* 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 "dird.h"
/* Commands sent to Storage daemon */
-static char jobcmd[] = "JobId=%d job=%s job_name=%s client_name=%s \
-type=%d level=%d FileSet=%s NoAttr=%d SpoolAttr=%d\n";
-static char use_device[] = "use device=%s media_type=%s pool_name=%s pool_type=%s\n";
+static char jobcmd[] = "JobId=%d job=%s job_name=%s client_name=%s "
+ "type=%d level=%d FileSet=%s NoAttr=%d SpoolAttr=%d FileSetMD5=%s "
+ "SpoolData=%d WritePartAfterJob=%d NewVol=%d\n";
+static char use_storage[] = "use storage=%s media_type=%s pool_name=%s "
+ "pool_type=%s append=%d\n";
+static char use_device[] = "use device=%s\n";
+//static char query_device[] = "query device=%s";
/* Response from Storage daemon */
static char OKjob[] = "3000 OK Job SDid=%d SDtime=%d Authorization=%100s\n";
-static char OK_device[] = "3000 OK use device\n";
+static char OK_device[] = "3000 OK use device device=%s\n";
/* Storage Daemon requests */
static char Job_start[] = "3010 Job %127s start\n";
-static char Job_end[] =
+static char Job_end[] =
"3099 Job %127s end JobStatus=%d JobFiles=%d JobBytes=%" lld "\n";
-static char Job_status[] = "3012 Job %127s jobstatus %d\n";
/* Forward referenced functions */
-static void *msg_thread(void *arg);
+extern "C" void *msg_thread(void *arg);
/*
* Establish a message channel connection with the Storage daemon
- * and perform authentication.
+ * and perform authentication.
*/
-int connect_to_storage_daemon(JCR *jcr, int retry_interval,
+bool connect_to_storage_daemon(JCR *jcr, int retry_interval,
int max_retry_time, int verbose)
{
BSOCK *sd;
+ STORE *store;
+
+ if (jcr->store_bsock) {
+ return true; /* already connected */
+ }
+ store = (STORE *)jcr->storage->first();
/*
- * Open message channel with the Storage daemon
+ * Open message channel with the Storage daemon
*/
- Dmsg2(200, "bnet_connect to Storage daemon %s:%d\n", jcr->store->address,
- jcr->store->SDport);
+ Dmsg2(100, "bnet_connect to Storage daemon %s:%d\n", store->address,
+ store->SDport);
sd = bnet_connect(jcr, retry_interval, max_retry_time,
- _("Storage daemon"), jcr->store->address,
- NULL, jcr->store->SDport, verbose);
+ _("Storage daemon"), store->address,
+ NULL, store->SDport, verbose);
if (sd == NULL) {
- return 0;
+ return false;
}
- sd->res = (RES *)jcr->store; /* save pointer to other end */
+ sd->res = (RES *)store; /* save pointer to other end */
jcr->store_bsock = sd;
- if (!authenticate_storage_daemon(jcr)) {
- return 0;
+ if (!authenticate_storage_daemon(jcr, store)) {
+ bnet_close(sd);
+ jcr->store_bsock = NULL;
+ return false;
}
- return 1;
+ return true;
+}
+
+/*
+ * Here we ask the SD to send us the info for a
+ * particular device resource.
+ */
+#ifdef needed
+bool update_device_res(JCR *jcr, DEVICE *dev)
+{
+ POOL_MEM device_name;
+ BSOCK *sd;
+ if (!connect_to_storage_daemon(jcr, 5, 30, 0)) {
+ return false;
+ }
+ sd = jcr->store_bsock;
+ pm_strcpy(device_name, dev->hdr.name);
+ bash_spaces(device_name);
+ bnet_fsend(sd, query_device, device_name.c_str());
+ Dmsg1(100, ">stored: %s\n", sd->msg);
+ /* The data is returned through Device_update */
+ if (bget_dirmsg(sd) <= 0) {
+ return false;
+ }
+ return true;
}
+#endif
/*
* Start a job with the Storage daemon
*/
-int start_storage_daemon_job(JCR *jcr)
+int start_storage_daemon_job(JCR *jcr, alist *store, int append)
{
- int status;
+ bool ok = false;
STORE *storage;
BSOCK *sd;
char auth_key[100];
- char *device_name, *pool_name, *pool_type, *media_type;
- int device_name_len, pool_name_len, pool_type_len, media_type_len;
+ POOL_MEM store_name, device_name, pool_name, pool_type, media_type;
+ char PoolId[50];
- storage = jcr->store;
sd = jcr->store_bsock;
/*
* Now send JobId and permissions, and get back the authorization key.
