bool setup_job(JCR *jcr)
{
int errstat;
jcr->lock();
Dsm_check(100);
init_msg(jcr, jcr->messages);
/* Initialize termination condition variable */
if ((errstat = pthread_cond_init(&jcr->term_wait, NULL)) != 0) {
berrno be;
Jmsg1(jcr, M_FATAL, 0, _("Unable to init job cond variable: ERR=%s\n"), be.bstrerror(errstat));
jcr->unlock();
goto bail_out;
}
jcr->term_wait_inited = true;
create_unique_job_name(jcr, jcr->job->name());
jcr->setJobStatus(JS_Created);
jcr->unlock();
/*
* Open database
*/
Dmsg0(100, "Open database\n");
jcr->db = db_init_database(jcr, jcr->catalog->db_driver, jcr->catalog->db_name,
jcr->catalog->db_user, jcr->catalog->db_password,
jcr->catalog->db_address, jcr->catalog->db_port,
jcr->catalog->db_socket, jcr->catalog->mult_db_connections,
jcr->catalog->disable_batch_insert);
if (!jcr->db || !db_open_database(jcr, jcr->db)) {
Jmsg(jcr, M_FATAL, 0, _("Could not open database \"%s\".\n"),
jcr->catalog->db_name);
if (jcr->db) {
Jmsg(jcr, M_FATAL, 0, "%s", db_strerror(jcr->db));
db_close_database(jcr, jcr->db);
}
goto bail_out;
}
Dmsg0(150, "DB opened\n");
if (!jcr->fname) {
jcr->fname = get_pool_memory(PM_FNAME);
}
if (!jcr->pool_source) {
jcr->pool_source = get_pool_memory(PM_MESSAGE);
pm_strcpy(jcr->pool_source, _("unknown source"));
}
if (jcr->JobReads()) {
if (!jcr->rpool_source) {
jcr->rpool_source = get_pool_memory(PM_MESSAGE);
pm_strcpy(jcr->rpool_source, _("unknown source"));
}
}
/*
* Create Job record
*/
init_jcr_job_record(jcr);
if (!get_or_create_client_record(jcr)) {
goto bail_out;
}
if (!db_create_job_record(jcr, jcr->db, &jcr->jr)) {
Jmsg(jcr, M_FATAL, 0, "%s", db_strerror(jcr->db));
goto bail_out;
}
jcr->JobId = jcr->jr.JobId;
Dmsg4(100, "Created job record JobId=%d Name=%s Type=%c Level=%c\n",
jcr->JobId, jcr->Job, jcr->jr.JobType, jcr->jr.JobLevel);
generate_daemon_event(jcr, "JobStart");
new_plugins(jcr); /* instantiate plugins for this jcr */
generate_plugin_event(jcr, bDirEventJobStart);
if (job_canceled(jcr)) {
goto bail_out;
}
if (jcr->JobReads() && !jcr->rstorage) {
if (jcr->job->storage) {
copy_rwstorage(jcr, jcr->job->storage, _("Job resource"));
} else {
copy_rwstorage(jcr, jcr->job->pool->storage, _("Pool resource"));
}
}
if (!jcr->JobReads()) {
free_rstorage(jcr);
}
/*
* Now, do pre-run stuff, like setting job level (Inc/diff, ...)
* this allows us to setup a proper job start record for restarting
* in case of later errors.
*/
switch (jcr->getJobType()) {
case JT_BACKUP:
if (!do_backup_init(jcr)) {
backup_cleanup(jcr, JS_ErrorTerminated);
goto bail_out;
}
break;
case JT_VERIFY:
if (!do_verify_init(jcr)) {
verify_cleanup(jcr, JS_ErrorTerminated);
goto bail_out;
}
break;
case JT_RESTORE:
if (!do_restore_init(jcr)) {
restore_cleanup(jcr, JS_ErrorTerminated);
goto bail_out;
}
break;
case JT_ADMIN:
if (!do_admin_init(jcr)) {
admin_cleanup(jcr, JS_ErrorTerminated);
goto bail_out;
}
break;
case JT_COPY:
case JT_MIGRATE:
if (!do_migration_init(jcr)) {
migration_cleanup(jcr, JS_ErrorTerminated);
goto bail_out;
}
break;
default:
Pmsg1(0, _("Unimplemented job type: %d\n"), jcr->getJobType());
jcr->setJobStatus(JS_ErrorTerminated);
goto bail_out;
}
generate_job_event(jcr, "JobInit");
generate_plugin_event(jcr, bDirEventJobInit);
Dsm_check(100);
return true;
bail_out:
return false;
}
/*
* This is the engine called by jobq.c:jobq_add() when we were pulled
* from the work queue.
