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\n";
40 static char Create_job_media[] = "CatReq Job=%s CreateJobMedia \
41 FirstIndex=%u LastIndex=%u StartFile=%u EndFile=%u \
42 StartBlock=%u EndBlock=%u\n";
43 static char FileAttributes[] = "UpdCat Job=%s FileAttributes ";
44 static char Job_status[] = "3012 Job %s jobstatus %d\n";
47 /* Responses received from the Director */
48 static char OK_media[] = "1000 OK VolName=%127s VolJobs=%u VolFiles=%u\
49 VolBlocks=%u VolBytes=%" lld " VolMounts=%u VolErrors=%u VolWrites=%u\
50 MaxVolBytes=%" lld " VolCapacityBytes=%" lld " VolStatus=%20s\
51 Slot=%d MaxVolJobs=%u MaxVolFiles=%u\n";
52 static char OK_update[] = "1000 OK UpdateMedia\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 static int do_request_volume_info(JCR *jcr)
73 BSOCK *dir = jcr->dir_bsock;
76 jcr->VolumeName[0] = 0; /* No volume */
77 if (bnet_recv(dir) <= 0) {
78 Dmsg0(200, "getvolname error bnet_recv\n");
79 Mmsg(&jcr->errmsg, _("Network error on bnet_recv in req_vol_info.\n"));
82 if (sscanf(dir->msg, OK_media, vol.VolCatName,
83 &vol.VolCatJobs, &vol.VolCatFiles,
84 &vol.VolCatBlocks, &vol.VolCatBytes,
85 &vol.VolCatMounts, &vol.VolCatErrors,
86 &vol.VolCatWrites, &vol.VolCatMaxBytes,
87 &vol.VolCatCapacityBytes, vol.VolCatStatus,
88 &vol.Slot, &vol.VolCatMaxJobs, &vol.VolCatMaxFiles) != 14) {
90 Dmsg1(200, "Bad response from Dir: %s\n", dir->msg);
91 Mmsg(&jcr->errmsg, _("Error scanning Dir response: %s\n"), dir->msg);
94 unbash_spaces(vol.VolCatName);
95 pm_strcpy(&jcr->VolumeName, vol.VolCatName); /* set desired VolumeName */
96 memcpy(&jcr->VolCatInfo, &vol, sizeof(jcr->VolCatInfo));
98 Dmsg2(200, "do_reqest_vol_info got slot=%d Volume=%s\n",
99 vol.Slot, vol.VolCatName);
105 * Get Volume info for a specific volume from the Director's Database
107 * Returns: 1 on success (not Director guarantees that Pool and MediaType
108 * are correct and VolStatus==Append or
109 * VolStatus==Recycle)
112 * Volume information returned in jcr
114 int dir_get_volume_info(JCR *jcr, enum get_vol_info_rw writing)
116 BSOCK *dir = jcr->dir_bsock;
118 bstrncpy(jcr->VolCatInfo.VolCatName, jcr->VolumeName, sizeof(jcr->VolCatInfo.VolCatName));
119 Dmsg1(200, "dir_get_volume_info=%s\n", jcr->VolCatInfo.VolCatName);
120 bash_spaces(jcr->VolCatInfo.VolCatName);
121 bnet_fsend(dir, Get_Vol_Info, jcr->Job, jcr->VolCatInfo.VolCatName,
122 writing==GET_VOL_INFO_FOR_WRITE?1:0);
123 return do_request_volume_info(jcr);
129 * Get info on the next appendable volume in the Director's database
130 * Returns: 1 on success
133 * Volume information returned in jcr
136 int dir_find_next_appendable_volume(JCR *jcr)
138 BSOCK *dir = jcr->dir_bsock;
140 Dmsg0(200, "dir_find_next_appendable_volume\n");
141 bnet_fsend(dir, Find_media, jcr->Job, 1);
142 return do_request_volume_info(jcr);
147 * After writing a Volume, send the updated statistics
148 * back to the director.
