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;
74 VOLUME_CAT_INFO *vol = &jcr->VolCatInfo;
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 */
97 Dmsg2(200, "do_reqest_vol_info got slot=%d Volume=%s\n",
98 vol->Slot, vol->VolCatName);
104 * Get Volume info for a specific volume from the Director's Database
106 * Returns: 1 on success (not Director guarantees that Pool and MediaType
107 * are correct and VolStatus==Append or
108 * VolStatus==Recycle)
111 * Volume information returned in jcr
113 int dir_get_volume_info(JCR *jcr, int writing)
115 BSOCK *dir = jcr->dir_bsock;
117 strcpy(jcr->VolCatInfo.VolCatName, jcr->VolumeName);
118 Dmsg1(200, "dir_get_volume_info=%s\n", jcr->VolCatInfo.VolCatName);
119 bash_spaces(jcr->VolCatInfo.VolCatName);
120 bnet_fsend(dir, Get_Vol_Info, jcr->Job, jcr->VolCatInfo.VolCatName, writing);
121 return do_request_volume_info(jcr);
127 * Get info on the next appendable volume in the Director's database
128 * Returns: 1 on success
131 * Volume information returned in jcr
134 int dir_find_next_appendable_volume(JCR *jcr)
136 BSOCK *dir = jcr->dir_bsock;
138 Dmsg0(200, "dir_find_next_appendable_volume\n");
139 bnet_fsend(dir, Find_media, jcr->Job, 1);
140 return do_request_volume_info(jcr);
145 * After writing a Volume, send the updated statistics
146 * back to the director.
148 int dir_update_volume_info(JCR *jcr, VOLUME_CAT_INFO *vol, int relabel)
150 BSOCK *dir = jcr->dir_bsock;
151 time_t EndTime = time(NULL);
152 char ed1[50], ed2[50];
154 if (vol->VolCatName[0] == 0) {
155 Jmsg0(jcr, M_ERROR, 0, _("NULL Volume name. This shouldn't happen!!!\n"));
158 bash_spaces(vol->VolCatName);
159 bnet_fsend(dir, Update_media, jcr->Job,
160 vol->VolCatName, vol->VolCatJobs, vol->VolCatFiles,
161 vol->VolCatBlocks, edit_uint64(vol->VolCatBytes, ed1),
162 vol->VolCatMounts, vol->VolCatErrors,
163 vol->VolCatWrites, edit_uint64(vol->VolCatMaxBytes, ed2),
164 EndTime, vol->VolCatStatus, vol->Slot, relabel);
165 Dmsg1(120, "update_volume_data(): %s", dir->msg);
166 unbash_spaces(vol->VolCatName);
167 if (bnet_recv(dir) <= 0) {
168 Dmsg0(190, "updateVolCatInfo error bnet_recv\n");
169 Jmsg(jcr, M_ERROR, 0, _("Error updating Volume Info: %s\n"),
173 Dmsg1(120, "Updatevol: %s", dir->msg);
174 if (strcmp(dir->msg, OK_update) != 0) {
175 Dmsg1(130, "Bad response from Dir: %s\n", dir->msg);
176 Jmsg(jcr, M_ERROR, 0, _("Error updating Volume Info: %s\n"), dir->msg);
183 * After writing a Volume, create the JobMedia record.
185 int dir_create_jobmedia_record(JCR *jcr)
187 BSOCK *dir = jcr->dir_bsock;
189 bnet_fsend(dir, Create_job_media, jcr->Job,
190 jcr->VolFirstFile, jcr->JobFiles,
191 jcr->StartFile, jcr->EndFile,
192 jcr->StartBlock, jcr->EndBlock);
193 Dmsg1(100, "create_jobmedia(): %s", dir->msg);
194 if (bnet_recv(dir) <= 0) {
195 Dmsg0(190, "create_jobmedia error bnet_recv\n");
196 Jmsg(jcr, M_ERROR, 0, _("Error creating JobMedia record: %s\n"),
200 Dmsg1(120, "Create_jobmedia: %s", dir->msg);
201 if (strcmp(dir->msg, OK_update) != 0) {
202 Dmsg1(130, "Bad response from Dir: %s\n", dir->msg);
203 Jmsg(jcr, M_ERROR, 0, _("Error creating JobMedia record: %s\n"), dir->msg);
211 * Update File Attribute data
213 int dir_update_file_attributes(JCR *jcr, DEV_RECORD *rec)
215 BSOCK *dir = jcr->dir_bsock;
218 dir->msglen = sprintf(dir->msg, FileAttributes, jcr->Job);
219 dir->msg = check_pool_memory_size(dir->msg, dir->msglen +
220 sizeof(DEV_RECORD) + rec->data_len);
221 ser_begin(dir->msg + dir->msglen, 0);
222 ser_uint32(rec->VolSessionId);
223 ser_uint32(rec->VolSessionTime);
224 ser_int32(rec->FileIndex);
225 ser_int32(rec->Stream);
226 ser_uint32(rec->data_len);
227 ser_bytes(rec->data, rec->data_len);
228 dir->msglen = ser_length(dir->msg);
229 return bnet_send(dir);
235 * Entered with device blocked.
