]> git.wh0rd.org - dump.git/blob - dump/tape.c
BLKSFBUF ioctl in dump.
[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 1.55 2001/09/06 09:00:32 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/time.h>
72 #include <sys/wait.h>
73 #include <sys/mtio.h>
74 #ifdef __linux__
75 #include <linux/fs.h>
76 #undef atomic_read /* this get wrongly defined in kernel */
77 /* headers and we don't want it */
78 #ifdef HAVE_EXT2FS_EXT2_FS_H
79 #include <ext2fs/ext2_fs.h>
80 #else
81 #include <linux/ext2_fs.h>
82 #endif
83 #include <ext2fs/ext2fs.h>
84 #include <bsdcompat.h>
85 #elif defined sunos
86 #include <sys/vnode.h>
87
88 #include <ufs/fs.h>
89 #include <ufs/inode.h>
90 #else
91 #include <ufs/ufs/dinode.h>
92 #include <ufs/ffs/fs.h>
93 #endif /* __linux__ */
94
95 #include <protocols/dumprestore.h>
96
97 #ifdef HAVE_ZLIB
98 #include <zlib.h>
99 #endif /* HAVE_ZLIB */
100
101 #ifdef HAVE_BZLIB
102 #include <bzlib.h>
103 #endif /* HAVE_BZLIB */
104
105 #include "dump.h"
106
107 int writesize; /* size of malloc()ed buffer for tape */
108 long lastspclrec = -1; /* tape block number of last written header */
109 int trecno = 0; /* next record to write in current block */
110 extern long blocksperfile; /* number of blocks per output file */
111 long blocksthisvol; /* number of blocks on current output file */
112 extern int ntrec; /* blocking factor on tape */
113 extern int cartridge;
114 extern char *host;
115 char *nexttape;
116 extern pid_t rshpid;
117 int eot_code = 1;
118 long long tapea_bytes = 0; /* bytes_written at start of current volume */
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
128 /*
129 * Concurrent dump mods (Caltech) - disk block reading and tape writing
130 * are exported to several slave processes. While one slave writes the
131 * tape, the others read disk blocks; they pass control of the tape in
132 * a ring via signals. The parent process traverses the filesystem and
133 * sends writeheader()'s and lists of daddr's to the slaves via pipes.
134 * The following structure defines the instruction packets sent to slaves.
135 */
136 struct req {
137 daddr_t dblk;
138 int count;
139 };
140 int reqsiz;
141
142 struct slave_results {
143 ssize_t unclen; /* uncompressed length */
144 ssize_t clen; /* compressed length */
145 };
146
147 #define SLAVES 3 /* 1 slave writing, 1 reading, 1 for slack */
148 struct slave {
149 int tapea; /* header number at start of this chunk */
150 int count; /* count to next header (used for TS_TAPE */
151 /* after EOT) */
152 int inode; /* inode that we are currently dealing with */
153 int fd; /* FD for this slave */
154 int pid; /* PID for this slave */
155 int sent; /* 1 == we've sent this slave requests */
156 int firstrec; /* record number of this block */
157 char (*tblock)[TP_BSIZE]; /* buffer for data blocks */
158 struct req *req; /* buffer for requests */
159 } slaves[SLAVES+1];
160 struct slave *slp;
161
162 char (*nextblock)[TP_BSIZE];
163
164 static time_t tstart_volume; /* time of volume start */
165 static int tapea_volume; /* value of spcl.c_tapea at volume start */
166
167 int master; /* pid of master, for sending error signals */
168 int tenths; /* length of tape overhead per block written */
169 static int caught; /* have we caught the signal to proceed? */
170 static int ready; /* have we reached the lock point without having */
171 /* received the SIGUSR2 signal from the prev slave? */
172 static sigjmp_buf jmpbuf; /* where to jump to if we are ready when the */
173 /* SIGUSR2 arrives from the previous slave */
174 #ifdef USE_QFA
175 static int gtperr = 0;
176 #endif
177
178 int
179 alloctape(void)
180 {
181 int pgoff = getpagesize() - 1;
182 char *buf;
183 int i;
184
185 writesize = ntrec * TP_BSIZE;
186 reqsiz = (ntrec + 1) * sizeof(struct req);
187 /*
188 * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode
189 * (see DEC TU80 User's Guide). The shorter gaps of 6250-bpi require
190 * repositioning after stopping, i.e, streaming mode, where the gap is
191 * variable, 0.30" to 0.45". The gap is maximal when the tape stops.
192 */
193 if (blocksperfile == 0 && !unlimited)
194 tenths = (cartridge ? 16 : density == 625 ? 5 : 8);
195 else {
196 tenths = 0;
197 density = 1;
198 }
199 /*
200 * Allocate tape buffer contiguous with the array of instruction
201 * packets, so flushtape() can write them together with one write().
202 * Align tape buffer on page boundary to speed up tape write().
