]> git.wh0rd.org - dump.git/blob - dump/tape.c
Updated and cleaned the -B compressed limit.
[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 1.53 2001/07/20 11:02:45 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 blockswritten += ntrec;
447 blocksthisvol += ntrec;
448 if (!pipeout && !unlimited) {
449 if (blocksperfile) {
450 if ( compressed ? (bytes_written - tapea_bytes + SLAVES * (writesize + sizeof(struct tapebuf))) >= blocksperfile * 1024
451 : blocksthisvol >= blocksperfile ) {
452 close_rewind();
453 startnewtape(0);
454 }
455 }
456 else if (asize > tsize) {
457 close_rewind();
458 startnewtape(0);
459 }
460 }
461 timeest();
462 }
463
464 time_t
465 trewind(void)
466 {
467 int f;
468 int got;
469 struct slave_results returned;
470
471 for (f = 0; f < SLAVES; f++) {
472 /*
473 * Drain the results, but unlike EOT we DO (or should) care
474 * what the return values were, since if we detect EOT after
475 * we think we've written the last blocks to the tape anyway,
476 * we have to replay those blocks with rollforward.
477 *
478 * fixme: punt for now.
479 */
480 if (slaves[f].sent) {
481 if (atomic_read( slaves[f].fd, (char *)&returned, sizeof returned)
482 != sizeof returned) {
483 perror(" DUMP: error reading command pipe in master");
484 dumpabort(0);
485 }
486 got = returned.unclen;
487 bytes_written += returned.clen;
488 if (returned.unclen == returned.clen)
489 uncomprblks++;
490 slaves[f].sent = 0;
491
492 if (got < 0)
493 tperror(-got);
494
495 if (got == 0) {
496 msg("EOT detected in last 2 tape records!\n");
497 msg("Use a longer tape, decrease the size estimate\n");
498 quit("or use no size estimate at all.\n");
499 }
500 }
501 (void) close(slaves[f].fd);
502 }
503 while (wait((int *)NULL) >= 0) /* wait for any signals from slaves */
504 /* void */;
505
506 if (!pipeout) {
507
508 msg("Closing %s\n", tape);
509
510 #ifdef RDUMP
511 if (host) {
512 rmtclose();
513 while (rmtopen(tape, 0) < 0)
514 sleep(10);
515 rmtclose();
516 }
517 else
518 #endif
519 {
520 (void) close(tapefd);
521 if (!fifoout) {
522 while ((f = OPEN(tape, 0)) < 0)
523 sleep (10);
524 (void) close(f);
525 }
526 }
527 eot_code = 1;
528 if (eot_script && spcl.c_type != TS_END) {
529 msg("Launching %s\n", eot_script);
530 eot_code = system_command(eot_script, tape, tapeno);
531 }
532 if (eot_code != 0 && eot_code != 1) {
533 msg("Dump aborted by the end of tape script\n");
534 dumpabort(0);
535 }
536 }
537 return do_stats();
538 }
539
540
541 void
542 close_rewind(void)
543 {
544 (void)trewind();
545 if (nexttape || Mflag || (eot_code == 0) )
546 return;
547 if (!nogripe) {
548 msg("Change Volumes: Mount volume #%d\n", tapeno+1);
549 broadcast("CHANGE DUMP VOLUMES!\7\7\n");
550 }
551 while (!query("Is the new volume mounted and ready to go?"))
552 if (query("Do you want to abort?")) {
553 dumpabort(0);
554 /*NOTREACHED*/
555 }
556 }
557
558 void
559 rollforward(void)
560 {
561 register struct req *p, *q, *prev;
562 register struct slave *tslp;
563 int i, size, savedtapea, got;
564 union u_spcl *ntb, *otb;
565 struct slave_results returned;
566 #ifdef __linux__
567 int blks;
568 long lastfirstrec;
569 #endif
570 tslp = &slaves[SLAVES];
571 ntb = (union u_spcl *)tslp->tblock[1];
572
573 /*
574 * Each of the N slaves should have requests that need to
575 * be replayed on the next tape. Use the extra slave buffers
576 * (slaves[SLAVES]) to construct request lists to be sent to
577 * each slave in turn.
