tape.c revision 90496
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 35#if 0 36static char sccsid[] = "@(#)tape.c 8.4 (Berkeley) 5/1/95"; 37#endif 38static const char rcsid[] = 39 "$FreeBSD: head/sbin/dump/tape.c 90496 2002-02-11 00:50:50Z iedowse $"; 40#endif /* not lint */ 41 42#include <sys/param.h> 43#include <sys/socket.h> 44#include <sys/time.h> 45#include <sys/wait.h> 46#include <sys/stat.h> 47 48#include <ufs/ufs/dinode.h> 49#include <ufs/ffs/fs.h> 50 51#include <protocols/dumprestore.h> 52 53#include <errno.h> 54#include <fcntl.h> 55#include <setjmp.h> 56#include <signal.h> 57#include <stdio.h> 58#ifdef __STDC__ 59#include <stdlib.h> 60#include <string.h> 61#include <unistd.h> 62#else 63int write(), read(); 64#endif 65 66#include "dump.h" 67 68int writesize; /* size of malloc()ed buffer for tape */ 69long lastspclrec = -1; /* tape block number of last written header */ 70int trecno = 0; /* next record to write in current block */ 71extern long blocksperfile; /* number of blocks per output file */ 72long blocksthisvol; /* number of blocks on current output file */ 73extern int ntrec; /* blocking factor on tape */ 74extern int cartridge; 75extern char *host; 76char *nexttape; 77 78static int atomic __P((ssize_t (*)(), int, char *, int)); 79static void doslave __P((int, int)); 80static void enslave __P((void)); 81static void flushtape __P((void)); 82static void killall __P((void)); 83static void rollforward __P((void)); 84 85/* 86 * Concurrent dump mods (Caltech) - disk block reading and tape writing 87 * are exported to several slave processes. While one slave writes the 88 * tape, the others read disk blocks; they pass control of the tape in 89 * a ring via signals. The parent process traverses the filesystem and 90 * sends writeheader()'s and lists of daddr's to the slaves via pipes. 91 * The following structure defines the instruction packets sent to slaves. 92 */ 93struct req { 94 daddr_t dblk; 95 int count; 96}; 97int reqsiz; 98 99#define SLAVES 3 /* 1 slave writing, 1 reading, 1 for slack */ 100struct slave { 101 int tapea; /* header number at start of this chunk */ 102 int count; /* count to next header (used for TS_TAPE */ 103 /* after EOT) */ 104 int inode; /* inode that we are currently dealing with */ 105 int fd; /* FD for this slave */ 106 int pid; /* PID for this slave */ 107 int sent; /* 1 == we've sent this slave requests */ 108 int firstrec; /* record number of this block */ 109 char (*tblock)[TP_BSIZE]; /* buffer for data blocks */ 110 struct req *req; /* buffer for requests */ 111} slaves[SLAVES+1]; 112struct slave *slp; 113 114char (*nextblock)[TP_BSIZE]; 115 116int master; /* pid of master, for sending error signals */ 117int tenths; /* length of tape used per block written */ 118static int caught; /* have we caught the signal to proceed? */ 119static int ready; /* have we reached the lock point without having */ 120 /* received the SIGUSR2 signal from the prev slave? */ 121static jmp_buf jmpbuf; /* where to jump to if we are ready when the */ 122 /* SIGUSR2 arrives from the previous slave */ 123 124int 125alloctape() 126{ 127 int pgoff = getpagesize() - 1; 128 char *buf; 129 int i; 130 131 writesize = ntrec * TP_BSIZE; 132 reqsiz = (ntrec + 1) * sizeof(struct req); 133 /* 134 * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode 135 * (see DEC TU80 User's Guide). The shorter gaps of 6250-bpi require 136 * repositioning after stopping, i.e, streaming mode, where the gap is 137 * variable, 0.30" to 0.45". The gap is maximal when the tape stops. 