3 * Higher Level Device routines.
4 * Knows about Bacula tape labels and such
6 * NOTE! In general, subroutines that have the word
7 * "device" in the name do locking. Subroutines
8 * that have the word "dev" in the name do not
9 * do locking. Thus if xxx_device() calls
10 * yyy_dev(), all is OK, but if xxx_device()
11 * calls yyy_device(), everything will hang.
12 * Obviously, no zzz_dev() is allowed to call
13 * a www_device() or everything falls apart.
15 * Concerning the routines lock_device() and block_device()
16 * see the end of this module for details. In general,
17 * blocking a device leaves it in a state where all threads
18 * other than the current thread block when they attempt to
19 * lock the device. They remain suspended (blocked) until the device
20 * is unblocked. So, a device is blocked during an operation
21 * that takes a long time (initialization, mounting a new
22 * volume, ...) locking a device is done for an operation
23 * that takes a short time such as writing data to the
27 * Kern Sibbald, MM, MMI
32 Copyright (C) 2000-2004 Kern Sibbald and John Walker
34 This program is free software; you can redistribute it and/or
35 modify it under the terms of the GNU General Public License as
36 published by the Free Software Foundation; either version 2 of
37 the License, or (at your option) any later version.
39 This program is distributed in the hope that it will be useful,
40 but WITHOUT ANY WARRANTY; without even the implied warranty of
41 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
42 General Public License for more details.
44 You should have received a copy of the GNU General Public
45 License along with this program; if not, write to the Free
46 Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
51 #include "bacula.h" /* pull in global headers */
52 #include "stored.h" /* pull in Storage Deamon headers */
54 /* Forward referenced functions */
56 extern char my_name[];
57 extern int debug_level;
60 * This is the dreaded moment. We either have an end of
61 * medium condition or worse, and error condition.
62 * Attempt to "recover" by obtaining a new Volume.
64 * Here are a few things to know:
65 * jcr->VolCatInfo contains the info on the "current" tape for this job.
66 * dev->VolCatInfo contains the info on the tape in the drive.
67 * The tape in the drive could have changed several times since
68 * the last time the job used it (jcr->VolCatInfo).
69 * jcr->VolumeName is the name of the current/desired tape in the drive.
71 * We enter with device locked, and
72 * exit with device locked.
74 * Note, we are called only from one place in block.c
76 * Returns: true on success
79 bool fixup_device_block_write_error(DCR *dcr, DEV_BLOCK *block)
82 char PrevVolName[MAX_NAME_LENGTH];
86 char dt[MAX_TIME_LENGTH];
88 DEVICE *dev = dcr->dev;
90 wait_time = time(NULL);
91 stat = status_dev(dev);
92 if (!(stat & BMT_EOD)) {
93 return false; /* this really shouldn't happen */
96 Dmsg0(100, "======= Got EOD ========\n");
98 block_device(dev, BST_DOING_ACQUIRE);
99 /* Unlock, but leave BLOCKED */
102 /* Create a jobmedia record for this job */
103 if (!dir_create_jobmedia_record(dcr)) {
104 Jmsg(jcr, M_ERROR, 0, _("Could not create JobMedia record for Volume=\"%s\" Job=%s\n"),
105 jcr->VolCatInfo.VolCatName, jcr->Job);
111 bstrncpy(dev->VolCatInfo.VolCatStatus, "Full", sizeof(dev->VolCatInfo.VolCatStatus));
112 Dmsg2(100, "Call update_vol_info Stat=%s Vol=%s\n",
113 dev->VolCatInfo.VolCatStatus, dev->VolCatInfo.VolCatName);
114 dev->VolCatInfo.VolCatFiles = dev->file; /* set number of files */
115 dev->VolCatInfo.VolCatJobs++; /* increment number of jobs */
116 if (!dir_update_volume_info(dcr, false)) { /* send Volume info to Director */
119 return false; /* device locked */
121 Dmsg0(100, "Back from update_vol_info\n");
123 bstrncpy(PrevVolName, dev->VolCatInfo.VolCatName, sizeof(PrevVolName));
124 bstrncpy(dev->VolHdr.PrevVolName, PrevVolName, sizeof(dev->VolHdr.PrevVolName));
126 label_blk = new_block(dev);
128 /* Inform User about end of medium */
129 Jmsg(jcr, M_INFO, 0, _("End of medium on Volume \"%s\" Bytes=%s Blocks=%s at %s.\n"),
130 PrevVolName, edit_uint64_with_commas(dev->VolCatInfo.VolCatBytes, b1),
131 edit_uint64_with_commas(dev->VolCatInfo.VolCatBlocks, b2),
132 bstrftime(dt, sizeof(dt), time(NULL)));
134 if (!mount_next_write_volume(dcr, label_blk, 1)) {
135 free_block(label_blk);
138 return false; /* device locked */
140 P(dev->mutex); /* lock again */
142 Jmsg(jcr, M_INFO, 0, _("New volume \"%s\" mounted on device %s at %s.\n"),
143 jcr->VolumeName, dev_name(dev), bstrftime(dt, sizeof(dt), time(NULL)));
146 * If this is a new tape, the label_blk will contain the
147 * label, so write it now. If this is a previously
148 * used tape, mount_next_write_volume() will return an
149 * empty label_blk, and nothing will be written.
