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) 2000-2003 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 Drive=%d InChanger=%d\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 Drive=%d InChanger=%d";
52 static char OK_create[] = "1000 OK CreateJobMedia\n";
54 /* Forward referenced functions */
55 static int wait_for_sysop(JCR *jcr, DEVICE *dev, int wait_sec);
58 * Send current JobStatus to Director
60 int dir_send_job_status(JCR *jcr)
62 return bnet_fsend(jcr->dir_bsock, Job_status, jcr->Job, jcr->JobStatus);
67 * dir_get_volume_info()
69 * dir_find_next_appendable_volume()
71 * Returns: 1 on success and vol info in jcr->VolCatInfo
74 static int do_get_volume_info(JCR *jcr)
76 BSOCK *dir = jcr->dir_bsock;
79 jcr->VolumeName[0] = 0; /* No volume */
80 if (bnet_recv(dir) <= 0) {
81 Dmsg0(200, "getvolname error bnet_recv\n");
82 Mmsg(&jcr->errmsg, _("Network error on bnet_recv in req_vol_info.\n"));
85 memset(&vol, 0, sizeof(vol));
86 if (sscanf(dir->msg, OK_media, vol.VolCatName,
87 &vol.VolCatJobs, &vol.VolCatFiles,
88 &vol.VolCatBlocks, &vol.VolCatBytes,
89 &vol.VolCatMounts, &vol.VolCatErrors,
90 &vol.VolCatWrites, &vol.VolCatMaxBytes,
91 &vol.VolCatCapacityBytes, vol.VolCatStatus,
92 &vol.Slot, &vol.VolCatMaxJobs, &vol.VolCatMaxFiles,
93 &vol.Drive, &vol.InChanger) != 16) {
95 Dmsg1(200, "Bad response from Dir: %s\n", dir->msg);
96 Mmsg(&jcr->errmsg, _("Error getting Volume info: %s\n"), dir->msg);
99 unbash_spaces(vol.VolCatName);
100 pm_strcpy(&jcr->VolumeName, vol.VolCatName); /* set desired VolumeName */
101 memcpy(&jcr->VolCatInfo, &vol, sizeof(jcr->VolCatInfo));
103 Dmsg2(200, "do_reqest_vol_info got slot=%d Volume=%s\n",
104 vol.Slot, vol.VolCatName);
110 * Get Volume info for a specific volume from the Director's Database
112 * Returns: 1 on success (not Director guarantees that Pool and MediaType
113 * are correct and VolStatus==Append or
114 * VolStatus==Recycle)
117 * Volume information returned in jcr
119 int dir_get_volume_info(JCR *jcr, enum get_vol_info_rw writing)
121 BSOCK *dir = jcr->dir_bsock;
123 bstrncpy(jcr->VolCatInfo.VolCatName, jcr->VolumeName, sizeof(jcr->VolCatInfo.VolCatName));
124 Dmsg1(200, "dir_get_volume_info=%s\n", jcr->VolCatInfo.VolCatName);
125 bash_spaces(jcr->VolCatInfo.VolCatName);
126 bnet_fsend(dir, Get_Vol_Info, jcr->Job, jcr->VolCatInfo.VolCatName,
127 writing==GET_VOL_INFO_FOR_WRITE?1:0);
128 return do_get_volume_info(jcr);
134 * Get info on the next appendable volume in the Director's database
135 * Returns: 1 on success
138 * Volume information returned in jcr
141 int dir_find_next_appendable_volume(JCR *jcr)
143 BSOCK *dir = jcr->dir_bsock;
145 Dmsg0(200, "dir_find_next_appendable_volume\n");
146 bnet_fsend(dir, Find_media, jcr->Job, 1);
147 return do_get_volume_info(jcr);
152 * After writing a Volume, send the updated statistics
153 * back to the director.
