1 /* $OpenBSD: tape.c,v 1.43 2015/08/20 22:02:20 deraadt Exp $ */ 2 /* $NetBSD: tape.c,v 1.11 1997/06/05 11:13:26 lukem Exp $ */ 3 4 /*- 5 * Copyright (c) 1980, 1991, 1993 6 * The Regents of the University of California. All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 */ 32 33 #include <sys/param.h> /* MAXBSIZE DEV_BSIZE */ 34 #include <sys/socket.h> 35 #include <sys/time.h> 36 #include <sys/wait.h> 37 #include <sys/stat.h> 38 #include <ufs/ffs/fs.h> 39 #include <ufs/ufs/dinode.h> 40 41 #include <protocols/dumprestore.h> 42 43 #include <errno.h> 44 #include <fcntl.h> 45 #include <signal.h> 46 #include <stdio.h> 47 #include <stdlib.h> 48 #include <string.h> 49 #include <time.h> 50 #include <unistd.h> 51 #include <limits.h> 52 53 #include "dump.h" 54 #include "pathnames.h" 55 56 #define MINIMUM(a, b) (((a) < (b)) ? (a) : (b)) 57 58 int writesize; /* size of malloc()ed buffer for tape */ 59 int64_t lastspclrec = -1; /* tape block number of last written header */ 60 int trecno = 0; /* next record to write in current block */ 61 extern int64_t blocksperfile; /* number of blocks per output file */ 62 int64_t blocksthisvol; /* number of blocks on current output file */ 63 extern int ntrec; /* blocking factor on tape */ 64 extern int cartridge; 65 extern char *host; 66 char *nexttape; 67 68 static ssize_t atomic(ssize_t (*)(int, void *, size_t), int, char *, int); 69 static void doslave(int, int); 70 static void enslave(void); 71 static void flushtape(void); 72 static void killall(void); 73 static void rollforward(void); 74 75 void tperror(int signo); 76 void sigpipe(int signo); 77 void proceed(int signo); 78 79 /* 80 * Concurrent dump mods (Caltech) - disk block reading and tape writing 81 * are exported to several slave processes. While one slave writes the 82 * tape, the others read disk blocks; they pass control of the tape in 83 * a ring via signals. The parent process traverses the filesystem and 84 * sends writeheader()'s and lists of daddr's to the slaves via pipes. 85 * The following structure defines the instruction packets sent to slaves. 86 */ 87 struct req { 88 daddr_t dblk; 89 int count; 90 }; 91 int reqsiz; 92 93 #define SLAVES 3 /* 1 slave writing, 1 reading, 1 for slack */ 94 struct slave { 95 int64_t tapea; /* header number at start of this chunk */ 96 int64_t firstrec; /* record number of this block */ 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 pid_t pid; /* PID for this slave */ 102 int sent; /* 1 == we've sent this slave requests */ 103 char (*tblock)[TP_BSIZE]; /* buffer for data blocks */ 104 struct req *req; /* buffer for requests */ 105 } slaves[SLAVES+1]; 106 struct slave *slp; 107 108 char (*nextblock)[TP_BSIZE]; 109 110 static time_t tstart_volume; /* time of volume start */ 111 static int64_t tapea_volume; /* value of spcl.c_tapea at volume start */ 112 113 pid_t master; /* pid of master, for sending error signals */ 114 int tenths; /* length of tape used per block written */ 115 static volatile sig_atomic_t caught; /* have we caught the signal to proceed? */ 116 117 int 118 alloctape(void) 119 { 120 int pgoff = getpagesize() - 1; 121 char *buf; 122 int i; 123 124 writesize = ntrec * TP_BSIZE; 125 reqsiz = (ntrec + 1) * sizeof(struct req); 126 /* 127 * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode 128 * (see DEC TU80 User's Guide). The shorter gaps of 6250-bpi require 129 * repositioning after stopping, i.e, streaming mode, where the gap is 130 * variable, 0.30" to 0.45". The gap is maximal when the tape stops. 