203 */
204 for (i = 0; i <= SLAVES; i++) {
205 buf = (char *)
206 malloc((unsigned)(reqsiz + writesize + pgoff + TP_BSIZE));
207 if (buf == NULL)
208 return(0);
209 slaves[i].tblock = (char (*)[TP_BSIZE])
210 #ifdef __linux__
211 (((long)&buf[reqsiz] + pgoff) &~ pgoff);
212 #else
213 (((long)&buf[ntrec + 1] + pgoff) &~ pgoff);
214 #endif
215 slaves[i].req = (struct req *)slaves[i].tblock - ntrec - 1;
216 }
217 slp = &slaves[0];
218 slp->count = 1;
219 slp->tapea = 0;
220 slp->firstrec = 0;
221 nextblock = slp->tblock;
222 return(1);
223 }
224
225 void
226 writerec(const void *dp, int isspcl)
227 {
228
229 slp->req[trecno].dblk = (daddr_t)0;
230 slp->req[trecno].count = 1;
231 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
232 *(union u_spcl *)(*(nextblock)) = *(union u_spcl *)dp;
233 nextblock++;
234 if (isspcl)
235 lastspclrec = spcl.c_tapea;
236 trecno++;
237 spcl.c_tapea++;
238 if (trecno >= ntrec)
239 flushtape();
240 }
241
242 void
243 dumpblock(daddr_t blkno, int size)
244 {
245 int avail, tpblks, dblkno;
246
247 dblkno = fsbtodb(sblock, blkno);
248 tpblks = size >> tp_bshift;
249 while ((avail = MIN(tpblks, ntrec - trecno)) > 0) {
250 slp->req[trecno].dblk = dblkno;
251 slp->req[trecno].count = avail;
252 trecno += avail;
253 spcl.c_tapea += avail;
254 if (trecno >= ntrec)
255 flushtape();
256 dblkno += avail << (tp_bshift - dev_bshift);
257 tpblks -= avail;
258 }
259 }
260
261 int nogripe = 0;
262
263 static void
264 tperror(int errnum)
265 {
266
267 if (pipeout) {
268 msg("write error on %s: %s\n", tape, strerror(errnum));
269 quit("Cannot recover\n");
270 /* NOTREACHED */
271 }
272 msg("write error %d blocks into volume %d: %s\n",
273 blocksthisvol, tapeno, strerror(errnum));
274 broadcast("DUMP WRITE ERROR!\n");
275 if (query("Do you want to rewrite this volume?")) {
276 msg("Closing this volume. Prepare to restart with new media;\n");
277 msg("this dump volume will be rewritten.\n");
278 killall();
279 nogripe = 1;
280 close_rewind();
281 Exit(X_REWRITE);
282 }
283 if (query("Do you want to start the next tape?"))
284 return;
285 dumpabort(0);
286 }
287
288 static void
289 sigpipe(int signo)
290 {
291
292 quit("Broken pipe\n");
293 }
294
295 /*
296 * do_stats --
297 * Update xferrate stats
298 */
299 time_t
300 do_stats(void)
301 {
302 time_t tnow, ttaken;
303 int blocks;
304
305 tnow = time(NULL);
306 ttaken = tnow - tstart_volume;
307 blocks = spcl.c_tapea - tapea_volume;
308 msg("Volume %d completed at: %s", tapeno, ctime(&tnow));
309 if (! compressed)
310 msg("Volume %d %ld tape blocks (%.2fMB)\n", tapeno,
311 blocks, ((double)blocks * TP_BSIZE / 1048576));
312 if (ttaken > 0) {
313 long volkb = (bytes_written - tapea_bytes) / 1024;
314 long txfrate = volkb / ttaken;
315 msg("Volume %d took %d:%02d:%02d\n", tapeno,
316 ttaken / 3600, (ttaken % 3600) / 60, ttaken % 60);
317 msg("Volume %d transfer rate: %ld kB/s\n", tapeno,
318 txfrate);
319 xferrate += txfrate;
320 if (compressed) {
321 double rate = .0005 + (double) blocks / (double) volkb;
322 msg("Volume %d %ldkB uncompressed, %ldkB compressed,"
323 " %1.3f:1\n",
324 tapeno, blocks, volkb, rate);
325 }
326 }
327 return(tnow);
328 }
329
330 char *
331 mktimeest(time_t tnow)
332 {
333 static char msgbuf[128];
334 time_t deltat;
335
336 msgbuf[0] = '\0';
337
338 if (blockswritten < 500)
339 return NULL;
340 if (blockswritten > tapesize)
341 tapesize = blockswritten;
342 deltat = tstart_writing - tnow + (1.0 * (tnow - tstart_writing))
343 / blockswritten * tapesize;
344 if (tnow > tstart_volume)
345 (void)snprintf(msgbuf, sizeof(msgbuf),
346 "%3.2f%% done at %ld kB/s, finished in %d:%02d\n",
347 (blockswritten * 100.0) / tapesize,
348 (spcl.c_tapea - tapea_volume) / (tnow - tstart_volume),
349 (int)(deltat / 3600), (int)((deltat % 3600) / 60));
350 else
351 (void)snprintf(msgbuf, sizeof(msgbuf),
352 "%3.2f%% done, finished in %d:%02d\n",
353 (blockswritten * 100.0) / tapesize,
354 (int)(deltat / 3600), (int)((deltat % 3600) / 60));
355
356 return msgbuf;
357 }
358
359 #if defined(SIGINFO)
360 /*
361 * statussig --
362 * information message upon receipt of SIGINFO
363 */
364 void
365 statussig(int notused)
366 {
367 int save_errno = errno;
368 char *buf;
369
370 buf = mktimeest(time(NULL));
371 if (buf)
372 write(STDERR_FILENO, buf, strlen(buf));
373 errno = save_errno;
374 }
375 #endif
376
377 static void
378 flushtape(void)
379 {
380 int i, blks, got;
381 long lastfirstrec;