578 */
579 for (i = 0; i < SLAVES; i++) {
580 q = &tslp->req[1];
581 otb = (union u_spcl *)slp->tblock;
582
583 /*
584 * For each request in the current slave, copy it to tslp.
585 */
586
587 prev = NULL;
588 for (p = slp->req; p->count > 0; p += p->count) {
589 *q = *p;
590 if (p->dblk == 0)
591 *ntb++ = *otb++; /* copy the datablock also */
592 prev = q;
593 q += q->count;
594 }
595 if (prev == NULL)
596 quit("rollforward: protocol botch");
597 if (prev->dblk != 0)
598 prev->count -= 1;
599 else
600 ntb--;
601 q -= 1;
602 q->count = 0;
603 q = &tslp->req[0];
604 if (i == 0) {
605 q->dblk = 0;
606 q->count = 1;
607 trecno = 0;
608 nextblock = tslp->tblock;
609 savedtapea = spcl.c_tapea;
610 spcl.c_tapea = slp->tapea;
611 startnewtape(0);
612 spcl.c_tapea = savedtapea;
613 lastspclrec = savedtapea - 1;
614 }
615 size = (char *)ntb - (char *)q;
616 if (atomic_write( slp->fd, (char *)q, size) != size) {
617 perror(" DUMP: error writing command pipe");
618 dumpabort(0);
619 }
620 slp->sent = 1;
621 #ifdef __linux__
622 lastfirstrec = slp->firstrec;
623 #endif
624 if (++slp >= &slaves[SLAVES])
625 slp = &slaves[0];
626
627 q->count = 1;
628
629 if (prev->dblk != 0) {
630 /*
631 * If the last one was a disk block, make the
632 * first of this one be the last bit of that disk
633 * block...
634 */
635 q->dblk = prev->dblk +
636 prev->count * (TP_BSIZE / DEV_BSIZE);
637 ntb = (union u_spcl *)tslp->tblock;
638 } else {
639 /*
640 * It wasn't a disk block. Copy the data to its
641 * new location in the buffer.
642 */
643 q->dblk = 0;
644 *((union u_spcl *)tslp->tblock) = *ntb;
645 ntb = (union u_spcl *)tslp->tblock[1];
646 }
647 }
648 slp->req[0] = *q;
649 nextblock = slp->tblock;
650 if (q->dblk == 0) {
651 #ifdef __linux__
652 /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
653 *(union u_spcl *)(*nextblock) = *(union u_spcl *)tslp->tblock;
654 #endif
655 nextblock++;
656 }
657 trecno = 1;
658
659 /*
660 * Clear the first slaves' response. One hopes that it
661 * worked ok, otherwise the tape is much too short!
662 */
663 if (slp->sent) {
664 if (atomic_read( slp->fd, (char *)&returned, sizeof returned)
665 != sizeof returned) {
666 perror(" DUMP: error reading command pipe in master");
667 dumpabort(0);
668 }
669 got = returned.unclen;
670 bytes_written += returned.clen;
671 if (returned.clen == returned.unclen)
672 uncomprblks++;
673 slp->sent = 0;
674
675 if (got < 0)
676 tperror(-got);
677
678 if (got == 0) {
679 quit("EOT detected at start of the tape!\n");
680 }
681 }
682
683 #ifdef __linux__
684 blks = 0;
685 if (spcl.c_type != TS_END) {
686 for (i = 0; i < spcl.c_count; i++)
687 if (spcl.c_addr[i] != 0)
688 blks++;
689 }
690
691 slp->firstrec = lastfirstrec + ntrec;
692 slp->count = lastspclrec + blks + 1 - spcl.c_tapea;
693 slp->inode = curino;
694 asize += tenths + returned.clen / density;
695 blockswritten += ntrec;
696 blocksthisvol += ntrec;
697 #endif
698 }
699
700 /*
701 * We implement taking and restoring checkpoints on the tape level.
702 * When each tape is opened, a new process is created by forking; this
703 * saves all of the necessary context in the parent. The child
704 * continues the dump; the parent waits around, saving the context.
