2 * Subroutines to handle Catalog reqests sent to the Director
3 * Reqests/commands from the Director are handled in dircmd.c
5 * Kern Sibbald, December 2000
10 Copyright (C) 2000-2004 Kern Sibbald and John Walker
12 This program is free software; you can redistribute it and/or
13 modify it under the terms of the GNU General Public License as
14 published by the Free Software Foundation; either version 2 of
15 the License, or (at your option) any later version.
17 This program is distributed in the hope that it will be useful,
18 but WITHOUT ANY WARRANTY; without even the implied warranty of
19 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20 General Public License for more details.
22 You should have received a copy of the GNU General Public
23 License along with this program; if not, write to the Free
24 Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
29 #include "bacula.h" /* pull in global headers */
30 #include "stored.h" /* pull in Storage Deamon headers */
32 /* Requests sent to the Director */
33 static char Find_media[] = "CatReq Job=%s FindMedia=%d\n";
34 static char Get_Vol_Info[] = "CatReq Job=%s GetVolInfo VolName=%s write=%d\n";
35 static char Update_media[] = "CatReq Job=%s UpdateMedia VolName=%s"
36 " VolJobs=%u VolFiles=%u VolBlocks=%u VolBytes=%s VolMounts=%u"
37 " VolErrors=%u VolWrites=%u MaxVolBytes=%s EndTime=%d VolStatus=%s"
38 " Slot=%d relabel=%d InChanger=%d VolReadTime=%s VolWriteTime=%s\n";
39 static char Create_job_media[] = "CatReq Job=%s CreateJobMedia"
40 " FirstIndex=%u LastIndex=%u StartFile=%u EndFile=%u"
41 " StartBlock=%u EndBlock=%u\n";
42 static char FileAttributes[] = "UpdCat Job=%s FileAttributes ";
43 static char Job_status[] = "3012 Job %s jobstatus %d\n";
46 /* Responses received from the Director */
47 static char OK_media[] = "1000 OK VolName=%127s VolJobs=%u VolFiles=%u"
48 " VolBlocks=%u VolBytes=%" lld " VolMounts=%u VolErrors=%u VolWrites=%u"
49 " MaxVolBytes=%" lld " VolCapacityBytes=%" lld " VolStatus=%20s"
50 " Slot=%d MaxVolJobs=%u MaxVolFiles=%u InChanger=%d"
51 " VolReadTime=%" lld " VolWriteTime=%" lld " EndFile=%u EndBlock=%u";
54 static char OK_create[] = "1000 OK CreateJobMedia\n";
56 /* Forward referenced functions */
57 static int wait_for_sysop(DCR *dcr);
60 * Send current JobStatus to Director
62 bool dir_send_job_status(JCR *jcr)
64 return bnet_fsend(jcr->dir_bsock, Job_status, jcr->Job, jcr->JobStatus);
69 * dir_get_volume_info()
71 * dir_find_next_appendable_volume()
73 * Returns: true on success and vol info in dcr->VolCatInfo
76 static bool do_get_volume_info(DCR *dcr)
79 BSOCK *dir = jcr->dir_bsock;
