tape.c revision 32758
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 34#ifndef lint 35static char sccsid[] = "@(#)tape.c 8.4 (Berkeley) 5/1/95"; 36#endif /* not lint */ 37 38#include <sys/param.h> 39#include <sys/socket.h> 40#include <sys/time.h> 41#include <sys/wait.h> 42#ifdef sunos 43#include <sys/vnode.h> 44 45#include <ufs/fs.h> 46#include <ufs/inode.h> 47#else 48#include <ufs/ufs/dinode.h> 49#include <ufs/ffs/fs.h> 50#endif 51 52#include <protocols/dumprestore.h> 53 54#include <errno.h> 55#include <fcntl.h> 56#include <setjmp.h> 57#include <signal.h> 58#include <stdio.h> 59#ifdef __STDC__ 60#include <stdlib.h> 61#include <string.h> 62#include <unistd.h> 63#else 64int write(), read(); 65#endif 66 67#include "dump.h" 68#include "pathnames.h" 69 70int writesize; /* size of malloc()ed buffer for tape */ 71long lastspclrec = -1; /* tape block number of last written header */ 72int trecno = 0; /* next record to write in current block */ 73extern long blocksperfile; /* number of blocks per output file */ 74long blocksthisvol; /* number of blocks on current output file */ 75extern int ntrec; /* blocking factor on tape */ 76extern int cartridge; 77extern char *host; 78char *nexttape; 79 80static int atomic __P((ssize_t (*)(), int, char *, int)); 81static void doslave __P((int, int)); 82static void enslave __P((void)); 83static void flushtape __P((void)); 84static void killall __P((void)); 85static 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 */ 95struct req { 96 daddr_t dblk; 97 int count; 98}; 99int reqsiz; 100 101#define SLAVES 3 /* 1 slave writing, 1 reading, 1 for slack */ 102struct 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]; 114struct slave *slp; 115 116char (*nextblock)[TP_BSIZE]; 117 118int master; /* pid of master, for sending error signals */ 119int tenths; /* length of tape used per block written */ 120static int caught; /* have we caught the signal to proceed? */ 121static int ready; /* have we reached the lock point without having */ 122 /* received the SIGUSR2 signal from the prev slave? */ 123static jmp_buf jmpbuf; /* where to jump to if we are ready when the */ 124 /* SIGUSR2 arrives from the previous slave */ 125 126int 127alloctape() 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 166void 167writerec(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 *(union u_spcl *)(*(nextblock)++) = *(union u_spcl *)dp; 175 if (isspcl) 176 lastspclrec = spcl.c_tapea; 177 trecno++; 178 spcl.c_tapea++; 179 if (trecno >= ntrec) 180 flushtape(); 181} 182 183void 184dumpblock(blkno, size) 185 daddr_t blkno; 186 int size; 187{ 188 int avail, tpblks, dblkno; 189 190 dblkno = fsbtodb(sblock, blkno); 191 tpblks = size >> tp_bshift; 192 while ((avail = MIN(tpblks, ntrec - trecno)) > 0) { 193 slp->req[trecno].dblk = dblkno; 194 slp->req[trecno].count = avail; 195 trecno += avail; 196 spcl.c_tapea += avail; 197 if (trecno >= ntrec) 198 flushtape(); 199 dblkno += avail << (tp_bshift - dev_bshift); 200 tpblks -= avail; 201 } 202} 203 204int nogripe = 0; 205 206void 207tperror(signo) 208 int signo; 209{ 210 211 if (pipeout) { 212 msg("write error on %s\n", tape); 213 quit("Cannot recover\n"); 214 /* NOTREACHED */ 215 } 216 msg("write error %d blocks into volume %d\n", blocksthisvol, tapeno); 217 broadcast("DUMP WRITE ERROR!