705 * If the child returns X_REWRITE, then it had problems writing that tape;
706 * this causes the parent to fork again, duplicating the context, and
707 * everything continues as if nothing had happened.
708 */
709 void
710 startnewtape(int top)
711 {
712 int parentpid;
713 int childpid;
714 int status;
715 int waitpid;
716 char *p;
717
718 #ifdef __linux__
719 sigset_t sigs;
720 sigemptyset(&sigs);
721 sigaddset(&sigs, SIGINT);
722 sigprocmask(SIG_BLOCK, &sigs, NULL);
723 #else /* __linux__ */
724 #ifdef sunos
725 void (*interrupt_save)();
726 #else
727 sig_t interrupt_save;
728 #endif
729 interrupt_save = signal(SIGINT, SIG_IGN);
730 #endif /* __linux__ */
731
732 parentpid = getpid();
733 tapea_volume = spcl.c_tapea;
734 tapea_bytes = bytes_written;
735 tstart_volume = time(NULL);
736
737 restore_check_point:
738 #ifdef __linux__
739 sigprocmask(SIG_UNBLOCK, &sigs, NULL);
740 #else
741 (void)signal(SIGINT, interrupt_save);
742 #endif
743 /*
744 * All signals are inherited...
745 */
746 childpid = fork();
747 if (childpid < 0) {
748 msg("Context save fork fails in parent %d\n", parentpid);
749 Exit(X_ABORT);
750 }
751 if (childpid != 0) {
752 /*
753 * PARENT:
754 * save the context by waiting
755 * until the child doing all of the work returns.
756 * don't catch the interrupt
757 */
758 #ifdef __linux__
759 sigprocmask(SIG_BLOCK, &sigs, NULL);
760 #else
761 signal(SIGINT, SIG_IGN);
762 #endif
763 #ifdef TDEBUG
764 msg("Tape: %d; parent process: %d child process %d\n",
765 tapeno+1, parentpid, childpid);
766 #endif /* TDEBUG */
767 while ((waitpid = wait(&status)) != childpid)
768 if (waitpid != rshpid)
769 msg("Parent %d waiting for child %d has another child %d return\n",
770 parentpid, childpid, waitpid);
771 if (status & 0xFF) {
772 msg("Child %d returns LOB status %o\n",
773 childpid, status&0xFF);
774 }
775 status = (status >> 8) & 0xFF;
776 #ifdef TDEBUG
777 switch(status) {
778 case X_FINOK:
779 msg("Child %d finishes X_FINOK\n", childpid);
780 break;
781 case X_ABORT:
782 msg("Child %d finishes X_ABORT\n", childpid);
783 break;
784 case X_REWRITE:
785 msg("Child %d finishes X_REWRITE\n", childpid);
786 break;
787 default:
788 msg("Child %d finishes unknown %d\n",
789 childpid, status);
790 break;
791 }
792 #endif /* TDEBUG */
793 switch(status) {
794 case X_FINOK:
795 Exit(X_FINOK);
796 case X_ABORT:
797 Exit(X_ABORT);
798 case X_REWRITE:
799 goto restore_check_point;
800 default:
801 msg("Bad return code from dump: %d\n", status);
802 Exit(X_ABORT);
803 }
804 /*NOTREACHED*/
805 } else { /* we are the child; just continue */
806 #ifdef TDEBUG
807 sleep(4); /* allow time for parent's message to get out */
808 msg("Child on Tape %d has parent %d, my pid = %d\n",
809 tapeno+1, parentpid, getpid());
810 #endif /* TDEBUG */
811 /*
812 * If we have a name like "/dev/rmt0,/dev/rmt1",
813 * use the name before the comma first, and save
814 * the remaining names for subsequent volumes.
