1 /* $NetBSD: tape.c,v 1.57 2021/06/19 13:56:34 christos Exp $ */
2
3 /*-
4 * Copyright (c) 1980, 1991, 1993
5 * The Regents of the University of California. All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 */
31
32 #include <sys/cdefs.h>
33 #ifndef lint
34 #if 0
35 static char sccsid[] = "@(#)tape.c 8.4 (Berkeley) 5/1/95";
36 #else
37 __RCSID("$NetBSD: tape.c,v 1.57 2021/06/19 13:56:34 christos Exp $");
38 #endif
39 #endif /* not lint */
40
41 #include <sys/param.h>
42 #include <sys/socket.h>
43 #include <sys/time.h>
44 #include <sys/wait.h>
45 #include <sys/ioctl.h>
46 #include <sys/mtio.h>
47
48 #include <errno.h>
49 #include <fcntl.h>
50 #include <signal.h>
51 #include <stdio.h>
52 #include <stdlib.h>
53 #include <string.h>
54 #include <time.h>
55 #include <unistd.h>
56
57 #include "dump.h"
58 #include "pathnames.h"
59
60 int writesize; /* size of malloc()ed buffer for tape */
61 int64_t lastspclrec = -1; /* tape block number of last written header */
62 int trecno = 0; /* next record to write in current block */
63 extern long blocksperfile; /* number of blocks per output file */
64 long blocksthisvol; /* number of blocks on current output file */
65 extern int ntrec; /* blocking factor on tape */
66 extern int cartridge;
67 extern const char *host;
68 char *nexttape;
69
70 static ssize_t atomic_read(int, void *, int);
71 static ssize_t atomic_write(int, const void *, int);
72 static void doworker(int, int);
73 static void create_workers(void);
74 static void flushtape(void);
75 static void killall(void);
76 static void proceed(int);
77 static void rollforward(void);
78 static void sigpipe(int);
79 static void tperror(int);
80
81 /*
82 * Concurrent dump mods (Caltech) - disk block reading and tape writing
83 * are exported to several worker processes. While one worker writes the
84 * tape, the others read disk blocks; they pass control of the tape in
85 * a ring via signals. The parent process traverses the file system and
86 * sends writeheader()'s and lists of daddr's to the workers via pipes.
87 * The following structure defines the instruction packets sent to workers.
88 */
89 struct req {
90 daddr_t dblk;
91 int count;
92 };
93 int reqsiz;
94
95 #define WORKERS 3 /* 1 worker writing, 1 reading, 1 for slack */
96 struct worker {
97 int64_t tapea; /* header number at start of this chunk */
98 int64_t firstrec; /* record number of this block */
99 int count; /* count to next header (used for TS_TAPE */
100 /* after EOT) */
101 int inode; /* inode that we are currently dealing with */
102 int fd; /* FD for this worker */
103 int pid; /* PID for this worker */
104 int sent; /* 1 == we've sent this worker requests */
105 char (*tblock)[TP_BSIZE]; /* buffer for data blocks */
106 struct req *req; /* buffer for requests */
107 } workers[WORKERS+1];
108 struct worker *wp;
109
110 char (*nextblock)[TP_BSIZE];
111
112 static int64_t tapea_volume; /* value of spcl.c_tapea at volume start */
113
114 int master; /* pid of master, for sending error signals */
115 int tenths; /* length of tape used per block written */
116 static volatile sig_atomic_t caught; /* have we caught the signal to proceed? */
117
118 int
alloctape(void)119 alloctape(void)
120 {
121 int pgoff = getpagesize() - 1;
122 char *buf;
123 int i;
124
125 writesize = ntrec * TP_BSIZE;
126 reqsiz = (ntrec + 1) * sizeof(struct req);
127 /*
128 * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode
129 * (see DEC TU80 User's Guide). The shorter gaps of 6250-bpi require
130 * repositioning after stopping, i.e, streaming mode, where the gap is
131 * variable, 0.30" to 0.45". The gap is maximal when the tape stops.