bash_spaces(jcr->job->hdr.name);
bash_spaces(jcr->client->hdr.name);
bash_spaces(jcr->fileset->hdr.name);
- bnet_fsend(sd, jobcmd, jcr->JobId, jcr->Job, jcr->job->hdr.name,
- jcr->client->hdr.name, jcr->JobType, jcr->JobLevel,
+ if (jcr->fileset->MD5[0] == 0) {
+ bstrncpy(jcr->fileset->MD5, "**Dummy**", sizeof(jcr->fileset->MD5));
+ }
+ bnet_fsend(sd, jobcmd, jcr->JobId, jcr->Job, jcr->job->hdr.name,
+ jcr->client->hdr.name, jcr->JobType, jcr->JobLevel,
jcr->fileset->hdr.name, !jcr->pool->catalog_files,
- jcr->job->SpoolAttributes);
+ jcr->job->SpoolAttributes, jcr->fileset->MD5, jcr->spool_data,
+ jcr->write_part_after_job, jcr->job->NewVolEachJob);
+ Dmsg1(100, ">stored: %s\n", sd->msg);
unbash_spaces(jcr->job->hdr.name);
unbash_spaces(jcr->client->hdr.name);
unbash_spaces(jcr->fileset->hdr.name);
- if (bnet_recv(sd) > 0) {
- Dmsg1(110, "<stored: %s", sd->msg);
- if (sscanf(sd->msg, OKjob, &jcr->VolSessionId,
+ if (bget_dirmsg(sd) > 0) {
+ Dmsg1(100, "<stored: %s", sd->msg);
+ if (sscanf(sd->msg, OKjob, &jcr->VolSessionId,
&jcr->VolSessionTime, &auth_key) != 3) {
+ Dmsg1(100, "BadJob=%s\n", sd->msg);
Jmsg(jcr, M_FATAL, 0, _("Storage daemon rejected Job command: %s\n"), sd->msg);
return 0;
} else {
return 0;
}
- /*
- * Send use device = xxx media = yyy pool = zzz
- */
- device_name_len = strlen(storage->dev_name) + 1;
- media_type_len = strlen(storage->media_type) + 1;
- pool_type_len = strlen(jcr->pool->pool_type) + 1;
- pool_name_len = strlen(jcr->pool->hdr.name) + 1;
- device_name = (char *) get_memory(device_name_len);
- pool_name = (char *) get_memory(pool_name_len);
- pool_type = (char *) get_memory(pool_type_len);
- media_type = (char *) get_memory(media_type_len);
- memcpy(device_name, storage->dev_name, device_name_len);
- memcpy(media_type, storage->media_type, media_type_len);
- memcpy(pool_type, jcr->pool->pool_type, pool_type_len);
- memcpy(pool_name, jcr->pool->hdr.name, pool_name_len);
- bash_spaces(device_name);
- bash_spaces(media_type);
+ pm_strcpy(pool_type, jcr->pool->pool_type);
+ pm_strcpy(pool_name, jcr->pool->hdr.name);
bash_spaces(pool_type);
bash_spaces(pool_name);
- sd->msg = check_pool_memory_size(sd->msg, sizeof(device_name) +
- device_name_len + media_type_len + pool_type_len + pool_name_len);
- bnet_fsend(sd, use_device, device_name, media_type, pool_name, pool_type);
- Dmsg1(110, ">stored: %s", sd->msg);
- status = response(sd, OK_device, "Use Device");
-
- free_memory(device_name);
- free_memory(media_type);
- free_memory(pool_name);
- free_memory(pool_type);
-
- return status;
+ edit_int64(jcr->PoolId, PoolId);
+
+ /*
+ * We have two loops here. The first comes from the
+ * Storage = associated with the Job, and we need
+ * to attach to each one.
