2 Bacula® - The Network Backup Solution
4 Copyright (C) 2000-2009 Free Software Foundation Europe e.V.
6 The main author of Bacula is Kern Sibbald, with contributions from
7 many others, a complete list can be found in the file AUTHORS.
8 This program is Free Software; you can redistribute it and/or
9 modify it under the terms of version two of the GNU General Public
10 License as published by the Free Software Foundation and included
13 This program is distributed in the hope that it will be useful, but
14 WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with this program; if not, write to the Free Software
20 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
23 Bacula® is a registered trademark of Kern Sibbald.
24 The licensor of Bacula is the Free Software Foundation Europe
25 (FSFE), Fiduciary Program, Sumatrastrasse 25, 8006 Zürich,
26 Switzerland, email:ftf@fsfeurope.org.
29 * Drive reservation functions for Storage Daemon
33 * Split from job.c and acquire.c June 2005
42 const int dbglvl = 150;
44 static brwlock_t reservation_lock;
46 /* Forward referenced functions */
47 static int can_reserve_drive(DCR *dcr, RCTX &rctx);
48 static int reserve_device(RCTX &rctx);
49 static bool reserve_device_for_read(DCR *dcr);
50 static bool reserve_device_for_append(DCR *dcr, RCTX &rctx);
51 static bool use_storage_cmd(JCR *jcr);
52 static void queue_reserve_message(JCR *jcr);
53 static void pop_reserve_messages(JCR *jcr);
54 //void switch_device(DCR *dcr, DEVICE *dev);
56 /* Requests from the Director daemon */
57 static char use_storage[] = "use storage=%127s media_type=%127s "
58 "pool_name=%127s pool_type=%127s append=%d copy=%d stripe=%d\n";
59 static char use_device[] = "use device=%127s\n";
61 /* Responses sent to Director daemon */
62 static char OK_device[] = "3000 OK use device device=%s\n";
63 static char NO_device[] = "3924 Device \"%s\" not in SD Device resources.\n";
64 static char BAD_use[] = "3913 Bad use command: %s\n";
66 bool use_cmd(JCR *jcr)
69 * Get the device, media, and pool information
71 if (!use_storage_cmd(jcr)) {
72 set_jcr_job_status(jcr, JS_ErrorTerminated);
73 memset(jcr->sd_auth_key, 0, strlen(jcr->sd_auth_key));
80 * This allows a given thread to recursively call lock_reservations.
81 * It must, of course, call unlock_... the same number of times.
83 void init_reservations_lock()
86 if ((errstat=rwl_init(&reservation_lock)) != 0) {
88 Emsg1(M_ABORT, 0, _("Unable to initialize reservation lock. ERR=%s\n"),
89 be.bstrerror(errstat));
95 void term_reservations_lock()
97 rwl_destroy(&reservation_lock);
101 int reservations_lock_count = 0;
103 /* This applies to a drive and to Volumes */
104 void _lock_reservations()
107 reservations_lock_count++;
108 if ((errstat=rwl_writelock(&reservation_lock)) != 0) {
110 Emsg2(M_ABORT, 0, "rwl_writelock failure. stat=%d: ERR=%s\n",
111 errstat, be.bstrerror(errstat));
115 void _unlock_reservations()
118 reservations_lock_count--;
119 if ((errstat=rwl_writeunlock(&reservation_lock)) != 0) {
121 Emsg2(M_ABORT, 0, "rwl_writeunlock failure. stat=%d: ERR=%s\n",
122 errstat, be.bstrerror(errstat));
126 void DCR::set_reserved()
129 Dmsg2(dbglvl, "Inc reserve=%d dev=%s\n", dev->num_reserved(), dev->print_name());
133 void DCR::clear_reserved()
138 Dmsg2(dbglvl, "Dec reserve=%d dev=%s\n", dev->num_reserved(), dev->print_name());
143 * Remove any reservation from a drive and tell the system
144 * that the volume is unused at least by us.
146 void DCR::unreserve_device()
151 reserved_volume = false;
152 /* If we set read mode in reserving, remove it */
153 if (dev->can_read()) {
156 if (dev->num_writers < 0) {
157 Jmsg1(jcr, M_ERROR, 0, _("Hey! num_writers=%d!!!!\n"), dev->num_writers);
158 dev->num_writers = 0;
160 if (dev->num_reserved() == 0 && dev->num_writers == 0) {
168 * We get the following type of information:
170 * use storage=xxx media_type=yyy pool_name=xxx pool_type=yyy append=1 copy=0 strip=0
174 * use storage=xxx media_type=yyy pool_name=xxx pool_type=yyy append=0 copy=0 strip=0
178 static bool use_storage_cmd(JCR *jcr)
180 POOL_MEM store_name, dev_name, media_type, pool_name, pool_type;
181 BSOCK *dir = jcr->dir_bsock;
189 memset(&rctx, 0, sizeof(RCTX));
192 * If there are multiple devices, the director sends us
193 * use_device for each device that it wants to use.