132 */
133 if (blocksperfile == 0 && !unlimited)
134 tenths = writesize / density +
135 (cartridge ? 16 : density == 625 ? 5 : 8);
136 /*
137 * Allocate tape buffer contiguous with the array of instruction
138 * packets, so flushtape() can write them together with one write().
139 * Align tape buffer on page boundary to speed up tape write().
140 */
141 for (i = 0; i <= WORKERS; i++) {
142 buf = (char *)
143 xmalloc((unsigned)(reqsiz + writesize + pgoff + TP_BSIZE));
144 workers[i].tblock = (char (*)[TP_BSIZE])
145 (((long)&buf[ntrec + 1] + pgoff) &~ pgoff);
146 workers[i].req = (struct req *)workers[i].tblock - ntrec - 1;
147 }
148 wp = &workers[0];
149 wp->count = 1;
150 wp->tapea = 0;
151 wp->firstrec = 0;
152 nextblock = wp->tblock;
153 return(1);
154 }
155
156 void
writerec(const char * dp,int isspcl)157 writerec(const char *dp, int isspcl)
158 {
159
160 wp->req[trecno].dblk = (daddr_t)0;
161 wp->req[trecno].count = 1;
162 *(union u_spcl *)(*(nextblock)++) = *(const union u_spcl *)dp;
163 if (isspcl)
164 lastspclrec = iswap64(spcl.c_tapea);
165 trecno++;
166 spcl.c_tapea = iswap64(iswap64(spcl.c_tapea) +1);
167 if (trecno >= ntrec)
168 flushtape();
169 }
170
171 void
dumpblock(daddr_t blkno,int size)172 dumpblock(daddr_t blkno, int size)
173 {
174 int avail, tpblks;
175 daddr_t dblkno;
176
177 dblkno = fsatoda(ufsib, blkno);
178 tpblks = size >> tp_bshift;
179 while ((avail = MIN(tpblks, ntrec - trecno)) > 0) {
180 wp->req[trecno].dblk = dblkno;
181 wp->req[trecno].count = avail;
182 trecno += avail;
183 spcl.c_tapea = iswap64(iswap64(spcl.c_tapea) + avail);
184 if (trecno >= ntrec)
185 flushtape();
186 dblkno += avail << (tp_bshift - dev_bshift);
187 tpblks -= avail;
188 }
189 }
190
191 int nogripe = 0;
192
193 static void
tperror(int signo __unused)194 tperror(int signo __unused)
195 {
196
197 if (pipeout) {
198 msg("write error on %s\n", tape);
199 quit("Cannot recover");
200 /* NOTREACHED */
201 }
202 msg("write error %ld blocks into volume %d\n", blocksthisvol, tapeno);
203 broadcast("DUMP WRITE ERROR!\n");
204 if (!query("Do you want to restart?"))
