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, 2001, 2002 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";
36 static char Update_media[] = "CatReq Job=%s UpdateMedia VolName=%s\
37 VolJobs=%u VolFiles=%u VolBlocks=%u VolBytes=%s VolMounts=%u\
38 VolErrors=%u VolWrites=%u MaxVolBytes=%s EndTime=%d VolStatus=%s\
39 Slot=%d relabel=%d\n";
41 static char Create_job_media[] = "CatReq Job=%s CreateJobMedia \
42 FirstIndex=%u LastIndex=%u StartFile=%u EndFile=%u \
43 StartBlock=%u EndBlock=%u\n";
46 static char FileAttributes[] = "UpdCat Job=%s FileAttributes ";
48 static char Job_status[] = "3012 Job %s jobstatus %d\n";
51 /* Responses received from the Director */
52 static char OK_media[] = "1000 OK VolName=%127s VolJobs=%u VolFiles=%u\
53 VolBlocks=%u VolBytes=%" lld " VolMounts=%u VolErrors=%u VolWrites=%u\
54 MaxVolBytes=%" lld " VolCapacityBytes=%" lld " VolStatus=%20s\
55 Slot=%d MaxVolJobs=%u MaxVolFiles=%u\n";
57 static char OK_update[] = "1000 OK UpdateMedia\n";
61 * Send current JobStatus to Director
63 int dir_send_job_status(JCR *jcr)
65 return bnet_fsend(jcr->dir_bsock, Job_status, jcr->Job, jcr->JobStatus);
70 * dir_get_volume_info()
72 * dir_find_next_appendable_volume()
74 static int do_request_volume_info(JCR *jcr)
76 BSOCK *dir = jcr->dir_bsock;
77 VOLUME_CAT_INFO *vol = &jcr->VolCatInfo;
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 if (sscanf(dir->msg, OK_media, vol->VolCatName,
86 &vol->VolCatJobs, &vol->VolCatFiles,
87 &vol->VolCatBlocks, &vol->VolCatBytes,
88 &vol->VolCatMounts, &vol->VolCatErrors,
89 &vol->VolCatWrites, &vol->VolCatMaxBytes,
90 &vol->VolCatCapacityBytes, vol->VolCatStatus,
91 &vol->Slot, &vol->VolCatMaxJobs, &vol->VolCatMaxFiles) != 14) {
93 Dmsg1(200, "Bad response from Dir: %s\n", dir->msg);
94 Mmsg(&jcr->errmsg, _("Error scanning Dir response: %s\n"), dir->msg);
97 unbash_spaces(vol->VolCatName);
98 strcpy(jcr->VolumeName, vol->VolCatName); /* set desired VolumeName */
100 Dmsg2(200, "do_reqest_vol_info got slot=%d Volume=%s\n",
101 vol->Slot, vol->VolCatName);
107 * Get Volume info for a specific volume from the Director's Database
109 * Returns: 1 on success (not Director guarantees that Pool and MediaType
110 * are correct and VolStatus==Append or
111 * VolStatus==Recycle)
114 * Volume information returned in jcr
116 int dir_get_volume_info(JCR *jcr, int writing)
118 BSOCK *dir = jcr->dir_bsock;
120 strcpy(jcr->VolCatInfo.VolCatName, jcr->VolumeName);
121 Dmsg1(200, "dir_get_volume_info=%s\n", jcr->VolCatInfo.VolCatName);
122 bash_spaces(jcr->VolCatInfo.VolCatName);
123 bnet_fsend(dir, Get_Vol_Info, jcr->Job, jcr->VolCatInfo.VolCatName, writing);
124 return do_request_volume_info(jcr);
130 * Get info on the next appendable volume in the Director's database
131 * Returns: 1 on success
134 * Volume information returned in jcr
137 int dir_find_next_appendable_volume(JCR *jcr)
139 BSOCK *dir = jcr->dir_bsock;
141 Dmsg0(200, "dir_find_next_appendable_volume\n");
142 bnet_fsend(dir, Find_media, jcr->Job, 1);
143 return do_request_volume_info(jcr);
148 * After writing a Volume, send the updated statistics
149 * back to the director.
