sbin/dump/tape.c

   1 /*-
   2  * Copyright (c) 1980, 1991, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  * 3. All advertising materials mentioning features or use of this software
  14  *    must display the following acknowledgement:
  15  *      This product includes software developed by the University of
  16  *      California, Berkeley and its contributors.
  17  * 4. Neither the name of the University nor the names of its contributors
  18  *    may be used to endorse or promote products derived from this software
  19  *    without specific prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  31  * SUCH DAMAGE.
  32  *
  33  * @(#)tape.c   8.4 (Berkeley) 5/1/95
  34  * $FreeBSD: src/sbin/dump/tape.c,v 1.12.2.3 2002/02/23 22:32:51 iedowse Exp $
  35  * $DragonFly: src/sbin/dump/tape.c,v 1.14 2005/08/28 04:35:12 dillon Exp $
  36  */
  37
  38 #include <sys/param.h>
  39 #include <sys/socket.h>
  40 #include <sys/time.h>
  41 #include <sys/wait.h>
  42 #include <sys/stat.h>
  43 #ifdef sunos
  44 #include <sys/vnode.h>
  45
  46 #include <ufs/fs.h>
  47 #include <ufs/inode.h>
  48 #else
  49 #include <vfs/ufs/dinode.h>
  50 #include <vfs/ufs/fs.h>
  51 #endif
  52
  53 #include <protocols/dumprestore.h>
  54
  55 #include <errno.h>
  56 #include <fcntl.h>
  57 #include <setjmp.h>
  58 #include <signal.h>
  59 #include <stdio.h>
  60 #ifdef __STDC__
  61 #include <stdlib.h>
  62 #include <string.h>
  63 #include <unistd.h>
  64 #else
  65 int     write(), read();
  66 #endif
  67
  68 #include "dump.h"
  69
  70 static int writesize;           /* size of malloc()ed buffer for tape */
  71 static long lastspclrec = -1;   /* tape block number of last written header */
  72 static int trecno = 0;          /* next record to write in current block */
  73 static long blocksthisvol;              /* number of blocks on current output file */
  74 static const char       *nexttape;
  75 static int tapeno = 0;  /* current tape number */
  76
  77 static int      atomic_read(int, void *, int);
  78 static int      atomic_write(int, const void *, int);
  79 #ifdef WRITEDEBUG
  80 static void     doslave(int, int);
  81 #else
  82 static void     doslave(int);
  83 #endif
  84 static void     enslave(void);
  85 static void     flushtape(void);
  86 static void     killall(void);
  87 static void     rollforward(void);
  88
  89 /*
  90  * Concurrent dump mods (Caltech) - disk block reading and tape writing
  91  * are exported to several slave processes.  While one slave writes the
  92  * tape, the others read disk blocks; they pass control of the tape in
  93  * a ring via signals. The parent process traverses the filesystem and
  94  * sends writeheader()'s and lists of daddr's to the slaves via pipes.
  95  * The following structure defines the instruction packets sent to slaves.