151 Dmsg0(190, "write label block to dev\n");
152 if (!write_block_to_dev(dcr, label_blk)) {
153 Pmsg1(0, "write_block_to_device Volume label failed. ERR=%s",
155 free_block(label_blk);
157 return false; /* device locked */
159 free_block(label_blk);
162 * Walk through all attached jcrs indicating the volume has changed
164 Dmsg1(100, "Walk attached jcrs. Volume=%s\n", dev->VolCatInfo.VolCatName);
165 // for (JCR *mjcr=NULL; (mjcr=next_attached_jcr(dev, mjcr)); ) {
167 foreach_dlist(mdcr, dev->attached_dcrs) {
168 JCR *mjcr = mdcr->jcr;
169 if (mjcr->JobId == 0) {
170 continue; /* ignore console */
174 pm_strcpy(&mjcr->VolumeName, jcr->VolumeName); /* get a copy of the new volume */
175 bstrncpy(mdcr->VolumeName, jcr->VolumeName, sizeof(mdcr->VolumeName));
179 /* Clear NewVol now because dir_get_volume_info() already done */
180 jcr->dcr->NewVol = false;
181 set_new_volume_parameters(dcr);
183 jcr->run_time += time(NULL) - wait_time; /* correct run time for mount wait */
185 /* Write overflow block to device */
186 Dmsg0(190, "Write overflow block to dev\n");
187 if (!write_block_to_dev(dcr, block)) {
188 Pmsg1(0, "write_block_to_device overflow block failed. ERR=%s",
191 return false; /* device locked */
195 return true; /* device locked */
199 * We have a new Volume mounted, so reset the Volume parameters
200 * concerning this job. The global changes were made earlier
201 * in the dev structure.
203 void set_new_volume_parameters(DCR *dcr)
206 DEVICE *dev = dcr->dev;
207 if (dcr->NewVol && !dir_get_volume_info(dcr, GET_VOL_INFO_FOR_WRITE)) {
208 Jmsg1(jcr, M_ERROR, 0, "%s", jcr->errmsg);
210 /* Set new start/end positions */
211 if (dev_state(dev, ST_TAPE)) {
212 dcr->StartBlock = dev->block_num;
213 dcr->StartFile = dev->file;
215 dcr->StartBlock = (uint32_t)dev->file_addr;
216 dcr->StartFile = (uint32_t)(dev->file_addr >> 32);
219 dcr->VolFirstIndex = 0;
220 dcr->VolLastIndex = 0;
223 dcr->WroteVol = false;
227 * We are now in a new Volume file, so reset the Volume parameters
228 * concerning this job. The global changes were made earlier
229 * in the dev structure.
231 void set_new_file_parameters(DCR *dcr)
233 DEVICE *dev = dcr->dev;
235 /* Set new start/end positions */
236 if (dev_state(dev, ST_TAPE)) {
237 dcr->StartBlock = dev->block_num;
238 dcr->StartFile = dev->file;
240 dcr->StartBlock = (uint32_t)dev->file_addr;
241 dcr->StartFile = (uint32_t)(dev->file_addr >> 32);
244 dcr->VolFirstIndex = 0;
245 dcr->VolLastIndex = 0;
246 dcr->NewFile = false;
247 dcr->WroteVol = false;
253 * First Open of the device. Expect dev to already be initialized.
255 * This routine is used only when the Storage daemon starts
256 * and always_open is set, and in the stand-alone utility
257 * routines such as bextract.
259 * Note, opening of a normal file is deferred to later so
260 * that we can get the filename; the device_name for
261 * a file is the directory only.
263 * Returns: false on failure
266 bool first_open_device(DEVICE *dev)
268 Dmsg0(120, "start open_output_device()\n");
275 /* Defer opening files */
276 if (!dev_is_tape(dev)) {
277 Dmsg0(129, "Device is file, deferring open.\n");
282 if (!(dev->state & ST_OPENED)) {
284 if (dev_cap(dev, CAP_STREAM)) {
285 mode = OPEN_WRITE_ONLY;
287 mode = OPEN_READ_WRITE;
289 Dmsg0(129, "Opening device.\n");
290 if (open_dev(dev, NULL, mode) < 0) {
291 Emsg1(M_FATAL, 0, _("dev open failed: %s\n"), dev->errmsg);
296 Dmsg1(129, "open_dev %s OK\n", dev_name(dev));
303 * Make sure device is open, if not do so
305 bool open_device(JCR *jcr, DEVICE *dev)
308 if (!(dev_state(dev, ST_OPENED))) {
310 if (dev_cap(dev, CAP_STREAM)) {
311 mode = OPEN_WRITE_ONLY;
313 mode = OPEN_READ_WRITE;
315 if (open_dev(dev, jcr->VolCatInfo.VolCatName, mode) < 0) {
316 /* If polling, ignore the error */
318 Jmsg2(jcr, M_FATAL, 0, _("Unable to open device %s. ERR=%s\n"),
319 dev_name(dev), strerror_dev(dev));
327 void dev_lock(DEVICE *dev)
330 if ((errstat=rwl_writelock(&dev->lock))) {
331 Emsg1(M_ABORT, 0, "Device write lock failure. ERR=%s\n", strerror(errstat));
335 void dev_unlock(DEVICE *dev)
338 if ((errstat=rwl_writeunlock(&dev->lock))) {
339 Emsg1(M_ABORT, 0, "Device write unlock failure. ERR=%s\n", strerror(errstat));
344 * When dev_blocked is set, all threads EXCEPT thread with id no_wait_id
345 * must wait. The no_wait_id thread is out obtaining a new volume
346 * and preparing the label.