205 dumpabort(0);
206 msg("Closing this volume. Prepare to restart with new media;\n");
207 msg("this dump volume will be rewritten.\n");
208 killall();
209 nogripe = 1;
210 close_rewind();
211 Exit(X_REWRITE);
212 }
213
214 static void
sigpipe(int signo __unused)215 sigpipe(int signo __unused)
216 {
217
218 quit("Broken pipe");
219 }
220
221 /*
222 * do_stats --
223 * Update xferrate stats
224 */
225 time_t
do_stats(void)226 do_stats(void)
227 {
228 time_t tnow, ttaken;
229 int64_t blocks;
230
231 (void)time(&tnow);
232 ttaken = tnow - tstart_volume;
233 blocks = iswap64(spcl.c_tapea) - tapea_volume;
234 msg("Volume %d completed at: %s", tapeno, ctime(&tnow));
235 if (ttaken > 0) {
236 msg("Volume %d took %d:%02d:%02d\n", tapeno,
237 (int) (ttaken / 3600), (int) ((ttaken % 3600) / 60),
238 (int) (ttaken % 60));
239 msg("Volume %d transfer rate: %d KB/s\n", tapeno,
240 (int) (blocks / ttaken));
241 xferrate += blocks / ttaken;
242 }
243 return(tnow);
244 }
245
246 /*
247 * statussig --
248 * information message upon receipt of SIGINFO
249 * (derived from optr.c::timeest())
250 */
251 void
statussig(int notused __unused)252 statussig(int notused __unused)
253 {
254 time_t tnow, deltat;
255 char msgbuf[128];
256 int errno_save;
257
258 if (blockswritten < 500)
259 return;
260 errno_save = errno;
261 (void) time((time_t *) &tnow);
262 if (tnow <= tstart_volume)
263 return;
264 deltat = tstart_writing - tnow +
265 (1.0 * (tnow - tstart_writing)) / blockswritten * tapesize;
266 (void)snprintf(msgbuf, sizeof(msgbuf),
267 "%3.2f%% done at %ld KB/s, finished in %d:%02d\n",
268 (blockswritten * 100.0) / tapesize,
269 (long)((iswap64(spcl.c_tapea) - tapea_volume) /
270 (tnow - tstart_volume)),
271 (int)(deltat / 3600), (int)((deltat % 3600) / 60));
272 write(STDERR_FILENO, msgbuf, strlen(msgbuf));
273 errno = errno_save;
274 }
275
276 static void
flushtape(void)277 flushtape(void)
278 {
279 int i, blks, got;
280 int64_t lastfirstrec;
281
282 int siz = (char *)nextblock - (char *)wp->req;
283
284 wp->req[trecno].count = 0; /* Sentinel */
285
286 if (atomic_write(wp->fd, wp->req, siz) != siz)
287 quite(errno, "error writing command pipe");
288 wp->sent = 1; /* we sent a request, read the response later */
289
290 lastfirstrec = wp->firstrec;
291
292 if (++wp >= &workers[WORKERS])
293 wp = &workers[0];
294
295 /* Read results back from next worker */
296 if (wp->sent) {
297 if (atomic_read(wp->fd, &got, sizeof got)
298 != sizeof got) {
299 perror(" DUMP: error reading command pipe in master");
300 dumpabort(0);
301 }
302 wp->sent = 0;
303
304 /* Check for end of tape */
305 if (got < writesize) {
306 msg("End of tape detected\n");
307
308 /*
309 * Drain the results, don't care what the values were.
310 * If we read them here then trewind won't...
311 */
312 for (i = 0; i < WORKERS; i++) {
313 if (workers[i].sent) {
314 if (atomic_read(workers[i].fd,
315 &got, sizeof got)
316 != sizeof got) {
317 perror(" DUMP: error reading command pipe in master");
318 dumpabort(0);
319 }
320 workers[i].sent = 0;
321 }
322 }
323
324 close_rewind();
325 rollforward();
326 return;
327 }
328 }
329
330 blks = 0;
331 if (iswap32(spcl.c_type) == TS_INODE ||
332 iswap32(spcl.c_type) == TS_ADDR) {
333 for (i = 0; i < iswap32(spcl.c_count); i++)
334 if (spcl.c_addr[i] != 0)
335 blks++;
336 }
337 wp->count = lastspclrec + blks + 1 - iswap64(spcl.c_tapea);
338 wp->tapea = iswap64(spcl.c_tapea);
339 wp->firstrec = lastfirstrec + ntrec;
340 wp->inode = curino;
341 nextblock = wp->tblock;
342 trecno = 0;
343 asize += tenths;
344 blockswritten += ntrec;
345 blocksthisvol += ntrec;
346 if (!pipeout && !unlimited && (blocksperfile ?
