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 1.2 2003/06/17 04:27:32 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 <ufs/ufs/dinode.h> 50 #include <ufs/ffs/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 int writesize; /* size of malloc()ed buffer for tape */ 71 long lastspclrec = -1; /* tape block number of last written header */ 72 int trecno = 0; /* next record to write in current block */ 73 extern long blocksperfile; /* number of blocks per output file */ 74 long blocksthisvol; /* number of blocks on current output file */ 75 extern int ntrec; /* blocking factor on tape */ 76 extern int cartridge; 77 extern char *host; 78 char *nexttape; 79 80 static int atomic __P((ssize_t (*)(), int, char *, int)); 81 static void doslave __P((int, int)); 82 static void enslave __P((void)); 83 static void flushtape __P((void)); 84 static void killall __P((void)); 85 static void rollforward __P((void)); 86 87 /* 88 * Concurrent dump mods (Caltech) - disk block reading and tape writing 89 * are exported to several slave processes. While one slave writes the 90 * tape, the others read disk blocks; they pass control of the tape in 91 * a ring via signals. The parent process traverses the filesystem and 92 * sends writeheader()'s and lists of daddr's to the slaves via pipes. 93 * The following structure defines the instruction packets sent to slaves. 94 */ 95 struct req { 96 daddr_t dblk; 97 int count; 98 }; 99 int reqsiz; 100 101 #define SLAVES 3 /* 1 slave writing, 1 reading, 1 for slack */ 102 struct slave { 103 int tapea; /* header number at start of this chunk */ 104 int count; /* count to next header (used for TS_TAPE */ 105 /* after EOT) */ 106 int inode; /* inode that we are currently dealing with */ 107 int fd; /* FD for this slave */ 108 int pid; /* PID for this slave */ 109 int sent; /* 1 == we've sent this slave requests */ 110 int firstrec; /* record number of this block */ 111 char (*tblock)[TP_BSIZE]; /* buffer for data blocks */ 112 struct req *req; /* buffer for requests */ 113 } slaves[SLAVES+1]; 114 struct slave *slp; 115 116 char (*nextblock)[TP_BSIZE]; 117 118 int master; /* pid of master, for sending error signals */ 119 int tenths; /* length of tape used per block written */ 120 static int caught; /* have we caught the signal to proceed? */ 121 static int ready; /* have we reached the lock point without having */ 122 /* received the SIGUSR2 signal from the prev slave? */ 123 static jmp_buf jmpbuf; /* where to jump to if we are ready when the */ 124 /* SIGUSR2 arrives from the previous slave */ 125 126 int 127 alloctape() 128 { 129 int pgoff = getpagesize() - 1; 130 char *buf; 131 int i; 132 133 writesize = ntrec * TP_BSIZE; 134 reqsiz = (ntrec + 1) * sizeof(struct req); 135 /* 136 * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode 137 * (see DEC TU80 User's Guide). The shorter gaps of 6250-bpi require 138 * repositioning after stopping, i.e, streaming mode, where the gap is 139 * variable, 0.30" to 0.45". The gap is maximal when the tape stops. 140 */ 141 if (blocksperfile == 0 && !unlimited) 142 tenths = writesize / density + 143 (cartridge ? 16 : density == 625 ? 5 : 8); 144 /* 145 * Allocate tape buffer contiguous with the array of instruction 146 * packets, so flushtape() can write them together with one write(). 147 * Align tape buffer on page boundary to speed up tape write(). 