
3098 int dump_to_disk_terminated;
3101 for(dumper = dmptable; dumper < (dmptable+inparallel); dumper++) {
3103 dumpers_size += sched(dumper->dp)->est_size;
3105 driver_debug(1, _("dumpers_size: %lld\n"), (long long)dumpers_size);
3108 for(dp = runq.head; dp != NULL; dp = dp->next) {
3109 runq_size += sched(dp)->est_size;
3111 driver_debug(1, _("runq_size: %lld\n"), (long long)runq_size);
3114 for(dp = directq.head; dp != NULL; dp = dp->next) {
3115 directq_size += sched(dp)->est_size;
3117 driver_debug(1, _("directq_size: %lld\n"), (long long)directq_size);
3120 for(dp = tapeq.head; dp != NULL; dp = dp->next) {
3121 tapeq_size += sched(dp)->act_size;
3124 tapeq_size += sched(taper_disk)->act_size - taper_written;
3126 driver_debug(1, _("tapeq_size: %lld\n"), (long long)tapeq_size);
3128 sched_size = runq_size + tapeq_size + dumpers_size;
3129 driver_debug(1, _("sched_size: %lld\n"), (long long)sched_size);
3131 dump_to_disk_size = dumpers_size + runq_size;
3132 driver_debug(1, _("dump_to_disk_size: %lld\n"), (long long)dump_to_disk_size);
3134 dump_to_disk_terminated = schedule_done && dump_to_disk_size == 0;
3136 // Changing conditionals can produce a driver hang, take care.
3138 // when to start writting to a new tape
3139 if ((taper_state & TAPER_STATE_WAIT_FOR_TAPE) &&
3140 ((taper_state & TAPER_STATE_DUMP_TO_TAPE) || // for dump to tape
3141 !empty(directq) || // if a dle is waiting for a dump to tape
3142 !empty(roomq) || // holding disk constraint
3143 idle_reason == IDLE_NO_DISKSPACE || // holding disk constraint
3144 (flush_threshold_dumped < tapeq_size && // flush-threshold-dumped &&
3145 flush_threshold_scheduled < sched_size) || // flush-threshold-scheduled
3146 (taperflush < tapeq_size && // taperflush
3147 (force_flush == 1 || // if force_flush
3148 dump_to_disk_terminated)) // or all dump to disk terminated
3150 result |= TAPE_ACTION_NEW_TAPE;
3151 // when to stop using new tape
3152 } else if ((taper_state & TAPER_STATE_WAIT_FOR_TAPE) &&
3153 (taperflush >= tapeq_size && // taperflush criteria not meet
3154 (force_flush == 1 || // if force_flush
3155 dump_to_disk_terminated)) // or all dump to disk terminated
3157 result |= TAPE_ACTION_NO_NEW_TAPE;
3160 // when to start a flush
3161 // We don't start a flush if taper_tape_started == 1 && dump_to_disk_terminated && force_flush == 0,
3162 // it is a criteria need to exit the first event_loop without flushing everything to tape,
3163 // they will be flush in another event_loop.
3164 if (!degraded_mode && !taper_busy && !empty(tapeq) &&
3165 (!((taper_state & TAPER_STATE_TAPE_STARTED) &&
3166 dump_to_disk_terminated && force_flush == 0) || // if tape already started and dump to disk not terminated
3167 ((taper_state & TAPER_STATE_TAPE_STARTED) &&
3168 force_flush == 1) || // if tape already started and force_flush
3169 !empty(roomq) || // holding disk constraint
3170 idle_reason == IDLE_NO_DISKSPACE || // holding disk constraint
3171 (flush_threshold_dumped < tapeq_size && // flush-threshold-dumped &&
3172 flush_threshold_scheduled < sched_size) || // flush-threshold-scheduled
3173 (force_flush == 1 && taperflush < tapeq_size))) { // taperflush if force_flush
3174 result |= TAPE_ACTION_START_A_FLUSH;
3188 g_printf("================\n");
3189 g_printf(_("driver state at time %s: %s\n"), walltime_str(curclock()), str);
3190 g_printf(_("free kps: %lu, space: %lld\n"),
3192 (long long)free_space());
3193 if(degraded_mode) g_printf(_("taper: DOWN\n"));
3194 else if(!taper_busy) g_printf(_("taper: idle\n"));
3195 else g_printf(_("taper: writing %s:%s.%d est size %lld\n"),
3196 taper_disk->host->hostname, taper_disk->name,
3197 sched(taper_disk)->level,
3198 (long long)sched(taper_disk)->est_size);
3199 for(i = 0; i < inparallel; i++) {
3200 dp = dmptable[i].dp;
3201 if(!dmptable[i].busy)
3202 g_printf(_("%s: idle\n"), dmptable[i].name);
3204 qname = quote_string(dp->name);
3205 g_printf(_("%s: dumping %s:%s.%d est kps %d size %lld time %lu\n"),
3206 dmptable[i].name, dp->host->hostname, qname, sched(dp)->level,
3207 sched(dp)->est_kps, (long long)sched(dp)->est_size, sched(dp)->est_time);
3210 dump_queue("TAPE", tapeq, 5, stdout);
3211 dump_queue("ROOM", roomq, 5, stdout);
3212 dump_queue("RUN ", runq, 5, stdout);
3213 g_printf("================\n");