]> git.wh0rd.org - dump.git/blob - dump/tape.c
Made -Q option work on regular files.
[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 <stelian@popies.net>, 1999-2000
6 * Stelian Pop <stelian@popies.net> - AlcĂ´ve <www.alcove.com>, 2000-2002
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 1.60 2002/01/22 11:12:28 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 #include <system.h>
56 #ifdef __STDC__
57 #include <stdlib.h>
58 #include <string.h>
59 #include <unistd.h>
60 #else
61 int write(), read();
62 #endif
63
64 #ifdef __linux__
65 #include <sys/types.h>
66 #include <sys/time.h>
67 #include <time.h>
68 #endif
69 #include <sys/param.h>
70 #include <sys/socket.h>
71 #include <sys/wait.h>
72 #include <sys/mtio.h>
73 #ifdef __linux__
74 #include <linux/fs.h>
75 #undef atomic_read /* this get wrongly defined in kernel */
76 /* headers and we don't want it */
77 #ifdef HAVE_EXT2FS_EXT2_FS_H
78 #include <ext2fs/ext2_fs.h>
79 #else
80 #include <linux/ext2_fs.h>
81 #endif
82 #include <ext2fs/ext2fs.h>
83 #include <bsdcompat.h>
84 #elif defined sunos
85 #include <sys/vnode.h>
86
87 #include <ufs/fs.h>
88 #include <ufs/inode.h>
89 #else
90 #include <ufs/ufs/dinode.h>
91 #include <ufs/ffs/fs.h>
92 #endif /* __linux__ */
93
94 #include <protocols/dumprestore.h>
95
96 #ifdef HAVE_ZLIB
97 #include <zlib.h>
98 #endif /* HAVE_ZLIB */
99
100 #ifdef HAVE_BZLIB
101 #include <bzlib.h>
102 #endif /* HAVE_BZLIB */
103
104 #include "dump.h"
105
106 int writesize; /* size of malloc()ed buffer for tape */
107 long lastspclrec = -1; /* tape block number of last written header */
108 int trecno = 0; /* next record to write in current block */
109 extern long blocksperfile; /* number of blocks per output file */
110 long blocksthisvol; /* number of blocks on current output file */
111 extern int ntrec; /* blocking factor on tape */
112 extern int cartridge;
113 extern char *host;
114 char *nexttape;
115 extern pid_t rshpid;
116 int eot_code = 1;
117 long long tapea_bytes = 0; /* bytes_written at start of current volume */
118 static int magtapeout; /* output is really a tape */
119
120 static ssize_t atomic_read __P((int, void *, size_t));
121 static ssize_t atomic_write __P((int, const void *, size_t));
122 static void doslave __P((int, int, int));
123 static void enslave __P((void));
124 static void flushtape __P((void));
125 static void killall __P((void));
126 static void rollforward __P((void));
127 #ifdef USE_QFA
128 static int GetTapePos __P((long *));
129 static void MkTapeString __P((struct s_spcl *, long));
130 #endif
131
132 /*
133 * Concurrent dump mods (Caltech) - disk block reading and tape writing
134 * are exported to several slave processes. While one slave writes the
135 * tape, the others read disk blocks; they pass control of the tape in
136 * a ring via signals. The parent process traverses the filesystem and
137 * sends writeheader()'s and lists of daddr's to the slaves via pipes.
138 * The following structure defines the instruction packets sent to slaves.
139 */
140 struct req {
141 daddr_t dblk;
142 int count;
143 };
144 int reqsiz;
145
146 struct slave_results {
147 ssize_t unclen; /* uncompressed length */
148 ssize_t clen; /* compressed length */
149 };
150
151 #define SLAVES 3 /* 1 slave writing, 1 reading, 1 for slack */
152 struct slave {
153 int tapea; /* header number at start of this chunk */
154 int count; /* count to next header (used for TS_TAPE */
155 /* after EOT) */
156 int inode; /* inode that we are currently dealing with */
157 int fd; /* FD for this slave */
158 int pid; /* PID for this slave */
159 int sent; /* 1 == we've sent this slave requests */
160 int firstrec; /* record number of this block */
161 char (*tblock)[TP_BSIZE]; /* buffer for data blocks */
162 struct req *req; /* buffer for requests */
163 } slaves[SLAVES+1];
164 struct slave *slp;
165
166 char (*nextblock)[TP_BSIZE];
167
168 static time_t tstart_volume; /* time of volume start */
169 static int tapea_volume; /* value of spcl.c_tapea at volume start */
170
171 int master; /* pid of master, for sending error signals */
172 int tenths; /* length of tape overhead per block written */
173 static int caught; /* have we caught the signal to proceed? */
174 static int ready; /* have we reached the lock point without having */
175 /* received the SIGUSR2 signal from the prev slave? */
176 static sigjmp_buf jmpbuf; /* where to jump to if we are ready when the */
177 /* SIGUSR2 arrives from the previous slave */
178 #ifdef USE_QFA
179 static int gtperr = 0;
180 #endif
181
182 int
183 alloctape(void)
184 {
185 int pgoff = getpagesize() - 1;
186 char *buf;
187 int i;
188
189 writesize = ntrec * TP_BSIZE;
190 reqsiz = (ntrec + 1) * sizeof(struct req);
191 /*
192 * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode
193 * (see DEC TU80 User's Guide). The shorter gaps of 6250-bpi require
194 * repositioning after stopping, i.e, streaming mode, where the gap is
195 * variable, 0.30" to 0.45". The gap is maximal when the tape stops.