347 (blocksthisvol >= blocksperfile) : (asize > tsize))) {
348 close_rewind();
349 startnewtape(0);
350 }
351 timeest();
352 }
353
354 void
trewind(int eject)355 trewind(int eject)
356 {
357 int f;
358 int got;
359
360 for (f = 0; f < WORKERS; f++) {
361 /*
362 * Drain the results, but unlike EOT we DO (or should) care
363 * what the return values were, since if we detect EOT after
364 * we think we've written the last blocks to the tape anyway,
365 * we have to replay those blocks with rollforward.
366 *
367 * fixme: punt for now.
368 */
369 if (workers[f].sent) {
370 if (atomic_read(workers[f].fd, &got, sizeof got)
371 != sizeof got) {
372 perror(" DUMP: error reading command pipe in master");
373 dumpabort(0);
374 }
375 workers[f].sent = 0;
376 if (got != writesize) {
377 msg("EOT detected in last 2 tape records!\n");
378 msg("Use a longer tape, decrease the size estimate\n");
379 quit("or use no size estimate at all");
380 }
381 }
382 (void) close(workers[f].fd);
383 }
384 while (wait(NULL) >= 0) /* wait for any signals from workers */
385 /* void */;
386
387 if (pipeout)
388 return;
389
390 msg("Closing %s\n", tape);
391
392 #ifdef RDUMP
393 if (host) {
394 rmtclose();
395 while (rmtopen(tape, 0, 0) < 0)
396 sleep(10);
397 if (eflag && eject) {
398 msg("Ejecting %s\n", tape);
399 (void) rmtioctl(MTOFFL, 0);
400 }
401 rmtclose();
402 return;
403 }
404 #endif
405 (void) close(tapefd);
406 while ((f = open(tape, 0)) < 0)
407 sleep (10);
408 if (eflag && eject) {
409 struct mtop offl;
410
411 msg("Ejecting %s\n", tape);
412 offl.mt_op = MTOFFL;
413 offl.mt_count = 0;
414 (void) ioctl(f, MTIOCTOP, &offl);
415 }
416 (void) close(f);
417 }
418
419 void
close_rewind(void)420 close_rewind(void)
421 {
422 int i, f;
423
424 trewind(1);
425 (void)do_stats();
426 if (nexttape)
427 return;
428 if (!nogripe) {
429 msg("Change Volumes: Mount volume #%d\n", tapeno+1);
430 broadcast("CHANGE DUMP VOLUMES!\a\a\n");
431 }
432 if (lflag) {
433 for (i = 0; i < lflag / 10; i++) { /* wait lflag seconds */
434 if (host) {
435 if (rmtopen(tape, 0, 0) >= 0) {
436 rmtclose();
437 return;
438 }
439 } else {
440 if ((f = open(tape, 0)) >= 0) {
441 close(f);
442 return;
443 }
444 }
445 sleep (10);
446 }
447 }
448
449 while (!query("Is the new volume mounted and ready to go?"))
450 if (query("Do you want to abort?")) {
451 dumpabort(0);
452 /*NOTREACHED*/
453 }
454 }
455
456 void
rollforward(void)457 rollforward(void)
458 {
459 struct req *p, *q, *prev;
460 struct worker *twp;
461 int i, size, got;
462 int64_t savedtapea;
463 union u_spcl *ntb, *otb;
464 twp = &workers[WORKERS];
465 ntb = (union u_spcl *)twp->tblock[1];
466
467 /*
468 * Each of the N workers should have requests that need to
469 * be replayed on the next tape. Use the extra worker buffers
470 * (workers[WORKERS]) to construct request lists to be sent to
471 * each worker in turn.
472 */
473 for (i = 0; i < WORKERS; i++) {
474 q = &twp->req[1];
475 otb = (union u_spcl *)wp->tblock;
476
477 /*
478 * For each request in the current worker, copy it to twp.