815 */
816 tapeno++; /* current tape sequence */
817 if (Mflag) {
818 snprintf(tape, MAXPATHLEN, "%s%03d", tapeprefix, tapeno);
819 tape[MAXPATHLEN - 1] = '\0';
820 msg("Dumping volume %d on %s\n", tapeno, tape);
821 }
822 else if (nexttape || strchr(tapeprefix, ',')) {
823 if (nexttape && *nexttape)
824 tapeprefix = nexttape;
825 if ((p = strchr(tapeprefix, ',')) != NULL) {
826 *p = '\0';
827 nexttape = p + 1;
828 } else
829 nexttape = NULL;
830 strncpy(tape, tapeprefix, MAXPATHLEN);
831 tape[MAXPATHLEN - 1] = '\0';
832 msg("Dumping volume %d on %s\n", tapeno, tape);
833 }
834 #ifdef RDUMP
835 while ((tapefd = (host ? rmtopen(tape, 2) : pipeout ?
836 fileno(stdout) :
837 OPEN(tape, O_WRONLY|O_CREAT, 0666))) < 0)
838 #else
839 while ((tapefd = (pipeout ? fileno(stdout) :
840 OPEN(tape, O_RDWR|O_CREAT, 0666))) < 0)
841 #endif
842 {
843 msg("Cannot open output \"%s\".\n", tape);
844 if (!query("Do you want to retry the open?"))
845 dumpabort(0);
846 }
847
848 enslave(); /* Share open tape file descriptor with slaves */
849
850 asize = 0;
851 blocksthisvol = 0;
852 if (top)
853 newtape++; /* new tape signal */
854 spcl.c_count = slp->count;
855 /*
856 * measure firstrec in TP_BSIZE units since restore doesn't
857 * know the correct ntrec value...
858 */
859 spcl.c_firstrec = slp->firstrec;
860 spcl.c_volume++;
861 spcl.c_type = TS_TAPE;
862 spcl.c_flags |= DR_NEWHEADER;
863 spcl.c_ntrec = ntrec;
864 if (compressed)
865 spcl.c_flags |= DR_COMPRESSED;
866 writeheader((dump_ino_t)slp->inode);
867 spcl.c_flags &=~ DR_NEWHEADER;
868 msg("Volume %d started with block %ld at: %s", tapeno,
869 spcl.c_tapea, ctime(&tstart_volume));
870 if (tapeno > 1)
871 msg("Volume %d begins with blocks from inode %d\n",
872 tapeno, slp->inode);
873 }
874 }
875
876 void
877 dumpabort(int signo)
878 {
879
880 if (master != 0 && master != getpid())
881 /* Signals master to call dumpabort */
882 (void) kill(master, SIGTERM);
883 else {
884 killall();
885 msg("The ENTIRE dump is aborted.\n");
886 }
887 #ifdef RDUMP
888 rmtclose();
889 #endif
890 Exit(X_ABORT);
891 }
892
893 void
894 Exit(int status)
895 {
896
897 #ifdef TDEBUG
898 msg("pid = %d exits with status %d\n", getpid(), status);
899 #endif /* TDEBUG */
900 exit(status);
901 }
902
903 /*
904 * proceed - handler for SIGUSR2, used to synchronize IO between the slaves.
