/*
* After receiving a connection (in dircmd.c) if it is
* from the Storage daemon, this routine is called.
*/
void *handle_stored_connection(BSOCK *sd, char *job_name)
{
JCR *jcr;
/*
* With the following bmicrosleep on, running the
* SD under the debugger fails.
*/
// bmicrosleep(0, 50000); /* wait 50 millisecs */
if (!(jcr = get_jcr_by_full_name(job_name))) {
Jmsg1(NULL, M_FATAL, 0, _("SD connect failed: Job name not found: %s\n"), job_name);
Dmsg1(3, "**** Job \"%s\" not found.\n", job_name);
sd->close();
delete sd;
return NULL;
}
Dmsg1(50, "Found Job %s\n", job_name);
if (jcr->authenticated) {
Jmsg2(jcr, M_FATAL, 0, _("Hey!!!! JobId %u Job %s already authenticated.\n"),
(uint32_t)jcr->JobId, jcr->Job);
Dmsg2(50, "Hey!!!! JobId %u Job %s already authenticated.\n",
(uint32_t)jcr->JobId, jcr->Job);
sd->close();
delete sd;
free_jcr(jcr);
return NULL;
}
jcr->store_bsock = sd;
jcr->store_bsock->set_jcr(jcr);
/*
* Authenticate the Storage daemon
*/
if (jcr->authenticated || !authenticate_storagedaemon(jcr)) {
Dmsg1(50, "Authentication failed Job %s\n", jcr->Job);
Jmsg(jcr, M_FATAL, 0, _("Unable to authenticate Storage daemon\n"));
} else {
jcr->authenticated = true;
Dmsg2(50, "OK Authentication jid=%u Job %s\n", (uint32_t)jcr->JobId, jcr->Job);
}
if (!jcr->authenticated) {
jcr->setJobStatus(JS_ErrorTerminated);
}
pthread_cond_signal(&jcr->job_start_wait); /* wake waiting job */
free_jcr(jcr);
return NULL;
}
/*
* After receiving a connection (in dircmd.c) if it is
* from the File daemon, this routine is called.
*/
void handle_filed_connection(BSOCK *fd, char *job_name, int fd_version,
int sd_version)
{
JCR *jcr;
if (!(jcr=get_jcr_by_full_name(job_name))) {
Jmsg1(NULL, M_FATAL, 0, _("FD connect failed: Job name not found: %s\n"), job_name);
Dmsg1(3, "**** Job \"%s\" not found.\n", job_name);
fd->destroy();
return;
}
Dmsg1(100, "Found Filed Job %s\n", job_name);
if (jcr->authenticated) {
Jmsg2(jcr, M_FATAL, 0, _("Hey!!!! JobId %u Job %s already authenticated.\n"),
(uint32_t)jcr->JobId, jcr->Job);
Dmsg2(050, "Hey!!!! JobId %u Job %s already authenticated.\n",
(uint32_t)jcr->JobId, jcr->Job);
fd->destroy();
free_jcr(jcr);
return;
}
jcr->file_bsock = fd;
jcr->file_bsock->set_jcr(jcr);
jcr->FDVersion = fd_version;
jcr->SDVersion = sd_version;
Dmsg2(050, "fd_version=%d sd_version=%d\n", fd_version, sd_version);
/*
* Authenticate the File daemon
*/
if (jcr->authenticated || !authenticate_filed(jcr)) {
Dmsg1(50, "Authentication failed Job %s\n", jcr->Job);
Jmsg(jcr, M_FATAL, 0, _("Unable to authenticate File daemon\n"));
} else {
jcr->authenticated = true;
Dmsg2(050, "OK Authentication jid=%u Job %s\n", (uint32_t)jcr->JobId, jcr->Job);
}
if (!jcr->authenticated) {
jcr->setJobStatus(JS_ErrorTerminated);
}
Dmsg3(050, "=== Auth OK, unblock Job %s jid=%d sd_ver=%d\n", job_name, jcr->JobId, sd_version);
if (sd_version > 0) {
jcr->sd_client = true;
}
pthread_cond_signal(&jcr->job_start_wait); /* wake waiting job */
free_jcr(jcr);
return;
}
/*
* Director requests us to start a job
* Basic tasks done here:
* - We pickup the JobId to be run from the Director.
