2 * This file handles accepting Director Commands
4 * Most Director commands are handled here, with the
5 * exception of the Job command command and subsequent
6 * subcommands that are handled
9 * N.B. in this file, in general we must use P(dev->mutex) rather
10 * than lock_device(dev) so that we can examine the blocked
11 * state rather than blocking ourselves. In some "safe" cases,
12 * we can do things to a blocked device. CAREFUL!!!!
14 * File daemon commands are handled in fdcmd.c
16 * Kern Sibbald, May MMI
22 Copyright (C) 2000, 2001, 2002 Kern Sibbald and John Walker
24 This program is free software; you can redistribute it and/or
25 modify it under the terms of the GNU General Public License as
26 published by the Free Software Foundation; either version 2 of
27 the License, or (at your option) any later version.
29 This program is distributed in the hope that it will be useful,
30 but WITHOUT ANY WARRANTY; without even the implied warranty of
31 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
32 General Public License for more details.
34 You should have received a copy of the GNU General Public
35 License along with this program; if not, write to the Free
36 Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
44 /* Exported variables */
46 /* Imported variables */
47 extern BSOCK *filed_chan;
48 extern int r_first, r_last;
49 extern struct s_res resources[];
50 extern char my_name[];
51 extern time_t daemon_start_time;
52 extern struct s_last_job last_job;
54 /* Static variables */
55 static char derrmsg[] = "3900 Invalid command\n";
56 static char OKsetdebug[] = "3000 OK setdebug=%d\n";
59 /* Imported functions */
60 extern void terminate_child();
61 extern int job_cmd(JCR *jcr);
63 /* Forward referenced functions */
64 static int label_cmd(JCR *jcr);
65 static int setdebug_cmd(JCR *jcr);
66 static int cancel_cmd(JCR *cjcr);
67 static int mount_cmd(JCR *jcr);
68 static int unmount_cmd(JCR *jcr);
69 static int status_cmd(JCR *sjcr);
70 static void label_volume_if_ok(JCR *jcr, DEVICE *dev, char *vname, char *poolname,
75 int (*func)(JCR *jcr);
79 * The following are the recognized commands from the Director.
81 static struct s_cmds cmds[] = {
82 {"JobId=", job_cmd}, /* start Job */
83 {"setdebug=", setdebug_cmd}, /* set debug level */
84 {"cancel", cancel_cmd},
85 {"label", label_cmd}, /* label a tape */
87 {"unmount", unmount_cmd},
88 {"status", status_cmd},
89 {NULL, NULL} /* list terminator */
94 * Connection request. We accept connections either from the
95 * Director or a Client.
97 * Note, we are running as a seperate thread of the Storage daemon.
98 * and it is because a Director has made a connection with
99 * us on the "Message" channel.
101 * Basic tasks done here:
102 * - Create a JCR record
103 * - Authenticate the Director
104 * - We wait for a command
105 * - We execute the command
106 * - We continue or exit depending on the return status
108 void connection_request(void *arg)
110 BSOCK *bs = (BSOCK *)arg;
114 char name[MAX_NAME_LENGTH];
116 if (bnet_recv(bs) <= 0) {
117 Emsg0(M_ERROR, 0, "Connection request failed.\n");
122 * See if this is a File daemon connection
124 if (sscanf(bs->msg, "Hello Start Job %127s calling\n", name) == 1) {
125 handle_filed_connection(bs, name);
129 jcr = new_jcr(sizeof(JCR), stored_free_jcr); /* create Job Control Record */
130 jcr->dir_bsock = bs; /* save Director bsock */
132 Dmsg0(1000, "stored in start_job\n");
135 * Authenticate the Director
137 if (!authenticate_director(jcr)) {
138 Jmsg(jcr, M_FATAL, 0, _("Unable to authenticate Director\n"));
142 Dmsg0(90, "Message channel init completed.\n");
144 for (quit=0; !quit;) {
147 if ((bnet_stat = bnet_recv(bs)) <= 0) {
148 break; /* connection terminated */
150 Dmsg1(9, "<dird: %s\n", bs->msg);
