* Version $Id$
*/
/*
- Copyright (C) 2000, 2001, 2002 Kern Sibbald and John Walker
+ Copyright (C) 2000-2005 Kern Sibbald
This program is free software; you can redistribute it and/or
modify it under the terms of the GNU General Public License as
#include "stored.h" /* pull in Storage Deamon headers */
/* Requests sent to the Director */
-static char Find_media[] = "CatReq Job=%s FindMedia=%d\n";
-static char Get_Vol_Info[] = "CatReq Job=%s GetVolInfo VolName=%s\n";
-
-static char Update_media[] = "CatReq Job=%s UpdateMedia VolName=%s\
- VolJobs=%d VolFiles=%d VolBlocks=%d VolBytes=%" lld " VolMounts=%d\
- VolErrors=%d VolWrites=%d VolMaxBytes=%" lld " EndTime=%d VolStatus=%s\
- Slot=%d relabel=%d\n";
-
-static char Create_job_media[] = "CatReq Job=%s CreateJobMedia \
- FirstIndex=%d LastIndex=%d StartFile=%d EndFile=%d \
- StartBlock=%d EndBlock=%d\n";
-
-
+static char Find_media[] = "CatReq Job=%s FindMedia=%d\n";
+static char Get_Vol_Info[] = "CatReq Job=%s GetVolInfo VolName=%s write=%d\n";
+static char Update_media[] = "CatReq Job=%s UpdateMedia VolName=%s"
+ " VolJobs=%u VolFiles=%u VolBlocks=%u VolBytes=%s VolMounts=%u"
+ " VolErrors=%u VolWrites=%u MaxVolBytes=%s EndTime=%d VolStatus=%s"
+ " Slot=%d relabel=%d InChanger=%d VolReadTime=%s VolWriteTime=%s\n";
+static char Create_job_media[] = "CatReq Job=%s CreateJobMedia"
+ " FirstIndex=%u LastIndex=%u StartFile=%u EndFile=%u"
+ " StartBlock=%u EndBlock=%u\n";
static char FileAttributes[] = "UpdCat Job=%s FileAttributes ";
-
-static char Job_status[] = "3012 Job %s jobstatus %d\n";
+static char Job_status[] = "Status Job=%s JobStatus=%d\n";
/* Responses received from the Director */
-static char OK_media[] = "1000 OK VolName=%127s VolJobs=%d VolFiles=%d\
- VolBlocks=%d VolBytes=%" lld " VolMounts=%d VolErrors=%d VolWrites=%d\
- VolMaxBytes=%" lld " VolCapacityBytes=%" lld " VolStatus=%20s\
- Slot=%d\n";
+static char OK_media[] = "1000 OK VolName=%127s VolJobs=%u VolFiles=%u"
+ " VolBlocks=%u VolBytes=%" lld " VolMounts=%u VolErrors=%u VolWrites=%u"
+ " MaxVolBytes=%" lld " VolCapacityBytes=%" lld " VolStatus=%20s"
+ " Slot=%d MaxVolJobs=%u MaxVolFiles=%u InChanger=%d"
+ " VolReadTime=%" lld " VolWriteTime=%" lld " EndFile=%u EndBlock=%u";
+
-static char OK_update[] = "1000 OK UpdateMedia\n";
+static char OK_create[] = "1000 OK CreateJobMedia\n";
+/* Forward referenced functions */
+static int wait_for_sysop(DCR *dcr);
/*
* Send current JobStatus to Director
*/
-int dir_send_job_status(JCR *jcr)
+bool dir_send_job_status(JCR *jcr)
{
return bnet_fsend(jcr->dir_bsock, Job_status, jcr->Job, jcr->JobStatus);
}
* dir_get_volume_info()
* and
* dir_find_next_appendable_volume()
+ *
+ * Returns: true on success and vol info in dcr->VolCatInfo
+ * false on failure
*/
-static int do_request_volume_info(JCR *jcr)
+static bool do_get_volume_info(DCR *dcr)
{
+ JCR *jcr = dcr->jcr;
BSOCK *dir = jcr->dir_bsock;
- VOLUME_CAT_INFO *vol = &jcr->VolCatInfo;
+ VOLUME_CAT_INFO vol;
+ int n;
+ int InChanger;
- jcr->VolumeName[0] = 0; /* No volume */
+ dcr->VolumeName[0] = 0; /* No volume */
if (bnet_recv(dir) <= 0) {
- Dmsg0(30, "getvolname error bnet_recv\n");
- return 0;
+ Dmsg0(200, "getvolname error bnet_recv\n");
+ Mmsg(jcr->errmsg, _("Network error on bnet_recv in req_vol_info.\n"));
+ return false;
}
- if (sscanf(dir->msg, OK_media, vol->VolCatName,
- &vol->VolCatJobs, &vol->VolCatFiles,
- &vol->VolCatBlocks, &vol->VolCatBytes,
- &vol->VolCatMounts, &vol->VolCatErrors,
- &vol->VolCatWrites, &vol->VolCatMaxBytes,
- &vol->VolCatCapacityBytes, vol->VolCatStatus,