382 struct slave_results returned;
383
384 int siz = (char *)nextblock - (char *)slp->req;
385
386 slp->req[trecno].count = 0; /* Sentinel */
387
388 if (atomic_write( slp->fd, (char *)slp->req, siz) != siz)
389 quit("error writing command pipe: %s\n", strerror(errno));
390 slp->sent = 1; /* we sent a request, read the response later */
391
392 lastfirstrec = slp->firstrec;
393
394 if (++slp >= &slaves[SLAVES])
395 slp = &slaves[0];
396
397 /* Read results back from next slave */
398 if (slp->sent) {
399 if (atomic_read( slp->fd, (char *)&returned, sizeof returned)
400 != sizeof returned) {
401 perror(" DUMP: error reading command pipe in master");
402 dumpabort(0);
403 }
404 got = returned.unclen;
405 bytes_written += returned.clen;
406 if (returned.unclen == returned.clen)
407 uncomprblks++;
408 slp->sent = 0;
409
410 /* Check for errors or end of tape */
411 if (got <= 0) {
412 /* Check for errors */
413 if (got < 0)
414 tperror(-got);
415 else
416 msg("End of tape detected\n");
417
418 /*
419 * Drain the results, don't care what the values were.
420 * If we read them here then trewind won't...
421 */
422 for (i = 0; i < SLAVES; i++) {
423 if (slaves[i].sent) {
424 if (atomic_read( slaves[i].fd,
425 (char *)&returned, sizeof returned)
426 != sizeof returned) {
427 perror(" DUMP: error reading command pipe in master");
428 dumpabort(0);
429 }
430 slaves[i].sent = 0;
431 }
432 }
433
434 close_rewind();
435 rollforward();
436 return;
437 }
438 }
439
440 blks = 0;
441 if (spcl.c_type != TS_END) {
442 for (i = 0; i < spcl.c_count; i++)
443 if (spcl.c_addr[i] != 0)
444 blks++;
445 }
446 slp->count = lastspclrec + blks + 1 - spcl.c_tapea;
447 slp->tapea = spcl.c_tapea;
448 slp->firstrec = lastfirstrec + ntrec;
449 slp->inode = curino;
450 nextblock = slp->tblock;
451 trecno = 0;
452 asize += tenths + returned.clen / density;
453 blockswritten += ntrec;
454 blocksthisvol += ntrec;
455 if (!pipeout && !unlimited) {
456 if (blocksperfile) {
457 if ( compressed ? (bytes_written - tapea_bytes + SLAVES * (writesize + sizeof(struct tapebuf))) >= blocksperfile * 1024
458 : blocksthisvol >= blocksperfile ) {
459 close_rewind();
460 startnewtape(0);
461 }
462 }
463 else if (asize > tsize) {
464 close_rewind();
465 startnewtape(0);
466 }
467 }
468 timeest();
469 }
470
471 time_t
472 trewind(void)
473 {
474 int f;
475 int got;
476 struct slave_results returned;
477
478 for (f = 0; f < SLAVES; f++) {
479 /*
480 * Drain the results, but unlike EOT we DO (or should) care
481 * what the return values were, since if we detect EOT after
482 * we think we've written the last blocks to the tape anyway,
483 * we have to replay those blocks with rollforward.
484 *
485 * fixme: punt for now.
486 */
487 if (slaves[f].sent) {
488 if (atomic_read( slaves[f].fd, (char *)&returned, sizeof returned)
489 != sizeof returned) {
490 perror(" DUMP: error reading command pipe in master");
491 dumpabort(0);
492 }
493 got = returned.unclen;
494 bytes_written += returned.clen;
495 if (returned.unclen == returned.clen)
496 uncomprblks++;
497 slaves[f].sent = 0;
498
499 if (got < 0)
500 tperror(-got);
501
502 if (got == 0) {
503 msg("EOT detected in last 2 tape records!\n");
504 msg("Use a longer tape, decrease the size estimate\n");
505 quit("or use no size estimate at all.\n");
506 }
507 }
508 (void) close(slaves[f].fd);
509 }
510 while (wait((int *)NULL) >= 0) /* wait for any signals from slaves */
511 /* void */;
512
513 if (!pipeout) {
514
515 msg("Closing %s\n", tape);
516
517 #ifdef RDUMP
518 if (host) {
519 rmtclose();
520 while (rmtopen(tape, 0) < 0)
521 sleep(10);
522 rmtclose();
523 }
524 else
525 #endif
526 {
527 (void) close(tapefd);
528 if (!fifoout) {
529 while ((f = OPEN(tape, 0)) < 0)
530 sleep (10);
531 (void) close(f);
532 }
533 }
534 eot_code = 1;
535 if (eot_script && spcl.c_type != TS_END) {
536 msg("Launching %s\n", eot_script);
537 eot_code = system_command(eot_script, tape, tapeno);
538 }
539 if (eot_code != 0 && eot_code != 1) {
540 msg("Dump aborted by the end of tape script\n");
541 dumpabort(0);
542 }
543 }
544 return do_stats();
545 }
546
547
548 void
549 close_rewind(void)
550 {
551 (void)trewind();
552 if (nexttape || Mflag || (eot_code == 0) )
553 return;
554 if (!nogripe) {
555 msg("Change Volumes: Mount volume #%d\n", tapeno+1);
556 broadcast("CHANGE DUMP VOLUMES!\7\7\n");
557 }
558 while (!query("Is the new volume mounted and ready to go?"))