196 */
197 if (blocksperfile == 0 && !unlimited)
198 tenths = (cartridge ? 16 : density == 625 ? 5 : 8);
199 else {
200 tenths = 0;
201 density = 1;
202 }
203 /*
204 * Allocate tape buffer contiguous with the array of instruction
205 * packets, so flushtape() can write them together with one write().
206 * Align tape buffer on page boundary to speed up tape write().
207 */
208 for (i = 0; i <= SLAVES; i++) {
209 buf = (char *)
210 malloc((unsigned)(reqsiz + writesize + pgoff + TP_BSIZE));
211 if (buf == NULL)
212 return(0);
213 slaves[i].tblock = (char (*)[TP_BSIZE])
214 #ifdef __linux__
215 (((long)&buf[reqsiz] + pgoff) &~ pgoff);
216 #else
217 (((long)&buf[ntrec + 1] + pgoff) &~ pgoff);
218 #endif
219 slaves[i].req = (struct req *)slaves[i].tblock - ntrec - 1;
220 }
221 slp = &slaves[0];
222 slp->count = 1;
223 slp->tapea = 0;
224 slp->firstrec = 0;
225 nextblock = slp->tblock;
226 return(1);
227 }
228
229 void
230 writerec(const void *dp, int isspcl)
231 {
232
233 slp->req[trecno].dblk = (daddr_t)0;
234 slp->req[trecno].count = 1;
235 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
236 *(union u_spcl *)(*(nextblock)) = *(union u_spcl *)dp;
237 nextblock++;
238 if (isspcl)
239 lastspclrec = spcl.c_tapea;
240 trecno++;
241 spcl.c_tapea++;
242 if (trecno >= ntrec)
243 flushtape();
244 }
245
246 void
247 dumpblock(daddr_t blkno, int size)
248 {
249 int avail, tpblks, dblkno;
250
251 dblkno = fsbtodb(sblock, blkno);
252 tpblks = size >> tp_bshift;
253 while ((avail = MIN(tpblks, ntrec - trecno)) > 0) {
254 slp->req[trecno].dblk = dblkno;
255 slp->req[trecno].count = avail;
256 trecno += avail;
257 spcl.c_tapea += avail;
258 if (trecno >= ntrec)
259 flushtape();
260 dblkno += avail << (tp_bshift - dev_bshift);
261 tpblks -= avail;
262 }
263 }
264
265 int nogripe = 0;
266
267 static void
268 tperror(int errnum)
269 {
270
271 if (pipeout) {
272 msg("write error on %s: %s\n", tape, strerror(errnum));
273 quit("Cannot recover\n");
274 /* NOTREACHED */
275 }
276 msg("write error %d blocks into volume %d: %s\n",
277 blocksthisvol, tapeno, strerror(errnum));
278 broadcast("DUMP WRITE ERROR!\n");
279 if (query("Do you want to rewrite this volume?")) {
280 msg("Closing this volume. Prepare to restart with new media;\n");
281 msg("this dump volume will be rewritten.\n");
282 killall();
283 nogripe = 1;
284 close_rewind();
285 Exit(X_REWRITE);
286 }
287 if (query("Do you want to start the next tape?"))
288 return;
289 dumpabort(0);
290 }
291
292 static void
293 sigpipe(int signo)
294 {
295
296 quit("Broken pipe\n");
297 }
298
299 /*
300 * do_stats --
301 * Update xferrate stats
302 */
303 time_t
304 do_stats(void)
305 {
306 time_t tnow, ttaken;
307 int blocks;
308
309 tnow = time(NULL);
310 ttaken = tnow - tstart_volume;
311 blocks = spcl.c_tapea - tapea_volume;
312 msg("Volume %d completed at: %s", tapeno, ctime(&tnow));
313 if (! compressed)
314 msg("Volume %d %ld tape blocks (%.2fMB)\n", tapeno,
315 blocks, ((double)blocks * TP_BSIZE / 1048576));
316 if (ttaken > 0) {
317 long volkb = (bytes_written - tapea_bytes) / 1024;
318 long txfrate = volkb / ttaken;
319 msg("Volume %d took %d:%02d:%02d\n", tapeno,
320 ttaken / 3600, (ttaken % 3600) / 60, ttaken % 60);
321 msg("Volume %d transfer rate: %ld kB/s\n", tapeno,
322 txfrate);
323 xferrate += txfrate;
324 if (compressed) {
325 double rate = .0005 + (double) blocks / (double) volkb;
326 msg("Volume %d %ldkB uncompressed, %ldkB compressed,"
327 " %1.3f:1\n",
328 tapeno, blocks, volkb, rate);
329 }
330 }
331 return(tnow);
332 }
333
334 char *
335 mktimeest(time_t tnow)
336 {
337 static char msgbuf[128];
338 time_t deltat;
339
340 msgbuf[0] = '\0';
341
342 if (blockswritten < 500)
343 return NULL;
344 if (blockswritten > tapesize)
345 tapesize = blockswritten;
346 deltat = tstart_writing - tnow + (1.0 * (tnow - tstart_writing))
347 / blockswritten * tapesize;
348 if (tnow > tstart_volume)
349 (void)snprintf(msgbuf, sizeof(msgbuf),
350 "%3.2f%% done at %ld kB/s, finished in %d:%02d\n",
351 (blockswritten * 100.0) / tapesize,
352 (spcl.c_tapea - tapea_volume) / (tnow - tstart_volume),
353 (int)(deltat / 3600), (int)((deltat % 3600) / 60));
354 else
355 (void)snprintf(msgbuf, sizeof(msgbuf),
356 "%3.2f%% done, finished in %d:%02d\n",
357 (blockswritten * 100.0) / tapesize,
358 (int)(deltat / 3600), (int)((deltat % 3600) / 60));
359
360 return msgbuf;
361 }
362
363 #if defined(SIGINFO)
364 /*
365 * statussig --
366 * information message upon receipt of SIGINFO
367 */
368 void
369 statussig(int notused)
370 {
371 int save_errno = errno;
372 char *buf;
373
374 buf = mktimeest(time(NULL));