148 */ 149 for (i = 0; i <= SLAVES; i++) { 150 buf = (char *) 151 malloc((unsigned)(reqsiz + writesize + pgoff + TP_BSIZE)); 152 if (buf == NULL) 153 return(0); 154 slaves[i].tblock = (char (*)[TP_BSIZE]) 155 (((long)&buf[ntrec + 1] + pgoff) &~ pgoff); 156 slaves[i].req = (struct req *)slaves[i].tblock - ntrec - 1; 157 } 158 slp = &slaves[0]; 159 slp->count = 1; 160 slp->tapea = 0; 161 slp->firstrec = 0; 162 nextblock = slp->tblock; 163 return(1); 164 } 165 166 void 167 writerec(dp, isspcl) 168 char *dp; 169 int isspcl; 170 { 171 172 slp->req[trecno].dblk = (daddr_t)0; 173 slp->req[trecno].count = 1; 174 #ifndef __alpha__ 175 *(union u_spcl *)(*(nextblock)++) = *(union u_spcl *)dp; 176 #else 177 bcopy(dp, *(nextblock)++, sizeof (union u_spcl)); 178 #endif 179 if (isspcl) 180 lastspclrec = spcl.c_tapea; 181 trecno++; 182 spcl.c_tapea++; 183 if (trecno >= ntrec) 184 flushtape(); 185 } 186 187 void 188 dumpblock(blkno, size) 189 daddr_t blkno; 190 int size; 191 { 192 int avail, tpblks, dblkno; 193 194 dblkno = fsbtodb(sblock, blkno); 195 tpblks = size >> tp_bshift; 196 while ((avail = MIN(tpblks, ntrec - trecno)) > 0) { 197 slp->req[trecno].dblk = dblkno; 198 slp->req[trecno].count = avail; 199 trecno += avail; 200 spcl.c_tapea += avail; 201 if (trecno >= ntrec) 202 flushtape(); 203 dblkno += avail << (tp_bshift - dev_bshift); 204 tpblks -= avail; 205 } 206 } 207 208 int nogripe = 0; 209 210 void 211 tperror(signo) 212 int signo; 213 { 214 215 if (pipeout) { 216 msg("write error on %s\n", tape); 217 quit("Cannot recover\n"); 218 /* NOTREACHED */ 219 } 220 msg("write error %d blocks into volume %d\n", blocksthisvol, tapeno); 221 broadcast("DUMP WRITE ERROR!\n"); 222 if (!query("Do you want to restart?")) 223 dumpabort(0); 224 msg("Closing this volume. Prepare to restart with new media;\n"); 225 msg("this dump volume will be rewritten.\n"); 226 killall(); 227 nogripe = 1; 228 close_rewind(); 229 Exit(X_REWRITE); 230 } 231 232 void 233 sigpipe(signo) 234 int signo; 235 { 236 237 quit("Broken pipe\n"); 238 } 239 240 static void 241 flushtape() 242 { 243 int i, blks, got; 244 long lastfirstrec; 245 246 int siz = (char *)nextblock - (char *)slp->req; 247 248 slp->req[trecno].count = 0; /* Sentinel */ 249 250 if (atomic(write, slp->fd, (char *)slp->req, siz) != siz) 251 quit("error writing command pipe: %s\n", strerror(errno)); 252 slp->sent = 1; /* we sent a request, read the response later */ 253 254 lastfirstrec = slp->firstrec; 255 256 if (++slp >= &slaves[SLAVES]) 257 slp = &slaves[0]; 258 259 /* Read results back from next slave */ 260 if (slp->sent) { 261 if (atomic(read, slp->fd, (char *)&got, sizeof got) 262 != sizeof got) { 263 perror(" DUMP: error reading command pipe in master"); 264 dumpabort(0); 265 } 266 slp->sent = 0; 267 268 /* Check for end of tape */ 269 if (got < writesize) { 270 msg("End of tape detected\n"); 271 272 /* 273 * Drain the results, don't care what the values were. 274 * If we read them here then trewind won't... 275 */ 276 for (i = 0; i < SLAVES; i++) { 277 if (slaves[i].sent) { 278 if (atomic(read, slaves[i].fd, 279 (char *)&got, sizeof got) 280 != sizeof got) { 281 perror(" DUMP: error reading command pipe in master"); 282 dumpabort(0); 283 } 284 slaves[i].sent = 0; 285 } 286 } 287 288 close_rewind(); 289 rollforward(); 290 return; 291 } 292 } 293 294 blks = 0; 295 if (spcl.c_type != TS_END) { 296 for (i = 0; i < spcl.c_count; i++) 297 if (spcl.c_addr[i] != 0) 298 blks++; 299 } 300 slp->count = lastspclrec + blks + 1 - spcl.c_tapea; 301 slp->tapea = spcl.c_tapea; 302 slp->firstrec = lastfirstrec + ntrec; 303 slp->inode = curino; 304 nextblock = slp->tblock; 305 trecno = 0; 306 asize += tenths; 307 blockswritten += ntrec; 308 blocksthisvol += ntrec; 309 if (!pipeout && !unlimited && (blocksperfile ? 