559 if (query("Do you want to abort?")) {
560 dumpabort(0);
561 /*NOTREACHED*/
562 }
563 }
564
565 void
566 rollforward(void)
567 {
568 register struct req *p, *q = NULL, *prev;
569 register struct slave *tslp;
570 int i, size, savedtapea, got;
571 union u_spcl *ntb, *otb;
572 struct slave_results returned;
573 #ifdef __linux__
574 int blks;
575 long lastfirstrec;
576 #endif
577 tslp = &slaves[SLAVES];
578 ntb = (union u_spcl *)tslp->tblock[1];
579
580 /*
581 * Each of the N slaves should have requests that need to
582 * be replayed on the next tape. Use the extra slave buffers
583 * (slaves[SLAVES]) to construct request lists to be sent to
584 * each slave in turn.
585 */
586 for (i = 0; i < SLAVES; i++) {
587 q = &tslp->req[1];
588 otb = (union u_spcl *)slp->tblock;
589
590 /*
591 * For each request in the current slave, copy it to tslp.
592 */
593
594 prev = NULL;
595 for (p = slp->req; p->count > 0; p += p->count) {
596 *q = *p;
597 if (p->dblk == 0)
598 *ntb++ = *otb++; /* copy the datablock also */
599 prev = q;
600 q += q->count;
601 }
602 if (prev == NULL)
603 quit("rollforward: protocol botch");
604 if (prev->dblk != 0)
605 prev->count -= 1;
606 else
607 ntb--;
608 q -= 1;
609 q->count = 0;
610 q = &tslp->req[0];
611 if (i == 0) {
612 q->dblk = 0;
613 q->count = 1;
614 trecno = 0;
615 nextblock = tslp->tblock;
616 savedtapea = spcl.c_tapea;
617 spcl.c_tapea = slp->tapea;
618 startnewtape(0);
619 spcl.c_tapea = savedtapea;
620 lastspclrec = savedtapea - 1;
621 }
622 size = (char *)ntb - (char *)q;
623 if (atomic_write( slp->fd, (char *)q, size) != size) {
624 perror(" DUMP: error writing command pipe");
625 dumpabort(0);
626 }
627 slp->sent = 1;
628 #ifdef __linux__
629 lastfirstrec = slp->firstrec;
630 #endif
631 if (++slp >= &slaves[SLAVES])
632 slp = &slaves[0];
633
634 q->count = 1;
635
636 if (prev->dblk != 0) {
637 /*
638 * If the last one was a disk block, make the
639 * first of this one be the last bit of that disk
640 * block...
641 */
642 q->dblk = prev->dblk +
643 prev->count * (TP_BSIZE / DEV_BSIZE);
644 ntb = (union u_spcl *)tslp->tblock;
645 } else {
646 /*
647 * It wasn't a disk block. Copy the data to its
648 * new location in the buffer.
649 */
650 q->dblk = 0;
651 *((union u_spcl *)tslp->tblock) = *ntb;
652 ntb = (union u_spcl *)tslp->tblock[1];
653 }
654 }
655 slp->req[0] = *q;
656 nextblock = slp->tblock;
657 if (q->dblk == 0) {
658 #ifdef __linux__
659 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
660 *(union u_spcl *)(*nextblock) = *(union u_spcl *)tslp->tblock;
661 #endif
662 nextblock++;
663 }
664 trecno = 1;
665
666 /*
667 * Clear the first slaves' response. One hopes that it
668 * worked ok, otherwise the tape is much too short!
669 */
670 if (slp->sent) {
671 if (atomic_read( slp->fd, (char *)&returned, sizeof returned)
672 != sizeof returned) {
673 perror(" DUMP: error reading command pipe in master");
674 dumpabort(0);
675 }
676 got = returned.unclen;
677 bytes_written += returned.clen;
678 if (returned.clen == returned.unclen)
679 uncomprblks++;
680 slp->sent = 0;
681
682 if (got < 0)
683 tperror(-got);
684
685 if (got == 0) {
686 quit("EOT detected at start of the tape!\n");
687 }
688 }
689
690 #ifdef __linux__
691 blks = 0;
692 if (spcl.c_type != TS_END) {
693 for (i = 0; i < spcl.c_count; i++)
694 if (spcl.c_addr[i] != 0)
695 blks++;
696 }
697
698 slp->firstrec = lastfirstrec + ntrec;
699 slp->count = lastspclrec + blks + 1 - spcl.c_tapea;
700 slp->inode = curino;
701 asize += tenths + returned.clen / density;
702 blockswritten += ntrec;
703 blocksthisvol += ntrec;
704 #endif
705 }
706
707 /*
708 * We implement taking and restoring checkpoints on the tape level.
