]> git.wh0rd.org - dump.git/blob - dump/tape.c
Discard newline when reading the tape path entered by the operator in restore.
[dump.git] / dump / tape.c
1 /*
2 * Ported to Linux's Second Extended File System as part of the
3 * dump and restore backup suit
4 * Remy Card <card@Linux.EU.Org>, 1994-1997
5 * Stelian Pop <pop@noos.fr>, 1999-2000
6 * Stelian Pop <pop@noos.fr> - AlcĂ´ve <www.alcove.fr>, 2000
7 */
8
9 /*-
10 * Copyright (c) 1980, 1991, 1993
11 * The Regents of the University of California. All rights reserved.
12 *
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
15 * are met:
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. All advertising materials mentioning features or use of this software
22 * must display the following acknowledgement:
23 * This product includes software developed by the University of
24 * California, Berkeley and its contributors.
25 * 4. Neither the name of the University nor the names of its contributors
26 * may be used to endorse or promote products derived from this software
27 * without specific prior written permission.
28 *
29 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
30 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
31 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
32 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
33 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
36 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
37 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
38 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39 * SUCH DAMAGE.
40 */
41
42 #ifndef lint
43 static const char rcsid[] =
44 "$Id: tape.c,v 1.49 2001/07/18 09:50:48 stelian Exp $";
45 #endif /* not lint */
46
47 #include <config.h>
48 #include <compatlfs.h>
49 #include <errno.h>
50 #include <fcntl.h>
51 #include <setjmp.h>
52 #include <signal.h>
53 #include <stdio.h>
54 #include <compaterr.h>
55 #ifdef __STDC__
56 #include <stdlib.h>
57 #include <string.h>
58 #include <unistd.h>
59 #else
60 int write(), read();
61 #endif
62
63 #ifdef __linux__
64 #include <sys/types.h>
65 #include <sys/time.h>
66 #include <time.h>
67 #endif
68 #include <sys/param.h>
69 #include <sys/socket.h>
70 #include <sys/time.h>
71 #include <sys/wait.h>
72 #include <sys/mtio.h>
73 #ifdef __linux__
74 #ifdef HAVE_EXT2FS_EXT2_FS_H
75 #include <ext2fs/ext2_fs.h>
76 #else
77 #include <linux/ext2_fs.h>
78 #endif
79 #include <ext2fs/ext2fs.h>
80 #include <bsdcompat.h>
81 #elif defined sunos
82 #include <sys/vnode.h>
83
84 #include <ufs/fs.h>
85 #include <ufs/inode.h>
86 #else
87 #include <ufs/ufs/dinode.h>
88 #include <ufs/ffs/fs.h>
89 #endif /* __linux__ */
90
91 #include <protocols/dumprestore.h>
92
93 #ifdef HAVE_ZLIB
94 #include <zlib.h>
95 #endif /* HAVE_ZLIB */
96
97 #include "dump.h"
98
99 int writesize; /* size of malloc()ed buffer for tape */
100 long lastspclrec = -1; /* tape block number of last written header */
101 int trecno = 0; /* next record to write in current block */
102 extern long blocksperfile; /* number of blocks per output file */
103 long blocksthisvol; /* number of blocks on current output file */
104 extern int ntrec; /* blocking factor on tape */
105 extern int cartridge;
106 extern char *host;
107 char *nexttape;
108 extern pid_t rshpid;
109 int eot_code = 1;
110 long long tapea_bytes = 0; /* bytes_written at start of current volume */
111
112 static ssize_t atomic_read __P((int, void *, size_t));
113 static ssize_t atomic_write __P((int, const void *, size_t));
114 static void doslave __P((int, int));
115 static void enslave __P((void));
116 static void flushtape __P((void));
117 static void killall __P((void));
118 static void rollforward __P((void));
119 static int system_command __P((const char *, const char *, int));
120
121 /*
122 * Concurrent dump mods (Caltech) - disk block reading and tape writing
123 * are exported to several slave processes. While one slave writes the
124 * tape, the others read disk blocks; they pass control of the tape in
125 * a ring via signals. The parent process traverses the filesystem and
126 * sends writeheader()'s and lists of daddr's to the slaves via pipes.
127 * The following structure defines the instruction packets sent to slaves.
128 */
129 struct req {
130 daddr_t dblk;
131 int count;
132 };
133 int reqsiz;
134
135 struct slave_results {
136 ssize_t unclen; /* uncompressed length */
137 ssize_t clen; /* compressed length */
138 };
139
140 #define SLAVES 3 /* 1 slave writing, 1 reading, 1 for slack */
141 struct slave {
142 int tapea; /* header number at start of this chunk */
143 int count; /* count to next header (used for TS_TAPE */
144 /* after EOT) */
145 int inode; /* inode that we are currently dealing with */
146 int fd; /* FD for this slave */
147 int pid; /* PID for this slave */
148 int sent; /* 1 == we've sent this slave requests */
149 int firstrec; /* record number of this block */
150 char (*tblock)[TP_BSIZE]; /* buffer for data blocks */
151 struct req *req; /* buffer for requests */
152 } slaves[SLAVES+1];
153 struct slave *slp;
154
155 char (*nextblock)[TP_BSIZE];
156
157 static time_t tstart_volume; /* time of volume start */
158 static int tapea_volume; /* value of spcl.c_tapea at volume start */
159
160 int master; /* pid of master, for sending error signals */
161 int tenths; /* length of tape overhead per block written */
162 static int caught; /* have we caught the signal to proceed? */
163 static int ready; /* have we reached the lock point without having */
164 /* received the SIGUSR2 signal from the prev slave? */
165 static sigjmp_buf jmpbuf; /* where to jump to if we are ready when the */
166 /* SIGUSR2 arrives from the previous slave */
167 #ifdef USE_QFA
168 static int gtperr = 0;
169 #endif
170
171 int
172 alloctape(void)
173 {
174 int pgoff = getpagesize() - 1;
175 char *buf;
176 int i;
177
178 writesize = ntrec * TP_BSIZE;
179 reqsiz = (ntrec + 1) * sizeof(struct req);
180 /*
181 * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode
182 * (see DEC TU80 User's Guide). The shorter gaps of 6250-bpi require
183 * repositioning after stopping, i.e, streaming mode, where the gap is
184 * variable, 0.30" to 0.45". The gap is maximal when the tape stops.
