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) 2000-2004 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 InChanger=%d VolReadTime=%s VolWriteTime=%s\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 InChanger=%d"
51 " VolReadTime=%" lld " VolWriteTime=%" lld;
54 static char OK_create[] = "1000 OK CreateJobMedia\n";
56 /* Forward referenced functions */
57 static int wait_for_sysop(JCR *jcr, DEVICE *dev);
60 * Send current JobStatus to Director
62 int dir_send_job_status(JCR *jcr)
64 return bnet_fsend(jcr->dir_bsock, Job_status, jcr->Job, jcr->JobStatus);
69 * dir_get_volume_info()
71 * dir_find_next_appendable_volume()
73 * Returns: 1 on success and vol info in jcr->VolCatInfo
76 static int do_get_volume_info(JCR *jcr)
78 BSOCK *dir = jcr->dir_bsock;
83 jcr->VolumeName[0] = 0; /* No volume */
84 dcr->VolumeName[0] = 0; /* No volume */
85 if (bnet_recv(dir) <= 0) {
86 Dmsg0(200, "getvolname error bnet_recv\n");
87 Mmsg(&jcr->errmsg, _("Network error on bnet_recv in req_vol_info.\n"));
90 memset(&vol, 0, sizeof(vol));
91 Dmsg1(200, "Get vol info=%s\n", dir->msg);
92 n = sscanf(dir->msg, OK_media, vol.VolCatName,
93 &vol.VolCatJobs, &vol.VolCatFiles,
94 &vol.VolCatBlocks, &vol.VolCatBytes,
95 &vol.VolCatMounts, &vol.VolCatErrors,
96 &vol.VolCatWrites, &vol.VolCatMaxBytes,
97 &vol.VolCatCapacityBytes, vol.VolCatStatus,
98 &vol.Slot, &vol.VolCatMaxJobs, &vol.VolCatMaxFiles,
99 &vol.InChanger, &vol.VolReadTime, &vol.VolWriteTime);
101 Dmsg2(100, "Bad response from Dir fields=%d: %s\n", n, dir->msg);
102 Mmsg(&jcr->errmsg, _("Error getting Volume info: %s\n"), dir->msg);
105 unbash_spaces(vol.VolCatName);
106 pm_strcpy(&jcr->VolumeName, vol.VolCatName); /* set desired VolumeName */
107 bstrncpy(dcr->VolumeName, vol.VolCatName, sizeof(dcr->VolumeName));
108 memcpy(&jcr->VolCatInfo, &vol, sizeof(jcr->VolCatInfo));
109 memcpy(&dcr->VolCatInfo, &vol, sizeof(dcr->VolCatInfo));
111 Dmsg2(200, "do_reqest_vol_info got slot=%d Volume=%s\n",
112 vol.Slot, vol.VolCatName);
118 * Get Volume info for a specific volume from the Director's Database
120 * Returns: 1 on success (not Director guarantees that Pool and MediaType
121 * are correct and VolStatus==Append or
122 * VolStatus==Recycle)
125 * Volume information returned in jcr
127 int dir_get_volume_info(JCR *jcr, enum get_vol_info_rw writing)
129 BSOCK *dir = jcr->dir_bsock;
131 bstrncpy(jcr->VolCatInfo.VolCatName, jcr->VolumeName, sizeof(jcr->VolCatInfo.VolCatName));
132 Dmsg1(200, "dir_get_volume_info=%s\n", jcr->VolCatInfo.VolCatName);
133 bash_spaces(jcr->VolCatInfo.VolCatName);
134 bnet_fsend(dir, Get_Vol_Info, jcr->Job, jcr->VolCatInfo.VolCatName,
135 writing==GET_VOL_INFO_FOR_WRITE?1:0);
136 return do_get_volume_info(jcr);
142 * Get info on the next appendable volume in the Director's database
143 * Returns: 1 on success
146 * Volume information returned in jcr
149 int dir_find_next_appendable_volume(JCR *jcr)
151 BSOCK *dir = jcr->dir_bsock;
153 Dmsg0(200, "dir_find_next_appendable_volume\n");
154 bnet_fsend(dir, Find_media, jcr->Job, 1);
155 return do_get_volume_info(jcr);
160 * After writing a Volume, send the updated statistics
161 * back to the director.
