/*
* Job control and execution for Storage Daemon
*
+ * Kern Sibbald, MM
+ *
* 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
extern uint32_t newVolSessionId();
/* Forward referenced functions */
-static int use_device_cmd(JCR *jcr);
+static bool use_device_cmd(JCR *jcr);
/* Requests from the Director daemon */
-static char jobcmd[] = "JobId=%d job=%127s job_name=%127s client_name=%127s \
-type=%d level=%d FileSet=%127s Allow=";
-static char use_device[] = "use device=%s media_type=%s pool_name=%s pool_type=%s\n";
+static char jobcmd[] = "JobId=%d job=%127s job_name=%127s client_name=%127s "
+ "type=%d level=%d FileSet=%127s NoAttr=%d SpoolAttr=%d FileSetMD5=%127s "
+ "SpoolData=%d WritePartAfterJob=%d";
+static char use_device[] = "use device=%127s media_type=%127s "
+ "pool_name=%127s pool_type=%127s append=%d";
+static char query_device[] = "query device=%127s";
+
/* Responses sent to Director daemon */
static char OKjob[] = "3000 OK Job SDid=%u SDtime=%u Authorization=%s\n";
static char OK_device[] = "3000 OK use device\n";
-static char NO_device[] = "3904 Device %s not available\n";
-static char BAD_use[] = "3903 Bad use command: %s\n";
-static char BAD_job[] = "3905 Bad Job command: %s\n";
-
-
+static char NO_device[] = "3914 Device \"%s\" not in SD Device resources.\n";
+static char BAD_use[] = "3913 Bad use command: %s\n";
+static char BAD_job[] = "3915 Bad Job command: %s\n";
+static char OK_query[] = "3001 OK query append=%d read=%d num_writers=%d "
+ "num_waiting=%d open=%d use_count=%d labeled=%d "
+ "media_type=%s volume_name=%s";
+static char BAD_query[] = "3917 Bad query command: %s\n";
/*
* Director requests us to start a job
* - Return when the connection is terminated or
* there is an error.
*/
-int job_cmd(JCR *jcr)
+bool job_cmd(JCR *jcr)
{
- int JobId, errstat;
+ int JobId;
char auth_key[100];
BSOCK *dir = jcr->dir_bsock;
- POOLMEM *job_name, *client_name, *job, *fileset_name;
- int JobType, level;
- struct timeval tv;
- struct timezone tz;
- struct timespec timeout;
+ POOL_MEM job_name, client_name, job, fileset_name, fileset_md5;
+ int JobType, level, spool_attributes, no_attributes, spool_data, write_part_after_job;
+ JCR *ojcr;
/*
* Get JobId and permissions from Director
*/
-
- Dmsg1(30, "Job_cmd: %s\n", dir->msg);
- job = get_memory(dir->msglen);
- job_name = get_memory(dir->msglen);
- client_name = get_memory(dir->msglen);
- fileset_name = get_memory(dir->msglen);
- if (sscanf(dir->msg, jobcmd, &JobId, job, job_name, client_name,
- &JobType, &level, fileset_name) != 7) {
- bnet_fsend(dir, BAD_job, dir->msg);
- Emsg1(M_FATAL, 0, _("Bad Job Command from Director: %s\n"), dir->msg);
- free_memory(job);
- free_memory(job_name);
- free_memory(client_name);
- free_memory(fileset_name);
- jcr->JobStatus = JS_ErrorTerminated;
- return 0;
+ Dmsg1(100, "Job_cmd: %s\n", dir->msg);
+ if (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) != 12) {
+ pm_strcpy(jcr->errmsg, dir->msg);
+ bnet_fsend(dir, BAD_job, jcr->errmsg);
+ Emsg1(M_FATAL, 0, _("Bad Job Command from Director: %s\n"), jcr->errmsg);
+ set_jcr_job_status(jcr, JS_ErrorTerminated);
+ return false;
+ }
+ /*
+ * 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)(long)ojcr, job.c_str());
+ free_jcr(ojcr);
}
jcr->JobId = JobId;
jcr->VolSessionId = newVolSessionId();
jcr->VolSessionTime = VolSessionTime;
- strcpy(jcr->Job, job);
+ bstrncpy(jcr->Job, job, sizeof(jcr->Job));