138 */ 139 if (blocksperfile == 0 && !unlimited) 140 tenths = writesize / density + 141 (cartridge ? 16 : density == 625 ? 5 : 8); 142 /* 143 * Allocate tape buffer contiguous with the array of instruction 144 * packets, so flushtape() can write them together with one write(). 145 * Align tape buffer on page boundary to speed up tape write(). 146 */ 147 for (i = 0; i <= SLAVES; i++) { 148 buf = (char *) 149 malloc((unsigned)(reqsiz + writesize + pgoff + TP_BSIZE)); 150 if (buf == NULL) 151 return(0); 152 slaves[i].tblock = (char (*)[TP_BSIZE]) 153 (((long)&buf[ntrec + 1] + pgoff) &~ pgoff); 154 slaves[i].req = (struct req *)slaves[i].tblock - ntrec - 1; 155 } 156 slp = &slaves[0]; 157 slp->count = 1; 158 slp->tapea = 0; 159 slp->firstrec = 0; 160 nextblock = slp->tblock; 161 return(1); 162} 163 164void 165writerec(dp, isspcl) 166 char *dp; 167 int isspcl; 168{ 169 170 slp->req[trecno].dblk = (daddr_t)0; 171 slp->req[trecno].count = 1; 172#ifndef __alpha__ 173 *(union u_spcl *)(*(nextblock)++) = *(union u_spcl *)dp; 174#else 175 bcopy(dp, *(nextblock)++, sizeof (union u_spcl)); 176#endif 177 if (isspcl) 178 lastspclrec = spcl.c_tapea; 179 trecno++; 180 spcl.c_tapea++; 181 if (trecno >= ntrec) 182 flushtape(); 183} 184 185void 186dumpblock(blkno, size) 187 daddr_t blkno; 188 int size; 189{ 190 int avail, tpblks, dblkno; 191 192 dblkno = fsbtodb(sblock, blkno); 193 tpblks = size >> tp_bshift; 194 while ((avail = MIN(tpblks, ntrec - trecno)) > 0) { 195 slp->req[trecno].dblk = dblkno; 196 slp->req[trecno].count = avail; 197 trecno += avail; 198 spcl.c_tapea += avail; 199 if (trecno >= ntrec) 200 flushtape(); 201 dblkno += avail << (tp_bshift - dev_bshift); 202 tpblks -= avail; 203 } 204} 205 206int nogripe = 0; 207 208void 209tperror(signo) 210 int signo; 211{ 212 213 if (pipeout) { 214 msg("write error on %s\n", tape); 215 quit("Cannot recover\n"); 216 /* NOTREACHED */ 217 } 218 msg("write error %d blocks into volume %d\n", blocksthisvol, tapeno); 219 broadcast("DUMP WRITE ERROR!\n"); 220 if (!query("Do you want to restart?")) 221 dumpabort(0); 222 msg("Closing this volume. Prepare to restart with new media;\n"); 223 msg("this dump volume will be rewritten.