155 int dir_update_volume_info(JCR *jcr, DEVICE *dev, int label)
157 BSOCK *dir = jcr->dir_bsock;
158 time_t EndTime = time(NULL);
159 char ed1[50], ed2[50];
160 VOLUME_CAT_INFO *vol = &dev->VolCatInfo;
162 if (vol->VolCatName[0] == 0) {
163 Jmsg0(jcr, M_ERROR, 0, _("NULL Volume name. This shouldn't happen!!!\n"));
166 if (dev_state(dev, ST_READ)) {
167 Jmsg0(jcr, M_ERROR, 0, _("Attempt to update_volume_info in read mode!!!\n"));
170 if (!dev_state(dev, ST_LABEL)) {
171 Jmsg0(jcr, M_ERROR, 0, _("Attempt to update_volume_info on non-labeled Volume!!!\n"));
175 Dmsg1(100, "Update cat VolFiles=%d\n", dev->file);
176 /* Just labeled or relabeled the tape */
178 bstrncpy(vol->VolCatStatus, "Append", sizeof(vol->VolCatStatus));
179 vol->VolCatBytes = 1; /* indicates tape labeled */
181 bash_spaces(vol->VolCatName);
182 bnet_fsend(dir, Update_media, jcr->Job,
183 vol->VolCatName, vol->VolCatJobs, vol->VolCatFiles,
184 vol->VolCatBlocks, edit_uint64(vol->VolCatBytes, ed1),
185 vol->VolCatMounts, vol->VolCatErrors,
186 vol->VolCatWrites, edit_uint64(vol->VolCatMaxBytes, ed2),
187 EndTime, vol->VolCatStatus, vol->Slot, label, vol->Drive,
189 Dmsg1(120, "update_volume_info(): %s", dir->msg);
190 unbash_spaces(vol->VolCatName);
192 if (!do_get_volume_info(jcr)) {
193 Jmsg(jcr, M_ERROR, 0, "%s", jcr->errmsg);
196 Dmsg1(120, "get_volume_info(): %s", dir->msg);
197 /* Update dev Volume info in case something changed (e.g. expired) */
198 memcpy(&dev->VolCatInfo, &jcr->VolCatInfo, sizeof(dev->VolCatInfo));
203 * After writing a Volume, create the JobMedia record.
205 int dir_create_jobmedia_record(JCR *jcr)
207 BSOCK *dir = jcr->dir_bsock;
209 if (!jcr->WroteVol) {
210 return 1; /* nothing written to tape */
213 jcr->WroteVol = false;
214 bnet_fsend(dir, Create_job_media, jcr->Job,
215 jcr->VolFirstIndex, jcr->VolLastIndex,
216 jcr->StartFile, jcr->EndFile,
217 jcr->StartBlock, jcr->EndBlock);
218 Dmsg1(100, "create_jobmedia(): %s", dir->msg);
219 if (bnet_recv(dir) <= 0) {
220 Dmsg0(190, "create_jobmedia error bnet_recv\n");
221 Jmsg(jcr, M_ERROR, 0, _("Error creating JobMedia record: ERR=%s\n"),
225 Dmsg1(120, "Create_jobmedia: %s", dir->msg);
226 if (strcmp(dir->msg, OK_create) != 0) {
227 Dmsg1(130, "Bad response from Dir: %s\n", dir->msg);
228 Jmsg(jcr, M_ERROR, 0, _("Error creating JobMedia record: %s\n"), dir->msg);
236 * Update File Attribute data
238 int dir_update_file_attributes(JCR *jcr, DEV_RECORD *rec)
240 BSOCK *dir = jcr->dir_bsock;
243 dir->msglen = sprintf(dir->msg, FileAttributes, jcr->Job);
244 dir->msg = check_pool_memory_size(dir->msg, dir->msglen +
245 sizeof(DEV_RECORD) + rec->data_len);
246 ser_begin(dir->msg + dir->msglen, 0);
247 ser_uint32(rec->VolSessionId);
248 ser_uint32(rec->VolSessionTime);
249 ser_int32(rec->FileIndex);
250 ser_int32(rec->Stream);
251 ser_uint32(rec->data_len);
252 ser_bytes(rec->data, rec->data_len);
253 dir->msglen = ser_length(dir->msg);
254 return bnet_send(dir);
259 * Request to mount next Volume, which Volume not specified
261 * Entered with device blocked.
262 * Leaves with device blocked.
264 * Returns: 1 on success (operator issues a mount command)
266 * Note, must create dev->errmsg on error return.
268 * On success, jcr->VolumeName and jcr->VolCatInfo contain
269 * information on suggested volume, but this may not be the
270 * same as what is actually mounted.