709 * When each tape is opened, a new process is created by forking; this
710 * saves all of the necessary context in the parent. The child
711 * continues the dump; the parent waits around, saving the context.
712 * If the child returns X_REWRITE, then it had problems writing that tape;
713 * this causes the parent to fork again, duplicating the context, and
714 * everything continues as if nothing had happened.
715 */
716 void
717 startnewtape(int top)
718 {
719 int parentpid;
720 int childpid;
721 int status;
722 int waitpid;
723 char *p;
724
725 #ifdef __linux__
726 sigset_t sigs;
727 sigemptyset(&sigs);
728 sigaddset(&sigs, SIGINT);
729 sigprocmask(SIG_BLOCK, &sigs, NULL);
730 #else /* __linux__ */
731 #ifdef sunos
732 void (*interrupt_save)();
733 #else
734 sig_t interrupt_save;
735 #endif
736 interrupt_save = signal(SIGINT, SIG_IGN);
737 #endif /* __linux__ */
738
739 parentpid = getpid();
740 tapea_volume = spcl.c_tapea;
741 tapea_bytes = bytes_written;
742 tstart_volume = time(NULL);
743
744 restore_check_point:
745 #ifdef __linux__
746 sigprocmask(SIG_UNBLOCK, &sigs, NULL);
747 #else
748 (void)signal(SIGINT, interrupt_save);
749 #endif
750 /*
751 * All signals are inherited...
752 */
753 childpid = fork();
754 if (childpid < 0) {
755 msg("Context save fork fails in parent %d\n", parentpid);
756 Exit(X_ABORT);
757 }
758 if (childpid != 0) {
759 /*
760 * PARENT:
761 * save the context by waiting
762 * until the child doing all of the work returns.
763 * don't catch the interrupt
764 */
765 #ifdef __linux__
766 sigprocmask(SIG_BLOCK, &sigs, NULL);
767 #else
768 signal(SIGINT, SIG_IGN);
769 #endif
770 #ifdef TDEBUG
771 msg("Tape: %d; parent process: %d child process %d\n",
772 tapeno+1, parentpid, childpid);
773 #endif /* TDEBUG */
774 while ((waitpid = wait(&status)) != childpid)
775 if (waitpid != rshpid)
776 msg("Parent %d waiting for child %d has another child %d return\n",
777 parentpid, childpid, waitpid);
778 if (status & 0xFF) {
779 msg("Child %d returns LOB status %o\n",
780 childpid, status&0xFF);
781 }
782 status = (status >> 8) & 0xFF;
783 #ifdef TDEBUG
784 switch(status) {
785 case X_FINOK:
786 msg("Child %d finishes X_FINOK\n", childpid);
787 break;
788 case X_ABORT:
789 msg("Child %d finishes X_ABORT\n", childpid);
790 break;
791 case X_REWRITE:
792 msg("Child %d finishes X_REWRITE\n", childpid);
793 break;
794 default:
795 msg("Child %d finishes unknown %d\n",
796 childpid, status);
797 break;
798 }
799 #endif /* TDEBUG */
800 switch(status) {
801 case X_FINOK:
802 Exit(X_FINOK);
803 case X_ABORT:
804 Exit(X_ABORT);
805 case X_REWRITE:
806 goto restore_check_point;
807 default:
808 msg("Bad return code from dump: %d\n", status);
809 Exit(X_ABORT);
810 }
811 /*NOTREACHED*/
812 } else { /* we are the child; just continue */
813 #ifdef TDEBUG
814 sleep(4); /* allow time for parent's message to get out */
815 msg("Child on Tape %d has parent %d, my pid = %d\n",
816 tapeno+1, parentpid, getpid());
817 #endif /* TDEBUG */
818 /*
819 * If we have a name like "/dev/rmt0,/dev/rmt1",
820 * use the name before the comma first, and save
821 * the remaining names for subsequent volumes.
822 */
823 tapeno++; /* current tape sequence */
824 if (Mflag) {
825 snprintf(tape, MAXPATHLEN, "%s%03d", tapeprefix, tapeno);
826 tape[MAXPATHLEN - 1] = '\0';
827 msg("Dumping volume %d on %s\n", tapeno, tape);
828 }
829 else if (nexttape || strchr(tapeprefix, ',')) {
830 if (nexttape && *nexttape)
831 tapeprefix = nexttape;
832 if ((p = strchr(tapeprefix, ',')) != NULL) {
833 *p = '\0';
834 nexttape = p + 1;
835 } else
836 nexttape = NULL;
837 strncpy(tape, tapeprefix, MAXPATHLEN);
838 tape[MAXPATHLEN - 1] = '\0';
839 msg("Dumping volume %d on %s\n", tapeno, tape);
840 }
841 #ifdef RDUMP
842 while ((tapefd = (host ? rmtopen(tape, 2) : pipeout ?