375 if (buf)
376 write(STDERR_FILENO, buf, strlen(buf));
377 errno = save_errno;
378 }
379 #endif
380
381 static void
382 flushtape(void)
383 {
384 int i, blks, got;
385 long lastfirstrec;
386 struct slave_results returned;
387
388 int siz = (char *)nextblock - (char *)slp->req;
389
390 slp->req[trecno].count = 0; /* Sentinel */
391
392 if (atomic_write( slp->fd, (char *)slp->req, siz) != siz)
393 quit("error writing command pipe: %s\n", strerror(errno));
394 slp->sent = 1; /* we sent a request, read the response later */
395
396 lastfirstrec = slp->firstrec;
397
398 if (++slp >= &slaves[SLAVES])
399 slp = &slaves[0];
400
401 /* Read results back from next slave */
402 if (slp->sent) {
403 if (atomic_read( slp->fd, (char *)&returned, sizeof returned)
404 != sizeof returned) {
405 perror(" DUMP: error reading command pipe in master");
406 dumpabort(0);
407 }
408 got = returned.unclen;
409 bytes_written += returned.clen;
410 if (returned.unclen == returned.clen)
411 uncomprblks++;
412 slp->sent = 0;
413
414 /* Check for errors or end of tape */
415 if (got <= 0) {
416 /* Check for errors */
417 if (got < 0)
418 tperror(-got);
419 else
420 msg("End of tape detected\n");
421
422 /*
423 * Drain the results, don't care what the values were.
424 * If we read them here then trewind won't...
425 */
426 for (i = 0; i < SLAVES; i++) {
427 if (slaves[i].sent) {
428 if (atomic_read( slaves[i].fd,
429 (char *)&returned, sizeof returned)
430 != sizeof returned) {
431 perror(" DUMP: error reading command pipe in master");
432 dumpabort(0);
433 }
434 slaves[i].sent = 0;
435 }
436 }
437
438 close_rewind();
439 rollforward();
440 return;
441 }
442 }
443
444 blks = 0;
445 if (spcl.c_type != TS_END) {
446 for (i = 0; i < spcl.c_count; i++)
447 if (spcl.c_addr[i] != 0)
448 blks++;
449 }
450 slp->count = lastspclrec + blks + 1 - spcl.c_tapea;
451 slp->tapea = spcl.c_tapea;
452 slp->firstrec = lastfirstrec + ntrec;
453 slp->inode = curino;
454 nextblock = slp->tblock;
455 trecno = 0;
456 asize += tenths + returned.clen / density;
457 blockswritten += ntrec;
458 blocksthisvol += ntrec;
459 if (!pipeout && !unlimited) {
460 if (blocksperfile) {
461 if ( compressed ? (bytes_written - tapea_bytes + SLAVES * (writesize + sizeof(struct tapebuf))) >= blocksperfile * 1024
462 : blocksthisvol >= blocksperfile ) {
463 close_rewind();
464 startnewtape(0);
465 }
466 }
467 else if (asize > tsize) {
468 close_rewind();
469 startnewtape(0);
470 }
471 }
472 timeest();
473 }
474
475 time_t
476 trewind(void)
477 {
478 int f;
479 int got;
480 struct slave_results returned;
481
482 for (f = 0; f < SLAVES; f++) {
483 /*
484 * Drain the results, but unlike EOT we DO (or should) care
485 * what the return values were, since if we detect EOT after
486 * we think we've written the last blocks to the tape anyway,
487 * we have to replay those blocks with rollforward.
488 *
489 * fixme: punt for now.
490 */
491 if (slaves[f].sent) {
492 if (atomic_read( slaves[f].fd, (char *)&returned, sizeof returned)
493 != sizeof returned) {
494 perror(" DUMP: error reading command pipe in master");
495 dumpabort(0);
496 }
497 got = returned.unclen;
498 bytes_written += returned.clen;
499 if (returned.unclen == returned.clen)
500 uncomprblks++;
501 slaves[f].sent = 0;
502
503 if (got < 0)
504 tperror(-got);
505
506 if (got == 0) {
507 msg("EOT detected in last 2 tape records!\n");
508 msg("Use a longer tape, decrease the size estimate\n");
509 quit("or use no size estimate at all.\n");
510 }
511 }
512 (void) close(slaves[f].fd);
513 }
514 while (wait((int *)NULL) >= 0) /* wait for any signals from slaves */
515 /* void */;
516
517 if (!pipeout) {
518
519 msg("Closing %s\n", tape);
520
521 #ifdef RDUMP
522 if (host) {
523 rmtclose();
524 while (rmtopen(tape, 0) < 0)
525 sleep(10);
526 rmtclose();
527 }
528 else
529 #endif
530 {
531 (void) close(tapefd);
532 if (!fifoout) {
533 while ((f = OPEN(tape, 0)) < 0)
534 sleep (10);
535 (void) close(f);
536 }
537 }
538 eot_code = 1;
539 if (eot_script && spcl.c_type != TS_END) {
540 msg("Launching %s\n", eot_script);
541 eot_code = system_command(eot_script, tape, tapeno);
542 }
543 if (eot_code != 0 && eot_code != 1) {
544 msg("Dump aborted by the end of tape script\n");
545 dumpabort(0);
546 }
547 }
548 return do_stats();
549 }
550
551
552 void
553 close_rewind(void)
554 {
555 (void)trewind();
556 if (nexttape || Mflag || (eot_code == 0) )
557 return;
558 if (!nogripe) {
559 msg("Change Volumes: Mount volume #%d\n", tapeno+1);
560 broadcast("CHANGE DUMP VOLUMES!\7\7\n");
561 }
562 while (!query("Is the new volume mounted and ready to go?"))