151 int dir_update_volume_info(JCR *jcr, VOLUME_CAT_INFO *vol, int relabel)
153 BSOCK *dir = jcr->dir_bsock;
154 time_t EndTime = time(NULL);
155 char ed1[50], ed2[50];
157 if (vol->VolCatName[0] == 0) {
158 Jmsg0(jcr, M_ERROR, 0, _("NULL Volume name. This shouldn't happen!!!\n"));
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 if (bnet_recv(dir) <= 0) {
169 Dmsg0(190, "updateVolCatInfo error bnet_recv\n");
170 Jmsg(jcr, M_ERROR, 0, _("Error updating Volume Info: %s\n"),
174 Dmsg1(120, "Updatevol: %s", dir->msg);
175 if (strcmp(dir->msg, OK_update) != 0) {
176 Dmsg1(130, "Bad response from Dir: %s\n", dir->msg);
177 Jmsg(jcr, M_ERROR, 0, _("Error updating Volume Info: %s\n"), dir->msg);
184 * After writing a Volume, create the JobMedia record.
186 int dir_create_jobmedia_record(JCR *jcr)
188 BSOCK *dir = jcr->dir_bsock;
190 bnet_fsend(dir, Create_job_media, jcr->Job,
191 jcr->VolFirstFile, jcr->JobFiles,
192 jcr->StartFile, jcr->EndFile,
193 jcr->StartBlock, jcr->EndBlock);
194 Dmsg1(100, "create_jobmedia(): %s", dir->msg);
195 if (bnet_recv(dir) <= 0) {
196 Dmsg0(190, "create_jobmedia error bnet_recv\n");
197 Jmsg(jcr, M_ERROR, 0, _("Error creating JobMedia record: %s\n"),
201 Dmsg1(120, "Create_jobmedia: %s", dir->msg);
202 if (strcmp(dir->msg, OK_update) != 0) {
203 Dmsg1(130, "Bad response from Dir: %s\n", dir->msg);
204 Jmsg(jcr, M_ERROR, 0, _("Error creating JobMedia record: %s\n"), dir->msg);
212 * Update File Attribute data
214 int dir_update_file_attributes(JCR *jcr, DEV_RECORD *rec)
216 BSOCK *dir = jcr->dir_bsock;
219 dir->msglen = sprintf(dir->msg, FileAttributes, jcr->Job);
220 dir->msg = check_pool_memory_size(dir->msg, dir->msglen +
221 sizeof(DEV_RECORD) + rec->data_len);
222 ser_begin(dir->msg + dir->msglen, 0);
223 ser_uint32(rec->VolSessionId);
224 ser_uint32(rec->VolSessionTime);
225 ser_int32(rec->FileIndex);
226 ser_int32(rec->Stream);
227 ser_uint32(rec->data_len);
228 ser_bytes(rec->data, rec->data_len);
229 dir->msglen = ser_length(dir->msg);
230 return bnet_send(dir);
236 * Entered with device blocked.
237 * Leaves with device blocked.
239 * Returns: 1 on success (operator issues a mount command)
241 * Note, must create dev->errmsg on error return.
243 * On success, jcr->VolumeName and jcr->VolCatInfo contain
244 * information on suggested volume, but this may not be the
245 * same as what is actually mounted.