195 dirstore = New(alist(10, not_owned_by_alist));
196 jcr->reserve_msgs = New(alist(10, not_owned_by_alist));
198 Dmsg1(dbglvl, "<dird: %s", dir->msg);
199 ok = sscanf(dir->msg, use_storage, store_name.c_str(),
200 media_type.c_str(), pool_name.c_str(),
201 pool_type.c_str(), &append, &Copy, &Stripe) == 7;
206 jcr->write_store = dirstore;
208 jcr->read_store = dirstore;
210 rctx.append = append;
211 unbash_spaces(store_name);
212 unbash_spaces(media_type);
213 unbash_spaces(pool_name);
214 unbash_spaces(pool_type);
215 store = new DIRSTORE;
216 dirstore->append(store);
217 memset(store, 0, sizeof(DIRSTORE));
218 store->device = New(alist(10));
219 bstrncpy(store->name, store_name, sizeof(store->name));
220 bstrncpy(store->media_type, media_type, sizeof(store->media_type));
221 bstrncpy(store->pool_name, pool_name, sizeof(store->pool_name));
222 bstrncpy(store->pool_type, pool_type, sizeof(store->pool_type));
223 store->append = append;
225 /* Now get all devices */
226 while (dir->recv() >= 0) {
227 Dmsg1(dbglvl, "<dird device: %s", dir->msg);
228 ok = sscanf(dir->msg, use_device, dev_name.c_str()) == 1;
232 unbash_spaces(dev_name);
233 store->device->append(bstrdup(dev_name.c_str()));
235 } while (ok && dir->recv() >= 0);
237 /* Developer debug code */
239 if (debug_level >= dbglvl) {
240 foreach_alist(store, dirstore) {
241 Dmsg5(dbglvl, "Storage=%s media_type=%s pool=%s pool_type=%s append=%d\n",
242 store->name, store->media_type, store->pool_name,
243 store->pool_type, store->append);
244 foreach_alist(device_name, store->device) {
245 Dmsg1(dbglvl, " Device=%s\n", device_name);
250 init_jcr_device_wait_timers(jcr);
251 jcr->dcr = new_dcr(jcr, NULL, NULL); /* get a dcr */
253 BSOCK *dir = jcr->dir_bsock;
254 dir->fsend(_("3939 Could not get dcr\n"));
255 Dmsg1(dbglvl, ">dird: %s", dir->msg);
259 * At this point, we have a list of all the Director's Storage
260 * resources indicated for this Job, which include Pool, PoolType,
261 * storage name, and Media type.
262 * Then for each of the Storage resources, we have a list of
263 * device names that were given.
265 * Wiffle through them and find one that can do the backup.
268 int wait_for_device_retries = 0;
271 rctx.notify_dir = true;
274 for ( ; !fail && !job_canceled(jcr); ) {
275 pop_reserve_messages(jcr);
276 rctx.suitable_device = false;
277 rctx.have_volume = false;
278 rctx.VolumeName[0] = 0;
279 rctx.any_drive = false;
280 if (!jcr->PreferMountedVols) {
282 * Here we try to find a drive that is not used.
283 * This will maximize the use of available drives.
286 rctx.num_writers = 20000000; /* start with impossible number */
287 rctx.low_use_drive = NULL;
288 rctx.PreferMountedVols = false;
289 rctx.exact_match = false;
290 rctx.autochanger_only = true;
291 if ((ok = find_suitable_device_for_job(jcr, rctx))) {
294 /* Look through all drives possibly for low_use drive */
295 if (rctx.low_use_drive) {
296 rctx.try_low_use_drive = true;
297 if ((ok = find_suitable_device_for_job(jcr, rctx))) {
300 rctx.try_low_use_drive = false;
302 rctx.autochanger_only = false;
303 if ((ok = find_suitable_device_for_job(jcr, rctx))) {
308 * Now we look for a drive that may or may not be in
311 /* Look for an exact Volume match all drives */
312 rctx.PreferMountedVols = true;
313 rctx.exact_match = true;
314 rctx.autochanger_only = false;
315 if ((ok = find_suitable_device_for_job(jcr, rctx))) {
318 /* Look for any mounted drive */
319 rctx.exact_match = false;
320 if ((ok = find_suitable_device_for_job(jcr, rctx))) {
324 rctx.any_drive = true;
325 if ((ok = find_suitable_device_for_job(jcr, rctx))) {
328 /* Keep reservations locked *except* during wait_for_device() */
329 unlock_reservations();
331 * The idea of looping on repeat a few times it to ensure
332 * that if there is some subtle timing problem between two
333 * jobs, we will simply try again, and most likely succeed.
