1 /* $NetBSD: tape.c,v 1.38 2002/08/18 08:03:35 yamt Exp $ */ 2 3 /*- 4 * Copyright (c) 1980, 1991, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. All advertising materials mentioning features or use of this software 16 * must display the following acknowledgement: 17 * This product includes software developed by the University of 18 * California, Berkeley and its contributors. 19 * 4. Neither the name of the University nor the names of its contributors 20 * may be used to endorse or promote products derived from this software 21 * without specific prior written permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 33 * SUCH DAMAGE. 34 */ 35 36 #include <sys/cdefs.h> 37 #ifndef lint 38 #if 0 39 static char sccsid[] = "@(#)tape.c 8.4 (Berkeley) 5/1/95"; 40 #else 41 __RCSID("$NetBSD: tape.c,v 1.38 2002/08/18 08:03:35 yamt Exp $"); 42 #endif 43 #endif /* not lint */ 44 45 #include <sys/param.h> 46 #include <sys/socket.h> 47 #include <sys/time.h> 48 #include <sys/wait.h> 49 #include <ufs/ufs/dinode.h> 50 #include <sys/ioctl.h> 51 #include <sys/mtio.h> 52 53 #include <protocols/dumprestore.h> 54 55 #include <errno.h> 56 #include <fcntl.h> 57 #include <setjmp.h> 58 #include <signal.h> 59 #include <stdio.h> 60 #include <stdlib.h> 61 #include <string.h> 62 #include <time.h> 63 #include <unistd.h> 64 65 #include "dump.h" 66 #include "pathnames.h" 67 68 int writesize; /* size of malloc()ed buffer for tape */ 69 long lastspclrec = -1; /* tape block number of last written header */ 70 int trecno = 0; /* next record to write in current block */ 71 extern long blocksperfile; /* number of blocks per output file */ 72 long blocksthisvol; /* number of blocks on current output file */ 73 extern int ntrec; /* blocking factor on tape */ 74 extern int cartridge; 75 extern char *host; 76 char *nexttape; 77 78 static ssize_t atomic_read(int, char *, int); 79 static ssize_t atomic_write(int, char *, int); 80 static void doslave(int, int); 81 static void enslave(void); 82 static void flushtape(void); 83 static void killall(void); 84 static void proceed(int); 85 static void rollforward(void); 86 static void sigpipe(int); 87 static void tperror(int); 88 89 /* 90 * Concurrent dump mods (Caltech) - disk block reading and tape writing 91 * are exported to several slave processes. While one slave writes the 92 * tape, the others read disk blocks; they pass control of the tape in 93 * a ring via signals. The parent process traverses the file system and 94 * sends writeheader()'s and lists of daddr's to the slaves via pipes. 95 * The following structure defines the instruction packets sent to slaves. 96 */ 97 struct req { 98 daddr_t dblk; 99 int count; 100 }; 101 int reqsiz; 102 103 #define SLAVES 3 /* 1 slave writing, 1 reading, 1 for slack */ 104 struct slave { 105 int tapea; /* header number at start of this chunk */ 106 int count; /* count to next header (used for TS_TAPE */ 107 /* after EOT) */ 108 int inode; /* inode that we are currently dealing with */ 109 int fd; /* FD for this slave */ 110 int pid; /* PID for this slave */ 111 int sent; /* 1 == we've sent this slave requests */ 112 int firstrec; /* record number of this block */ 113 char (*tblock)[TP_BSIZE]; /* buffer for data