2 * Routines to acquire and release a device for read/write
4 * Kern Sibbald, August MMII
9 Copyright (C) 2002-2005 Kern Sibbald
11 This program is free software; you can redistribute it and/or
12 modify it under the terms of the GNU General Public License as
13 published by the Free Software Foundation; either version 2 of
14 the License, or (at your option) any later version.
16 This program is distributed in the hope that it will be useful,
17 but WITHOUT ANY WARRANTY; without even the implied warranty of
18 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 General Public License for more details.
21 You should have received a copy of the GNU General Public
22 License along with this program; if not, write to the Free
23 Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
28 #include "bacula.h" /* pull in global headers */
29 #include "stored.h" /* pull in Storage Deamon headers */
31 static int can_reserve_drive(DCR *dcr);
34 * Create a new Device Control Record and attach
35 * it to the device (if this is a real job).
37 DCR *new_dcr(JCR *jcr, DEVICE *dev)
39 if (jcr && jcr->dcr) {
42 DCR *dcr = (DCR *)malloc(sizeof(DCR));
43 memset(dcr, 0, sizeof(DCR));
50 dcr->device = dev->device;
52 dcr->block = new_block(dev);
53 dcr->rec = new_record();
55 dcr->max_spool_size = dev->device->max_spool_size;
56 /* Attach this dcr only if dev is initialized */
57 if (dev->fd != 0 && jcr && jcr->JobType != JT_SYSTEM) {
58 dev->attached_dcrs->append(dcr); /* attach dcr to device */
59 // jcr->dcrs->append(dcr); /* put dcr in list for Job */
65 * Search the dcrs list for the given dcr. If it is found,
66 * as it should be, then remove it. Also zap the jcr pointer
67 * to the dcr if it is the same one.
70 static void remove_dcr_from_dcrs(DCR *dcr)
76 int num = jcr->dcrs->size();
77 for (i=0; i < num; i++) {
78 ldcr = (DCR *)jcr->dcrs->get(i);
81 if (jcr->dcr == dcr) {
91 * Free up all aspects of the given dcr -- i.e. dechain it,
92 * release allocated memory, zap pointers, ...
94 void free_dcr(DCR *dcr)
97 DEVICE *dev = dcr->dev;
100 * If we reserved the device, we must decrement the
103 if (dcr->reserved_device) {
106 if (dev->num_writers < 0) {
107 Jmsg1(dcr->jcr, M_ERROR, 0, _("Hey! num_writers=%d!!!!\n"), dev->num_writers);
108 dev->num_writers = 0;
109 dcr->reserved_device = false;
114 /* Detach this dcr only if the dev is initialized */
115 if (dev->fd != 0 && jcr && jcr->JobType != JT_SYSTEM) {
116 dev->attached_dcrs->remove(dcr); /* detach dcr from device */
117 // remove_dcr_from_dcrs(dcr); /* remove dcr from jcr list */
120 free_block(dcr->block);
123 free_record(dcr->rec);
126 dcr->jcr->dcr = NULL;
133 * We "reserve" the drive by setting the ST_READ bit. No one else
134 * should touch the drive until that is cleared.
135 * This allows the DIR to "reserve" the device before actually
136 * starting the job. If the device is not available, the DIR
137 * can wait (to be implemented 1/05).
139 bool reserve_device_for_read(DCR *dcr)
141 DEVICE *dev = dcr->dev;
147 init_device_wait_timers(dcr);
149 dev->block(BST_DOING_ACQUIRE);
151 Mmsg(jcr->errmsg, _("Device %s is BLOCKED due to user unmount.\n"),
153 for (first=true; device_is_unmounted(dev); first=false) {
155 if (!wait_for_device(dcr, jcr->errmsg, first)) {
158 dev->block(BST_DOING_ACQUIRE);
161 Mmsg2(jcr->errmsg, _("Device %s is busy. Job %d canceled.\n"),
162 dev->print_name(), jcr->JobId);
163 for (first=true; dev->is_busy(); first=false) {
165 if (!wait_for_device(dcr, jcr->errmsg, first)) {
168 dev->block(BST_DOING_ACQUIRE);
178 /*********************************************************************
179 * Acquire device for reading.