185 */
186 if (blocksperfile == 0 && !unlimited)
187 tenths = (cartridge ? 16 : density == 625 ? 5 : 8);
188 else {
189 tenths = 0;
190 density = 1;
191 }
192 /*
193 * Allocate tape buffer contiguous with the array of instruction
194 * packets, so flushtape() can write them together with one write().
195 * Align tape buffer on page boundary to speed up tape write().
196 */
197 for (i = 0; i <= SLAVES; i++) {
198 buf = (char *)
199 malloc((unsigned)(reqsiz + writesize + pgoff + TP_BSIZE));
200 if (buf == NULL)
201 return(0);
202 slaves[i].tblock = (char (*)[TP_BSIZE])
203 #ifdef __linux__
204 (((long)&buf[reqsiz] + pgoff) &~ pgoff);
205 #else
206 (((long)&buf[ntrec + 1] + pgoff) &~ pgoff);
207 #endif
208 slaves[i].req = (struct req *)slaves[i].tblock - ntrec - 1;
209 }
210 slp = &slaves[0];
211 slp->count = 1;
212 slp->tapea = 0;
213 slp->firstrec = 0;
214 nextblock = slp->tblock;
215 return(1);
216 }
217
218 void
219 writerec(const void *dp, int isspcl)
220 {
221
222 slp->req[trecno].dblk = (daddr_t)0;
223 slp->req[trecno].count = 1;
224 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
225 *(union u_spcl *)(*(nextblock)) = *(union u_spcl *)dp;
226 nextblock++;
227 if (isspcl)
228 lastspclrec = spcl.c_tapea;
229 trecno++;
230 spcl.c_tapea++;
231 if (trecno >= ntrec)
232 flushtape();
233 }
234
235 void
236 dumpblock(daddr_t blkno, int size)
237 {
238 int avail, tpblks, dblkno;
239
240 dblkno = fsbtodb(sblock, blkno);
241 tpblks = size >> tp_bshift;
242 while ((avail = MIN(tpblks, ntrec - trecno)) > 0) {
243 slp->req[trecno].dblk = dblkno;
244 slp->req[trecno].count = avail;
245 trecno += avail;
246 spcl.c_tapea += avail;
247 if (trecno >= ntrec)
248 flushtape();
249 dblkno += avail << (tp_bshift - dev_bshift);
250 tpblks -= avail;
251 }
252 }
253
254 int nogripe = 0;
255
256 static void
257 tperror(int errnum)
258 {
259
260 if (pipeout) {
261 msg("write error on %s: %s\n", tape, strerror(errnum));
262 quit("Cannot recover\n");
263 /* NOTREACHED */
264 }
265 msg("write error %d blocks into volume %d: %s\n",
266 blocksthisvol, tapeno, strerror(errnum));
267 broadcast("DUMP WRITE ERROR!\n");
268 if (query("Do you want to rewrite this volume?")) {
269 msg("Closing this volume. Prepare to restart with new media;\n");
270 msg("this dump volume will be rewritten.\n");
271 killall();
272 nogripe = 1;
273 close_rewind();
274 Exit(X_REWRITE);
275 }
276 if (query("Do you want to start the next tape?"))