163 int dir_update_volume_info(JCR *jcr, DEVICE *dev, int label)
165 BSOCK *dir = jcr->dir_bsock;
166 time_t LastWritten = time(NULL);
167 char ed1[50], ed2[50], ed3[50], ed4[50];
168 VOLUME_CAT_INFO *vol = &dev->VolCatInfo;
170 if (vol->VolCatName[0] == 0) {
171 Jmsg0(jcr, M_ERROR, 0, _("NULL Volume name. This shouldn't happen!!!\n"));
174 if (dev_state(dev, ST_READ)) {
175 Jmsg0(jcr, M_ERROR, 0, _("Attempt to update_volume_info in read mode!!!\n"));
178 if (!dev_state(dev, ST_LABEL)) {
179 Jmsg0(jcr, M_ERROR, 0, _("Attempt to update_volume_info on non-labeled Volume!!!\n"));
183 Dmsg1(100, "Update cat VolFiles=%d\n", dev->file);
184 /* Just labeled or relabeled the tape */
186 bstrncpy(vol->VolCatStatus, "Append", sizeof(vol->VolCatStatus));
187 vol->VolCatBytes = 1; /* indicates tape labeled */
189 bash_spaces(vol->VolCatName);
190 bnet_fsend(dir, Update_media, jcr->Job,
191 vol->VolCatName, vol->VolCatJobs, vol->VolCatFiles,
192 vol->VolCatBlocks, edit_uint64(vol->VolCatBytes, ed1),
193 vol->VolCatMounts, vol->VolCatErrors,
194 vol->VolCatWrites, edit_uint64(vol->VolCatMaxBytes, ed2),
195 LastWritten, vol->VolCatStatus, vol->Slot, label,
197 edit_uint64(vol->VolReadTime, ed3),
198 edit_uint64(vol->VolWriteTime, ed4) );
200 Dmsg1(120, "update_volume_info(): %s", dir->msg);
201 unbash_spaces(vol->VolCatName);
203 if (!do_get_volume_info(jcr)) {
204 Jmsg(jcr, M_ERROR, 0, "%s", jcr->errmsg);
207 Dmsg1(120, "get_volume_info(): %s", dir->msg);
208 /* Update dev Volume info in case something changed (e.g. expired) */
209 memcpy(&dev->VolCatInfo, &jcr->VolCatInfo, sizeof(dev->VolCatInfo));
214 * After writing a Volume, create the JobMedia record.
216 int dir_create_jobmedia_record(JCR *jcr)
218 BSOCK *dir = jcr->dir_bsock;
221 if (!dcr->WroteVol) {
222 return 1; /* nothing written to tape */
225 dcr->WroteVol = false;
226 bnet_fsend(dir, Create_job_media, jcr->Job,
227 dcr->VolFirstIndex, dcr->VolLastIndex,
228 dcr->StartFile, dcr->EndFile,
229 dcr->StartBlock, dcr->EndBlock);
230 Dmsg1(100, "create_jobmedia(): %s", dir->msg);
231 if (bnet_recv(dir) <= 0) {
232 Dmsg0(190, "create_jobmedia error bnet_recv\n");
233 Jmsg(jcr, M_ERROR, 0, _("Error creating JobMedia record: ERR=%s\n"),
237 Dmsg1(120, "Create_jobmedia: %s", dir->msg);
238 if (strcmp(dir->msg, OK_create) != 0) {
239 Dmsg1(130, "Bad response from Dir: %s\n", dir->msg);
240 Jmsg(jcr, M_ERROR, 0, _("Error creating JobMedia record: %s\n"), dir->msg);
248 * Update File Attribute data
250 int dir_update_file_attributes(JCR *jcr, DEV_RECORD *rec)
252 BSOCK *dir = jcr->dir_bsock;
255 dir->msglen = sprintf(dir->msg, FileAttributes, jcr->Job);
256 dir->msg = check_pool_memory_size(dir->msg, dir->msglen +
257 sizeof(DEV_RECORD) + rec->data_len);
258 ser_begin(dir->msg + dir->msglen, 0);
259 ser_uint32(rec->VolSessionId);
260 ser_uint32(rec->VolSessionTime);
261 ser_int32(rec->FileIndex);
262 ser_int32(rec->Stream);
263 ser_uint32(rec->data_len);
264 ser_bytes(rec->data, rec->data_len);
265 dir->msglen = ser_length(dir->msg);
266 return bnet_send(dir);
271 * Request the sysop to create an appendable volume
273 * Entered with device blocked.