84 dcr->VolumeName[0] = 0; /* No volume */
85 if (bnet_recv(dir) <= 0) {
86 Dmsg0(200, "getvolname error bnet_recv\n");
87 Mmsg(jcr->errmsg, _("Network error on bnet_recv in req_vol_info.\n"));
90 memset(&vol, 0, sizeof(vol));
91 Dmsg1(300, "Get vol info=%s", dir->msg);
92 n = sscanf(dir->msg, OK_media, vol.VolCatName,
93 &vol.VolCatJobs, &vol.VolCatFiles,
94 &vol.VolCatBlocks, &vol.VolCatBytes,
95 &vol.VolCatMounts, &vol.VolCatErrors,
96 &vol.VolCatWrites, &vol.VolCatMaxBytes,
97 &vol.VolCatCapacityBytes, vol.VolCatStatus,
98 &vol.Slot, &vol.VolCatMaxJobs, &vol.VolCatMaxFiles,
99 &InChanger, &vol.VolReadTime, &vol.VolWriteTime,
100 &vol.EndFile, &vol.EndBlock);
102 Dmsg2(100, "Bad response from Dir fields=%d: %s\n", n, dir->msg);
103 Mmsg(jcr->errmsg, _("Error getting Volume info: %s\n"), dir->msg);
106 vol.InChanger = InChanger; /* bool in structure */
107 unbash_spaces(vol.VolCatName);
108 bstrncpy(dcr->VolumeName, vol.VolCatName, sizeof(dcr->VolumeName));
109 memcpy(&dcr->VolCatInfo, &vol, sizeof(dcr->VolCatInfo));
111 Dmsg2(300, "do_reqest_vol_info got slot=%d Volume=%s\n",
112 vol.Slot, vol.VolCatName);
118 * Get Volume info for a specific volume from the Director's Database
120 * Returns: true on success (not Director guarantees that Pool and MediaType
121 * are correct and VolStatus==Append or
122 * VolStatus==Recycle)
125 * Volume information returned in jcr
127 bool dir_get_volume_info(DCR *dcr, enum get_vol_info_rw writing)
130 BSOCK *dir = jcr->dir_bsock;
132 bstrncpy(dcr->VolCatInfo.VolCatName, dcr->VolumeName, sizeof(dcr->VolCatInfo.VolCatName));
133 Dmsg1(300, "dir_get_volume_info=%s\n", dcr->VolCatInfo.VolCatName);
134 bash_spaces(dcr->VolCatInfo.VolCatName);
135 bnet_fsend(dir, Get_Vol_Info, jcr->Job, dcr->VolCatInfo.VolCatName,
136 writing==GET_VOL_INFO_FOR_WRITE?1:0);
137 return do_get_volume_info(dcr);
143 * Get info on the next appendable volume in the Director's database
144 * Returns: true on success
147 * Volume information returned in dcr
150 bool dir_find_next_appendable_volume(DCR *dcr)
153 BSOCK *dir = jcr->dir_bsock;
156 Dmsg0(200, "dir_find_next_appendable_volume\n");
158 * Try the three oldest or most available volumes. Note,
159 * the most available could already be mounted on another
160 * drive, so we continue looking for a not in use Volume.
162 for (int vol_index=1; vol_index < 3; vol_index++) {
163 bnet_fsend(dir, Find_media, jcr->Job, vol_index);
164 if (do_get_volume_info(dcr)) {
165 Dmsg2(300, "JobId=%d got possible Vol=%s\n", jcr->JobId, dcr->VolumeName);
168 * Walk through all jobs and see if the volume is
169 * already mounted. If so, try a different one.