334 * This can happen if one job reserves a drive or finishes using
335 * a drive at the same time a second job wants it.
337 if (repeat++ > 1) { /* try algorithm 3 times */
338 bmicrosleep(30, 0); /* wait a bit */
339 Dmsg0(dbglvl, "repeat reserve algorithm\n");
340 } else if (!rctx.suitable_device || !wait_for_device(jcr, wait_for_device_retries)) {
341 Dmsg0(dbglvl, "Fail. !suitable_device || !wait_for_device\n");
345 dir->signal(BNET_HEARTBEAT); /* Inform Dir that we are alive */
347 unlock_reservations();
350 * If we get here, there are no suitable devices available, which
351 * means nothing configured. If a device is suitable but busy
352 * with another Volume, we will not come here.
354 unbash_spaces(dir->msg);
355 pm_strcpy(jcr->errmsg, dir->msg);
356 Jmsg(jcr, M_FATAL, 0, _("Device reservation failed for JobId=%d: %s\n"),
357 jcr->JobId, jcr->errmsg);
358 dir->fsend(NO_device, dev_name.c_str());
360 Dmsg1(dbglvl, ">dird: %s", dir->msg);
363 unbash_spaces(dir->msg);
364 pm_strcpy(jcr->errmsg, dir->msg);
365 Jmsg(jcr, M_FATAL, 0, _("Failed command: %s\n"), jcr->errmsg);
366 dir->fsend(BAD_use, jcr->errmsg);
367 Dmsg1(dbglvl, ">dird: %s", dir->msg);
370 release_reserve_messages(jcr);
376 * Walk through the autochanger resources and check if
377 * the volume is in one of them.
379 * Returns: true if volume is in device
382 static bool is_vol_in_autochanger(RCTX &rctx, VOLRES *vol)
384 AUTOCHANGER *changer = vol->dev->device->changer_res;
386 /* Find resource, and make sure we were able to open it */
387 if (strcmp(rctx.device_name, changer->hdr.name) == 0) {
388 Dmsg1(dbglvl, "Found changer device %s\n", vol->dev->device->hdr.name);
391 Dmsg1(dbglvl, "Incorrect changer device %s\n", changer->hdr.name);
396 * Search for a device suitable for this job.
397 * Note, this routine sets sets rctx.suitable_device if any
398 * device exists within the SD. The device may not be actually
400 * It also returns if it finds a useable device.
402 bool find_suitable_device_for_job(JCR *jcr, RCTX &rctx)
411 dirstore = jcr->write_store;
413 dirstore = jcr->read_store;
415 Dmsg5(dbglvl, "Start find_suit_dev PrefMnt=%d exact=%d suitable=%d chgronly=%d any=%d\n",
416 rctx.PreferMountedVols, rctx.exact_match, rctx.suitable_device,
417 rctx.autochanger_only, rctx.any_drive);
420 * If the appropriate conditions of this if are met, namely that
421 * we are appending and the user wants mounted drive (or we
422 * force try a mounted drive because they are all busy), we
423 * start by looking at all the Volumes in the volume list.
425 if (!is_vol_list_empty() && rctx.append && rctx.PreferMountedVols) {
426 dlist *temp_vol_list;
428 temp_vol_list = dup_vol_list(jcr);
430 /* Look through reserved volumes for one we can use */
431 Dmsg0(dbglvl, "look for vol in vol list\n");
432 foreach_dlist(vol, temp_vol_list) {
434 Dmsg1(dbglvl, "vol=%s no dev\n", vol->vol_name);
437 /* Check with Director if this Volume is OK */
438 bstrncpy(dcr->VolumeName, vol->vol_name, sizeof(dcr->VolumeName));
439 if (!dir_get_volume_info(dcr, GET_VOL_INFO_FOR_WRITE)) {
443 Dmsg1(dbglvl, "vol=%s OK for this job\n", vol->vol_name);
444 foreach_alist(store, dirstore) {
447 foreach_alist(device_name, store->device) {
448 /* Found a device, try to use it */
449 rctx.device_name = device_name;
450 rctx.device = vol->dev->device;
452 if (vol->dev->is_autochanger()) {
453 Dmsg1(dbglvl, "vol=%s is in changer\n", vol->vol_name);
454 if (!is_vol_in_autochanger(rctx, vol)) {
457 } else if (strcmp(device_name, vol->dev->device->hdr.name) != 0) {
458 Dmsg2(dbglvl, "device=%s not suitable want %s\n",
459 vol->dev->device->hdr.name, device_name);
463 bstrncpy(rctx.VolumeName, vol->vol_name, sizeof(rctx.VolumeName));
464 rctx.have_volume = true;
465 /* Try reserving this device and volume */
466 Dmsg2(dbglvl, "try vol=%s on device=%s\n", rctx.VolumeName, device_name);
467 stat = reserve_device(rctx);
468 if (stat == 1) { /* found available device */
469 Dmsg1(dbglvl, "Suitable device found=%s\n", device_name);
472 } else if (stat == 0) { /* device busy */
473 Dmsg1(dbglvl, "Suitable device=%s, busy: not use\n", device_name);
475 /* otherwise error */
476 Dmsg0(dbglvl, "No suitable device found.\n");
478 rctx.have_volume = false;
479 rctx.VolumeName[0] = 0;
488 } /* end for loop over reserved volumes */
490 Dmsg0(dbglvl, "lock volumes\n");
491 free_temp_vol_list(temp_vol_list);
494 Dmsg1(dbglvl, "OK dev found. Vol=%s from in-use vols list\n", rctx.VolumeName);
499 * No reserved volume we can use, so now search for an available device.