- &vol->Slot) != 12) {
- Dmsg1(30, "Bad response from Dir: %s\n", dir->msg);
- return 0;
+ memset(&vol, 0, sizeof(vol));
+ Dmsg1(300, "Get vol info=%s", dir->msg);
+ n = sscanf(dir->msg, OK_media, vol.VolCatName,
+ &vol.VolCatJobs, &vol.VolCatFiles,
+ &vol.VolCatBlocks, &vol.VolCatBytes,
+ &vol.VolCatMounts, &vol.VolCatErrors,
+ &vol.VolCatWrites, &vol.VolCatMaxBytes,
+ &vol.VolCatCapacityBytes, vol.VolCatStatus,
+ &vol.Slot, &vol.VolCatMaxJobs, &vol.VolCatMaxFiles,
+ &InChanger, &vol.VolReadTime, &vol.VolWriteTime,
+ &vol.EndFile, &vol.EndBlock);
+ if (n != 19) {
+ Dmsg2(100, "Bad response from Dir fields=%d: %s\n", n, dir->msg);
+ Mmsg(jcr->errmsg, _("Error getting Volume info: %s\n"), dir->msg);
+ return false;
}
- unbash_spaces(vol->VolCatName);
- strcpy(jcr->VolumeName, vol->VolCatName); /* set desired VolumeName */
+ vol.InChanger = InChanger; /* bool in structure */
+ unbash_spaces(vol.VolCatName);
+ bstrncpy(dcr->VolumeName, vol.VolCatName, sizeof(dcr->VolumeName));
+ memcpy(&dcr->VolCatInfo, &vol, sizeof(dcr->VolCatInfo));
- Dmsg2(030, "do_reqest_vol_info got slot=%d Volume=%s\n",
- vol->Slot, vol->VolCatName);
- return 1;
+ Dmsg2(300, "do_reqest_vol_info got slot=%d Volume=%s\n",
+ vol.Slot, vol.VolCatName);
+ return true;
}
/*
* Get Volume info for a specific volume from the Director's Database
*
- * Returns: 1 on success (not Director guarantees that Pool and MediaType
- * are correct and VolStatus==Append or
- * VolStatus==Recycle)
- * 0 on failure
+ * Returns: true on success (not Director guarantees that Pool and MediaType
+ * are correct and VolStatus==Append or
+ * VolStatus==Recycle)
+ * false on failure
*
* Volume information returned in jcr
*/
-int dir_get_volume_info(JCR *jcr)
+bool dir_get_volume_info(DCR *dcr, enum get_vol_info_rw writing)
{
+ JCR *jcr = dcr->jcr;
BSOCK *dir = jcr->dir_bsock;
- strcpy(jcr->VolCatInfo.VolCatName, jcr->VolumeName);
- Dmsg1(200, "dir_get_volume_info=%s\n", jcr->VolCatInfo.VolCatName);
- bash_spaces(jcr->VolCatInfo.VolCatName);
- bnet_fsend(dir, Get_Vol_Info, jcr->Job, jcr->VolCatInfo.VolCatName);
- return do_request_volume_info(jcr);
+ bstrncpy(dcr->VolCatInfo.VolCatName, dcr->VolumeName, sizeof(dcr->VolCatInfo.VolCatName));
+ Dmsg1(300, "dir_get_volume_info=%s\n", dcr->VolCatInfo.VolCatName);
+ bash_spaces(dcr->VolCatInfo.VolCatName);
+ bnet_fsend(dir, Get_Vol_Info, jcr->Job, dcr->VolCatInfo.VolCatName,
+ writing==GET_VOL_INFO_FOR_WRITE?1:0);
+ return do_get_volume_info(dcr);
}
/*
* Get info on the next appendable volume in the Director's database
- * Returns: 1 on success
- * 0 on failure
+ * Returns: true on success
+ * false on failure
*
- * Volume information returned in jcr
+ * Volume information returned in dcr
*
*/
-int dir_find_next_appendable_volume(JCR *jcr)
+bool dir_find_next_appendable_volume(DCR *dcr)
{
+ JCR *jcr = dcr->jcr;
BSOCK *dir = jcr->dir_bsock;
+ JCR *njcr;
Dmsg0(200, "dir_find_next_appendable_volume\n");
- bnet_fsend(dir, Find_media, jcr->Job, 1);
- return do_request_volume_info(jcr);
+ /*
+ * Try the three oldest or most available volumes. Note,
+ * the most available could already be mounted on another
+ * drive, so we continue looking for a not in use Volume.
+ */
+ for (int vol_index=1; vol_index < 3; vol_index++) {
+ bnet_fsend(dir, Find_media, jcr->Job, vol_index);
+ if (do_get_volume_info(dcr)) {
+ Dmsg2(300, "JobId=%d got possible Vol=%s\n", jcr->JobId, dcr->VolumeName);
+ bool found = false;
+ /*
+ * Walk through all jobs and see if the volume is
+ * already mounted. If so, try a different one.