\n"); 218 if (!query("Do you want to restart?")) 219 dumpabort(0); 220 msg("Closing this volume. Prepare to restart with new media;\n"); 221 msg("this dump volume will be rewritten.\n"); 222 killall(); 223 nogripe = 1; 224 close_rewind(); 225 Exit(X_REWRITE); 226} 227 228void 229sigpipe(signo) 230 int signo; 231{ 232 233 quit("Broken pipe\n"); 234} 235 236static void 237flushtape() 238{ 239 int i, blks, got; 240 long lastfirstrec; 241 242 int siz = (char *)nextblock - (char *)slp->req; 243 244 slp->req[trecno].count = 0; /* Sentinel */ 245 246 if (atomic(write, slp->fd, (char *)slp->req, siz) != siz) 247 quit("error writing command pipe: %s\n", strerror(errno)); 248 slp->sent = 1; /* we sent a request, read the response later */ 249 250 lastfirstrec = slp->firstrec; 251 252 if (++slp >= &slaves[SLAVES]) 253 slp = &slaves[0]; 254 255 /* Read results back from next slave */ 256 if (slp->sent) { 257 if (atomic(read, slp->fd, (char *)&got, sizeof got) 258 != sizeof got) { 259 perror(" DUMP: error reading command pipe in master"); 260 dumpabort(0); 261 } 262 slp->sent = 0; 263 264 /* Check for end of tape */ 265 if (got < writesize) { 266 msg("End of tape detected\n"); 267 268 /* 269 * Drain the results, don't care what the values were. 270 * If we read them here then trewind won't... 271 */ 272 for (i = 0; i < SLAVES; i++) { 273 if (slaves[i].sent) { 274 if (atomic(read, slaves[i].fd, 275 (char *)&got, sizeof got) 276 != sizeof got) { 277 perror(" DUMP: error reading command pipe in master"); 278 dumpabort(0); 279 } 280 slaves[i].sent = 0; 281 } 282 } 283 284 close_rewind(); 285 rollforward(); 286 return; 287 } 288 } 289 290 blks = 0; 291 if (spcl.c_type != TS_END) { 292 for (i = 0; i < spcl.c_count; i++) 293 if (spcl.c_addr[i] != 0) 294 blks++; 295 } 296 slp->count = lastspclrec + blks + 1 - spcl.c_tapea; 297 slp->tapea = spcl.c_tapea; 298 slp->firstrec = lastfirstrec + ntrec; 299 slp->inode = curino; 300 nextblock = slp->tblock; 301 trecno = 0; 302 asize += tenths; 303 blockswritten += ntrec; 304 blocksthisvol += ntrec; 305 if (!pipeout && !unlimited && (blocksperfile ? 306 (blocksthisvol >= blocksperfile) : (asize > tsize))) { 307 close_rewind(); 308 startnewtape(0); 309 } 310 timeest(); 311} 312 313void 314trewind() 315{ 316 int f; 317 int got; 318 319 for (f = 0; f < SLAVES; f++) { 320 /* 321 * Drain the results, but unlike EOT we DO (or should) care 322 * what the return values were, since if we detect EOT after 323 * we think we've written the last blocks to the tape anyway, 324 * we have to replay those blocks with rollforward. 325 * 326 * fixme: punt for now. 327 */ 328 if (slaves[f].sent) { 329 if (atomic(read, slaves[f].fd, (char *)&got, sizeof got) 330 != sizeof got) { 331 perror(" DUMP: error reading command pipe in master"); 332 dumpabort(0); 333 } 334 slaves[f].sent = 0; 335 if (got != writesize) { 336 msg("EOT detected in last 2 tape records!\n"); 337 msg("Use a longer tape, decrease the size estimate\n"); 338 quit("or use no size estimate at all.