277 return;
278 dumpabort(0);
279 }
280
281 static void
282 sigpipe(int signo)
283 {
284
285 quit("Broken pipe\n");
286 }
287
288 /*
289 * do_stats --
290 * Update xferrate stats
291 */
292 time_t
293 do_stats(void)
294 {
295 time_t tnow, ttaken;
296 int blocks;
297
298 tnow = time(NULL);
299 ttaken = tnow - tstart_volume;
300 blocks = spcl.c_tapea - tapea_volume;
301 msg("Volume %d completed at: %s", tapeno, ctime(&tnow));
302 if (! compressed)
303 msg("Volume %d %ld tape blocks (%.2fMB)\n", tapeno,
304 blocks, ((double)blocks * TP_BSIZE / 1048576));
305 if (ttaken > 0) {
306 long volkb = (bytes_written - tapea_bytes) / 1024;
307 long txfrate = volkb / ttaken;
308 msg("Volume %d took %d:%02d:%02d\n", tapeno,
309 ttaken / 3600, (ttaken % 3600) / 60, ttaken % 60);
310 msg("Volume %d transfer rate: %ld kB/s\n", tapeno,
311 txfrate);
312 xferrate += txfrate;
313 if (compressed) {
314 double rate = .0005 + (double) blocks / (double) volkb;
315 msg("Volume %d %ldkB uncompressed, %ldkB compressed,"
316 " %1.3f:1\n",
317 tapeno, blocks, volkb, rate);
318 }
319 }
320 return(tnow);
321 }
322
323 char *
324 mktimeest(time_t tnow)
325 {
326 static char msgbuf[128];
327 time_t deltat;
328
329 msgbuf[0] = '\0';
330
331 if (blockswritten < 500)
332 return NULL;
333 if (blockswritten > tapesize)
334 tapesize = blockswritten;
335 deltat = tstart_writing - tnow + (1.0 * (tnow - tstart_writing))
336 / blockswritten * tapesize;
337 if (tnow > tstart_volume)
338 (void)snprintf(msgbuf, sizeof(msgbuf),
339 "%3.2f%% done at %ld kB/s, finished in %d:%02d\n",
340 (blockswritten * 100.0) / tapesize,
341 (spcl.c_tapea - tapea_volume) / (tnow - tstart_volume),
342 (int)(deltat / 3600), (int)((deltat % 3600) / 60));
343 else
344 (void)snprintf(msgbuf, sizeof(msgbuf),
345 "%3.2f%% done, finished in %d:%02d\n",
346 (blockswritten * 100.0) / tapesize,
347 (int)(deltat / 3600), (int)((deltat % 3600) / 60));
348
349 return msgbuf;
350 }
351
352 #if defined(SIGINFO)
353 /*
354 * statussig --
355 * information message upon receipt of SIGINFO
356 */
357 void
358 statussig(int notused)
359 {
360 int save_errno = errno;
361 char *buf;
362
363 buf = mktimeest(time(NULL));
364 if (buf)
365 write(STDERR_FILENO, buf, strlen(buf));
366 errno = save_errno;
367 }
368 #endif
369
370 static void
371 flushtape(void)
372 {
373 int i, blks, got;
374 long lastfirstrec;
375 struct slave_results returned;
376
377 int siz = (char *)nextblock - (char *)slp->req;
378
379 slp->req[trecno].count = 0; /* Sentinel */
380
381 if (atomic_write( slp->fd, (char *)slp->req, siz) != siz)
382 quit("error writing command pipe: %s\n", strerror(errno));
383 slp->sent = 1; /* we sent a request, read the response later */
384
385 lastfirstrec = slp->firstrec;
386
387 if (++slp >= &slaves[SLAVES])
388 slp = &slaves[0];
389
390 /* Read results back from next slave */
391 if (slp->sent) {
392 if (atomic_read( slp->fd, (char *)&returned, sizeof returned)
393 != sizeof returned) {
394 perror(" DUMP: error reading command pipe in master");
395 dumpabort(0);
396 }
397 got = returned.unclen;
398 bytes_written += returned.clen;
399 if (returned.unclen == returned.clen)
400 uncomprblks++;
401 slp->sent = 0;
402
403 /* Check for errors or end of tape */
404 if (got <= 0) {
405 /* Check for errors */
406 if (got < 0)
407 tperror(-got);
408 else
409 msg("End of tape detected\n");
410
411 /*
412 * Drain the results, don't care what the values were.
413 * If we read them here then trewind won't...
414 */
415 for (i = 0; i < SLAVES; i++) {
416 if (slaves[i].sent) {
417 if (atomic_read( slaves[i].fd,
418 (char *)&returned, sizeof returned)
419 != sizeof returned) {
420 perror(" DUMP: error reading command pipe in master");
421 dumpabort(0);
422 }
423 slaves[i].sent = 0;
424 }
425 }
426
427 close_rewind();
428 rollforward();
429 return;
430 }
431 }
432
433 blks = 0;
434 if (spcl.c_type != TS_END) {
435 for (i = 0; i < spcl.c_count; i++)
436 if (spcl.c_addr[i] != 0)
437 blks++;
438 }
439 slp->count = lastspclrec + blks + 1 - spcl.c_tapea;
440 slp->tapea = spcl.c_tapea;
441 slp->firstrec = lastfirstrec + ntrec;
442 slp->inode = curino;
443 nextblock = slp->tblock;
444 trecno = 0;
445 asize += tenths + returned.clen / density;
446 csize += returned.clen;
447 blockswritten += ntrec;
448 blocksthisvol += ntrec;
449 if (!pipeout && !unlimited) {
450 if (blocksperfile) {
451 if ( compressed ? csize >= blocksperfile * 1024
452 : blocksthisvol >= blocksperfile ) {
453 close_rewind();
454 startnewtape(0);
455 }
456 }
457 else if (asize > tsize) {
458 close_rewind();
459 startnewtape(0);
460 }
461 }
462 timeest();
463 }
464
465 /*
466 * Executes the command in a shell.
467 * Returns -1 if an error occured, the exit status of
468 * the command on success.
469 */
470 int system_command(const char *command, const char *device, int volnum) {
471 int pid, status;
472 char commandstr[4096];
473
474 pid = fork();
475 if (pid == -1) {
476 perror(" DUMP: unable to fork");
477 return -1;
478 }
479 if (pid == 0) {
480 setuid(getuid());
481 setgid(getgid());
482 #if OLD_STYLE_FSCRIPT
483 snprintf(commandstr, sizeof(commandstr), "%s", command);
484 #else
485 snprintf(commandstr, sizeof(commandstr), "%s %s %d", command, device, volnum);
486 #endif
487 commandstr[sizeof(commandstr) - 1] = '\0';
488 execl("/bin/sh", "sh", "-c", commandstr, NULL);
489 perror(" DUMP: unable to execute shell");
490 exit(-1);
491 }
492 do {
493 if (waitpid(pid, &status, 0) == -1) {
494 if (errno != EINTR) {
495 perror(" DUMP: waitpid error");
496 return -1;
497 }
498 } else {
499 if (WIFEXITED(status))
500 return WEXITSTATUS(status);
501 else
502 return -1;
503 }
504 } while(1);
505 }
506
507 time_t
508 trewind(void)
509 {
510 int f;
511 int got;
512 struct slave_results returned;
513
514 for (f = 0; f < SLAVES; f++) {
515 /*
516 * Drain the results, but unlike EOT we DO (or should) care
517 * what the return values were, since if we detect EOT after
518 * we think we've written the last blocks to the tape anyway,
519 * we have to replay those blocks with rollforward.