310 (blocksthisvol >= blocksperfile) : (asize > tsize))) { 311 close_rewind(); 312 startnewtape(0); 313 } 314 timeest(); 315 } 316 317 void 318 trewind() 319 { 320 struct stat sb; 321 int f; 322 int got; 323 324 for (f = 0; f < SLAVES; f++) { 325 /* 326 * Drain the results, but unlike EOT we DO (or should) care 327 * what the return values were, since if we detect EOT after 328 * we think we've written the last blocks to the tape anyway, 329 * we have to replay those blocks with rollforward. 330 * 331 * fixme: punt for now. 332 */ 333 if (slaves[f].sent) { 334 if (atomic(read, slaves[f].fd, (char *)&got, sizeof got) 335 != sizeof got) { 336 perror(" DUMP: error reading command pipe in master"); 337 dumpabort(0); 338 } 339 slaves[f].sent = 0; 340 if (got != writesize) { 341 msg("EOT detected in last 2 tape records!\n"); 342 msg("Use a longer tape, decrease the size estimate\n"); 343 quit("or use no size estimate at all.\n"); 344 } 345 } 346 (void) close(slaves[f].fd); 347 } 348 while (wait((int *)NULL) >= 0) /* wait for any signals from slaves */ 349 /* void */; 350 351 if (pipeout) 352 return; 353 354 msg("Closing %s\n", tape); 355 356 #ifdef RDUMP 357 if (host) { 358 rmtclose(); 359 while (rmtopen(tape, 0) < 0) 360 sleep(10); 361 rmtclose(); 362 return; 363 } 364 #endif 365 if (fstat(tapefd, &sb) == 0 && S_ISFIFO(sb.st_mode)) { 366 (void)close(tapefd); 367 return; 368 } 369 (void) close(tapefd); 370 while ((f = open(tape, 0)) < 0) 371 sleep (10); 372 (void) close(f); 373 } 374 375 void 376 close_rewind() 377 { 378 time_t tstart_changevol, tend_changevol; 379 380 trewind(); 381 if (nexttape) 382 return; 383 (void)time((time_t *)&(tstart_changevol)); 384 if (!nogripe) { 385 msg("Change Volumes: Mount volume #%d\n", tapeno+1); 386 broadcast("CHANGE DUMP VOLUMES!\a\a\n"); 387 } 388 while (!query("Is the new volume mounted and ready to go?")) 389 if (query("Do you want to abort?")) { 390 dumpabort(0); 391 /*NOTREACHED*/ 392 } 393 (void)time((time_t *)&(tend_changevol)); 394 if ((tstart_changevol != (time_t)-1) && (tend_changevol != (time_t)-1)) 395 tstart_writing += (tend_changevol - tstart_changevol); 396 } 397 398 void 399 rollforward() 400 { 401 register struct req *p, *q, *prev; 402 register struct slave *tslp; 403 int i, size, savedtapea, got; 404 union u_spcl *ntb, *otb; 405 tslp = &slaves[SLAVES]; 406 ntb = (union u_spcl *)tslp->tblock[1]; 407 408 /* 409 * Each of the N slaves should have requests that need to 410 * be replayed on the next tape. Use the extra slave buffers 411 * (slaves[SLAVES]) to construct request lists to be sent to 412 * each slave in turn. 413 */ 414 for (i = 0; i < SLAVES; i++) { 415 q = &tslp->req[1]; 416 otb = (union u_spcl *)slp->tblock; 417 418 /* 419 * For each request in the current slave, copy it to tslp. 420 */ 421 422 prev = NULL; 423 for (p = slp->req; p->count > 0; p += p->count) { 424 *q = *p; 425 if (p->dblk == 0) 426 *ntb++ = *otb++; /* copy the datablock also */ 427 prev = q; 428 q += q->count; 429 } 430 if (prev == NULL) 431 quit("rollforward: protocol botch"); 432 if (prev->dblk != 0) 433 prev->count -= 1; 434 else 435 ntb--; 436 q -= 1; 437 q->count = 0; 438 q = &tslp->req[0]; 439 if (i == 0) { 440 q->dblk = 0; 441 q->count = 1; 442 trecno = 0; 443 nextblock = tslp->tblock; 444 savedtapea = spcl.c_tapea; 445 spcl.c_tapea = slp->tapea; 446 startnewtape(0); 447 spcl.c_tapea = savedtapea; 448 lastspclrec = savedtapea - 1; 449 } 450 size = (char *)ntb - (char *)q; 451 if (atomic(write, slp->fd, (char *)q, size) != size) { 452 perror(" DUMP: error writing command pipe"); 453 dumpabort(0); 454 } 455 slp->sent = 1; 456 if (++slp >= &slaves[SLAVES]) 457 slp = &slaves[0]; 458 459 q->count = 1; 460 461 if (prev->dblk != 0) { 462 /* 463 * If the last one was a disk block, make the 464 * first of this one be the last bit of that disk 465 * block... 