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) 2000-2005 Kern Sibbald
34 This program is free software; you can redistribute it and/or
35 modify it under the terms of the GNU General Public License
36 version 2 as amended with additional clauses defined in the
37 file LICENSE in the main source directory.
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
42 the file LICENSE for additional details.
46 #include "bacula.h" /* pull in global headers */
47 #include "stored.h" /* pull in Storage Deamon headers */
49 /* Forward referenced functions */
51 extern char my_name[];
52 extern int debug_level;
55 * This is the dreaded moment. We either have an end of
56 * medium condition or worse, and error condition.
57 * Attempt to "recover" by obtaining a new Volume.
59 * Here are a few things to know:
60 * dcr->VolCatInfo contains the info on the "current" tape for this job.
61 * dev->VolCatInfo contains the info on the tape in the drive.
62 * The tape in the drive could have changed several times since
63 * the last time the job used it (jcr->VolCatInfo).
64 * dcr->VolumeName is the name of the current/desired tape in the drive.
66 * We enter with device locked, and
67 * exit with device locked.
69 * Note, we are called only from one place in block.c
71 * Returns: true on success
74 bool fixup_device_block_write_error(DCR *dcr)
77 char PrevVolName[MAX_NAME_LENGTH];
79 DEV_BLOCK *block = dcr->block;
82 char dt[MAX_TIME_LENGTH];
84 DEVICE *dev = dcr->dev;
86 wait_time = time(NULL);
87 stat = status_dev(dev);
88 if (!(stat & BMT_EOD)) {
89 return false; /* this really shouldn't happen */
92 Dmsg0(100, "======= Got EOD ========\n");
94 block_device(dev, BST_DOING_ACQUIRE);
95 /* Unlock, but leave BLOCKED */
98 bstrncpy(PrevVolName, dev->VolCatInfo.VolCatName, sizeof(PrevVolName));
99 bstrncpy(dev->VolHdr.PrevVolumeName, PrevVolName, sizeof(dev->VolHdr.PrevVolumeName));
101 label_blk = new_block(dev);
102 dcr->block = label_blk;
104 /* Inform User about end of medium */
105 Jmsg(jcr, M_INFO, 0, _("End of medium on Volume \"%s\" Bytes=%s Blocks=%s at %s.\n"),
106 PrevVolName, edit_uint64_with_commas(dev->VolCatInfo.VolCatBytes, b1),
107 edit_uint64_with_commas(dev->VolCatInfo.VolCatBlocks, b2),
108 bstrftime(dt, sizeof(dt), time(NULL)));
110 if (!mount_next_write_volume(dcr, 1)) {
111 free_block(label_blk);
115 return false; /* device locked */
117 P(dev->mutex); /* lock again */
119 Jmsg(jcr, M_INFO, 0, _("New volume \"%s\" mounted on device %s at %s.\n"),
120 dcr->VolumeName, dev->print_name(), bstrftime(dt, sizeof(dt), time(NULL)));
123 * If this is a new tape, the label_blk will contain the
124 * label, so write it now. If this is a previously
125 * used tape, mount_next_write_volume() will return an
126 * empty label_blk, and nothing will be written.
128 Dmsg0(190, "write label block to dev\n");
129 if (!write_block_to_dev(dcr)) {
131 Pmsg1(0, "write_block_to_device Volume label failed. ERR=%s",
132 be.strerror(dev->dev_errno));
133 free_block(label_blk);
136 return false; /* device locked */
138 free_block(label_blk);
142 * Walk through all attached jcrs indicating the volume has changed
144 Dmsg1(100, "Walk attached jcrs. Volume=%s\n", dev->VolCatInfo.VolCatName);
145 // for (JCR *mjcr=NULL; (mjcr=next_attached_jcr(dev, mjcr)); ) {
147 foreach_dlist(mdcr, dev->attached_dcrs) {
148 JCR *mjcr = mdcr->jcr;
149 if (mjcr->JobId == 0) {
150 continue; /* ignore console */
154 bstrncpy(mdcr->VolumeName, dcr->VolumeName, sizeof(mdcr->VolumeName));
158 /* Clear NewVol now because dir_get_volume_info() already done */
159 jcr->dcr->NewVol = false;
160 set_new_volume_parameters(dcr);
162 jcr->run_time += time(NULL) - wait_time; /* correct run time for mount wait */
164 /* Write overflow block to device */
165 Dmsg0(190, "Write overflow block to dev\n");
166 if (!write_block_to_dev(dcr)) {
168 Pmsg1(0, "write_block_to_device overflow block failed. ERR=%s",
169 be.strerror(dev->dev_errno));
171 return false; /* device locked */
175 return true; /* device locked */
179 * We have a new Volume mounted, so reset the Volume parameters
180 * concerning this job. The global changes were made earlier
181 * in the dev structure.