+ * This would be better done by walking through
+ * all the devices.
+ */
+ lock_jcr_chain();
+ foreach_jcr(njcr) {
+ if (jcr == njcr) {
+ free_locked_jcr(njcr);
+ continue; /* us */
+ }
+ Dmsg2(300, "Compare to JobId=%d using Vol=%s\n", njcr->JobId, njcr->dcr->VolumeName);
+ if (njcr->dcr && strcmp(dcr->VolumeName, njcr->dcr->VolumeName) == 0) {
+ found = true;
+ Dmsg1(400, "Vol in use by JobId=%u\n", njcr->JobId);
+ free_locked_jcr(njcr);
+ break;
+ }
+ free_locked_jcr(njcr);
+ }
+ unlock_jcr_chain();
+ if (!found) {
+ Dmsg0(400, "dir_find_next_appendable_volume return true\n");
+ return true; /* Got good Volume */
+ }
+ } else {
+ Dmsg0(200, "No volume info, return false\n");
+ return false;
+ }
+ }
+ Dmsg0(400, "dir_find_next_appendable_volume return true\n");
+ return true;
}
/*
* After writing a Volume, send the updated statistics
- * back to the director.
+ * back to the director. The information comes from the
+ * dev record.
*/
-int dir_update_volume_info(JCR *jcr, VOLUME_CAT_INFO *vol, int relabel)
+bool dir_update_volume_info(DCR *dcr, bool label)
{
+ JCR *jcr = dcr->jcr;
BSOCK *dir = jcr->dir_bsock;
- time_t EndTime = time(NULL);
+ DEVICE *dev = dcr->dev;
+ time_t LastWritten = time(NULL);
+ char ed1[50], ed2[50], ed3[50], ed4[50];
+ VOLUME_CAT_INFO *vol = &dev->VolCatInfo;
+ int InChanger;
if (vol->VolCatName[0] == 0) {
- Jmsg0(jcr, M_ERROR, 0, _("NULL Volume name. This shouldn't happen!!!\n"));
- return 0;
+ Jmsg0(jcr, M_FATAL, 0, _("NULL Volume name. This shouldn't happen!!!\n"));
+ Dmsg0(000, "NULL Volume name. This shouldn't happen!!!\n");
+ return false;
+ }
+ if (dev_state(dev, ST_READ)) {
+ Jmsg0(jcr, M_FATAL, 0, _("Attempt to update_volume_info in read mode!!!\n"));
+ Dmsg0(000, "Attempt to update_volume_info in read mode!!!\n");
+ return false;
}
+
+ Dmsg1(300, "Update cat VolFiles=%d\n", dev->file);
+ /* Just labeled or relabeled the tape */
+ if (label) {
+ bstrncpy(vol->VolCatStatus, "Append", sizeof(vol->VolCatStatus));
+ vol->VolCatBytes = 1; /* indicates tape labeled */
+ }
+ bash_spaces(vol->VolCatName);
+ InChanger = vol->InChanger;
bnet_fsend(dir, Update_media, jcr->Job,
vol->VolCatName, vol->VolCatJobs, vol->VolCatFiles,
- vol->VolCatBlocks, vol->VolCatBytes,
+ vol->VolCatBlocks, edit_uint64(vol->VolCatBytes, ed1),
vol->VolCatMounts, vol->VolCatErrors,
- vol->VolCatWrites, vol->VolCatMaxBytes, EndTime,
- vol->VolCatStatus, vol->Slot, relabel);
- Dmsg1(20, "update_volume_data(): %s", dir->msg);
- if (bnet_recv(dir) <= 0) {
- Dmsg0(90, "updateVolCatInfo error bnet_recv\n");
- return 0;
+ vol->VolCatWrites, edit_uint64(vol->VolCatMaxBytes, ed2),
+ LastWritten, vol->VolCatStatus, vol->Slot, label,
+ InChanger, /* bool in structure */
+ edit_uint64(vol->VolReadTime, ed3),
+ edit_uint64(vol->VolWriteTime, ed4) );
+
+ Dmsg1(300, "update_volume_info(): %s", dir->msg);
+ unbash_spaces(vol->VolCatName);
+
+ if (!do_get_volume_info(dcr)) {
+ Jmsg(jcr, M_FATAL, 0, "%s", jcr->errmsg);
+ Dmsg1(000, "Didn't get vol info: %s", jcr->errmsg);
+ return false;
}
- Dmsg1(20, "Updatevol: %s", dir->msg);
- if (strcmp(dir->msg, OK_update) != 0) {
- Dmsg1(30, "Bad response from Dir: %s\n", dir->msg);
- Jmsg(jcr, M_ERROR, 0, _("Error updating Volume Info: %s\n"), dir->msg);
- return 0;
- }
- return 1;
+ Dmsg1(420, "get_volume_info(): %s", dir->msg);
+ /* Update dev Volume info in case something changed (e.g. expired) */
+ memcpy(&dev->VolCatInfo, &dcr->VolCatInfo, sizeof(dev->VolCatInfo));
+ return true;
}
/*
* After writing a Volume, create the JobMedia record.