236 * Leaves with device blocked.
238 * Returns: 1 on success (operator issues a mount command)
240 * Note, must create dev->errmsg on error return.
242 * On success, jcr->VolumeName and jcr->VolCatInfo contain
243 * information on suggested volume, but this may not be the
244 * same as what is actually mounted.
246 * When we return with success, the correct tape may or may not
247 * actually be mounted. The calling routine must read it and
250 int dir_ask_sysop_to_mount_next_volume(JCR *jcr, DEVICE *dev)
253 /* ******FIXME******* put these on config variable */
254 int min_wait = 60 * 60;
255 int max_wait = 24 * 60 * 60;
256 int max_num_wait = 9; /* 5 waits =~ 1 day, then 1 day at a time */
261 Dmsg0(130, "enter dir_ask_sysop_to_mount_next_volume\n");
262 ASSERT(dev->dev_blocked);
265 if (job_canceled(jcr)) {
266 Mmsg(&dev->errmsg, _("Job %s canceled while waiting for mount on Storage Device \"%s\".\n"),
267 jcr->Job, jcr->dev_name);
268 Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
271 if (dir_find_next_appendable_volume(jcr)) { /* get suggested volume */
272 jstat = JS_WaitMount;
274 * If we have a valid volume name and we are not
275 * removable media, return now, otherwise wait
276 * for the operator to mount the media.
278 if (jcr->VolumeName[0] && !dev_cap(dev, CAP_REM) && dev_cap(dev, CAP_LABEL)) {
279 Dmsg0(190, "Return 1 from mount without wait.\n");
282 Jmsg(jcr, M_MOUNT, 0, _(
283 "Please mount Volume \"%s\" on Storage Device \"%s\" for Job %s\n"
284 "Use \"mount\" command to release Job.\n"),
285 jcr->VolumeName, jcr->dev_name, jcr->Job);
286 Dmsg3(190, "Mount %s on %s for Job %s\n",
287 jcr->VolumeName, jcr->dev_name, jcr->Job);
289 jstat = JS_WaitMedia;
290 Jmsg(jcr, M_MOUNT, 0, _(
291 "Job %s waiting. Cannot find any appendable volumes.\n\
292 Please use the \"label\" command to create a new Volume for:\n\
302 jcr->JobStatus = jstat;
303 dir_send_job_status(jcr);
305 stat = wait_for_sysop(jcr, dev, wait_sec);
307 if (stat == ETIMEDOUT) {
308 wait_sec *= 2; /* double wait time */
309 if (wait_sec > max_wait) { /* but not longer than maxtime */
313 if (num_wait >= max_num_wait) {
314 Mmsg(&dev->errmsg, _("Gave up waiting to mount Storage Device \"%s\" for Job %s\n"),
315 jcr->dev_name, jcr->Job);
316 Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
317 Dmsg1(190, "Gave up waiting on device %s\n", dev->dev_name);
318 return 0; /* exceeded maximum waits */
322 if (stat == EINVAL) {
323 Mmsg2(&dev->errmsg, _("pthread error in mount_next_volume stat=%d ERR=%s\n"),
324 stat, strerror(stat));
325 Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
329 Jmsg(jcr, M_WARNING, 0, _("pthread error in mount_next_volume stat=%d ERR=%s\n"), stat,
332 Dmsg1(190, "Someone woke me for device %s\n", dev->dev_name);
334 /* Restart wait counters */
337 /* If no VolumeName, and cannot get one, try again */
338 if (jcr->VolumeName[0] == 0 &&
339 !dir_find_next_appendable_volume(jcr)) {
340 Jmsg(jcr, M_MOUNT, 0, _(
341 "Someone woke me up, but I cannot find any appendable\n\
342 volumes for Job=%s.\n"), jcr->Job);
347 set_jcr_job_status(jcr, JS_Running);
348 dir_send_job_status(jcr);
349 Dmsg0(130, "leave dir_ask_sysop_to_mount_next_volume\n");
355 * Entered with device blocked and jcr->VolumeName is desired
357 * Leaves with device blocked.
359 * Returns: 1 on success (operator issues a mount command)
361 * Note, must create dev->errmsg on error return.