170 * This would be better done by walking through
176 free_locked_jcr(njcr);
179 Dmsg2(300, "Compare to JobId=%d using Vol=%s\n", njcr->JobId, njcr->dcr->VolumeName);
180 if (njcr->dcr && strcmp(dcr->VolumeName, njcr->dcr->VolumeName) == 0) {
182 Dmsg1(400, "Vol in use by JobId=%u\n", njcr->JobId);
183 free_locked_jcr(njcr);
186 free_locked_jcr(njcr);
190 Dmsg0(400, "dir_find_next_appendable_volume return true\n");
191 return true; /* Got good Volume */
194 Dmsg0(200, "No volume info, return false\n");
198 Dmsg0(400, "dir_find_next_appendable_volume return true\n");
204 * After writing a Volume, send the updated statistics
205 * back to the director. The information comes from the
208 bool dir_update_volume_info(DCR *dcr, bool label)
211 BSOCK *dir = jcr->dir_bsock;
212 DEVICE *dev = dcr->dev;
213 time_t LastWritten = time(NULL);
214 char ed1[50], ed2[50], ed3[50], ed4[50];
215 VOLUME_CAT_INFO *vol = &dev->VolCatInfo;
218 if (vol->VolCatName[0] == 0) {
219 Jmsg0(jcr, M_FATAL, 0, _("NULL Volume name. This shouldn't happen!!!\n"));
220 Dmsg0(000, "NULL Volume name. This shouldn't happen!!!\n");
223 if (dev_state(dev, ST_READ)) {
224 Jmsg0(jcr, M_FATAL, 0, _("Attempt to update_volume_info in read mode!!!\n"));
225 Dmsg0(000, "Attempt to update_volume_info in read mode!!!\n");
229 Dmsg1(300, "Update cat VolFiles=%d\n", dev->file);
230 /* Just labeled or relabeled the tape */
232 bstrncpy(vol->VolCatStatus, "Append", sizeof(vol->VolCatStatus));
233 vol->VolCatBytes = 1; /* indicates tape labeled */
235 bash_spaces(vol->VolCatName);
236 InChanger = vol->InChanger;
237 bnet_fsend(dir, Update_media, jcr->Job,
238 vol->VolCatName, vol->VolCatJobs, vol->VolCatFiles,
239 vol->VolCatBlocks, edit_uint64(vol->VolCatBytes, ed1),
240 vol->VolCatMounts, vol->VolCatErrors,
241 vol->VolCatWrites, edit_uint64(vol->VolCatMaxBytes, ed2),
242 LastWritten, vol->VolCatStatus, vol->Slot, label,
243 InChanger, /* bool in structure */
244 edit_uint64(vol->VolReadTime, ed3),
245 edit_uint64(vol->VolWriteTime, ed4) );
247 Dmsg1(300, "update_volume_info(): %s", dir->msg);
248 unbash_spaces(vol->VolCatName);
250 if (!do_get_volume_info(dcr)) {
251 Jmsg(jcr, M_FATAL, 0, "%s", jcr->errmsg);
252 Dmsg1(000, "Didn't get vol info: %s", jcr->errmsg);
255 Dmsg1(420, "get_volume_info(): %s", dir->msg);
256 /* Update dev Volume info in case something changed (e.g. expired) */
257 memcpy(&dev->VolCatInfo, &dcr->VolCatInfo, sizeof(dev->VolCatInfo));
262 * After writing a Volume, create the JobMedia record.
264 bool dir_create_jobmedia_record(DCR *dcr)
267 BSOCK *dir = jcr->dir_bsock;
269 if (!dcr->WroteVol) {
270 return true; /* nothing written to tape */
273 dcr->WroteVol = false;
274 bnet_fsend(dir, Create_job_media, jcr->Job,
275 dcr->VolFirstIndex, dcr->VolLastIndex,
276 dcr->StartFile, dcr->EndFile,
277 dcr->StartBlock, dcr->EndBlock);
278 Dmsg1(400, "create_jobmedia(): %s", dir->msg);
279 if (bnet_recv(dir) <= 0) {
280 Dmsg0(190, "create_jobmedia error bnet_recv\n");
281 Jmsg(jcr, M_FATAL, 0, _("Error creating JobMedia record: ERR=%s\n"),
285 Dmsg1(400, "Create_jobmedia: %s", dir->msg);
286 if (strcmp(dir->msg, OK_create) != 0) {
287 Dmsg1(130, "Bad response from Dir: %s\n", dir->msg);
288 Jmsg(jcr, M_FATAL, 0, _("Error creating JobMedia record: %s\n"), dir->msg);
296 * Update File Attribute data
298 bool dir_update_file_attributes(DCR *dcr, DEV_RECORD *rec)
301 BSOCK *dir = jcr->dir_bsock;
304 dir->msglen = sprintf(dir->msg, FileAttributes, jcr->Job);
305 dir->msg = check_pool_memory_size(dir->msg, dir->msglen +
306 sizeof(DEV_RECORD) + rec->data_len);
307 ser_begin(dir->msg + dir->msglen, 0);
308 ser_uint32(rec->VolSessionId);
309 ser_uint32(rec->VolSessionTime);
310 ser_int32(rec->FileIndex);
311 ser_int32(rec->Stream);
312 ser_uint32(rec->data_len);
313 ser_bytes(rec->data, rec->data_len);
314 dir->msglen = ser_length(dir->msg);
315 return bnet_send(dir);
320 * Request the sysop to create an appendable volume
322 * Entered with device blocked.