*/
-int dir_create_job_media_record(JCR *jcr)
+bool dir_create_jobmedia_record(DCR *dcr)
{
+ JCR *jcr = dcr->jcr;
BSOCK *dir = jcr->dir_bsock;
+ if (!dcr->WroteVol) {
+ return true; /* nothing written to tape */
+ }
+
+ dcr->WroteVol = false;
bnet_fsend(dir, Create_job_media, jcr->Job,
- jcr->VolFirstFile, jcr->JobFiles,
- jcr->start_file, jcr->end_file,
- jcr->start_block, jcr->end_block);
- Dmsg1(20, "create_job_media(): %s", dir->msg);
+ dcr->VolFirstIndex, dcr->VolLastIndex,
+ dcr->StartFile, dcr->EndFile,
+ dcr->StartBlock, dcr->EndBlock);
+ Dmsg1(400, "create_jobmedia(): %s", dir->msg);
if (bnet_recv(dir) <= 0) {
- Dmsg0(90, "create_jobmedia error bnet_recv\n");
- return 0;
+ Dmsg0(190, "create_jobmedia error bnet_recv\n");
+ Jmsg(jcr, M_FATAL, 0, _("Error creating JobMedia record: ERR=%s\n"),
+ bnet_strerror(dir));
+ return false;
}
- Dmsg1(20, "Create_jobmedia: %s", dir->msg);
- if (strcmp(dir->msg, OK_update) != 0) {
- Dmsg1(30, "Bad response from Dir: %s\n", dir->msg);
- Jmsg(jcr, M_ERROR, 0, _("Error creating JobMedia record: %s\n"), dir->msg);
- return 0;
+ Dmsg1(400, "Create_jobmedia: %s", dir->msg);
+ if (strcmp(dir->msg, OK_create) != 0) {
+ Dmsg1(130, "Bad response from Dir: %s\n", dir->msg);
+ Jmsg(jcr, M_FATAL, 0, _("Error creating JobMedia record: %s\n"), dir->msg);
+ return false;
}
- return 1;
+ return true;
}
/*
* Update File Attribute data
*/
-int dir_update_file_attributes(JCR *jcr, DEV_RECORD *rec)
+bool dir_update_file_attributes(DCR *dcr, DEV_RECORD *rec)
{
+ JCR *jcr = dcr->jcr;
BSOCK *dir = jcr->dir_bsock;
ser_declare;
/*
- *
+ * Request the sysop to create an appendable volume
+ *
* Entered with device blocked.
* Leaves with device blocked.
*
- * Returns: 1 on success (operator issues a mount command)
- * 0 on failure
+ * Returns: true on success (operator issues a mount command)
+ * false on failure
* Note, must create dev->errmsg on error return.
*
- * On success, jcr->VolumeName and jcr->VolCatInfo contain
+ * On success, dcr->VolumeName and dcr->VolCatInfo contain
* information on suggested volume, but this may not be the
* same as what is actually mounted.
*
* actually be mounted. The calling routine must read it and
* verify the label.
*/
-int dir_ask_sysop_to_mount_next_volume(JCR *jcr, DEVICE *dev)
+bool dir_ask_sysop_to_create_appendable_volume(DCR *dcr)
{
- struct timeval tv;
- struct timezone tz;
- struct timespec timeout;
- int stat, jstat;
- /* ******FIXME******* put these on config variable */
- int min_wait = 60 * 60;
- int max_wait = 24 * 60 * 60;
- int max_num_wait = 9; /* 5 waits =~ 1 day, then 1 day at a time */
-
- int wait_sec;
- int num_wait = 0;
- int dev_blocked;
- char *msg;
-
- Dmsg0(30, "enter dir_ask_sysop_to_mount_next_volume\n");
+ int stat = 0, jstat;
+ bool unmounted;
+ bool first = true;
+ DEVICE *dev = dcr->dev;
+ JCR *jcr = dcr->jcr;
+
+ Dmsg0(400, "enter dir_ask_sysop_to_create_appendable_volume\n");
ASSERT(dev->dev_blocked);
- wait_sec = min_wait;
for ( ;; ) {
- if (job_cancelled(jcr)) {
- Mmsg(&dev->errmsg, _("Job %s cancelled while waiting for mount on Storage Device \"%s\".\n"),
- jcr->Job, jcr->dev_name);
- Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
- return 0;
+ if (job_canceled(jcr)) {
+ Mmsg(dev->errmsg,
+ _("Job %s canceled while waiting for mount on Storage Device \"%s\".\n"),
+ jcr->Job, dcr->dev_name);
+ Jmsg(jcr, M_INFO, 0, "%s", dev->errmsg);
+ return false;
}
- if (dir_find_next_appendable_volume(jcr)) { /* get suggested volume */
- jstat = JS_WaitMount;
+ /* First pass, we *know* there are no appendable volumes, so no need to call */
+ if (!first && dir_find_next_appendable_volume(dcr)) { /* get suggested volume */
+ unmounted = (dev->dev_blocked == BST_UNMOUNTED) ||
+ (dev->dev_blocked == BST_UNMOUNTED_WAITING_FOR_SYSOP);
/*
* If we have a valid volume name and we are not
- * removable media, return now, otherwise wait
- * for the operator to mount the media.
