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: 1 on success
79 int fixup_device_block_write_error(JCR *jcr, DEVICE *dev, DEV_BLOCK *block)
82 char PrevVolName[MAX_NAME_LENGTH];
86 char dt[MAX_TIME_LENGTH];
88 wait_time = time(NULL);
89 stat = status_dev(dev);
90 if (!(stat & BMT_EOD)) {
91 return 0; /* this really shouldn't happen */
94 Dmsg0(100, "======= Got EOD ========\n");
96 block_device(dev, BST_DOING_ACQUIRE);
97 /* Unlock, but leave BLOCKED */
100 /* Create a jobmedia record for this job */
101 if (!dir_create_jobmedia_record(jcr)) {
102 Jmsg(jcr, M_ERROR, 0, _("Could not create JobMedia record for Volume=\"%s\" Job=%s\n"),
103 jcr->VolCatInfo.VolCatName, jcr->Job);
109 bstrncpy(dev->VolCatInfo.VolCatStatus, "Full", sizeof(dev->VolCatInfo.VolCatStatus));
110 Dmsg2(100, "Call update_vol_info Stat=%s Vol=%s\n",
111 dev->VolCatInfo.VolCatStatus, dev->VolCatInfo.VolCatName);
112 dev->VolCatInfo.VolCatFiles = dev->file; /* set number of files */
113 dev->VolCatInfo.VolCatJobs++; /* increment number of jobs */
114 if (!dir_update_volume_info(jcr, dev, 0)) { /* send Volume info to Director */
117 return 0; /* device locked */
119 Dmsg0(100, "Back from update_vol_info\n");
121 bstrncpy(PrevVolName, dev->VolCatInfo.VolCatName, sizeof(PrevVolName));
122 bstrncpy(dev->VolHdr.PrevVolName, PrevVolName, sizeof(dev->VolHdr.PrevVolName));
124 label_blk = new_block(dev);
126 /* Inform User about end of medium */
127 Jmsg(jcr, M_INFO, 0, _("End of medium on Volume \"%s\" Bytes=%s Blocks=%s at %s.\n"),
128 PrevVolName, edit_uint64_with_commas(dev->VolCatInfo.VolCatBytes, b1),
129 edit_uint64_with_commas(dev->VolCatInfo.VolCatBlocks, b2),
130 bstrftime(dt, sizeof(dt), time(NULL)));
132 if (!mount_next_write_volume(jcr, dev, label_blk, 1)) {
133 free_block(label_blk);
136 return 0; /* device locked */
138 P(dev->mutex); /* lock again */
140 Jmsg(jcr, M_INFO, 0, _("New volume \"%s\" mounted on device %s at %s.\n"),
141 jcr->VolumeName, dev_name(dev), bstrftime(dt, sizeof(dt), time(NULL)));
144 * If this is a new tape, the label_blk will contain the
145 * label, so write it now. If this is a previously
146 * used tape, mount_next_write_volume() will return an
147 * empty label_blk, and nothing will be written.
149 Dmsg0(190, "write label block to dev\n");
150 if (!write_block_to_dev(jcr->dcr, label_blk)) {
151 Pmsg1(0, "write_block_to_device Volume label failed. ERR=%s",
153 free_block(label_blk);
155 return 0; /* device locked */
157 free_block(label_blk);
160 * Walk through all attached jcrs indicating the volume has changed
162 Dmsg1(100, "Walk attached jcrs. Volume=%s\n", dev->VolCatInfo.VolCatName);
163 for (JCR *mjcr=NULL; (mjcr=next_attached_jcr(dev, mjcr)); ) {
164 if (mjcr->JobId == 0) {
165 continue; /* ignore console */
167 mjcr->dcr->NewVol = true;
169 pm_strcpy(&mjcr->VolumeName, jcr->VolumeName); /* get a copy of the new volume */
170 bstrncpy(mjcr->dcr->VolumeName, jcr->VolumeName, sizeof(mjcr->dcr->VolumeName));
174 /* Clear NewVol now because dir_get_volume_info() already done */
175 jcr->dcr->NewVol = false;
176 set_new_volume_parameters(jcr, dev);
178 jcr->run_time += time(NULL) - wait_time; /* correct run time for mount wait */
180 /* Write overflow block to device */
181 Dmsg0(190, "Write overflow block to dev\n");
182 if (!write_block_to_dev(jcr->dcr, block)) {
183 Pmsg1(0, "write_block_to_device overflow block failed. ERR=%s",
186 return 0; /* device locked */
190 return 1; /* device locked */
194 * We have a new Volume mounted, so reset the Volume parameters
195 * concerning this job. The global changes were made earlier
196 * in the dev structure.
198 void set_new_volume_parameters(JCR *jcr, DEVICE *dev)
201 if (dcr->NewVol && !dir_get_volume_info(jcr, GET_VOL_INFO_FOR_WRITE)) {
202 Jmsg1(jcr, M_ERROR, 0, "%s", jcr->errmsg);
204 /* Set new start/end positions */
205 if (dev_state(dev, ST_TAPE)) {
206 dcr->StartBlock = dev->block_num;
207 dcr->StartFile = dev->file;
209 dcr->StartBlock = (uint32_t)dev->file_addr;
210 dcr->StartFile = (uint32_t)(dev->file_addr >> 32);
213 dcr->VolFirstIndex = 0;
214 dcr->VolLastIndex = 0;
217 dcr->WroteVol = false;
221 * We are now in a new Volume file, so reset the Volume parameters
222 * concerning this job. The global changes were made earlier
223 * in the dev structure.