152 for (i=0; cmds[i].cmd; i++) {
153 if (strncmp(cmds[i].cmd, bs->msg, strlen(cmds[i].cmd)) == 0) {
154 if (!cmds[i].func(jcr)) { /* do command */
155 quit = TRUE; /* error, get out */
156 Dmsg1(90, "Command %s requsts quit\n", cmds[i].cmd);
158 found = TRUE; /* indicate command found */
162 if (!found) { /* command not found */
163 bnet_fsend(bs, derrmsg);
168 if (bnet_stat != BNET_TERMINATE) {
169 bnet_sig(bs, BNET_TERMINATE);
176 * Set debug level as requested by the Director
179 static int setdebug_cmd(JCR *jcr)
181 BSOCK *dir = jcr->dir_bsock;
184 Dmsg1(10, "setdebug_cmd: %s", dir->msg);
185 if (sscanf(dir->msg, "setdebug=%d", &level) != 1 || level < 0) {
186 bnet_fsend(dir, "3991 Bad setdebug command: %s\n", dir->msg);
190 return bnet_fsend(dir, OKsetdebug, level);
197 static int cancel_cmd(JCR *cjcr)
199 BSOCK *dir = cjcr->dir_bsock;
201 char Job[MAX_NAME_LENGTH];
204 if (sscanf(dir->msg, "cancel Job=%127s", Job) == 1) {
205 if (!(jcr=get_jcr_by_full_name(Job))) {
206 bnet_fsend(dir, _("3992 Job %s not found.\n"), Job);
209 oldStatus = jcr->JobStatus;
210 jcr->JobStatus = JS_Cancelled;
211 if (!jcr->authenticated && jcr->JobStatus == JS_WaitFD) {
212 pthread_cond_signal(&jcr->job_start_wait); /* wake waiting thread */
215 if (jcr->file_bsock) {
216 bnet_sig(jcr->file_bsock, BNET_TERMINATE);
218 bnet_fsend(dir, _("3000 Job %s Status=%c marked to be cancelled.\n"),
219 jcr->Job, oldStatus);
223 bnet_fsend(dir, _("3993 Error scanning cancel command.\n"));
225 bnet_sig(dir, BNET_EOD);
233 static int label_cmd(JCR *jcr)
235 POOLMEM *dname, *volname, *poolname, *mtype;
236 BSOCK *dir = jcr->dir_bsock;
242 dname = get_memory(dir->msglen+1);
243 volname = get_memory(dir->msglen+1);
244 poolname = get_memory(dir->msglen+1);
245 mtype = get_memory(dir->msglen+1);
246 if (sscanf(dir->msg, "label %s VolumeName=%s PoolName=%s MediaType=%s Slot=%d",
247 dname, volname, poolname, mtype, &slot) == 5) {
248 unbash_spaces(dname);
249 unbash_spaces(volname);
250 unbash_spaces(poolname);
251 unbash_spaces(mtype);
254 while ((device=(DEVRES *)GetNextRes(R_DEVICE, (RES *)device))) {
255 /* Find resource, and make sure we were able to open it */
256 if (strcmp(device->hdr.name, dname) == 0 && device->dev) {
257 Dmsg1(20, "Found device %s\n", device->hdr.name);
264 /******FIXME**** compare MediaTypes */
265 jcr->device = device;
268 P(dev->mutex); /* Use P to avoid indefinite block */
269 if (!(dev->state & ST_OPENED)) {
270 if (open_dev(dev, volname, READ_WRITE) < 0) {
271 bnet_fsend(dir, _("3994 Connot open device: %s\n"), strerror_dev(dev));
273 label_volume_if_ok(jcr, dev, volname, poolname, slot);
274 force_close_dev(dev);
276 /* Under certain "safe" conditions, we can steal the lock */
277 } else if (dev->dev_blocked &&
278 (dev->dev_blocked == BST_UNMOUNTED ||
279 dev->dev_blocked == BST_WAITING_FOR_SYSOP ||
280 dev->dev_blocked == BST_UNMOUNTED_WAITING_FOR_SYSOP)) {
281 label_volume_if_ok(jcr, dev, volname, poolname, slot);
282 } else if (dev->state & ST_READ || dev->num_writers) {
283 if (dev->state & ST_READ) {
284 bnet_fsend(dir, _("3901 Device %s is busy with 1 reader.\n"),
287 bnet_fsend(dir, _("3902 Device %s is busy with %d writer(s).\n"),
288 dev_name(dev), dev->num_writers);
290 } else { /* device not being used */
291 label_volume_if_ok(jcr, dev, volname, poolname, slot);
295 bnet_fsend(dir, _("3999 Device %s not found\n"), dname);
298 /* NB dir->msg gets clobbered in bnet_fsend, so save command */
299 strcpy(dname, dir->msg);
300 bnet_fsend(dir, _("3903 Error scanning label command: %s\n"), dname);
303 free_memory(volname);
304 free_memory(poolname);
306 bnet_sig(dir, BNET_EOD);
311 * Read the tape label and determine if we can safely
312 * label the tape (not a Bacula volume), then label it.