+ * removable media, return now, or if we have a
+ * Slot for an autochanger, otherwise wait
+ * for the operator to mount the media.
*/
- if (jcr->VolumeName[0] && !(dev->capabilities & CAP_REM) &&
- dev->capabilities & CAP_LABEL) {
- Dmsg0(90, "Return 1 from mount without wait.\n");
- return 1;
+ if (!unmounted && ((dcr->VolumeName[0] && !dev_cap(dev, CAP_REM) &&
+ dev_cap(dev, CAP_LABEL)) ||
+ (dcr->VolumeName[0] && dcr->VolCatInfo.Slot))) {
+ Dmsg0(400, "Return 1 from mount without wait.\n");
+ return true;
}
- if (dev->capabilities & CAP_ANONVOLS) {
- msg = "Suggest mounting";
- } else {
- msg = "Please mount";
+ jstat = JS_WaitMount;
+ if (!dev->poll) {
+ Jmsg(jcr, M_MOUNT, 0, _(
+"Please mount Volume \"%s\" on Storage Device \"%s\" for Job %s\n"
+"Use \"mount\" command to release Job.\n"),
+ dcr->VolumeName, dcr->dev_name, jcr->Job);
+ Dmsg3(400, "Mount %s on %s for Job %s\n",
+ dcr->VolumeName, dcr->dev_name, jcr->Job);
}
- Jmsg(jcr, M_MOUNT, 0, _("%s Volume \"%s\" on Storage Device \"%s\" for Job %s\n"),
- msg, jcr->VolumeName, jcr->dev_name, jcr->Job);
- Dmsg3(90, "Mount %s on %s for Job %s\n",
- jcr->VolumeName, jcr->dev_name, jcr->Job);
} else {
jstat = JS_WaitMedia;
- Jmsg(jcr, M_MOUNT, 0, _(
+ if (!dev->poll) {
+ Jmsg(jcr, M_MOUNT, 0, _(
"Job %s waiting. Cannot find any appendable volumes.\n\
-Please use the \"label\" command to create new Volumes for:\n\
- Storage Device \"%s\" with Pool \"%s\" and Media type \"%s\".\n\
-Use \"mount\" to resume the job.\n"),
- jcr->Job, jcr->dev_name, jcr->pool_name, jcr->media_type);
+Please use the \"label\" command to create a new Volume for:\n\
+ Storage: %s\n\
+ Media type: %s\n\
+ Pool: %s\n"),
+ jcr->Job,
+ dcr->dev_name,
+ dcr->media_type,
+ dcr->pool_name);
+ }
}
- /*
- * Wait then send message again
- */
- gettimeofday(&tv, &tz);
- timeout.tv_nsec = tv.tv_usec * 1000;
- timeout.tv_sec = tv.tv_sec + wait_sec;
+ first = false;
- P(dev->mutex);
- dev_blocked = dev->dev_blocked;
- dev->dev_blocked = BST_WAITING_FOR_SYSOP; /* indicate waiting for mount */
jcr->JobStatus = jstat;
dir_send_job_status(jcr);
- for ( ;!job_cancelled(jcr); ) {
- Dmsg1(90, "I'm going to sleep on device %s\n", dev->dev_name);
- stat = pthread_cond_timedwait(&dev->wait_next_vol, &dev->mutex, &timeout);
- if (dev->dev_blocked == BST_WAITING_FOR_SYSOP) {
- break;
- }
- /*
- * Someone other than us blocked the device (probably the
- * user via the Console program.
- * So, we continue waiting.