501 * For each storage device that the user specified, we
502 * search and see if there is a resource for that device.
504 foreach_alist(store, dirstore) {
506 foreach_alist(device_name, store->device) {
508 rctx.device_name = device_name;
509 stat = search_res_for_device(rctx);
510 if (stat == 1) { /* found available device */
511 Dmsg1(dbglvl, "available device found=%s\n", device_name);
514 } else if (stat == 0) { /* device busy */
515 Dmsg1(dbglvl, "No usable device=%s, busy: not use\n", device_name);
517 /* otherwise error */
518 Dmsg0(dbglvl, "No usable device found.\n");
526 Dmsg1(dbglvl, "OK dev found. Vol=%s\n", rctx.VolumeName);
528 Dmsg0(dbglvl, "Leave find_suit_dev: no dev found.\n");
534 * Search for a particular storage device with particular storage
535 * characteristics (MediaType).
537 int search_res_for_device(RCTX &rctx)
539 AUTOCHANGER *changer;
542 Dmsg1(dbglvl, "search res for %s\n", rctx.device_name);
543 /* Look through Autochangers first */
544 foreach_res(changer, R_AUTOCHANGER) {
545 Dmsg1(dbglvl, "Try match changer res=%s\n", changer->hdr.name);
546 /* Find resource, and make sure we were able to open it */
547 if (strcmp(rctx.device_name, changer->hdr.name) == 0) {
548 /* Try each device in this AutoChanger */
549 foreach_alist(rctx.device, changer->device) {
550 Dmsg1(dbglvl, "Try changer device %s\n", rctx.device->hdr.name);
551 if (!rctx.device->autoselect) {
552 Dmsg1(100, "Device %s not autoselect skipped.\n",
553 rctx.device->hdr.name);
554 continue; /* device is not available */
556 stat = reserve_device(rctx);
557 if (stat != 1) { /* try another device */
561 if (rctx.store->append == SD_APPEND) {
562 Dmsg2(dbglvl, "Device %s reserved=%d for append.\n",
563 rctx.device->hdr.name, rctx.jcr->dcr->dev->num_reserved());
565 Dmsg2(dbglvl, "Device %s reserved=%d for read.\n",
566 rctx.device->hdr.name, rctx.jcr->read_dcr->dev->num_reserved());
573 /* Now if requested look through regular devices */
574 if (!rctx.autochanger_only) {
575 foreach_res(rctx.device, R_DEVICE) {
576 Dmsg1(dbglvl, "Try match res=%s\n", rctx.device->hdr.name);
577 /* Find resource, and make sure we were able to open it */
578 if (strcmp(rctx.device_name, rctx.device->hdr.name) == 0) {
579 stat = reserve_device(rctx);
580 if (stat != 1) { /* try another device */
584 if (rctx.store->append == SD_APPEND) {
585 Dmsg2(dbglvl, "Device %s reserved=%d for append.\n",
586 rctx.device->hdr.name, rctx.jcr->dcr->dev->num_reserved());
588 Dmsg2(dbglvl, "Device %s reserved=%d for read.\n",
589 rctx.device->hdr.name, rctx.jcr->read_dcr->dev->num_reserved());
595 return -1; /* nothing found */
599 * Try to reserve a specific device.