563 if (query("Do you want to abort?")) {
564 dumpabort(0);
565 /*NOTREACHED*/
566 }
567 }
568
569 void
570 rollforward(void)
571 {
572 register struct req *p, *q = NULL, *prev;
573 register struct slave *tslp;
574 int i, size, savedtapea, got;
575 union u_spcl *ntb, *otb;
576 struct slave_results returned;
577 #ifdef __linux__
578 int blks;
579 long lastfirstrec;
580 #endif
581 tslp = &slaves[SLAVES];
582 ntb = (union u_spcl *)tslp->tblock[1];
583
584 /*
585 * Each of the N slaves should have requests that need to
586 * be replayed on the next tape. Use the extra slave buffers
587 * (slaves[SLAVES]) to construct request lists to be sent to
588 * each slave in turn.
589 */
590 for (i = 0; i < SLAVES; i++) {
591 q = &tslp->req[1];
592 otb = (union u_spcl *)slp->tblock;
593
594 /*
595 * For each request in the current slave, copy it to tslp.
596 */
597
598 prev = NULL;
599 for (p = slp->req; p->count > 0; p += p->count) {
600 *q = *p;
601 if (p->dblk == 0)
602 *ntb++ = *otb++; /* copy the datablock also */
603 prev = q;
604 q += q->count;
605 }
606 if (prev == NULL)
607 quit("rollforward: protocol botch");
608 if (prev->dblk != 0)
609 prev->count -= 1;
610 else
611 ntb--;
612 q -= 1;
613 q->count = 0;
614 q = &tslp->req[0];
615 if (i == 0) {
616 q->dblk = 0;
617 q->count = 1;
618 trecno = 0;
619 nextblock = tslp->tblock;
620 savedtapea = spcl.c_tapea;
621 spcl.c_tapea = slp->tapea;
622 startnewtape(0);
623 spcl.c_tapea = savedtapea;
624 lastspclrec = savedtapea - 1;
625 }
626 size = (char *)ntb - (char *)q;
627 if (atomic_write( slp->fd, (char *)q, size) != size) {
628 perror(" DUMP: error writing command pipe");
629 dumpabort(0);
630 }
631 slp->sent = 1;
632 #ifdef __linux__
633 lastfirstrec = slp->firstrec;
634 #endif
635 if (++slp >= &slaves[SLAVES])
636 slp = &slaves[0];
637
638 q->count = 1;
639
640 if (prev->dblk != 0) {
641 /*
642 * If the last one was a disk block, make the
643 * first of this one be the last bit of that disk
644 * block...
645 */
646 q->dblk = prev->dblk +
647 prev->count * (TP_BSIZE / DEV_BSIZE);
648 ntb = (union u_spcl *)tslp->tblock;
649 } else {
650 /*
651 * It wasn't a disk block. Copy the data to its
652 * new location in the buffer.
653 */
654 q->dblk = 0;
655 *((union u_spcl *)tslp->tblock) = *ntb;
656 ntb = (union u_spcl *)tslp->tblock[1];
657 }
658 }
659 slp->req[0] = *q;
660 nextblock = slp->tblock;
661 if (q->dblk == 0) {
662 #ifdef __linux__
663 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
664 *(union u_spcl *)(*nextblock) = *(union u_spcl *)tslp->tblock;
665 #endif
666 nextblock++;
667 }
668 trecno = 1;
669
670 /*
671 * Clear the first slaves' response. One hopes that it
672 * worked ok, otherwise the tape is much too short!
673 */
674 if (slp->sent) {
675 if (atomic_read( slp->fd, (char *)&returned, sizeof returned)
676 != sizeof returned) {
677 perror(" DUMP: error reading command pipe in master");
678 dumpabort(0);
679 }
680 got = returned.unclen;
681 bytes_written += returned.clen;
682 if (returned.clen == returned.unclen)
683 uncomprblks++;
684 slp->sent = 0;
685
686 if (got < 0)
687 tperror(-got);
688
689 if (got == 0) {
690 quit("EOT detected at start of the tape!\n");
691 }
692 }
693
694 #ifdef __linux__
695 blks = 0;
696 if (spcl.c_type != TS_END) {
697 for (i = 0; i < spcl.c_count; i++)
698 if (spcl.c_addr[i] != 0)
699 blks++;
700 }
701
702 slp->firstrec = lastfirstrec + ntrec;
703 slp->count = lastspclrec + blks + 1 - spcl.c_tapea;
704 slp->inode = curino;
705 asize += tenths + returned.clen / density;
706 blockswritten += ntrec;
707 blocksthisvol += ntrec;
708 #endif
709 }
710
711 /*
712 * We implement taking and restoring checkpoints on the tape level.
713 * When each tape is opened, a new process is created by forking; this
714 * saves all of the necessary context in the parent. The child
715 * continues the dump; the parent waits around, saving the context.
716 * If the child returns X_REWRITE, then it had problems writing that tape;
717 * this causes the parent to fork again, duplicating the context, and
718 * everything continues as if nothing had happened.