348 void _lock_device(const char *file, int line, DEVICE *dev)
351 Dmsg3(500, "lock %d from %s:%d\n", dev->dev_blocked, file, line);
353 if (dev->dev_blocked && !pthread_equal(dev->no_wait_id, pthread_self())) {
354 dev->num_waiting++; /* indicate that I am waiting */
355 while (dev->dev_blocked) {
356 if ((stat = pthread_cond_wait(&dev->wait, &dev->mutex)) != 0) {
358 Emsg1(M_ABORT, 0, _("pthread_cond_wait failure. ERR=%s\n"),
362 dev->num_waiting--; /* no longer waiting */
367 * Check if the device is blocked or not
369 bool device_is_unmounted(DEVICE *dev)
372 int blocked = dev->dev_blocked;
373 stat = (blocked == BST_UNMOUNTED) ||
374 (blocked == BST_UNMOUNTED_WAITING_FOR_SYSOP);
378 const char *edit_blocked_reason(DEVICE *dev)
380 switch (dev->dev_blocked) {
381 case BST_NOT_BLOCKED:
382 return "not blocked";
384 return "user unmounted device";
385 case BST_WAITING_FOR_SYSOP:
386 return "waiting for operator action";
387 case BST_DOING_ACQUIRE:
388 return "opening, validating, or positioning tape";
389 case BST_WRITING_LABEL:
390 return "labeling tape";
391 case BST_UNMOUNTED_WAITING_FOR_SYSOP:
392 return "closed by user during mount request";
394 return "mount request";
396 return "unknown blocked code";
400 void _unlock_device(const char *file, int line, DEVICE *dev)
402 Dmsg2(500, "unlock from %s:%d\n", file, line);
407 * Block all other threads from using the device
408 * Device must already be locked. After this call,
409 * the device is blocked to any thread calling lock_device(),
410 * but the device is not locked (i.e. no P on device). Also,
411 * the current thread can do slip through the lock_device()
412 * calls without blocking.
414 void _block_device(const char *file, int line, DEVICE *dev, int state)
416 Dmsg3(500, "block set %d from %s:%d\n", state, file, line);
417 ASSERT(dev->dev_blocked == BST_NOT_BLOCKED);
418 dev->dev_blocked = state; /* make other threads wait */
419 dev->no_wait_id = pthread_self(); /* allow us to continue */
425 * Unblock the device, and wake up anyone who went to sleep.
427 void _unblock_device(const char *file, int line, DEVICE *dev)
429 Dmsg3(500, "unblock %d from %s:%d\n", dev->dev_blocked, file, line);
430 ASSERT(dev->dev_blocked);
431 dev->dev_blocked = BST_NOT_BLOCKED;
433 if (dev->num_waiting > 0) {
434 pthread_cond_broadcast(&dev->wait); /* wake them up */
439 * Enter with device locked and blocked
440 * Exit with device unlocked and blocked by us.
442 void _steal_device_lock(const char *file, int line, DEVICE *dev, bsteal_lock_t *hold, int state)
444 Dmsg4(500, "steal lock. old=%d new=%d from %s:%d\n", dev->dev_blocked, state,
446 hold->dev_blocked = dev->dev_blocked;
447 hold->dev_prev_blocked = dev->dev_prev_blocked;
448 hold->no_wait_id = dev->no_wait_id;
449 dev->dev_blocked = state;
450 dev->no_wait_id = pthread_self();
455 * Enter with device blocked by us but not locked
456 * Exit with device locked, and blocked by previous owner
458 void _give_back_device_lock(const char *file, int line, DEVICE *dev, bsteal_lock_t *hold)
460 Dmsg4(500, "return lock. old=%d new=%d from %s:%d\n",
461 dev->dev_blocked, hold->dev_blocked, file, line);
463 dev->dev_blocked = hold->dev_blocked;
464 dev->dev_prev_blocked = hold->dev_prev_blocked;
465 dev->no_wait_id = hold->no_wait_id;
466 if (dev->dev_blocked == BST_NOT_BLOCKED && dev->num_waiting > 0) {
467 pthread_cond_broadcast(&dev->wait); /* wake them up */