905 */
906 static void
907 proceed(int signo)
908 {
909 if (ready)
910 siglongjmp(jmpbuf, 1);
911 caught++;
912 }
913
914 void
915 enslave(void)
916 {
917 int cmd[2];
918 #ifdef LINUX_FORK_BUG
919 int i, j;
920 #else
921 register int i, j;
922 #endif
923
924 master = getpid();
925
926 { struct sigaction sa;
927 memset(&sa, 0, sizeof sa);
928 sigemptyset(&sa.sa_mask);
929 sa.sa_handler = dumpabort;
930 sigaction(SIGTERM, &sa, NULL); /* Slave sends SIGTERM on dumpabort() */
931 sa.sa_handler = sigpipe;
932 sigaction(SIGPIPE, &sa, NULL);
933 sa.sa_handler = proceed;
934 sa.sa_flags = SA_RESTART;
935 sigaction(SIGUSR2, &sa, NULL); /* Slave sends SIGUSR2 to next slave */
936 }
937
938 for (i = 0; i < SLAVES; i++) {
939 if (i == slp - &slaves[0]) {
940 caught = 1;
941 } else {
942 caught = 0;
943 }
944
945 if (socketpair(AF_UNIX, SOCK_STREAM, 0, cmd) < 0 ||
946 (slaves[i].pid = fork()) < 0)
947 quit("too many slaves, %d (recompile smaller): %s\n",
948 i, strerror(errno));
949
950 slaves[i].fd = cmd[1];
951 slaves[i].sent = 0;
952 if (slaves[i].pid == 0) { /* Slave starts up here */
953 sigset_t sigs;
954 for (j = 0; j <= i; j++)
955 (void) close(slaves[j].fd);
956 sigemptyset(&sigs);
957 sigaddset(&sigs, SIGINT); /* Master handles this */
958 #if defined(SIGINFO)
959 sigaddset(&sigs, SIGINFO);
960 #endif
961 sigprocmask(SIG_BLOCK, &sigs, NULL);
962
963 #ifdef LINUX_FORK_BUG
964 if (atomic_write( cmd[0], (char *) &i, sizeof i)
965 != sizeof i)
966 quit("master/slave protocol botched 3\n");
967 #endif
968 doslave(cmd[0], i, (slaves[i].pid == slp->pid));
969 Exit(X_FINOK);
970 }
971 else
972 close(cmd[0]);
973 }
974
975 #ifdef LINUX_FORK_BUG
976 /*
977 * Wait for all slaves to _actually_ start to circumvent a bug in
978 * Linux kernels >= 2.1.3 where a signal sent to a child that hasn't
979 * returned from fork() causes a SEGV in the child process
980 */
981 for (i = 0; i < SLAVES; i++)
982 if (atomic_read( slaves[i].fd, (char *) &j, sizeof j) != sizeof j)
983 quit("master/slave protocol botched 4\n");
984 #endif
985
986 for (i = 0; i < SLAVES; i++)
987 (void) atomic_write( slaves[i].fd,
988 (char *) &slaves[(i + 1) % SLAVES].pid,
989 sizeof slaves[0].pid);
990
991 master = 0;
992 }
993
994 void
995 killall(void)
996 {
997 register int i;
998
999 for (i = 0; i < SLAVES; i++)
1000 if (slaves[i].pid > 0) {
1001 (void) kill(slaves[i].pid, SIGKILL);
1002 slaves[i].sent = 0;
1003 }
1004 }
1005
1006 /*
1007 * Synchronization - each process waits for a SIGUSR2 from the
1008 * previous process before writing to the tape, and sends SIGUSR2
1009 * to the next process when the tape write completes. On tape errors
1010 * a SIGUSR1 is sent to the master which then terminates all of the
1011 * slaves.
1012 */
1013 static void
1014 doslave(int cmd, int slave_number, int first)
1015 {
1016 register int nread;
1017 int nextslave, size, eot_count, bufsize;
1018 volatile int wrote = 0;
1019 char *buffer;
1020 #ifdef HAVE_ZLIB
1021 struct tapebuf *comp_buf = NULL;
1022 int compresult, do_compress = !first;
1023 unsigned long worklen;
1024 #endif /* HAVE_ZLIB */
1025 struct slave_results returns;
1026 #ifdef __linux__
1027 errcode_t retval;
1028 #endif
1029 #ifdef USE_QFA
1030 long curtapepos;
1031 union u_spcl *uspclptr;
1032 struct s_spcl *spclptr;
1033 #endif /* USE_QFA */
1034
1035 /*
1036 * Need our own seek pointer.
1037 */
1038 (void) close(diskfd);
1039 if ((diskfd = OPEN(disk, O_RDONLY)) < 0)
1040 quit("slave couldn't reopen disk: %s\n", strerror(errno));
1041 #ifdef __linux__
1042 ext2fs_close(fs);
1043 retval = dump_fs_open(disk, &fs);
1044 if (retval)
1045 quit("slave couldn't reopen disk: %s\n", error_message(retval));
1046 #endif /* __linux__ */
1047
1048 /*
1049 * Need the pid of the next slave in the loop...