719 */
720 void
721 startnewtape(int top)
722 {
723 int parentpid;
724 int childpid;
725 int status;
726 int waitpid;
727 char *p;
728
729 #ifdef __linux__
730 sigset_t sigs;
731 sigemptyset(&sigs);
732 sigaddset(&sigs, SIGINT);
733 sigprocmask(SIG_BLOCK, &sigs, NULL);
734 #else /* __linux__ */
735 #ifdef sunos
736 void (*interrupt_save)();
737 #else
738 sig_t interrupt_save;
739 #endif
740 interrupt_save = signal(SIGINT, SIG_IGN);
741 #endif /* __linux__ */
742
743 parentpid = getpid();
744 tapea_volume = spcl.c_tapea;
745 tapea_bytes = bytes_written;
746 tstart_volume = time(NULL);
747
748 restore_check_point:
749 #ifdef __linux__
750 sigprocmask(SIG_UNBLOCK, &sigs, NULL);
751 #else
752 (void)signal(SIGINT, interrupt_save);
753 #endif
754 /*
755 * All signals are inherited...
756 */
757 childpid = fork();
758 if (childpid < 0) {
759 msg("Context save fork fails in parent %d\n", parentpid);
760 Exit(X_ABORT);
761 }
762 if (childpid != 0) {
763 /*
764 * PARENT:
765 * save the context by waiting
766 * until the child doing all of the work returns.
767 * don't catch the interrupt
768 */
769 #ifdef __linux__
770 sigprocmask(SIG_BLOCK, &sigs, NULL);
771 #else
772 signal(SIGINT, SIG_IGN);
773 #endif
774 #ifdef TDEBUG
775 msg("Tape: %d; parent process: %d child process %d\n",
776 tapeno+1, parentpid, childpid);
777 #endif /* TDEBUG */
778 while ((waitpid = wait(&status)) != childpid)
779 if (waitpid != rshpid)
780 msg("Parent %d waiting for child %d has another child %d return\n",
781 parentpid, childpid, waitpid);
782 if (status & 0xFF) {
783 msg("Child %d returns LOB status %o\n",
784 childpid, status&0xFF);
785 }
786 status = (status >> 8) & 0xFF;
787 #ifdef TDEBUG
788 switch(status) {
789 case X_FINOK:
790 msg("Child %d finishes X_FINOK\n", childpid);
791 break;
792 case X_ABORT:
793 msg("Child %d finishes X_ABORT\n", childpid);
794 break;
795 case X_REWRITE:
796 msg("Child %d finishes X_REWRITE\n", childpid);
797 break;
798 default:
799 msg("Child %d finishes unknown %d\n",
800 childpid, status);
801 break;
802 }
803 #endif /* TDEBUG */
804 switch(status) {
805 case X_FINOK:
806 Exit(X_FINOK);
807 case X_ABORT:
808 Exit(X_ABORT);
809 case X_REWRITE:
810 goto restore_check_point;
811 default:
812 msg("Bad return code from dump: %d\n", status);
813 Exit(X_ABORT);
814 }
815 /*NOTREACHED*/
816 } else { /* we are the child; just continue */
817 #ifdef TDEBUG
818 sleep(4); /* allow time for parent's message to get out */
819 msg("Child on Tape %d has parent %d, my pid = %d\n",
820 tapeno+1, parentpid, getpid());
821 #endif /* TDEBUG */
822 /*
823 * If we have a name like "/dev/rmt0,/dev/rmt1",
824 * use the name before the comma first, and save
825 * the remaining names for subsequent volumes.
826 */
827 tapeno++; /* current tape sequence */
828 if (Mflag) {
829 snprintf(tape, MAXPATHLEN, "%s%03d", tapeprefix, tapeno);
830 tape[MAXPATHLEN - 1] = '\0';
831 msg("Dumping volume %d on %s\n", tapeno, tape);
832 }
833 else if (nexttape || strchr(tapeprefix, ',')) {
834 if (nexttape && *nexttape)
835 tapeprefix = nexttape;
836 if ((p = strchr(tapeprefix, ',')) != NULL) {
837 *p = '\0';
838 nexttape = p + 1;
839 } else
840 nexttape = NULL;
841 strncpy(tape, tapeprefix, MAXPATHLEN);
842 tape[MAXPATHLEN - 1] = '\0';
843 msg("Dumping volume %d on %s\n", tapeno, tape);
844 }
845 #ifdef RDUMP
846 while ((tapefd = (host ? rmtopen(tape, 2) : pipeout ?
847 fileno(stdout) :
848 OPEN(tape, O_WRONLY|O_CREAT, 0666))) < 0)
849 #else
850 while ((tapefd = (pipeout ? fileno(stdout) :
851 OPEN(tape, O_RDWR|O_CREAT, 0666))) < 0)
852 #endif
853 {
854 msg("Cannot open output \"%s\".\n", tape);
855 if (!query("Do you want to retry the open?"))
856 dumpabort(0);
857 }
858 #ifdef RDUMP
859 if (!host)
860 #endif
861 {
862 struct mtget mt_stat;
863 magtapeout = ioctl(tapefd, MTIOCGET, (char *)&mt_stat) == 0;
864 /*
865 msg("Output is to %s\n",
866 magtapeout ? "tape" : "file/pipe");
867 */
868 }
869
870 enslave(); /* Share open tape file descriptor with slaves */
871
872 asize = 0;
873 blocksthisvol = 0;
874 if (top)
875 newtape++; /* new tape signal */
876 spcl.c_count = slp->count;
877 /*
878 * measure firstrec in TP_BSIZE units since restore doesn't
879 * know the correct ntrec value...