150 int dir_update_volume_info(JCR *jcr, VOLUME_CAT_INFO *vol, int relabel)
152 BSOCK *dir = jcr->dir_bsock;
153 time_t EndTime = time(NULL);
154 char ed1[50], ed2[50];
156 if (vol->VolCatName[0] == 0) {
157 Jmsg0(jcr, M_ERROR, 0, _("NULL Volume name. This shouldn't happen!!!\n"));
160 bash_spaces(vol->VolCatName);
161 bnet_fsend(dir, Update_media, jcr->Job,
162 vol->VolCatName, vol->VolCatJobs, vol->VolCatFiles,
163 vol->VolCatBlocks, edit_uint64(vol->VolCatBytes, ed1),
164 vol->VolCatMounts, vol->VolCatErrors,
165 vol->VolCatWrites, edit_uint64(vol->VolCatMaxBytes, ed2),
166 EndTime, vol->VolCatStatus, vol->Slot, relabel);
167 Dmsg1(120, "update_volume_data(): %s", dir->msg);
168 unbash_spaces(vol->VolCatName);
169 if (bnet_recv(dir) <= 0) {
170 Dmsg0(190, "updateVolCatInfo error bnet_recv\n");
171 Jmsg(jcr, M_ERROR, 0, _("Error updating Volume info Vol=\"%s\": ERR=%s\n"),
172 vol->VolCatName, bnet_strerror(dir));
175 Dmsg1(120, "Updatevol: %s", dir->msg);
176 if (strcmp(dir->msg, OK_update) != 0) {
177 Dmsg1(130, "Bad response from Dir: %s\n", dir->msg);
178 Jmsg(jcr, M_ERROR, 0, _("Error updating Volume info Vol=\"%s\": %s\n"),
179 vol->VolCatName, dir->msg);
186 * After writing a Volume, create the JobMedia record.
188 int dir_create_jobmedia_record(JCR *jcr)
190 BSOCK *dir = jcr->dir_bsock;
192 if (!jcr->WroteVol) {
193 return 1; /* nothing written to tape */
196 jcr->WroteVol = false;
197 bnet_fsend(dir, Create_job_media, jcr->Job,
198 jcr->VolFirstIndex, jcr->VolLastIndex,
199 jcr->StartFile, jcr->EndFile,
200 jcr->StartBlock, jcr->EndBlock);
201 Dmsg1(100, "create_jobmedia(): %s", dir->msg);
202 if (bnet_recv(dir) <= 0) {
203 Dmsg0(190, "create_jobmedia error bnet_recv\n");
204 Jmsg(jcr, M_ERROR, 0, _("Error creating JobMedia record: ERR=%s\n"),
208 Dmsg1(120, "Create_jobmedia: %s", dir->msg);
209 if (strcmp(dir->msg, OK_update) != 0) {
210 Dmsg1(130, "Bad response from Dir: %s\n", dir->msg);
211 Jmsg(jcr, M_ERROR, 0, _("Error creating JobMedia record: %s\n"), dir->msg);
219 * Update File Attribute data
221 int dir_update_file_attributes(JCR *jcr, DEV_RECORD *rec)
223 BSOCK *dir = jcr->dir_bsock;
226 dir->msglen = sprintf(dir->msg, FileAttributes, jcr->Job);
227 dir->msg = check_pool_memory_size(dir->msg, dir->msglen +
228 sizeof(DEV_RECORD) + rec->data_len);
229 ser_begin(dir->msg + dir->msglen, 0);
230 ser_uint32(rec->VolSessionId);
231 ser_uint32(rec->VolSessionTime);
232 ser_int32(rec->FileIndex);
233 ser_int32(rec->Stream);
234 ser_uint32(rec->data_len);
235 ser_bytes(rec->data, rec->data_len);
236 dir->msglen = ser_length(dir->msg);
237 return bnet_send(dir);
243 * Entered with device blocked.
244 * Leaves with device blocked.
246 * Returns: 1 on success (operator issues a mount command)
248 * Note, must create dev->errmsg on error return.
250 * On success, jcr->VolumeName and jcr->VolCatInfo contain
251 * information on suggested volume, but this may not be the
252 * same as what is actually mounted.
254 * When we return with success, the correct tape may or may not
255 * actually be mounted. The calling routine must read it and
258 int dir_ask_sysop_to_mount_next_volume(JCR *jcr, DEVICE *dev)
261 /* ******FIXME******* put these on config variable */
262 int min_wait = 60 * 60;
263 int max_wait = 24 * 60 * 60;
264 int max_num_wait = 9; /* 5 waits =~ 1 day, then 1 day at a time */
269 Dmsg0(130, "enter dir_ask_sysop_to_mount_next_volume\n");
270 ASSERT(dev->dev_blocked);
273 if (job_canceled(jcr)) {
274 Mmsg(&dev->errmsg, _("Job %s canceled while waiting for mount on Storage Device \"%s\".\n"),
275 jcr->Job, jcr->dev_name);
276 Jmsg(jcr, M_INFO, 0, "%s", dev->errmsg);
279 if (dir_find_next_appendable_volume(jcr)) { /* get suggested volume */
280 jstat = JS_WaitMount;
282 * If we have a valid volume name and we are not
283 * removable media, return now, otherwise wait
284 * for the operator to mount the media.