131 */ 132 if (blocksperfile == 0 && !unlimited) 133 tenths = writesize / density + 134 (cartridge ? 16 : density == 625 ? 5 : 8); 135 /* 136 * Allocate tape buffer contiguous with the array of instruction 137 * packets, so flushtape() can write them together with one write(). 138 * Align tape buffer on page boundary to speed up tape write(). 139 */ 140 for (i = 0; i <= SLAVES; i++) { 141 buf = malloc((unsigned)(reqsiz + writesize + pgoff + TP_BSIZE)); 142 if (buf == NULL) 143 return(0); 144 slaves[i].tblock = (char (*)[TP_BSIZE]) 145 (((long)&buf[ntrec + 1] + pgoff) &~ pgoff); 146 slaves[i].req = (struct req *)slaves[i].tblock - ntrec - 1; 147 } 148 slp = &slaves[0]; 149 slp->count = 1; 150 slp->tapea = 0; 151 slp->firstrec = 0; 152 nextblock = slp->tblock; 153 return(1); 154 } 155 156 void 157 writerec(char *dp, int isspcl) 158 { 159 160 slp->req[trecno].dblk = 0; 161 slp->req[trecno].count = 1; 162 *(union u_spcl *)(*(nextblock)++) = *(union u_spcl *)dp; 163 if (isspcl) 164 lastspclrec = spcl.c_tapea; 165 trecno++; 166 spcl.c_tapea++; 167 if (trecno >= ntrec) 168 flushtape(); 169 } 170 171 void 172 dumpblock(daddr_t blkno, int size) 173 { 174 int avail, tpblks; 175 daddr_t dblkno; 176 177 dblkno = fsbtodb(sblock, blkno); 178 tpblks = size >> tp_bshift; 179 while ((avail = MINIMUM(tpblks, ntrec - trecno)) > 0) { 180 slp->req[trecno].dblk = dblkno; 181 slp->req[trecno].count = avail; 182 trecno += avail; 183 spcl.c_tapea += avail; 184 if (trecno >= ntrec) 185 flushtape(); 186 dblkno += avail << (tp_bshift - (ffs(DEV_BSIZE) - 1)); 187 tpblks -= avail; 188 } 189 } 190 191 int nogripe = 0; 192 193 /* ARGSUSED */ 194 void 195 tperror(int signo) 196 { 197 /* XXX - signal races */ 198 199 if (pipeout) { 200 msg("write error on %s\n", tape); 201 quit("Cannot recover\n"); 202 /* NOTREACHED */ 203 } 204 msg("write error %lld blocks into volume %d\n", 205 (long long)blocksthisvol, tapeno); 206 broadcast("DUMP WRITE ERROR!\n"); 207 if (!query("Do you want to restart?")) 208 dumpabort(0); 209 msg("Closing this volume. Prepare to restart with new media;\n"); 210 msg("this dump volume will be rewritten.\n"); 211 killall(); 212 nogripe = 1; 213 close_rewind(); 214 Exit(X_REWRITE); 215 } 216 217 /* ARGSUSED */ 218 void 219 sigpipe(int signo) 220 { 221 222 quit("Broken pipe\n"); 223 } 224 225 /* 226 * do_stats -- 227 * Update xferrate stats 228 */ 229 time_t 230 do_stats(void) 231 { 232 time_t tnow, ttaken; 233 int64_t blocks; 234 235 (void)time(&tnow); 236 ttaken = tnow - tstart_volume; 237 blocks = spcl.c_tapea - tapea_volume; 238 msg("Volume %d completed at: %s", tapeno, ctime(&tnow)); 239 if (ttaken > 0) { 240 msg("Volume %d took %lld:%02lld:%02lld\n", tapeno, 241 (long long)ttaken / 3600, ((long long)ttaken % 3600) / 60, 242 (long long)ttaken % 60); 243 blocks /= ttaken; 244 msg("Volume %d transfer rate: %lld KB/s\n", tapeno, blocks); 245 xferrate += blocks; 246 } 247 return(tnow); 248 } 249 250 /* 251 * statussig -- 252 * information message upon receipt of SIGINFO 253 * (derived from optr.c::timeest()) 254 * XXX not safe 255 */ 256 /* ARGSUSED */ 257 void 258 statussig(int signo) 259 { 260 time_t tnow, deltat; 261 char msgbuf[128]; 262 int save_errno = errno; 263 264 if (blockswritten < 500) 265 return; 266 (void) time(&tnow); 267 deltat = tstart_writing - tnow + (1.0 * (tnow - tstart_writing)) 268 / blockswritten * tapesize; 269 (void)snprintf(msgbuf, sizeof(msgbuf), 270 "dump: %s %3.2f%% done at %lld KB/s, finished in %d:%02d\n", 271 tape, (blockswritten * 100.0) / tapesize, 272 (spcl.c_tapea - tapea_volume) / (tnow - tstart_volume), 273 (int)(deltat / 3600), (int)((deltat % 3600) / 60)); 274 write(STDERR_FILENO, msgbuf, strlen(msgbuf)); 275 errno = save_errno; 276 } 277 278 static void 279 flushtape(void) 280 { 281 int i, blks, got; 282 int64_t lastfirstrec; 283 284 int siz = (char *)nextblock - (char *)slp->req; 285 286 slp->req[trecno].count = 0; /* Sentinel */ 287 288 if (atomic((ssize_t (*)(int, void *, size_t))write, slp->fd, 289 (char *)slp->req, siz) != siz) 290 quit("error writing command pipe: %s\n", strerror(errno)); 291 slp->sent = 1; /* we sent a request, read the response later */ 292 293 lastfirstrec = slp->firstrec; 294 295 if (++slp >= &slaves[SLAVES]) 296 slp = &slaves[0]; 297 298 /* Read results back from next slave */ 299 if (slp->sent) { 300 if (atomic(read, slp->fd, (char *)&got, sizeof(got)) 301 != sizeof(got)) { 302 perror(" DUMP: error reading command pipe in master"); 303 dumpabort(0); 304 } 305 slp->sent = 0; 306 307 /* Check for end of tape */ 308 if (got < writesize) { 309 msg("End of tape detected\n"); 310 311 /* 312 * Drain the results, don't care what the values were. 313 * If we read them here then trewind won't... 314 */ 315 for (i = 0; i < SLAVES; i++) { 316 if (slaves[i].sent) { 317 if (atomic(read, slaves[i].fd, 318 (char *)&got, sizeof(got)) 319 != sizeof(got)) { 320 perror(" DUMP: error reading command pipe in master"); 321 dumpabort(0); 322 } 323 slaves[i].sent = 0; 324 } 325 } 326 327 close_rewind(); 328 rollforward(); 329 return; 330 } 331 } 332 333 blks = 0; 334 if (spcl.c_type != TS_END) { 335 for (i = 0; i < spcl.c_count; i++) 336 if (spcl.c_addr[i] != 0) 337 blks++; 338 } 339 slp->count = lastspclrec + blks + 1 - spcl.c_tapea; 340 slp->tapea = spcl.c_tapea; 341 slp->firstrec = lastfirstrec + ntrec; 342 slp->inode = curino; 343 nextblock = slp->tblock; 344 trecno = 0; 345 asize += tenths; 346 blockswritten += ntrec; 347 blocksthisvol += ntrec; 348 if (!pipeout && !unlimited && (blocksperfile ? 349 (blocksthisvol >= blocksperfile) : (asize > tsize))) { 350 close_rewind(); 351 startnewtape(0); 352 } 353 timeest(); 354 } 355 356 void 357 trewind(void) 358 { 359 struct stat sb; 360 int f, got; 361 362 for (f = 0; f < SLAVES; f++) { 363 /* 364 * Drain the results, but unlike EOT we DO (or should) care 365 * what the return values were, since if we detect EOT after 366 * we think we've written the last blocks to the tape anyway, 367 * we have to replay those blocks with rollforward. 368 * 369 * fixme: punt for now. 370 */ 371 if (slaves[f].sent) { 372 if (atomic(read, slaves[f].fd, (char *)&got, sizeof(got)) 373 != sizeof(got)) { 374 perror(" DUMP: error reading command pipe in master"); 375 dumpabort(0); 376 } 377 slaves[f].sent = 0; 378 if (got != writesize) { 379 msg("EOT detected in last 2 tape records!\n"); 380 msg("Use a longer tape, decrease the size estimate\n"); 381 quit("or use no size estimate at all.