unbash_spaces(job_name);
- jcr->job_name = get_memory(strlen(job_name) + 1);
- strcpy(jcr->job_name, 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_memory(strlen(client_name) + 1);
- strcpy(jcr->client_name, 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_memory(strlen(fileset_name) + 1);
- strcpy(jcr->fileset_name, fileset_name);
+ jcr->fileset_name = get_pool_memory(PM_NAME);
+ pm_strcpy(jcr->fileset_name, fileset_name);
jcr->JobType = JobType;
jcr->JobLevel = level;
- free_memory(job);
- free_memory(job_name);
- free_memory(client_name);
- free_memory(fileset_name);
-
- /* Initialize FD start condition variable */
- if ((errstat = pthread_cond_init(&jcr->job_start_wait, NULL)) != 0) {
- Emsg1(M_FATAL, 0, _("Unable to init job cond variable: ERR=%s\n"), strerror(errstat));
- jcr->JobStatus = JS_ErrorTerminated;
- return 0;
- }
- jcr->authenticated = FALSE;
+ jcr->no_attributes = no_attributes;
+ jcr->spool_attributes = spool_attributes;
+ jcr->spool_data = spool_data;
+ 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->authenticated = false;
/*
* Pass back an authorization key for the File daemon
*/
- gettimeofday(&tv, &tz);
- srandom(tv.tv_usec + tv.tv_sec);
- sprintf(auth_key, "%ld", (long)random());
+ make_session_key(auth_key, NULL, 1);
bnet_fsend(dir, OKjob, jcr->VolSessionId, jcr->VolSessionTime, auth_key);
- Dmsg1(10, ">dird: %s", dir->msg);
+ Dmsg1(110, ">dird: %s", dir->msg);
jcr->sd_auth_key = bstrdup(auth_key);
+ memset(auth_key, 0, sizeof(auth_key));
+ return true;
+}
+bool use_cmd(JCR *jcr)
+{
/*
* Wait for the device, media, and pool information
*/
if (!use_device_cmd(jcr)) {
- jcr->JobStatus = JS_ErrorTerminated;
- return 0;
+ set_jcr_job_status(jcr, JS_ErrorTerminated);
+ memset(jcr->sd_auth_key, 0, strlen(jcr->sd_auth_key));
+ return false;
}
+ return true;
+}
- jcr->JobStatus = JS_WaitFD; /* wait for FD to connect */
+bool run_cmd(JCR *jcr)
+{
+ struct timeval tv;
+ struct timezone tz;
+ struct timespec timeout;
+ int errstat;
+
+ /* The following jobs don't need the FD */
+ switch (jcr->JobType) {
+ case JT_MIGRATION:
+ case JT_COPY:
+ case JT_ARCHIVE:
+ jcr->authenticated = true;
+ run_job(jcr);
+ return false;
+ }
+
+ set_jcr_job_status(jcr, JS_WaitFD); /* wait for FD to connect */
dir_send_job_status(jcr);
gettimeofday(&tv, &tz);
- timeout.tv_nsec = tv.tv_usec * 1000;
+ timeout.tv_nsec = tv.tv_usec * 1000;
timeout.tv_sec = tv.tv_sec + 30 * 60; /* wait 30 minutes */
-
- Dmsg1(200, "%s waiting on job_start_wait\n", jcr->Job);
+ Dmsg1(100, "%s waiting on FD to contact SD\n", jcr->Job);
+ /*
+ * Wait for the File daemon to contact us to start the Job,
+ * when he does, we will be released, unless the 30 minutes
+ * expires.
+ */
P(jcr->mutex);
- for ( ;!job_cancelled(jcr); ) {
+ for ( ;!job_canceled(jcr); ) {
errstat = pthread_cond_timedwait(&jcr->job_start_wait, &jcr->mutex, &timeout);
if (errstat == 0 || errstat == ETIMEDOUT) {
break;
}
V(jcr->mutex);
- if (jcr->authenticated && !job_cancelled(jcr)) {
+ memset(jcr->sd_auth_key, 0, strlen(jcr->sd_auth_key));
+
+ if (jcr->authenticated && !job_canceled(jcr)) {
+ Dmsg1(100, "Running job %s\n", jcr->Job);
run_job(jcr); /* Run the job */
}
- return 0;
+ return false;
}
/*
- * This entry point is only called if we have a separate
- * Storage Daemon Data port. Otherwise, the connection
- * is made to the main port, and if it is a File daemon
- * calling, handl_filed_connection() is called directly.