274 * Leaves with device blocked.
276 * Returns: 1 on success (operator issues a mount command)
278 * Note, must create dev->errmsg on error return.
280 * On success, jcr->VolumeName and jcr->VolCatInfo contain
281 * information on suggested volume, but this may not be the
282 * same as what is actually mounted.
284 * When we return with success, the correct tape may or may not
285 * actually be mounted. The calling routine must read it and
288 int dir_ask_sysop_to_create_appendable_volume(JCR *jcr, DEVICE *dev)
294 Dmsg0(130, "enter dir_ask_sysop_to_create_appendable_volume\n");
295 ASSERT(dev->dev_blocked);
297 if (job_canceled(jcr)) {
299 _("Job %s canceled while waiting for mount on Storage Device \"%s\".\n"),
300 jcr->Job, jcr->dev_name);
301 Jmsg(jcr, M_INFO, 0, "%s", dev->errmsg);
304 /* First pass, we *know* there are no appendable volumes, so no need to call */
305 if (!first && dir_find_next_appendable_volume(jcr)) { /* get suggested volume */
306 jstat = JS_WaitMount;
307 unmounted = (dev->dev_blocked == BST_UNMOUNTED) ||
308 (dev->dev_blocked == BST_UNMOUNTED_WAITING_FOR_SYSOP);
310 * If we have a valid volume name and we are not
311 * removable media, return now, or if we have a
312 * Slot for an autochanger, otherwise wait
313 * for the operator to mount the media.
315 if (!unmounted && ((jcr->VolumeName[0] && !dev_cap(dev, CAP_REM) &&
316 dev_cap(dev, CAP_LABEL)) ||
317 (jcr->VolumeName[0] && jcr->VolCatInfo.Slot))) {
318 Dmsg0(100, "Return 1 from mount without wait.\n");
322 Jmsg(jcr, M_MOUNT, 0, _(
323 "Please mount Volume \"%s\" on Storage Device \"%s\" for Job %s\n"
324 "Use \"mount\" command to release Job.\n"),
325 jcr->VolumeName, jcr->dev_name, jcr->Job);
326 Dmsg3(190, "Mount %s on %s for Job %s\n",
327 jcr->VolumeName, jcr->dev_name, jcr->Job);
330 jstat = JS_WaitMedia;
332 Jmsg(jcr, M_MOUNT, 0, _(
333 "Job %s waiting. Cannot find any appendable volumes.\n\
334 Please use the \"label\" command to create a new Volume for:\n\
346 jcr->JobStatus = jstat;
347 dir_send_job_status(jcr);
349 stat = wait_for_sysop(jcr, dev);
351 Dmsg1(200, "Poll timeout in create append vol on device %s\n", dev_name(dev));
355 if (stat == ETIMEDOUT) {
356 if (!double_dev_wait_time(dev)) {
357 Mmsg(&dev->errmsg, _("Gave up waiting to mount Storage Device \"%s\" for Job %s\n"),
358 dev_name(dev), jcr->Job);
359 Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
360 Dmsg1(190, "Gave up waiting on device %s\n", dev_name(dev));
361 return 0; /* exceeded maximum waits */
365 if (stat == EINVAL) {
366 Mmsg2(&dev->errmsg, _("pthread error in mount_next_volume stat=%d ERR=%s\n"),
367 stat, strerror(stat));
368 Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
372 Jmsg(jcr, M_WARNING, 0, _("pthread error in mount_next_volume stat=%d ERR=%s\n"), stat,
375 Dmsg1(190, "Someone woke me for device %s\n", dev_name(dev));
377 /* If no VolumeName, and cannot get one, try again */