880 */
881 spcl.c_firstrec = slp->firstrec;
882 spcl.c_volume++;
883 spcl.c_type = TS_TAPE;
884 spcl.c_flags |= DR_NEWHEADER;
885 spcl.c_ntrec = ntrec;
886 if (compressed)
887 spcl.c_flags |= DR_COMPRESSED;
888 writeheader((dump_ino_t)slp->inode);
889 spcl.c_flags &=~ DR_NEWHEADER;
890 msg("Volume %d started with block %ld at: %s", tapeno,
891 spcl.c_tapea, ctime(&tstart_volume));
892 if (tapeno > 1)
893 msg("Volume %d begins with blocks from inode %d\n",
894 tapeno, slp->inode);
895 }
896 }
897
898 void
899 dumpabort(int signo)
900 {
901
902 if (master != 0 && master != getpid())
903 /* Signals master to call dumpabort */
904 (void) kill(master, SIGTERM);
905 else {
906 killall();
907 msg("The ENTIRE dump is aborted.\n");
908 }
909 #ifdef RDUMP
910 rmtclose();
911 #endif
912 Exit(X_ABORT);
913 }
914
915 void
916 Exit(int status)
917 {
918
919 #ifdef TDEBUG
920 msg("pid = %d exits with status %d\n", getpid(), status);
921 #endif /* TDEBUG */
922 exit(status);
923 }
924
925 /*
926 * proceed - handler for SIGUSR2, used to synchronize IO between the slaves.
927 */
928 static void
929 proceed(int signo)
930 {
931 if (ready)
932 siglongjmp(jmpbuf, 1);
933 caught++;
934 }
935
936 void
937 enslave(void)
938 {
939 int cmd[2];
940 #ifdef LINUX_FORK_BUG
941 int i, j;
942 #else
943 register int i, j;
944 #endif
945
946 master = getpid();
947
948 { struct sigaction sa;
949 memset(&sa, 0, sizeof sa);
950 sigemptyset(&sa.sa_mask);
951 sa.sa_handler = dumpabort;
952 sigaction(SIGTERM, &sa, NULL); /* Slave sends SIGTERM on dumpabort() */
953 sa.sa_handler = sigpipe;
954 sigaction(SIGPIPE, &sa, NULL);
955 sa.sa_handler = proceed;
956 sa.sa_flags = SA_RESTART;
957 sigaction(SIGUSR2, &sa, NULL); /* Slave sends SIGUSR2 to next slave */
958 }
959
960 for (i = 0; i < SLAVES; i++) {
961 if (i == slp - &slaves[0]) {
962 caught = 1;
963 } else {
964 caught = 0;
965 }
966
967 if (socketpair(AF_UNIX, SOCK_STREAM, 0, cmd) < 0 ||
968 (slaves[i].pid = fork()) < 0)
969 quit("too many slaves, %d (recompile smaller): %s\n",
970 i, strerror(errno));
971
972 slaves[i].fd = cmd[1];
973 slaves[i].sent = 0;
974 if (slaves[i].pid == 0) { /* Slave starts up here */
975 sigset_t sigs;
976 for (j = 0; j <= i; j++)
977 (void) close(slaves[j].fd);
978 sigemptyset(&sigs);
979 sigaddset(&sigs, SIGINT); /* Master handles this */
980 #if defined(SIGINFO)
981 sigaddset(&sigs, SIGINFO);
982 #endif
983 sigprocmask(SIG_BLOCK, &sigs, NULL);
984
985 #ifdef LINUX_FORK_BUG
986 if (atomic_write( cmd[0], (char *) &i, sizeof i)
987 != sizeof i)
988 quit("master/slave protocol botched 3\n");
989 #endif
990 doslave(cmd[0], i, (slaves[i].pid == slp->pid));
991 Exit(X_FINOK);
992 }
993 else
994 close(cmd[0]);
995 }
996
997 #ifdef LINUX_FORK_BUG
998 /*
999 * Wait for all slaves to _actually_ start to circumvent a bug in
1000 * Linux kernels >= 2.1.3 where a signal sent to a child that hasn't
1001 * returned from fork() causes a SEGV in the child process
1002 */
1003 for (i = 0; i < SLAVES; i++)
1004 if (atomic_read( slaves[i].fd, (char *) &j, sizeof j) != sizeof j)
1005 quit("master/slave protocol botched 4\n");
1006 #endif
1007
1008 for (i = 0; i < SLAVES; i++)
1009 (void) atomic_write( slaves[i].fd,
1010 (char *) &slaves[(i + 1) % SLAVES].pid,
1011 sizeof slaves[0].pid);
1012
1013 master = 0;
1014 }
1015
1016 void
1017 killall(void)
1018 {
1019 register int i;
1020
1021 for (i = 0; i < SLAVES; i++)
1022 if (slaves[i].pid > 0) {
1023 (void) kill(slaves[i].pid, SIGKILL);
1024 slaves[i].sent = 0;
1025 }
1026 }
1027
1028 /*
1029 * Synchronization - each process waits for a SIGUSR2 from the
1030 * previous process before writing to the tape, and sends SIGUSR2
1031 * to the next process when the tape write completes. On tape errors
1032 * a SIGUSR1 is sent to the master which then terminates all of the
1033 * slaves.
