]> git.wh0rd.org - dump.git/blob - dump/tape.c
f048a78bb6f7fc1b4f7787ad5ce9bf872d8fbbec
[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 1.52 2001/07/20 09:01:46 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 #ifdef HAVE_EXT2FS_EXT2_FS_H
76 #include <ext2fs/ext2_fs.h>
77 #else
78 #include <linux/ext2_fs.h>
79 #endif
80 #include <ext2fs/ext2fs.h>
81 #include <bsdcompat.h>
82 #elif defined sunos
83 #include <sys/vnode.h>
84
85 #include <ufs/fs.h>
86 #include <ufs/inode.h>
87 #else
88 #include <ufs/ufs/dinode.h>
89 #include <ufs/ffs/fs.h>
90 #endif /* __linux__ */
91
92 #include <protocols/dumprestore.h>
93
94 #ifdef HAVE_ZLIB
95 #include <zlib.h>
96 #endif /* HAVE_ZLIB */
97
98 #include "dump.h"
99
100 int writesize; /* size of malloc()ed buffer for tape */
101 long lastspclrec = -1; /* tape block number of last written header */
102 int trecno = 0; /* next record to write in current block */
103 extern long blocksperfile; /* number of blocks per output file */
104 long blocksthisvol; /* number of blocks on current output file */
105 extern int ntrec; /* blocking factor on tape */
106 extern int cartridge;
107 extern char *host;
108 char *nexttape;
109 extern pid_t rshpid;
110 int eot_code = 1;
111 long long tapea_bytes = 0; /* bytes_written at start of current volume */
112
113 static ssize_t atomic_read __P((int, void *, size_t));
114 static ssize_t atomic_write __P((int, const void *, size_t));
115 static void doslave __P((int, int, int));
116 static void enslave __P((void));
117 static void flushtape __P((void));
118 static void killall __P((void));
119 static void rollforward __P((void));
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 + writesize >= 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 time_t
466 trewind(void)
467 {
468 int f;
469 int got;
470 struct slave_results returned;
471
472 for (f = 0; f < SLAVES; f++) {
473 /*
474 * Drain the results, but unlike EOT we DO (or should) care
475 * what the return values were, since if we detect EOT after
476 * we think we've written the last blocks to the tape anyway,
477 * we have to replay those blocks with rollforward.
478 *
479 * fixme: punt for now.
480 */
481 if (slaves[f].sent) {
482 if (atomic_read( slaves[f].fd, (char *)&returned, sizeof returned)
483 != sizeof returned) {
484 perror(" DUMP: error reading command pipe in master");
485 dumpabort(0);
486 }
487 got = returned.unclen;
488 bytes_written += returned.clen;
489 if (returned.unclen == returned.clen)
490 uncomprblks++;
491 slaves[f].sent = 0;
492
493 if (got < 0)
494 tperror(-got);
495
496 if (got == 0) {
497 msg("EOT detected in last 2 tape records!\n");
498 msg("Use a longer tape, decrease the size estimate\n");
499 quit("or use no size estimate at all.\n");
500 }
501 }
502 (void) close(slaves[f].fd);
503 }
504 while (wait((int *)NULL) >= 0) /* wait for any signals from slaves */
505 /* void */;
506
507 if (!pipeout) {
508
509 msg("Closing %s\n", tape);
510
511 #ifdef RDUMP
512 if (host) {
513 rmtclose();
514 while (rmtopen(tape, 0) < 0)
515 sleep(10);
516 rmtclose();
517 }
518 else
519 #endif
520 {
521 (void) close(tapefd);
522 if (!fifoout) {
523 while ((f = OPEN(tape, 0)) < 0)
524 sleep (10);
525 (void) close(f);
526 }
527 }
528 eot_code = 1;
529 if (eot_script && spcl.c_type != TS_END) {
530 msg("Launching %s\n", eot_script);
531 eot_code = system_command(eot_script, tape, tapeno);
532 }
533 if (eot_code != 0 && eot_code != 1) {
534 msg("Dump aborted by the end of tape script\n");
535 dumpabort(0);
536 }
537 }
538 return do_stats();
539 }
540
541
542 void
543 close_rewind(void)
544 {
545 (void)trewind();
546 if (nexttape || Mflag || (eot_code == 0) )
547 return;
548 if (!nogripe) {
549 msg("Change Volumes: Mount volume #%d\n", tapeno+1);
550 broadcast("CHANGE DUMP VOLUMES!\7\7\n");
551 }
552 while (!query("Is the new volume mounted and ready to go?"))