+ * After receiving a connection (in job.c) if it is
+ * from the File daemon, this routine is called.
*/
-void connection_from_filed(void *arg)
-{
- BSOCK *fd = (BSOCK *)arg;
- char job_name[MAX_NAME_LENGTH];
-
- Dmsg0(200, "enter connection_from_filed\n");
- if (bnet_recv(fd) <= 0) {
- Emsg0(M_FATAL, 0, _("Unable to authenticate Client.\n"));
- return;
- }
- Dmsg1(100, "got: %s\n", fd->msg);
-
- if (sscanf(fd->msg, "Hello Start Job %127s\n", job_name) != 1) {
- Emsg1(M_FATAL, 0, _("Authentication failure: %s\n"), fd->msg);
- return;
- }
- handle_filed_connection(fd, job_name);
- return;
-}
-
void handle_filed_connection(BSOCK *fd, char *job_name)
{
JCR *jcr;
+ bmicrosleep(0, 50000); /* wait 50 millisecs */
if (!(jcr=get_jcr_by_full_name(job_name))) {
- Emsg1(M_FATAL, 0, _("Job name not found: %s\n"), job_name);
+ Jmsg1(NULL, M_FATAL, 0, _("Job name not found: %s\n"), job_name);
+ Dmsg1(100, "Job name not found: %s\n", job_name);
return;
}
jcr->file_bsock = fd;
+ jcr->file_bsock->jcr = jcr;
- Dmsg1(100, "Found Job %s\n", job_name);
+ Dmsg1(110, "Found Job %s\n", job_name);
if (jcr->authenticated) {
- Pmsg2(000, "Hey!!!! JobId %d Job %s already authenticated.\n",
+ Jmsg2(jcr, M_FATAL, 0, "Hey!!!! JobId %u Job %s already authenticated.\n",
jcr->JobId, jcr->Job);
+ free_jcr(jcr);
+ return;
}
-
+
/*
* Authenticate the File daemon
*/
Dmsg1(100, "Authentication failed Job %s\n", jcr->Job);
Jmsg(jcr, M_FATAL, 0, _("Unable to authenticate File daemon\n"));
} else {
- jcr->authenticated = TRUE;
- Dmsg1(100, "OK Authentication Job %s\n", jcr->Job);
+ jcr->authenticated = true;
+ Dmsg1(110, "OK Authentication Job %s\n", jcr->Job);
}
P(jcr->mutex);
if (!jcr->authenticated) {
- jcr->JobStatus = JS_ErrorTerminated;
+ set_jcr_job_status(jcr, JS_ErrorTerminated);
}
pthread_cond_signal(&jcr->job_start_wait); /* wake waiting job */
V(jcr->mutex);
}
-/*
+/*
* Use Device command from Director
- * He tells is what Device Name to use, the Media Type,
+ * He tells is what Device Name to use, the Media Type,
* the Pool Name, and the Pool Type.
*
* Ensure that the device exists and is opened, then store
* the media and pool info in the JCR.
*/
-static int use_device_cmd(JCR *jcr)
+static bool use_device_cmd(JCR *jcr)
{
- POOLMEM *dev_name, *media_type, *pool_name, *pool_type;
+ POOL_MEM dev_name, media_type, pool_name, pool_type;
BSOCK *dir = jcr->dir_bsock;
DEVRES *device;
+ int append;
+ bool ok;
- if (bnet_recv(dir) <= 0) {
- Emsg0(M_FATAL, 0, "No Device from Director\n");
- return 0;
- }
-
- Dmsg1(20, "Use device: %s", dir->msg);
- dev_name = get_memory(dir->msglen);
- media_type = get_memory(dir->msglen);
- pool_name = get_memory(dir->msglen);
- pool_type = get_memory(dir->msglen);
- if (sscanf(dir->msg, use_device, dev_name, media_type, pool_name, pool_type) == 4) {
+ Dmsg1(120, "Use device: %s", dir->msg);
+ /*
+ * If there are multiple devices, the director sends us
+ * use_device for each device that it wants to use.