\n"); 382 } 383 } 384 (void) close(slaves[f].fd); 385 } 386 while (wait((int *)NULL) >= 0) /* wait for any signals from slaves */ 387 /* void */; 388 389 if (pipeout) 390 return; 391 392 msg("Closing %s\n", tape); 393 394 #ifdef RDUMP 395 if (host) { 396 rmtclose(); 397 while (rmtopen(tape, O_RDONLY) < 0) 398 sleep(10); 399 rmtclose(); 400 return; 401 } 402 #endif 403 /* 404 * st(4) says: "Bit 1 of the minor number specifies whether an eject is 405 * attempted when the device is closed. When it is set, the device 406 * will attempt to eject its media on close ...". 407 * 408 * If the tape has been ejected, looping on open() will generate 'Media 409 * not present' errors until a tape is loaded. Once loaded the tape 410 * will be immediately ejected as a result of the second close(). 411 * 412 * So if the tape will be ejected, just close and return. 413 */ 414 if ((fstat(tapefd, &sb) == 0) && (minor(sb.st_rdev) & 0x02)) { 415 (void) close(tapefd); 416 return; 417 } 418 419 (void) close(tapefd); 420 while ((f = open(tape, O_RDONLY)) < 0) 421 sleep (10); 422 (void) close(f); 423 } 424 425 void 426 close_rewind(void) 427 { 428 trewind(); 429 (void)do_stats(); 430 if (nexttape) 431 return; 432 if (!nogripe) { 433 msg("Change Volumes: Mount volume #%d\n", tapeno+1); 434 broadcast("CHANGE DUMP VOLUMES!\7\7\n"); 435 } 436 while (!query("Is the new volume mounted and ready to go?")) 437 if (query("Do you want to abort?")) { 438 dumpabort(0); 439 /*NOTREACHED*/ 440 } 441 } 442 443 void 444 rollforward(void) 445 { 446 struct req *p, *q, *prev; 447 struct slave *tslp; 448 int i, size, got; 449 int64_t savedtapea; 450 union u_spcl *ntb, *otb; 451 tslp = &slaves[SLAVES]; 452 ntb = (union u_spcl *)tslp->tblock[1]; 453 454 /* 455 * Each of the N slaves should have requests that need to 456 * be replayed on the next tape. Use the extra slave buffers 457 * (slaves[SLAVES]) to construct request lists to be sent to 458 * each slave in turn. 459 */ 460 for (i = 0; i < SLAVES; i++) { 461 q = &tslp->req[1]; 462 otb = (union u_spcl *)slp->tblock; 463 464 /* 465 * For each request in the current slave, copy it to tslp. 466 */ 467 468 prev = NULL; 469 for (p = slp->req; p->count > 0; p += p->count) { 470 *q = *p; 471 if (p->dblk == 0) 472 *ntb++ = *otb++; /* copy the datablock also */ 473 prev = q; 474 q += q->count; 475 } 476 if (prev == NULL) 477 quit("rollforward: protocol botch\n"); 478 if (prev->dblk != 0) 479 prev->count -= 1; 480 else 481 ntb--; 482 q -= 1; 483 q->count = 0; 484 q = &tslp->req[0]; 485 if (i == 0) { 486 q->dblk = 0; 487 q->count = 1; 488 trecno = 0; 489 nextblock = tslp->tblock; 490 savedtapea = spcl.c_tapea; 491 spcl.c_tapea = slp->tapea; 492 startnewtape(0); 493 spcl.c_tapea = savedtapea; 494 lastspclrec = savedtapea - 1; 495 } 496 size = (char *)ntb - (char *)q; 497 if (atomic((ssize_t (*)(int, void *, size_t))write, 498 slp->fd, (char *)q, size) != size) { 499 perror(" DUMP: error writing command pipe"); 500 dumpabort(0); 501 } 502 slp->sent = 1; 503 if (++slp >= &slaves[SLAVES]) 504 slp = &slaves[0]; 505 506 q->count = 1; 507 508 if (prev->dblk != 0) { 509 /* 510 * If the last one was a disk block, make the 511 * first of this one be the last bit of that disk 512 * block... 513 */ 514 q->dblk = prev->dblk + 515 prev->count * (TP_BSIZE / DEV_BSIZE); 516 ntb = (union u_spcl *)tslp->tblock; 517 } else { 518 /* 519 * It wasn't a disk block. Copy the data to its 520 * new location in the buffer. 521 */ 522 q->dblk = 0; 523 *((union u_spcl *)tslp->tblock) = *ntb; 524 ntb = (union u_spcl *)tslp->tblock[1]; 525 } 526 } 527 slp->req[0] = *q; 528 nextblock = slp->tblock; 529 if (q->dblk == 0) 530 nextblock++; 531 trecno = 1; 532 533 /* 534 * Clear the first slaves' response. One hopes that it 535 * worked ok, otherwise the tape is much too short! 