247 * When we return with success, the correct tape may or may not
248 * actually be mounted. The calling routine must read it and
251 int dir_ask_sysop_to_mount_next_volume(JCR *jcr, DEVICE *dev)
255 struct timespec timeout;
257 /* ******FIXME******* put these on config variable */
258 int min_wait = 60 * 60;
259 int max_wait = 24 * 60 * 60;
260 int max_num_wait = 9; /* 5 waits =~ 1 day, then 1 day at a time */
266 Dmsg0(130, "enter dir_ask_sysop_to_mount_next_volume\n");
267 ASSERT(dev->dev_blocked);
270 if (job_cancelled(jcr)) {
271 Mmsg(&dev->errmsg, _("Job %s cancelled while waiting for mount on Storage Device \"%s\".\n"),
272 jcr->Job, jcr->dev_name);
273 Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
276 if (dir_find_next_appendable_volume(jcr)) { /* get suggested volume */
277 jstat = JS_WaitMount;
279 * If we have a valid volume name and we are not
280 * removable media, return now, otherwise wait
281 * for the operator to mount the media.
283 if (jcr->VolumeName[0] && !(dev->capabilities & CAP_REM) &&
284 dev->capabilities & CAP_LABEL) {
285 Dmsg0(190, "Return 1 from mount without wait.\n");
288 Jmsg(jcr, M_MOUNT, 0, _(
289 "Please mount Volume \"%s\" on Storage Device \"%s\" for Job %s\n"
290 "Use \"mount\" command to release Job.\n"),
291 jcr->VolumeName, jcr->dev_name, jcr->Job);
292 Dmsg3(190, "Mount %s on %s for Job %s\n",
293 jcr->VolumeName, jcr->dev_name, jcr->Job);
295 jstat = JS_WaitMedia;
296 Jmsg(jcr, M_MOUNT, 0, _(
297 "Job %s waiting. Cannot find any appendable volumes.\n\
298 Please use the \"label\" command to create a new Volume for:\n\
308 * Wait then send message again
310 gettimeofday(&tv, &tz);
311 timeout.tv_nsec = tv.tv_usec * 1000;
312 timeout.tv_sec = tv.tv_sec + wait_sec;
315 dev_blocked = dev->dev_blocked;
316 dev->dev_blocked = BST_WAITING_FOR_SYSOP; /* indicate waiting for mount */
317 jcr->JobStatus = jstat;
318 dir_send_job_status(jcr);
320 for ( ;!job_cancelled(jcr); ) {
321 Dmsg1(190, "I'm going to sleep on device %s\n", dev->dev_name);
322 stat = pthread_cond_timedwait(&dev->wait_next_vol, &dev->mutex, &timeout);
323 if (dev->dev_blocked == BST_WAITING_FOR_SYSOP) {
327 * Someone other than us blocked the device (probably the
328 * user via the Console program.
329 * So, we continue waiting.
331 gettimeofday(&tv, &tz);
333 timeout.tv_sec = tv.tv_sec + 10; /* wait 10 seconds */
335 dev->dev_blocked = dev_blocked;
338 if (stat == ETIMEDOUT) {
339 wait_sec *= 2; /* double wait time */
340 if (wait_sec > max_wait) { /* but not longer than maxtime */
344 if (num_wait >= max_num_wait) {
345 Mmsg(&dev->errmsg, _("Gave up waiting to mount Storage Device \"%s\" for Job %s\n"),
346 jcr->dev_name, jcr->Job);
347 Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
348 Dmsg1(190, "Gave up waiting on device %s\n", dev->dev_name);
349 return 0; /* exceeded maximum waits */
353 if (stat == EINVAL) {
354 Mmsg2(&dev->errmsg, _("pthread error in mount_next_volume stat=%d ERR=%s\n"),
355 stat, strerror(stat));
356 Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
360 Jmsg(jcr, M_WARNING, 0, _("pthread error in mount_next_volume stat=%d ERR=%s\n"), stat,
363 Dmsg1(190, "Someone woke me for device %s\n", dev->dev_name);
365 /* Restart wait counters */
368 /* If no VolumeName, and cannot get one, try again */
369 if (jcr->VolumeName[0] == 0 &&
370 !dir_find_next_appendable_volume(jcr)) {
371 Jmsg(jcr, M_MOUNT, 0, _(
372 "Someone woke me up, but I cannot find any appendable\n\
373 volumes for Job=%s.\n"), jcr->Job);
378 jcr->JobStatus = JS_Running;
379 dir_send_job_status(jcr);
380 Dmsg0(130, "leave dir_ask_sysop_to_mount_next_volume\n");
386 * Entered with device blocked and jcr->VolumeName is desired
388 * Leaves with device blocked.