\n"); 224 killall(); 225 nogripe = 1; 226 close_rewind(); 227 Exit(X_REWRITE); 228} 229 230void 231sigpipe(signo) 232 int signo; 233{ 234 235 quit("Broken pipe\n"); 236} 237 238static void 239flushtape() 240{ 241 int i, blks, got; 242 long lastfirstrec; 243 244 int siz = (char *)nextblock - (char *)slp->req; 245 246 slp->req[trecno].count = 0; /* Sentinel */ 247 248 if (atomic(write, slp->fd, (char *)slp->req, siz) != siz) 249 quit("error writing command pipe: %s\n", strerror(errno)); 250 slp->sent = 1; /* we sent a request, read the response later */ 251 252 lastfirstrec = slp->firstrec; 253 254 if (++slp >= &slaves[SLAVES]) 255 slp = &slaves[0]; 256 257 /* Read results back from next slave */ 258 if (slp->sent) { 259 if (atomic(read, slp->fd, (char *)&got, sizeof got) 260 != sizeof got) { 261 perror(" DUMP: error reading command pipe in master"); 262 dumpabort(0); 263 } 264 slp->sent = 0; 265 266 /* Check for end of tape */ 267 if (got < writesize) { 268 msg("End of tape detected\n"); 269 270 /* 271 * Drain the results, don't care what the values were. 272 * If we read them here then trewind won't... 273 */ 274 for (i = 0; i < SLAVES; i++) { 275 if (slaves[i].sent) { 276 if (atomic(read, slaves[i].fd, 277 (char *)&got, sizeof got) 278 != sizeof got) { 279 perror(" DUMP: error reading command pipe in master"); 280 dumpabort(0); 281 } 282 slaves[i].sent = 0; 283 } 284 } 285 286 close_rewind(); 287 rollforward(); 288 return; 289 } 290 } 291 292 blks = 0; 293 if (spcl.c_type != TS_END) { 294 for (i = 0; i < spcl.c_count; i++) 295 if (spcl.c_addr[i] != 0) 296 blks++; 297 } 298 slp->count = lastspclrec + blks + 1 - spcl.c_tapea; 299 slp->tapea = spcl.c_tapea; 300 slp->firstrec = lastfirstrec + ntrec; 301 slp->inode = curino; 302 nextblock = slp->tblock; 303 trecno = 0; 304 asize += tenths; 305 blockswritten += ntrec; 306 blocksthisvol += ntrec; 307 if (!pipeout && !unlimited && (blocksperfile ? 308 (blocksthisvol >= blocksperfile) : (asize > tsize))) { 309 close_rewind(); 310 startnewtape(0); 311 } 312 timeest(); 313} 314 315void 316trewind() 317{ 318 struct stat sb; 319 int f; 320 int got; 321 322 for (f = 0; f < SLAVES; f++) { 323 /* 324 * Drain the results, but unlike EOT we DO (or should) care 325 * what the return values were, since if we detect EOT after 326 * we think we've written the last blocks to the tape anyway, 327 * we have to replay those blocks with rollforward. 328 * 329 * fixme: punt for now. 330 */ 331 if (slaves[f].sent) { 332 if (atomic(read, slaves[f].fd, (char *)&got, sizeof got) 333 != sizeof got) { 334 perror(" DUMP: error reading command pipe in master"); 335 dumpabort(0); 336 } 337 slaves[f].sent = 0; 338 if (got != writesize) { 339 msg("EOT detected in last 2 tape records!\n"); 340 msg("Use a longer tape, decrease the size estimate\n"); 341 quit("or use no size estimate at all.\n"); 342 } 343 } 344 (void) close(slaves[f].fd); 345 } 346 while (wait((int *)NULL) >= 0) /* wait for any signals from slaves */ 347 /* void */; 348 349 if (pipeout) 350 return; 351 352 msg("Closing %s\n", tape); 353 354#ifdef RDUMP 355 if (host) { 356 rmtclose(); 357 while (rmtopen(tape, 0) < 0) 358 sleep(10); 359 rmtclose(); 360 return; 361 } 362#endif 363 if (fstat(tapefd, &sb) == 0 && S_ISFIFO(sb.st_mode)) { 364 (void)close(tapefd); 365 return; 366 } 367 (void) close(tapefd); 368 while ((f = open(tape, 0)) < 0) 369 sleep (10); 370 (void) close(f); 371} 372 373void 374close_rewind() 375{ 376 time_t tstart_changevol, tend_changevol; 377 378 trewind(); 379 if (nexttape) 380 return; 381 (void)time((time_t *)&(tstart_changevol)); 382 if (!nogripe) { 383 msg("Change Volumes: Mount volume #%d\n", tapeno+1); 384 broadcast("CHANGE DUMP VOLUMES!