  96  */
  97 struct req {
  98         daddr_t dblk;
  99         int count;
 100 };
 101 int reqsiz;
 102
 103 #define SLAVES 3                /* 1 slave writing, 1 reading, 1 for slack */
 104 struct slave {
 105         int tapea;              /* header number at start of this chunk */
 106         int count;              /* count to next header (used for TS_TAPE */
 107                                 /* after EOT) */
 108         int inode;              /* inode that we are currently dealing with */
 109         int fd;                 /* FD for this slave */
 110         int pid;                /* PID for this slave */
 111         int sent;               /* 1 == we've sent this slave requests */
 112         int firstrec;           /* record number of this block */
 113         char (*tblock)[TP_BSIZE]; /* buffer for data blocks */
 114         struct req *req;        /* buffer for requests */
 115 } slaves[SLAVES+1];
 116 struct slave *slp;
 117
 118 char    (*nextblock)[TP_BSIZE];
 119
 120 int master;             /* pid of master, for sending error signals */
 121 int tenths;             /* length of tape used per block written */
 122 static int caught;      /* have we caught the signal to proceed? */
 123 static int ready;       /* have we reached the lock point without having */
 124                         /* received the SIGUSR2 signal from the prev slave? */
 125 static jmp_buf jmpbuf;  /* where to jump to if we are ready when the */
 126                         /* SIGUSR2 arrives from the previous slave */
 127
 128 int
 129 alloctape(void)
 130 {
 131         int pgoff = getpagesize() - 1;
 132         char *buf;
 133         int i;
 134
 135         writesize = ntrec * TP_BSIZE;
 136         reqsiz = (ntrec + 1) * sizeof(struct req);
 137         /*
 138          * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode
 139          * (see DEC TU80 User's Guide).  The shorter gaps of 6250-bpi require
 140          * repositioning after stopping, i.e, streaming mode, where the gap is
 141          * variable, 0.30" to 0.45".  The gap is maximal when the tape stops.
 142          */
 143         if (blocksperfile == 0 && !unlimited)
 144                 tenths = writesize / density +
 145                     (cartridge ? 16 : density == 625 ? 5 : 8);
 146         /*
 147          * Allocate tape buffer contiguous with the array of instruction
 148          * packets, so flushtape() can write them together with one write().
 149          * Align tape buffer on page boundary to speed up tape write().
 150          */
 151         for (i = 0; i <= SLAVES; i++) {
 152                 buf = (char *)
 153                     malloc((unsigned)(reqsiz + writesize + pgoff + TP_BSIZE));
 154                 if (buf == NULL)
 155                         return(0);
 156                 slaves[i].tblock = (char (*)[TP_BSIZE])
 157                     (((long)&buf[ntrec + 1] + pgoff) &~ pgoff);
 158                 slaves[i].req = (struct req *)slaves[i].tblock - ntrec - 1;
 159         }
 160         slp = &slaves[0];
 161         slp->count = 1;
 162         slp->tapea = 0;
 163         slp->firstrec = 0;
 164         nextblock = slp->tblock;
 165         return(1);
 166 }
 167
 168 void
 169 writerec(const void *dp, int isspcl)
 170 {
 171
 172         slp->req[trecno].dblk = (daddr_t)0;
 173         slp->req[trecno].count = 1;
 174         bcopy(dp, *(nextblock)++, sizeof (union u_spcl));
 175         if (isspcl)
 176                 lastspclrec = spcl.c_tapea;
 177         trecno++;
 178         spcl.c_tapea++;
 179         if (trecno >= ntrec)
 180                 flushtape();
 181 }
 182
 183 void
 184 dumpblock(daddr_t blkno, int size)
 185 {
 186         int avail, tpblks, dblkno;
 187
 188         dblkno = fsbtodb(sblock, blkno);
 189         tpblks = size >> tp_bshift;
 190         while ((avail = MIN(tpblks, ntrec - trecno)) > 0) {
 191                 slp->req[trecno].dblk = dblkno;
 192                 slp->req[trecno].count = avail;
 193                 trecno += avail;
 194                 spcl.c_tapea += avail;
 195                 if (trecno >= ntrec)
 196                         flushtape();
 197                 dblkno += avail << (tp_bshift - dev_bshift);
 198                 tpblks -= avail;
 199         }
 200 }
 201
 202 int     nogripe = 0;
 203
 204 static void
 205 tperror(int signo __unused)
 206 {
 207
 208         if (pipeout) {
 209                 msg("write error on %s\n", tape);
 210                 quit("Cannot recover\n");
 211                 /* NOTREACHED */
 212         }
 213         msg("write error %ld blocks into volume %d\n", blocksthisvol, tapeno);
 214         broadcast("DUMP WRITE ERROR!\n");
 215         if (!query("Do you want to restart?"))