1034 */
1035 static void
1036 doslave(int cmd, int slave_number, int first)
1037 {
1038 register int nread;
1039 int nextslave, size, eot_count, bufsize;
1040 volatile int wrote = 0;
1041 char *buffer;
1042 #if defined(HAVE_ZLIB) || defined(HAVE_BZLIB)
1043 struct tapebuf *comp_buf = NULL;
1044 int compresult, do_compress = !first;
1045 unsigned long worklen;
1046 #ifdef HAVE_BZLIB
1047 unsigned int worklen2;
1048 #endif
1049 #endif /* HAVE_ZLIB || HAVE_BZLIB */
1050 struct slave_results returns;
1051 #ifdef __linux__
1052 errcode_t retval;
1053 #endif
1054 #ifdef USE_QFA
1055 long curtapepos;
1056 union u_spcl *uspclptr;
1057 struct s_spcl *spclptr;
1058 #endif /* USE_QFA */
1059 sigset_t set;
1060
1061 sigemptyset(&set);
1062 sigaddset(&set, SIGUSR2);
1063 sigprocmask(SIG_BLOCK, &set, NULL);
1064 sigemptyset(&set);
1065
1066 /*
1067 * Need our own seek pointer.
1068 */
1069 (void) close(diskfd);
1070 if ((diskfd = OPEN(disk, O_RDONLY)) < 0)
1071 quit("slave couldn't reopen disk: %s\n", strerror(errno));
1072 #ifdef __linux__
1073 #ifdef BLKFLSBUF
1074 (void)ioctl(diskfd, BLKFLSBUF);
1075 #endif
1076 ext2fs_close(fs);
1077 retval = dump_fs_open(disk, &fs);
1078 if (retval)
1079 quit("slave couldn't reopen disk: %s\n", error_message(retval));
1080 #endif /* __linux__ */
1081
1082 /*
1083 * Need the pid of the next slave in the loop...
1084 */
1085 if ((nread = atomic_read( cmd, (char *)&nextslave, sizeof nextslave))
1086 != sizeof nextslave) {
1087 quit("master/slave protocol botched - didn't get pid of next slave.\n");
1088 }
1089
1090 #if defined(HAVE_ZLIB) || defined(HAVE_BZLIB)
1091 /* if we're doing a compressed dump, allocate the compress buffer */
1092 if (compressed) {
1093 comp_buf = malloc(sizeof(struct tapebuf) + TP_BSIZE + writesize);
1094 if (comp_buf == NULL)
1095 quit("couldn't allocate a compress buffer.\n");
1096 if (bzipflag)
1097 comp_buf->flags = COMPRESS_BZLIB;
1098 else
1099 comp_buf->flags = COMPRESS_ZLIB;
1100 }
1101 #endif /* HAVE_ZLIB || HAVE_BZLIB */
1102
1103 /*
1104 * Get list of blocks to dump, read the blocks into tape buffer
1105 */
1106 while ((nread = atomic_read( cmd, (char *)slp->req, reqsiz)) == reqsiz) {
1107 register struct req *p = slp->req;
1108
1109 for (trecno = 0; trecno < ntrec;
1110 trecno += p->count, p += p->count) {
1111 if (p->dblk) { /* read a disk block */
1112 bread(p->dblk, slp->tblock[trecno],
1113 p->count * TP_BSIZE);
1114 } else { /* read record from pipe */
1115 if (p->count != 1 || atomic_read( cmd,
1116 (char *)slp->tblock[trecno],
1117 TP_BSIZE) != TP_BSIZE)
1118 quit("master/slave protocol botched.\n");
1119 }
1120 }
1121
1122 /* Try to write the data... */
1123 wrote = 0;
1124 eot_count = 0;
1125 size = 0;
1126 buffer = (char *) slp->tblock[0]; /* set write pointer */
1127 bufsize = writesize; /* length to write */
1128 returns.clen = returns.unclen = bufsize;
1129
1130 #if defined(HAVE_ZLIB) || defined(HAVE_BZLIB)
1131 /*
1132 * When writing a compressed dump, each block except
1133 * the first one on each tape is written
1134 * from struct tapebuf with an 4 byte prefix
1135 * followed by the data. This can be less than
1136 * writesize. Restore, on a short read, can compare the
1137 * length read to the compressed length in the header
1138 * to verify that the read was good. Blocks which don't
1139 * compress well are written uncompressed.
1140 * The first block written by each slave is not compressed
1141 * and does not have a prefix.
1142 */
1143
1144 if (compressed && do_compress) {
1145 comp_buf->length = bufsize;
1146 worklen = TP_BSIZE + writesize;
1147 compresult = 1;
1148 #ifdef HAVE_ZLIB
1149 if (!bzipflag) {
1150 compresult = compress2(comp_buf->buf,
1151 &worklen,
1152 (char *)slp->tblock[0],
1153 writesize,
1154 compressed);
1155 if (compresult == Z_OK)
1156 compresult = 1;
1157 else
1158 compresult = 0;
1159 }
1160 #endif /* HAVE_ZLIB */
1161 #ifdef HAVE_BZLIB
1162 if (bzipflag) {
1163 worklen2 = worklen;
1164 compresult = BZ2_bzBuffToBuffCompress(
1165 comp_buf->buf,
1166 &worklen2,
1167 (char *)slp->tblock[0],
1168 writesize,
1169 compressed,
1170 0, 30);
1171 worklen = worklen2;
1172 if (compresult == BZ_OK)
1173 compresult = 1;
1174 else
1175 compresult = 0;
1176 }
1177
1178 #endif /* HAVE_BZLIB */
1179 if (compresult && worklen <= (writesize - 16)) {
1180 /* write the compressed buffer */
1181 comp_buf->length = worklen;
1182 comp_buf->compressed = 1;
1183 buffer = (char *) comp_buf;
1184 returns.clen = bufsize = worklen + sizeof(struct tapebuf);