323 * Leaves with device blocked.
325 * Returns: true on success (operator issues a mount command)
327 * Note, must create dev->errmsg on error return.
329 * On success, dcr->VolumeName and dcr->VolCatInfo contain
330 * information on suggested volume, but this may not be the
331 * same as what is actually mounted.
333 * When we return with success, the correct tape may or may not
334 * actually be mounted. The calling routine must read it and
337 bool dir_ask_sysop_to_create_appendable_volume(DCR *dcr)
342 DEVICE *dev = dcr->dev;
345 Dmsg0(400, "enter dir_ask_sysop_to_create_appendable_volume\n");
346 ASSERT(dev->dev_blocked);
348 if (job_canceled(jcr)) {
350 _("Job %s canceled while waiting for mount on Storage Device \"%s\".\n"),
351 jcr->Job, dcr->dev_name);
352 Jmsg(jcr, M_INFO, 0, "%s", dev->errmsg);
355 /* First pass, we *know* there are no appendable volumes, so no need to call */
356 if (!first && dir_find_next_appendable_volume(dcr)) { /* get suggested volume */
357 unmounted = (dev->dev_blocked == BST_UNMOUNTED) ||
358 (dev->dev_blocked == BST_UNMOUNTED_WAITING_FOR_SYSOP);
360 * If we have a valid volume name and we are not
361 * removable media, return now, or if we have a
362 * Slot for an autochanger, otherwise wait
363 * for the operator to mount the media.
365 if (!unmounted && ((dcr->VolumeName[0] && !dev_cap(dev, CAP_REM) &&
366 dev_cap(dev, CAP_LABEL)) ||
367 (dcr->VolumeName[0] && dcr->VolCatInfo.Slot))) {
368 Dmsg0(400, "Return 1 from mount without wait.\n");
371 jstat = JS_WaitMount;
373 Jmsg(jcr, M_MOUNT, 0, _(
374 "Please mount Volume \"%s\" on Storage Device \"%s\" for Job %s\n"
375 "Use \"mount\" command to release Job.\n"),
376 dcr->VolumeName, dcr->dev_name, jcr->Job);
377 Dmsg3(400, "Mount %s on %s for Job %s\n",
378 dcr->VolumeName, dcr->dev_name, jcr->Job);
381 jstat = JS_WaitMedia;
383 Jmsg(jcr, M_MOUNT, 0, _(
384 "Job %s waiting. Cannot find any appendable volumes.\n"
385 "Please use the \"label\" command to create a new Volume for:\n"
397 jcr->JobStatus = jstat;
398 dir_send_job_status(jcr);
400 stat = wait_for_sysop(dcr);
402 Dmsg1(400, "Poll timeout in create append vol on device %s\n", dev_name(dev));
406 if (stat == ETIMEDOUT) {
407 if (!double_dev_wait_time(dev)) {
408 Mmsg(dev->errmsg, _("Max time exceeded waiting to mount Storage Device \"%s\" for Job %s\n"),
409 dev_name(dev), jcr->Job);
410 Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
411 Dmsg1(400, "Gave up waiting on device %s\n", dev_name(dev));
412 return false; /* exceeded maximum waits */
416 if (stat == EINVAL) {
418 Mmsg2(dev->errmsg, _("pthread error in mount_next_volume stat=%d ERR=%s\n"),
419 stat, be.strerror(stat));
420 Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
425 Jmsg(jcr, M_WARNING, 0, _("pthread error in mount_next_volume stat=%d ERR=%s\n"), stat,
428 Dmsg1(400, "Someone woke me for device %s\n", dev_name(dev));
430 /* If no VolumeName, and cannot get one, try again */
431 if (dcr->VolumeName[0] == 0 && !job_canceled(jcr) &&
432 !dir_find_next_appendable_volume(dcr)) {
433 Jmsg(jcr, M_MOUNT, 0, _(
434 "Someone woke me up, but I cannot find any appendable\n"
435 "volumes for Job=%s.\n"), jcr->Job);
436 /* Restart wait counters after user interaction */