 216                 dumpabort(0);
 217         msg("Closing this volume.  Prepare to restart with new media;\n");
 218         msg("this dump volume will be rewritten.\n");
 219         killall();
 220         nogripe = 1;
 221         close_rewind();
 222         Exit(X_REWRITE);
 223 }
 224
 225 static void
 226 sigpipe(int signo __unused)
 227 {
 228
 229         quit("Broken pipe\n");
 230 }
 231
 232 static void
 233 flushtape(void)
 234 {
 235         int i, blks, got;
 236         long lastfirstrec;
 237
 238         int siz = (char *)nextblock - (char *)slp->req;
 239
 240         slp->req[trecno].count = 0;                     /* Sentinel */
 241
 242         if (atomic_write(slp->fd, slp->req, siz) != siz)
 243                 quit("error writing command pipe: %s\n", strerror(errno));
 244         slp->sent = 1; /* we sent a request, read the response later */
 245
 246         lastfirstrec = slp->firstrec;
 247
 248         if (++slp >= &slaves[SLAVES])
 249                 slp = &slaves[0];
 250
 251         /* Read results back from next slave */
 252         if (slp->sent) {
 253                 if (atomic_read(slp->fd, &got, sizeof got)
 254                     != sizeof got) {
 255                         perror("  DUMP: error reading command pipe in master");
 256                         dumpabort(0);
 257                 }
 258                 slp->sent = 0;
 259
 260                 /* Check for end of tape */
 261                 if (got < writesize) {
 262                         msg("End of tape detected\n");
 263
 264                         /*
 265                          * Drain the results, don't care what the values were.
 266                          * If we read them here then trewind won't...
 267                          */
 268                         for (i = 0; i < SLAVES; i++) {
 269                                 if (slaves[i].sent) {
 270                                         if (atomic_read(slaves[i].fd,
 271                                             &got, sizeof got)
 272                                             != sizeof got) {
 273                                                 perror("  DUMP: error reading command pipe in master");
 274                                                 dumpabort(0);
 275                                         }
 276                                         slaves[i].sent = 0;
 277                                 }
 278                         }
 279
 280                         close_rewind();
 281                         rollforward();
 282                         return;
 283                 }
 284         }
 285
 286         blks = 0;
 287         if (spcl.c_type != TS_END) {
 288                 for (i = 0; i < spcl.c_count; i++)
 289                         if (spcl.c_addr[i] != 0)
 290                                 blks++;
 291         }
 292         slp->count = lastspclrec + blks + 1 - spcl.c_tapea;
 293         slp->tapea = spcl.c_tapea;
 294         slp->firstrec = lastfirstrec + ntrec;
 295         slp->inode = curino;
 296         nextblock = slp->tblock;
 297         trecno = 0;
 298         asize += tenths;
 299         blockswritten += ntrec;
 300         blocksthisvol += ntrec;
 301         if (!pipeout && !unlimited && (blocksperfile ?
 302             (blocksthisvol >= blocksperfile) : (asize > tsize))) {
 303                 close_rewind();
 304                 startnewtape(0);
 305         }
 306         timeest();
 307 }
 308
 309 void
 310 trewind(void)
 311 {
 312         struct stat sb;
 313         int f;
 314         int got;
 315
 316         for (f = 0; f < SLAVES; f++) {
 317                 /*
 318                  * Drain the results, but unlike EOT we DO (or should) care
 319                  * what the return values were, since if we detect EOT after
 320                  * we think we've written the last blocks to the tape anyway,
 321                  * we have to replay those blocks with rollforward.
 322                  *
 323                  * fixme: punt for now.