blocks */ 114 struct req *req; /* buffer for requests */ 115 } slaves[SLAVES+1]; 116 struct slave *slp; 117 118 char (*nextblock)[TP_BSIZE]; 119 120 static time_t tstart_volume; /* time of volume start */ 121 static int tapea_volume; /* value of spcl.c_tapea at volume start */ 122 123 int master; /* pid of master, for sending error signals */ 124 int tenths; /* length of tape used per block written */ 125 static int caught; /* have we caught the signal to proceed? */ 126 static int ready; /* have we reached the lock point without having */ 127 /* received the SIGUSR2 signal from the prev slave? */ 128 static jmp_buf jmpbuf; /* where to jump to if we are ready when the */ 129 /* SIGUSR2 arrives from the previous slave */ 130 131 int 132 alloctape(void) 133 { 134 int pgoff = getpagesize() - 1; 135 char *buf; 136 int i; 137 138 writesize = ntrec * TP_BSIZE; 139 reqsiz = (ntrec + 1) * sizeof(struct req); 140 /* 141 * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode 142 * (see DEC TU80 User's Guide). The shorter gaps of 6250-bpi require 143 * repositioning after stopping, i.e, streaming mode, where the gap is 144 * variable, 0.30" to 0.45". The gap is maximal when the tape stops. 145 */ 146 if (blocksperfile == 0 && !unlimited) 147 tenths = writesize / density + 148 (cartridge ? 16 : density == 625 ? 5 : 8); 149 /* 150 * Allocate tape buffer contiguous with the array of instruction 151 * packets, so flushtape() can write them together with one write(). 152 * Align tape buffer on page boundary to speed up tape write(). 153 */ 154 for (i = 0; i <= SLAVES; i++) { 155 buf = (char *) 156 xmalloc((unsigned)(reqsiz + writesize + pgoff + TP_BSIZE)); 157 slaves[i].tblock = (char (*)[TP_BSIZE]) 158 (((long)&buf[ntrec + 1] + pgoff) &~ pgoff); 159 slaves[i].req = (struct req *)slaves[i].tblock - ntrec - 1; 160 } 161 slp = &slaves[0]; 162 slp->count = 1; 163 slp->tapea = 0; 164 slp->firstrec = 0; 165 nextblock = slp->tblock; 166 return(1); 167 } 168 169 void 170 writerec(char *dp, int isspcl) 171 { 172 173 slp->req[trecno].dblk = (daddr_t)0; 174 slp->req[trecno].count = 1; 175 *(union u_spcl *)(*(nextblock)++) = *(union u_spcl *)dp; 176 if (isspcl) 177 lastspclrec = iswap32(spcl.c_tapea); 178 trecno++; 179 spcl.c_tapea = iswap32(iswap32(spcl.c_tapea) +1); 180 if (trecno >= ntrec) 181 flushtape(); 182 } 183 184 void 185 dumpblock(daddr_t blkno, int size) 186 { 187 int avail, tpblks, dblkno; 188 189 dblkno = fsatoda(ufsib, blkno); 190 tpblks = size >> tp_bshift; 191 while ((avail = MIN(tpblks, ntrec - trecno)) > 0) { 192 slp->req[trecno].dblk = dblkno; 193 slp->req[trecno].count = avail; 194 trecno += avail; 195 spcl.c_tapea = iswap32(iswap32(spcl.c_tapea) + avail); 196 if (trecno >= ntrec) 197 flushtape(); 198 dblkno += avail << (tp_bshift - dev_bshift); 199 tpblks -= avail; 200 } 201 } 202 203 int nogripe = 0; 204 205 static void 206 tperror(int signo) 207 { 208 209 if (pipeout) { 210 msg("write error on %s\n", tape); 211 quit("Cannot recover\n"); 212 /* NOTREACHED */ 213 } 214 msg("write error %ld blocks into volume %d\n", blocksthisvol, tapeno); 215 broadcast("DUMP WRITE ERROR!\n"); 216 if (!query("Do you want to restart?")) 