1185 }
1186 else {
1187 /* write the data uncompressed */
1188 comp_buf->length = writesize;
1189 comp_buf->compressed = 0;
1190 buffer = (char *) comp_buf;
1191 returns.clen = bufsize = writesize + sizeof(struct tapebuf);
1192 returns.unclen = returns.clen;
1193 memcpy(comp_buf->buf, (char *)slp->tblock[0], writesize);
1194 }
1195 }
1196 /* compress the remaining blocks if we're compressing */
1197 do_compress = compressed;
1198 #endif /* HAVE_ZLIB || HAVE_BZLIB */
1199
1200 if (sigsetjmp(jmpbuf, 1) == 0) {
1201 ready = 1;
1202 if (!caught)
1203 sigsuspend(&set);
1204 }
1205 ready = 0;
1206 caught = 0;
1207
1208 #ifdef USE_QFA
1209 if (gTapeposfd >= 0) {
1210 int i;
1211 for (i = 0; i < ntrec; ++i) {
1212 uspclptr = (union u_spcl *)&slp->tblock[i];
1213 spclptr = &uspclptr->s_spcl;
1214 if ((spclptr->c_magic == NFS_MAGIC) &&
1215 (spclptr->c_type == TS_INODE) &&
1216 (spclptr->c_date == gThisDumpDate)) {
1217 /* if an error occured previously don't
1218 * try again */
1219 if (gtperr == 0) {
1220 if ((gtperr = GetTapePos(&curtapepos)) == 0)
1221 MkTapeString(spclptr, curtapepos);
1222 }
1223 }
1224 }
1225 }
1226 #endif /* USE_QFA */
1227
1228 while (eot_count < 10 && size < bufsize) {
1229 #ifdef RDUMP
1230 if (host)
1231 wrote = rmtwrite(buffer + size, bufsize - size);
1232 else
1233 #endif
1234 wrote = write(tapefd, buffer + size, bufsize - size);
1235 #ifdef WRITEDEBUG
1236 printf("slave %d wrote %d\n", slave_number, wrote);
1237 #endif
1238 if (wrote < 0)
1239 break;
1240 if (wrote == 0)
1241 eot_count++;
1242 size += wrote;
1243 }
1244
1245 #ifdef WRITEDEBUG
1246 if (size != bufsize)
1247 printf("slave %d only wrote %d out of %d bytes and gave up.\n",
1248 slave_number, size, bufsize);
1249 #endif
1250
1251 /*
1252 * Handle ENOSPC as an EOT condition.
1253 */
1254 if (wrote < 0 && errno == ENOSPC) {
1255 wrote = 0;
1256 eot_count++;
1257 }
1258
1259 if (eot_count > 0)
1260 returns.clen = returns.unclen = 0;
1261
1262 /*
1263 * pass errno back to master for special handling
1264 */
1265 if (wrote < 0)
1266 returns.unclen = -errno;
1267
1268 /*
1269 * pass size of data and size of write back to master
1270 * (for EOT handling)
1271 */
1272 (void) atomic_write( cmd, (char *)&returns, sizeof returns);
1273
1274 /*
1275 * Signal the next slave to go.
1276 */
1277 (void) kill(nextslave, SIGUSR2);
1278 }
1279 if (nread != 0)
1280 quit("error reading command pipe: %s\n", strerror(errno));
1281 }
1282
1283 /*
1284 * Since a read from a pipe may not return all we asked for,
1285 * or a write may not write all we ask if we get a signal,
1286 * loop until the count is satisfied (or error).
1287 */
1288 static ssize_t
1289 atomic_read(int fd, void *buf, size_t count)
1290 {
1291 int got, need = count;
1292
1293 do {
1294 while ((got = read(fd, buf, need)) > 0 && (need -= got) > 0)
1295 (char *)buf += got;
1296 } while (got == -1 && errno == EINTR);
1297 return (got < 0 ? got : count - need);
1298 }
1299
1300 /*
1301 * Since a read from a pipe may not return all we asked for,
1302 * or a write may not write all we ask if we get a signal,
1303 * loop until the count is satisfied (or error).
1304 */
1305 static ssize_t
1306 atomic_write(int fd, const void *buf, size_t count)
1307 {
1308 int got, need = count;
1309
1310 do {
1311 while ((got = write(fd, buf, need)) > 0 && (need -= got) > 0)
1312 (char *)buf += got;
1313 } while (got == -1 && errno == EINTR);
1314 return (got < 0 ? got : count - need);
1315 }
1316
1317
1318 #ifdef USE_QFA
1319 /*
1320 * read the current tape position
1321 */
1322 static int
1323 GetTapePos(long *pos)
1324 {
1325 int err = 0;
1326
1327 if (magtapeout) {
1328 *pos = 0;
1329 err = (ioctl(tapefd, MTIOCPOS, pos) < 0);
1330 }
1331 else {
1332 *pos = lseek(tapefd, 0, SEEK_CUR);
1333 err = (*pos < 0);
1334 }
1335 if (err) {
1336 err = errno;
1337 msg("[%ld] error: %d (getting tapepos: %ld)\n", getpid(),
1338 err, *pos);
1339 return err;
1340 }
1341 return err;
1342 }
1343
1344 static void
1345 MkTapeString(struct s_spcl *spclptr, long curtapepos) {
1346
1347 #ifdef DEBUG_QFA
1348 msg("inode %ld at tapepos %ld\n", spclptr->c_inumber, curtapepos);
1349 #endif
1350
1351 snprintf(gTps, sizeof(gTps), "%ld\t%d\t%ld\n",
1352 (unsigned long)spclptr->c_inumber,
1353 tapeno,
1354 curtapepos);
1355 gTps[sizeof(gTps) - 1] = '\0';
1356 if (write(gTapeposfd, gTps, strlen(gTps)) != strlen(gTps)) {
1357 warn("error writing tapepos file.\n");
1358 }
1359 }
1360 #endif /* USE_QFA */