 324                  */
 325                 if (slaves[f].sent) {
 326                         if (atomic_read(slaves[f].fd, &got, sizeof got)
 327                             != sizeof got) {
 328                                 perror("  DUMP: error reading command pipe in master");
 329                                 dumpabort(0);
 330                         }
 331                         slaves[f].sent = 0;
 332                         if (got != writesize) {
 333                                 msg("EOT detected in last 2 tape records!\n");
 334                                 msg("Use a longer tape, decrease the size estimate\n");
 335                                 quit("or use no size estimate at all.\n");
 336                         }
 337                 }
 338                 close(slaves[f].fd);
 339         }
 340         while (wait((int *)NULL) >= 0)  /* wait for any signals from slaves */
 341                 /* void */;
 342
 343         if (pipeout)
 344                 return;
 345
 346         msg("Closing %s\n", tape);
 347
 348 #ifdef RDUMP
 349         if (host) {
 350                 rmtclose();
 351                 while (rmtopen(tape, 0) < 0)
 352                         sleep(10);
 353                 rmtclose();
 354                 return;
 355         }
 356 #endif
 357         if (fstat(tapefd, &sb) == 0 && S_ISFIFO(sb.st_mode)) {
 358                 close(tapefd);
 359                 return;
 360         }
 361         close(tapefd);
 362         while ((f = open(tape, 0)) < 0)
 363                 sleep (10);
 364         close(f);
 365 }
 366
 367 void
 368 close_rewind(void)
 369 {
 370         time_t tstart_changevol, tend_changevol;
 371
 372         trewind();
 373         if (nexttape)
 374                 return;
 375         time((time_t *)&(tstart_changevol));
 376         if (!nogripe) {
 377                 msg("Change Volumes: Mount volume #%d\n", tapeno+1);
 378                 broadcast("CHANGE DUMP VOLUMES!\a\a\n");
 379         }
 380         while (!query("Is the new volume mounted and ready to go?"))
 381                 if (query("Do you want to abort?")) {
 382                         dumpabort(0);
 383                         /*NOTREACHED*/
 384                 }
 385         time((time_t *)&(tend_changevol));
 386         if ((tstart_changevol != (time_t)-1) && (tend_changevol != (time_t)-1))
 387                 tstart_writing += (tend_changevol - tstart_changevol);
 388 }
 389
 390 void
 391 rollforward(void)
 392 {
 393         struct req *p, *q, *prev;
 394         struct slave *tslp;
 395         int i, size, savedtapea, got;
 396         union u_spcl *ntb, *otb;
 397         tslp = &slaves[SLAVES];
 398         ntb = (union u_spcl *)tslp->tblock[1];
 399
 400         /*
 401          * Each of the N slaves should have requests that need to
 402          * be replayed on the next tape.  Use the extra slave buffers
 403          * (slaves[SLAVES]) to construct request lists to be sent to
 404          * each slave in turn.
 405          */
 406         for (i = 0; i < SLAVES; i++) {
 407                 q = &tslp->req[1];
 408                 otb = (union u_spcl *)slp->tblock;
 409
 410                 /*
 411                  * For each request in the current slave, copy it to tslp.
 412                  */
 413
 414                 prev = NULL;
 415                 for (p = slp->req; p->count > 0; p += p->count) {
 416                         *q = *p;
 417                         if (p->dblk == 0)
 418                                 *ntb++ = *otb++; /* copy the datablock also */
 419                         prev = q;
 420                         q += q->count;
 421                 }
 422                 if (prev == NULL)
 423                         quit("rollforward: protocol botch");
 424                 if (prev->dblk != 0)
 425                         prev->count -= 1;
 426                 else
 427                         ntb--;
 428                 q -= 1;
 429                 q->count = 0;
 430                 q = &tslp->req[0];
 431                 if (i == 0) {
 432                         q->dblk = 0;
 433                         q->count = 1;
 434                         trecno = 0;
 435                         nextblock = tslp->tblock;
 436                         savedtapea = spcl.c_tapea;
 437                         spcl.c_tapea = slp->tapea;
 438                         startnewtape(0);
 439                         spcl.c_tapea = savedtapea;
 440                         lastspclrec = savedtapea - 1;
 441                 }
 442                 size = (char *)ntb - (char *)q;
 443                 if (atomic_write(slp->fd, q, size) != size) {
 444                         perror("  DUMP: error writing command pipe");
 445                         dumpabort(0);
 446                 }
 447                 slp->sent = 1;
 448                 if (++slp >= &slaves[SLAVES])
 449                         slp = &slaves[0];
 450
 451                 q->count = 1;
 452
 453                 if (prev->dblk != 0) {
 454                         /*
 455                          * If the last one was a disk block, make the
 456                          * first of this one be the last bit of that disk
 457                          * block...