601 * Returns: 1 -- OK, have DCR
605 static int reserve_device(RCTX &rctx)
609 const int name_len = MAX_NAME_LENGTH;
611 /* Make sure MediaType is OK */
612 Dmsg2(dbglvl, "chk MediaType device=%s request=%s\n",
613 rctx.device->media_type, rctx.store->media_type);
614 if (strcmp(rctx.device->media_type, rctx.store->media_type) != 0) {
618 /* Make sure device exists -- i.e. we can stat() it */
619 if (!rctx.device->dev) {
620 rctx.device->dev = init_dev(rctx.jcr, rctx.device);
622 if (!rctx.device->dev) {
623 if (rctx.device->changer_res) {
624 Jmsg(rctx.jcr, M_WARNING, 0, _("\n"
625 " Device \"%s\" in changer \"%s\" requested by DIR could not be opened or does not exist.\n"),
626 rctx.device->hdr.name, rctx.device_name);
628 Jmsg(rctx.jcr, M_WARNING, 0, _("\n"
629 " Device \"%s\" requested by DIR could not be opened or does not exist.\n"),
632 return -1; /* no use waiting */
635 rctx.suitable_device = true;
636 Dmsg1(dbglvl, "try reserve %s\n", rctx.device->hdr.name);
637 rctx.jcr->dcr = dcr = new_dcr(rctx.jcr, rctx.jcr->dcr, rctx.device->dev);
639 BSOCK *dir = rctx.jcr->dir_bsock;
640 dir->fsend(_("3926 Could not get dcr for device: %s\n"), rctx.device_name);
641 Dmsg1(dbglvl, ">dird: %s", dir->msg);
644 bstrncpy(dcr->pool_name, rctx.store->pool_name, name_len);
645 bstrncpy(dcr->pool_type, rctx.store->pool_type, name_len);
646 bstrncpy(dcr->media_type, rctx.store->media_type, name_len);
647 bstrncpy(dcr->dev_name, rctx.device_name, name_len);
648 if (rctx.store->append == SD_APPEND) {
649 Dmsg2(dbglvl, "call reserve for append: have_vol=%d vol=%s\n", rctx.have_volume, rctx.VolumeName);
650 ok = reserve_device_for_append(dcr, rctx);
656 Dmsg5(dbglvl, "Reserved=%d dev_name=%s mediatype=%s pool=%s ok=%d\n",
657 dcr->dev->num_reserved(),
658 dcr->dev_name, dcr->media_type, dcr->pool_name, ok);
659 Dmsg3(dbglvl, "Vol=%s num_writers=%d, have_vol=%d\n",
660 rctx.VolumeName, dcr->dev->num_writers, rctx.have_volume);
661 if (rctx.have_volume) {
662 Dmsg0(dbglvl, "Call reserve_volume\n");
663 if (reserve_volume(dcr, rctx.VolumeName)) {
664 Dmsg1(dbglvl, "Reserved vol=%s\n", rctx.VolumeName);
666 Dmsg1(dbglvl, "Could not reserve vol=%s\n", rctx.VolumeName);
670 dcr->any_volume = true;
671 Dmsg0(dbglvl, "no vol, call find_next_appendable_vol.\n");
672 if (dir_find_next_appendable_volume(dcr)) {
673 bstrncpy(rctx.VolumeName, dcr->VolumeName, sizeof(rctx.VolumeName));
674 rctx.have_volume = true;
675 Dmsg1(dbglvl, "looking for Volume=%s\n", rctx.VolumeName);
677 Dmsg0(dbglvl, "No next volume found\n");
678 rctx.have_volume = false;
679 rctx.VolumeName[0] = 0;
681 * If there is at least one volume that is valid and in use,
682 * but we get here, check if we are running with prefers
683 * non-mounted drives. In that case, we have selected a
684 * non-used drive and our one and only volume is mounted
685 * elsewhere, so we bail out and retry using that drive.
687 if (dcr->found_in_use() && !rctx.PreferMountedVols) {
688 rctx.PreferMountedVols = true;
689 if (dcr->VolumeName[0]) {
690 dcr->unreserve_device();
695 * Note. Under some circumstances, the Director can hand us
696 * a Volume name that is not the same as the one on the current
697 * drive, and in that case, the call above to find the next
698 * volume will fail because in attempting to reserve the Volume
699 * the code will realize that we already have a tape mounted,
700 * and it will fail. This *should* only happen if there are
701 * writers, thus the following test. In that case, we simply
702 * bail out, and continue waiting, rather than plunging on
703 * and hoping that the operator can resolve the problem.
705 if (dcr->dev->num_writers != 0) {
706 if (dcr->VolumeName[0]) {
707 dcr->unreserve_device();
714 ok = reserve_device_for_read(dcr);
716 rctx.jcr->read_dcr = dcr;
717 Dmsg5(dbglvl, "Read reserved=%d dev_name=%s mediatype=%s pool=%s ok=%d\n",
718 dcr->dev->num_reserved(),
719 dcr->dev_name, dcr->media_type, dcr->pool_name, ok);
726 if (rctx.notify_dir) {
728 BSOCK *dir = rctx.jcr->dir_bsock;
729 pm_strcpy(dev_name, rctx.device->hdr.name);
730 bash_spaces(dev_name);
731 ok = dir->fsend(OK_device, dev_name.c_str()); /* Return real device name */
732 Dmsg1(dbglvl, ">dird: %s", dir->msg);
739 rctx.have_volume = false;
740 rctx.VolumeName[0] = 0;
741 Dmsg0(dbglvl, "Not OK.\n");
746 * We "reserve" the drive by setting the ST_READ bit. No one else
747 * should touch the drive until that is cleared.