272 * When we return with success, the correct tape may or may not
273 * actually be mounted. The calling routine must read it and
276 int dir_ask_sysop_to_mount_next_volume(JCR *jcr, DEVICE *dev)
279 /* ******FIXME******* put these on config variable */
280 int min_wait = 60 * 60;
281 int max_wait = 24 * 60 * 60;
282 int max_num_wait = 9; /* 5 waits =~ 1 day, then 1 day at a time */
287 Dmsg0(130, "enter dir_ask_sysop_to_mount_next_volume\n");
288 ASSERT(dev->dev_blocked);
291 if (job_canceled(jcr)) {
292 Mmsg(&dev->errmsg, _("Job %s canceled while waiting for mount on Storage Device \"%s\".\n"),
293 jcr->Job, jcr->dev_name);
294 Jmsg(jcr, M_INFO, 0, "%s", dev->errmsg);
297 if (dir_find_next_appendable_volume(jcr)) { /* get suggested volume */
298 jstat = JS_WaitMount;
300 * If we have a valid volume name and we are not
301 * removable media, return now, or if we have a
302 * Slot for an autochanger, otherwise wait
303 * for the operator to mount the media.
305 if ((jcr->VolumeName[0] && !dev_cap(dev, CAP_REM) && dev_cap(dev, CAP_LABEL)) ||
306 (jcr->VolumeName[0] && jcr->VolCatInfo.Slot)) {
307 Dmsg0(100, "Return 1 from mount without wait.\n");
310 Jmsg(jcr, M_MOUNT, 0, _(
311 "Please mount Volume \"%s\" on Storage Device \"%s\" for Job %s\n"
312 "Use \"mount\" command to release Job.\n"),
313 jcr->VolumeName, jcr->dev_name, jcr->Job);
314 Dmsg3(190, "Mount %s on %s for Job %s\n",
315 jcr->VolumeName, jcr->dev_name, jcr->Job);
317 jstat = JS_WaitMedia;
318 Jmsg(jcr, M_MOUNT, 0, _(
319 "Job %s waiting. Cannot find any appendable volumes.\n\
320 Please use the \"label\" command to create a new Volume for:\n\
330 jcr->JobStatus = jstat;
331 dir_send_job_status(jcr);
333 stat = wait_for_sysop(jcr, dev, wait_sec);
335 if (stat == ETIMEDOUT) {
336 wait_sec *= 2; /* double wait time */
337 if (wait_sec > max_wait) { /* but not longer than maxtime */
341 if (num_wait >= max_num_wait) {
342 Mmsg(&dev->errmsg, _("Gave up waiting to mount Storage Device \"%s\" for Job %s\n"),
343 jcr->dev_name, jcr->Job);
344 Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
345 Dmsg1(190, "Gave up waiting on device %s\n", dev_name(dev));
346 return 0; /* exceeded maximum waits */
350 if (stat == EINVAL) {
351 Mmsg2(&dev->errmsg, _("pthread error in mount_next_volume stat=%d ERR=%s\n"),
352 stat, strerror(stat));
353 Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
357 Jmsg(jcr, M_WARNING, 0, _("pthread error in mount_next_volume stat=%d ERR=%s\n"), stat,
360 Dmsg1(190, "Someone woke me for device %s\n", dev_name(dev));
362 /* Restart wait counters */
365 /* If no VolumeName, and cannot get one, try again */
366 if (jcr->VolumeName[0] == 0 && !job_canceled(jcr) &&
367 !dir_find_next_appendable_volume(jcr)) {
368 Jmsg(jcr, M_MOUNT, 0, _(
369 "Someone woke me up, but I cannot find any appendable\n\
370 volumes for Job=%s.\n"), jcr->Job);
375 set_jcr_job_status(jcr, JS_Running);
376 dir_send_job_status(jcr);
377 Dmsg0(130, "leave dir_ask_sysop_to_mount_next_volume\n");
382 * Request to mount specific Volume
384 * Entered with device blocked and jcr->VolumeName is desired
386 * Leaves with device blocked.
388 * Returns: 1 on success (operator issues a mount command)
390 * Note, must create dev->errmsg on error return.