\n"); 339 } 340 } 341 (void) close(slaves[f].fd); 342 } 343 while (wait((int *)NULL) >= 0) /* wait for any signals from slaves */ 344 /* void */; 345 346 if (pipeout) 347 return; 348 349 msg("Closing %s\n", tape); 350 351#ifdef RDUMP 352 if (host) { 353 rmtclose(); 354 while (rmtopen(tape, 0) < 0) 355 sleep(10); 356 rmtclose(); 357 return; 358 } 359#endif 360 (void) close(tapefd); 361 while ((f = open(tape, 0)) < 0) 362 sleep (10); 363 (void) close(f); 364} 365 366void 367close_rewind() 368{ 369 time_t tstart_changevol, tend_changevol; 370 371 trewind(); 372 if (nexttape) 373 return; 374 (void)time((time_t *)&(tstart_changevol)); 375 if (!nogripe) { 376 msg("Change Volumes: Mount volume #%d\n", tapeno+1); 377 broadcast("CHANGE DUMP VOLUMES!\7\7\n"); 378 } 379 while (!query("Is the new volume mounted and ready to go?")) 380 if (query("Do you want to abort?")) { 381 dumpabort(0); 382 /*NOTREACHED*/ 383 } 384 (void)time((time_t *)&(tend_changevol)); 385 if ((tstart_changevol != (time_t)-1) && (tend_changevol != (time_t)-1)) 386 tstart_writing += (tend_changevol - tstart_changevol); 387} 388 389void 390rollforward() 391{ 392 register struct req *p, *q, *prev; 393 register struct slave *tslp; 394 int i, size, savedtapea, got; 395 union u_spcl *ntb, *otb; 396 tslp = &slaves[SLAVES]; 397 ntb = (union u_spcl *)tslp->tblock[1]; 398 399 /* 400 * Each of the N slaves should have requests that need to 401 * be replayed on the next tape. Use the extra slave buffers 402 * (slaves[SLAVES]) to construct request lists to be sent to 403 * each slave in turn. 404 */ 405 for (i = 0; i < SLAVES; i++) { 406 q = &tslp->req[1]; 407 otb = (union u_spcl *)slp->tblock; 408 409 /* 410 * For each request in the current slave, copy it to tslp. 411 */ 412 413 prev = NULL; 414 for (p = slp->req; p->count > 0; p += p->count) { 415 *q = *p; 416 if (p->dblk == 0) 417 *ntb++ = *otb++; /* copy the datablock also */ 418 prev = q; 419 q += q->count; 420 } 421 if (prev == NULL) 422 quit("rollforward: protocol botch"); 423 if (prev->dblk != 0) 424 prev->count -= 1; 425 else 426 ntb--; 427 q -= 1; 428 q->count = 0; 429 q = &tslp->req[0]; 430 if (i == 0) { 431 q->dblk = 0; 432 q->count = 1; 433 trecno = 0; 434 nextblock = tslp->tblock; 435 savedtapea = spcl.c_tapea; 436 spcl.c_tapea = slp->tapea; 437 startnewtape(0); 438 spcl.c_tapea = savedtapea; 439 lastspclrec = savedtapea - 1; 440 } 441 size = (char *)ntb - (char *)q; 442 if (atomic(write, slp->fd, (char *)q, size) != size) { 443 perror(" DUMP: error writing command pipe"); 444 dumpabort(0); 445 } 446 slp->sent = 1; 447 if (++slp >= &slaves[SLAVES]) 448 slp = &slaves[0]; 449 450 q->count = 1; 451 452 if (prev->dblk != 0) { 453 /* 454 * If the last one was a disk block, make the 455 * first of this one be the last bit of that disk 456 * block... 457 */ 458 q->dblk = prev->dblk + 459 prev->count * (TP_BSIZE / DEV_BSIZE); 460 ntb = (union u_spcl *)tslp->tblock; 461 } else { 462 /* 463 * It wasn't a disk block. Copy the data to its 464 * new location in the buffer. 465 */ 466 q->dblk = 0; 467 *((union u_spcl *)tslp->tblock) = *ntb; 468 ntb = (union u_spcl *)tslp->tblock[1]; 469 } 470 } 471 slp->req[0] = *q; 472 nextblock = slp->tblock; 473 if (q->dblk == 0) 474 nextblock++; 475 trecno = 1; 476 477 /* 478 * Clear the first slaves' response. One hopes that it 479 * worked ok, otherwise the tape is much too short! 