843 fileno(stdout) :
844 OPEN(tape, O_WRONLY|O_CREAT, 0666))) < 0)
845 #else
846 while ((tapefd = (pipeout ? fileno(stdout) :
847 OPEN(tape, O_RDWR|O_CREAT, 0666))) < 0)
848 #endif
849 {
850 msg("Cannot open output \"%s\".\n", tape);
851 if (!query("Do you want to retry the open?"))
852 dumpabort(0);
853 }
854
855 enslave(); /* Share open tape file descriptor with slaves */
856
857 asize = 0;
858 blocksthisvol = 0;
859 if (top)
860 newtape++; /* new tape signal */
861 spcl.c_count = slp->count;
862 /*
863 * measure firstrec in TP_BSIZE units since restore doesn't
864 * know the correct ntrec value...
865 */
866 spcl.c_firstrec = slp->firstrec;
867 spcl.c_volume++;
868 spcl.c_type = TS_TAPE;
869 spcl.c_flags |= DR_NEWHEADER;
870 spcl.c_ntrec = ntrec;
871 if (compressed)
872 spcl.c_flags |= DR_COMPRESSED;
873 writeheader((dump_ino_t)slp->inode);
874 spcl.c_flags &=~ DR_NEWHEADER;
875 msg("Volume %d started with block %ld at: %s", tapeno,
876 spcl.c_tapea, ctime(&tstart_volume));
877 if (tapeno > 1)
878 msg("Volume %d begins with blocks from inode %d\n",
879 tapeno, slp->inode);
880 }
881 }
882
883 void
884 dumpabort(int signo)
885 {
886
887 if (master != 0 && master != getpid())
888 /* Signals master to call dumpabort */
889 (void) kill(master, SIGTERM);
890 else {
891 killall();
892 msg("The ENTIRE dump is aborted.\n");
893 }
894 #ifdef RDUMP
895 rmtclose();
896 #endif
897 Exit(X_ABORT);
898 }
899
900 void
901 Exit(int status)
902 {
903
904 #ifdef TDEBUG
905 msg("pid = %d exits with status %d\n", getpid(), status);
906 #endif /* TDEBUG */
907 exit(status);
908 }
909
910 /*
911 * proceed - handler for SIGUSR2, used to synchronize IO between the slaves.
912 */
913 static void
914 proceed(int signo)
915 {
916 if (ready)
917 siglongjmp(jmpbuf, 1);
918 caught++;
919 }
920
921 void
922 enslave(void)
923 {
924 int cmd[2];
925 #ifdef LINUX_FORK_BUG
926 int i, j;
927 #else
928 register int i, j;
929 #endif
930
931 master = getpid();
932
933 { struct sigaction sa;
934 memset(&sa, 0, sizeof sa);
935 sigemptyset(&sa.sa_mask);
936 sa.sa_handler = dumpabort;
937 sigaction(SIGTERM, &sa, NULL); /* Slave sends SIGTERM on dumpabort() */
938 sa.sa_handler = sigpipe;
939 sigaction(SIGPIPE, &sa, NULL);
940 sa.sa_handler = proceed;
941 sa.sa_flags = SA_RESTART;
942 sigaction(SIGUSR2, &sa, NULL); /* Slave sends SIGUSR2 to next slave */
943 }
944
945 for (i = 0; i < SLAVES; i++) {
946 if (i == slp - &slaves[0]) {
947 caught = 1;
948 } else {
949 caught = 0;
950 }
951
952 if (socketpair(AF_UNIX, SOCK_STREAM, 0, cmd) < 0 ||
953 (slaves[i].pid = fork()) < 0)
954 quit("too many slaves, %d (recompile smaller): %s\n",
955 i, strerror(errno));
956
957 slaves[i].fd = cmd[1];
958 slaves[i].sent = 0;
959 if (slaves[i].pid == 0) { /* Slave starts up here */
960 sigset_t sigs;
961 for (j = 0; j <= i; j++)
962 (void) close(slaves[j].fd);
963 sigemptyset(&sigs);
964 sigaddset(&sigs, SIGINT); /* Master handles this */
965 #if defined(SIGINFO)
966 sigaddset(&sigs, SIGINFO);
967 #endif
968 sigprocmask(SIG_BLOCK, &sigs, NULL);
969
970 #ifdef LINUX_FORK_BUG
971 if (atomic_write( cmd[0], (char *) &i, sizeof i)
972 != sizeof i)
973 quit("master/slave protocol botched 3\n");
974 #endif
975 doslave(cmd[0], i, (slaves[i].pid == slp->pid));
976 Exit(X_FINOK);
977 }
978 else
979 close(cmd[0]);
980 }
981
982 #ifdef LINUX_FORK_BUG
983 /*
984 * Wait for all slaves to _actually_ start to circumvent a bug in
985 * Linux kernels >= 2.1.3 where a signal sent to a child that hasn't
986 * returned from fork() causes a SEGV in the child process
987 */
988 for (i = 0; i < SLAVES; i++)
989 if (atomic_read( slaves[i].fd, (char *) &j, sizeof j) != sizeof j)
990 quit("master/slave protocol botched 4\n");
991 #endif
992
993 for (i = 0; i < SLAVES; i++)
994 (void) atomic_write( slaves[i].fd,
995 (char *) &slaves[(i + 1) % SLAVES].pid,
996 sizeof slaves[0].pid);
997
998 master = 0;
999 }
1000
1001 void
1002 killall(void)
1003 {
1004 register int i;
1005
1006 for (i = 0; i < SLAVES; i++)
1007 if (slaves[i].pid > 0) {
1008 (void) kill(slaves[i].pid, SIGKILL);
1009 slaves[i].sent = 0;
1010 }
1011 }
1012
1013 /*
1014 * Synchronization - each process waits for a SIGUSR2 from the
1015 * previous process before writing to the tape, and sends SIGUSR2
1016 * to the next process when the tape write completes. On tape errors
1017 * a SIGUSR1 is sent to the master which then terminates all of the
1018 * slaves.