- */
- gettimeofday(&tv, &tz);
- timeout.tv_nsec = 0;
- timeout.tv_sec = tv.tv_sec + 10; /* wait 10 seconds */
+ stat = wait_for_sysop(dcr);
+ if (dev->poll) {
+ Dmsg1(400, "Poll timeout in create append vol on device %s\n", dev_name(dev));
+ continue;
}
- dev->dev_blocked = dev_blocked;
- V(dev->mutex);
if (stat == ETIMEDOUT) {
- wait_sec *= 2; /* double wait time */
- if (wait_sec > max_wait) { /* but not longer than maxtime */
- wait_sec = max_wait;
- }
- num_wait++;
- if (num_wait >= max_num_wait) {
- Mmsg(&dev->errmsg, _("Gave up waiting to mount Storage Device \"%s\" for Job %s\n"),
- jcr->dev_name, jcr->Job);
+ if (!double_dev_wait_time(dev)) {
+ Mmsg(dev->errmsg, _("Max time exceeded waiting to mount Storage Device \"%s\" for Job %s\n"),
+ dev_name(dev), jcr->Job);
Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
- Dmsg1(90, "Gave up waiting on device %s\n", dev->dev_name);
- return 0; /* exceeded maximum waits */
+ Dmsg1(400, "Gave up waiting on device %s\n", dev_name(dev));
+ return false; /* exceeded maximum waits */
}
continue;
}
if (stat == EINVAL) {
- Mmsg2(&dev->errmsg, _("pthread error in mount_next_volume stat=%d ERR=%s\n"),
- stat, strerror(stat));
+ berrno be;
+ Mmsg2(dev->errmsg, _("pthread error in mount_next_volume stat=%d ERR=%s\n"),
+ stat, be.strerror(stat));
Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
- return 0;
+ return false;
}
if (stat != 0) {
+ berrno be;
Jmsg(jcr, M_WARNING, 0, _("pthread error in mount_next_volume stat=%d ERR=%s\n"), stat,
- strerror(stat));
+ be.strerror(stat));
}
- Dmsg1(90, "Someone woke me for device %s\n", dev->dev_name);
+ Dmsg1(400, "Someone woke me for device %s\n", dev_name(dev));
- /* Restart wait counters */
- wait_sec = min_wait;
- num_wait = 0;
/* If no VolumeName, and cannot get one, try again */
- if (jcr->VolumeName[0] == 0 &&
- !dir_find_next_appendable_volume(jcr)) {
+ if (dcr->VolumeName[0] == 0 && !job_canceled(jcr) &&
+ !dir_find_next_appendable_volume(dcr)) {
Jmsg(jcr, M_MOUNT, 0, _(
"Someone woke me up, but I cannot find any appendable\n\
volumes for Job=%s.\n"), jcr->Job);
+ /* Restart wait counters after user interaction */
+ init_dev_wait_timers(dev);
continue;
}
+ unmounted = (dev->dev_blocked == BST_UNMOUNTED) ||
+ (dev->dev_blocked == BST_UNMOUNTED_WAITING_FOR_SYSOP);
+ if (unmounted) {
+ continue; /* continue to wait */
+ }
+
+ /*
+ * Device mounted, we have a volume, break and return
+ */
break;
}
- jcr->JobStatus = JS_Running;
+ set_jcr_job_status(jcr, JS_Running);
dir_send_job_status(jcr);
- Dmsg0(30, "leave dir_ask_sysop_to_mount_next_volume\n");
- return 1;
+ Dmsg0(400, "leave dir_ask_sysop_to_mount_create_appendable_volume\n");
+ return true;
}
/*
- *
- * Entered with device blocked and jcr->VolumeName is desired
+ * Request to mount specific Volume
+ *
+ * Entered with device blocked and dcr->VolumeName is desired
* volume.
* Leaves with device blocked.
*
- * Returns: 1 on success (operator issues a mount command)
- * 0 on failure
- * Note, must create dev->errmsg on error return.
+ * Returns: true on success (operator issues a mount command)
+ * false on failure
+ * Note, must create dev->errmsg on error return.
*
*/
-int dir_ask_sysop_to_mount_volume(JCR *jcr, DEVICE *dev)
+bool dir_ask_sysop_to_mount_volume(DCR *dcr)
{
- int stat, jstat;
- /* ******FIXME******* put these on config variable */
- int min_wait = 60 * 60;
- int max_wait = 24 * 60 * 60;
- int max_num_wait = 9; /* 5 waits =~ 1 day, then 1 day at a time */
- int wait_sec;
- int num_wait = 0;
- int dev_blocked;
- char *msg;
- struct timeval tv;
- struct timezone tz;
- struct timespec timeout;
-
- Dmsg0(30, "enter dir_ask_sysop_to_mount_next_volume\n");
- if (!jcr->VolumeName[0]) {
- Mmsg0(&dev->errmsg, _("Cannot request another volume: no volume name given.\n"));
- return 0;
+ int stat = 0;
+ const char *msg;
+ DEVICE *dev = dcr->dev;
+ JCR *jcr = dcr->jcr;
+
+ Dmsg0(400, "enter dir_ask_sysop_to_mount_volume\n");
+ if (!dcr->VolumeName[0]) {
+ Mmsg0(dev->errmsg, _("Cannot request another volume: no volume name given.\n"));
+ return false;
}
ASSERT(dev->dev_blocked);
- wait_sec = min_wait;
for ( ;; ) {
- if (job_cancelled(jcr)) {
- Mmsg(&dev->errmsg, _("Job %s cancelled while waiting for mount on Storage Device \"%s\".\n"),
- jcr->Job, jcr->dev_name);