180 * The drive should have previously been reserved by calling
181 * reserve_device_for_read(). We read the Volume label from the block and
182 * leave the block pointers just after the label.
184 * Returns: NULL if failed for any reason
187 DCR *acquire_device_for_read(DCR *dcr)
189 DEVICE *dev = dcr->dev;
192 bool tape_previously_mounted;
193 bool tape_initially_mounted;
195 bool try_autochanger = true;
197 int vol_label_status;
199 dev->block(BST_DOING_ACQUIRE);
201 if (dev->num_writers > 0) {
202 Jmsg2(jcr, M_FATAL, 0, _("Num_writers=%d not zero. Job %d canceled.\n"),
203 dev->num_writers, jcr->JobId);
207 /* Find next Volume, if any */
210 Jmsg(jcr, M_FATAL, 0, _("No volumes specified. Job %d canceled.\n"), jcr->JobId);
214 for (i=1; i<jcr->CurVolume; i++) {
218 goto get_out; /* should not happen */
220 bstrncpy(dcr->VolumeName, vol->VolumeName, sizeof(dcr->VolumeName));
222 init_device_wait_timers(dcr);
224 tape_previously_mounted = dev->can_read() ||
227 tape_initially_mounted = tape_previously_mounted;
230 /* Volume info is always needed because of VolParts */
231 Dmsg0(200, "dir_get_volume_info\n");
232 if (!dir_get_volume_info(dcr, GET_VOL_INFO_FOR_READ)) {
233 Jmsg1(jcr, M_WARNING, 0, "%s", jcr->errmsg);
236 dev->num_parts = dcr->VolCatInfo.VolCatParts;
238 for (i=0; i<5; i++) {
239 dev->clear_labeled(); /* force reread of label */
240 if (job_canceled(jcr)) {
241 Mmsg1(dev->errmsg, _("Job %d canceled.\n"), jcr->JobId);
242 goto get_out; /* error return */
245 * This code ensures that the device is ready for
246 * reading. If it is a file, it opens it.
247 * If it is a tape, it checks the volume name
249 for ( ; !dev->is_open(); ) {
250 Dmsg1(120, "bstored: open vol=%s\n", dcr->VolumeName);
251 if (open_dev(dev, dcr->VolumeName, OPEN_READ_ONLY) < 0) {
252 if (dev->dev_errno == EIO) { /* no tape loaded */
253 Jmsg3(jcr, M_WARNING, 0, _("Open device %s Volume \"%s\" failed: ERR=%s\n"),
254 dev->print_name(), dcr->VolumeName, strerror_dev(dev));
258 /* If we have a dvd that requires mount,
259 * we need to try to open the label, so the info can be reported
260 * if a wrong volume has been mounted. */
261 if (dev->is_dvd() && (dcr->VolCatInfo.VolCatParts > 0)) {
265 Jmsg3(jcr, M_FATAL, 0, _("Open device %s Volume \"%s\" failed: ERR=%s\n"),
266 dev->print_name(), dcr->VolumeName, strerror_dev(dev));
269 Dmsg1(129, "open_dev %s OK\n", dev->print_name());
273 vol_label_status = read_dev_volume_label_guess(dcr, 0);
275 vol_label_status = read_dev_volume_label(dcr);
278 Dmsg0(200, "calling read-vol-label\n");
279 switch (vol_label_status) {
282 memcpy(&dev->VolCatInfo, &dcr->VolCatInfo, sizeof(dev->VolCatInfo));
286 * Send error message generated by read_dev_volume_label()
287 * only we really had a tape mounted. This supresses superfluous
288 * error messages when nothing is mounted.