378 if (jcr->VolumeName[0] == 0 && !job_canceled(jcr) &&
379 !dir_find_next_appendable_volume(jcr)) {
380 Jmsg(jcr, M_MOUNT, 0, _(
381 "Someone woke me up, but I cannot find any appendable\n\
382 volumes for Job=%s.\n"), jcr->Job);
383 /* Restart wait counters after user interaction */
384 init_dev_wait_timers(dev);
387 unmounted = (dev->dev_blocked == BST_UNMOUNTED) ||
388 (dev->dev_blocked == BST_UNMOUNTED_WAITING_FOR_SYSOP);
390 continue; /* continue to wait */
394 * Device mounted, we have a volume, break and return
398 set_jcr_job_status(jcr, JS_Running);
399 dir_send_job_status(jcr);
400 Dmsg0(130, "leave dir_ask_sysop_to_mount_create_appendable_volume\n");
405 * Request to mount specific Volume
407 * Entered with device blocked and jcr->VolumeName is desired
409 * Leaves with device blocked.
411 * Returns: 1 on success (operator issues a mount command)
413 * Note, must create dev->errmsg on error return.
416 int dir_ask_sysop_to_mount_volume(JCR *jcr, DEVICE *dev)
421 Dmsg0(130, "enter dir_ask_sysop_to_mount_volume\n");
422 if (!jcr->VolumeName[0]) {
423 Mmsg0(&dev->errmsg, _("Cannot request another volume: no volume name given.\n"));
426 ASSERT(dev->dev_blocked);
428 if (job_canceled(jcr)) {
429 Mmsg(&dev->errmsg, _("Job %s canceled while waiting for mount on Storage Device \"%s\".\n"),
430 jcr->Job, jcr->dev_name);
435 * If we have a valid volume name and we are not
436 * removable media, return now, or if we have a
437 * Slot for an autochanger, otherwise wait
438 * for the operator to mount the media.
440 if ((jcr->VolumeName[0] && !dev_cap(dev, CAP_REM) && dev_cap(dev, CAP_LABEL)) ||
441 (jcr->VolumeName[0] && jcr->VolCatInfo.Slot)) {
442 Dmsg0(100, "Return 1 from mount without wait.\n");
447 msg = _("Please mount");
448 Jmsg(jcr, M_MOUNT, 0, _("%s Volume \"%s\" on Storage Device \"%s\" for Job %s\n"),
449 msg, jcr->VolumeName, jcr->dev_name, jcr->Job);
450 Dmsg3(190, "Mount %s on %s for Job %s\n",
451 jcr->VolumeName, jcr->dev_name, jcr->Job);
454 jcr->JobStatus = JS_WaitMount;
455 dir_send_job_status(jcr);
457 stat = wait_for_sysop(jcr, dev); /* wait on device */
459 Dmsg1(200, "Poll timeout in mount vol on device %s\n", dev_name(dev));
460 Dmsg1(200, "Blocked=%d\n", dev->dev_blocked);
464 if (stat == ETIMEDOUT) {
465 if (!double_dev_wait_time(dev)) {
466 Mmsg(&dev->errmsg, _("Gave up waiting to mount Storage Device \"%s\" for Job %s\n"),
467 dev_name(dev), jcr->Job);
468 Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
469 Dmsg1(190, "Gave up waiting on device %s\n", dev_name(dev));
470 return 0; /* exceeded maximum waits */
474 if (stat == EINVAL) {
475 Mmsg2(&dev->errmsg, _("pthread error in mount_volume stat=%d ERR=%s\n"),
476 stat, strerror(stat));
477 Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
481 Jmsg(jcr, M_ERROR, 0, _("pthread error in mount_next_volume stat=%d ERR=%s\n"), stat,
484 Dmsg1(190, "Someone woke me for device %s\n", dev_name(dev));
487 set_jcr_job_status(jcr, JS_Running);