390 * Returns: 1 on success (operator issues a mount command)
392 * Note, must create dev->errmsg on error return.
395 int dir_ask_sysop_to_mount_volume(JCR *jcr, DEVICE *dev)
398 /* ******FIXME******* put these on config variable */
399 int min_wait = 60 * 60;
400 int max_wait = 24 * 60 * 60;
401 int max_num_wait = 9; /* 5 waits =~ 1 day, then 1 day at a time */
408 struct timespec timeout;
410 Dmsg0(130, "enter dir_ask_sysop_to_mount_next_volume\n");
411 if (!jcr->VolumeName[0]) {
412 Mmsg0(&dev->errmsg, _("Cannot request another volume: no volume name given.\n"));
415 ASSERT(dev->dev_blocked);
418 if (job_cancelled(jcr)) {
419 Mmsg(&dev->errmsg, _("Job %s cancelled while waiting for mount on Storage Device \"%s\".\n"),
420 jcr->Job, jcr->dev_name);
423 msg = _("Please mount");
424 Jmsg(jcr, M_MOUNT, 0, _("%s Volume \"%s\" on Storage Device \"%s\" for Job %s\n"),
425 msg, jcr->VolumeName, jcr->dev_name, jcr->Job);
426 Dmsg3(190, "Mount %s on %s for Job %s\n",
427 jcr->VolumeName, jcr->dev_name, jcr->Job);
430 * Wait then send message again
432 gettimeofday(&tv, &tz);
433 timeout.tv_nsec = tv.tv_usec * 1000;
434 timeout.tv_sec = tv.tv_sec + wait_sec;
437 dev_blocked = dev->dev_blocked;
438 dev->dev_blocked = BST_WAITING_FOR_SYSOP; /* indicate waiting for mount */
439 jcr->JobStatus = JS_WaitMount;
440 dir_send_job_status(jcr);
442 for ( ;!job_cancelled(jcr); ) {
443 Dmsg1(190, "I'm going to sleep on device %s\n", dev->dev_name);
444 stat = pthread_cond_timedwait(&dev->wait_next_vol, &dev->mutex, &timeout);
445 if (dev->dev_blocked == BST_WAITING_FOR_SYSOP) {
449 * Someone other than us blocked the device (probably the
450 * user via the Console program.
451 * So, we continue waiting.
453 gettimeofday(&tv, &tz);
455 timeout.tv_sec = tv.tv_sec + 10; /* wait 10 seconds */
457 dev->dev_blocked = dev_blocked;
460 if (stat == ETIMEDOUT) {
461 wait_sec *= 2; /* double wait time */
462 if (wait_sec > max_wait) { /* but not longer than maxtime */
466 if (num_wait >= max_num_wait) {
467 Mmsg(&dev->errmsg, _("Gave up waiting to mount Storage Device \"%s\" for Job %s\n"),
468 jcr->dev_name, jcr->Job);
469 Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
470 Dmsg1(190, "Gave up waiting on device %s\n", dev->dev_name);
471 return 0; /* exceeded maximum waits */
475 if (stat == EINVAL) {
476 Mmsg2(&dev->errmsg, _("pthread error in mount_volume stat=%d ERR=%s\n"),
477 stat, strerror(stat));
478 Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
482 Jmsg(jcr, M_ERROR, 0, _("pthread error in mount_next_volume stat=%d ERR=%s\n"), stat,
485 Dmsg1(190, "Someone woke me for device %s\n", dev->dev_name);
487 /* Restart wait counters */
492 jcr->JobStatus = JS_Running;
493 dir_send_job_status(jcr);
494 Dmsg0(130, "leave dir_ask_sysop_to_mount_next_volume\n");