553 if (query("Do you want to abort?")) {
554 dumpabort(0);
555 /*NOTREACHED*/
556 }
557 }
558
559 void
560 rollforward(void)
561 {
562 register struct req *p, *q, *prev;
563 register struct slave *tslp;
564 int i, size, savedtapea, got;
565 union u_spcl *ntb, *otb;
566 struct slave_results returned;
567 #ifdef __linux__
568 int blks;
569 long lastfirstrec;
570 #endif
571 tslp = &slaves[SLAVES];
572 ntb = (union u_spcl *)tslp->tblock[1];
573
574 /*
575 * Each of the N slaves should have requests that need to
576 * be replayed on the next tape. Use the extra slave buffers
577 * (slaves[SLAVES]) to construct request lists to be sent to
578 * each slave in turn.
579 */
580 for (i = 0; i < SLAVES; i++) {
581 q = &tslp->req[1];
582 otb = (union u_spcl *)slp->tblock;
583
584 /*
585 * For each request in the current slave, copy it to tslp.
586 */
587
588 prev = NULL;
589 for (p = slp->req; p->count > 0; p += p->count) {
590 *q = *p;
591 if (p->dblk == 0)
592 *ntb++ = *otb++; /* copy the datablock also */
593 prev = q;
594 q += q->count;
595 }
596 if (prev == NULL)
597 quit("rollforward: protocol botch");
598 if (prev->dblk != 0)
599 prev->count -= 1;
600 else
601 ntb--;
602 q -= 1;
603 q->count = 0;
604 q = &tslp->req[0];
605 if (i == 0) {
606 q->dblk = 0;
607 q->count = 1;
608 trecno = 0;
609 nextblock = tslp->tblock;
610 savedtapea = spcl.c_tapea;
611 spcl.c_tapea = slp->tapea;
612 startnewtape(0);
613 spcl.c_tapea = savedtapea;
614 lastspclrec = savedtapea - 1;
615 }
616 size = (char *)ntb - (char *)q;
617 if (atomic_write( slp->fd, (char *)q, size) != size) {
618 perror(" DUMP: error writing command pipe");
619 dumpabort(0);
620 }
621 slp->sent = 1;
622 #ifdef __linux__
623 lastfirstrec = slp->firstrec;
624 #endif
625 if (++slp >= &slaves[SLAVES])
626 slp = &slaves[0];
627
628 q->count = 1;
629
630 if (prev->dblk != 0) {
631 /*
632 * If the last one was a disk block, make the
633 * first of this one be the last bit of that disk
634 * block...
635 */
636 q->dblk = prev->dblk +
637 prev->count * (TP_BSIZE / DEV_BSIZE);
638 ntb = (union u_spcl *)tslp->tblock;
639 } else {
640 /*
641 * It wasn't a disk block. Copy the data to its
642 * new location in the buffer.
643 */
644 q->dblk = 0;
645 *((union u_spcl *)tslp->tblock) = *ntb;
646 ntb = (union u_spcl *)tslp->tblock[1];
647 }
648 }
649 slp->req[0] = *q;
650 nextblock = slp->tblock;
651 if (q->dblk == 0) {
652 #ifdef __linux__
653 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
654 *(union u_spcl *)(*nextblock) = *(union u_spcl *)tslp->tblock;
655 #endif
656 nextblock++;
657 }
658 trecno = 1;
659
660 /*
661 * Clear the first slaves' response. One hopes that it
662 * worked ok, otherwise the tape is much too short!