479 */
480
481 prev = NULL;
482 for (p = wp->req; p->count > 0; p += p->count) {
483 *q = *p;
484 if (p->dblk == 0)
485 *ntb++ = *otb++; /* copy the datablock also */
486 prev = q;
487 q += q->count;
488 }
489 if (prev == NULL)
490 quit("rollforward: protocol botch");
491 if (prev->dblk != 0)
492 prev->count -= 1;
493 else
494 ntb--;
495 q -= 1;
496 q->count = 0;
497 q = &twp->req[0];
498 if (i == 0) {
499 q->dblk = 0;
500 q->count = 1;
501 trecno = 0;
502 nextblock = twp->tblock;
503 savedtapea = iswap64(spcl.c_tapea);
504 spcl.c_tapea = iswap64(wp->tapea);
505 startnewtape(0);
506 spcl.c_tapea = iswap64(savedtapea);
507 lastspclrec = savedtapea - 1;
508 }
509 size = (char *)ntb - (char *)q;
510 if (atomic_write(wp->fd, q, size) != size) {
511 perror(" DUMP: error writing command pipe");
512 dumpabort(0);
513 }
514 wp->sent = 1;
515 if (++wp >= &workers[WORKERS])
516 wp = &workers[0];
517
518 q->count = 1;
519
520 if (prev->dblk != 0) {
521 /*
522 * If the last one was a disk block, make the
523 * first of this one be the last bit of that disk
524 * block...
525 */
526 q->dblk = prev->dblk +
527 prev->count * (TP_BSIZE / DEV_BSIZE);
528 ntb = (union u_spcl *)twp->tblock;
529 } else {
530 /*
531 * It wasn't a disk block. Copy the data to its
532 * new location in the buffer.
533 */
534 q->dblk = 0;
535 *((union u_spcl *)twp->tblock) = *ntb;
536 ntb = (union u_spcl *)twp->tblock[1];
537 }
538 }
539 wp->req[0] = *q;
540 nextblock = wp->tblock;
541 if (q->dblk == 0)
542 nextblock++;
543 trecno = 1;
544
545 /*
546 * Clear the first workers' response. One hopes that it
547 * worked ok, otherwise the tape is much too short!
548 */
549 if (wp->sent) {
550 if (atomic_read(wp->fd, &got, sizeof got)
551 != sizeof got) {
552 perror(" DUMP: error reading command pipe in master");
553 dumpabort(0);
554 }
555 wp->sent = 0;
556
557 if (got != writesize) {
558 quit("EOT detected at start of the tape");
559 }
560 }
561 }
562
563 /*
564 * We implement taking and restoring checkpoints on the tape level.
565 * When each tape is opened, a new process is created by forking; this
566 * saves all of the necessary context in the parent. The child
567 * continues the dump; the parent waits around, saving the context.
568 * If the child returns X_REWRITE, then it had problems writing that tape;
569 * this causes the parent to fork again, duplicating the context, and
570 * everything continues as if nothing had happened.
571 */
572 void
startnewtape(int top)573 startnewtape(int top)
574 {
575 int parentpid;
576 int childpid;
577 int status;
578 int waitforpid;
579 char *p;
580 sig_t interrupt_save;
581
582 interrupt_save = signal(SIGINT, SIG_IGN);
583 parentpid = getpid();
584 tapea_volume = iswap64(spcl.c_tapea);
585 (void)time(&tstart_volume);
586
587 restore_check_point:
588 (void)signal(SIGINT, interrupt_save);
589 /*
590 * All signals are inherited...
591 */
592 childpid = fork();
593 if (childpid < 0) {
594 msg("Context save fork fails in parent %d\n", parentpid);
595 Exit(X_ABORT);
596 }
597 if (childpid != 0) {
598 /*
599 * PARENT:
600 * save the context by waiting
601 * until the child doing all of the work returns.