364 int dir_ask_sysop_to_mount_volume(JCR *jcr, DEVICE *dev)
367 /* ******FIXME******* put these on config variable */
368 int min_wait = 60 * 60;
369 int max_wait = 24 * 60 * 60;
370 int max_num_wait = 9; /* 5 waits =~ 1 day, then 1 day at a time */
375 Dmsg0(130, "enter dir_ask_sysop_to_mount_next_volume\n");
376 if (!jcr->VolumeName[0]) {
377 Mmsg0(&dev->errmsg, _("Cannot request another volume: no volume name given.\n"));
380 ASSERT(dev->dev_blocked);
383 if (job_canceled(jcr)) {
384 Mmsg(&dev->errmsg, _("Job %s canceled while waiting for mount on Storage Device \"%s\".\n"),
385 jcr->Job, jcr->dev_name);
388 msg = _("Please mount");
389 Jmsg(jcr, M_MOUNT, 0, _("%s Volume \"%s\" on Storage Device \"%s\" for Job %s\n"),
390 msg, jcr->VolumeName, jcr->dev_name, jcr->Job);
391 Dmsg3(190, "Mount %s on %s for Job %s\n",
392 jcr->VolumeName, jcr->dev_name, jcr->Job);
394 jcr->JobStatus = JS_WaitMount;
395 dir_send_job_status(jcr);
397 stat = wait_for_sysop(jcr, dev, wait_sec); /* wait on device */
399 if (stat == ETIMEDOUT) {
400 wait_sec *= 2; /* double wait time */
401 if (wait_sec > max_wait) { /* but not longer than maxtime */
405 if (num_wait >= max_num_wait) {
406 Mmsg(&dev->errmsg, _("Gave up waiting to mount Storage Device \"%s\" for Job %s\n"),
407 jcr->dev_name, jcr->Job);
408 Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
409 Dmsg1(190, "Gave up waiting on device %s\n", dev->dev_name);
410 return 0; /* exceeded maximum waits */
414 if (stat == EINVAL) {
415 Mmsg2(&dev->errmsg, _("pthread error in mount_volume stat=%d ERR=%s\n"),
416 stat, strerror(stat));
417 Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
421 Jmsg(jcr, M_ERROR, 0, _("pthread error in mount_next_volume stat=%d ERR=%s\n"), stat,
424 Dmsg1(190, "Someone woke me for device %s\n", dev->dev_name);
426 /* Restart wait counters */
431 set_jcr_job_status(jcr, JS_Running);
432 dir_send_job_status(jcr);
433 Dmsg0(130, "leave dir_ask_sysop_to_mount_next_volume\n");
438 * Wait for SysOp to mount a tape
440 static int wait_for_sysop(JCR *jcr, DEVICE *dev, int wait_sec)
444 struct timespec timeout;
446 time_t start = time(NULL);
447 time_t last_heartbeat = 0;
452 * Wait requested time (wait_sec). However, we also wake up every
453 * HB_TIME seconds and send a heartbeat to the FD and the Director
454 * to keep stateful firewalls from closing them down while waiting
458 if (me->heartbeat_interval && add_wait > me->heartbeat_interval) {
459 add_wait = me->heartbeat_interval;
461 gettimeofday(&tv, &tz);
462 timeout.tv_nsec = tv.tv_usec * 1000;
463 timeout.tv_sec = tv.tv_sec + add_wait;
466 dev_blocked = dev->dev_blocked;
467 dev->dev_blocked = BST_WAITING_FOR_SYSOP; /* indicate waiting for mount */
469 for ( ; !job_canceled(jcr); ) {
472 Dmsg3(100, "I'm going to sleep on device %s. HB=%d wait=%d\n", dev->dev_name,
473 (int)me->heartbeat_interval, wait_sec);
474 stat = pthread_cond_timedwait(&dev->wait_next_vol, &dev->mutex, &timeout);
475 Dmsg1(100, "Wokeup from sleep on device stat=%d\n", stat);
479 /* Note, this always triggers the first time. We want that. */
480 if (me->heartbeat_interval) {
481 if (now - last_heartbeat >= me->heartbeat_interval) {
482 /* send heartbeats */
483 if (jcr->file_bsock) {
484 bnet_sig(jcr->file_bsock, BNET_HEARTBEAT);
485 Dmsg0(100, "Send heartbeat to FD.\n");
487 if (jcr->dir_bsock) {
488 bnet_sig(jcr->dir_bsock, BNET_HEARTBEAT);
490 last_heartbeat = now;
494 if (dev->dev_blocked == BST_MOUNT) { /* mount request ? */
499 if (stat != ETIMEDOUT) { /* we blocked the device */
500 break; /* on error return */
502 if (now - start >= wait_sec) { /* on exceeding wait time return */
503 Dmsg0(100, "Exceed wait time.\n");
506 add_wait = wait_sec - (now - start);
507 if (me->heartbeat_interval && add_wait > me->heartbeat_interval) {
508 add_wait = me->heartbeat_interval;
510 gettimeofday(&tv, &tz);
511 timeout.tv_nsec = tv.tv_usec * 1000;
512 timeout.tv_sec = tv.tv_sec + add_wait; /* additional wait */
513 Dmsg1(100, "Additional wait %d sec.\n", add_wait);
516 dev->dev_blocked = dev_blocked;