536 */ 537 if (slp->sent) { 538 if (atomic(read, slp->fd, (char *)&got, sizeof(got)) 539 != sizeof(got)) { 540 perror(" DUMP: error reading command pipe in master"); 541 dumpabort(0); 542 } 543 slp->sent = 0; 544 545 if (got != writesize) { 546 quit("EOT detected at start of the tape!\n"); 547 } 548 } 549 } 550 551 /* 552 * We implement taking and restoring checkpoints on the tape level. 553 * When each tape is opened, a new process is created by forking; this 554 * saves all of the necessary context in the parent. The child 555 * continues the dump; the parent waits around, saving the context. 556 * If the child returns X_REWRITE, then it had problems writing that tape; 557 * this causes the parent to fork again, duplicating the context, and 558 * everything continues as if nothing had happened. 559 */ 560 void 561 startnewtape(int top) 562 { 563 pid_t parentpid; 564 pid_t childpid; 565 int status; 566 pid_t waitingpid; 567 char *p; 568 sig_t interrupt_save; 569 570 interrupt_save = signal(SIGINT, SIG_IGN); 571 parentpid = getpid(); 572 tapea_volume = spcl.c_tapea; 573 (void)time(&tstart_volume); 574 575 restore_check_point: 576 (void)signal(SIGINT, interrupt_save); 577 /* 578 * All signals are inherited... 579 */ 580 childpid = fork(); 581 if (childpid < 0) { 582 msg("Context save fork fails in parent %d\n", parentpid); 583 Exit(X_ABORT); 584 } 585 if (childpid != 0) { 586 /* 587 * PARENT: 588 * save the context by waiting 589 * until the child doing all of the work returns. 590 * don't catch the interrupt 591 */ 592 signal(SIGINT, SIG_IGN); 593 #ifdef TDEBUG 594 msg("Tape: %d; parent process: %d child process %d\n", 595 tapeno+1, parentpid, childpid); 596 #endif /* TDEBUG */ 597 while ((waitingpid = wait(&status)) != childpid) 598 msg("Parent %d waiting for child %d has another child %d return\n", 599 parentpid, childpid, waitingpid); 600 if (status & 0xFF) { 601 msg("Child %d returns LOB status %o\n", 602 childpid, status&0xFF); 603 } 604 status = (status >> 8) & 0xFF; 605 #ifdef TDEBUG 606 switch(status) { 607 case X_FINOK: 608 msg("Child %d finishes X_FINOK\n", childpid); 609 break; 610 case X_ABORT: 611 msg("Child %d finishes X_ABORT\n", childpid); 612 break; 613 case X_REWRITE: 614 msg("Child %d finishes X_REWRITE\n", childpid); 615 break; 616 default: 617 msg("Child %d finishes unknown %d\n", 618 childpid, status); 619 break; 620 } 621 #endif /* TDEBUG */ 622 switch(status) { 623 case X_FINOK: 624 Exit(X_FINOK); 625 break; 626 case X_ABORT: 627 Exit(X_ABORT); 628 break; 629 case X_REWRITE: 630 goto restore_check_point; 631 default: 632 msg("Bad return code from dump: %d\n", status); 633 Exit(X_ABORT); 634 } 635 /*NOTREACHED*/ 636 } else { /* we are the child; just continue */ 637 #ifdef TDEBUG 638 sleep(4); /* allow time for parent's message to get out */ 639 msg("Child on Tape %d has parent %d, my pid = %d\n", 640 tapeno+1, parentpid, getpid()); 641 #endif /* TDEBUG */ 642 /* 643 * If we have a name like "/dev/rst0,/dev/rst1", 644 * use the name before the comma first, and save 645 * the remaining names for subsequent volumes. 646 */ 647 tapeno++; /* current tape sequence */ 648 if (nexttape || strchr(tape, ',')) { 649 if (nexttape && *nexttape) 650 tape = nexttape; 651 if ((p = strchr(tape, ',')) != NULL) { 652 *p = '\0'; 653 nexttape = p + 1; 654 } else 655 nexttape = NULL; 656 msg("Dumping volume %d on %s\n", tapeno, tape); 657 } 658 #ifdef RDUMP 659 while ((tapefd = (host ? rmtopen(tape, O_WRONLY|O_CREAT) : 660 pipeout ? 