520 *
521 * fixme: punt for now.
522 */
523 if (slaves[f].sent) {
524 if (atomic_read( slaves[f].fd, (char *)&returned, sizeof returned)
525 != sizeof returned) {
526 perror(" DUMP: error reading command pipe in master");
527 dumpabort(0);
528 }
529 got = returned.unclen;
530 bytes_written += returned.clen;
531 if (returned.unclen == returned.clen)
532 uncomprblks++;
533 slaves[f].sent = 0;
534
535 if (got < 0)
536 tperror(-got);
537
538 if (got == 0) {
539 msg("EOT detected in last 2 tape records!\n");
540 msg("Use a longer tape, decrease the size estimate\n");
541 quit("or use no size estimate at all.\n");
542 }
543 }
544 (void) close(slaves[f].fd);
545 }
546 while (wait((int *)NULL) >= 0) /* wait for any signals from slaves */
547 /* void */;
548
549 if (!pipeout) {
550
551 msg("Closing %s\n", tape);
552
553 #ifdef RDUMP
554 if (host) {
555 rmtclose();
556 while (rmtopen(tape, 0) < 0)
557 sleep(10);
558 rmtclose();
559 }
560 else
561 #endif
562 {
563 (void) close(tapefd);
564 if (!fifoout) {
565 while ((f = OPEN(tape, 0)) < 0)
566 sleep (10);
567 (void) close(f);
568 }
569 }
570 eot_code = 1;
571 if (eot_script && spcl.c_type != TS_END) {
572 msg("Launching %s\n", eot_script);
573 eot_code = system_command(eot_script, tape, tapeno);
574 }
575 if (eot_code != 0 && eot_code != 1) {
576 msg("Dump aborted by the end of tape script\n");
577 dumpabort(0);
578 }
579 }
580 return do_stats();
581 }
582
583
584 void
585 close_rewind(void)
586 {
587 (void)trewind();
588 if (nexttape || Mflag || (eot_code == 0) )
589 return;
590 if (!nogripe) {
591 msg("Change Volumes: Mount volume #%d\n", tapeno+1);
592 broadcast("CHANGE DUMP VOLUMES!\7\7\n");
593 }
594 while (!query("Is the new volume mounted and ready to go?"))
595 if (query("Do you want to abort?")) {
596 dumpabort(0);
597 /*NOTREACHED*/
598 }
599 }
600
601 void
602 rollforward(void)
603 {
604 register struct req *p, *q, *prev;
605 register struct slave *tslp;
606 int i, size, savedtapea, got;
607 union u_spcl *ntb, *otb;
608 struct slave_results returned;
609 #ifdef __linux__
610 int blks;
611 long lastfirstrec;
612 #endif
613 tslp = &slaves[SLAVES];
614 ntb = (union u_spcl *)tslp->tblock[1];
615
616 /*
617 * Each of the N slaves should have requests that need to
618 * be replayed on the next tape. Use the extra slave buffers
619 * (slaves[SLAVES]) to construct request lists to be sent to
620 * each slave in turn.
621 */
622 for (i = 0; i < SLAVES; i++) {
623 q = &tslp->req[1];
624 otb = (union u_spcl *)slp->tblock;
625
626 /*
627 * For each request in the current slave, copy it to tslp.
628 */
629
630 prev = NULL;
631 for (p = slp->req; p->count > 0; p += p->count) {
632 *q = *p;
633 if (p->dblk == 0)
634 *ntb++ = *otb++; /* copy the datablock also */
635 prev = q;
636 q += q->count;
637 }
638 if (prev == NULL)
639 quit("rollforward: protocol botch");
640 if (prev->dblk != 0)
641 prev->count -= 1;
642 else
643 ntb--;
644 q -= 1;
645 q->count = 0;
646 q = &tslp->req[0];
647 if (i == 0) {
648 q->dblk = 0;
649 q->count = 1;
650 trecno = 0;
651 nextblock = tslp->tblock;
652 savedtapea = spcl.c_tapea;
653 spcl.c_tapea = slp->tapea;
654 startnewtape(0);
655 spcl.c_tapea = savedtapea;
656 lastspclrec = savedtapea - 1;
657 }
658 size = (char *)ntb - (char *)q;
659 if (atomic_write( slp->fd, (char *)q, size) != size) {
660 perror(" DUMP: error writing command pipe");
661 dumpabort(0);
662 }
663 slp->sent = 1;
664 #ifdef __linux__
665 lastfirstrec = slp->firstrec;
666 #endif
667 if (++slp >= &slaves[SLAVES])
668 slp = &slaves[0];
669
670 q->count = 1;
671
672 if (prev->dblk != 0) {
673 /*
674 * If the last one was a disk block, make the
675 * first of this one be the last bit of that disk
676 * block...
677 */
678 q->dblk = prev->dblk +
679 prev->count * (TP_BSIZE / DEV_BSIZE);
680 ntb = (union u_spcl *)tslp->tblock;
681 } else {
682 /*
683 * It wasn't a disk block. Copy the data to its
684 * new location in the buffer.
