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 dev->VolCatInfo.VolCatFiles = dev->file; /* set number of files */
176 Dmsg1(100, "Update cat VolFiles=%d\n", dev->file);
177 /* Just labeled or relabeled the tape */
179 bstrncpy(vol->VolCatStatus, "Append", sizeof(vol->VolCatStatus));
180 vol->VolCatBytes = 1; /* indicates tape labeled */
182 bash_spaces(vol->VolCatName);
183 bnet_fsend(dir, Update_media, jcr->Job,
184 vol->VolCatName, vol->VolCatJobs, vol->VolCatFiles,
185 vol->VolCatBlocks, edit_uint64(vol->VolCatBytes, ed1),
186 vol->VolCatMounts, vol->VolCatErrors,
187 vol->VolCatWrites, edit_uint64(vol->VolCatMaxBytes, ed2),
188 EndTime, vol->VolCatStatus, vol->Slot, label, vol->Drive,
190 Dmsg1(120, "update_volume_info(): %s", dir->msg);
191 unbash_spaces(vol->VolCatName);
193 if (!do_get_volume_info(jcr)) {
194 Jmsg(jcr, M_ERROR, 0, "%s", jcr->errmsg);
197 Dmsg1(120, "get_volume_info(): %s", dir->msg);
198 /* Update dev Volume info in case something changed (e.g. expired) */
199 memcpy(&dev->VolCatInfo, &jcr->VolCatInfo, sizeof(dev->VolCatInfo));
204 * After writing a Volume, create the JobMedia record.
206 int dir_create_jobmedia_record(JCR *jcr)
208 BSOCK *dir = jcr->dir_bsock;
210 if (!jcr->WroteVol) {
211 return 1; /* nothing written to tape */
214 jcr->WroteVol = false;
215 bnet_fsend(dir, Create_job_media, jcr->Job,
216 jcr->VolFirstIndex, jcr->VolLastIndex,
217 jcr->StartFile, jcr->EndFile,
218 jcr->StartBlock, jcr->EndBlock);
219 Dmsg1(100, "create_jobmedia(): %s", dir->msg);
220 if (bnet_recv(dir) <= 0) {
221 Dmsg0(190, "create_jobmedia error bnet_recv\n");
222 Jmsg(jcr, M_ERROR, 0, _("Error creating JobMedia record: ERR=%s\n"),
226 Dmsg1(120, "Create_jobmedia: %s", dir->msg);
227 if (strcmp(dir->msg, OK_create) != 0) {
228 Dmsg1(130, "Bad response from Dir: %s\n", dir->msg);
229 Jmsg(jcr, M_ERROR, 0, _("Error creating JobMedia record: %s\n"), dir->msg);
237 * Update File Attribute data
239 int dir_update_file_attributes(JCR *jcr, DEV_RECORD *rec)
241 BSOCK *dir = jcr->dir_bsock;
244 dir->msglen = sprintf(dir->msg, FileAttributes, jcr->Job);
245 dir->msg = check_pool_memory_size(dir->msg, dir->msglen +
246 sizeof(DEV_RECORD) + rec->data_len);
247 ser_begin(dir->msg + dir->msglen, 0);
248 ser_uint32(rec->VolSessionId);
249 ser_uint32(rec->VolSessionTime);
250 ser_int32(rec->FileIndex);
251 ser_int32(rec->Stream);
252 ser_uint32(rec->data_len);
253 ser_bytes(rec->data, rec->data_len);
254 dir->msglen = ser_length(dir->msg);
255 return bnet_send(dir);
260 * Request to mount next Volume, which Volume not specified
262 * Entered with device blocked.
263 * Leaves with device blocked.
265 * Returns: 1 on success (operator issues a mount command)
267 * Note, must create dev->errmsg on error return.
269 * On success, jcr->VolumeName and jcr->VolCatInfo contain
270 * information on suggested volume, but this may not be the
271 * same as what is actually mounted.