1 : open(tape, O_WRONLY|O_CREAT, 0666))) < 0) 661 #else 662 while ((tapefd = (pipeout ? 1 : 663 open(tape, O_WRONLY|O_CREAT, 0666))) < 0) 664 #endif 665 { 666 msg("Cannot open output \"%s\".\n", tape); 667 if (!query("Do you want to retry the open?")) 668 dumpabort(0); 669 } 670 671 enslave(); /* Share open tape file descriptor with slaves */ 672 673 asize = 0; 674 blocksthisvol = 0; 675 if (top) 676 newtape++; /* new tape signal */ 677 spcl.c_count = slp->count; 678 /* 679 * measure firstrec in TP_BSIZE units since restore doesn't 680 * know the correct ntrec value... 681 */ 682 spcl.c_firstrec = slp->firstrec; 683 spcl.c_volume++; 684 spcl.c_type = TS_TAPE; 685 if (sblock->fs_magic != FS_UFS2_MAGIC) 686 spcl.c_flags |= DR_NEWHEADER; 687 writeheader((ino_t)slp->inode); 688 if (sblock->fs_magic != FS_UFS2_MAGIC) 689 spcl.c_flags &=~ DR_NEWHEADER; 690 msg("Volume %d started at: %s", tapeno, ctime(&tstart_volume)); 691 if (tapeno > 1) 692 msg("Volume %d begins with blocks from inode %llu\n", 693 tapeno, (unsigned long long)slp->inode); 694 } 695 } 696 697 /* ARGSUSED */ 698 void 699 dumpabort(int signo) 700 { 701 702 if (master != 0 && master != getpid()) 703 /* Signals master to call dumpabort */ 704 (void) kill(master, SIGTERM); 705 else { 706 killall(); 707 msg("The ENTIRE dump is aborted.\n"); 708 } 709 #ifdef RDUMP 710 rmtclose(); 711 #endif 712 Exit(X_ABORT); 713 } 714 715 __dead void 716 Exit(int status) 717 { 718 719 #ifdef TDEBUG 720 msg("pid = %d exits with status %d\n", getpid(), status); 721 #endif /* TDEBUG */ 722 exit(status); 723 } 724 725 /* 726 * proceed - handler for SIGUSR2, used to synchronize IO between the slaves. 727 */ 728 /* ARGSUSED */ 729 void 730 proceed(int signo) 731 { 732 caught++; 733 } 734 735 void 736 enslave(void) 737 { 738 int cmd[2]; 739 int i, j; 740 741 master = getpid(); 742 743 signal(SIGTERM, dumpabort); /* Slave sends SIGTERM on dumpabort() */ 744 signal(SIGPIPE, sigpipe); 745 signal(SIGUSR1, tperror); /* Slave sends SIGUSR1 on tape errors */ 746 signal(SIGUSR2, proceed); /* Slave sends SIGUSR2 to next slave */ 747 748 for (i = 0; i < SLAVES; i++) { 749 if (i == slp - &slaves[0]) { 750 caught = 1; 751 } else { 752 caught = 0; 753 } 754 755 if (socketpair(AF_UNIX, SOCK_STREAM, 0, cmd) < 0 || 756 (slaves[i].pid = fork()) < 0) 757 quit("too many slaves, %d (recompile smaller): %s\n", 758 i, strerror(errno)); 759 760 slaves[i].fd = cmd[1]; 761 slaves[i].sent = 0; 762 if (slaves[i].pid == 0) { /* Slave starts up here */ 763 for (j = 0; j <= i; j++) 764 (void) close(slaves[j].fd); 765 signal(SIGINT, SIG_IGN); /* Master handles this */ 766 signal(SIGINFO, SIG_IGN); 767 doslave(cmd[0], i); 768 Exit(X_FINOK); 769 } 770 } 771 772 for (i = 0; i < SLAVES; i++) 773 (void) atomic((ssize_t (*)(int, void *, size_t))write, 774 slaves[i].fd, (char *) &slaves[(i + 1) % SLAVES].pid, 775 sizeof(slaves[0].pid)); 776 master = 0; 777 } 778 779 void 780 killall(void) 781 { 782 int i; 783 784 for (i = 0; i < SLAVES; i++) 785 if (slaves[i].pid > 0) { 786 (void) kill(slaves[i].pid, SIGKILL); 787 slaves[i].pid = 0; 788 } 789 } 790 791 /* 792 * Synchronization - each process has a lockfile, and shares file 793 * descriptors to the following process's lockfile. When our write 794 * completes, we release our lock on the following process's lock- 795 * file, allowing the following process to lock it and proceed. We 796 * get the lock back for the next cycle by swapping descriptors. 