748 * This allows the DIR to "reserve" the device before actually
751 static bool reserve_device_for_read(DCR *dcr)
753 DEVICE *dev = dcr->dev;
758 if (job_canceled(jcr)) {
764 if (dev->is_device_unmounted()) {
765 Dmsg1(dbglvl, "Device %s is BLOCKED due to user unmount.\n", dev->print_name());
766 Mmsg(jcr->errmsg, _("3601 JobId=%u device %s is BLOCKED due to user unmount.\n"),
767 jcr->JobId, dev->print_name());
768 queue_reserve_message(jcr);
772 if (dev->is_busy()) {
773 Dmsg4(dbglvl, "Device %s is busy ST_READ=%d num_writers=%d reserved=%d.\n",
775 dev->state & ST_READ?1:0, dev->num_writers, dev->num_reserved());
776 Mmsg(jcr->errmsg, _("3602 JobId=%u device %s is busy (already reading/writing).\n"),
777 jcr->JobId, dev->print_name());
778 queue_reserve_message(jcr);
794 * We reserve the device for appending by incrementing
795 * num_reserved(). We do virtually all the same work that
796 * is done in acquire_device_for_append(), but we do
797 * not attempt to mount the device. This routine allows
798 * the DIR to reserve multiple devices before *really*
799 * starting the job. It also permits the SD to refuse
800 * certain devices (not up, ...).
802 * Note, in reserving a device, if the device is for the
803 * same pool and the same pool type, then it is acceptable.
804 * The Media Type has already been checked. If we are
805 * the first tor reserve the device, we put the pool
806 * name and pool type in the device record.
808 static bool reserve_device_for_append(DCR *dcr, RCTX &rctx)
811 DEVICE *dev = dcr->dev;
815 if (job_canceled(jcr)) {
821 /* If device is being read, we cannot write it */
822 if (dev->can_read()) {
823 Mmsg(jcr->errmsg, _("3603 JobId=%u device %s is busy reading.\n"),
824 jcr->JobId, dev->print_name());
825 Dmsg1(dbglvl, "%s", jcr->errmsg);
826 queue_reserve_message(jcr);
830 /* If device is unmounted, we are out of luck */
831 if (dev->is_device_unmounted()) {
832 Mmsg(jcr->errmsg, _("3604 JobId=%u device %s is BLOCKED due to user unmount.\n"),
833 jcr->JobId, dev->print_name());
834 Dmsg1(dbglvl, "%s", jcr->errmsg);
835 queue_reserve_message(jcr);
839 Dmsg1(dbglvl, "reserve_append device is %s\n", dev->print_name());
841 /* Now do detailed tests ... */
842 if (can_reserve_drive(dcr, rctx) != 1) {
843 Dmsg0(dbglvl, "can_reserve_drive!=1\n");
855 static int is_pool_ok(DCR *dcr)
857 DEVICE *dev = dcr->dev;
860 /* Now check if we want the same Pool and pool type */
861 if (strcmp(dev->pool_name, dcr->pool_name) == 0 &&
862 strcmp(dev->pool_type, dcr->pool_type) == 0) {
863 /* OK, compatible device */
864 Dmsg1(dbglvl, "OK dev: %s num_writers=0, reserved, pool matches\n", dev->print_name());
867 /* Drive Pool not suitable for us */
869 "3608 JobId=%u wants Pool=\"%s\" but have Pool=\"%s\" nreserve=%d on drive %s.\n"),
870 (uint32_t)jcr->JobId, dcr->pool_name, dev->pool_name,
871 dev->num_reserved(), dev->print_name());
872 queue_reserve_message(jcr);
873 Dmsg2(dbglvl, "failed: busy num_writers=0, reserved, pool=%s wanted=%s\n",
874 dev->pool_name, dcr->pool_name);
879 static bool is_max_jobs_ok(DCR *dcr)
881 DEVICE *dev = dcr->dev;
884 Dmsg5(dbglvl, "MaxJobs=%d Jobs=%d reserves=%d Status=%s Vol=%s\n",
885 dcr->VolCatInfo.VolCatMaxJobs,
886 dcr->VolCatInfo.VolCatJobs, dev->num_reserved(),
887 dcr->VolCatInfo.VolCatStatus,
889 /* Limit max concurrent jobs on this drive */
890 if (dev->max_concurrent_jobs > 0 && dev->max_concurrent_jobs <=
891 (uint32_t)(dev->num_writers + dev->num_reserved())) {
892 /* Max Concurrent Jobs depassed or already reserved */
