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 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 * dcr->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 * dcr->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)
82 char PrevVolName[MAX_NAME_LENGTH];
84 DEV_BLOCK *block = dcr->block;
87 char dt[MAX_TIME_LENGTH];
89 DEVICE *dev = dcr->dev;
91 wait_time = time(NULL);
92 stat = status_dev(dev);
93 if (!(stat & BMT_EOD)) {
94 return false; /* this really shouldn't happen */
97 Dmsg0(100, "======= Got EOD ========\n");
99 block_device(dev, BST_DOING_ACQUIRE);
100 /* Unlock, but leave BLOCKED */
103 bstrncpy(PrevVolName, dev->VolCatInfo.VolCatName, sizeof(PrevVolName));
104 bstrncpy(dev->VolHdr.PrevVolName, PrevVolName, sizeof(dev->VolHdr.PrevVolName));
106 label_blk = new_block(dev);
107 dcr->block = label_blk;
109 /* Inform User about end of medium */
110 Jmsg(jcr, M_INFO, 0, _("End of medium on Volume \"%s\" Bytes=%s Blocks=%s at %s.\n"),
111 PrevVolName, edit_uint64_with_commas(dev->VolCatInfo.VolCatBytes, b1),
112 edit_uint64_with_commas(dev->VolCatInfo.VolCatBlocks, b2),
113 bstrftime(dt, sizeof(dt), time(NULL)));
115 if (!mount_next_write_volume(dcr, 1)) {
116 free_block(label_blk);
120 return false; /* device locked */
122 P(dev->mutex); /* lock again */
124 Jmsg(jcr, M_INFO, 0, _("New volume \"%s\" mounted on device %s at %s.\n"),
125 dcr->VolumeName, dev->print_name(), bstrftime(dt, sizeof(dt), time(NULL)));
128 * If this is a new tape, the label_blk will contain the
129 * label, so write it now. If this is a previously
130 * used tape, mount_next_write_volume() will return an
131 * empty label_blk, and nothing will be written.
133 Dmsg0(190, "write label block to dev\n");
134 if (!write_block_to_dev(dcr)) {
136 Pmsg1(0, "write_block_to_device Volume label failed. ERR=%s",
137 be.strerror(dev->dev_errno));
138 free_block(label_blk);
141 return false; /* device locked */
143 free_block(label_blk);
147 * Walk through all attached jcrs indicating the volume has changed
149 Dmsg1(100, "Walk attached jcrs. Volume=%s\n", dev->VolCatInfo.VolCatName);
150 // for (JCR *mjcr=NULL; (mjcr=next_attached_jcr(dev, mjcr)); ) {
152 foreach_dlist(mdcr, dev->attached_dcrs) {
153 JCR *mjcr = mdcr->jcr;
154 if (mjcr->JobId == 0) {
155 continue; /* ignore console */
159 bstrncpy(mdcr->VolumeName, dcr->VolumeName, sizeof(mdcr->VolumeName));
163 /* Clear NewVol now because dir_get_volume_info() already done */
164 jcr->dcr->NewVol = false;
165 set_new_volume_parameters(dcr);
167 jcr->run_time += time(NULL) - wait_time; /* correct run time for mount wait */
169 /* Write overflow block to device */
170 Dmsg0(190, "Write overflow block to dev\n");
171 if (!write_block_to_dev(dcr)) {
173 Pmsg1(0, "write_block_to_device overflow block failed. ERR=%s",
174 be.strerror(dev->dev_errno));
176 return false; /* device locked */
180 return true; /* device locked */
184 * We have a new Volume mounted, so reset the Volume parameters
185 * concerning this job. The global changes were made earlier
186 * in the dev structure.