466 */ 467 q->dblk = prev->dblk + 468 prev->count * (TP_BSIZE / DEV_BSIZE); 469 ntb = (union u_spcl *)tslp->tblock; 470 } else { 471 /* 472 * It wasn't a disk block. Copy the data to its 473 * new location in the buffer. 474 */ 475 q->dblk = 0; 476 *((union u_spcl *)tslp->tblock) = *ntb; 477 ntb = (union u_spcl *)tslp->tblock[1]; 478 } 479 } 480 slp->req[0] = *q; 481 nextblock = slp->tblock; 482 if (q->dblk == 0) 483 nextblock++; 484 trecno = 1; 485 486 /* 487 * Clear the first slaves' response. One hopes that it 488 * worked ok, otherwise the tape is much too short! 489 */ 490 if (slp->sent) { 491 if (atomic(read, slp->fd, (char *)&got, sizeof got) 492 != sizeof got) { 493 perror(" DUMP: error reading command pipe in master"); 494 dumpabort(0); 495 } 496 slp->sent = 0; 497 498 if (got != writesize) { 499 quit("EOT detected at start of the tape!\n"); 500 } 501 } 502 } 503 504 /* 505 * We implement taking and restoring checkpoints on the tape level. 506 * When each tape is opened, a new process is created by forking; this 507 * saves all of the necessary context in the parent. The child 508 * continues the dump; the parent waits around, saving the context. 509 * If the child returns X_REWRITE, then it had problems writing that tape; 510 * this causes the parent to fork again, duplicating the context, and 511 * everything continues as if nothing had happened. 512 */ 513 void 514 startnewtape(top) 515 int top; 516 { 517 int parentpid; 518 int childpid; 519 int status; 520 int waitpid; 521 char *p; 522 #ifdef sunos 523 void (*interrupt_save)(); 524 #else 525 sig_t interrupt_save; 526 #endif 527 528 interrupt_save = signal(SIGINT, SIG_IGN); 529 parentpid = getpid(); 530 531 restore_check_point: 532 (void)signal(SIGINT, interrupt_save); 533 /* 534 * All signals are inherited... 535 */ 536 setproctitle(NULL); /* Restore the proctitle. */ 537 childpid = fork(); 538 if (childpid < 0) { 539 msg("Context save fork fails in parent %d\n", parentpid); 540 Exit(X_ABORT); 541 } 542 if (childpid != 0) { 543 /* 544 * PARENT: 545 * save the context by waiting 546 * until the child doing all of the work returns. 547 * don't catch the interrupt 548 */ 549 signal(SIGINT, SIG_IGN); 550 #ifdef TDEBUG 551 msg("Tape: %d; parent process: %d child process %d\n", 552 tapeno+1, parentpid, childpid); 553 #endif /* TDEBUG */ 554 while ((waitpid = wait(&status)) != childpid) 555 msg("Parent %d waiting for child %d has another child %d return\n", 556 parentpid, childpid, waitpid); 557 if (status & 0xFF) { 558 msg("Child %d returns LOB status %o\n", 559 childpid, status&0xFF); 560 } 561 status = (status >> 8) & 0xFF; 562 #ifdef TDEBUG 563 switch(status) { 564 case X_FINOK: 565 msg("Child %d finishes X_FINOK\n", childpid); 566 break; 567 case X_ABORT: 568 msg("Child %d finishes X_ABORT\n", childpid); 569 break; 570 case X_REWRITE: 571 msg("Child %d finishes X_REWRITE\n", childpid); 572 break; 573 default: 574 msg("Child %d finishes unknown %d\n", 575 childpid, status); 576 break; 577 } 578 #endif /* TDEBUG */ 579 switch(status) { 580 case X_FINOK: 581 Exit(X_FINOK); 582 case X_ABORT: 583 Exit(X_ABORT); 584 case X_REWRITE: 585 goto restore_check_point; 586 default: 587 msg("Bad return code from dump: %d\n", status); 588 Exit(X_ABORT); 589 } 590 /*NOTREACHED*/ 591 } else { /* we are the child; just continue */ 592 #ifdef TDEBUG 593 sleep(4); /* allow time for parent's message to get out */ 594 msg("Child on Tape %d has parent %d, my pid = %d\n", 595 tapeno+1, parentpid, getpid()); 596 #endif /* TDEBUG */ 597 /* 598 * If we have a name like "/dev/rmt0,/dev/rmt1", 599 * use the name before the comma first, and save 600 * the remaining names for subsequent volumes. 