183 void set_new_volume_parameters(DCR *dcr)
186 DEVICE *dev = dcr->dev;
187 if (dcr->NewVol && !dir_get_volume_info(dcr, GET_VOL_INFO_FOR_WRITE)) {
188 Jmsg1(jcr, M_ERROR, 0, "%s", jcr->errmsg);
190 /* Set new start/end positions */
191 if (dev->is_tape()) {
192 dcr->StartBlock = dev->block_num;
193 dcr->StartFile = dev->file;
195 dcr->StartBlock = (uint32_t)dev->file_addr;
196 dcr->StartFile = (uint32_t)(dev->file_addr >> 32);
199 dcr->VolFirstIndex = 0;
200 dcr->VolLastIndex = 0;
203 dcr->WroteVol = false;
207 * We are now in a new Volume file, so reset the Volume parameters
208 * concerning this job. The global changes were made earlier
209 * in the dev structure.
211 void set_new_file_parameters(DCR *dcr)
213 DEVICE *dev = dcr->dev;
215 /* Set new start/end positions */
216 if (dev->is_tape()) {
217 dcr->StartBlock = dev->block_num;
218 dcr->StartFile = dev->file;
220 dcr->StartBlock = (uint32_t)dev->file_addr;
221 dcr->StartFile = (uint32_t)(dev->file_addr >> 32);
224 dcr->VolFirstIndex = 0;
225 dcr->VolLastIndex = 0;
226 dcr->NewFile = false;
227 dcr->WroteVol = false;
233 * First Open of the device. Expect dev to already be initialized.
235 * This routine is used only when the Storage daemon starts
236 * and always_open is set, and in the stand-alone utility
237 * routines such as bextract.
239 * Note, opening of a normal file is deferred to later so
240 * that we can get the filename; the device_name for
241 * a file is the directory only.
243 * Returns: false on failure
246 bool first_open_device(DCR *dcr)
248 DEVICE *dev = dcr->dev;
250 Dmsg0(120, "start open_output_device()\n");
257 /* Defer opening files */
258 if (!dev->is_tape()) {
259 Dmsg0(129, "Device is file, deferring open.\n");
265 if (dev_cap(dev, CAP_STREAM)) {
266 mode = OPEN_WRITE_ONLY;
268 mode = OPEN_READ_WRITE;
270 Dmsg0(129, "Opening device.\n");
271 dev->open_nowait = true;
272 if (dev->open(dcr, mode) < 0) {
273 Emsg1(M_FATAL, 0, _("dev open failed: %s\n"), dev->errmsg);
274 dev->open_nowait = false;
278 Dmsg1(129, "open dev %s OK\n", dev->print_name());
279 dev->open_nowait = false;
285 * Make sure device is open, if not do so
287 bool open_device(DCR *dcr)
289 DEVICE *dev = dcr->dev;
292 if (dev_cap(dev, CAP_STREAM)) {
293 mode = OPEN_WRITE_ONLY;
295 mode = OPEN_READ_WRITE;
297 if (dev->open(dcr, mode) < 0) {
298 /* If polling, ignore the error */
299 if ((!dev->poll) && (!dev->is_dvd())) {
300 Jmsg2(dcr->jcr, M_FATAL, 0, _("Unable to open device %s: ERR=%s\n"),
301 dev->print_name(), strerror_dev(dev));
302 Pmsg2(000, "Unable to open archive %s: ERR=%s\n",
303 dev->print_name(), strerror_dev(dev));
311 * Release any Volume attached to this device
312 * then close the device.