188 void set_new_volume_parameters(DCR *dcr)
191 DEVICE *dev = dcr->dev;
192 if (dcr->NewVol && !dir_get_volume_info(dcr, GET_VOL_INFO_FOR_WRITE)) {
193 Jmsg1(jcr, M_ERROR, 0, "%s", jcr->errmsg);
195 /* Set new start/end positions */
196 if (dev_state(dev, ST_TAPE)) {
197 dcr->StartBlock = dev->block_num;
198 dcr->StartFile = dev->file;
200 dcr->StartBlock = (uint32_t)dev->file_addr;
201 dcr->StartFile = (uint32_t)(dev->file_addr >> 32);
204 dcr->VolFirstIndex = 0;
205 dcr->VolLastIndex = 0;
208 dcr->WroteVol = false;
212 * We are now in a new Volume file, so reset the Volume parameters
213 * concerning this job. The global changes were made earlier
214 * in the dev structure.
216 void set_new_file_parameters(DCR *dcr)
218 DEVICE *dev = dcr->dev;
220 /* Set new start/end positions */
221 if (dev_state(dev, ST_TAPE)) {
222 dcr->StartBlock = dev->block_num;
223 dcr->StartFile = dev->file;
225 dcr->StartBlock = (uint32_t)dev->file_addr;
226 dcr->StartFile = (uint32_t)(dev->file_addr >> 32);
229 dcr->VolFirstIndex = 0;
230 dcr->VolLastIndex = 0;
231 dcr->NewFile = false;
232 dcr->WroteVol = false;
238 * First Open of the device. Expect dev to already be initialized.
240 * This routine is used only when the Storage daemon starts
241 * and always_open is set, and in the stand-alone utility
242 * routines such as bextract.
244 * Note, opening of a normal file is deferred to later so
245 * that we can get the filename; the device_name for
246 * a file is the directory only.
248 * Returns: false on failure
251 bool first_open_device(DEVICE *dev)
253 Dmsg0(120, "start open_output_device()\n");
260 /* Defer opening files */
261 if (!dev->is_tape()) {
262 Dmsg0(129, "Device is file, deferring open.\n");
267 if (!dev->is_open()) {
269 if (dev_cap(dev, CAP_STREAM)) {
270 mode = OPEN_WRITE_ONLY;
272 mode = OPEN_READ_WRITE;
274 Dmsg0(129, "Opening device.\n");
275 dev->open_nowait = true;
276 if (open_dev(dev, NULL, mode) < 0) {
277 Emsg1(M_FATAL, 0, _("dev open failed: %s\n"), dev->errmsg);
278 dev->open_nowait = false;
283 Dmsg1(129, "open_dev %s OK\n", dev->print_name());
284 dev->open_nowait = false;
290 * Make sure device is open, if not do so
292 bool open_device(DCR *dcr)
294 DEVICE *dev = dcr->dev;
296 if (!dev->is_open()) {
298 if (dev_cap(dev, CAP_STREAM)) {
299 mode = OPEN_WRITE_ONLY;
301 mode = OPEN_READ_WRITE;
303 if (open_dev(dev, dcr->VolCatInfo.VolCatName, mode) < 0) {
304 /* If polling, ignore the error */
306 Jmsg2(dcr->jcr, M_FATAL, 0, _("Unable to open device %s: ERR=%s\n"),
307 dev->print_name(), strerror_dev(dev));
308 Pmsg2(000, "Unable to open archive %s: ERR=%s\n",
309 dev->print_name(), strerror_dev(dev));
317 void dev_lock(DEVICE *dev)
320 if ((errstat=rwl_writelock(&dev->lock))) {
321 Emsg1(M_ABORT, 0, "Device write lock failure. ERR=%s\n", strerror(errstat));
325 void dev_unlock(DEVICE *dev)
328 if ((errstat=rwl_writeunlock(&dev->lock))) {
329 Emsg1(M_ABORT, 0, "Device write unlock failure. ERR=%s\n", strerror(errstat));
334 * When dev_blocked is set, all threads EXCEPT thread with id no_wait_id
335 * must wait. The no_wait_id thread is out obtaining a new volume
336 * and preparing the label.