797 */ 798 static void 799 doslave(int cmd, int slave_number) 800 { 801 int nread, nextslave, size, wrote = 0, eot_count; 802 sigset_t nsigset, osigset; 803 804 /* 805 * Need our own seek pointer. 806 */ 807 (void) close(diskfd); 808 if ((diskfd = open(disk, O_RDONLY)) < 0) 809 quit("slave couldn't reopen disk: %s\n", strerror(errno)); 810 811 /* 812 * Need the pid of the next slave in the loop... 813 */ 814 if ((nread = atomic(read, cmd, (char *)&nextslave, sizeof(nextslave))) 815 != sizeof(nextslave)) { 816 quit("master/slave protocol botched - didn't get pid of next slave.\n"); 817 } 818 819 /* 820 * Get list of blocks to dump, read the blocks into tape buffer 821 */ 822 while ((nread = atomic(read, cmd, (char *)slp->req, reqsiz)) == reqsiz) { 823 struct req *p = slp->req; 824 825 for (trecno = 0; trecno < ntrec; 826 trecno += p->count, p += p->count) { 827 if (p->dblk) { 828 bread(p->dblk, slp->tblock[trecno], 829 p->count * TP_BSIZE); 830 } else { 831 if (p->count != 1 || atomic(read, cmd, 832 (char *)slp->tblock[trecno], 833 TP_BSIZE) != TP_BSIZE) 834 quit("master/slave protocol botched.\n"); 835 } 836 } 837 838 sigemptyset(&nsigset); 839 sigaddset(&nsigset, SIGUSR2); 840 sigprocmask(SIG_BLOCK, &nsigset, &osigset); 841 while (!caught) 842 sigsuspend(&osigset); 843 caught = 0; 844 sigprocmask(SIG_SETMASK, &osigset, NULL); 845 846 /* Try to write the data... */ 847 eot_count = 0; 848 size = 0; 849 850 while (eot_count < 10 && size < writesize) { 851 #ifdef RDUMP 852 if (host) 853 wrote = rmtwrite(slp->tblock[0]+size, 854 writesize-size); 855 else 856 #endif 857 wrote = write(tapefd, slp->tblock[0]+size, 858 writesize-size); 859 #ifdef WRITEDEBUG 860 printf("slave %d wrote %d\n", slave_number, wrote); 861 #endif 862 if (wrote < 0) 863 break; 864 if (wrote == 0) 865 eot_count++; 866 size += wrote; 867 } 868 869 #ifdef WRITEDEBUG 870 if (size != writesize) 871 printf("slave %d only wrote %d out of %d bytes and gave up.\n", 872 slave_number, size, writesize); 873 #endif 874 875 if (eot_count > 0) 876 size = 0; 877 878 /* 879 * Handle ENOSPC as an EOT condition 880 */ 881 if (wrote < 0 && errno == ENOSPC) { 882 wrote = 0; 883 eot_count++; 884 } 885 886 if (size < 0) { 887 (void) kill(master, SIGUSR1); 888 sigemptyset(&nsigset); 889 for (;;) 890 sigsuspend(&nsigset); 891 } else { 892 /* 893 * pass size of write back to master 894 * (for EOT handling) 895 */ 896 (void) atomic((ssize_t (*)(int, void *, size_t))write, 897 cmd, (char *)&size, sizeof(size)); 898 } 899 900 /* 901 * If partial write, don't want next slave to go. 902 * Also jolts him awake. 903 */ 904 (void) kill(nextslave, SIGUSR2); 905 } 906 if (nread != 0) 907 quit("error reading command pipe: %s\n", strerror(errno)); 908 } 909 910 /* 911 * Since a read from a pipe may not return all we asked for, 912 * or a write may not write all we ask if we get a signal, 913 * loop until the count is satisfied (or error). 914 */ 915 static ssize_t 916 atomic(ssize_t (*func)(int, void *, size_t), int fd, char *buf, int count) 917 { 918 ssize_t got, need = count; 919 920 while ((got = (*func)(fd, buf, need)) > 0 && (need -= got) > 0) 921 buf += got; 922 return (got < 0 ? got : count - need); 923 } 924