1050 */
1051 if ((nread = atomic_read( cmd, (char *)&nextslave, sizeof nextslave))
1052 != sizeof nextslave) {
1053 quit("master/slave protocol botched - didn't get pid of next slave.\n");
1054 }
1055
1056 #ifdef HAVE_ZLIB
1057 /* if we're doing a compressed dump, allocate the compress buffer */
1058 if (compressed) {
1059 comp_buf = malloc(sizeof(struct tapebuf) + TP_BSIZE + writesize);
1060 if (comp_buf == NULL)
1061 quit("couldn't allocate a compress buffer.\n");
1062 comp_buf->flags = 0;
1063 }
1064 #endif /* HAVE_ZLIB */
1065
1066 /*
1067 * Get list of blocks to dump, read the blocks into tape buffer
1068 */
1069 while ((nread = atomic_read( cmd, (char *)slp->req, reqsiz)) == reqsiz) {
1070 register struct req *p = slp->req;
1071
1072 for (trecno = 0; trecno < ntrec;
1073 trecno += p->count, p += p->count) {
1074 if (p->dblk) { /* read a disk block */
1075 bread(p->dblk, slp->tblock[trecno],
1076 p->count * TP_BSIZE);
1077 } else { /* read record from pipe */
1078 if (p->count != 1 || atomic_read( cmd,
1079 (char *)slp->tblock[trecno],
1080 TP_BSIZE) != TP_BSIZE)
1081 quit("master/slave protocol botched.\n");
1082 }
1083 }
1084
1085 /* Try to write the data... */
1086 wrote = 0;
1087 eot_count = 0;
1088 size = 0;
1089 buffer = (char *) slp->tblock[0]; /* set write pointer */
1090 bufsize = writesize; /* length to write */
1091 returns.clen = returns.unclen = bufsize;
1092
1093 #ifdef HAVE_ZLIB
1094 /*
1095 * When writing a compressed dump, each block except
1096 * the first one on each tape is written
1097 * from struct tapebuf with an 4 byte prefix
1098 * followed by the data. This can be less than
1099 * writesize. Restore, on a short read, can compare the
1100 * length read to the compressed length in the header
1101 * to verify that the read was good. Blocks which don't
1102 * compress well are written uncompressed.
1103 * The first block written by each slave is not compressed
1104 * and does not have a prefix.
1105 */
1106
1107 if (compressed && do_compress) {
1108 comp_buf->length = bufsize;
1109 worklen = TP_BSIZE + writesize;
1110 compresult = compress2(comp_buf->buf, &worklen,
1111 (char *)slp->tblock[0], writesize, compressed);
1112 if (compresult == Z_OK && worklen <= (writesize - 16)) {
1113 /* write the compressed buffer */
1114 comp_buf->length = worklen;
1115 comp_buf->compressed = 1;
1116 buffer = (char *) comp_buf;
1117 returns.clen = bufsize = worklen + sizeof(struct tapebuf);
1118 }
1119 else {
1120 /* write the data uncompressed */
1121 comp_buf->length = writesize;
1122 comp_buf->compressed = 0;
1123 buffer = (char *) comp_buf;
1124 returns.clen = bufsize = writesize + sizeof(struct tapebuf);
1125 returns.unclen = returns.clen;
1126 memcpy(comp_buf->buf, (char *)slp->tblock[0], writesize);
1127 }
1128 }
1129 /* compress the remaining blocks if we're compressing */
1130 do_compress = compressed;
1131 #endif /* HAVE_ZLIB */
1132
1133 if (sigsetjmp(jmpbuf, 1) == 0) {
1134 ready = 1;
1135 if (!caught)
1136 (void) pause();
1137 }
1138 ready = 0;
1139 caught = 0;
1140
1141 #ifdef USE_QFA
1142 if (gTapeposfd >= 0) {
1143 uspclptr = (union u_spcl *)&slp->tblock[0];
1144 spclptr = &uspclptr->s_spcl;
1145 if ((spclptr->c_magic == NFS_MAGIC) &&
1146 (spclptr->c_type == TS_INODE) &&