602 * don't catch the interrupt
603 */
604 signal(SIGINT, SIG_IGN);
605 signal(SIGINFO, SIG_IGN); /* only want child's stats */
606 #ifdef TDEBUG
607 msg("Tape: %d; parent process: %d child process %d\n",
608 tapeno+1, parentpid, childpid);
609 #endif /* TDEBUG */
610 while ((waitforpid = wait(&status)) != childpid)
611 msg("Parent %d waiting for child %d has another child %d return\n",
612 parentpid, childpid, waitforpid);
613 if (status & 0xFF) {
614 msg("Child %d returns LOB status %o\n",
615 childpid, status&0xFF);
616 }
617 status = (status >> 8) & 0xFF;
618 #ifdef TDEBUG
619 switch(status) {
620 case X_FINOK:
621 msg("Child %d finishes X_FINOK\n", childpid);
622 break;
623 case X_ABORT:
624 msg("Child %d finishes X_ABORT\n", childpid);
625 break;
626 case X_REWRITE:
627 msg("Child %d finishes X_REWRITE\n", childpid);
628 break;
629 default:
630 msg("Child %d finishes unknown %d\n",
631 childpid, status);
632 break;
633 }
634 #endif /* TDEBUG */
635 switch(status) {
636 case X_FINOK:
637 Exit(X_FINOK);
638 case X_ABORT:
639 Exit(X_ABORT);
640 case X_REWRITE:
641 goto restore_check_point;
642 default:
643 msg("Bad return code from dump: %d\n", status);
644 Exit(X_ABORT);
645 }
646 /*NOTREACHED*/
647 } else { /* we are the child; just continue */
648 signal(SIGINFO, statussig); /* now want child's stats */
649 #ifdef TDEBUG
650 sleep(4); /* allow time for parent's message to get out */
651 msg("Child on Tape %d has parent %d, my pid = %d\n",
652 tapeno+1, parentpid, getpid());
653 #endif /* TDEBUG */
654 /*
655 * If we have a name like "/dev/rst0,/dev/rst1",
656 * use the name before the comma first, and save
657 * the remaining names for subsequent volumes.
658 */
659 tapeno++; /* current tape sequence */
660 if (nexttape || strchr(tape, ',')) {
661 if (nexttape && *nexttape)
662 tape = nexttape;
663 if ((p = strchr(tape, ',')) != NULL) {
664 *p = '\0';
665 nexttape = p + 1;
666 } else
667 nexttape = NULL;
668 msg("Dumping volume %d on %s\n", tapeno, tape);
669 }
670 #ifdef RDUMP
671 while ((tapefd = (host ? rmtopen(tape, 2, 1) :
672 pipeout ? 1 : open(tape, O_WRONLY|O_CREAT, 0666))) < 0)
673 #else
674 while ((tapefd = (pipeout ? 1 :
675 open(tape, O_WRONLY|O_CREAT, 0666))) < 0)
676 #endif
677 {
678 msg("Cannot open output \"%s\".\n", tape);
679 if (!query("Do you want to retry the open?"))
680 dumpabort(0);
681 }
682
683 create_workers(); /* Share open tape file descriptor with workers */
684
685 asize = 0;
686 blocksthisvol = 0;
687 if (top)
688 newtape++; /* new tape signal */
689 spcl.c_count = iswap32(wp->count);
690 /*
691 * measure firstrec in TP_BSIZE units since restore doesn't
692 * know the correct ntrec value...