* - We pickup the device, media, and pool from the Director
* - Wait for a connection from the File Daemon (FD)
* - Accept commands from the FD (i.e. run the job)
* - Return when the connection is terminated or
* there is an error.
*/
bool job_cmd(JCR *jcr)
{
int32_t JobId;
char auth_key[MAX_NAME_LENGTH];
char seed[MAX_NAME_LENGTH];
char spool_size[MAX_NAME_LENGTH];
BSOCK *dir = jcr->dir_bsock;
POOL_MEM job_name, client_name, job, fileset_name, fileset_md5, backup_format;
int32_t JobType, level, spool_attributes, no_attributes, spool_data;
int32_t PreferMountedVols, rerunning, protocol, dumplevel;
int status;
uint64_t quota = 0;
JCR *ojcr;
/*
* Get JobId and permissions from Director
*/
Dmsg1(100, "<dird: %s", dir->msg);
bstrncpy(spool_size, "0", sizeof(spool_size));
status = sscanf(dir->msg, jobcmd, &JobId, job.c_str(), job_name.c_str(),
client_name.c_str(), &JobType, &level, fileset_name.c_str(),
&no_attributes, &spool_attributes, fileset_md5.c_str(),
&spool_data, &PreferMountedVols, spool_size, &rerunning,
&jcr->VolSessionId, &jcr->VolSessionTime, "a, &protocol,
backup_format.c_str(), &dumplevel);
if (status != 20) {
pm_strcpy(jcr->errmsg, dir->msg);
dir->fsend(BAD_job, status, jcr->errmsg);
Dmsg1(100, ">dird: %s", dir->msg);
jcr->setJobStatus(JS_ErrorTerminated);
return false;
}
jcr->rerunning = (rerunning) ? true : false;
jcr->setJobProtocol(protocol);
Dmsg4(100, "rerunning=%d VolSesId=%d VolSesTime=%d Protocol=%d\n",
jcr->rerunning, jcr->VolSessionId, jcr->VolSessionTime, jcr->getJobProtocol());
/*
* Since this job could be rescheduled, we
* check to see if we have it already. If so
* free the old jcr and use the new one.
*/
ojcr = get_jcr_by_full_name(job.c_str());
if (ojcr && !ojcr->authenticated) {
Dmsg2(100, "Found ojcr=0x%x Job %s\n", (unsigned)(intptr_t)ojcr, job.c_str());
free_jcr(ojcr);
}
jcr->JobId = JobId;
Dmsg2(800, "Start JobId=%d %p\n", JobId, jcr);
/*
* If job rescheduled because previous was incomplete,
* the Resched flag is set and VolSessionId and VolSessionTime
* are given to us (same as restarted job).
*/
if (!jcr->rerunning) {
jcr->VolSessionId = newVolSessionId();
jcr->VolSessionTime = VolSessionTime;
}
bstrncpy(jcr->Job, job, sizeof(jcr->Job));
unbash_spaces(job_name);
jcr->job_name = get_pool_memory(PM_NAME);
pm_strcpy(jcr->job_name, job_name);
unbash_spaces(client_name);
jcr->client_name = get_pool_memory(PM_NAME);
pm_strcpy(jcr->client_name, client_name);
unbash_spaces(fileset_name);
jcr->fileset_name = get_pool_memory(PM_NAME);
pm_strcpy(jcr->fileset_name, fileset_name);
jcr->setJobType(JobType);
jcr->setJobLevel(level);
jcr->no_attributes = no_attributes;
jcr->spool_attributes = spool_attributes;
jcr->spool_data = spool_data;
jcr->spool_size = str_to_int64(spool_size);
jcr->fileset_md5 = get_pool_memory(PM_NAME);
pm_strcpy(jcr->fileset_md5, fileset_md5);