290 if (tape_previously_mounted) {
291 Jmsg(jcr, M_WARNING, 0, "%s", jcr->errmsg);
295 if (tape_initially_mounted) {
296 tape_initially_mounted = false;
301 Jmsg1(jcr, M_WARNING, 0, "%s", jcr->errmsg);
303 tape_previously_mounted = true;
305 /* If the device requires mount, close it, so the device can be ejected.
306 * FIXME: This should perhaps be done for all devices. */
307 if (dev_cap(dev, CAP_REQMOUNT)) {
308 force_close_dev(dev);
311 /* Call autochanger only once unless ask_sysop called */
312 if (try_autochanger) {
314 Dmsg2(200, "calling autoload Vol=%s Slot=%d\n",
315 dcr->VolumeName, dcr->VolCatInfo.Slot);
316 stat = autoload_device(dcr, 0, NULL);
318 try_autochanger = false;
319 continue; /* try reading volume mounted */
323 /* Mount a specific volume and no other */
324 Dmsg0(200, "calling dir_ask_sysop\n");
325 if (!dir_ask_sysop_to_mount_volume(dcr)) {
326 goto get_out; /* error return */
328 try_autochanger = true; /* permit using autochanger again */
329 continue; /* try reading again */
334 Jmsg1(jcr, M_FATAL, 0, _("Too many errors trying to mount device %s.\n"),
341 set_jcr_job_status(jcr, JS_Running);
342 dir_send_job_status(jcr);
343 Jmsg(jcr, M_INFO, 0, _("Ready to read from volume \"%s\" on device %s.\n"),
344 dcr->VolumeName, dev->print_name());
356 * We reserve the device for appending by incrementing the
357 * reserved_device. We do virtually all the same work that
358 * is done in acquire_device_for_append(), but we do
359 * not attempt to mount the device. This routine allows
360 * the DIR to reserve multiple devices before *really*
361 * starting the job. It also permits the SD to refuse
362 * certain devices (not up, ...).
364 * Note, in reserving a device, if the device is for the
365 * same pool and the same pool type, then it is acceptable.
366 * The Media Type has already been checked. If we are
367 * the first tor reserve the device, we put the pool
368 * name and pool type in the device record.
370 bool reserve_device_for_append(DCR *dcr)
373 DEVICE *dev = dcr->dev;
379 init_device_wait_timers(dcr);
381 dev->block(BST_DOING_ACQUIRE);
383 Mmsg1(jcr->errmsg, _("Device %s is busy reading.\n"),
385 for (first=true; dev->can_read(); first=false) {
387 if (!wait_for_device(dcr, jcr->errmsg, first)) {
390 dev->block(BST_DOING_ACQUIRE);
394 Mmsg(jcr->errmsg, _("Device %s is BLOCKED due to user unmount.\n"),
396 for (first=true; device_is_unmounted(dev); first=false) {
398 if (!wait_for_device(dcr, jcr->errmsg, first)) {
401 dev->block(BST_DOING_ACQUIRE);
404 Dmsg1(190, "reserve_append device is %s\n", dev->is_tape()?"tape":"disk");
407 switch (can_reserve_drive(dcr)) {
409 Mmsg1(jcr->errmsg, _("Device %s is busy writing on another Volume.\n"), dev->print_name());
411 if (!wait_for_device(dcr, jcr->errmsg, first)) {
414 dev->block(BST_DOING_ACQUIRE);
417 goto bail_out; /* error */
419 break; /* OK, reserve drive */
425 dev->reserved_device++;
426 dcr->reserved_device = true;
435 * Returns: 1 if drive can be reserved
436 * 0 if we should wait
439 static int can_reserve_drive(DCR *dcr)
441 DEVICE *dev = dcr->dev;
444 * First handle the case that the drive is not yet in append mode
446 if (!dev->can_append() && dev->num_writers == 0) {