338 void _lock_device(const char *file, int line, DEVICE *dev)
341 Dmsg3(500, "lock %d from %s:%d\n", dev->dev_blocked, file, line);
343 if (dev->dev_blocked && !pthread_equal(dev->no_wait_id, pthread_self())) {
344 dev->num_waiting++; /* indicate that I am waiting */
345 while (dev->dev_blocked) {
346 if ((stat = pthread_cond_wait(&dev->wait, &dev->mutex)) != 0) {
348 Emsg1(M_ABORT, 0, _("pthread_cond_wait failure. ERR=%s\n"),
352 dev->num_waiting--; /* no longer waiting */
357 * Check if the device is blocked or not
359 bool device_is_unmounted(DEVICE *dev)
362 int blocked = dev->dev_blocked;
363 stat = (blocked == BST_UNMOUNTED) ||
364 (blocked == BST_UNMOUNTED_WAITING_FOR_SYSOP);
368 const char *edit_blocked_reason(DEVICE *dev)
370 switch (dev->dev_blocked) {
371 case BST_NOT_BLOCKED:
372 return "not blocked";
374 return "user unmounted device";
375 case BST_WAITING_FOR_SYSOP:
376 return "waiting for operator action";
377 case BST_DOING_ACQUIRE:
378 return "opening, validating, or positioning tape";
379 case BST_WRITING_LABEL:
380 return "labeling tape";
381 case BST_UNMOUNTED_WAITING_FOR_SYSOP:
382 return "closed by user during mount request";
384 return "mount request";
386 return "unknown blocked code";
390 void _unlock_device(const char *file, int line, DEVICE *dev)
392 Dmsg2(500, "unlock from %s:%d\n", file, line);
397 * Block all other threads from using the device
398 * Device must already be locked. After this call,
399 * the device is blocked to any thread calling lock_device(),
400 * but the device is not locked (i.e. no P on device). Also,
401 * the current thread can do slip through the lock_device()
402 * calls without blocking.
404 void _block_device(const char *file, int line, DEVICE *dev, int state)
406 Dmsg3(500, "block set %d from %s:%d\n", state, file, line);
407 ASSERT(dev->dev_blocked == BST_NOT_BLOCKED);
408 dev->dev_blocked = state; /* make other threads wait */
409 dev->no_wait_id = pthread_self(); /* allow us to continue */
415 * Unblock the device, and wake up anyone who went to sleep.
417 void _unblock_device(const char *file, int line, DEVICE *dev)
419 Dmsg3(500, "unblock %d from %s:%d\n", dev->dev_blocked, file, line);
420 ASSERT(dev->dev_blocked);
421 dev->dev_blocked = BST_NOT_BLOCKED;
423 if (dev->num_waiting > 0) {
424 pthread_cond_broadcast(&dev->wait); /* wake them up */
429 * Enter with device locked and blocked
430 * Exit with device unlocked and blocked by us.
432 void _steal_device_lock(const char *file, int line, DEVICE *dev, bsteal_lock_t *hold, int state)
434 Dmsg4(500, "steal lock. old=%d new=%d from %s:%d\n", dev->dev_blocked, state,
436 hold->dev_blocked = dev->dev_blocked;
437 hold->dev_prev_blocked = dev->dev_prev_blocked;
438 hold->no_wait_id = dev->no_wait_id;
439 dev->dev_blocked = state;
440 dev->no_wait_id = pthread_self();
445 * Enter with device blocked by us but not locked
446 * Exit with device locked, and blocked by previous owner
448 void _give_back_device_lock(const char *file, int line, DEVICE *dev, bsteal_lock_t *hold)
450 Dmsg4(500, "return lock. old=%d new=%d from %s:%d\n",
451 dev->dev_blocked, hold->dev_blocked, file, line);
453 dev->dev_blocked = hold->dev_blocked;
454 dev->dev_prev_blocked = hold->dev_prev_blocked;
455 dev->no_wait_id = hold->no_wait_id;
456 if (dev->dev_blocked == BST_NOT_BLOCKED && dev->num_waiting > 0) {
457 pthread_cond_broadcast(&dev->wait); /* wake them up */