jcr->PreferMountedVols = PreferMountedVols;
jcr->RemainingQuota = quota;
unbash_spaces(backup_format);
jcr->backup_format = get_pool_memory(PM_NAME);
pm_strcpy(jcr->backup_format, backup_format);
jcr->DumpLevel = dumplevel;
jcr->authenticated = false;
Dmsg1(50, "Quota set as %llu\n", quota);
/*
* Pass back an authorization key for the File daemon
*/
bsnprintf(seed, sizeof(seed), "%p%d", jcr, JobId);
make_session_key(auth_key, seed, 1);
dir->fsend(OK_job, jcr->VolSessionId, jcr->VolSessionTime, auth_key);
Dmsg2(50, ">dird jid=%u: %s", (uint32_t)jcr->JobId, dir->msg);
jcr->sd_auth_key = bstrdup(auth_key);
memset(auth_key, 0, sizeof(auth_key));
dispatch_new_plugin_options(jcr);
generate_plugin_event(jcr, bsdEventJobStart, (void *)"JobStart");
return true;
}
void *handle_stored_connection(BSOCK *sd)
{
JCR *jcr;
char job_name[MAX_NAME_LENGTH];
/*
* Do a sanity check on the message received
*/
if (sd->msglen < 25 || sd->msglen > 256) {
Dmsg1(000, "<filed: %s", sd->msg);
Emsg2(M_ERROR, 0, _("Invalid connection from %s. Len=%d\n"), sd->who(), sd->msglen);
bmicrosleep(5, 0); /* make user wait 5 seconds */
sd->close();
return NULL;
}
if (sscanf(sd->msg, "Hello Storage calling Start Job %127s", job_name) != 1) {
char addr[64];
char *who = bnet_get_peer(sd, addr, sizeof(addr)) ? sd->who() : addr;
sd->msg[100] = 0;
Dmsg2(dbglvl, "Bad Hello command from Director at %s: %s\n", sd->who(), sd->msg);
Jmsg2(NULL, M_FATAL, 0, _("Bad Hello command from Director at %s: %s\n"), who, sd->msg);
sd->close();
return NULL;
}
if (!(jcr = get_jcr_by_full_name(job_name))) {
Jmsg1(NULL, M_FATAL, 0, _("SD connect failed: Job name not found: %s\n"), job_name);
Dmsg1(3, "**** Job \"%s\" not found.\n", job_name);
sd->close();
return NULL;
}
Dmsg1(50, "Found Job %s\n", job_name);
jcr->store_bsock = sd;
jcr->store_bsock->set_jcr(jcr);
/*
* Authenticate the Storage Daemon.
*/
if (!authenticate_storagedaemon(jcr)) {
Dmsg1(50, "Authentication failed Job %s\n", jcr->Job);
Jmsg(jcr, M_FATAL, 0, _("Unable to authenticate File daemon\n"));
jcr->setJobStatus(JS_ErrorTerminated);
} else {
Dmsg2(50, "OK Authentication jid=%u Job %s\n", (uint32_t)jcr->JobId, jcr->Job);
}
if (!jcr->max_bandwidth) {
if (jcr->director->max_bandwidth_per_job) {
jcr->max_bandwidth = jcr->director->max_bandwidth_per_job;
} else if (me->max_bandwidth_per_job) {
jcr->max_bandwidth = me->max_bandwidth_per_job;
}
}
sd->set_bwlimit(jcr->max_bandwidth);
if (me->allow_bw_bursting) {
sd->set_bwlimit_bursting();
}
free_jcr(jcr);
return NULL;
}
/*
* Director requests us to start a job
* Basic tasks done here:
* - We pickup the JobId to be run from the Director.
* - We pickup the device, media, and pool from the Director
* - Wait for a connection from the File Daemon (FD)
* - Accept commands from the FD (i.e. run the job)
* - Return when the connection is terminated or
* there is an error.