\a\a\n"); 385 } 386 while (!query("Is the new volume mounted and ready to go?")) 387 if (query("Do you want to abort?")) { 388 dumpabort(0); 389 /*NOTREACHED*/ 390 } 391 (void)time((time_t *)&(tend_changevol)); 392 if ((tstart_changevol != (time_t)-1) && (tend_changevol != (time_t)-1)) 393 tstart_writing += (tend_changevol - tstart_changevol); 394} 395 396void 397rollforward() 398{ 399 struct req *p, *q, *prev; 400 struct slave *tslp; 401 int i, size, savedtapea, got; 402 union u_spcl *ntb, *otb; 403 tslp = &slaves[SLAVES]; 404 ntb = (union u_spcl *)tslp->tblock[1]; 405 406 /* 407 * Each of the N slaves should have requests that need to 408 * be replayed on the next tape. Use the extra slave buffers 409 * (slaves[SLAVES]) to construct request lists to be sent to 410 * each slave in turn. 411 */ 412 for (i = 0; i < SLAVES; i++) { 413 q = &tslp->req[1]; 414 otb = (union u_spcl *)slp->tblock; 415 416 /* 417 * For each request in the current slave, copy it to tslp. 418 */ 419 420 prev = NULL; 421 for (p = slp->req; p->count > 0; p += p->count) { 422 *q = *p; 423 if (p->dblk == 0) 424 *ntb++ = *otb++; /* copy the datablock also */ 425 prev = q; 426 q += q->count; 427 } 428 if (prev == NULL) 429 quit("rollforward: protocol botch"); 430 if (prev->dblk != 0) 431 prev->count -= 1; 432 else 433 ntb--; 434 q -= 1; 435 q->count = 0; 436 q = &tslp->req[0]; 437 if (i == 0) { 438 q->dblk = 0; 439 q->count = 1; 440 trecno = 0; 441 nextblock = tslp->tblock; 442 savedtapea = spcl.c_tapea; 443 spcl.c_tapea = slp->tapea; 444 startnewtape(0); 445 spcl.c_tapea = savedtapea; 446 lastspclrec = savedtapea - 1; 447 } 448 size = (char *)ntb - (char *)q; 449 if (atomic(write, slp->fd, (char *)q, size) != size) { 450 perror(" DUMP: error writing command pipe"); 451 dumpabort(0); 452 } 453 slp->sent = 1; 454 if (++slp >= &slaves[SLAVES]) 455 slp = &slaves[0]; 456 457 q->count = 1; 458 459 if (prev->dblk != 0) { 460 /* 461 * If the last one was a disk block, make the 462 * first of this one be the last bit of that disk 463 * block... 464 */ 465 q->dblk = prev->dblk + 466 prev->count * (TP_BSIZE / DEV_BSIZE); 467 ntb = (union u_spcl *)tslp->tblock; 468 } else { 469 /* 470 * It wasn't a disk block. Copy the data to its 471 * new location in the buffer. 472 */ 473 q->dblk = 0; 474 *((union u_spcl *)tslp->tblock) = *ntb; 475 ntb = (union u_spcl *)tslp->tblock[1]; 476 } 477 } 478 slp->req[0] = *q; 479 nextblock = slp->tblock; 480 if (q->dblk == 0) 481 nextblock++; 482 trecno = 1; 483 484 /* 485 * Clear the first slaves' response. One hopes that it 486 * worked ok, otherwise the tape is much too short! 487 */ 488 if (slp->sent) { 489 if (atomic(read, slp->fd, (char *)&got, sizeof got) 490 != sizeof got) { 491 perror(" DUMP: error reading command pipe in master"); 492 dumpabort(0); 493 } 494 slp->sent = 0; 495 496 if (got != writesize) { 497 quit("EOT detected at start of the tape!