 458                          */
 459                         q->dblk = prev->dblk +
 460                                 prev->count * (TP_BSIZE / DEV_BSIZE);
 461                         ntb = (union u_spcl *)tslp->tblock;
 462                 } else {
 463                         /*
 464                          * It wasn't a disk block.  Copy the data to its
 465                          * new location in the buffer.
 466                          */
 467                         q->dblk = 0;
 468                         *((union u_spcl *)tslp->tblock) = *ntb;
 469                         ntb = (union u_spcl *)tslp->tblock[1];
 470                 }
 471         }
 472         slp->req[0] = *q;
 473         nextblock = slp->tblock;
 474         if (q->dblk == 0)
 475                 nextblock++;
 476         trecno = 1;
 477
 478         /*
 479          * Clear the first slaves' response.  One hopes that it
 480          * worked ok, otherwise the tape is much too short!
 481          */
 482         if (slp->sent) {
 483                 if (atomic_read(slp->fd, &got, sizeof got)
 484                     != sizeof got) {
 485                         perror("  DUMP: error reading command pipe in master");
 486                         dumpabort(0);
 487                 }
 488                 slp->sent = 0;
 489
 490                 if (got != writesize) {
 491                         quit("EOT detected at start of the tape!\n");
 492                 }
 493         }
 494 }
 495
 496 /*
 497  * We implement taking and restoring checkpoints on the tape level.
 498  * When each tape is opened, a new process is created by forking; this
 499  * saves all of the necessary context in the parent.  The child
 500  * continues the dump; the parent waits around, saving the context.
 501  * If the child returns X_REWRITE, then it had problems writing that tape;
 502  * this causes the parent to fork again, duplicating the context, and
 503  * everything continues as if nothing had happened.
 504  */
 505 void
 506 startnewtape(int top)
 507 {
 508         int     parentpid;
 509         int     childpid;
 510         int     status;
 511         int     wait_pid;
 512         char    *p;
 513 #ifdef sunos
 514         void    (*interrupt_save)();
 515 #else
 516         sig_t   interrupt_save;
 517 #endif
 518
 519         interrupt_save = signal(SIGINT, SIG_IGN);
 520         parentpid = getpid();
 521
 522 restore_check_point:
 523         signal(SIGINT, interrupt_save);
 524         /*
 525          *      All signals are inherited...
 526          */
 527         setproctitle(NULL);     /* Restore the proctitle. */
 528         childpid = fork();
 529         if (childpid < 0) {
 530                 msg("Context save fork fails in parent %d\n", parentpid);
 531                 Exit(X_ABORT);
 532         }
 533         if (childpid != 0) {
 534                 /*
 535                  *      PARENT:
 536                  *      save the context by waiting
 537                  *      until the child doing all of the work returns.