893 Mmsg(jcr->errmsg, _("3609 JobId=%u Max concurrent jobs exceeded on drive %s.\n"),
894 (uint32_t)jcr->JobId, dev->print_name());
895 queue_reserve_message(jcr);
896 Dmsg1(dbglvl, "reserve dev failed: %s", jcr->errmsg);
899 if (strcmp(dcr->VolCatInfo.VolCatStatus, "Recycle") == 0) {
902 if (dcr->VolCatInfo.VolCatMaxJobs > 0 && dcr->VolCatInfo.VolCatMaxJobs <=
903 (dcr->VolCatInfo.VolCatJobs + dev->num_reserved())) {
904 /* Max Job Vols depassed or already reserved */
905 Mmsg(jcr->errmsg, _("3610 JobId=%u Volume max jobs exceeded on drive %s.\n"),
906 (uint32_t)jcr->JobId, dev->print_name());
907 queue_reserve_message(jcr);
908 Dmsg1(dbglvl, "reserve dev failed: %s", jcr->errmsg);
909 return false; /* wait */
915 * Returns: 1 if drive can be reserved
916 * 0 if we should wait
917 * -1 on error or impossibility
919 static int can_reserve_drive(DCR *dcr, RCTX &rctx)
921 DEVICE *dev = dcr->dev;
924 Dmsg5(dbglvl, "PrefMnt=%d exact=%d suitable=%d chgronly=%d any=%d\n",
925 rctx.PreferMountedVols, rctx.exact_match, rctx.suitable_device,
926 rctx.autochanger_only, rctx.any_drive);
928 /* Check for max jobs on this Volume */
929 if (!is_max_jobs_ok(dcr)) {
933 /* setting any_drive overrides PreferMountedVols flag */
934 if (!rctx.any_drive) {
936 * When PreferMountedVols is set, we keep track of the
937 * drive in use that has the least number of writers, then if
938 * no unmounted drive is found, we try that drive. This
939 * helps spread the load to the least used drives.
941 if (rctx.try_low_use_drive && dev == rctx.low_use_drive) {
942 Dmsg2(dbglvl, "OK dev=%s == low_drive=%s.\n",
943 dev->print_name(), rctx.low_use_drive->print_name());
946 /* If he wants a free drive, but this one is busy, no go */
947 if (!rctx.PreferMountedVols && dev->is_busy()) {
948 /* Save least used drive */
949 if ((dev->num_writers + dev->num_reserved()) < rctx.num_writers) {
950 rctx.num_writers = dev->num_writers + dev->num_reserved();
951 rctx.low_use_drive = dev;
952 Dmsg2(dbglvl, "set low use drive=%s num_writers=%d\n",
953 dev->print_name(), rctx.num_writers);
955 Dmsg1(dbglvl, "not low use num_writers=%d\n", dev->num_writers+dev->num_reserved());
957 Dmsg0(dbglvl, "failed: !prefMnt && busy.\n");
958 Mmsg(jcr->errmsg, _("3605 JobId=%u wants free drive but device %s is busy.\n"),
959 jcr->JobId, dev->print_name());
960 queue_reserve_message(jcr);
964 /* Check for prefer mounted volumes */
965 if (rctx.PreferMountedVols && !dev->vol && dev->is_tape()) {
966 Mmsg(jcr->errmsg, _("3606 JobId=%u prefers mounted drives, but drive %s has no Volume.\n"),
967 jcr->JobId, dev->print_name());
968 queue_reserve_message(jcr);
969 Dmsg0(dbglvl, "failed: want mounted -- no vol\n");
970 return 0; /* No volume mounted */
973 /* Check for exact Volume name match */
974 /* ***FIXME*** for Disk, we can accept any volume that goes with this
977 if (rctx.exact_match && rctx.have_volume) {
979 Dmsg5(dbglvl, "PrefMnt=%d exact=%d suitable=%d chgronly=%d any=%d\n",
980 rctx.PreferMountedVols, rctx.exact_match, rctx.suitable_device,
981 rctx.autochanger_only, rctx.any_drive);
982 Dmsg4(dbglvl, "have_vol=%d have=%s resvol=%s want=%s\n",
983 rctx.have_volume, dev->VolHdr.VolumeName,
984 dev->vol?dev->vol->vol_name:"*none*", rctx.VolumeName);
985 ok = strcmp(dev->VolHdr.VolumeName, rctx.VolumeName) == 0 ||
986 (dev->vol && strcmp(dev->vol->vol_name, rctx.VolumeName) == 0);
988 Mmsg(jcr->errmsg, _("3607 JobId=%u wants Vol=\"%s\" drive has Vol=\"%s\" on drive %s.\n"),
989 jcr->JobId, rctx.VolumeName, dev->VolHdr.VolumeName,