480 */ 481 if (slp->sent) { 482 if (atomic(read, slp->fd, (char *)&got, sizeof got) 483 != sizeof got) { 484 perror(" DUMP: error reading command pipe in master"); 485 dumpabort(0); 486 } 487 slp->sent = 0; 488 489 if (got != writesize) { 490 quit("EOT detected at start of the tape!\n"); 491 } 492 } 493} 494 495/* 496 * We implement taking and restoring checkpoints on the tape level. 497 * When each tape is opened, a new process is created by forking; this 498 * saves all of the necessary context in the parent. The child 499 * continues the dump; the parent waits around, saving the context. 500 * If the child returns X_REWRITE, then it had problems writing that tape; 501 * this causes the parent to fork again, duplicating the context, and 502 * everything continues as if nothing had happened. 503 */ 504void 505startnewtape(top) 506 int top; 507{ 508 int parentpid; 509 int childpid; 510 int status; 511 int waitpid; 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 522restore_check_point: 523 (void)signal(SIGINT, interrupt_save); 524 /* 525 * All signals are inherited... 526 */ 527 childpid = fork(); 528 if (childpid < 0) { 529 msg("Context save fork fails in parent %d\n", parentpid); 530 Exit(X_ABORT); 531 } 532 if (childpid != 0) { 533 /* 534 * PARENT: 535 * save the context by waiting 536 * until the child doing all of the work returns. 537 * don't catch the interrupt 538 */ 539 signal(SIGINT, SIG_IGN); 540#ifdef TDEBUG 541 msg("Tape: %d; parent process: %d child process %d\n", 542 tapeno+1, parentpid, childpid); 543#endif /* TDEBUG */ 544 while ((waitpid = wait(&status)) != childpid) 545 msg("Parent %d waiting for child %d has another child %d return\n", 546 parentpid, childpid, waitpid); 547 if (status & 0xFF) { 548 msg("Child %d returns LOB status %o\n", 549 childpid, status&0xFF); 550 } 551 status = (status >> 8) & 0xFF; 552#ifdef TDEBUG 553 switch(status) { 554 case X_FINOK: 555 msg("Child %d finishes X_FINOK\n", childpid); 556 break; 557 case X_ABORT: 558 msg("Child %d finishes X_ABORT\n", childpid); 559 break; 560 case X_REWRITE: 561 msg("Child %d finishes X_REWRITE\n", childpid); 562 break; 563 default: 564 msg("Child %d finishes unknown %d\n", 565 childpid, status); 566 break; 567 } 568#endif /* TDEBUG */ 569 switch(status) { 570 case X_FINOK: 571 Exit(X_FINOK); 572 case X_ABORT: 573 Exit(X_ABORT); 574 case X_REWRITE: 575 goto restore_check_point; 576 default: 577 msg("Bad return code from dump: %d\n", status); 578 Exit(X_ABORT); 579 } 580 /*NOTREACHED*/ 581 } else { /* we are the child; just continue */ 582#ifdef TDEBUG 583 sleep(4); /* allow time for parent's message to get out */ 584 msg("Child on Tape %d has parent %d, my pid = %d\n", 585 tapeno+1, parentpid, getpid()); 586#endif /* TDEBUG */ 587 /* 588 * If we have a name like "/dev/rmt0,/dev/rmt1", 589 * use the name before the comma first, and save 590 * the remaining names for subsequent volumes. 