\n"); 498 } 499 } 500} 501 502/* 503 * We implement taking and restoring checkpoints on the tape level. 504 * When each tape is opened, a new process is created by forking; this 505 * saves all of the necessary context in the parent. The child 506 * continues the dump; the parent waits around, saving the context. 507 * If the child returns X_REWRITE, then it had problems writing that tape; 508 * this causes the parent to fork again, duplicating the context, and 509 * everything continues as if nothing had happened. 510 */ 511void 512startnewtape(top) 513 int top; 514{ 515 int parentpid; 516 int childpid; 517 int status; 518 int waitpid; 519 char *p; 520 sig_t interrupt_save; 521 522 interrupt_save = signal(SIGINT, SIG_IGN); 523 parentpid = getpid(); 524 525restore_check_point: 526 (void)signal(SIGINT, interrupt_save); 527 /* 528 * All signals are inherited... 529 */ 530 childpid = fork(); 531 if (childpid < 0) { 532 msg("Context save fork fails in parent %d\n", parentpid); 533 Exit(X_ABORT); 534 } 535 if (childpid != 0) { 536 /* 537 * PARENT: 538 * save the context by waiting 539 * until the child doing all of the work returns. 540 * don't catch the interrupt 541 */ 542 signal(SIGINT, SIG_IGN); 543#ifdef TDEBUG 544 msg("Tape: %d; parent process: %d child process %d\n", 545 tapeno+1, parentpid, childpid); 546#endif /* TDEBUG */ 547 while ((waitpid = wait(&status)) != childpid) 548 msg("Parent %d waiting for child %d has another child %d return\n", 549 parentpid, childpid, waitpid); 550 if (status & 0xFF) { 551 msg("Child %d returns LOB status %o\n", 552 childpid, status&0xFF); 553 } 554 status = (status >> 8) & 0xFF; 555#ifdef TDEBUG 556 switch(status) { 557 case X_FINOK: 558 msg("Child %d finishes X_FINOK\n", childpid); 559 break; 560 case X_ABORT: 561 msg("Child %d finishes X_ABORT\n", childpid); 562 break; 563 case X_REWRITE: 564 msg("Child %d finishes X_REWRITE\n", childpid); 565 break; 566 default: 567 msg("Child %d finishes unknown %d\n", 568 childpid, status); 569 break; 570 } 571#endif /* TDEBUG */ 572 switch(status) { 573 case X_FINOK: 574 Exit(X_FINOK); 575 case X_ABORT: 576 Exit(X_ABORT); 577 case X_REWRITE: 578 goto restore_check_point; 579 default: 580 msg("Bad return code from dump: %d\n", status); 581 Exit(X_ABORT); 582 } 583 /*NOTREACHED*/ 584 } else { /* we are the child; just continue */ 585#ifdef TDEBUG 586 sleep(4); /* allow time for parent's message to get out */ 587 msg("Child on Tape %d has parent %d, my pid = %d\n", 588 tapeno+1, parentpid, getpid()); 589#endif /* TDEBUG */ 590 /* 591 * If we have a name like "/dev/rmt0,/dev/rmt1", 592 * use the name before the comma first, and save 593 * the remaining names for subsequent volumes. 594 */ 595 tapeno++; /* current tape sequence */ 596 if (nexttape || strchr(tape, ',')) { 597 if (nexttape && *nexttape) 598 tape = nexttape; 599 if ((p = strchr(tape, ',')) != NULL) { 600 *p = '\0'; 601 nexttape = p + 1; 602 } else 603 nexttape = NULL; 604 msg("Dumping volume %d on %s\n", tapeno, tape); 605 } 606#ifdef RDUMP 607 while ((tapefd = (host ? rmtopen(tape, 2) : 608 pipeout ? 1 : open(tape, O_WRONLY|O_CREAT, 0666))) < 0) 609#else 610 while ((tapefd = (pipeout ? 