 538                  *      don't catch the interrupt
 539                  */
 540                 signal(SIGINT, SIG_IGN);
 541 #ifdef TDEBUG
 542                 msg("Tape: %d; parent process: %d child process %d\n",
 543                         tapeno+1, parentpid, childpid);
 544 #endif /* TDEBUG */
 545                 while ((wait_pid = wait(&status)) != childpid)
 546                         msg("Parent %d waiting for child %d has another child %d return\n",
 547                                 parentpid, childpid, wait_pid);
 548                 if (status & 0xFF) {
 549                         msg("Child %d returns LOB status %o\n",
 550                                 childpid, status&0xFF);
 551                 }
 552                 status = (status >> 8) & 0xFF;
 553 #ifdef TDEBUG
 554                 switch(status) {
 555                         case X_FINOK:
 556                                 msg("Child %d finishes X_FINOK\n", childpid);
 557                                 break;
 558                         case X_ABORT:
 559                                 msg("Child %d finishes X_ABORT\n", childpid);
 560                                 break;
 561                         case X_REWRITE:
 562                                 msg("Child %d finishes X_REWRITE\n", childpid);
 563                                 break;
 564                         default:
 565                                 msg("Child %d finishes unknown %d\n",
 566                                         childpid, status);
 567                                 break;
 568                 }
 569 #endif /* TDEBUG */
 570                 switch(status) {
 571                         case X_FINOK:
 572                                 Exit(X_FINOK);
 573                         case X_ABORT:
 574                                 Exit(X_ABORT);
 575                         case X_REWRITE:
 576                                 goto restore_check_point;
 577                         default:
 578                                 msg("Bad return code from dump: %d\n", status);
 579                                 Exit(X_ABORT);
 580                 }
 581                 /*NOTREACHED*/
 582         } else {        /* we are the child; just continue */
 583 #ifdef TDEBUG
 584                 sleep(4);       /* allow time for parent's message to get out */
 585                 msg("Child on Tape %d has parent %d, my pid = %d\n",
 586                         tapeno+1, parentpid, getpid());
 587 #endif /* TDEBUG */
 588                 /*
 589                  * If we have a name like "/dev/rmt0,/dev/rmt1",
 590                  * use the name before the comma first, and save
 591                  * the remaining names for subsequent volumes.
 592                  */
 593                 tapeno++;               /* current tape sequence */
 594                 if (nexttape || strchr(tape, ',')) {
 595                         if (nexttape && *nexttape)
 596                                 tape = nexttape;
 597                         if ((p = strchr(tape, ',')) != NULL) {
 598                                 *p = '\0';
 599                                 nexttape = p + 1;
 600                         } else
 601                                 nexttape = NULL;
 602                         msg("Dumping volume %d on %s\n", tapeno, tape);
 603                 }
 604 #ifdef RDUMP
 605                 while ((tapefd = (host ? rmtopen(tape, 2) :
 606                         pipeout ? 1 : open(tape, O_WRONLY|O_CREAT, 0666))) < 0)
 607 #else
 608                 while ((tapefd = (pipeout ? 1 :
 609                                   open(tape, O_WRONLY|O_CREAT, 0666))) < 0)
 610 #endif
 611                     {
 612                         msg("Cannot open output \"%s\".\n", tape);
 613                         if (!query("Do you want to retry the open?"))
 614                                 dumpabort(0);
 615                 }
 616
 617                 enslave();  /* Share open tape file descriptor with slaves */
 618                 signal(SIGINFO, infosch);
 619
 620                 asize = 0;
 621                 blocksthisvol = 0;
 622                 if (top)
 623                         newtape++;              /* new tape signal */
 624                 spcl.c_count = slp->count;
 625                 /*
 626                  * measure firstrec in TP_BSIZE units since restore doesn't
 627                  * know the correct ntrec value...
 628                  */
 629                 spcl.c_firstrec = slp->firstrec;
 630                 spcl.c_volume++;
 631                 spcl.c_type = TS_TAPE;
 632                 spcl.c_flags |= DR_NEWHEADER;
 633                 writeheader((ufs1_ino_t)slp->inode);
 634                 spcl.c_flags &=~ DR_NEWHEADER;
 635                 if (tapeno > 1)
 636                         msg("Volume %d begins with blocks from inode %d\n",
 637                                 tapeno, slp->inode);
 638         }
 639 }
 640
 641 void
 642 dumpabort(int signo __unused)
 643 {
 644
 645         if (master != 0 && master != getpid())
 646                 /* Signals master to call dumpabort */
 647                 kill(master, SIGTERM);
 648         else {
 649                 killall();
 650                 msg("The ENTIRE dump is aborted.\n");
 651         }
 652 #ifdef RDUMP
 653         rmtclose();
 654 #endif
 655         Exit(X_ABORT);
 656 }
 657
 658 void
 659 Exit(int status)
 660 {
 661
 662 #ifdef TDEBUG
 663         msg("pid = %d exits with status %d\n", getpid(), status);
 664 #endif /* TDEBUG */
 665         exit(status);
 666 }
 667
 668 /*
 669  * proceed - handler for SIGUSR2, used to synchronize IO between the slaves.