663 */
664 if (slp->sent) {
665 if (atomic_read( slp->fd, (char *)&returned, sizeof returned)
666 != sizeof returned) {
667 perror(" DUMP: error reading command pipe in master");
668 dumpabort(0);
669 }
670 got = returned.unclen;
671 bytes_written += returned.clen;
672 if (returned.clen == returned.unclen)
673 uncomprblks++;
674 slp->sent = 0;
675
676 if (got < 0)
677 tperror(-got);
678
679 if (got == 0) {
680 quit("EOT detected at start of the tape!\n");
681 }
682 }
683
684 #ifdef __linux__
685 blks = 0;
686 if (spcl.c_type != TS_END) {
687 for (i = 0; i < spcl.c_count; i++)
688 if (spcl.c_addr[i] != 0)
689 blks++;
690 }
691
692 slp->firstrec = lastfirstrec + ntrec;
693 slp->count = lastspclrec + blks + 1 - spcl.c_tapea;
694 slp->inode = curino;
695 asize += tenths + returned.clen / density;
696 csize += returned.clen;
697 blockswritten += ntrec;
698 blocksthisvol += ntrec;
699 #endif
700 }
701
702 /*
703 * We implement taking and restoring checkpoints on the tape level.
704 * When each tape is opened, a new process is created by forking; this
705 * saves all of the necessary context in the parent. The child
706 * continues the dump; the parent waits around, saving the context.
707 * If the child returns X_REWRITE, then it had problems writing that tape;
708 * this causes the parent to fork again, duplicating the context, and
709 * everything continues as if nothing had happened.
710 */
711 void
712 startnewtape(int top)
713 {
714 int parentpid;
715 int childpid;
716 int status;
717 int waitpid;
718 char *p;
719
720 #ifdef __linux__
721 sigset_t sigs;
722 sigemptyset(&sigs);
723 sigaddset(&sigs, SIGINT);
724 sigprocmask(SIG_BLOCK, &sigs, NULL);
725 #else /* __linux__ */
726 #ifdef sunos
727 void (*interrupt_save)();
728 #else
729 sig_t interrupt_save;
730 #endif
731 interrupt_save = signal(SIGINT, SIG_IGN);
732 #endif /* __linux__ */
733
734 parentpid = getpid();
735 tapea_volume = spcl.c_tapea;
736 tapea_bytes = bytes_written;
737 tstart_volume = time(NULL);
738
739 restore_check_point:
740 #ifdef __linux__
741 sigprocmask(SIG_UNBLOCK, &sigs, NULL);
742 #else
743 (void)signal(SIGINT, interrupt_save);
744 #endif
745 /*
746 * All signals are inherited...
747 */
748 childpid = fork();
749 if (childpid < 0) {
750 msg("Context save fork fails in parent %d\n", parentpid);
751 Exit(X_ABORT);
752 }
753 if (childpid != 0) {
754 /*
755 * PARENT:
756 * save the context by waiting
757 * until the child doing all of the work returns.
758 * don't catch the interrupt
759 */
760 #ifdef __linux__
761 sigprocmask(SIG_BLOCK, &sigs, NULL);
762 #else
763 signal(SIGINT, SIG_IGN);
764 #endif
765 #ifdef TDEBUG
766 msg("Tape: %d; parent process: %d child process %d\n",
767 tapeno+1, parentpid, childpid);
768 #endif /* TDEBUG */
769 while ((waitpid = wait(&status)) != childpid)
770 if (waitpid != rshpid)
771 msg("Parent %d waiting for child %d has another child %d return\n",
772 parentpid, childpid, waitpid);
773 if (status & 0xFF) {
774 msg("Child %d returns LOB status %o\n",
775 childpid, status&0xFF);
776 }
777 status = (status >> 8) & 0xFF;
778 #ifdef TDEBUG
779 switch(status) {
780 case X_FINOK:
781 msg("Child %d finishes X_FINOK\n", childpid);
782 break;
783 case X_ABORT:
784 msg("Child %d finishes X_ABORT\n", childpid);
785 break;
786 case X_REWRITE:
787 msg("Child %d finishes X_REWRITE\n", childpid);
788 break;
789 default:
790 msg("Child %d finishes unknown %d\n",
791 childpid, status);
792 break;
793 }
794 #endif /* TDEBUG */
795 switch(status) {
796 case X_FINOK:
797 Exit(X_FINOK);
798 case X_ABORT:
799 Exit(X_ABORT);
800 case X_REWRITE:
801 goto restore_check_point;
802 default:
803 msg("Bad return code from dump: %d\n", status);
804 Exit(X_ABORT);
805 }
806 /*NOTREACHED*/
807 } else { /* we are the child; just continue */
808 #ifdef TDEBUG
809 sleep(4); /* allow time for parent's message to get out */
810 msg("Child on Tape %d has parent %d, my pid = %d\n",
811 tapeno+1, parentpid, getpid());
812 #endif /* TDEBUG */
813 /*
814 * If we have a name like "/dev/rmt0,/dev/rmt1",
815 * use the name before the comma first, and save
816 * the remaining names for subsequent volumes.