447 /* Now check if there are any reservations on the drive */
448 if (dev->reserved_device) {
449 /* Yes, now check if we want the same Pool and pool type */
450 if (strcmp(dev->pool_name, dcr->pool_name) == 0 &&
451 strcmp(dev->pool_type, dcr->pool_type) == 0) {
452 /* OK, compatible device */
454 /* Drive not suitable for us */
458 /* Device is available but not yet reserved, reserve it for us */
459 bstrncpy(dev->pool_name, dcr->pool_name, sizeof(dev->pool_name));
460 bstrncpy(dev->pool_type, dcr->pool_type, sizeof(dev->pool_type));
462 return 1; /* reserve drive */
466 * Now check if the device is in append mode
468 if (dev->can_append() || dev->num_writers > 0) {
469 Dmsg0(190, "device already in append.\n");
470 /* Yes, now check if we want the same Pool and pool type */
471 if (strcmp(dev->pool_name, dcr->pool_name) == 0 &&
472 strcmp(dev->pool_type, dcr->pool_type) == 0) {
473 /* OK, compatible device */
475 /* Drive not suitable for us */
476 Jmsg(jcr, M_WARNING, 0, _("Device %s is busy writing on another Volume.\n"), dev->print_name());
480 Pmsg0(000, "Logic error!!!! Should not get here.\n");
481 Jmsg0(jcr, M_FATAL, 0, _("Logic error!!!! Should not get here.\n"));
482 return -1; /* error, should not get here */
484 return 1; /* reserve drive */
488 * Acquire device for writing. We permit multiple writers.
489 * If this is the first one, we read the label.
491 * Returns: NULL if failed for any reason
493 * Note, normally reserve_device_for_append() is called
494 * before this routine.
496 DCR *acquire_device_for_append(DCR *dcr)
498 bool release = false;
499 bool recycle = false;
500 bool do_mount = false;
501 DEVICE *dev = dcr->dev;
504 init_device_wait_timers(dcr);
506 dev->block(BST_DOING_ACQUIRE);
507 Dmsg1(190, "acquire_append device is %s\n", dev->is_tape()?"tape":"disk");
509 if (dcr->reserved_device) {
510 dev->reserved_device--;
511 dcr->reserved_device = false;
515 * With the reservation system, this should not happen
517 if (dev->can_read()) {
518 Jmsg(jcr, M_FATAL, 0, _("Device %s is busy reading.\n"), dev->print_name());
522 if (dev->can_append()) {
523 Dmsg0(190, "device already in append.\n");
525 * Device already in append mode
527 * Check if we have the right Volume mounted
528 * OK if current volume info OK
529 * OK if next volume matches current volume
530 * otherwise mount desired volume obtained from
531 * dir_find_next_appendable_volume
533 bstrncpy(dcr->VolumeName, dev->VolHdr.VolName, sizeof(dcr->VolumeName));
534 if (!dir_get_volume_info(dcr, GET_VOL_INFO_FOR_WRITE) &&
535 !(dir_find_next_appendable_volume(dcr) &&
536 strcmp(dev->VolHdr.VolName, dcr->VolumeName) == 0)) { /* wrong tape mounted */
537 Dmsg0(190, "Wrong tape mounted.\n");
538 if (dev->num_writers != 0 || dev->reserved_device) {
539 Jmsg(jcr, M_FATAL, 0, _("Device %s is busy writing on another Volume.\n"), dev->print_name());
542 /* Wrong tape mounted, release it, then fall through to get correct one */
543 Dmsg0(190, "Wrong tape mounted, release and try mount.\n");
548 * At this point, the correct tape is already mounted, so
549 * we do not need to do mount_next_write_volume(), unless
550 * we need to recycle the tape.