217 dumpabort(0); 218 msg("Closing this volume. Prepare to restart with new media;\n"); 219 msg("this dump volume will be rewritten.\n"); 220 killall(); 221 nogripe = 1; 222 close_rewind(); 223 Exit(X_REWRITE); 224 } 225 226 static void 227 sigpipe(int signo) 228 { 229 230 quit("Broken pipe\n"); 231 } 232 233 /* 234 * do_stats -- 235 * Update xferrate stats 236 */ 237 time_t 238 do_stats(void) 239 { 240 time_t tnow, ttaken; 241 int blocks; 242 243 (void)time(&tnow); 244 ttaken = tnow - tstart_volume; 245 blocks = iswap32(spcl.c_tapea) - tapea_volume; 246 msg("Volume %d completed at: %s", tapeno, ctime(&tnow)); 247 if (ttaken > 0) { 248 msg("Volume %d took %d:%02d:%02d\n", tapeno, 249 (int) (ttaken / 3600), (int) ((ttaken % 3600) / 60), 250 (int) (ttaken % 60)); 251 msg("Volume %d transfer rate: %d KB/s\n", tapeno, 252 (int) (blocks / ttaken)); 253 xferrate += blocks / ttaken; 254 } 255 return(tnow); 256 } 257 258 /* 259 * statussig -- 260 * information message upon receipt of SIGINFO 261 * (derived from optr.c::timeest()) 262 */ 263 void 264 statussig(int notused) 265 { 266 time_t tnow, deltat; 267 char msgbuf[128]; 268 int errno_save; 269 270 if (blockswritten < 500) 271 return; 272 errno_save = errno; 273 (void) time((time_t *) &tnow); 274 if (tnow <= tstart_volume) 275 return; 276 deltat = tstart_writing - tnow + 277 (1.0 * (tnow - tstart_writing)) / blockswritten * tapesize; 278 (void)snprintf(msgbuf, sizeof(msgbuf), 279 "%3.2f%% done at %ld KB/s, finished in %d:%02d\n", 280 (blockswritten * 100.0) / tapesize, 281 (long)((iswap32(spcl.c_tapea) - tapea_volume) / 282 (tnow - tstart_volume)), 283 (int)(deltat / 3600), (int)((deltat % 3600) / 60)); 284 write(STDERR_FILENO, msgbuf, strlen(msgbuf)); 285 errno = errno_save; 286 } 287 288 static void 289 flushtape(void) 290 { 291 int i, blks, got; 292 long lastfirstrec; 293 294 int siz = (char *)nextblock - (char *)slp->req; 295 296 slp->req[trecno].count = 0; /* Sentinel */ 297 298 if (atomic_write(slp->fd, (char *)slp->req, siz) != siz) 299 quit("error writing command pipe: %s\n", strerror(errno)); 300 slp->sent = 1; /* we sent a request, read the response later */ 301 302 lastfirstrec = slp->firstrec; 303 304 if (++slp >= &slaves[SLAVES]) 305 slp = &slaves[0]; 306 307 /* Read results back from next slave */ 308 if (slp->sent) { 309 if (atomic_read(slp->fd, (char *)&got, sizeof got) 310 != sizeof got) { 311 perror(" DUMP: error reading command pipe in master"); 312 dumpabort(0); 313 } 314 slp->sent = 0; 315 316 /* Check for end of tape */ 317 if (got < writesize) { 318 msg("End of tape detected\n"); 319 320 /* 321 * Drain the results, don't care what the values were. 322 * If we read them here then trewind won't... 323 */ 324 for (i = 0; i < SLAVES; i++) { 325 if (slaves[i].sent) { 326 if (atomic_read(slaves[i].fd, 327 (char *)&got, sizeof got) 328 != sizeof got) { 329 perror(" DUMP: error reading command pipe in master"); 330 dumpabort(0); 331 } 332 slaves[i].sent = 0; 333 } 334 } 335 336 close_rewind(); 337 rollforward(); 338 return; 339 } 340 } 341 342 blks = 0; 343 if (iswap32(spcl.c_type) != TS_END) { 344 for (i = 0; i < iswap32(spcl.c_count); i++) 345 if (spcl.c_addr[i] != 0) 346 blks++; 347 } 348 slp->count = lastspclrec + blks + 1 - iswap32(spcl.c_tapea); 349 slp->tapea = iswap32(spcl.c_tapea); 350 slp->firstrec = lastfirstrec + ntrec; 351 slp->inode = curino; 352 nextblock = slp->tblock; 353 trecno = 0; 354 asize += tenths; 355 blockswritten += ntrec; 356 blocksthisvol += ntrec; 357 if (!pipeout && !unlimited && (blocksperfile ? 