601 */ 602 tapeno++; /* current tape sequence */ 603 if (nexttape || strchr(tape, ',')) { 604 if (nexttape && *nexttape) 605 tape = nexttape; 606 if ((p = strchr(tape, ',')) != NULL) { 607 *p = '\0'; 608 nexttape = p + 1; 609 } else 610 nexttape = NULL; 611 msg("Dumping volume %d on %s\n", tapeno, tape); 612 } 613 #ifdef RDUMP 614 while ((tapefd = (host ? rmtopen(tape, 2) : 615 pipeout ? 1 : open(tape, O_WRONLY|O_CREAT, 0666))) < 0) 616 #else 617 while ((tapefd = (pipeout ? 1 : 618 open(tape, O_WRONLY|O_CREAT, 0666))) < 0) 619 #endif 620 { 621 msg("Cannot open output \"%s\".\n", tape); 622 if (!query("Do you want to retry the open?")) 623 dumpabort(0); 624 } 625 626 enslave(); /* Share open tape file descriptor with slaves */ 627 signal(SIGINFO, infosch); 628 629 asize = 0; 630 blocksthisvol = 0; 631 if (top) 632 newtape++; /* new tape signal */ 633 spcl.c_count = slp->count; 634 /* 635 * measure firstrec in TP_BSIZE units since restore doesn't 636 * know the correct ntrec value... 637 */ 638 spcl.c_firstrec = slp->firstrec; 639 spcl.c_volume++; 640 spcl.c_type = TS_TAPE; 641 spcl.c_flags |= DR_NEWHEADER; 642 writeheader((ino_t)slp->inode); 643 spcl.c_flags &=~ DR_NEWHEADER; 644 if (tapeno > 1) 645 msg("Volume %d begins with blocks from inode %d\n", 646 tapeno, slp->inode); 647 } 648 } 649 650 void 651 dumpabort(signo) 652 int signo; 653 { 654 655 if (master != 0 && master != getpid()) 656 /* Signals master to call dumpabort */ 657 (void) kill(master, SIGTERM); 658 else { 659 killall(); 660 msg("The ENTIRE dump is aborted.\n"); 661 } 662 #ifdef RDUMP 663 rmtclose(); 664 #endif 665 Exit(X_ABORT); 666 } 667 668 void 669 Exit(status) 670 int status; 671 { 672 673 #ifdef TDEBUG 674 msg("pid = %d exits with status %d\n", getpid(), status); 675 #endif /* TDEBUG */ 676 exit(status); 677 } 678 679 /* 680 * proceed - handler for SIGUSR2, used to synchronize IO between the slaves. 681 */ 682 void 683 proceed(signo) 684 int signo; 685 { 686 687 if (ready) 688 longjmp(jmpbuf, 1); 689 caught++; 690 } 691 692 void 693 enslave() 694 { 695 int cmd[2]; 696 register int i, j; 697 698 master = getpid(); 699 700 signal(SIGTERM, dumpabort); /* Slave sends SIGTERM on dumpabort() */ 701 signal(SIGPIPE, sigpipe); 702 signal(SIGUSR1, tperror); /* Slave sends SIGUSR1 on tape errors */ 703 signal(SIGUSR2, proceed); /* Slave sends SIGUSR2 to next slave */ 704 705 for (i = 0; i < SLAVES; i++) { 706 if (i == slp - &slaves[0]) { 707 caught = 1; 708 } else { 709 caught = 0; 710 } 711 712 if (socketpair(AF_UNIX, SOCK_STREAM, 0, cmd) < 0 || 713 (slaves[i].pid = fork()) < 0) 714 quit("too many slaves, %d (recompile smaller): %s\n", 715 i, strerror(errno)); 716 717 slaves[i].fd = cmd[1]; 718 slaves[i].sent = 0; 719 if (slaves[i].pid == 0) { /* Slave starts up here */ 720 for (j = 0; j <= i; j++) 721 (void) close(slaves[j].fd); 722 signal(SIGINT, SIG_IGN); /* Master handles this */ 723 doslave(cmd[0], i); 724 Exit(X_FINOK); 725 } 726 } 727 728 for (i = 0; i < SLAVES; i++) 729 (void) atomic(write, slaves[i].fd, 730 (char *) &slaves[(i + 1) % SLAVES].pid, 731 sizeof slaves[0].pid); 732 733 master = 0; 734 } 735 736 void 737 killall() 738 { 739 register int i; 740 741 for (i = 0; i < SLAVES; i++) 742 if (slaves[i].pid > 0) { 743 (void) kill(slaves[i].pid, SIGKILL); 744 slaves[i].sent = 0; 745 } 746 } 747 748 /* 749 * Synchronization - each process has a lockfile, and shares file 750 * descriptors to the following process's lockfile. When our write 751 * completes, we release our lock on the following process's lock- 752 * file, allowing the following process to lock it and proceed. We 753 * get the lock back for the next cycle by swapping descriptors. 