552 recycle = strcmp(dcr->VolCatInfo.VolCatStatus, "Recycle") == 0;
553 Dmsg1(190, "Correct tape mounted. recycle=%d\n", recycle);
554 if (recycle && dev->num_writers != 0) {
555 Jmsg(jcr, M_FATAL, 0, _("Cannot recycle volume \"%s\""
556 " on device %s because it is in use by another job.\n"),
557 dev->VolHdr.VolName, dev->print_name());
560 if (dev->num_writers == 0) {
561 memcpy(&dev->VolCatInfo, &dcr->VolCatInfo, sizeof(dev->VolCatInfo));
565 /* Not already in append mode, so mount the device */
566 Dmsg0(190, "Not in append mode, try mount.\n");
567 ASSERT(dev->num_writers == 0);
571 if (do_mount || recycle) {
572 Dmsg0(190, "Do mount_next_write_vol\n");
573 bool mounted = mount_next_write_volume(dcr, release);
575 if (!job_canceled(jcr)) {
576 /* Reduce "noise" -- don't print if job canceled */
577 Jmsg(jcr, M_FATAL, 0, _("Could not ready device %s for append.\n"),
584 dev->num_writers++; /* we are now a writer */
585 if (jcr->NumVolumes == 0) {
591 * If we jump here, it is an error return because
592 * rtn_dev will still be NULL
603 * This job is done, so release the device. From a Unix standpoint,
604 * the device remains open.
607 bool release_device(DCR *dcr)
610 DEVICE *dev = dcr->dev;
614 Dmsg1(100, "release_device device is %s\n", dev->is_tape()?"tape":"disk");
616 /* if device is reserved, job never started, so release the reserve here */
617 if (dcr->reserved_device) {
618 dev->reserved_device--;
619 dcr->reserved_device = false;
622 if (dev->can_read()) {
623 dev->clear_read(); /* clear read bit */
625 /******FIXME**** send read volume usage statistics to director */
627 } else if (dev->num_writers > 0) {
629 * Note if WEOT is set, we are at the end of the tape
630 * and may not be positioned correctly, so the
631 * job_media_record and update_vol_info have already been
632 * done, which means we skip them here.
635 Dmsg1(100, "There are %d writers in release_device\n", dev->num_writers);
636 if (dev->is_labeled()) {
637 Dmsg0(100, "dir_create_jobmedia_record. Release\n");
638 if (!dev->at_weot() && !dir_create_jobmedia_record(dcr)) {
639 Jmsg(jcr, M_FATAL, 0, _("Could not create JobMedia record for Volume=\"%s\" Job=%s\n"),
640 dcr->VolCatInfo.VolCatName, jcr->Job);
642 /* If no more writers, write an EOF */
643 if (!dev->num_writers && dev->can_write()) {
645 write_ansi_ibm_labels(dcr, ANSI_EOF_LABEL, dev->VolHdr.VolName);
646 Dmsg0(100, "==== write ansi eof label \n");
648 if (!dev->at_weot()) {
649 dev->VolCatInfo.VolCatFiles = dev->file; /* set number of files */
650 dev->VolCatInfo.VolCatJobs++; /* increment number of jobs */
651 /* Note! do volume update before close, which zaps VolCatInfo */
652 Dmsg0(100, "dir_update_vol_info. Release0\n");
653 dir_update_volume_info(dcr, false); /* send Volume info to Director */
659 * If we reach here, it is most likely because the job
660 * has failed, since the device is not in read mode and
661 * there are no writers. It was probably reserved.
665 /* If no writers, close if file or !CAP_ALWAYS_OPEN */
666 if (dev->num_writers == 0 && (!dev->is_tape() || !dev_cap(dev, CAP_ALWAYSOPEN))) {
667 offline_or_rewind_dev(dev);
671 /* Fire off Alert command and include any output */
672 if (!job_canceled(jcr) && dcr->device->alert_command) {
676 char line[MAXSTRING];
677 alert = get_pool_memory(PM_FNAME);
678 alert = edit_device_codes(dcr, alert, "");
679 bpipe = open_bpipe(alert, 0, "r");
681 while (fgets(line, sizeof(line), bpipe->rfd)) {
682 Jmsg(jcr, M_ALERT, 0, _("Alert: %s"), line);
684 status = close_bpipe(bpipe);
690 Jmsg(jcr, M_ALERT, 0, _("3997 Bad alert command: %s: ERR=%s.\n"),
691 alert, be.strerror(status));
694 Dmsg1(400, "alert status=%d\n", status);
695 free_pool_memory(alert);
700 pthread_cond_broadcast(&wait_device_release);