685 */
686 q->dblk = 0;
687 *((union u_spcl *)tslp->tblock) = *ntb;
688 ntb = (union u_spcl *)tslp->tblock[1];
689 }
690 }
691 slp->req[0] = *q;
692 nextblock = slp->tblock;
693 if (q->dblk == 0) {
694 #ifdef __linux__
695 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
696 *(union u_spcl *)(*nextblock) = *(union u_spcl *)tslp->tblock;
697 #endif
698 nextblock++;
699 }
700 trecno = 1;
701
702 /*
703 * Clear the first slaves' response. One hopes that it
704 * worked ok, otherwise the tape is much too short!
705 */
706 if (slp->sent) {
707 if (atomic_read( slp->fd, (char *)&returned, sizeof returned)
708 != sizeof returned) {
709 perror(" DUMP: error reading command pipe in master");
710 dumpabort(0);
711 }
712 got = returned.unclen;
713 bytes_written += returned.clen;
714 if (returned.clen == returned.unclen)
715 uncomprblks++;
716 slp->sent = 0;
717
718 if (got < 0)
719 tperror(-got);
720
721 if (got == 0) {
722 quit("EOT detected at start of the tape!\n");
723 }
724 }
725
726 #ifdef __linux__
727 blks = 0;
728 if (spcl.c_type != TS_END) {
729 for (i = 0; i < spcl.c_count; i++)
730 if (spcl.c_addr[i] != 0)
731 blks++;
732 }
733
734 slp->firstrec = lastfirstrec + ntrec;
735 slp->count = lastspclrec + blks + 1 - spcl.c_tapea;
736 slp->inode = curino;
737 asize += tenths + returned.clen / density;
738 csize += returned.clen;
739 blockswritten += ntrec;
740 blocksthisvol += ntrec;
741 #endif
742 }
743
744 /*
745 * We implement taking and restoring checkpoints on the tape level.
746 * When each tape is opened, a new process is created by forking; this
747 * saves all of the necessary context in the parent. The child
748 * continues the dump; the parent waits around, saving the context.
749 * If the child returns X_REWRITE, then it had problems writing that tape;
750 * this causes the parent to fork again, duplicating the context, and
751 * everything continues as if nothing had happened.
752 */
753 void
754 startnewtape(int top)
755 {
756 int parentpid;
757 int childpid;
758 int status;
759 int waitpid;
760 char *p;
761
762 #ifdef __linux__
763 sigset_t sigs;
764 sigemptyset(&sigs);
765 sigaddset(&sigs, SIGINT);
766 sigprocmask(SIG_BLOCK, &sigs, NULL);
767 #else /* __linux__ */
768 #ifdef sunos
769 void (*interrupt_save)();
770 #else
771 sig_t interrupt_save;
772 #endif
773 interrupt_save = signal(SIGINT, SIG_IGN);
774 #endif /* __linux__ */
775
776 parentpid = getpid();
777 tapea_volume = spcl.c_tapea;
778 tapea_bytes = bytes_written;
779 tstart_volume = time(NULL);
780
781 restore_check_point:
782 #ifdef __linux__
783 sigprocmask(SIG_UNBLOCK, &sigs, NULL);
784 #else
785 (void)signal(SIGINT, interrupt_save);
786 #endif
787 /*
788 * All signals are inherited...
789 */
790 childpid = fork();
791 if (childpid < 0) {
792 msg("Context save fork fails in parent %d\n", parentpid);
793 Exit(X_ABORT);
794 }
795 if (childpid != 0) {
796 /*
797 * PARENT:
798 * save the context by waiting
799 * until the child doing all of the work returns.
800 * don't catch the interrupt
801 */
802 #ifdef __linux__
803 sigprocmask(SIG_BLOCK, &sigs, NULL);
804 #else
805 signal(SIGINT, SIG_IGN);
806 #endif
807 #ifdef TDEBUG
808 msg("Tape: %d; parent process: %d child process %d\n",
809 tapeno+1, parentpid, childpid);
810 #endif /* TDEBUG */
811 while ((waitpid = wait(&status)) != childpid)
812 if (waitpid != rshpid)
813 msg("Parent %d waiting for child %d has another child %d return\n",
814 parentpid, childpid, waitpid);
815 if (status & 0xFF) {
816 msg("Child %d returns LOB status %o\n",
817 childpid, status&0xFF);
818 }
819 status = (status >> 8) & 0xFF;
820 #ifdef TDEBUG
821 switch(status) {
822 case X_FINOK:
823 msg("Child %d finishes X_FINOK\n", childpid);
824 break;
825 case X_ABORT:
826 msg("Child %d finishes X_ABORT\n", childpid);
827 break;
828 case X_REWRITE:
829 msg("Child %d finishes X_REWRITE\n", childpid);
830 break;
831 default:
832 msg("Child %d finishes unknown %d\n",
833 childpid, status);
834 break;
835 }
836 #endif /* TDEBUG */
837 switch(status) {
838 case X_FINOK:
839 Exit(X_FINOK);
840 case X_ABORT:
841 Exit(X_ABORT);
842 case X_REWRITE:
843 goto restore_check_point;
844 default:
845 msg("Bad return code from dump: %d\n", status);
846 Exit(X_ABORT);
847 }
848 /*NOTREACHED*/
849 } else { /* we are the child; just continue */
850 #ifdef TDEBUG
851 sleep(4); /* allow time for parent's message to get out */
852 msg("Child on Tape %d has parent %d, my pid = %d\n",
853 tapeno+1, parentpid, getpid());
854 #endif /* TDEBUG */
855 /*
856 * If we have a name like "/dev/rmt0,/dev/rmt1",
857 * use the name before the comma first, and save
858 * the remaining names for subsequent volumes.