286 if (jcr->VolumeName[0] && !dev_cap(dev, CAP_REM) && dev_cap(dev, CAP_LABEL)) {
287 Dmsg0(190, "Return 1 from mount without wait.\n");
290 Jmsg(jcr, M_MOUNT, 0, _(
291 "Please mount Volume \"%s\" on Storage Device \"%s\" for Job %s\n"
292 "Use \"mount\" command to release Job.\n"),
293 jcr->VolumeName, jcr->dev_name, jcr->Job);
294 Dmsg3(190, "Mount %s on %s for Job %s\n",
295 jcr->VolumeName, jcr->dev_name, jcr->Job);
297 jstat = JS_WaitMedia;
298 Jmsg(jcr, M_MOUNT, 0, _(
299 "Job %s waiting. Cannot find any appendable volumes.\n\
300 Please use the \"label\" command to create a new Volume for:\n\
310 jcr->JobStatus = jstat;
311 dir_send_job_status(jcr);
313 stat = wait_for_sysop(jcr, dev, wait_sec);
315 if (stat == ETIMEDOUT) {
316 wait_sec *= 2; /* double wait time */
317 if (wait_sec > max_wait) { /* but not longer than maxtime */
321 if (num_wait >= max_num_wait) {
322 Mmsg(&dev->errmsg, _("Gave up waiting to mount Storage Device \"%s\" for Job %s\n"),
323 jcr->dev_name, jcr->Job);
324 Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
325 Dmsg1(190, "Gave up waiting on device %s\n", dev_name(dev));
326 return 0; /* exceeded maximum waits */
330 if (stat == EINVAL) {
331 Mmsg2(&dev->errmsg, _("pthread error in mount_next_volume stat=%d ERR=%s\n"),
332 stat, strerror(stat));
333 Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
337 Jmsg(jcr, M_WARNING, 0, _("pthread error in mount_next_volume stat=%d ERR=%s\n"), stat,
340 Dmsg1(190, "Someone woke me for device %s\n", dev_name(dev));
342 /* Restart wait counters */
345 /* If no VolumeName, and cannot get one, try again */
346 if (jcr->VolumeName[0] == 0 &&
347 !dir_find_next_appendable_volume(jcr) && !job_canceled(jcr)) {
348 Jmsg(jcr, M_MOUNT, 0, _(
349 "Someone woke me up, but I cannot find any appendable\n\
350 volumes for Job=%s.\n"), jcr->Job);
355 set_jcr_job_status(jcr, JS_Running);
356 dir_send_job_status(jcr);
357 Dmsg0(130, "leave dir_ask_sysop_to_mount_next_volume\n");
363 * Entered with device blocked and jcr->VolumeName is desired
365 * Leaves with device blocked.
367 * Returns: 1 on success (operator issues a mount command)
369 * Note, must create dev->errmsg on error return.