+ */
+ ok = sscanf(dir->msg, use_device, dev_name.c_str(), media_type.c_str(),
+ pool_name.c_str(), pool_type.c_str(), &append) == 5;
+ if (ok) {
unbash_spaces(dev_name);
unbash_spaces(media_type);
unbash_spaces(pool_name);
unbash_spaces(pool_type);
- device = NULL;
LockRes();
- while ((device=(DEVRES *)GetNextRes(R_DEVICE, (RES *)device))) {
+ foreach_res(device, R_DEVICE) {
/* Find resource, and make sure we were able to open it */
- if (strcmp(device->hdr.name, dev_name) == 0 && device->dev) {
- Dmsg1(20, "Found device %s\n", device->hdr.name);
- jcr->pool_name = get_memory(strlen(pool_name) + 1);
- strcpy(jcr->pool_name, pool_name);
- jcr->pool_type = get_memory(strlen(pool_type) + 1);
- strcpy(jcr->pool_type, pool_type);
- jcr->media_type = get_memory(strlen(media_type) + 1);
- strcpy(jcr->media_type, media_type);
- jcr->dev_name = get_memory(strlen(dev_name) + 1);
- strcpy(jcr->dev_name, dev_name);
- jcr->device = device;
- Dmsg4(20, use_device, dev_name, media_type, pool_name, pool_type);
- free_memory(dev_name);
- free_memory(media_type);
- free_memory(pool_name);
- free_memory(pool_type);
+ if (fnmatch(dev_name.c_str(), device->hdr.name, 0) == 0 &&
+ device->dev && strcmp(device->media_type, media_type.c_str()) == 0) {
+ const int name_len = MAX_NAME_LENGTH;
+ DCR *dcr = new_dcr(jcr, device->dev);
UnlockRes();
+ Dmsg1(120, "Found device %s\n", device->hdr.name);
+ bstrncpy(dcr->pool_name, pool_name, name_len);
+ bstrncpy(dcr->pool_type, pool_type, name_len);
+ bstrncpy(dcr->media_type, media_type, name_len);
+ bstrncpy(dcr->dev_name, dev_name, name_len);
+ jcr->dcr = dcr;
+ if (!dcr) {
+ bnet_fsend(dir, _("Could not get dcr for device: %s\n"), dev_name.c_str());
+ return false;
+ }
+ if (append == SD_APPEND) {
+ ok = reserve_device_for_append(jcr, device->dev);
+ } else {
+ ok = reserve_device_for_read(jcr, device->dev);
+ }
+ if (!ok) {
+ bnet_fsend(dir, _("Could not get dcr for device: %s\n"), dev_name.c_str());
+ free_dcr(jcr->dcr);
+ return false;
+ }
+ Dmsg1(220, "Got: %s", dir->msg);
return bnet_fsend(dir, OK_device);
}
}
UnlockRes();
- Jmsg(jcr, M_FATAL, 0, _("Requested device %s not found. Cannot continue.\n"),
- dev_name);
- bnet_fsend(dir, NO_device, dev_name);
+ if (verbose) {
+ unbash_spaces(dir->msg);
+ pm_strcpy(jcr->errmsg, dir->msg);
+ Jmsg(jcr, M_INFO, 0, _("Failed command: %s\n"), jcr->errmsg);
+ }
+ Jmsg(jcr, M_FATAL, 0, _("\n"
+ " Device \"%s\" with MediaType \"%s\" requested by DIR not found in SD Device resources.\n"),
+ dev_name.c_str(), media_type.c_str());
+ bnet_fsend(dir, NO_device, dev_name.c_str());
} else {
- Jmsg(jcr, M_FATAL, 0, _("store<dir: Bad Use Device command: %s\n"), dir->msg);
- bnet_fsend(dir, BAD_use, dir->msg);
+ unbash_spaces(dir->msg);
+ pm_strcpy(jcr->errmsg, dir->msg);
+ if (verbose) {
+ Jmsg(jcr, M_INFO, 0, _("Failed command: %s\n"), jcr->errmsg);
+ }
+ Jmsg(jcr, M_FATAL, 0, _("Bad Use Device command: %s\n"), jcr->errmsg);
+ bnet_fsend(dir, BAD_use, jcr->errmsg);
}
- free_memory(dev_name);
- free_memory(media_type);
- free_memory(pool_name);
- free_memory(pool_type);
- return 0; /* ERROR return */
+ return false; /* ERROR return */
+}
+
+/*
+ * Query Device command from Director
+ * Sends Storage Daemon's information on the device to the
+ * caller (presumably the Director).