* At this point, we are running in our own thread and all
* necessary resources are allocated -- see jobq.c
*/
static void *job_thread(void *arg)
{
JCR *jcr = (JCR *)arg;
pthread_detach(pthread_self());
Dsm_check(100);
Dmsg0(200, "=====Start Job=========\n");
jcr->setJobStatus(JS_Running); /* this will be set only if no error */
jcr->start_time = time(NULL); /* set the real start time */
jcr->jr.StartTime = jcr->start_time;
if (jcr->job->MaxStartDelay != 0 && jcr->job->MaxStartDelay <
(utime_t)(jcr->start_time - jcr->sched_time)) {
jcr->setJobStatus(JS_Canceled);
Jmsg(jcr, M_FATAL, 0, _("Job canceled because max start delay time exceeded.\n"));
}
if (job_check_maxrunschedtime(jcr)) {
jcr->setJobStatus(JS_Canceled);
Jmsg(jcr, M_FATAL, 0, _("Job canceled because max run sched time exceeded.\n"));
}
/* TODO : check if it is used somewhere */
if (jcr->job->RunScripts == NULL) {
Dmsg0(200, "Warning, job->RunScripts is empty\n");
jcr->job->RunScripts = New(alist(10, not_owned_by_alist));
}
if (!db_update_job_start_record(jcr, jcr->db, &jcr->jr)) {
Jmsg(jcr, M_FATAL, 0, "%s", db_strerror(jcr->db));
}
/* Run any script BeforeJob on dird */
run_scripts(jcr, jcr->job->RunScripts, "BeforeJob");
/*
* We re-update the job start record so that the start
* time is set after the run before job. This avoids
* that any files created by the run before job will
* be saved twice. They will be backed up in the current
* job, but not in the next one unless they are changed.
* Without this, they will be backed up in this job and
* in the next job run because in that case, their date
* is after the start of this run.
*/
jcr->start_time = time(NULL);
jcr->jr.StartTime = jcr->start_time;
if (!db_update_job_start_record(jcr, jcr->db, &jcr->jr)) {
Jmsg(jcr, M_FATAL, 0, "%s", db_strerror(jcr->db));
}
generate_job_event(jcr, "JobRun");
generate_plugin_event(jcr, bDirEventJobRun);
switch (jcr->getJobType()) {
case JT_BACKUP:
if (!job_canceled(jcr) && do_backup(jcr)) {
do_autoprune(jcr);
} else {
backup_cleanup(jcr, JS_ErrorTerminated);
}
break;
case JT_VERIFY:
if (!job_canceled(jcr) && do_verify(jcr)) {
do_autoprune(jcr);
} else {
verify_cleanup(jcr, JS_ErrorTerminated);
}
break;
case JT_RESTORE:
if (!job_canceled(jcr) && do_restore(jcr)) {
do_autoprune(jcr);
} else {
restore_cleanup(jcr, JS_ErrorTerminated);
}
break;
case JT_ADMIN:
if (!job_canceled(jcr) && do_admin(jcr)) {
do_autoprune(jcr);
} else {
admin_cleanup(jcr, JS_ErrorTerminated);
}
break;
case JT_COPY:
case JT_MIGRATE:
if (!job_canceled(jcr) && do_migration(jcr)) {
do_autoprune(jcr);
} else {
migration_cleanup(jcr, JS_ErrorTerminated);
}
break;
default:
Pmsg1(0, _("Unimplemented job type: %d\n"), jcr->getJobType());
break;
}
run_scripts(jcr, jcr->job->RunScripts, "AfterJob");
/* Send off any queued messages */
if (jcr->msg_queue && jcr->msg_queue->size() > 0) {
dequeue_messages(jcr);
}
generate_daemon_event(jcr, "JobEnd");
generate_plugin_event(jcr, bDirEventJobEnd);
Dmsg1(50, "======== End Job stat=%c ==========\n", jcr->JobStatus);
Dsm_check(100);
return NULL;
}
/*
* Do a backup of the specified FileSet
*
* Returns: false on failure
* true on success
*/
bool do_backup(JCR *jcr)
{
int stat;
int tls_need = BNET_TLS_NONE;
BSOCK *fd;
STORE *store;
char ed1[100];
if (jcr->get_JobLevel() == L_VIRTUAL_FULL) {
return do_vbackup(jcr);
}
/* Print Job Start message */
Jmsg(jcr, M_INFO, 0, _("Start Backup JobId %s, Job=%s\n"),
edit_uint64(jcr->JobId, ed1), jcr->Job);
set_jcr_job_status(jcr, JS_Running);
Dmsg2(100, "JobId=%d JobLevel=%c\n", jcr->jr.JobId, jcr->jr.JobLevel);
if (!db_update_job_start_record(jcr, jcr->db, &jcr->jr)) {
Jmsg(jcr, M_FATAL, 0, "%s", db_strerror(jcr->db));
return false;
}
/*
* Open a message channel connection with the Storage
* daemon. This is to let him know that our client
* will be contacting him for a backup session.