1 : 611 open(tape, O_WRONLY|O_CREAT, 0666))) < 0) 612#endif 613 { 614 msg("Cannot open output \"%s\".\n", tape); 615 if (!query("Do you want to retry the open?")) 616 dumpabort(0); 617 } 618 619 enslave(); /* Share open tape file descriptor with slaves */ 620 621 asize = 0; 622 blocksthisvol = 0; 623 if (top) 624 newtape++; /* new tape signal */ 625 spcl.c_count = slp->count; 626 /* 627 * measure firstrec in TP_BSIZE units since restore doesn't 628 * know the correct ntrec value... 629 */ 630 spcl.c_firstrec = slp->firstrec; 631 spcl.c_volume++; 632 spcl.c_type = TS_TAPE; 633 spcl.c_flags |= DR_NEWHEADER; 634 writeheader((ino_t)slp->inode); 635 spcl.c_flags &=~ DR_NEWHEADER; 636 if (tapeno > 1) 637 msg("Volume %d begins with blocks from inode %d\n", 638 tapeno, slp->inode); 639 } 640} 641 642void 643dumpabort(signo) 644 int signo; 645{ 646 647 if (master != 0 && master != getpid()) 648 /* Signals master to call dumpabort */ 649 (void) kill(master, SIGTERM); 650 else { 651 killall(); 652 msg("The ENTIRE dump is aborted.\n"); 653 } 654#ifdef RDUMP 655 rmtclose(); 656#endif 657 Exit(X_ABORT); 658} 659 660void 661Exit(status) 662 int status; 663{ 664 665#ifdef TDEBUG 666 msg("pid = %d exits with status %d\n", getpid(), status); 667#endif /* TDEBUG */ 668 exit(status); 669} 670 671/* 672 * proceed - handler for SIGUSR2, used to synchronize IO between the slaves. 673 */ 674void 675proceed(signo) 676 int signo; 677{ 678 679 if (ready) 680 longjmp(jmpbuf, 1); 681 caught++; 682} 683 684void 685enslave() 686{ 687 int cmd[2]; 688 int i, j; 689 690 master = getpid(); 691 692 signal(SIGTERM, dumpabort); /* Slave sends SIGTERM on dumpabort() */ 693 signal(SIGPIPE, sigpipe); 694 signal(SIGUSR1, tperror); /* Slave sends SIGUSR1 on tape errors */ 695 signal(SIGUSR2, proceed); /* Slave sends SIGUSR2 to next slave */ 696 697 for (i = 0; i < SLAVES; i++) { 698 if (i == slp - &slaves[0]) { 699 caught = 1; 700 } else { 701 caught = 0; 702 } 703 704 if (socketpair(AF_UNIX, SOCK_STREAM, 0, cmd) < 0 || 705 (slaves[i].pid = fork()) < 0) 706 quit("too many slaves, %d (recompile smaller): %s\n", 707 i, strerror(errno)); 708 709 slaves[i].fd = cmd[1]; 710 slaves[i].sent = 0; 711 if (slaves[i].pid == 0) { /* Slave starts up here */ 712 for (j = 0; j <= i; j++) 713 (void) close(slaves[j].fd); 714 signal(SIGINT, SIG_IGN); /* Master handles this */ 715 doslave(cmd[0], i); 716 Exit(X_FINOK); 717 } 718 } 719 720 for (i = 0; i < SLAVES; i++) 721 (void) atomic(write, slaves[i].fd, 722 (char *) &slaves[(i + 1) % SLAVES].pid, 723 sizeof slaves[0].pid); 724 725 master = 0; 726} 727 728void 729killall() 730{ 731 int i; 732 733 for (i = 0; i < SLAVES; i++) 734 if (slaves[i].pid > 0) { 735 (void) kill(slaves[i].pid, SIGKILL); 736 slaves[i].sent = 0; 737 } 738} 739 740/* 741 * Synchronization - each process has a lockfile, and shares file 742 * descriptors to the following process's lockfile. When our write 743 * completes, we release our lock on the following process's lock- 744 * file, allowing the following process to lock it and proceed. We 745 * get the lock back for the next cycle by swapping descriptors. 