437 init_dev_wait_timers(dev);
440 unmounted = (dev->dev_blocked == BST_UNMOUNTED) ||
441 (dev->dev_blocked == BST_UNMOUNTED_WAITING_FOR_SYSOP);
443 continue; /* continue to wait */
447 * Device mounted, we have a volume, break and return
451 set_jcr_job_status(jcr, JS_Running);
452 dir_send_job_status(jcr);
453 Dmsg0(400, "leave dir_ask_sysop_to_mount_create_appendable_volume\n");
458 * Request to mount specific Volume
460 * Entered with device blocked and dcr->VolumeName is desired
462 * Leaves with device blocked.
464 * Returns: true on success (operator issues a mount command)
466 * Note, must create dev->errmsg on error return.
469 bool dir_ask_sysop_to_mount_volume(DCR *dcr)
473 DEVICE *dev = dcr->dev;
476 Dmsg0(400, "enter dir_ask_sysop_to_mount_volume\n");
477 if (!dcr->VolumeName[0]) {
478 Mmsg0(dev->errmsg, _("Cannot request another volume: no volume name given.\n"));
481 ASSERT(dev->dev_blocked);
483 if (job_canceled(jcr)) {
484 Mmsg(dev->errmsg, _("Job %s canceled while waiting for mount on Storage Device \"%s\".\n"),
485 jcr->Job, dcr->dev_name);
490 msg = _("Please mount");
491 Jmsg(jcr, M_MOUNT, 0, _("%s Volume \"%s\" on Storage Device \"%s\" for Job %s\n"),
492 msg, dcr->VolumeName, dcr->dev_name, jcr->Job);
493 Dmsg3(400, "Mount \"%s\" on device \"%s\" for Job %s\n",
494 dcr->VolumeName, dcr->dev_name, jcr->Job);
497 jcr->JobStatus = JS_WaitMount;
498 dir_send_job_status(jcr);
500 stat = wait_for_sysop(dcr); ; /* wait on device */
502 Dmsg1(400, "Poll timeout in mount vol on device %s\n", dev_name(dev));
503 Dmsg1(400, "Blocked=%s\n", edit_blocked_reason(dev));
507 if (stat == ETIMEDOUT) {
508 if (!double_dev_wait_time(dev)) {
509 Mmsg(dev->errmsg, _("Max time exceeded waiting to mount Storage Device \"%s\" for Job %s\n"),
510 dev_name(dev), jcr->Job);
511 Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
512 Dmsg1(400, "Gave up waiting on device %s\n", dev_name(dev));
513 return false; /* exceeded maximum waits */
517 if (stat == EINVAL) {
519 Mmsg2(dev->errmsg, _("pthread error in mount_volume stat=%d ERR=%s\n"),
520 stat, be.strerror(stat));
521 Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
526 Jmsg(jcr, M_FATAL, 0, _("pthread error in mount_next_volume stat=%d ERR=%s\n"), stat,
529 Dmsg1(400, "Someone woke me for device %s\n", dev_name(dev));
532 set_jcr_job_status(jcr, JS_Running);
533 dir_send_job_status(jcr);
534 Dmsg0(400, "leave dir_ask_sysop_to_mount_volume\n");
539 * Wait for SysOp to mount a tape
541 static int wait_for_sysop(DCR *dcr)
545 struct timespec timeout;
546 time_t last_heartbeat = 0;
547 time_t first_start = time(NULL);
551 DEVICE *dev = dcr->dev;
555 unmounted = (dev->dev_blocked == BST_UNMOUNTED) ||
556 (dev->dev_blocked == BST_UNMOUNTED_WAITING_FOR_SYSOP);
560 * Wait requested time (dev->rem_wait_sec). However, we also wake up every
561 * HB_TIME seconds and send a heartbeat to the FD and the Director
562 * to keep stateful firewalls from closing them down while waiting
565 add_wait = dev->rem_wait_sec;
566 if (me->heartbeat_interval && add_wait > me->heartbeat_interval) {
567 add_wait = me->heartbeat_interval;
569 /* If the user did not unmount the tape and we are polling, ensure
570 * that we poll at the correct interval.