1019 */
1020 static void
1021 doslave(int cmd, int slave_number, int first)
1022 {
1023 register int nread;
1024 int nextslave, size, eot_count, bufsize;
1025 volatile int wrote = 0;
1026 char *buffer;
1027 #if defined(HAVE_ZLIB) || defined(HAVE_BZLIB)
1028 struct tapebuf *comp_buf = NULL;
1029 int compresult, do_compress = !first;
1030 unsigned long worklen;
1031 #ifdef HAVE_BZLIB
1032 unsigned int worklen2;
1033 #endif
1034 #endif /* HAVE_ZLIB || HAVE_BZLIB */
1035 struct slave_results returns;
1036 #ifdef __linux__
1037 errcode_t retval;
1038 #endif
1039 #ifdef USE_QFA
1040 long curtapepos;
1041 union u_spcl *uspclptr;
1042 struct s_spcl *spclptr;
1043 #endif /* USE_QFA */
1044
1045 /*
1046 * Need our own seek pointer.
1047 */
1048 (void) close(diskfd);
1049 if ((diskfd = OPEN(disk, O_RDONLY)) < 0)
1050 quit("slave couldn't reopen disk: %s\n", strerror(errno));
1051 #ifdef __linux__
1052 #ifdef BLKFLSBUF
1053 (void)ioctl(diskfd, BLKFLSBUF);
1054 #endif
1055 ext2fs_close(fs);
1056 retval = dump_fs_open(disk, &fs);
1057 if (retval)
1058 quit("slave couldn't reopen disk: %s\n", error_message(retval));
1059 #endif /* __linux__ */
1060
1061 /*
1062 * Need the pid of the next slave in the loop...
1063 */
1064 if ((nread = atomic_read( cmd, (char *)&nextslave, sizeof nextslave))
1065 != sizeof nextslave) {
1066 quit("master/slave protocol botched - didn't get pid of next slave.\n");
1067 }
1068
1069 #if defined(HAVE_ZLIB) || defined(HAVE_BZLIB)
1070 /* if we're doing a compressed dump, allocate the compress buffer */
1071 if (compressed) {
1072 comp_buf = malloc(sizeof(struct tapebuf) + TP_BSIZE + writesize);
1073 if (comp_buf == NULL)
1074 quit("couldn't allocate a compress buffer.\n");
1075 if (bzipflag)
1076 comp_buf->flags = COMPRESS_BZLIB;
1077 else
1078 comp_buf->flags = COMPRESS_ZLIB;
1079 }
1080 #endif /* HAVE_ZLIB || HAVE_BZLIB */
1081
1082 /*
1083 * Get list of blocks to dump, read the blocks into tape buffer
1084 */
1085 while ((nread = atomic_read( cmd, (char *)slp->req, reqsiz)) == reqsiz) {
1086 register struct req *p = slp->req;
1087
1088 for (trecno = 0; trecno < ntrec;
1089 trecno += p->count, p += p->count) {
1090 if (p->dblk) { /* read a disk block */
1091 bread(p->dblk, slp->tblock[trecno],
1092 p->count * TP_BSIZE);
1093 } else { /* read record from pipe */
1094 if (p->count != 1 || atomic_read( cmd,
1095 (char *)slp->tblock[trecno],
1096 TP_BSIZE) != TP_BSIZE)
1097 quit("master/slave protocol botched.\n");
1098 }
1099 }
1100
1101 /* Try to write the data... */
1102 wrote = 0;
1103 eot_count = 0;
1104 size = 0;
1105 buffer = (char *) slp->tblock[0]; /* set write pointer */
1106 bufsize = writesize; /* length to write */
1107 returns.clen = returns.unclen = bufsize;
1108
1109 #if defined(HAVE_ZLIB) || defined(HAVE_BZLIB)
1110 /*
1111 * When writing a compressed dump, each block except
1112 * the first one on each tape is written
1113 * from struct tapebuf with an 4 byte prefix
1114 * followed by the data. This can be less than
1115 * writesize. Restore, on a short read, can compare the
1116 * length read to the compressed length in the header
1117 * to verify that the read was good. Blocks which don't
1118 * compress well are written uncompressed.
1119 * The first block written by each slave is not compressed
1120 * and does not have a prefix.