*/
bool job_cmd(JCR *jcr)
{
int32_t JobId;
char sd_auth_key[200];
char spool_size[30];
char seed[100];
BSOCK *dir = jcr->dir_bsock;
POOL_MEM job_name, client_name, job, fileset_name, fileset_md5;
int32_t JobType, level, spool_attributes, no_attributes, spool_data;
int32_t write_part_after_job, PreferMountedVols;
int32_t rerunning;
int32_t is_client;
int stat;
JCR *ojcr;
/*
* Get JobId and permissions from Director
*/
Dmsg1(100, "<dird: %s", dir->msg);
bstrncpy(spool_size, "0", sizeof(spool_size));
stat = sscanf(dir->msg, jobcmd, &JobId, job.c_str(), job_name.c_str(),
client_name.c_str(),
&JobType, &level, fileset_name.c_str(), &no_attributes,
&spool_attributes, fileset_md5.c_str(), &spool_data,
&write_part_after_job, &PreferMountedVols, spool_size,
&rerunning, &jcr->VolSessionId, &jcr->VolSessionTime,
&is_client, &sd_auth_key);
if (stat != 19) {
pm_strcpy(jcr->errmsg, dir->msg);
dir->fsend(BAD_job, stat, jcr->errmsg);
Dmsg1(100, ">dird: %s", dir->msg);
jcr->setJobStatus(JS_ErrorTerminated);
return false;
}
jcr->rerunning = rerunning;
jcr->sd_client = is_client;
if (is_client) {
jcr->sd_auth_key = bstrdup(sd_auth_key);
}
Dmsg3(100, "rerunning=%d VolSesId=%d VolSesTime=%d\n", jcr->rerunning,
jcr->VolSessionId, jcr->VolSessionTime);
/*
* Since this job could be rescheduled, we
* check to see if we have it already. If so
* free the old jcr and use the new one.
*/
ojcr = get_jcr_by_full_name(job.c_str());
if (ojcr && !ojcr->authenticated) {
Dmsg2(100, "Found ojcr=0x%x Job %s\n", (unsigned)(intptr_t)ojcr, job.c_str());
free_jcr(ojcr);
}
jcr->JobId = JobId;
Dmsg2(800, "Start JobId=%d %p\n", JobId, jcr);
set_jcr_in_tsd(jcr);
/*
* If job rescheduled because previous was incomplete,
* the Resched flag is set and VolSessionId and VolSessionTime
* are given to us (same as restarted job).
*/
if (!jcr->rerunning) {
jcr->VolSessionId = newVolSessionId();
jcr->VolSessionTime = VolSessionTime;
}
bstrncpy(jcr->Job, job, sizeof(jcr->Job));
unbash_spaces(job_name);
jcr->job_name = get_pool_memory(PM_NAME);
pm_strcpy(jcr->job_name, job_name);
unbash_spaces(client_name);
jcr->client_name = get_pool_memory(PM_NAME);
pm_strcpy(jcr->client_name, client_name);
unbash_spaces(fileset_name);
jcr->fileset_name = get_pool_memory(PM_NAME);
pm_strcpy(jcr->fileset_name, fileset_name);
jcr->setJobType(JobType);
jcr->setJobLevel(level);
jcr->no_attributes = no_attributes;
jcr->spool_attributes = spool_attributes;
jcr->spool_data = spool_data;
jcr->spool_size = str_to_int64(spool_size);
jcr->write_part_after_job = write_part_after_job;
jcr->fileset_md5 = get_pool_memory(PM_NAME);
pm_strcpy(jcr->fileset_md5, fileset_md5);
jcr->PreferMountedVols = PreferMountedVols;
jcr->authenticated = false;
/*
* Pass back an authorization key for the File daemon
*/
if (jcr->sd_client) {
bstrncpy(sd_auth_key, "xxx", 3);
} else {
bsnprintf(seed, sizeof(seed), "%p%d", jcr, JobId);
make_session_key(sd_auth_key, seed, 1);
}
dir->fsend(OKjob, jcr->VolSessionId, jcr->VolSessionTime, sd_auth_key);
Dmsg2(150, ">dird jid=%u: %s", (uint32_t)jcr->JobId, dir->msg);
/* If not client, set key, otherwise it is already set */
if (!jcr->sd_client) {
jcr->sd_auth_key = bstrdup(sd_auth_key);
memset(sd_auth_key, 0, sizeof(sd_auth_key));
}
new_plugins(jcr); /* instantiate the plugins */
generate_daemon_event(jcr, "JobStart");
generate_plugin_event(jcr, bsdEventJobStart, (void *)"JobStart");
return true;
}