746 */ 747static void 748doslave(cmd, slave_number) 749 int cmd; 750 int slave_number; 751{ 752 int nread; 753 int nextslave, size, wrote, eot_count; 754 755 /* 756 * Need our own seek pointer. 757 */ 758 (void) close(diskfd); 759 if ((diskfd = open(disk, O_RDONLY)) < 0) 760 quit("slave couldn't reopen disk: %s\n", strerror(errno)); 761 762 /* 763 * Need the pid of the next slave in the loop... 764 */ 765 if ((nread = atomic(read, cmd, (char *)&nextslave, sizeof nextslave)) 766 != sizeof nextslave) { 767 quit("master/slave protocol botched - didn't get pid of next slave.\n"); 768 } 769 770 /* 771 * Get list of blocks to dump, read the blocks into tape buffer 772 */ 773 while ((nread = atomic(read, cmd, (char *)slp->req, reqsiz)) == reqsiz) { 774 struct req *p = slp->req; 775 776 for (trecno = 0; trecno < ntrec; 777 trecno += p->count, p += p->count) { 778 if (p->dblk) { 779 bread(p->dblk, slp->tblock[trecno], 780 p->count * TP_BSIZE); 781 } else { 782 if (p->count != 1 || atomic(read, cmd, 783 (char *)slp->tblock[trecno], 784 TP_BSIZE) != TP_BSIZE) 785 quit("master/slave protocol botched.\n"); 786 } 787 } 788 if (setjmp(jmpbuf) == 0) { 789 ready = 1; 790 if (!caught) 791 (void) pause(); 792 } 793 ready = 0; 794 caught = 0; 795 796 /* Try to write the data... */ 797 eot_count = 0; 798 size = 0; 799 800 while (eot_count < 10 && size < writesize) { 801#ifdef RDUMP 802 if (host) 803 wrote = rmtwrite(slp->tblock[0]+size, 804 writesize-size); 805 else 806#endif 807 wrote = write(tapefd, slp->tblock[0]+size, 808 writesize-size); 809#ifdef WRITEDEBUG 810 printf("slave %d wrote %d\n", slave_number, wrote); 811#endif 812 if (wrote < 0) 813 break; 814 if (wrote == 0) 815 eot_count++; 816 size += wrote; 817 } 818 819#ifdef WRITEDEBUG 820 if (size != writesize) 821 printf("slave %d only wrote %d out of %d bytes and gave up.\n", 822 slave_number, size, writesize); 823#endif 824 825 /* 826 * Handle ENOSPC as an EOT condition. 827 */ 828 if (wrote < 0 && errno == ENOSPC) { 829 wrote = 0; 830 eot_count++; 831 } 832 833 if (eot_count > 0) 834 size = 0; 835 836 if (wrote < 0) { 837 (void) kill(master, SIGUSR1); 838 for (;;) 839 (void) sigpause(0); 840 } else { 841 /* 842 * pass size of write back to master 843 * (for EOT handling) 844 */ 845 (void) atomic(write, cmd, (char *)&size, sizeof size); 846 } 847 848 /* 849 * If partial write, don't want next slave to go. 850 * Also jolts him awake. 851 */ 852 (void) kill(nextslave, SIGUSR2); 853 } 854 if (nread != 0) 855 quit("error reading command pipe: %s\n", strerror(errno)); 856} 857 858/* 859 * Since a read from a pipe may not return all we asked for, 860 * or a write may not write all we ask if we get a signal, 861 * loop until the count is satisfied (or error). 862 */ 863static int 864atomic(func, fd, buf, count) 865 ssize_t (*func)(); 866 int fd; 867 char *buf; 868 int count; 869{ 870 int got, need = count; 871 872 while ((got = (*func)(fd, buf, need)) > 0 && (need -= got) > 0) 873 buf += got; 874 return (got < 0 ? got : count - need); 875} 876