314 * Enter with the mutex set
316 static void label_volume_if_ok(JCR *jcr, DEVICE *dev, char *vname, char *poolname,
319 BSOCK *dir = jcr->dir_bsock;
323 steal_device_lock(dev, &hold, BST_WRITING_LABEL);
325 strcpy(jcr->VolumeName, vname);
326 jcr->VolCatInfo.Slot = slot;
327 autoload_device(jcr, dev, 0, dir); /* autoload if possible */
328 block = new_block(dev);
330 /* See what we have for a Volume */
331 switch (read_dev_volume_label(jcr, dev, block)) {
333 case VOL_VERSION_ERROR:
334 case VOL_LABEL_ERROR:
336 bnet_fsend(dir, _("3901 Cannot label Volume because it is \
337 already labeled: %s\n"), dev->VolHdr.VolName);
341 write_volume_label_to_dev(jcr, jcr->device, vname, poolname);
342 strcpy(jcr->VolumeName, vname);
343 bnet_fsend(dir, _("3000 OK label. Volume=%s Device=%s\n"),
344 vname, dev->dev_name);
347 bnet_fsend(dir, _("3902 Cannot label Volume. \
348 Unknown status %d from read_volume_label()\n"), jcr->label_status);
352 return_device_lock(dev, &hold);
357 * Read the tape label
359 * Enter with the mutex set
361 static int read_label(JCR *jcr, DEVICE *dev)
364 BSOCK *dir = jcr->dir_bsock;
369 steal_device_lock(dev, &hold, BST_DOING_ACQUIRE);
372 jcr->VolumeName[0] = 0;
373 block = new_block(dev);
374 dev->state &= ~ST_LABEL; /* force read of label */
375 switch (read_dev_volume_label(jcr, dev, block)) {
377 bnet_fsend(dir, _("3001 Mounted Volume: %s\n"), dev->VolHdr.VolName);
381 bnet_fsend(dir, _("3902 Cannot mount Volume on Storage Device \"%s\" because:\n%s\n"),
382 dev->dev_name, jcr->errmsg);
388 return_device_lock(dev, &hold);
394 * Mount command from Director
396 static int mount_cmd(JCR *jcr)
399 BSOCK *dir = jcr->dir_bsock;
404 dev_name = (char *) get_memory(dir->msglen);
405 if (sscanf(dir->msg, "mount %s", dev_name) == 1) {
406 unbash_spaces(dev_name);
409 while ((device=(DEVRES *)GetNextRes(R_DEVICE, (RES *)device))) {
410 /* Find resource, and make sure we were able to open it */
411 if (strcmp(device->hdr.name, dev_name) == 0 && device->dev) {
412 Dmsg1(20, "Found device %s\n", device->hdr.name);
419 jcr->device = device;
421 P(dev->mutex); /* Use P to avoid indefinite block */
422 switch (dev->dev_blocked) { /* device blocked? */
424 case BST_WAITING_FOR_SYSOP:
425 /* Someone is waiting, wake him */
426 Dmsg0(90, "Waiting for mount attempt to wake thread\n");
427 pthread_cond_signal(&dev->wait_next_vol);
428 bnet_fsend(dir, "3001 OK mount. Device=%s\n", dev->dev_name);
431 case BST_UNMOUNTED_WAITING_FOR_SYSOP:
433 /* We freed the device, so reopen it and wake any waiting threads */
434 if (open_dev(dev, NULL, READ_WRITE) < 0) {
435 bnet_fsend(dir, _("3901 open device failed: ERR=%s\n"),
439 block = new_block(dev);
440 read_dev_volume_label(jcr, dev, block);
442 if (dev->dev_blocked == BST_UNMOUNTED) {
443 Dmsg0(90, "Unmounted unblocking device\n");
444 read_label(jcr, dev);
447 Dmsg0(90, "Unmounted waiting for mount attempt to wake thread\n");
448 dev->dev_blocked = BST_WAITING_FOR_SYSOP;
449 pthread_cond_signal(&dev->wait_next_vol);
451 if (dev->state & ST_LABEL) {
452 bnet_fsend(dir, _("3001 Device %s is mounted with Volume %s\n"),
453 dev->dev_name, dev->VolHdr.VolName);
455 bnet_fsend(dir, _("3905 Device %s open but no Bacula volume is mounted.\n"),
460 case BST_DOING_ACQUIRE:
461 bnet_fsend(dir, _("3001 Device %s is mounted; doing acquire.\n"),
465 case BST_WRITING_LABEL:
466 bnet_fsend(dir, _("3903 Device %s is being labeled.\n"), dev->dev_name);
469 case BST_NOT_BLOCKED:
470 if (dev->state & ST_OPENED) {
471 if (dev->state & ST_LABEL) {
472 bnet_fsend(dir, _("3001 Device %s is mounted with Volume %s\n"),
473 dev->dev_name, dev->VolHdr.VolName);
475 bnet_fsend(dir, _("3905 Device %s open but no Bacula volume is mounted.\n"),
479 if (!dev_is_tape(dev)) {
480 bnet_fsend(dir, _("3906 cannot mount non-tape.\n"));
483 if (open_dev(dev, NULL, READ_WRITE) < 0) {
484 bnet_fsend(dir, _("3901 open device failed: ERR=%s\n"),
488 read_label(jcr, dev);
489 if (dev->state & ST_LABEL) {
490 bnet_fsend(dir, _("3001 Device %s is mounted with Volume %s\n"),
491 dev->dev_name, dev->VolHdr.VolName);
493 bnet_fsend(dir, _("3905 Device %s open but no Bacula volume is mounted.\n"),
500 bnet_fsend(dir, _("3905 Bizarre wait state %d\n"), dev->dev_blocked);
505 bnet_fsend(dir, _("3999 Device %s not found\n"), dev_name);
508 strcpy(dev_name, dir->msg);
509 bnet_fsend(dir, _("3906 Error scanning mount command: %s\n"), dev_name);
511 free_memory(dev_name);
512 bnet_sig(dir, BNET_EOD);
517 * unmount command from Director
519 static int unmount_cmd(JCR *jcr)
522 BSOCK *dir = jcr->dir_bsock;
527 dname = (char *) get_memory(dir->msglen+1);
528 if (sscanf(dir->msg, "unmount %s", dname) == 1) {
529 unbash_spaces(dname);
532 while ((device=(DEVRES *)GetNextRes(R_DEVICE, (RES *)device))) {
533 /* Find resource, and make sure we were able to open it */
534 if (strcmp(device->hdr.name, dname) == 0 && device->dev) {
535 Dmsg1(20, "Found device %s\n", device->hdr.name);
542 jcr->device = device;
544 P(dev->mutex); /* Use P to avoid indefinite block */
545 if (!(dev->state & ST_OPENED)) {
546 Dmsg0(90, "Device already unmounted\n");
547 bnet_fsend(dir, _("3901 Device %s is already unmounted.\n"), dev_name(dev));
549 } else if (dev->dev_blocked == BST_WAITING_FOR_SYSOP) {
550 Dmsg2(90, "%d waiter dev_block=%d. doing unmount\n", dev->num_waiting,
552 if (dev->capabilities & CAP_OFFLINEUNMOUNT) {
555 force_close_dev(dev);
556 dev->dev_blocked = BST_UNMOUNTED_WAITING_FOR_SYSOP;
557 bnet_fsend(dir, _("3001 Device %s unmounted.\n"), dev_name(dev));
559 } else if (dev->dev_blocked == BST_DOING_ACQUIRE) {
560 bnet_fsend(dir, _("3902 Device %s is busy in acquire.\n"),
563 } else if (dev->dev_blocked == BST_WRITING_LABEL) {
564 bnet_fsend(dir, _("3903 Device %s is being labeled.\n"),
567 } else if (dev->state & ST_READ || dev->num_writers) {
568 if (dev->state & ST_READ) {
569 Dmsg0(90, "Device in read mode\n");
570 bnet_fsend(dir, _("3904 Device %s is busy with 1 reader.\n"),
573 Dmsg1(90, "Device busy with %d writers\n", dev->num_writers);
574 bnet_fsend(dir, _("3905 Device %s is busy with %d writer(s).\n"),
575 dev_name(dev), dev->num_writers);
578 } else { /* device not being used */
579 Dmsg0(90, "Device not in use, unmounting\n");
580 block_device(dev, BST_UNMOUNTED);
581 if (dev->capabilities & CAP_OFFLINEUNMOUNT) {
584 force_close_dev(dev);
585 bnet_fsend(dir, _("3002 Device %s unmounted.\n"), dev_name(dev));
589 bnet_fsend(dir, _("3999 Device %s not found\n"), dname);
592 /* NB dir->msg gets clobbered in bnet_fsend, so save command */
593 strcpy(dname, dir->msg);
594 bnet_fsend(dir, _("3907 Error scanning unmount command: %s\n"), dname);
597 bnet_sig(dir, BNET_EOD);
602 * Status command from Director
604 static int status_cmd(JCR *jcr)
608 int found, bps, sec, bpb;