*
*/
Dmsg0(110, "Open connection with storage daemon\n");
set_jcr_job_status(jcr, JS_WaitSD);
/*
* Start conversation with Storage daemon
*/
if (!connect_to_storage_daemon(jcr, 10, SDConnectTimeout, 1)) {
return false;
}
/*
* Now start a job with the Storage daemon
*/
if (!start_storage_daemon_job(jcr, NULL, jcr->wstorage)) {
return false;
}
/*
* Start the job prior to starting the message thread below
* to avoid two threads from using the BSOCK structure at
* the same time.
*/
if (!bnet_fsend(jcr->store_bsock, "run")) {
return false;
}
/*
* Now start a Storage daemon message thread. Note,
* this thread is used to provide the catalog services
* for the backup job, including inserting the attributes
* into the catalog. See catalog_update() in catreq.c
*/
if (!start_storage_daemon_message_thread(jcr)) {
return false;
}
Dmsg0(150, "Storage daemon connection OK\n");
set_jcr_job_status(jcr, JS_WaitFD);
if (!connect_to_file_daemon(jcr, 10, FDConnectTimeout, 1)) {
goto bail_out;
}
set_jcr_job_status(jcr, JS_Running);
fd = jcr->file_bsock;
if (!send_include_list(jcr)) {
goto bail_out;
}
if (!send_exclude_list(jcr)) {
goto bail_out;
}
if (!send_level_command(jcr)) {
goto bail_out;
}
/*
* send Storage daemon address to the File daemon
*/
store = jcr->wstore;
if (store->SDDport == 0) {
store->SDDport = store->SDport;
}
/* TLS Requirement */
if (store->tls_enable) {
if (store->tls_require) {
tls_need = BNET_TLS_REQUIRED;
} else {
tls_need = BNET_TLS_OK;
}
}
fd->fsend(storaddr, store->address, store->SDDport, tls_need);
if (!response(jcr, fd, OKstore, "Storage", DISPLAY_ERROR)) {
goto bail_out;
}
if (!send_runscripts_commands(jcr)) {
goto bail_out;
}
/*
* We re-update the job start record so that the start
* time is set after the run before job. This avoids
* that any files created by the run before job will
* be saved twice. They will be backed up in the current
* job, but not in the next one unless they are changed.
* Without this, they will be backed up in this job and
* in the next job run because in that case, their date
* is after the start of this run.
*/
jcr->start_time = time(NULL);
jcr->jr.StartTime = jcr->start_time;
if (!db_update_job_start_record(jcr, jcr->db, &jcr->jr)) {
Jmsg(jcr, M_FATAL, 0, "%s", db_strerror(jcr->db));
}
/*
* If backup is in accurate mode, we send the list of
* all files to FD.
*/
if (!send_accurate_current_files(jcr)) {
goto bail_out;
}
/* Send backup command */
fd->fsend(backupcmd);
if (!response(jcr, fd, OKbackup, "backup", DISPLAY_ERROR)) {
goto bail_out;
}
/* Pickup Job termination data */
stat = wait_for_job_termination(jcr);
db_write_batch_file_records(jcr); /* used by bulk batch file insert */
if (stat == JS_Terminated) {
backup_cleanup(jcr, stat);
return true;
}
return false;
/* Come here only after starting SD thread */
bail_out:
set_jcr_job_status(jcr, JS_ErrorTerminated);
Dmsg1(400, "wait for sd. use=%d\n", jcr->use_count());
/* Cancel SD */
wait_for_job_termination(jcr, FDConnectTimeout);
Dmsg1(400, "after wait for sd. use=%d\n", jcr->use_count());
return false;
}