991 queue_reserve_message(jcr);
992 Dmsg3(dbglvl, "not OK: dev have=%s resvol=%s want=%s\n",
993 dev->VolHdr.VolumeName, dev->vol?dev->vol->vol_name:"*none*", rctx.VolumeName);
996 if (!dcr->can_i_use_volume()) {
997 return 0; /* fail if volume on another drive */
1002 /* Check for unused autochanger drive */
1003 if (rctx.autochanger_only && !dev->is_busy() &&
1004 dev->VolHdr.VolumeName[0] == 0) {
1005 /* Device is available but not yet reserved, reserve it for us */
1006 Dmsg1(dbglvl, "OK Res Unused autochanger %s.\n", dev->print_name());
1007 bstrncpy(dev->pool_name, dcr->pool_name, sizeof(dev->pool_name));
1008 bstrncpy(dev->pool_type, dcr->pool_type, sizeof(dev->pool_type));
1009 return 1; /* reserve drive */
1013 * Handle the case that there are no writers
1015 if (dev->num_writers == 0) {
1016 /* Now check if there are any reservations on the drive */
1017 if (dev->num_reserved()) {
1018 return is_pool_ok(dcr);
1019 } else if (dev->can_append()) {
1020 if (is_pool_ok(dcr)) {
1023 /* Changing pool, unload old tape if any in drive */
1024 Dmsg0(dbglvl, "OK dev: num_writers=0, not reserved, pool change, unload changer\n");
1025 /* ***FIXME*** use set_unload() */
1026 unload_autochanger(dcr, -1);
1029 /* Device is available but not yet reserved, reserve it for us */
1030 Dmsg1(dbglvl, "OK Dev avail reserved %s\n", dev->print_name());
1031 bstrncpy(dev->pool_name, dcr->pool_name, sizeof(dev->pool_name));
1032 bstrncpy(dev->pool_type, dcr->pool_type, sizeof(dev->pool_type));
1033 return 1; /* reserve drive */
1037 * Check if the device is in append mode with writers (i.e.
1038 * available if pool is the same).
1040 if (dev->can_append() || dev->num_writers > 0) {
1041 return is_pool_ok(dcr);
1043 Pmsg1(000, _("Logic error!!!! JobId=%u Should not get here.\n"), (int)jcr->JobId);
1044 Mmsg(jcr->errmsg, _("3910 JobId=%u Logic error!!!! drive %s Should not get here.\n"),
1045 jcr->JobId, dev->print_name());
1046 queue_reserve_message(jcr);
1047 Jmsg0(jcr, M_FATAL, 0, _("Logic error!!!! Should not get here.\n"));
1048 return -1; /* error, should not get here */
1050 Mmsg(jcr->errmsg, _("3911 JobId=%u failed reserve drive %s.\n"),
1051 jcr->JobId, dev->print_name());
1052 queue_reserve_message(jcr);
1053 Dmsg1(dbglvl, "failed: No reserve %s\n", dev->print_name());
1061 * Queue a reservation error or failure message for this jcr
1063 static void queue_reserve_message(JCR *jcr)
1071 msgs = jcr->reserve_msgs;
1076 * Look for duplicate message. If found, do
1079 for (i=msgs->size()-1; i >= 0; i--) {
1080 msg = (char *)msgs->get(i);
1084 /* Comparison based on 4 digit message number */
1085 if (strncmp(msg, jcr->errmsg, 4) == 0) {
1089 /* Message unique, so insert it */
1090 jcr->reserve_msgs->push(bstrdup(jcr->errmsg));
1097 * Send any reservation messages queued for this jcr
1099 void send_drive_reserve_messages(JCR *jcr, void sendit(const char *msg, int len, void *sarg), void *arg)
1106 msgs = jcr->reserve_msgs;
1107 if (!msgs || msgs->size() == 0) {
1110 for (i=msgs->size()-1; i >= 0; i--) {
1111 msg = (char *)msgs->get(i);
1113 sendit(" ", 3, arg);
1114 sendit(msg, strlen(msg), arg);
1125 * Pop and release any reservations messages
1127 static void pop_reserve_messages(JCR *jcr)
1133 msgs = jcr->reserve_msgs;
1137 while ((msg = (char *)msgs->pop())) {
1145 * Also called from acquire.c
1147 void release_reserve_messages(JCR *jcr)
1149 pop_reserve_messages(jcr);
1151 if (!jcr->reserve_msgs) {
1154 delete jcr->reserve_msgs;
1155 jcr->reserve_msgs = NULL;