859 */
860 tapeno++; /* current tape sequence */
861 if (Mflag) {
862 snprintf(tape, MAXPATHLEN, "%s%03d", tapeprefix, tapeno);
863 tape[MAXPATHLEN - 1] = '\0';
864 msg("Dumping volume %d on %s\n", tapeno, tape);
865 }
866 else if (nexttape || strchr(tapeprefix, ',')) {
867 if (nexttape && *nexttape)
868 tapeprefix = nexttape;
869 if ((p = strchr(tapeprefix, ',')) != NULL) {
870 *p = '\0';
871 nexttape = p + 1;
872 } else
873 nexttape = NULL;
874 strncpy(tape, tapeprefix, MAXPATHLEN);
875 tape[MAXPATHLEN - 1] = '\0';
876 msg("Dumping volume %d on %s\n", tapeno, tape);
877 }
878 #ifdef RDUMP
879 while ((tapefd = (host ? rmtopen(tape, 2) : pipeout ?
880 fileno(stdout) :
881 OPEN(tape, O_WRONLY|O_CREAT, 0666))) < 0)
882 #else
883 while ((tapefd = (pipeout ? fileno(stdout) :
884 OPEN(tape, O_RDWR|O_CREAT, 0666))) < 0)
885 #endif
886 {
887 msg("Cannot open output \"%s\".\n", tape);
888 if (!query("Do you want to retry the open?"))
889 dumpabort(0);
890 }
891
892 enslave(); /* Share open tape file descriptor with slaves */
893
894 asize = 0;
895 csize = 0;
896 blocksthisvol = 0;
897 if (top)
898 newtape++; /* new tape signal */
899 spcl.c_count = slp->count;
900 /*
901 * measure firstrec in TP_BSIZE units since restore doesn't
902 * know the correct ntrec value...
903 */
904 spcl.c_firstrec = slp->firstrec;
905 spcl.c_volume++;
906 spcl.c_type = TS_TAPE;
907 spcl.c_flags |= DR_NEWHEADER;
908 spcl.c_ntrec = ntrec;
909 if (compressed)
910 spcl.c_flags |= DR_COMPRESSED;
911 writeheader((dump_ino_t)slp->inode);
912 spcl.c_flags &=~ DR_NEWHEADER;
913 msg("Volume %d started with block %ld at: %s", tapeno,
914 spcl.c_tapea, ctime(&tstart_volume));
915 if (tapeno > 1)
916 msg("Volume %d begins with blocks from inode %d\n",
917 tapeno, slp->inode);
918 }
919 }
920
921 void
922 dumpabort(int signo)
923 {
924
925 if (master != 0 && master != getpid())
926 /* Signals master to call dumpabort */
927 (void) kill(master, SIGTERM);
928 else {
929 killall();
930 msg("The ENTIRE dump is aborted.\n");
931 }
932 #ifdef RDUMP
933 rmtclose();
934 #endif
935 Exit(X_ABORT);
936 }
937
938 void
939 Exit(int status)
940 {
941
942 #ifdef TDEBUG
943 msg("pid = %d exits with status %d\n", getpid(), status);
944 #endif /* TDEBUG */
945 exit(status);
946 }
947
948 /*
949 * proceed - handler for SIGUSR2, used to synchronize IO between the slaves.
950 */
951 static void
952 proceed(int signo)
953 {
954 if (ready)
955 siglongjmp(jmpbuf, 1);
956 caught++;
957 }
958
959 void
960 enslave(void)
961 {
962 int cmd[2];
963 #ifdef LINUX_FORK_BUG
964 int i, j;
965 #else
966 register int i, j;
967 #endif
968
969 master = getpid();
970
971 { struct sigaction sa;
972 memset(&sa, 0, sizeof sa);
973 sigemptyset(&sa.sa_mask);
974 sa.sa_handler = dumpabort;
975 sigaction(SIGTERM, &sa, NULL); /* Slave sends SIGTERM on dumpabort() */
976 sa.sa_handler = sigpipe;
977 sigaction(SIGPIPE, &sa, NULL);
978 sa.sa_handler = proceed;
979 sa.sa_flags = SA_RESTART;
980 sigaction(SIGUSR2, &sa, NULL); /* Slave sends SIGUSR2 to next slave */
981 }
982
983 for (i = 0; i < SLAVES; i++) {
984 if (i == slp - &slaves[0]) {
985 caught = 1;
986 } else {
987 caught = 0;
988 }
989
990 if (socketpair(AF_UNIX, SOCK_STREAM, 0, cmd) < 0 ||
991 (slaves[i].pid = fork()) < 0)
992 quit("too many slaves, %d (recompile smaller): %s\n",
993 i, strerror(errno));
994
995 slaves[i].fd = cmd[1];
996 slaves[i].sent = 0;
997 if (slaves[i].pid == 0) { /* Slave starts up here */
998 sigset_t sigs;
999 for (j = 0; j <= i; j++)
1000 (void) close(slaves[j].fd);
1001 sigemptyset(&sigs);
1002 sigaddset(&sigs, SIGINT); /* Master handles this */
1003 #if defined(SIGINFO)
1004 sigaddset(&sigs, SIGINFO);
1005 #endif
1006 sigprocmask(SIG_BLOCK, &sigs, NULL);
1007
1008 #ifdef LINUX_FORK_BUG
1009 if (atomic_write( cmd[0], (char *) &i, sizeof i)
1010 != sizeof i)
1011 quit("master/slave protocol botched 3\n");
1012 #endif
1013 doslave(cmd[0], i);
1014 Exit(X_FINOK);
1015 }
1016 else
1017 close(cmd[0]);
1018 }
1019
1020 #ifdef LINUX_FORK_BUG
1021 /*
1022 * Wait for all slaves to _actually_ start to circumvent a bug in
1023 * Linux kernels >= 2.1.3 where a signal sent to a child that hasn't
1024 * returned from fork() causes a SEGV in the child process
1025 */
1026 for (i = 0; i < SLAVES; i++)
1027 if (atomic_read( slaves[i].fd, (char *) &j, sizeof j) != sizeof j)
1028 quit("master/slave protocol botched 4\n");
1029 #endif
1030
1031 for (i = 0; i < SLAVES; i++)
1032 (void) atomic_write( slaves[i].fd,
1033 (char *) &slaves[(i + 1) % SLAVES].pid,
1034 sizeof slaves[0].pid);
1035
1036 master = 0;
1037 }
1038
1039 void
1040 killall(void)
1041 {
1042 register int i;
1043
1044 for (i = 0; i < SLAVES; i++)
1045 if (slaves[i].pid > 0) {
1046 (void) kill(slaves[i].pid, SIGKILL);
1047 slaves[i].sent = 0;
1048 }
1049 }
1050
1051 /*
1052 * Synchronization - each process waits for a SIGUSR2 from the
1053 * previous process before writing to the tape, and sends SIGUSR2
1054 * to the next process when the tape write completes. On tape errors
1055 * a SIGUSR1 is sent to the master which then terminates all of the
1056 * slaves.
