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