273 * When we return with success, the correct tape may or may not
274 * actually be mounted. The calling routine must read it and
277 int dir_ask_sysop_to_mount_next_volume(JCR *jcr, DEVICE *dev)
280 /* ******FIXME******* put these on config variable */
281 int min_wait = 60 * 60;
282 int max_wait = 24 * 60 * 60;
283 int max_num_wait = 9; /* 5 waits =~ 1 day, then 1 day at a time */
288 Dmsg0(130, "enter dir_ask_sysop_to_mount_next_volume\n");
289 ASSERT(dev->dev_blocked);
292 if (job_canceled(jcr)) {
293 Mmsg(&dev->errmsg, _("Job %s canceled while waiting for mount on Storage Device \"%s\".\n"),
294 jcr->Job, jcr->dev_name);
295 Jmsg(jcr, M_INFO, 0, "%s", dev->errmsg);
298 if (dir_find_next_appendable_volume(jcr)) { /* get suggested volume */
299 jstat = JS_WaitMount;
301 * If we have a valid volume name and we are not
302 * removable media, return now, or if we have a
303 * Slot for an autochanger, otherwise wait
304 * for the operator to mount the media.
306 if ((jcr->VolumeName[0] && !dev_cap(dev, CAP_REM) && dev_cap(dev, CAP_LABEL)) ||
307 (jcr->VolumeName[0] && jcr->VolCatInfo.Slot)) {
308 Dmsg0(100, "Return 1 from mount without wait.\n");
311 Jmsg(jcr, M_MOUNT, 0, _(
312 "Please mount Volume \"%s\" on Storage Device \"%s\" for Job %s\n"
313 "Use \"mount\" command to release Job.\n"),
314 jcr->VolumeName, jcr->dev_name, jcr->Job);
315 Dmsg3(190, "Mount %s on %s for Job %s\n",
316 jcr->VolumeName, jcr->dev_name, jcr->Job);
318 jstat = JS_WaitMedia;
319 Jmsg(jcr, M_MOUNT, 0, _(
320 "Job %s waiting. Cannot find any appendable volumes.\n\
321 Please use the \"label\" command to create a new Volume for:\n\
331 jcr->JobStatus = jstat;
332 dir_send_job_status(jcr);
334 stat = wait_for_sysop(jcr, dev, wait_sec);
336 if (stat == ETIMEDOUT) {
337 wait_sec *= 2; /* double wait time */
338 if (wait_sec > max_wait) { /* but not longer than maxtime */
342 if (num_wait >= max_num_wait) {
343 Mmsg(&dev->errmsg, _("Gave up waiting to mount Storage Device \"%s\" for Job %s\n"),
344 jcr->dev_name, jcr->Job);
345 Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
346 Dmsg1(190, "Gave up waiting on device %s\n", dev_name(dev));
347 return 0; /* exceeded maximum waits */
351 if (stat == EINVAL) {
352 Mmsg2(&dev->errmsg, _("pthread error in mount_next_volume stat=%d ERR=%s\n"),
353 stat, strerror(stat));
354 Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
358 Jmsg(jcr, M_WARNING, 0, _("pthread error in mount_next_volume stat=%d ERR=%s\n"), stat,
361 Dmsg1(190, "Someone woke me for device %s\n", dev_name(dev));
363 /* Restart wait counters */
366 /* If no VolumeName, and cannot get one, try again */
367 if (jcr->VolumeName[0] == 0 && !job_canceled(jcr) &&
368 !dir_find_next_appendable_volume(jcr)) {
369 Jmsg(jcr, M_MOUNT, 0, _(
370 "Someone woke me up, but I cannot find any appendable\n\
371 volumes for Job=%s.\n"), jcr->Job);
376 set_jcr_job_status(jcr, JS_Running);
377 dir_send_job_status(jcr);
378 Dmsg0(130, "leave dir_ask_sysop_to_mount_next_volume\n");
383 * Request to mount specific Volume
385 * Entered with device blocked and jcr->VolumeName is desired
387 * Leaves with device blocked.
389 * Returns: 1 on success (operator issues a mount command)
391 * Note, must create dev->errmsg on error return.