+ * The inner loop loops over all the alternative devices
+ * associated with each Storage. It selects the first
+ * available one.
+ *
+ * Note, the outer loop is not yet implemented.
+ */
+// foreach_alist(storage, store) {
+ storage = (STORE *)store->first();
+ pm_strcpy(store_name, storage->hdr.name);
+ bash_spaces(store_name);
+ pm_strcpy(media_type, storage->media_type);
+ bash_spaces(media_type);
+ bnet_fsend(sd, use_storage, store_name.c_str(), media_type.c_str(),
+ pool_name.c_str(), pool_type.c_str(), append);
+
+ DEVICE *dev;
+ /* Loop over alternative storage Devices until one is OK */
+ foreach_alist(dev, storage->device) {
+ pm_strcpy(device_name, dev->hdr.name);
+ bash_spaces(device_name);
+ bnet_fsend(sd, use_device, device_name.c_str());
+ Dmsg1(100, ">stored: %s", sd->msg);
+ }
+ bnet_sig(sd, BNET_EOD);
+ if (bget_dirmsg(sd) > 0) {
+ Dmsg1(100, "<stored: %s", sd->msg);
+ /* ****FIXME**** save actual device name */
+ ok = sscanf(sd->msg, OK_device, device_name.c_str()) == 1;
+ } else {
+ POOL_MEM err_msg;
+ pm_strcpy(err_msg, sd->msg); /* save message */
+ Jmsg(jcr, M_WARNING, 0, _("\n"
+ " Storage daemon didn't accept Device \"%s\" because:\n %s"),
+ device_name.c_str(), err_msg.c_str()/* sd->msg */);
+ }
+// if (!ok) {
+// break;
+// }
+// }
+ if (ok) {
+ ok = bnet_fsend(sd, "run");
+ Dmsg1(100, ">stored: %s\n", sd->msg);
+ }
+ return ok;
}
-/*
+/*
* Start a thread to handle Storage daemon messages and
* Catalog requests.
*/
P(jcr->mutex);
jcr->use_count++; /* mark in use by msg thread */
+ jcr->sd_msg_thread_done = false;
+ jcr->SD_msg_chan = 0;
V(jcr->mutex);
+ Dmsg0(100, "Start SD msg_thread.\n");
if ((status=pthread_create(&thid, NULL, msg_thread, (void *)jcr)) != 0) {
- Jmsg1(jcr, M_ABORT, 0, _("Cannot create message thread: %s\n"), strerror(status));
- }
- jcr->SD_msg_chan = thid;
+ berrno be;
+ Jmsg1(jcr, M_ABORT, 0, _("Cannot create message thread: %s\n"), be.strerror(status));
+ }
+ Dmsg0(100, "SD msg_thread started.\n");
+ /* Wait for thread to start */
+ while (jcr->SD_msg_chan == 0) {
+ bmicrosleep(0, 50);
+ }
return 1;
}
-static void msg_thread_cleanup(void *arg)
+extern "C" void msg_thread_cleanup(void *arg)
{
JCR *jcr = (JCR *)arg;
Dmsg0(200, "End msg_thread\n");
- db_end_transaction(jcr->db); /* terminate any open transaction */
+ db_end_transaction(jcr, jcr->db); /* terminate any open transaction */
P(jcr->mutex);
- jcr->msg_thread_done = TRUE;
+ jcr->sd_msg_thread_done = true;
pthread_cond_broadcast(&jcr->term_wait); /* wakeup any waiting threads */
+ jcr->SD_msg_chan = 0;
V(jcr->mutex);
free_jcr(jcr); /* release jcr */
}
* Storage daemon).
* Note, we are running in a separate thread.
*/
-static void *msg_thread(void *arg)
+extern "C" void *msg_thread(void *arg)
{
JCR *jcr = (JCR *)arg;
BSOCK *sd;
uint64_t JobBytes;
int stat;
+ pthread_detach(pthread_self());
+ jcr->SD_msg_chan = pthread_self();
pthread_cleanup_push(msg_thread_cleanup, arg);
- Dmsg0(200, "msg_thread\n");
sd = jcr->store_bsock;
- pthread_detach(pthread_self());
/* Read the Storage daemon's output.