817 */
818 tapeno++; /* current tape sequence */
819 if (Mflag) {
820 snprintf(tape, MAXPATHLEN, "%s%03d", tapeprefix, tapeno);
821 tape[MAXPATHLEN - 1] = '\0';
822 msg("Dumping volume %d on %s\n", tapeno, tape);
823 }
824 else if (nexttape || strchr(tapeprefix, ',')) {
825 if (nexttape && *nexttape)
826 tapeprefix = nexttape;
827 if ((p = strchr(tapeprefix, ',')) != NULL) {
828 *p = '\0';
829 nexttape = p + 1;
830 } else
831 nexttape = NULL;
832 strncpy(tape, tapeprefix, MAXPATHLEN);
833 tape[MAXPATHLEN - 1] = '\0';
834 msg("Dumping volume %d on %s\n", tapeno, tape);
835 }
836 #ifdef RDUMP
837 while ((tapefd = (host ? rmtopen(tape, 2) : pipeout ?
838 fileno(stdout) :
839 OPEN(tape, O_WRONLY|O_CREAT, 0666))) < 0)
840 #else
841 while ((tapefd = (pipeout ? fileno(stdout) :
842 OPEN(tape, O_RDWR|O_CREAT, 0666))) < 0)
843 #endif
844 {
845 msg("Cannot open output \"%s\".\n", tape);
846 if (!query("Do you want to retry the open?"))
847 dumpabort(0);
848 }
849
850 enslave(); /* Share open tape file descriptor with slaves */
851
852 asize = 0;
853 csize = 0;
854 blocksthisvol = 0;
855 if (top)
856 newtape++; /* new tape signal */
857 spcl.c_count = slp->count;
858 /*
859 * measure firstrec in TP_BSIZE units since restore doesn't
860 * know the correct ntrec value...
861 */
862 spcl.c_firstrec = slp->firstrec;
863 spcl.c_volume++;
864 spcl.c_type = TS_TAPE;
865 spcl.c_flags |= DR_NEWHEADER;
866 spcl.c_ntrec = ntrec;
867 if (compressed)
868 spcl.c_flags |= DR_COMPRESSED;
869 writeheader((dump_ino_t)slp->inode);
870 spcl.c_flags &=~ DR_NEWHEADER;
871 msg("Volume %d started with block %ld at: %s", tapeno,
872 spcl.c_tapea, ctime(&tstart_volume));
873 if (tapeno > 1)
874 msg("Volume %d begins with blocks from inode %d\n",
875 tapeno, slp->inode);
876 }
877 }
878
879 void
880 dumpabort(int signo)
881 {
882
883 if (master != 0 && master != getpid())
884 /* Signals master to call dumpabort */
885 (void) kill(master, SIGTERM);
886 else {
887 killall();
888 msg("The ENTIRE dump is aborted.\n");
889 }
890 #ifdef RDUMP
891 rmtclose();
892 #endif
893 Exit(X_ABORT);
894 }
895
896 void
897 Exit(int status)
898 {
899
900 #ifdef TDEBUG
901 msg("pid = %d exits with status %d\n", getpid(), status);
902 #endif /* TDEBUG */
903 exit(status);
904 }
905
906 /*
907 * proceed - handler for SIGUSR2, used to synchronize IO between the slaves.
