int
ndmca_monitor_backup_tape_tcp (struct ndm_session *sess)
{
struct ndm_control_agent *ca = sess->control_acb;
int count;
ndmp9_data_state ds;
char *estb;
ndmalogf (sess, 0, 3, "Monitoring backup");
for (count = 0; count < 10; count++) {
ndmca_mon_wait_for_something (sess, count <= 1 ? 30 : 10);
if (ndmca_monitor_get_states(sess) < 0)
break;
#if 0
if (count > 2)
ndmca_mon_show_states(sess);
#endif
ds = ca->data_state.state;
estb = ndmca_data_est(ca);
ndmalogf (sess, 0, 1,
"DATA: bytes %lldKB%s",
ca->data_state.bytes_processed/1024LL,
estb ? estb : "");
ca->job.bytes_written = ca->data_state.bytes_processed;
if (ds == NDMP9_DATA_STATE_ACTIVE) {
count = 0;
continue;
}
/*
* If DATA has halted, the show is over.
*/
if (ds == NDMP9_DATA_STATE_HALTED) {
ndmalogf (sess, 0, 2, "Operation done, cleaning up");
ndmca_monitor_get_post_backup_env (sess);
return 0;
}
}
ndmalogf (sess, 0, 0, "Operation monitoring mishandled, cancelling");
return -1;
}
int
ndmca_monitor_backup (struct ndm_session *sess)
{
struct ndm_control_agent *ca = sess->control_acb;
int count;
ndmp9_data_state ds;
ndmp9_mover_state ms;
char *estb;
if (ca->job.tape_tcp) {
return ndmca_monitor_backup_tape_tcp(sess);
}
ndmalogf (sess, 0, 3, "Monitoring backup");
for (count = 0; count < 10; count++) {
ndmca_mon_wait_for_something (sess, count <= 1 ? 30 : 10);
if (ndmca_monitor_get_states(sess) < 0)
break;
#if 0
if (count > 2)
ndmca_mon_show_states(sess);
#endif
ds = ca->data_state.state;
ms = ca->mover_state.state;
estb = ndmca_data_est(ca);
ndmalogf (sess, 0, 1,
"DATA: bytes %lldKB%s MOVER: written %lldKB record %d",
ca->data_state.bytes_processed/1024LL,
estb ? estb : "",
ca->mover_state.bytes_moved/1024LL,
ca->mover_state.record_num);
ca->job.bytes_written = ca->data_state.bytes_processed;
if (ds == NDMP9_DATA_STATE_ACTIVE
&& ms == NDMP9_MOVER_STATE_ACTIVE) {
count = 0;
continue;
}
/*
* Check MOVER for needed tape change during DATA_FLOW_TO_TAPE.
* Have to do this before checking DATA. Even if DATA halted,
* MOVER may be holding unwritten data. Have to perform
* the tape change.
*/
if (ms == NDMP9_MOVER_STATE_PAUSED) {
ndmp9_mover_pause_reason pr;
pr = ca->mover_state.pause_reason;
if (!ca->pending_notify_mover_paused) {
/* count=count */
continue; /* wait for notice */
}
ca->pending_notify_mover_paused = 0;
ndmalogf (sess, 0, 3, "Mover paused, reason=%s",
ndmp9_mover_pause_reason_to_str (pr));
/* backups are different then recoverys... When
* we reach the end of a window, we signal EOW
* except in V2 where we signal EOF. EOM occurs
* at EOT (or EOF does).
* This is based on reading comments in the email
* archives...
*/
if ((pr == NDMP9_MOVER_PAUSE_EOM) ||
(pr == NDMP9_MOVER_PAUSE_EOW)) {
if (ndmca_monitor_load_next(sess) == 0) {
/* count=count */
continue; /* Happy */
}
/* Something went wrong with tape change. */
} else if ((sess->plumb.tape->protocol_version <= 2) &&
pr == NDMP9_MOVER_PAUSE_EOF) {
if (ndmca_monitor_load_next(sess) == 0) {
/* count=count */
continue; /* Happy */
}
/* Something went wrong with tape change. */
} else {
/* All other pause reasons
* are critically bogus. */
}
ndmalogf (sess, 0, 0,
"Operation paused w/o remedy, cancelling");
ndmca_mover_abort (sess);
return -1;
}
/*
* If DATA has halted, the show is over.
*/
if (ds == NDMP9_DATA_STATE_HALTED) {
if (ms != NDMP9_MOVER_STATE_HALTED) {
ndmalogf (sess, 0, 3,
"DATA halted, MOVER active");
/*
* MOVER still occupied. It might be a
* heartbeat away from asking for another
* tape. Give it a chance.
*/
continue;
}
ndmalogf (sess, 0, 2, "Operation done, cleaning up");
ndmca_monitor_get_post_backup_env (sess);
return 0;
}
#if 1
if (ms == NDMP9_MOVER_STATE_HALTED) {
if (ds == NDMP9_DATA_STATE_ACTIVE) {
ndmalogf (sess, 0, 3,
"MOVER halted, DATA active");
/*
* DATA still occupied.