393 int dir_ask_sysop_to_mount_volume(JCR *jcr, DEVICE *dev)
396 /* ******FIXME******* put these on config variable */
397 int min_wait = 60 * 60;
398 int max_wait = 24 * 60 * 60;
399 int max_num_wait = 9; /* 5 waits =~ 1 day, then 1 day at a time */
404 Dmsg0(130, "enter dir_ask_sysop_to_mount_next_volume\n");
405 if (!jcr->VolumeName[0]) {
406 Mmsg0(&dev->errmsg, _("Cannot request another volume: no volume name given.\n"));
409 ASSERT(dev->dev_blocked);
412 if (job_canceled(jcr)) {
413 Mmsg(&dev->errmsg, _("Job %s canceled while waiting for mount on Storage Device \"%s\".\n"),
414 jcr->Job, jcr->dev_name);
417 msg = _("Please mount");
418 Jmsg(jcr, M_MOUNT, 0, _("%s Volume \"%s\" on Storage Device \"%s\" for Job %s\n"),
419 msg, jcr->VolumeName, jcr->dev_name, jcr->Job);
420 Dmsg3(190, "Mount %s on %s for Job %s\n",
421 jcr->VolumeName, jcr->dev_name, jcr->Job);
423 jcr->JobStatus = JS_WaitMount;
424 dir_send_job_status(jcr);
426 stat = wait_for_sysop(jcr, dev, wait_sec); /* wait on device */
428 if (stat == ETIMEDOUT) {
429 wait_sec *= 2; /* double wait time */
430 if (wait_sec > max_wait) { /* but not longer than maxtime */
434 if (num_wait >= max_num_wait) {
435 Mmsg(&dev->errmsg, _("Gave up waiting to mount Storage Device \"%s\" for Job %s\n"),
436 jcr->dev_name, jcr->Job);
437 Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
438 Dmsg1(190, "Gave up waiting on device %s\n", dev_name(dev));
439 return 0; /* exceeded maximum waits */
443 if (stat == EINVAL) {
444 Mmsg2(&dev->errmsg, _("pthread error in mount_volume stat=%d ERR=%s\n"),
445 stat, strerror(stat));
446 Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
450 Jmsg(jcr, M_ERROR, 0, _("pthread error in mount_next_volume stat=%d ERR=%s\n"), stat,
453 Dmsg1(190, "Someone woke me for device %s\n", dev_name(dev));
455 /* Restart wait counters */
460 set_jcr_job_status(jcr, JS_Running);
461 dir_send_job_status(jcr);
462 Dmsg0(130, "leave dir_ask_sysop_to_mount_next_volume\n");
467 * Wait for SysOp to mount a tape
469 static int wait_for_sysop(JCR *jcr, DEVICE *dev, int wait_sec)
473 struct timespec timeout;
475 time_t start = time(NULL);
476 time_t last_heartbeat = 0;
481 * Wait requested time (wait_sec). However, we also wake up every
482 * HB_TIME seconds and send a heartbeat to the FD and the Director
483 * to keep stateful firewalls from closing them down while waiting
487 if (me->heartbeat_interval && add_wait > me->heartbeat_interval) {
488 add_wait = me->heartbeat_interval;
490 gettimeofday(&tv, &tz);
491 timeout.tv_nsec = tv.tv_usec * 1000;
492 timeout.tv_sec = tv.tv_sec + add_wait;
495 dev_blocked = dev->dev_blocked;
496 dev->dev_blocked = BST_WAITING_FOR_SYSOP; /* indicate waiting for mount */
498 for ( ; !job_canceled(jcr); ) {
501 Dmsg3(100, "I'm going to sleep on device %s. HB=%d wait=%d\n", dev_name(dev),
502 (int)me->heartbeat_interval, wait_sec);
503 stat = pthread_cond_timedwait(&dev->wait_next_vol, &dev->mutex, &timeout);
504 Dmsg1(100, "Wokeup from sleep on device stat=%d\n", stat);
508 /* Note, this always triggers the first time. We want that. */
509 if (me->heartbeat_interval) {
510 if (now - last_heartbeat >= me->heartbeat_interval) {
511 /* send heartbeats */
512 if (jcr->file_bsock) {
513 bnet_sig(jcr->file_bsock, BNET_HEARTBEAT);
514 Dmsg0(100, "Send heartbeat to FD.\n");
516 if (jcr->dir_bsock) {
517 bnet_sig(jcr->dir_bsock, BNET_HEARTBEAT);
519 last_heartbeat = now;
523 if (dev->dev_blocked == BST_MOUNT) { /* mount request ? */
528 if (stat != ETIMEDOUT) { /* we blocked the device */
529 break; /* on error return */
531 if (now - start >= wait_sec) { /* on exceeding wait time return */
532 Dmsg0(100, "Exceed wait time.\n");
535 add_wait = wait_sec - (now - start);
536 if (me->heartbeat_interval && add_wait > me->heartbeat_interval) {
537 add_wait = me->heartbeat_interval;
539 gettimeofday(&tv, &tz);
540 timeout.tv_nsec = tv.tv_usec * 1000;
541 timeout.tv_sec = tv.tv_sec + add_wait; /* additional wait */
542 Dmsg1(100, "Additional wait %d sec.\n", add_wait);
545 dev->dev_blocked = dev_blocked; /* restore entry state */