908 */
909 static void
910 proceed(int signo)
911 {
912 if (ready)
913 siglongjmp(jmpbuf, 1);
914 caught++;
915 }
916
917 void
918 enslave(void)
919 {
920 int cmd[2];
921 #ifdef LINUX_FORK_BUG
922 int i, j;
923 #else
924 register int i, j;
925 #endif
926
927 master = getpid();
928
929 { struct sigaction sa;
930 memset(&sa, 0, sizeof sa);
931 sigemptyset(&sa.sa_mask);
932 sa.sa_handler = dumpabort;
933 sigaction(SIGTERM, &sa, NULL); /* Slave sends SIGTERM on dumpabort() */
934 sa.sa_handler = sigpipe;
935 sigaction(SIGPIPE, &sa, NULL);
936 sa.sa_handler = proceed;
937 sa.sa_flags = SA_RESTART;
938 sigaction(SIGUSR2, &sa, NULL); /* Slave sends SIGUSR2 to next slave */
939 }
940
941 for (i = 0; i < SLAVES; i++) {
942 if (i == slp - &slaves[0]) {
943 caught = 1;
944 } else {
945 caught = 0;
946 }
947
948 if (socketpair(AF_UNIX, SOCK_STREAM, 0, cmd) < 0 ||
949 (slaves[i].pid = fork()) < 0)
950 quit("too many slaves, %d (recompile smaller): %s\n",
951 i, strerror(errno));
952
953 slaves[i].fd = cmd[1];
954 slaves[i].sent = 0;
955 if (slaves[i].pid == 0) { /* Slave starts up here */
956 sigset_t sigs;
957 for (j = 0; j <= i; j++)
958 (void) close(slaves[j].fd);
959 sigemptyset(&sigs);
960 sigaddset(&sigs, SIGINT); /* Master handles this */
961 #if defined(SIGINFO)
962 sigaddset(&sigs, SIGINFO);
963 #endif
964 sigprocmask(SIG_BLOCK, &sigs, NULL);
965
966 #ifdef LINUX_FORK_BUG
967 if (atomic_write( cmd[0], (char *) &i, sizeof i)
968 != sizeof i)
969 quit("master/slave protocol botched 3\n");
970 #endif
971 doslave(cmd[0], i, (slaves[i].pid == slp->pid));
972 Exit(X_FINOK);
973 }
974 else
975 close(cmd[0]);
976 }
977
978 #ifdef LINUX_FORK_BUG
979 /*
980 * Wait for all slaves to _actually_ start to circumvent a bug in
981 * Linux kernels >= 2.1.3 where a signal sent to a child that hasn't
982 * returned from fork() causes a SEGV in the child process
983 */
984 for (i = 0; i < SLAVES; i++)
985 if (atomic_read( slaves[i].fd, (char *) &j, sizeof j) != sizeof j)
986 quit("master/slave protocol botched 4\n");
987 #endif
988
989 for (i = 0; i < SLAVES; i++)
990 (void) atomic_write( slaves[i].fd,
991 (char *) &slaves[(i + 1) % SLAVES].pid,
992 sizeof slaves[0].pid);
993
994 master = 0;
995 }
996
997 void
998 killall(void)
999 {
1000 register int i;
1001
1002 for (i = 0; i < SLAVES; i++)
1003 if (slaves[i].pid > 0) {
1004 (void) kill(slaves[i].pid, SIGKILL);
1005 slaves[i].sent = 0;
1006 }
1007 }
1008
1009 /*
1010 * Synchronization - each process waits for a SIGUSR2 from the
1011 * previous process before writing to the tape, and sends SIGUSR2
1012 * to the next process when the tape write completes. On tape errors
1013 * a SIGUSR1 is sent to the master which then terminates all of the
1014 * slaves.
1015 */
1016 static void
1017 doslave(int cmd, int slave_number, int first)
1018 {
1019 register int nread;
1020 int nextslave, size, eot_count, bufsize;
1021 volatile int wrote = 0;
1022 char *buffer;
1023 #ifdef HAVE_ZLIB
1024 struct tapebuf *comp_buf = NULL;
1025 int compresult, do_compress = !first;
1026 unsigned long worklen;
1027 #endif /* HAVE_ZLIB */
1028 struct slave_results returns;
1029 #ifdef __linux__
1030 errcode_t retval;
1031 #endif
1032 #ifdef USE_QFA
1033 long curtapepos;
1034 union u_spcl *uspclptr;
1035 struct s_spcl *spclptr;
1036 #endif /* USE_QFA */
1037
1038 /*
1039 * Need our own seek pointer.
1040 */
1041 (void) close(diskfd);
1042 if ((diskfd = OPEN(disk, O_RDONLY)) < 0)
1043 quit("slave couldn't reopen disk: %s\n", strerror(errno));
1044 #ifdef __linux__
1045 ext2fs_close(fs);
1046 retval = dump_fs_open(disk, &fs);
1047 if (retval)
1048 quit("slave couldn't reopen disk: %s\n", error_message(retval));
1049 #endif /* __linux__ */
1050
1051 /*
1052 * Need the pid of the next slave in the loop...