1121 */
1122
1123 if (compressed && do_compress) {
1124 comp_buf->length = bufsize;
1125 worklen = TP_BSIZE + writesize;
1126 #ifdef HAVE_ZLIB
1127 if (!bzipflag) {
1128 compresult = compress2(comp_buf->buf,
1129 &worklen,
1130 (char *)slp->tblock[0],
1131 writesize,
1132 compressed);
1133 if (compresult == Z_OK)
1134 compresult = 1;
1135 else
1136 compresult = 0;
1137 }
1138 #endif /* HAVE_ZLIB */
1139 #ifdef HAVE_BZLIB
1140 if (bzipflag) {
1141 worklen2 = worklen;
1142 compresult = BZ2_bzBuffToBuffCompress(
1143 comp_buf->buf,
1144 &worklen2,
1145 (char *)slp->tblock[0],
1146 writesize,
1147 compressed,
1148 0, 30);
1149 worklen = worklen2;
1150 if (compresult == BZ_OK)
1151 compresult = 1;
1152 else
1153 compresult = 0;
1154 }
1155
1156 #endif /* HAVE_BZLIB */
1157 if (compresult && 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 || HAVE_BZLIB */
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 (spclptr->c_date == gThisDumpDate)) {
1193 /* if an error occured previously don't
1194 * try again */
1195 if (gtperr == 0) {
1196 if ((gtperr = GetTapePos(&curtapepos)) == 0) {
1197 #ifdef DEBUG_QFA
1198 msg("inode %ld at tapepos %ld\n", spclptr->c_inumber, curtapepos);
1199 #endif
1200 sprintf(gTps, "%ld\t%d\t%ld\n", (unsigned long)spclptr->c_inumber, tapeno, curtapepos);
1201 if (write(gTapeposfd, gTps, strlen(gTps)) != strlen(gTps)) {
1202 warn("error writing tapepos file.\n");
1203 }
1204 }
1205 }
1206 }
1207 }
1208 #endif /* USE_QFA */
1209
1210 while (eot_count < 10 && size < bufsize) {
1211 #ifdef RDUMP
1212 if (host)
1213 wrote = rmtwrite(buffer + size, bufsize - size);
1214 else
1215 #endif
1216 wrote = write(tapefd, buffer + size, bufsize - size);
1217 #ifdef WRITEDEBUG
1218 printf("slave %d wrote %d\n", slave_number, wrote);
1219 #endif
1220 if (wrote < 0)
1221 break;
1222 if (wrote == 0)
1223 eot_count++;
1224 size += wrote;
1225 }
1226
1227 #ifdef WRITEDEBUG
1228 if (size != bufsize)
1229 printf("slave %d only wrote %d out of %d bytes and gave up.\n",
1230 slave_number, size, bufsize);
1231 #endif
1232
1233 /*
1234 * Handle ENOSPC as an EOT condition.
1235 */
1236 if (wrote < 0 && errno == ENOSPC) {
1237 wrote = 0;
1238 eot_count++;
1239 }
1240
1241 if (eot_count > 0)
1242 returns.clen = returns.unclen = 0;
1243
1244 /*
1245 * pass errno back to master for special handling
1246 */
1247 if (wrote < 0)
1248 returns.unclen = -errno;
1249
1250 /*
1251 * pass size of data and size of write back to master
1252 * (for EOT handling)
1253 */
1254 (void) atomic_write( cmd, (char *)&returns, sizeof returns);
1255
1256 /*
1257 * Signal the next slave to go.
1258 */
1259 (void) kill(nextslave, SIGUSR2);
1260 }
1261 if (nread != 0)
1262 quit("error reading command pipe: %s\n", strerror(errno));
1263 }
1264
1265 /*
1266 * Since a read from a pipe may not return all we asked for,
1267 * or a write may not write all we ask if we get a signal,
1268 * loop until the count is satisfied (or error).
1269 */
1270 static ssize_t
1271 atomic_read(int fd, void *buf, size_t count)
1272 {
1273 int got, need = count;
1274
1275 do {
1276 while ((got = read(fd, buf, need)) > 0 && (need -= got) > 0)
1277 (char *)buf += got;
1278 } while (got == -1 && errno == EINTR);
1279 return (got < 0 ? got : count - need);
1280 }
1281
1282 /*
1283 * Since a read from a pipe may not return all we asked for,
1284 * or a write may not write all we ask if we get a signal,
1285 * loop until the count is satisfied (or error).
1286 */
1287 static ssize_t
1288 atomic_write(int fd, const void *buf, size_t count)
1289 {
1290 int got, need = count;
1291
1292 do {
1293 while ((got = write(fd, buf, need)) > 0 && (need -= got) > 0)
1294 (char *)buf += got;
1295 } while (got == -1 && errno == EINTR);
1296 return (got < 0 ? got : count - need);
1297 }
1298
1299
1300 #ifdef USE_QFA
1301 /*
1302 * read the current tape position
1303 */
1304 int
1305 GetTapePos(long *pos)
1306 {
1307 int err = 0;
1308
1309 *pos = 0;
1310 if (ioctl(tapefd, MTIOCPOS, pos) == -1) {
1311 err = errno;
1312 msg("[%ld] error: %d (getting tapepos: %ld)\n", getpid(),
1313 err, *pos);
1314 return err;
1315 }
1316 return err;
1317 }
1318 #endif /* USE_QFA */