 670  */
 671 static void
 672 proceed(int signo __unused)
 673 {
 674
 675         if (ready)
 676                 longjmp(jmpbuf, 1);
 677         caught++;
 678 }
 679
 680 static void
 681 enslave(void)
 682 {
 683         int cmd[2];
 684         int i, j;
 685
 686         master = getpid();
 687
 688         signal(SIGTERM, dumpabort);  /* Slave sends SIGTERM on dumpabort() */
 689         signal(SIGPIPE, sigpipe);
 690         signal(SIGUSR1, tperror);    /* Slave sends SIGUSR1 on tape errors */
 691         signal(SIGUSR2, proceed);    /* Slave sends SIGUSR2 to next slave */
 692
 693         for (i = 0; i < SLAVES; i++) {
 694                 if (i == slp - &slaves[0]) {
 695                         caught = 1;
 696                 } else {
 697                         caught = 0;
 698                 }
 699
 700                 if (socketpair(AF_UNIX, SOCK_STREAM, 0, cmd) < 0 ||
 701                     (slaves[i].pid = fork()) < 0)
 702                         quit("too many slaves, %d (recompile smaller): %s\n",
 703                             i, strerror(errno));
 704
 705                 slaves[i].fd = cmd[1];
 706                 slaves[i].sent = 0;
 707                 if (slaves[i].pid == 0) {           /* Slave starts up here */
 708                         for (j = 0; j <= i; j++)
 709                                 close(slaves[j].fd);
 710                         signal(SIGINT, SIG_IGN);    /* Master handles this */
 711 #ifdef WRITEDEBUG
 712                         doslave(cmd[0], i);
 713 #else
 714                         doslave(cmd[0]);
 715 #endif
 716                         Exit(X_FINOK);
 717                 }
 718         }
 719
 720         for (i = 0; i < SLAVES; i++)
 721                 atomic_write(slaves[i].fd, &slaves[(i + 1) % SLAVES].pid,
 722                     sizeof slaves[0].pid);
 723
 724         master = 0;
 725 }
 726
 727 void
 728 killall(void)
 729 {
 730         int i;
 731
 732         for (i = 0; i < SLAVES; i++)
 733                 if (slaves[i].pid > 0) {
 734                         kill(slaves[i].pid, SIGKILL);
 735                         slaves[i].sent = 0;
 736                 }
 737 }
 738
 739 /*
 740  * Synchronization - each process has a lockfile, and shares file
 741  * descriptors to the following process's lockfile.  When our write
 742  * completes, we release our lock on the following process's lock-
 743  * file, allowing the following process to lock it and proceed. We
 744  * get the lock back for the next cycle by swapping descriptors.
 745  */
 746 static void
 747 #ifdef WRITEDEBUG
 748 doslave(int cmd, int slave_number)
 749 #else
 750 doslave(int cmd)
 751 #endif
 752 {
 753         int nread;
 754         int nextslave, size, wrote, eot_count;
 755
 756         /*
 757          * Need our own seek pointer.
 758          */
 759         close(diskfd);
 760         if ((diskfd = open(disk, O_RDONLY)) < 0)
 761                 quit("slave couldn't reopen disk: %s\n", strerror(errno));
 762
 763         /*
 764          * Need the pid of the next slave in the loop...