- return 0;
+ if (job_canceled(jcr)) {
+ Mmsg(dev->errmsg, _("Job %s canceled while waiting for mount on Storage Device \"%s\".\n"),
+ jcr->Job, dcr->dev_name);
+ return false;
}
- msg = _("Please mount");
- Jmsg(jcr, M_MOUNT, 0, _("%s Volume \"%s\" on Storage Device \"%s\" for Job %s\n"),
- msg, jcr->VolumeName, jcr->dev_name, jcr->Job);
- Dmsg3(90, "Mount %s on %s for Job %s\n",
- jcr->VolumeName, jcr->dev_name, jcr->Job);
- /*
- * Wait then send message again
- */
- gettimeofday(&tv, &tz);
- timeout.tv_nsec = tv.tv_usec * 1000;
- timeout.tv_sec = tv.tv_sec + wait_sec;
+ if (!dev->poll) {
+ msg = _("Please mount");
+ Jmsg(jcr, M_MOUNT, 0, _("%s Volume \"%s\" on Storage Device \"%s\" for Job %s\n"),
+ msg, dcr->VolumeName, dcr->dev_name, jcr->Job);
+ Dmsg3(400, "Mount \"%s\" on device \"%s\" for Job %s\n",
+ dcr->VolumeName, dcr->dev_name, jcr->Job);
+ }
- P(dev->mutex);
- dev_blocked = dev->dev_blocked;
- dev->dev_blocked = BST_WAITING_FOR_SYSOP; /* indicate waiting for mount */
- jcr->JobStatus = jstat;
+ jcr->JobStatus = JS_WaitMount;
dir_send_job_status(jcr);
- for ( ;!job_cancelled(jcr); ) {
- Dmsg1(90, "I'm going to sleep on device %s\n", dev->dev_name);
- stat = pthread_cond_timedwait(&dev->wait_next_vol, &dev->mutex, &timeout);
- if (dev->dev_blocked == BST_WAITING_FOR_SYSOP) {
- break;
- }
- /*
- * Someone other than us blocked the device (probably the
- * user via the Console program.
- * So, we continue waiting.
- */
- gettimeofday(&tv, &tz);
- timeout.tv_nsec = 0;
- timeout.tv_sec = tv.tv_sec + 10; /* wait 10 seconds */
+ stat = wait_for_sysop(dcr); ; /* wait on device */
+ if (dev->poll) {
+ Dmsg1(400, "Poll timeout in mount vol on device %s\n", dev_name(dev));
+ Dmsg1(400, "Blocked=%s\n", edit_blocked_reason(dev));
+ return true;
}
- dev->dev_blocked = dev_blocked;
- V(dev->mutex);
if (stat == ETIMEDOUT) {
- wait_sec *= 2; /* double wait time */
- if (wait_sec > max_wait) { /* but not longer than maxtime */
- wait_sec = max_wait;
- }
- num_wait++;
- if (num_wait >= max_num_wait) {
- Mmsg(&dev->errmsg, _("Gave up waiting to mount Storage Device \"%s\" for Job %s\n"),
- jcr->dev_name, jcr->Job);
+ if (!double_dev_wait_time(dev)) {
+ Mmsg(dev->errmsg, _("Max time exceeded waiting to mount Storage Device \"%s\" for Job %s\n"),
+ dev_name(dev), jcr->Job);
Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
- Dmsg1(90, "Gave up waiting on device %s\n", dev->dev_name);
- return 0; /* exceeded maximum waits */
+ Dmsg1(400, "Gave up waiting on device %s\n", dev_name(dev));
+ return false; /* exceeded maximum waits */
}
continue;
}
if (stat == EINVAL) {
- Mmsg2(&dev->errmsg, _("pthread error in mount_volume stat=%d ERR=%s\n"),
- stat, strerror(stat));
+ berrno be;
+ Mmsg2(dev->errmsg, _("pthread error in mount_volume stat=%d ERR=%s\n"),
+ stat, be.strerror(stat));
Jmsg(jcr, M_FATAL, 0, "%s", dev->errmsg);
- return 0;
+ return false;
}
if (stat != 0) {
- Jmsg(jcr, M_ERROR, 0, _("pthread error in mount_next_volume stat=%d ERR=%s\n"), stat,
- strerror(stat));
+ berrno be;
+ Jmsg(jcr, M_FATAL, 0, _("pthread error in mount_next_volume stat=%d ERR=%s\n"), stat,
+ be.strerror(stat));
}
- Dmsg1(90, "Someone woke me for device %s\n", dev->dev_name);
-
- /* Restart wait counters */
- wait_sec = min_wait;
- num_wait = 0;
+ Dmsg1(400, "Someone woke me for device %s\n", dev_name(dev));
break;
}
- jcr->JobStatus = JS_Running;
+ set_jcr_job_status(jcr, JS_Running);
dir_send_job_status(jcr);
- Dmsg0(30, "leave dir_ask_sysop_to_mount_next_volume\n");
- return 1;
+ Dmsg0(400, "leave dir_ask_sysop_to_mount_volume\n");
+ return true;
+}
+
+/*
+ * Wait for SysOp to mount a tape
+ */
+static int wait_for_sysop(DCR *dcr)
+{
+ struct timeval tv;
+ struct timezone tz;
+ struct timespec timeout;
+ time_t last_heartbeat = 0;
+ time_t first_start = time(NULL);
+ int stat = 0;
+ int add_wait;
+ bool unmounted;
+ DEVICE *dev = dcr->dev;
+ JCR *jcr = dcr->jcr;
+
+ P(dev->mutex);
+ unmounted = (dev->dev_blocked == BST_UNMOUNTED) ||
+ (dev->dev_blocked == BST_UNMOUNTED_WAITING_FOR_SYSOP);
+
+ dev->poll = false;
+ /*
+ * Wait requested time (dev->rem_wait_sec). However, we also wake up every
+ * HB_TIME seconds and send a heartbeat to the FD and the Director
+ * to keep stateful firewalls from closing them down while waiting
+ * for the operator.