1053 */
1054 if ((nread = atomic_read( cmd, (char *)&nextslave, sizeof nextslave))
1055 != sizeof nextslave) {
1056 quit("master/slave protocol botched - didn't get pid of next slave.\n");
1057 }
1058
1059 #ifdef HAVE_ZLIB
1060 /* if we're doing a compressed dump, allocate the compress buffer */
1061 if (compressed) {
1062 comp_buf = malloc(sizeof(struct tapebuf) + TP_BSIZE + writesize);
1063 if (comp_buf == NULL)
1064 quit("couldn't allocate a compress buffer.\n");
1065 comp_buf->flags = 0;
1066 }
1067 #endif /* HAVE_ZLIB */
1068
1069 /*
1070 * Get list of blocks to dump, read the blocks into tape buffer
1071 */
1072 while ((nread = atomic_read( cmd, (char *)slp->req, reqsiz)) == reqsiz) {
1073 register struct req *p = slp->req;
1074
1075 for (trecno = 0; trecno < ntrec;
1076 trecno += p->count, p += p->count) {
1077 if (p->dblk) { /* read a disk block */
1078 bread(p->dblk, slp->tblock[trecno],
1079 p->count * TP_BSIZE);
1080 } else { /* read record from pipe */
1081 if (p->count != 1 || atomic_read( cmd,
1082 (char *)slp->tblock[trecno],
1083 TP_BSIZE) != TP_BSIZE)
1084 quit("master/slave protocol botched.\n");
1085 }
1086 }
1087
1088 /* Try to write the data... */
1089 wrote = 0;
1090 eot_count = 0;
1091 size = 0;
1092 buffer = (char *) slp->tblock[0]; /* set write pointer */
1093 bufsize = writesize; /* length to write */
1094 returns.clen = returns.unclen = bufsize;
1095
1096 #ifdef HAVE_ZLIB
1097 /*
1098 * When writing a compressed dump, each block except
1099 * the first one on each tape is written
1100 * from struct tapebuf with an 4 byte prefix
1101 * followed by the data. This can be less than
1102 * writesize. Restore, on a short read, can compare the
1103 * length read to the compressed length in the header
1104 * to verify that the read was good. Blocks which don't
1105 * compress well are written uncompressed.
1106 * The first block written by each slave is not compressed
1107 * and does not have a prefix.
1108 */
1109
1110 if (compressed && do_compress) {
1111 comp_buf->length = bufsize;
1112 worklen = TP_BSIZE + writesize;
1113 compresult = compress2(comp_buf->buf, &worklen,
1114 (char *)slp->tblock[0], writesize, compressed);
1115 if (compresult == Z_OK && worklen <= (writesize - 16)) {
1116 /* write the compressed buffer */
1117 comp_buf->length = worklen;
1118 comp_buf->compressed = 1;
1119 buffer = (char *) comp_buf;
1120 returns.clen = bufsize = worklen + sizeof(struct tapebuf);
1121 }
1122 else {
1123 /* write the data uncompressed */
1124 comp_buf->length = writesize;
1125 comp_buf->compressed = 0;
1126 buffer = (char *) comp_buf;
1127 returns.clen = bufsize = writesize + sizeof(struct tapebuf);
1128 returns.unclen = returns.clen;
1129 memcpy(comp_buf->buf, (char *)slp->tblock[0], writesize);
1130 }
1131 }
1132 /* compress the remaining blocks if we're compressing */
1133 do_compress = compressed;
1134 #endif /* HAVE_ZLIB */
1135
1136 if (sigsetjmp(jmpbuf, 1) == 0) {
1137 ready = 1;
1138 if (!caught)
1139 (void) pause();
1140 }
1141 ready = 0;
1142 caught = 0;
1143
1144 #ifdef USE_QFA
1145 if (gTapeposfd >= 0) {
1146 uspclptr = (union u_spcl *)&slp->tblock[0];
1147 spclptr = &uspclptr->s_spcl;
1148 if ((spclptr->c_magic == NFS_MAGIC) &&
1149 (spclptr->c_type == TS_INODE) &&
