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