+ */
+ add_wait = dev->rem_wait_sec;
+ if (me->heartbeat_interval && add_wait > me->heartbeat_interval) {
+ add_wait = me->heartbeat_interval;
+ }
+ /* If the user did not unmount the tape and we are polling, ensure
+ * that we poll at the correct interval.
+ */
+ if (!unmounted && dev->vol_poll_interval && add_wait > dev->vol_poll_interval) {
+ add_wait = dev->vol_poll_interval;
+ }
+ gettimeofday(&tv, &tz);
+ timeout.tv_nsec = tv.tv_usec * 1000;
+ timeout.tv_sec = tv.tv_sec + add_wait;
+
+ if (!unmounted) {
+ dev->dev_prev_blocked = dev->dev_blocked;
+ dev->dev_blocked = BST_WAITING_FOR_SYSOP; /* indicate waiting for mount */
+ }
+
+ for ( ; !job_canceled(jcr); ) {
+ time_t now, start;
+
+ Dmsg3(400, "I'm going to sleep on device %s. HB=%d wait=%d\n", dev_name(dev),
+ (int)me->heartbeat_interval, dev->wait_sec);
+ start = time(NULL);
+ /* Wait required time */
+ stat = pthread_cond_timedwait(&dev->wait_next_vol, &dev->mutex, &timeout);
+ Dmsg1(400, "Wokeup from sleep on device stat=%d\n", stat);
+
+ now = time(NULL);
+ dev->rem_wait_sec -= (now - start);
+
+ /* Note, this always triggers the first time. We want that. */
+ if (me->heartbeat_interval) {
+ if (now - last_heartbeat >= me->heartbeat_interval) {
+ /* send heartbeats */
+ if (jcr->file_bsock) {
+ bnet_sig(jcr->file_bsock, BNET_HEARTBEAT);
+ Dmsg0(400, "Send heartbeat to FD.\n");
+ }
+ if (jcr->dir_bsock) {
+ bnet_sig(jcr->dir_bsock, BNET_HEARTBEAT);
+ }
+ last_heartbeat = now;
+ }
+ }
+
+ /*
+ * Check if user unmounted the device while we were waiting
+ */
+ unmounted = (dev->dev_blocked == BST_UNMOUNTED) ||
+ (dev->dev_blocked == BST_UNMOUNTED_WAITING_FOR_SYSOP);
+
+ if (stat != ETIMEDOUT) { /* we blocked the device */
+ break; /* on error return */
+ }
+ if (dev->rem_wait_sec <= 0) { /* on exceeding wait time return */
+ Dmsg0(400, "Exceed wait time.\n");
+ break;
+ }
+
+ if (!unmounted && dev->vol_poll_interval &&
+ (now - first_start >= dev->vol_poll_interval)) {
+ Dmsg1(400, "In wait blocked=%s\n", edit_blocked_reason(dev));
+ dev->poll = true; /* returning a poll event */
+ break;
+ }
+ /*
+ * Check if user mounted the device while we were waiting
+ */
+ if (dev->dev_blocked == BST_MOUNT) { /* mount request ? */
+ stat = 0;
+ break;
+ }
+
+ add_wait = dev->wait_sec - (now - start);
+ if (add_wait < 0) {
+ add_wait = 0;
+ }
+ if (me->heartbeat_interval && add_wait > me->heartbeat_interval) {
+ add_wait = me->heartbeat_interval;
+ }
+ gettimeofday(&tv, &tz);
+ timeout.tv_nsec = tv.tv_usec * 1000;
+ timeout.tv_sec = tv.tv_sec + add_wait; /* additional wait */
+ Dmsg1(400, "Additional wait %d sec.\n", add_wait);
+ }
+
+ if (!unmounted) {
+ dev->dev_blocked = dev->dev_prev_blocked; /* restore entry state */
+ }
+ V(dev->mutex);
+ return stat;
}
projects / bacula / bacula / blobdiff