1057 */
1058 static void
1059 doslave(int cmd, int slave_number)
1060 {
1061 register int nread;
1062 int nextslave, size, eot_count, bufsize;
1063 volatile int wrote = 0;
1064 char *buffer;
1065 #ifdef HAVE_ZLIB
1066 struct tapebuf *comp_buf = NULL;
1067 int compresult, do_compress = slave_number > 0;
1068 unsigned long worklen;
1069 #endif /* HAVE_ZLIB */
1070 struct slave_results returns;
1071 #ifdef __linux__
1072 errcode_t retval;
1073 #endif
1074 #ifdef USE_QFA
1075 long curtapepos;
1076 union u_spcl *uspclptr;
1077 struct s_spcl *spclptr;
1078 #endif /* USE_QFA */
1079
1080 /*
1081 * Need our own seek pointer.
1082 */
1083 (void) close(diskfd);
1084 if ((diskfd = OPEN(disk, O_RDONLY)) < 0)
1085 quit("slave couldn't reopen disk: %s\n", strerror(errno));
1086 #ifdef __linux__
1087 ext2fs_close(fs);
1088 retval = dump_fs_open(disk, &fs);
1089 if (retval)
1090 quit("slave couldn't reopen disk: %s\n", error_message(retval));
1091 #endif /* __linux__ */
1092
1093 /*
1094 * Need the pid of the next slave in the loop...
1095 */
1096 if ((nread = atomic_read( cmd, (char *)&nextslave, sizeof nextslave))
1097 != sizeof nextslave) {
1098 quit("master/slave protocol botched - didn't get pid of next slave.\n");
1099 }
1100
1101 #ifdef HAVE_ZLIB
1102 /* if we're doing a compressed dump, allocate the compress buffer */
1103 if (compressed) {
1104 comp_buf = malloc(sizeof(struct tapebuf) + TP_BSIZE + writesize);
1105 if (comp_buf == NULL)
1106 quit("couldn't allocate a compress buffer.\n");
1107 comp_buf->flags = 0;
1108 }
1109 #endif /* HAVE_ZLIB */
1110
1111 /*
1112 * Get list of blocks to dump, read the blocks into tape buffer
1113 */
1114 while ((nread = atomic_read( cmd, (char *)slp->req, reqsiz)) == reqsiz) {
1115 register struct req *p = slp->req;
1116
1117 for (trecno = 0; trecno < ntrec;
1118 trecno += p->count, p += p->count) {
1119 if (p->dblk) { /* read a disk block */
1120 bread(p->dblk, slp->tblock[trecno],
1121 p->count * TP_BSIZE);
1122 } else { /* read record from pipe */
1123 if (p->count != 1 || atomic_read( cmd,
1124 (char *)slp->tblock[trecno],
1125 TP_BSIZE) != TP_BSIZE)
1126 quit("master/slave protocol botched.\n");
1127 }
1128 }
1129
1130 /* Try to write the data... */
1131 wrote = 0;
1132 eot_count = 0;
1133 size = 0;
1134 buffer = (char *) slp->tblock[0]; /* set write pointer */
1135 bufsize = writesize; /* length to write */
1136 returns.clen = returns.unclen = bufsize;
1137
1138 #ifdef HAVE_ZLIB
1139 /*
1140 * When writing a compressed dump, each block except
1141 * the first one on each tape is written
1142 * from struct tapebuf with an 4 byte prefix
1143 * followed by the data. This can be less than
1144 * writesize. Restore, on a short read, can compare the
1145 * length read to the compressed length in the header
1146 * to verify that the read was good. Blocks which don't
1147 * compress well are written uncompressed.
1148 * The first block written by each slave is not compressed
1149 * and does not have a prefix.