488 dir_send_job_status(jcr);
489 Dmsg0(130, "leave dir_ask_sysop_to_mount_volume\n");
494 * Wait for SysOp to mount a tape
496 static int wait_for_sysop(JCR *jcr, DEVICE *dev)
500 struct timespec timeout;
501 time_t last_heartbeat = 0;
502 time_t first_start = time(NULL);
508 unmounted = (dev->dev_blocked == BST_UNMOUNTED) ||
509 (dev->dev_blocked == BST_UNMOUNTED_WAITING_FOR_SYSOP);
513 * Wait requested time (dev->rem_wait_sec). However, we also wake up every
514 * HB_TIME seconds and send a heartbeat to the FD and the Director
515 * to keep stateful firewalls from closing them down while waiting
518 add_wait = dev->rem_wait_sec;
519 if (me->heartbeat_interval && add_wait > me->heartbeat_interval) {
520 add_wait = me->heartbeat_interval;
522 if (!unmounted && dev->vol_poll_interval && add_wait > dev->vol_poll_interval) {
523 add_wait = dev->vol_poll_interval;
525 gettimeofday(&tv, &tz);
526 timeout.tv_nsec = tv.tv_usec * 1000;
527 timeout.tv_sec = tv.tv_sec + add_wait;
530 dev->dev_prev_blocked = dev->dev_blocked;
531 dev->dev_blocked = BST_WAITING_FOR_SYSOP; /* indicate waiting for mount */
534 for ( ; !job_canceled(jcr); ) {
537 Dmsg3(100, "I'm going to sleep on device %s. HB=%d wait=%d\n", dev_name(dev),
538 (int)me->heartbeat_interval, dev->wait_sec);
540 stat = pthread_cond_timedwait(&dev->wait_next_vol, &dev->mutex, &timeout);
541 Dmsg1(100, "Wokeup from sleep on device stat=%d\n", stat);
544 dev->rem_wait_sec -= (now - start);
546 /* Note, this always triggers the first time. We want that. */
547 if (me->heartbeat_interval) {
548 if (now - last_heartbeat >= me->heartbeat_interval) {
549 /* send heartbeats */
550 if (jcr->file_bsock) {
551 bnet_sig(jcr->file_bsock, BNET_HEARTBEAT);
552 Dmsg0(100, "Send heartbeat to FD.\n");
554 if (jcr->dir_bsock) {
555 bnet_sig(jcr->dir_bsock, BNET_HEARTBEAT);
557 last_heartbeat = now;
562 * Check if user unmounted the device while we were waiting
564 unmounted = (dev->dev_blocked == BST_UNMOUNTED) ||
565 (dev->dev_blocked == BST_UNMOUNTED_WAITING_FOR_SYSOP);
567 if (stat != ETIMEDOUT) { /* we blocked the device */
568 break; /* on error return */
570 if (dev->rem_wait_sec <= 0) { /* on exceeding wait time return */
571 Dmsg0(100, "Exceed wait time.\n");
575 if (!unmounted && dev->vol_poll_interval &&
576 (now - first_start >= dev->vol_poll_interval)) {
577 Dmsg1(200, "In wait blocked=%d\n", dev->dev_blocked);
582 * Check if user mounted the device while we were waiting
584 if (dev->dev_blocked == BST_MOUNT) { /* mount request ? */
589 add_wait = dev->wait_sec - (now - start);
593 if (me->heartbeat_interval && add_wait > me->heartbeat_interval) {
594 add_wait = me->heartbeat_interval;
596 gettimeofday(&tv, &tz);
597 timeout.tv_nsec = tv.tv_usec * 1000;
598 timeout.tv_sec = tv.tv_sec + add_wait; /* additional wait */
599 Dmsg1(100, "Additional wait %d sec.\n", add_wait);
603 dev->dev_blocked = dev->dev_prev_blocked; /* restore entry state */