693 */
694 spcl.c_firstrec = iswap32(wp->firstrec);
695 spcl.c_volume = iswap32(iswap32(spcl.c_volume) + 1);
696 spcl.c_type = iswap32(TS_TAPE);
697 if (!is_ufs2)
698 spcl.c_flags = iswap32(iswap32(spcl.c_flags)
699 | DR_NEWHEADER);
700 writeheader((ino_t)wp->inode);
701 if (!is_ufs2)
702 spcl.c_flags = iswap32(iswap32(spcl.c_flags) &
703 ~ DR_NEWHEADER);
704 msg("Volume %d started at: %s", tapeno, ctime(&tstart_volume));
705 if (tapeno > 1)
706 msg("Volume %d begins with blocks from inode %d\n",
707 tapeno, wp->inode);
708 }
709 }
710
711 void
dumpabort(int signo __unused)712 dumpabort(int signo __unused)
713 {
714
715 if (master != 0 && master != getpid())
716 /* Signals master to call dumpabort */
717 (void) kill(master, SIGTERM);
718 else {
719 #ifdef DUMP_LFS
720 lfs_wrap_go();
721 #endif
722 killall();
723 msg("The ENTIRE dump is aborted.\n");
724 }
725 #ifdef RDUMP
726 rmtclose();
727 #endif
728 Exit(X_ABORT);
729 }
730
731 void
Exit(int status)732 Exit(int status)
733 {
734
735 #ifdef TDEBUG
736 msg("pid = %d exits with status %d\n", getpid(), status);
737 #endif /* TDEBUG */
738 exit(status);
739 }
740
741 /*
742 * proceed - handler for SIGUSR2, used to synchronize IO between the workers.
743 */
744 static void
proceed(int signo __unused)745 proceed(int signo __unused)
746 {
747 caught++;
748 }
749
750 void
create_workers(void)751 create_workers(void)
752 {
753 int cmd[2];
754 int i, j;
755
756 master = getpid();
757
758 signal(SIGTERM, dumpabort); /* Slave sends SIGTERM on dumpabort() */
759 signal(SIGPIPE, sigpipe);
760 signal(SIGUSR1, tperror); /* Slave sends SIGUSR1 on tape errors */
761 signal(SIGUSR2, proceed); /* Slave sends SIGUSR2 to next worker */
762
763 for (i = 0; i < WORKERS; i++) {
764 if (i == wp - &workers[0]) {
765 caught = 1;
766 } else {
767 caught = 0;
768 }
769
770 if (socketpair(AF_LOCAL, SOCK_STREAM, 0, cmd) < 0 ||
771 (workers[i].pid = fork()) < 0)
772 quite(errno, "too many workers, %d (recompile smaller)",
773 i);
774
775 workers[i].fd = cmd[1];
776 workers[i].sent = 0;
777 if (workers[i].pid == 0) { /* Slave starts up here */
778 for (j = 0; j <= i; j++)
779 (void) close(workers[j].fd);
780 signal(SIGINT, SIG_IGN); /* Master handles this */
781 signal(SIGINFO, SIG_IGN);
782 doworker(cmd[0], i);
783 Exit(X_FINOK);
784 }
785 }
786
787 for (i = 0; i < WORKERS; i++)
788 (void) atomic_write(workers[i].fd,
789 &workers[(i + 1) % WORKERS].pid,
790 sizeof workers[0].pid);
791
792 master = 0;
793 }
794
795 void
killall(void)796 killall(void)
797 {
798 int i;
799
800 for (i = 0; i < WORKERS; i++)
801 if (workers[i].pid > 0) {
802 (void) kill(workers[i].pid, SIGKILL);
803 workers[i].sent = 0;
804 }
805 }
806
807 /*
808 * Synchronization - each process has a lockfile, and shares file
809 * descriptors to the following process's lockfile. When our write
810 * completes, we release our lock on the following process's lock-
811 * file, allowing the following process to lock it and proceed. We
812 * get the lock back for the next cycle by swapping descriptors.
813 */
814 static void
doworker(int cmd,int worker_number __unused)815 doworker(int cmd, int worker_number __unused)
816 {
817 int nread, nextworker, size, wrote, eot_count, werror;
818 sigset_t nsigset, osigset;
819
820 wrote = 0;
821 /*
822 * Need our own seek pointer.
823 */
824 (void) close(diskfd);
825 if ((diskfd = open(disk_dev, O_RDONLY)) < 0)
826 quite(errno, "worker couldn't reopen disk");
827
828 /*
829 * Need the pid of the next worker in the loop...