372 int dir_ask_sysop_to_mount_volume(JCR *jcr, DEVICE *dev)
375 /* ******FIXME******* put these on config variable */
376 int min_wait = 60 * 60;
377 int max_wait = 24 * 60 * 60;
378 int max_num_wait = 9; /* 5 waits =~ 1 day, then 1 day at a time */
383 Dmsg0(130, "enter dir_ask_sysop_to_mount_next_volume\n");
384 if (!jcr->VolumeName[0]) {
385 Mmsg0(&dev->errmsg, _("Cannot request another volume: no volume name given.\n"));
388 ASSERT(dev->dev_blocked);
391 if (job_canceled(jcr)) {
392 Mmsg(&dev->errmsg, _("Job %s canceled while waiting for mount on Storage Device \"%s\".\n"),
393 jcr->Job, jcr->dev_name);
396 msg = _("Please mount");
397 Jmsg(jcr, M_MOUNT, 0, _("%s Volume \"%s\" on Storage Device \"%s\" for Job %s\n"),
398 msg, jcr->VolumeName, jcr->dev_name, jcr->Job);
399 Dmsg3(190, "Mount %s on %s for Job %s\n",
400 jcr->VolumeName, jcr->dev_name, jcr->Job);
402 jcr->JobStatus = JS_WaitMount;
403 dir_send_job_status(jcr);
405 stat = wait_for_sysop(jcr, dev, wait_sec); /* wait on device */
407 if (stat == ETIMEDOUT) {
408 wait_sec *= 2; /* double wait time */
409 if (wait_sec > max_wait) { /* but not longer than maxtime */
413 if (num_wait >= max_num_wait) {
414 Mmsg(&dev->errmsg, _("Gave up waiting to mount Storage Device \"%s\" for Job %s\n"),
415 jcr->dev_name, jcr->Job);
416 Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
417 Dmsg1(190, "Gave up waiting on device %s\n", dev_name(dev));
418 return 0; /* exceeded maximum waits */
422 if (stat == EINVAL) {
423 Mmsg2(&dev->errmsg, _("pthread error in mount_volume stat=%d ERR=%s\n"),
424 stat, strerror(stat));
425 Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
429 Jmsg(jcr, M_ERROR, 0, _("pthread error in mount_next_volume stat=%d ERR=%s\n"), stat,
432 Dmsg1(190, "Someone woke me for device %s\n", dev_name(dev));
434 /* Restart wait counters */
439 set_jcr_job_status(jcr, JS_Running);
440 dir_send_job_status(jcr);
441 Dmsg0(130, "leave dir_ask_sysop_to_mount_next_volume\n");
446 * Wait for SysOp to mount a tape
448 static int wait_for_sysop(JCR *jcr, DEVICE *dev, int wait_sec)
452 struct timespec timeout;
454 time_t start = time(NULL);
455 time_t last_heartbeat = 0;
460 * Wait requested time (wait_sec). However, we also wake up every
461 * HB_TIME seconds and send a heartbeat to the FD and the Director
462 * to keep stateful firewalls from closing them down while waiting
466 if (me->heartbeat_interval && add_wait > me->heartbeat_interval) {
467 add_wait = me->heartbeat_interval;
469 gettimeofday(&tv, &tz);
470 timeout.tv_nsec = tv.tv_usec * 1000;
471 timeout.tv_sec = tv.tv_sec + add_wait;
474 dev_blocked = dev->dev_blocked;
475 dev->dev_blocked = BST_WAITING_FOR_SYSOP; /* indicate waiting for mount */
477 for ( ; !job_canceled(jcr); ) {
480 Dmsg3(100, "I'm going to sleep on device %s. HB=%d wait=%d\n", dev_name(dev),
481 (int)me->heartbeat_interval, wait_sec);
482 stat = pthread_cond_timedwait(&dev->wait_next_vol, &dev->mutex, &timeout);
483 Dmsg1(100, "Wokeup from sleep on device stat=%d\n", stat);
487 /* Note, this always triggers the first time. We want that. */
488 if (me->heartbeat_interval) {
489 if (now - last_heartbeat >= me->heartbeat_interval) {
490 /* send heartbeats */
491 if (jcr->file_bsock) {
492 bnet_sig(jcr->file_bsock, BNET_HEARTBEAT);
493 Dmsg0(100, "Send heartbeat to FD.\n");
495 if (jcr->dir_bsock) {
496 bnet_sig(jcr->dir_bsock, BNET_HEARTBEAT);
498 last_heartbeat = now;
502 if (dev->dev_blocked == BST_MOUNT) { /* mount request ? */
507 if (stat != ETIMEDOUT) { /* we blocked the device */
508 break; /* on error return */
510 if (now - start >= wait_sec) { /* on exceeding wait time return */
511 Dmsg0(100, "Exceed wait time.\n");
514 add_wait = wait_sec - (now - start);
515 if (me->heartbeat_interval && add_wait > me->heartbeat_interval) {
516 add_wait = me->heartbeat_interval;
518 gettimeofday(&tv, &tz);
519 timeout.tv_nsec = tv.tv_usec * 1000;
520 timeout.tv_sec = tv.tv_sec + add_wait; /* additional wait */
521 Dmsg1(100, "Additional wait %d sec.\n", add_wait);
524 dev->dev_blocked = dev_blocked;