358 (blocksthisvol >= blocksperfile) : (asize > tsize))) { 359 close_rewind(); 360 startnewtape(0); 361 } 362 timeest(); 363 } 364 365 void 366 trewind(int eject) 367 { 368 int f; 369 int got; 370 371 for (f = 0; f < SLAVES; f++) { 372 /* 373 * Drain the results, but unlike EOT we DO (or should) care 374 * what the return values were, since if we detect EOT after 375 * we think we've written the last blocks to the tape anyway, 376 * we have to replay those blocks with rollforward. 377 * 378 * fixme: punt for now. 379 */ 380 if (slaves[f].sent) { 381 if (atomic_read(slaves[f].fd, (char *)&got, sizeof got) 382 != sizeof got) { 383 perror(" DUMP: error reading command pipe in master"); 384 dumpabort(0); 385 } 386 slaves[f].sent = 0; 387 if (got != writesize) { 388 msg("EOT detected in last 2 tape records!\n"); 389 msg("Use a longer tape, decrease the size estimate\n"); 390 quit("or use no size estimate at all.\n"); 391 } 392 } 393 (void) close(slaves[f].fd); 394 } 395 while (wait((int *)NULL) >= 0) /* wait for any signals from slaves */ 396 /* void */; 397 398 if (pipeout) 399 return; 400 401 msg("Closing %s\n", tape); 402 403 #ifdef RDUMP 404 if (host) { 405 rmtclose(); 406 while (rmtopen(tape, 0, 0) < 0) 407 sleep(10); 408 if (eflag && eject) { 409 msg("Ejecting %s\n", tape); 410 (void) rmtioctl(MTOFFL, 0); 411 } 412 rmtclose(); 413 return; 414 } 415 #endif 416 (void) close(tapefd); 417 while ((f = open(tape, 0)) < 0) 418 sleep (10); 419 if (eflag && eject) { 420 struct mtop offl; 421 422 msg("Ejecting %s\n", tape); 423 offl.mt_op = MTOFFL; 424 offl.mt_count = 0; 425 (void) ioctl(f, MTIOCTOP, &offl); 426 } 427 (void) close(f); 428 } 429 430 void 431 close_rewind(void) 432 { 433 int i, f; 434 435 trewind(1); 436 (void)do_stats(); 437 if (nexttape) 438 return; 439 if (!nogripe) { 440 msg("Change Volumes: Mount volume #%d\n", tapeno+1); 441 broadcast("CHANGE DUMP VOLUMES!\a\a\n"); 442 } 443 if (lflag) { 444 for (i = 0; i < lflag / 10; i++) { /* wait lflag seconds */ 445 if (host) { 446 if (rmtopen(tape, 0, 0) >= 0) { 447 rmtclose(); 448 return; 449 } 450 } else { 451 if ((f = open(tape, 0)) >= 0) { 452 close(f); 453 return; 454 } 455 } 456 sleep (10); 457 } 458 } 459 460 while (!query("Is the new volume mounted and ready to go?")) 461 if (query("Do you want to abort?")) { 462 dumpabort(0); 463 /*NOTREACHED*/ 464 } 465 } 466 467 void 468 rollforward(void) 469 { 470 struct req *p, *q, *prev; 471 struct slave *tslp; 472 int i, size, savedtapea, got; 473 union u_spcl *ntb, *otb; 474 tslp = &slaves[SLAVES]; 475 ntb = (union u_spcl *)tslp->tblock[1]; 476 477 /* 478 * Each of the N slaves should have requests that need to 479 * be replayed on the next tape. Use the extra slave buffers 480 * (slaves[SLAVES]) to construct request lists to be sent to 481 * each slave in turn. 482 */ 483 for (i = 0; i < SLAVES; i++) { 484 q = &tslp->req[1]; 485 otb = (union u_spcl *)slp->tblock; 486 487 /* 488 * For each request in the