*/
continue;
}
}
#endif
if (ms != NDMP9_MOVER_STATE_ACTIVE && count == 0) {
/* Not active. Not paused. Something wrong */
ndmalogf (sess, 0, 0,
"Operation in unreasonable state, cancelling");
return -1;
}
}
ndmalogf (sess, 0, 0, "Operation monitoring mishandled, cancelling");
return -1;
}
int
ndmca_monitor_recover (struct ndm_session *sess)
{
struct ndm_control_agent *ca = sess->control_acb;
int count, rc;
ndmp9_data_state ds;
ndmp9_mover_state ms;
char *estb;
int last_state_print = 0;
if (ca->job.tape_tcp) {
return (ndmca_monitor_recover_tape_tcp(sess));
}
ndmalogf (sess, 0, 3, "Monitoring recover");
for (count = 0; count < 10; count++) {
if (ca->pending_notify_data_read) {
ca->pending_notify_data_read = 0;
rc = ndmca_mover_read (sess,
ca->last_notify_data_read.offset,
ca->last_notify_data_read.length);
if (rc) {
ndmalogf (sess, 0, 0, "data-read failed");
return -1;
}
if (count < 5)
continue;
}
ndmca_mon_wait_for_something (sess, count <= 1 ? 30 : 10);
if (ndmca_monitor_get_states(sess) < 0)
break;
#if 0
if (count > 2)
ndmca_mon_show_states(sess);
#endif
ds = ca->data_state.state;
ms = ca->mover_state.state;
estb = ndmca_data_est(ca);
if ((ds != NDMP9_DATA_STATE_ACTIVE) ||
(ms != NDMP9_MOVER_STATE_ACTIVE) ||
((time(0) - last_state_print) >= 5)) {
ndmalogf (sess, 0, 1,
"DATA: bytes %lldKB%s MOVER: read %lldKB record %d",
ca->data_state.bytes_processed/1024LL,
estb ? estb : "",
ca->mover_state.bytes_moved/1024LL,
ca->mover_state.record_num);
last_state_print = time(0);
}
ca->job.bytes_read = ca->data_state.bytes_processed;
if (ds == NDMP9_DATA_STATE_ACTIVE
&& ms == NDMP9_MOVER_STATE_ACTIVE) {
count = 0;
continue;
}
/*
* Check MOVER for needed tape change during DATA_FLOW_TO_TAPE.
* Have to do this before checking DATA. Even if DATA halted,
* MOVER may be holding unwritten data. Have to perform
* the tape change.
*/
if (ms == NDMP9_MOVER_STATE_PAUSED) {
ndmp9_mover_pause_reason pr;
pr = ca->mover_state.pause_reason;
if (!ca->pending_notify_mover_paused) {
/* count=count */
continue; /* wait for notice */
}
ca->pending_notify_mover_paused = 0;
ndmalogf (sess, 0, 3, "Mover paused, reason=%s",
ndmp9_mover_pause_reason_to_str (pr));
if (((pr == NDMP9_MOVER_PAUSE_EOF) ||
(pr == NDMP9_MOVER_PAUSE_SEEK))
&& (ca->cur_media_ix == ca->job.media_tab.n_media)) {
/*
* Last tape consumed by tape agent.
* The DATA agent may be just shy
* of done, but there is no way for
* us to tell. So, close the
* image stream from the TAPE
* agent side, thus indicating
* EOF to the DATA agent.
*/
ndmalogf (sess, 0, 2, "End of tapes");
ndmca_mover_close (sess);
/* count=count */
continue;
}
if (pr == NDMP9_MOVER_PAUSE_EOM
|| pr == NDMP9_MOVER_PAUSE_EOF) {
if (ndmca_monitor_load_next(sess) == 0) {
/* count=count */
continue; /* Happy */
}
/* Something went wrong with tape change. */
} else if (pr == NDMP9_MOVER_PAUSE_SEEK) {
if (ndmca_monitor_seek_tape(sess) == 0) {
/* count=count */
continue; /* Happy */
}
/* Something went wrong with tape change. */
} else {
/* All other pause reasons
* are critically bogus. */
}
ndmalogf (sess, 0, 0,
"Operation paused w/o remedy, cancelling");
ndmca_mover_abort (sess);
return -1;
}
/*
* If DATA has halted, the show is over.
*/
if (ds == NDMP9_DATA_STATE_HALTED) {
if (ms != NDMP9_MOVER_STATE_HALTED) {
ndmalogf (sess, 0, 3,
"DATA halted, MOVER active");
/*
* MOVER still occupied. It might
* figure it out. Then again, it might
* be awaiting a MOVER_READ. The NDMP
* design does not provide a state
* for awaiting MOVER_READ, so we have
* to guess.
*/
if (count > 0) {
ndmca_mover_close(sess);
}
continue;
}
ndmalogf (sess, 0, 2, "Operation done, cleaning up");
ndmca_monitor_get_post_backup_env (sess);
return 0;
}
if (ms != NDMP9_MOVER_STATE_ACTIVE && count == 0) {
/* Not active. Not paused. Something wrong */
ndmalogf (sess, 0, 0,
"Operation in unreasonable state, cancelling");
return -1;
}
}
ndmalogf (sess, 0, 0, "Operation monitoring mishandled, cancelling");
return -1;
}