*/
- Dmsg0(200, "Start msg_thread loop\n");
- while ((stat=bget_msg(sd, 0)) > 0) {
+ Dmsg0(100, "Start msg_thread loop\n");
+ while ((stat=bget_dirmsg(sd)) >= 0) {
Dmsg1(200, "<stored: %s", sd->msg);
if (sscanf(sd->msg, Job_start, &Job) == 1) {
continue;
if (sscanf(sd->msg, Job_end, &Job, &JobStatus, &JobFiles,
&JobBytes) == 4) {
jcr->SDJobStatus = JobStatus; /* termination status */
- jcr->JobFiles = JobFiles;
- jcr->JobBytes = JobBytes;
+ jcr->SDJobFiles = JobFiles;
+ jcr->SDJobBytes = JobBytes;
break;
- }
- if (sscanf(sd->msg, Job_status, &Job, &JobStatus) == 2) {
- jcr->SDJobStatus = JobStatus; /* current status */
- continue;
}
}
- if (stat < 0) {
+ if (is_bnet_error(sd)) {
jcr->SDJobStatus = JS_ErrorTerminated;
}
pthread_cleanup_pop(1);
void wait_for_storage_daemon_termination(JCR *jcr)
{
+ int cancel_count = 0;
/* Now wait for Storage daemon to terminate our message thread */
+ set_jcr_job_status(jcr, JS_WaitSD);
P(jcr->mutex);
- jcr->JobStatus = JS_WaitSD;
- while (!jcr->msg_thread_done && !job_cancelled(jcr)) {
+ while (!jcr->sd_msg_thread_done) {
struct timeval tv;
struct timezone tz;
struct timespec timeout;
timeout.tv_sec = tv.tv_sec + 10; /* wait 10 seconds */
Dmsg0(300, "I'm waiting for message thread termination.\n");
pthread_cond_timedwait(&jcr->term_wait, &jcr->mutex, &timeout);
+ if (job_canceled(jcr)) {
+ cancel_count++;
+ }
+ /* Give SD 30 seconds to clean up after cancel */
+ if (cancel_count == 3) {
+ break;
+ }
}
V(jcr->mutex);
- jcr->JobStatus = jcr->SDJobStatus;
+ set_jcr_job_status(jcr, JS_Terminated);
+}
+
+#ifdef needed
+#define MAX_TRIES 30
+#define WAIT_TIME 2
+extern "C" void *device_thread(void *arg)
+{
+ int i;
+ JCR *jcr;
+ DEVICE *dev;
+
+
+ pthread_detach(pthread_self());
+ jcr = new_control_jcr("*DeviceInit*", JT_SYSTEM);
+ for (i=0; i < MAX_TRIES; i++) {
+ if (!connect_to_storage_daemon(jcr, 10, 30, 1)) {
+ Dmsg0(000, "Failed connecting to SD.\n");
+ continue;
+ }
+ LockRes();
+ foreach_res(dev, R_DEVICE) {
+ if (!update_device_res(jcr, dev)) {
+ Dmsg1(900, "Error updating device=%s\n", dev->hdr.name);
+ } else {
+ Dmsg1(900, "Updated Device=%s\n", dev->hdr.name);
+ }
+ }
+ UnlockRes();
+ bnet_close(jcr->store_bsock);
+ jcr->store_bsock = NULL;
+ break;
+
+ }
+ free_jcr(jcr);
+ return NULL;
+}
+
+/*
+ * Start a thread to handle getting Device resource information
+ * from SD. This is called once at startup of the Director.
+ */
+void init_device_resources()
+{
+ int status;
+ pthread_t thid;
+
+ Dmsg0(100, "Start Device thread.\n");
+ if ((status=pthread_create(&thid, NULL, device_thread, NULL)) != 0) {
+ berrno be;
+ Jmsg1(NULL, M_ABORT, 0, _("Cannot create message thread: %s\n"), be.strerror(status));
+ }
}
+#endif
projects / bacula / bacula / blobdiff