1150 */
1151
1152 if (compressed && do_compress) {
1153 comp_buf->length = bufsize;
1154 worklen = TP_BSIZE + writesize;
1155 compresult = compress2(comp_buf->buf, &worklen,
1156 (char *)slp->tblock[0], writesize, compressed);
1157 if (compresult == Z_OK && worklen <= (writesize - 16)) {
1158 /* write the compressed buffer */
1159 comp_buf->length = worklen;
1160 comp_buf->compressed = 1;
1161 buffer = (char *) comp_buf;
1162 returns.clen = bufsize = worklen + sizeof(struct tapebuf);
1163 }
1164 else {
1165 /* write the data uncompressed */
1166 comp_buf->length = writesize;
1167 comp_buf->compressed = 0;
1168 buffer = (char *) comp_buf;
1169 returns.clen = bufsize = writesize + sizeof(struct tapebuf);
1170 returns.unclen = returns.clen;
1171 memcpy(comp_buf->buf, (char *)slp->tblock[0], writesize);
1172 }
1173 }
1174 /* compress the remaining blocks if we're compressing */
1175 do_compress = compressed;
1176 #endif /* HAVE_ZLIB */
1177
1178 if (sigsetjmp(jmpbuf, 1) == 0) {
1179 ready = 1;
1180 if (!caught)
1181 (void) pause();
1182 }
1183 ready = 0;
1184 caught = 0;
1185
1186 #ifdef USE_QFA
1187 if (gTapeposfd >= 0) {
1188 uspclptr = (union u_spcl *)&slp->tblock[0];
1189 spclptr = &uspclptr->s_spcl;
1190 if ((spclptr->c_magic == NFS_MAGIC) &&
1191 (spclptr->c_type == TS_INODE)) {
1192 /* if an error occured previously don't
1193 * try again */
1194 if (gtperr == 0) {
1195 if ((gtperr = GetTapePos(&curtapepos)) == 0) {
1196 #ifdef DEBUG_QFA
1197 msg("inode %ld at tapepos %ld\n", spclptr->c_inumber, curtapepos);
1198 #endif
1199 sprintf(gTps, "%ld\t%d\t%ld\n", (unsigned long)spclptr->c_inumber, tapeno, curtapepos);
1200 if (write(gTapeposfd, gTps, strlen(gTps)) != strlen(gTps)) {
1201 warn("error writing tapepos file.\n");
1202 }
1203 }
1204 }
1205 }
1206 }
1207 #endif /* USE_QFA */
1208
1209 while (eot_count < 10 && size < bufsize) {
1210 #ifdef RDUMP
1211 if (host)
1212 wrote = rmtwrite(buffer + size, bufsize - size);
1213 else
1214 #endif
1215 wrote = write(tapefd, buffer + size, bufsize - size);
1216 #ifdef WRITEDEBUG
1217 printf("slave %d wrote %d\n", slave_number, wrote);
1218 #endif
1219 if (wrote < 0)
1220 break;
1221 if (wrote == 0)
1222 eot_count++;
1223 size += wrote;
1224 }
1225
1226 #ifdef WRITEDEBUG
1227 if (size != bufsize)
1228 printf("slave %d only wrote %d out of %d bytes and gave up.\n",
1229 slave_number, size, bufsize);
1230 #endif
1231
1232 /*
1233 * Handle ENOSPC as an EOT condition.
1234 */
1235 if (wrote < 0 && errno == ENOSPC) {
1236 wrote = 0;
1237 eot_count++;
1238 }
1239
1240 if (eot_count > 0)
1241 returns.clen = returns.unclen = 0;
1242
1243 /*
1244 * pass errno back to master for special handling
1245 */
1246 if (wrote < 0)
1247 returns.unclen = -errno;
1248
1249 /*
1250 * pass size of data and size of write back to master
1251 * (for EOT handling)
1252 */
1253 (void) atomic_write( cmd, (char *)&returns, sizeof returns);
1254
1255 /*
1256 * Signal the next slave to go.
1257 */
1258 (void) kill(nextslave, SIGUSR2);
1259 }
1260 if (nread != 0)
1261 quit("error reading command pipe: %s\n", strerror(errno));
1262 }
1263
1264 /*
1265 * Since a read from a pipe may not return all we asked for,
1266 * or a write may not write all we ask if we get a signal,
1267 * loop until the count is satisfied (or error).
1268 */
1269 static ssize_t
1270 atomic_read(int fd, void *buf, size_t count)
1271 {
1272 int got, need = count;
1273
1274 do {
1275 while ((got = read(fd, buf, need)) > 0 && (need -= got) > 0)
1276 (char *)buf += got;
1277 } while (got == -1 && errno == EINTR);
1278 return (got < 0 ? got : count - need);
1279 }
1280
1281 /*
1282 * Since a read from a pipe may not return all we asked for,
1283 * or a write may not write all we ask if we get a signal,
1284 * loop until the count is satisfied (or error).
1285 */
1286 static ssize_t
1287 atomic_write(int fd, const void *buf, size_t count)
1288 {
1289 int got, need = count;
1290
1291 do {
1292 while ((got = write(fd, buf, need)) > 0 && (need -= got) > 0)
1293 (char *)buf += got;
1294 } while (got == -1 && errno == EINTR);
1295 return (got < 0 ? got : count - need);
1296 }
1297
1298
1299 #ifdef USE_QFA
1300 /*
1301 * read the current tape position
1302 */
1303 int
1304 GetTapePos(long *pos)
1305 {
1306 int err = 0;
1307
1308 *pos = 0;
1309 if (ioctl(tapefd, MTIOCPOS, pos) == -1) {
1310 err = errno;
1311 msg("[%ld] error: %d (getting tapepos: %ld)\n", getpid(),
1312 err, *pos);
1313 return err;
1314 }
1315 return err;
1316 }
1317 #endif /* USE_QFA */