1147 (spclptr->c_date == gThisDumpDate)) {
1148 /* if an error occured previously don't
1149 * try again */
1150 if (gtperr == 0) {
1151 if ((gtperr = GetTapePos(&curtapepos)) == 0) {
1152 #ifdef DEBUG_QFA
1153 msg("inode %ld at tapepos %ld\n", spclptr->c_inumber, curtapepos);
1154 #endif
1155 sprintf(gTps, "%ld\t%d\t%ld\n", (unsigned long)spclptr->c_inumber, tapeno, curtapepos);
1156 if (write(gTapeposfd, gTps, strlen(gTps)) != strlen(gTps)) {
1157 warn("error writing tapepos file.\n");
1158 }
1159 }
1160 }
1161 }
1162 }
1163 #endif /* USE_QFA */
1164
1165 while (eot_count < 10 && size < bufsize) {
1166 #ifdef RDUMP
1167 if (host)
1168 wrote = rmtwrite(buffer + size, bufsize - size);
1169 else
1170 #endif
1171 wrote = write(tapefd, buffer + size, bufsize - size);
1172 #ifdef WRITEDEBUG
1173 printf("slave %d wrote %d\n", slave_number, wrote);
1174 #endif
1175 if (wrote < 0)
1176 break;
1177 if (wrote == 0)
1178 eot_count++;
1179 size += wrote;
1180 }
1181
1182 #ifdef WRITEDEBUG
1183 if (size != bufsize)
1184 printf("slave %d only wrote %d out of %d bytes and gave up.\n",
1185 slave_number, size, bufsize);
1186 #endif
1187
1188 /*
1189 * Handle ENOSPC as an EOT condition.
1190 */
1191 if (wrote < 0 && errno == ENOSPC) {
1192 wrote = 0;
1193 eot_count++;
1194 }
1195
1196 if (eot_count > 0)
1197 returns.clen = returns.unclen = 0;
1198
1199 /*
1200 * pass errno back to master for special handling
1201 */
1202 if (wrote < 0)
1203 returns.unclen = -errno;
1204
1205 /*
1206 * pass size of data and size of write back to master
1207 * (for EOT handling)
1208 */
1209 (void) atomic_write( cmd, (char *)&returns, sizeof returns);
1210
1211 /*
1212 * Signal the next slave to go.
1213 */
1214 (void) kill(nextslave, SIGUSR2);
1215 }
1216 if (nread != 0)
1217 quit("error reading command pipe: %s\n", strerror(errno));
1218 }
1219
1220 /*
1221 * Since a read from a pipe may not return all we asked for,
1222 * or a write may not write all we ask if we get a signal,
1223 * loop until the count is satisfied (or error).
1224 */
1225 static ssize_t
1226 atomic_read(int fd, void *buf, size_t count)
1227 {
1228 int got, need = count;
1229
1230 do {
1231 while ((got = read(fd, buf, need)) > 0 && (need -= got) > 0)
1232 (char *)buf += got;
1233 } while (got == -1 && errno == EINTR);
1234 return (got < 0 ? got : count - need);
1235 }
1236
1237 /*
1238 * Since a read from a pipe may not return all we asked for,
1239 * or a write may not write all we ask if we get a signal,
1240 * loop until the count is satisfied (or error).
1241 */
1242 static ssize_t
1243 atomic_write(int fd, const void *buf, size_t count)
1244 {
1245 int got, need = count;
1246
1247 do {
1248 while ((got = write(fd, buf, need)) > 0 && (need -= got) > 0)
1249 (char *)buf += got;
1250 } while (got == -1 && errno == EINTR);
1251 return (got < 0 ? got : count - need);
1252 }
1253
1254
1255 #ifdef USE_QFA
1256 /*
1257 * read the current tape position
1258 */
1259 int
1260 GetTapePos(long *pos)
1261 {
1262 int err = 0;
1263
1264 *pos = 0;
1265 if (ioctl(tapefd, MTIOCPOS, pos) == -1) {
1266 err = errno;
1267 msg("[%ld] error: %d (getting tapepos: %ld)\n", getpid(),
1268 err, *pos);
1269 return err;
1270 }
1271 return err;
1272 }
1273 #endif /* USE_QFA */