572 if (!unmounted && dev->vol_poll_interval && add_wait > dev->vol_poll_interval) {
573 add_wait = dev->vol_poll_interval;
575 gettimeofday(&tv, &tz);
576 timeout.tv_nsec = tv.tv_usec * 1000;
577 timeout.tv_sec = tv.tv_sec + add_wait;
580 dev->dev_prev_blocked = dev->dev_blocked;
581 dev->dev_blocked = BST_WAITING_FOR_SYSOP; /* indicate waiting for mount */
584 for ( ; !job_canceled(jcr); ) {
587 Dmsg3(400, "I'm going to sleep on device %s. HB=%d wait=%d\n", dev_name(dev),
588 (int)me->heartbeat_interval, dev->wait_sec);
590 /* Wait required time */
591 stat = pthread_cond_timedwait(&dev->wait_next_vol, &dev->mutex, &timeout);
592 Dmsg1(400, "Wokeup from sleep on device stat=%d\n", stat);
595 dev->rem_wait_sec -= (now - start);
597 /* Note, this always triggers the first time. We want that. */
598 if (me->heartbeat_interval) {
599 if (now - last_heartbeat >= me->heartbeat_interval) {
600 /* send heartbeats */
601 if (jcr->file_bsock) {
602 bnet_sig(jcr->file_bsock, BNET_HEARTBEAT);
603 Dmsg0(400, "Send heartbeat to FD.\n");
605 if (jcr->dir_bsock) {
606 bnet_sig(jcr->dir_bsock, BNET_HEARTBEAT);
608 last_heartbeat = now;
613 * Check if user unmounted the device while we were waiting
615 unmounted = (dev->dev_blocked == BST_UNMOUNTED) ||
616 (dev->dev_blocked == BST_UNMOUNTED_WAITING_FOR_SYSOP);
618 if (stat != ETIMEDOUT) { /* we blocked the device */
619 break; /* on error return */
621 if (dev->rem_wait_sec <= 0) { /* on exceeding wait time return */
622 Dmsg0(400, "Exceed wait time.\n");
626 if (!unmounted && dev->vol_poll_interval &&
627 (now - first_start >= dev->vol_poll_interval)) {
628 Dmsg1(400, "In wait blocked=%s\n", edit_blocked_reason(dev));
629 dev->poll = true; /* returning a poll event */
633 * Check if user mounted the device while we were waiting
635 if (dev->dev_blocked == BST_MOUNT) { /* mount request ? */
640 add_wait = dev->wait_sec - (now - start);
644 if (me->heartbeat_interval && add_wait > me->heartbeat_interval) {
645 add_wait = me->heartbeat_interval;
647 gettimeofday(&tv, &tz);
648 timeout.tv_nsec = tv.tv_usec * 1000;
649 timeout.tv_sec = tv.tv_sec + add_wait; /* additional wait */
650 Dmsg1(400, "Additional wait %d sec.\n", add_wait);
654 dev->dev_blocked = dev->dev_prev_blocked; /* restore entry state */