314 void close_device(DEVICE *dev)
321 * Used when unmounting the device, ignore use_count
323 void force_close_device(DEVICE *dev)
326 Mmsg0(dev->errmsg, _("Bad call to force_close_dev. Device not open\n"));
327 Emsg0(M_FATAL, 0, dev->errmsg);
330 Dmsg1(29, "Force close_dev %s\n", dev->print_name());
335 ASSERT(dev->use_count >= 0);
340 void dev_lock(DEVICE *dev)
343 if ((errstat=rwl_writelock(&dev->lock))) {
344 Emsg1(M_ABORT, 0, "Device write lock failure. ERR=%s\n", strerror(errstat));
348 void dev_unlock(DEVICE *dev)
351 if ((errstat=rwl_writeunlock(&dev->lock))) {
352 Emsg1(M_ABORT, 0, "Device write unlock failure. ERR=%s\n", strerror(errstat));
357 * When dev_blocked is set, all threads EXCEPT thread with id no_wait_id
358 * must wait. The no_wait_id thread is out obtaining a new volume
359 * and preparing the label.
361 void _lock_device(const char *file, int line, DEVICE *dev)
364 Dmsg3(500, "lock %d from %s:%d\n", dev->dev_blocked, file, line);
366 if (dev->dev_blocked && !pthread_equal(dev->no_wait_id, pthread_self())) {
367 dev->num_waiting++; /* indicate that I am waiting */
368 while (dev->dev_blocked) {
369 if ((stat = pthread_cond_wait(&dev->wait, &dev->mutex)) != 0) {
371 Emsg1(M_ABORT, 0, _("pthread_cond_wait failure. ERR=%s\n"),
375 dev->num_waiting--; /* no longer waiting */
380 * Check if the device is blocked or not
382 bool is_device_unmounted(DEVICE *dev)
385 int blocked = dev->dev_blocked;
386 stat = (blocked == BST_UNMOUNTED) ||
387 (blocked == BST_UNMOUNTED_WAITING_FOR_SYSOP);
391 void _unlock_device(const char *file, int line, DEVICE *dev)
393 Dmsg2(500, "unlock from %s:%d\n", file, line);
398 * Block all other threads from using the device
399 * Device must already be locked. After this call,
400 * the device is blocked to any thread calling lock_device(),
401 * but the device is not locked (i.e. no P on device). Also,
402 * the current thread can do slip through the lock_device()
403 * calls without blocking.
405 void _block_device(const char *file, int line, DEVICE *dev, int state)
407 Dmsg3(500, "block set %d from %s:%d\n", state, file, line);
408 ASSERT(dev->get_blocked() == BST_NOT_BLOCKED);
409 dev->set_blocked(state); /* make other threads wait */
410 dev->no_wait_id = pthread_self(); /* allow us to continue */
416 * Unblock the device, and wake up anyone who went to sleep.
418 void _unblock_device(const char *file, int line, DEVICE *dev)
420 Dmsg3(500, "unblock %s from %s:%d\n", dev->print_blocked(), file, line);
421 ASSERT(dev->dev_blocked);
422 dev->set_blocked(BST_NOT_BLOCKED);
424 if (dev->num_waiting > 0) {
425 pthread_cond_broadcast(&dev->wait); /* wake them up */
430 * Enter with device locked and blocked
431 * Exit with device unlocked and blocked by us.
433 void _steal_device_lock(const char *file, int line, DEVICE *dev, bsteal_lock_t *hold, int state)
436 Dmsg3(400, "steal lock. old=%s from %s:%d\n", dev->print_blocked(),
438 hold->dev_blocked = dev->get_blocked();
439 hold->dev_prev_blocked = dev->dev_prev_blocked;
440 hold->no_wait_id = dev->no_wait_id;
441 dev->set_blocked(state);
442 Dmsg1(400, "steal lock. new=%s\n", dev->print_blocked());
443 dev->no_wait_id = pthread_self();
448 * Enter with device blocked by us but not locked
449 * Exit with device locked, and blocked by previous owner
451 void _give_back_device_lock(const char *file, int line, DEVICE *dev, bsteal_lock_t *hold)
453 Dmsg3(400, "return lock. old=%s from %s:%d\n",
454 dev->print_blocked(), file, line);
456 dev->dev_blocked = hold->dev_blocked;
457 dev->dev_prev_blocked = hold->dev_prev_blocked;
458 dev->no_wait_id = hold->no_wait_id;
459 Dmsg1(400, "return lock. new=%s\n", dev->print_blocked());
460 if (dev->num_waiting > 0) {
461 Dmsg0(400, "Broadcast\n");
462 pthread_cond_broadcast(&dev->wait); /* wake them up */