225 void set_new_file_parameters(JCR *jcr, DEVICE *dev)
229 /* Set new start/end positions */
230 if (dev_state(dev, ST_TAPE)) {
231 dcr->StartBlock = dev->block_num;
232 dcr->StartFile = dev->file;
234 dcr->StartBlock = (uint32_t)dev->file_addr;
235 dcr->StartFile = (uint32_t)(dev->file_addr >> 32);
238 dcr->VolFirstIndex = 0;
239 dcr->VolLastIndex = 0;
240 dcr->NewFile = false;
241 dcr->WroteVol = false;
247 * First Open of the device. Expect dev to already be initialized.
249 * This routine is used only when the Storage daemon starts
250 * and always_open is set, and in the stand-alone utility
251 * routines such as bextract.
253 * Note, opening of a normal file is deferred to later so
254 * that we can get the filename; the device_name for
255 * a file is the directory only.
257 * Retuns: 0 on failure
260 int first_open_device(DEVICE *dev)
262 Dmsg0(120, "start open_output_device()\n");
269 /* Defer opening files */
270 if (!dev_is_tape(dev)) {
271 Dmsg0(129, "Device is file, deferring open.\n");
276 if (!(dev->state & ST_OPENED)) {
278 if (dev_cap(dev, CAP_STREAM)) {
279 mode = OPEN_WRITE_ONLY;
281 mode = OPEN_READ_WRITE;
283 Dmsg0(129, "Opening device.\n");
284 if (open_dev(dev, NULL, mode) < 0) {
285 Emsg1(M_FATAL, 0, _("dev open failed: %s\n"), dev->errmsg);
290 Dmsg1(129, "open_dev %s OK\n", dev_name(dev));
297 * Make sure device is open, if not do so
299 int open_device(JCR *jcr, DEVICE *dev)
302 if (!(dev_state(dev, ST_OPENED))) {
304 if (dev_cap(dev, CAP_STREAM)) {
305 mode = OPEN_WRITE_ONLY;
307 mode = OPEN_READ_WRITE;
309 if (open_dev(dev, jcr->VolCatInfo.VolCatName, mode) < 0) {
310 Jmsg2(jcr, M_FATAL, 0, _("Unable to open device %s. ERR=%s\n"),
311 dev_name(dev), strerror_dev(dev));
319 * When dev_blocked is set, all threads EXCEPT thread with id no_wait_id
320 * must wait. The no_wait_id thread is out obtaining a new volume
321 * and preparing the label.
323 void _lock_device(char *file, int line, DEVICE *dev)
326 Dmsg3(500, "lock %d from %s:%d\n", dev->dev_blocked, file, line);
328 if (dev->dev_blocked && !pthread_equal(dev->no_wait_id, pthread_self())) {
329 dev->num_waiting++; /* indicate that I am waiting */
330 while (dev->dev_blocked) {
331 if ((stat = pthread_cond_wait(&dev->wait, &dev->mutex)) != 0) {
333 Emsg1(M_ABORT, 0, _("pthread_cond_wait failure. ERR=%s\n"),
337 dev->num_waiting--; /* no longer waiting */
342 * Check if the device is blocked or not
344 int device_is_unmounted(DEVICE *dev)
348 stat = (dev->dev_blocked == BST_UNMOUNTED) ||
349 (dev->dev_blocked == BST_UNMOUNTED_WAITING_FOR_SYSOP);
354 void _unlock_device(char *file, int line, DEVICE *dev)
356 Dmsg2(500, "unlock from %s:%d\n", file, line);
361 * Block all other threads from using the device
362 * Device must already be locked. After this call,
363 * the device is blocked to any thread calling lock_device(),
364 * but the device is not locked (i.e. no P on device). Also,
365 * the current thread can do slip through the lock_device()
366 * calls without blocking.
368 void _block_device(char *file, int line, DEVICE *dev, int state)
370 Dmsg3(500, "block set %d from %s:%d\n", state, file, line);
371 ASSERT(dev->dev_blocked == BST_NOT_BLOCKED);
372 dev->dev_blocked = state; /* make other threads wait */
373 dev->no_wait_id = pthread_self(); /* allow us to continue */
379 * Unblock the device, and wake up anyone who went to sleep.
381 void _unblock_device(char *file, int line, DEVICE *dev)
383 Dmsg3(500, "unblock %d from %s:%d\n", dev->dev_blocked, file, line);
384 ASSERT(dev->dev_blocked);
385 dev->dev_blocked = BST_NOT_BLOCKED;
387 if (dev->num_waiting > 0) {
388 pthread_cond_broadcast(&dev->wait); /* wake them up */
393 * Enter with device locked and blocked
394 * Exit with device unlocked and blocked by us.
396 void _steal_device_lock(char *file, int line, DEVICE *dev, bsteal_lock_t *hold, int state)
398 Dmsg4(500, "steal lock. old=%d new=%d from %s:%d\n", dev->dev_blocked, state,
400 hold->dev_blocked = dev->dev_blocked;
401 hold->dev_prev_blocked = dev->dev_prev_blocked;
402 hold->no_wait_id = dev->no_wait_id;
403 dev->dev_blocked = state;
404 dev->no_wait_id = pthread_self();
409 * Enter with device blocked by us but not locked
410 * Exit with device locked, and blocked by previous owner
412 void _give_back_device_lock(char *file, int line, DEVICE *dev, bsteal_lock_t *hold)
414 Dmsg4(500, "return lock. old=%d new=%d from %s:%d\n",
415 dev->dev_blocked, hold->dev_blocked, file, line);
417 dev->dev_blocked = hold->dev_blocked;
418 dev->dev_prev_blocked = hold->dev_prev_blocked;
419 dev->no_wait_id = hold->no_wait_id;