591 */ 592 tapeno++; /* current tape sequence */ 593 if (nexttape || strchr(tape, ',')) { 594 if (nexttape && *nexttape) 595 tape = nexttape; 596 if ((p = strchr(tape, ',')) != NULL) { 597 *p = '\0'; 598 nexttape = p + 1; 599 } else 600 nexttape = NULL; 601 msg("Dumping volume %d on %s\n", tapeno, tape); 602 } 603#ifdef RDUMP 604 while ((tapefd = (host ? rmtopen(tape, 2) : 605 pipeout ? 1 : open(tape, O_WRONLY|O_CREAT, 0666))) < 0) 606#else 607 while ((tapefd = (pipeout ? 1 : 608 open(tape, O_WRONLY|O_CREAT, 0666))) < 0) 609#endif 610 { 611 msg("Cannot open output \"%s\".\n", tape); 612 if (!query("Do you want to retry the open?")) 613 dumpabort(0); 614 } 615 616 enslave(); /* Share open tape file descriptor with slaves */ 617 618 asize = 0; 619 blocksthisvol = 0; 620 if (top) 621 newtape++; /* new tape signal */ 622 spcl.c_count = slp->count; 623 /* 624 * measure firstrec in TP_BSIZE units since restore doesn't 625 * know the correct ntrec value... 626 */ 627 spcl.c_firstrec = slp->firstrec; 628 spcl.c_volume++; 629 spcl.c_type = TS_TAPE; 630 spcl.c_flags |= DR_NEWHEADER; 631 writeheader((ino_t)slp->inode); 632 spcl.c_flags &=~ DR_NEWHEADER; 633 if (tapeno > 1) 634 msg("Volume %d begins with blocks from inode %d\n", 635 tapeno, slp->inode); 636 } 637} 638 639void 640dumpabort(signo) 641 int signo; 642{ 643 644 if (master != 0 && master != getpid()) 645 /* Signals master to call dumpabort */ 646 (void) kill(master, SIGTERM); 647 else { 648 killall(); 649 msg("The ENTIRE dump is aborted.\n"); 650 } 651#ifdef RDUMP 652 rmtclose(); 653#endif 654 Exit(X_ABORT); 655} 656 657void 658Exit(status) 659 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 */ 671void 672proceed(signo) 673 int signo; 674{ 675 676 if (ready) 677 longjmp(jmpbuf, 1); 678 caught++; 679} 680 681void 682enslave() 683{ 684 int cmd[2]; 685 register int i, j; 686 687 master = getpid(); 688 689 signal(SIGTERM, dumpabort); /* Slave sends SIGTERM on dumpabort() */ 690 signal(SIGPIPE, sigpipe); 691 signal(SIGUSR1, tperror); /* Slave sends SIGUSR1 on tape errors */ 692 signal(SIGUSR2, proceed); /* Slave sends SIGUSR2 to next slave */ 693 694 for (i = 0; i < SLAVES; i++) { 695 if (i == slp - &slaves[0]) { 696 caught = 1; 697 } else { 698 caught = 0; 699 } 700 701 if (socketpair(AF_UNIX, SOCK_STREAM, 0, cmd) < 0 || 702 (slaves[i].pid = fork()) < 0) 703 quit("too many slaves, %d (recompile smaller): %s\n", 704 i, strerror(errno)); 705 706 slaves[i].fd = cmd[1]; 707 slaves[i].sent = 0; 708 if (slaves[i].pid == 0) { /* Slave starts up here */ 709 for (j = 0; j <= i; j++) 710 (void) close(slaves[j].fd); 711 signal(SIGINT, SIG_IGN); /* Master handles this */ 712 doslave(cmd[0], i); 713 Exit(X_FINOK); 714 } 715 } 716 717 for (i = 0; i < SLAVES; i++) 718 (void) atomic(write, slaves[i].fd, 719 (char *) &slaves[(i + 1) % SLAVES].pid, 720 sizeof slaves[0].pid); 721 722 master = 0; 723} 724 725void 726killall() 727{ 728 register int i; 729 730 for (i = 0; i < SLAVES; i++) 731 if (slaves[i].pid > 0) { 732 (void) kill(slaves[i].pid, SIGKILL); 733 slaves[i].sent = 0; 734 } 735} 736 737/* 738 * Synchronization - each process has a lockfile, and shares file 739 * descriptors to the following process's lockfile. When our write 740 * completes, we release our lock on the following process's lock- 741 * file, allowing the following process to lock it and proceed. We 742 * get the lock back for the next cycle by swapping descriptors. 