830 */
831 if ((nread = atomic_read(cmd, &nextworker, sizeof nextworker))
832 != sizeof nextworker) {
833 quit("master/worker protocol botched - didn't get pid"
834 " of next worker");
835 }
836
837 /*
838 * Get list of blocks to dump, read the blocks into tape buffer
839 */
840 while ((nread = atomic_read(cmd, wp->req, reqsiz)) == reqsiz) {
841 struct req *p = wp->req;
842
843 for (trecno = 0; trecno < ntrec;
844 trecno += p->count, p += p->count) {
845 if (p->dblk) {
846 bread(p->dblk, wp->tblock[trecno],
847 p->count * TP_BSIZE);
848 } else {
849 if (p->count != 1 || atomic_read(cmd,
850 wp->tblock[trecno],
851 TP_BSIZE) != TP_BSIZE)
852 quit("master/worker protocol botched");
853 }
854 }
855
856 sigemptyset(&nsigset);
857 sigaddset(&nsigset, SIGUSR2);
858 sigprocmask(SIG_BLOCK, &nsigset, &osigset);
859 while (!caught)
860 sigsuspend(&osigset);
861 caught = 0;
862 sigprocmask(SIG_SETMASK, &osigset, NULL);
863
864 /* Try to write the data... */
865 eot_count = 0;
866 size = 0;
867 werror = 0;
868
869 while (eot_count < 10 && size < writesize) {
870 #ifdef RDUMP
871 if (host)
872 wrote = rmtwrite(wp->tblock[0]+size,
873 writesize-size);
874 else
875 #endif
876 wrote = write(tapefd, wp->tblock[0]+size,
877 writesize-size);
878 werror = errno;
879 #ifdef WRITEDEBUG
880 fprintf(stderr, "worker %d wrote %d werror %d\n",
881 worker_number, wrote, werror);
882 #endif
883 if (wrote < 0)
884 break;
885 if (wrote == 0)
886 eot_count++;
887 size += wrote;
888 }
889
890 #ifdef WRITEDEBUG
891 if (size != writesize)
892 fprintf(stderr,
893 "worker %d only wrote %d out of %d bytes and gave up.\n",
894 worker_number, size, writesize);
895 #endif
896
897 /*
898 * Handle ENOSPC as an EOT condition.
899 */
900 if (wrote < 0 && werror == ENOSPC) {
901 wrote = 0;
902 eot_count++;
903 }
904
905 if (eot_count > 0)
906 size = 0;
907
908 if (wrote < 0) {
909 (void) kill(master, SIGUSR1);
910 sigemptyset(&nsigset);
911 for (;;)
912 sigsuspend(&nsigset);
913 } else {
914 /*
915 * pass size of write back to master
916 * (for EOT handling)
917 */
918 (void) atomic_write(cmd, &size, sizeof size);
919 }
920
921 /*
922 * If partial write, don't want next worker to go.
923 * Also jolts him awake.
924 */
925 (void) kill(nextworker, SIGUSR2);
926 }
927 printcachestats();
928 if (nread != 0)
929 quite(errno, "error reading command pipe");
930 }
931
932 /*
933 * Since a read from a pipe may not return all we asked for,
934 * loop until the count is satisfied (or error).
935 */
936 static ssize_t
atomic_read(int fd,void * buf,int count)937 atomic_read(int fd, void *buf, int count)
938 {
939 ssize_t got, need = count;
940
941 while ((got = read(fd, buf, need)) > 0 && (need -= got) > 0)
942 buf = (char *)buf + got;
943 return (got < 0 ? got : count - need);
944 }
945
946 /*
947 * Since a write may not write all we ask if we get a signal,
948 * loop until the count is satisfied (or error).
949 */
950 static ssize_t
atomic_write(int fd,const void * buf,int count)951 atomic_write(int fd, const void *buf, int count)
952 {
953 ssize_t got, need = count;
954
955 while ((got = write(fd, buf, need)) > 0 && (need -= got) > 0)
956 buf = (const char *)buf + got;
957 return (got < 0 ? got : count - need);
958 }
959