754 */ 755 static void 756 doslave(cmd, slave_number) 757 register int cmd; 758 int slave_number; 759 { 760 register int nread; 761 int nextslave, size, wrote, eot_count; 762 763 /* 764 * Need our own seek pointer. 765 */ 766 (void) close(diskfd); 767 if ((diskfd = open(disk, O_RDONLY)) < 0) 768 quit("slave couldn't reopen disk: %s\n", strerror(errno)); 769 770 /* 771 * Need the pid of the next slave in the loop... 772 */ 773 if ((nread = atomic(read, cmd, (char *)&nextslave, sizeof nextslave)) 774 != sizeof nextslave) { 775 quit("master/slave protocol botched - didn't get pid of next slave.\n"); 776 } 777 778 /* 779 * Get list of blocks to dump, read the blocks into tape buffer 780 */ 781 while ((nread = atomic(read, cmd, (char *)slp->req, reqsiz)) == reqsiz) { 782 register struct req *p = slp->req; 783 784 for (trecno = 0; trecno < ntrec; 785 trecno += p->count, p += p->count) { 786 if (p->dblk) { 787 bread(p->dblk, slp->tblock[trecno], 788 p->count * TP_BSIZE); 789 } else { 790 if (p->count != 1 || atomic(read, cmd, 791 (char *)slp->tblock[trecno], 792 TP_BSIZE) != TP_BSIZE) 793 quit("master/slave protocol botched.\n"); 794 } 795 } 796 if (setjmp(jmpbuf) == 0) { 797 ready = 1; 798 if (!caught) 799 (void) pause(); 800 } 801 ready = 0; 802 caught = 0; 803 804 /* Try to write the data... */ 805 eot_count = 0; 806 size = 0; 807 808 while (eot_count < 10 && size < writesize) { 809 #ifdef RDUMP 810 if (host) 811 wrote = rmtwrite(slp->tblock[0]+size, 812 writesize-size); 813 else 814 #endif 815 wrote = write(tapefd, slp->tblock[0]+size, 816 writesize-size); 817 #ifdef WRITEDEBUG 818 printf("slave %d wrote %d\n", slave_number, wrote); 819 #endif 820 if (wrote < 0) 821 break; 822 if (wrote == 0) 823 eot_count++; 824 size += wrote; 825 } 826 827 #ifdef WRITEDEBUG 828 if (size != writesize) 829 printf("slave %d only wrote %d out of %d bytes and gave up.\n", 830 slave_number, size, writesize); 831 #endif 832 833 /* 834 * Handle ENOSPC as an EOT condition. 835 */ 836 if (wrote < 0 && errno == ENOSPC) { 837 wrote = 0; 838 eot_count++; 839 } 840 841 if (eot_count > 0) 842 size = 0; 843 844 if (wrote < 0) { 845 (void) kill(master, SIGUSR1); 846 for (;;) 847 (void) sigpause(0); 848 } else { 849 /* 850 * pass size of write back to master 851 * (for EOT handling) 852 */ 853 (void) atomic(write, cmd, (char *)&size, sizeof size); 854 } 855 856 /* 857 * If partial write, don't want next slave to go. 858 * Also jolts him awake. 859 */ 860 (void) kill(nextslave, SIGUSR2); 861 } 862 if (nread != 0) 863 quit("error reading command pipe: %s\n", strerror(errno)); 864 } 865 866 /* 867 * Since a read from a pipe may not return all we asked for, 868 * or a write may not write all we ask if we get a signal, 869 * loop until the count is satisfied (or error). 870 */ 871 static int 872 atomic(func, fd, buf, count) 873 ssize_t (*func)(); 874 int fd; 875 char *buf; 876 int count; 877 { 878 int got, need = count; 879 880 while ((got = (*func)(fd, buf, need)) > 0 && (need -= got) > 0) 881 buf += got; 882 return (got < 0 ? got : count - need); 883 } 884