+ * This command always returns "true" so that the line is
+ * not closed on an error.
+ *
+ */
+bool query_cmd(JCR *jcr)
+{
+ POOL_MEM dev_name;
+ BSOCK *dir = jcr->dir_bsock;
+ DEVRES *device;
+ bool ok;
+
+ Dmsg1(120, "Query: %s", dir->msg);
+
+ ok = sscanf(dir->msg, query_device, dev_name.c_str()) == 1;
+ if (ok) {
+ unbash_spaces(dev_name);
+ LockRes();
+ foreach_res(device, R_DEVICE) {
+ /* Find resource, and make sure we were able to open it */
+ if (fnmatch(dev_name.c_str(), device->hdr.name, 0) == 0 &&
+ device->dev) {
+ DEVICE *dev = device->dev;
+ POOL_MEM VolumeName, MediaType;
+ UnlockRes();
+ if (dev->is_labeled()) {
+ pm_strcpy(VolumeName, dev->VolHdr.VolName);
+ } else {
+ pm_strcpy(VolumeName, "");
+ }
+ bash_spaces(VolumeName);
+ pm_strcpy(MediaType, device->media_type);
+ bash_spaces(MediaType);
+ return bnet_fsend(dir, OK_query, dev->can_append()!=0,
+ dev->can_read()!=0, dev->num_writers, dev->num_waiting,
+ dev->is_open()!=0, dev->use_count, dev->is_labeled()!=0,
+ MediaType.c_str(), VolumeName.c_str());
+ }
+ }
+ UnlockRes();
+ unbash_spaces(dir->msg);
+ pm_strcpy(jcr->errmsg, dir->msg);
+ bnet_fsend(dir, NO_device, dev_name.c_str());
+ } else {
+ unbash_spaces(dir->msg);
+ pm_strcpy(jcr->errmsg, dir->msg);
+ bnet_fsend(dir, BAD_query, jcr->errmsg);
+ }
+ return true;
}
-/*
+
+/*
* Destroy the Job Control Record and associated
* resources (sockets).
*/
-void stored_free_jcr(JCR *jcr)
+void stored_free_jcr(JCR *jcr)
{
if (jcr->file_bsock) {
bnet_close(jcr->file_bsock);
- }
- if (jcr->pool_name) {
- free_memory(jcr->pool_name);
- }
- if (jcr->pool_type) {
- free_memory(jcr->pool_type);
- }
- if (jcr->media_type) {
- free_memory(jcr->media_type);
- }
- if (jcr->dev_name) {
- free_memory(jcr->dev_name);
+ jcr->file_bsock = NULL;
}
if (jcr->job_name) {
free_pool_memory(jcr->job_name);
if (jcr->fileset_name) {
free_memory(jcr->fileset_name);
}
+ if (jcr->fileset_md5) {
+ free_memory(jcr->fileset_md5);
+ }
if (jcr->bsr) {
free_bsr(jcr->bsr);
+ jcr->bsr = NULL;
}
if (jcr->RestoreBootstrap) {
unlink(jcr->RestoreBootstrap);
free_pool_memory(jcr->RestoreBootstrap);
jcr->RestoreBootstrap = NULL;
}
-
+ if (jcr->next_dev || jcr->prev_dev) {
+ Emsg0(M_FATAL, 0, _("In free_jcr(), but still attached to device!!!!\n"));
+ }
+ pthread_cond_destroy(&jcr->job_start_wait);
+ if (jcr->dcr) {
+ free_dcr(jcr->dcr);
+ jcr->dcr = NULL;
+ }
return;
}
projects / bacula / bacula / blobdiff