609 BSOCK *user = jcr->dir_bsock;
610 char dt[MAX_TIME_LENGTH];
611 char b1[30], b2[30], b3[30];
613 bnet_fsend(user, "\n%s Version: " VERSION " (" DATE ")\n", my_name);
614 bstrftime(dt, sizeof(dt), daemon_start_time);
615 bnet_fsend(user, _("Daemon started %s, %d Job%s run.\n"), dt, last_job.NumJobs,
616 last_job.NumJobs == 1 ? "" : "s");
617 if (last_job.NumJobs > 0) {
620 bstrftime(dt, sizeof(dt), last_job.end_time);
621 bnet_fsend(user, _("Last Job %s finished at %s\n"), last_job.Job, dt);
623 jobstatus_to_ascii(last_job.JobStatus, termstat, sizeof(termstat));
624 bnet_fsend(user, _(" Files=%s Bytes=%s Termination Status=%s\n"),
625 edit_uint64_with_commas(last_job.JobFiles, b1),
626 edit_uint64_with_commas(last_job.JobBytes, b2),
631 for (device=NULL; (device=(DEVRES *)GetNextRes(R_DEVICE, (RES *)device)); ) {
634 if (dev->state & ST_OPENED) {
635 if (dev->state & ST_LABEL) {
636 bnet_fsend(user, _("Device %s is mounted with Volume %s\n"),
637 dev_name(dev), dev->VolHdr.VolName);
639 bnet_fsend(user, _("Device %s open but no Bacula volume is mounted.\n"), dev_name(dev));
641 switch (dev->dev_blocked) {
643 bnet_fsend(user, _(" Deviced is blocked. User unmounted.\n"));
645 case BST_UNMOUNTED_WAITING_FOR_SYSOP:
646 bnet_fsend(user, _(" Deviced is blocked. User unmounted during wait for media/mount.\n"));
648 case BST_WAITING_FOR_SYSOP:
649 if (jcr->JobStatus == JS_WaitMount) {
650 bnet_fsend(user, _(" Device is blocked waiting for mount.\n"));
652 bnet_fsend(user, _(" Device is blocked waiting for appendable media.\n"));
655 case BST_DOING_ACQUIRE:
656 bnet_fsend(user, _(" Device is being initialized.\n"));
658 case BST_WRITING_LABEL:
659 bnet_fsend(user, _(" Device is blocked labeling a Volume.\n"));
664 bpb = dev->VolCatInfo.VolCatBlocks;
668 bpb = dev->VolCatInfo.VolCatBytes / bpb;
669 bnet_fsend(user, _(" Total Bytes=%s Blocks=%s Bytes/block=%s\n"),
670 edit_uint64_with_commas(dev->VolCatInfo.VolCatBytes, b1),
671 edit_uint64_with_commas(dev->VolCatInfo.VolCatBlocks, b2),
672 edit_uint64_with_commas(bpb, b3));
673 bnet_fsend(user, _(" Positioned at File=%s Block=%s\n"),
674 edit_uint64_with_commas(dev->file, b1),
675 edit_uint64_with_commas(dev->block_num, b2));
678 bnet_fsend(user, _("Device %s is not open.\n"), dev_name(dev));
686 /* NOTE, we reuse a calling argument jcr. Be warned! */
687 for (jcr=NULL; (jcr=get_next_jcr(jcr)); ) {
688 if (jcr->JobStatus == JS_WaitFD) {
689 bnet_fsend(user, _("%s Job %s waiting for Client connection.\n"),
690 job_type_to_str(jcr->JobType), jcr->Job);
693 bnet_fsend(user, _("%s %s job %s is using device %s\n"),
694 job_level_to_str(jcr->JobLevel),
695 job_type_to_str(jcr->JobType),
696 jcr->Job, jcr->device->device_name);
697 sec = time(NULL) - jcr->run_time;
701 bps = jcr->JobBytes / sec;
702 bnet_fsend(user, _(" Files=%s Bytes=%s Bytes/sec=%s\n"),
703 edit_uint64_with_commas(jcr->JobFiles, b1),
704 edit_uint64_with_commas(jcr->JobBytes, b2),
705 edit_uint64_with_commas(bps, b3));
708 if (jcr->file_bsock) {
709 bnet_fsend(user, " FDReadSeqNo=%" lld " fd=%d\n",
710 jcr->file_bsock->read_seqno, jcr->file_bsock->fd);
712 bnet_fsend(user, " FDSocket closed\n");
716 free_locked_jcr(jcr);
720 bnet_fsend(user, _("No jobs running.\n"));
724 bnet_fsend(user, "\n\n");
725 dump_resource(R_DEVICE, resources[R_DEVICE-r_first].res_head, sendit, user);
727 bnet_fsend(user, "====\n");
729 bnet_sig(user, BNET_EOD);