 765          */
 766         if ((nread = atomic_read(cmd, &nextslave, sizeof nextslave))
 767             != sizeof nextslave) {
 768                 quit("master/slave protocol botched - didn't get pid of next slave.\n");
 769         }
 770
 771         /*
 772          * Get list of blocks to dump, read the blocks into tape buffer
 773          */
 774         while ((nread = atomic_read(cmd, slp->req, reqsiz)) == reqsiz) {
 775                 struct req *p = slp->req;
 776
 777                 for (trecno = 0; trecno < ntrec;
 778                      trecno += p->count, p += p->count) {
 779                         if (p->dblk) {
 780                                 bread(p->dblk, slp->tblock[trecno],
 781                                         p->count * TP_BSIZE);
 782                         } else {
 783                                 if (p->count != 1 ||
 784                                     atomic_read(cmd, slp->tblock[trecno],
 785                                     TP_BSIZE) != TP_BSIZE)
 786                                        quit("master/slave protocol botched.\n");
 787                         }
 788                 }
 789                 if (setjmp(jmpbuf) == 0) {
 790                         ready = 1;
 791                         if (!caught)
 792                                 pause();
 793                 }
 794                 ready = 0;
 795                 caught = 0;
 796
 797                 /* Try to write the data... */
 798                 eot_count = 0;
 799                 size = 0;
 800                 wrote = 0;
 801
 802                 while (eot_count < 10 && size < writesize) {
 803 #ifdef RDUMP
 804                         if (host)
 805                                 wrote = rmtwrite(slp->tblock[0]+size,
 806                                     writesize-size);
 807                         else
 808 #endif
 809                                 wrote = write(tapefd, slp->tblock[0]+size,
 810                                     writesize-size);
 811 #ifdef WRITEDEBUG
 812                         printf("slave %d wrote %d\n", slave_number, wrote);
 813 #endif
 814                         if (wrote < 0)
 815                                 break;
 816                         if (wrote == 0)
 817                                 eot_count++;
 818                         size += wrote;
 819                 }
 820
 821 #ifdef WRITEDEBUG
 822                 if (size != writesize)
 823                  printf("slave %d only wrote %d out of %d bytes and gave up.\n",
 824                      slave_number, size, writesize);
 825 #endif
 826
 827                 /*
 828                  * Handle ENOSPC as an EOT condition.
 829                  */
 830                 if (wrote < 0 && errno == ENOSPC) {
 831                         wrote = 0;
 832                         eot_count++;
 833                 }
 834
 835                 if (eot_count > 0)
 836                         size = 0;
 837
 838                 if (wrote < 0) {
 839                         kill(master, SIGUSR1);
 840                         for (;;)
 841                                 sigpause(0);
 842                 } else {
 843                         /*
 844                          * pass size of write back to master
 845                          * (for EOT handling)
 846                          */
 847                         atomic_write(cmd, &size, sizeof size);
 848                 }
 849
 850                 /*
 851                  * If partial write, don't want next slave to go.
 852                  * Also jolts him awake.
 853                  */
 854                 kill(nextslave, SIGUSR2);
 855         }
 856         if (nread != 0)
 857                 quit("error reading command pipe: %s\n", strerror(errno));
 858 }
 859
 860 static int
 861 atomic_read(int fd, void *buf, int count)
 862 {
 863         int got, need = count;
 864
 865         while ((got = read(fd, buf, need)) > 0 && (need -= got) > 0)
 866                 buf = (uint8_t *)buf + got;
 867         return (got < 0 ? got : count - need);
 868 }
 869
 870 static int
 871 atomic_write(int fd, const void *buf, int count)
 872 {
 873         int got, need = count;
 874
 875         while ((got = write(fd, buf, need)) > 0 && (need -= got) > 0)
 876                 buf = (const uint8_t *)buf + got;
 877         return (got < 0 ? got : count - need);
 878 }