743 */ 744static void 745doslave(cmd, slave_number) 746 register int cmd; 747 int slave_number; 748{ 749 register int nread; 750 int nextslave, size, wrote, eot_count; 751 752 /* 753 * Need our own seek pointer. 754 */ 755 (void) close(diskfd); 756 if ((diskfd = open(disk, O_RDONLY)) < 0) 757 quit("slave couldn't reopen disk: %s\n", strerror(errno)); 758 759 /* 760 * Need the pid of the next slave in the loop... 761 */ 762 if ((nread = atomic(read, cmd, (char *)&nextslave, sizeof nextslave)) 763 != sizeof nextslave) { 764 quit("master/slave protocol botched - didn't get pid of next slave.\n"); 765 } 766 767 /* 768 * Get list of blocks to dump, read the blocks into tape buffer 769 */ 770 while ((nread = atomic(read, cmd, (char *)slp->req, reqsiz)) == reqsiz) { 771 register struct req *p = slp->req; 772 773 for (trecno = 0; trecno < ntrec; 774 trecno += p->count, p += p->count) { 775 if (p->dblk) { 776 bread(p->dblk, slp->tblock[trecno], 777 p->count * TP_BSIZE); 778 } else { 779 if (p->count != 1 || atomic(read, cmd, 780 (char *)slp->tblock[trecno], 781 TP_BSIZE) != TP_BSIZE) 782 quit("master/slave protocol botched.\n"); 783 } 784 } 785 if (setjmp(jmpbuf) == 0) { 786 ready = 1; 787 if (!caught) 788 (void) pause(); 789 } 790 ready = 0; 791 caught = 0; 792 793 /* Try to write the data... */ 794 eot_count = 0; 795 size = 0; 796 797 while (eot_count < 10 && size < writesize) { 798#ifdef RDUMP 799 if (host) 800 wrote = rmtwrite(slp->tblock[0]+size, 801 writesize-size); 802 else 803#endif 804 wrote = write(tapefd, slp->tblock[0]+size, 805 writesize-size); 806#ifdef WRITEDEBUG 807 printf("slave %d wrote %d\n", slave_number, wrote); 808#endif 809 if (wrote < 0) 810 break; 811 if (wrote == 0) 812 eot_count++; 813 size += wrote; 814 } 815 816#ifdef WRITEDEBUG 817 if (size != writesize) 818 printf("slave %d only wrote %d out of %d bytes and gave up.\n", 819 slave_number, size, writesize); 820#endif 821 822 if (eot_count > 0) 823 size = 0; 824 825 /* 826 * fixme: Pyramids running OSx return ENOSPC 827 * at EOT on 1/2 inch drives. 828 */ 829 if (wrote < 0) { 830 (void) kill(master, SIGUSR1); 831 for (;;) 832 (void) sigpause(0); 833 } else { 834 /* 835 * pass size of write back to master 836 * (for EOT handling) 837 */ 838 (void) atomic(write, cmd, (char *)&size, sizeof size); 839 } 840 841 /* 842 * If partial write, don't want next slave to go. 843 * Also jolts him awake. 844 */ 845 (void) kill(nextslave, SIGUSR2); 846 } 847 if (nread != 0) 848 quit("error reading command pipe: %s\n", strerror(errno)); 849} 850 851/* 852 * Since a read from a pipe may not return all we asked for, 853 * or a write may not write all we ask if we get a signal, 854 * loop until the count is satisfied (or error). 855 */ 856static int 857atomic(func, fd, buf, count) 858 ssize_t (*func)(); 859 int fd; 860 char *buf; 861 int count; 862{ 863 int got, need = count; 864 865 while ((got = (*func)(fd, buf, need)) > 0 && (need -= got) > 0) 866 buf += got; 867 return (got < 0 ? got : count - need); 868} 869