394 int dir_ask_sysop_to_mount_volume(JCR *jcr, DEVICE *dev)
397 /* ******FIXME******* put these on config variable */
398 int min_wait = 60 * 60;
399 int max_wait = 24 * 60 * 60;
400 int max_num_wait = 9; /* 5 waits =~ 1 day, then 1 day at a time */
405 Dmsg0(130, "enter dir_ask_sysop_to_mount_next_volume\n");
406 if (!jcr->VolumeName[0]) {
407 Mmsg0(&dev->errmsg, _("Cannot request another volume: no volume name given.\n"));
410 ASSERT(dev->dev_blocked);
413 if (job_canceled(jcr)) {
414 Mmsg(&dev->errmsg, _("Job %s canceled while waiting for mount on Storage Device \"%s\".\n"),
415 jcr->Job, jcr->dev_name);
418 msg = _("Please mount");
419 Jmsg(jcr, M_MOUNT, 0, _("%s Volume \"%s\" on Storage Device \"%s\" for Job %s\n"),
420 msg, jcr->VolumeName, jcr->dev_name, jcr->Job);
421 Dmsg3(190, "Mount %s on %s for Job %s\n",
422 jcr->VolumeName, jcr->dev_name, jcr->Job);
424 jcr->JobStatus = JS_WaitMount;
425 dir_send_job_status(jcr);
427 stat = wait_for_sysop(jcr, dev, wait_sec); /* wait on device */
429 if (stat == ETIMEDOUT) {
430 wait_sec *= 2; /* double wait time */
431 if (wait_sec > max_wait) { /* but not longer than maxtime */
435 if (num_wait >= max_num_wait) {
436 Mmsg(&dev->errmsg, _("Gave up waiting to mount Storage Device \"%s\" for Job %s\n"),
437 jcr->dev_name, jcr->Job);
438 Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
439 Dmsg1(190, "Gave up waiting on device %s\n", dev_name(dev));
440 return 0; /* exceeded maximum waits */
444 if (stat == EINVAL) {
445 Mmsg2(&dev->errmsg, _("pthread error in mount_volume stat=%d ERR=%s\n"),
446 stat, strerror(stat));
447 Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
451 Jmsg(jcr, M_ERROR, 0, _("pthread error in mount_next_volume stat=%d ERR=%s\n"), stat,
454 Dmsg1(190, "Someone woke me for device %s\n", dev_name(dev));
456 /* Restart wait counters */
461 set_jcr_job_status(jcr, JS_Running);
462 dir_send_job_status(jcr);
463 Dmsg0(130, "leave dir_ask_sysop_to_mount_next_volume\n");
468 * Wait for SysOp to mount a tape
470 static int wait_for_sysop(JCR *jcr, DEVICE *dev, int wait_sec)
474 struct timespec timeout;
476 time_t start = time(NULL);
477 time_t last_heartbeat = 0;
482 * Wait requested time (wait_sec). However, we also wake up every
483 * HB_TIME seconds and send a heartbeat to the FD and the Director
484 * to keep stateful firewalls from closing them down while waiting
488 if (me->heartbeat_interval && add_wait > me->heartbeat_interval) {
489 add_wait = me->heartbeat_interval;
491 gettimeofday(&tv, &tz);
492 timeout.tv_nsec = tv.tv_usec * 1000;
493 timeout.tv_sec = tv.tv_sec + add_wait;
496 dev_blocked = dev->dev_blocked;
497 dev->dev_blocked = BST_WAITING_FOR_SYSOP; /* indicate waiting for mount */
499 for ( ; !job_canceled(jcr); ) {
502 Dmsg3(100, "I'm going to sleep on device %s. HB=%d wait=%d\n", dev_name(dev),
503 (int)me->heartbeat_interval, wait_sec);
504 stat = pthread_cond_timedwait(&dev->wait_next_vol, &dev->mutex, &timeout);
505 Dmsg1(100, "Wokeup from sleep on device stat=%d\n", stat);
509 /* Note, this always triggers the first time. We want that. */
510 if (me->heartbeat_interval) {
511 if (now - last_heartbeat >= me->heartbeat_interval) {
512 /* send heartbeats */
513 if (jcr->file_bsock) {
514 bnet_sig(jcr->file_bsock, BNET_HEARTBEAT);
515 Dmsg0(100, "Send heartbeat to FD.\n");
517 if (jcr->dir_bsock) {
518 bnet_sig(jcr->dir_bsock, BNET_HEARTBEAT);
520 last_heartbeat = now;
524 if (dev->dev_blocked == BST_MOUNT) { /* mount request ? */
529 if (stat != ETIMEDOUT) { /* we blocked the device */
530 break; /* on error return */
532 if (now - start >= wait_sec) { /* on exceeding wait time return */
533 Dmsg0(100, "Exceed wait time.\n");
536 add_wait = wait_sec - (now - start);
537 if (me->heartbeat_interval && add_wait > me->heartbeat_interval) {
538 add_wait = me->heartbeat_interval;
540 gettimeofday(&tv, &tz);
541 timeout.tv_nsec = tv.tv_usec * 1000;
542 timeout.tv_sec = tv.tv_sec + add_wait; /* additional wait */
543 Dmsg1(100, "Additional wait %d sec.\n", add_wait);
546 dev->dev_blocked = dev_blocked; /* restore entry state */