1150 (spclptr->c_date == gThisDumpDate)) {
1151 /* if an error occured previously don't
1152 * try again */
1153 if (gtperr == 0) {
1154 if ((gtperr = GetTapePos(&curtapepos)) == 0) {
1155 #ifdef DEBUG_QFA
1156 msg("inode %ld at tapepos %ld\n", spclptr->c_inumber, curtapepos);
1157 #endif
1158 sprintf(gTps, "%ld\t%d\t%ld\n", (unsigned long)spclptr->c_inumber, tapeno, curtapepos);
1159 if (write(gTapeposfd, gTps, strlen(gTps)) != strlen(gTps)) {
1160 warn("error writing tapepos file.\n");
1161 }
1162 }
1163 }
1164 }
1165 }
1166 #endif /* USE_QFA */
1167
1168 while (eot_count < 10 && size < bufsize) {
1169 #ifdef RDUMP
1170 if (host)
1171 wrote = rmtwrite(buffer + size, bufsize - size);
1172 else
1173 #endif
1174 wrote = write(tapefd, buffer + size, bufsize - size);
1175 #ifdef WRITEDEBUG
1176 printf("slave %d wrote %d\n", slave_number, wrote);
1177 #endif
1178 if (wrote < 0)
1179 break;
1180 if (wrote == 0)
1181 eot_count++;
1182 size += wrote;
1183 }
1184
1185 #ifdef WRITEDEBUG
1186 if (size != bufsize)
1187 printf("slave %d only wrote %d out of %d bytes and gave up.\n",
1188 slave_number, size, bufsize);
1189 #endif
1190
1191 /*
1192 * Handle ENOSPC as an EOT condition.
1193 */
1194 if (wrote < 0 && errno == ENOSPC) {
1195 wrote = 0;
1196 eot_count++;
1197 }
1198
1199 if (eot_count > 0)
1200 returns.clen = returns.unclen = 0;
1201
1202 /*
1203 * pass errno back to master for special handling
1204 */
1205 if (wrote < 0)
1206 returns.unclen = -errno;
1207
1208 /*
1209 * pass size of data and size of write back to master
1210 * (for EOT handling)
1211 */
1212 (void) atomic_write( cmd, (char *)&returns, sizeof returns);
1213
1214 /*
1215 * Signal the next slave to go.
1216 */
1217 (void) kill(nextslave, SIGUSR2);
1218 }
1219 if (nread != 0)
1220 quit("error reading command pipe: %s\n", strerror(errno));
1221 }
1222
1223 /*
1224 * Since a read from a pipe may not return all we asked for,
1225 * or a write may not write all we ask if we get a signal,
1226 * loop until the count is satisfied (or error).
1227 */
1228 static ssize_t
1229 atomic_read(int fd, void *buf, size_t count)
1230 {
1231 int got, need = count;
1232
1233 do {
1234 while ((got = read(fd, buf, need)) > 0 && (need -= got) > 0)
1235 (char *)buf += got;
1236 } while (got == -1 && errno == EINTR);
1237 return (got < 0 ? got : count - need);
1238 }
1239
1240 /*
1241 * Since a read from a pipe may not return all we asked for,
1242 * or a write may not write all we ask if we get a signal,
1243 * loop until the count is satisfied (or error).
1244 */
1245 static ssize_t
1246 atomic_write(int fd, const void *buf, size_t count)
1247 {
1248 int got, need = count;
1249
1250 do {
1251 while ((got = write(fd, buf, need)) > 0 && (need -= got) > 0)
1252 (char *)buf += got;
1253 } while (got == -1 && errno == EINTR);
1254 return (got < 0 ? got : count - need);
1255 }
1256
1257
1258 #ifdef USE_QFA
1259 /*
1260 * read the current tape position
1261 */
1262 int
1263 GetTapePos(long *pos)
1264 {
1265 int err = 0;
1266
1267 *pos = 0;
1268 if (ioctl(tapefd, MTIOCPOS, pos) == -1) {
1269 err = errno;
1270 msg("[%ld] error: %d (getting tapepos: %ld)\n", getpid(),
1271 err, *pos);
1272 return err;
1273 }
1274 return err;
1275 }
1276 #endif /* USE_QFA */