current slave, copy it to tslp. 489 */ 490 491 prev = NULL; 492 for (p = slp->req; p->count > 0; p += p->count) { 493 *q = *p; 494 if (p->dblk == 0) 495 *ntb++ = *otb++; /* copy the datablock also */ 496 prev = q; 497 q += q->count; 498 } 499 if (prev == NULL) 500 quit("rollforward: protocol botch"); 501 if (prev->dblk != 0) 502 prev->count -= 1; 503 else 504 ntb--; 505 q -= 1; 506 q->count = 0; 507 q = &tslp->req[0]; 508 if (i == 0) { 509 q->dblk = 0; 510 q->count = 1; 511 trecno = 0; 512 nextblock = tslp->tblock; 513 savedtapea = iswap32(spcl.c_tapea); 514 spcl.c_tapea = iswap32(slp->tapea); 515 startnewtape(0); 516 spcl.c_tapea = iswap32(savedtapea); 517 lastspclrec = savedtapea - 1; 518 } 519 size = (char *)ntb - (char *)q; 520 if (atomic_write(slp->fd, (char *)q, size) != size) { 521 perror(" DUMP: error writing command pipe"); 522 dumpabort(0); 523 } 524 slp->sent = 1; 525 if (++slp >= &slaves[SLAVES]) 526 slp = &slaves[0]; 527 528 q->count = 1; 529 530 if (prev->dblk != 0) { 531 /* 532 * If the last one was a disk block, make the 533 * first of this one be the last bit of that disk 534 * block... 535 */ 536 q->dblk = prev->dblk + 537 prev->count * (TP_BSIZE / DEV_BSIZE); 538 ntb = (union u_spcl *)tslp->tblock; 539 } else { 540 /* 541 * It wasn't a disk block. Copy the data to its 542 * new location in the buffer. 543 */ 544 q->dblk = 0; 545 *((union u_spcl *)tslp->tblock) = *ntb; 546 ntb = (union u_spcl *)tslp->tblock[1]; 547 } 548 } 549 slp->req[0] = *q; 550 nextblock = slp->tblock; 551 if (q->dblk == 0) 552 nextblock++; 553 trecno = 1; 554 555 /* 556 * Clear the first slaves' response. One hopes that it 557 * worked ok, otherwise the tape is much too short! 558 */ 559 if (slp->sent) { 560 if (atomic_read(slp->fd, (char *)&got, sizeof got) 561 != sizeof got) { 562 perror(" DUMP: error reading command pipe in master"); 563 dumpabort(0); 564 } 565 slp->sent = 0; 566 567 if (got != writesize) { 568 quit("EOT detected at start of the tape!\n"); 569 } 570 } 571 } 572 573 /* 574 * We implement taking and restoring checkpoints on the tape level. 575 * When each tape is opened, a new process is created by forking; this 576 * saves all of the necessary context in the parent. The child 577 * continues the dump; the parent waits around, saving the context. 578 * If the child returns X_REWRITE, then it had problems writing that tape; 579 * this causes the parent to fork again, duplicating the context, and 580 * everything continues as if nothing had happened. 581 */ 582 void 583 startnewtape(int top) 584 { 585 int parentpid; 586 int childpid; 587 int status; 588 int waitforpid; 589 char *p; 590 sig_t interrupt_save; 591 592 interrupt_save = signal(SIGINT, SIG_IGN); 593 parentpid = getpid(); 594 tapea_volume = iswap32(spcl.c_tapea); 595 (void)time(&tstart_volume); 596 597 restore_check_point: 598 (void)signal(SIGINT, interrupt_save); 599 /* 600 * All signals are inherited... 601 */ 602 childpid = fork(); 603 if (childpid < 0) { 604 msg("Context save fork fails in parent %d\n", parentpid); 605 Exit(X_ABORT); 606 } 607 if (childpid != 0) { 608 /* 609 * PARENT: 610 * save the context by waiting 611 * until the child doing all of the work returns. 612 * don't catch the interrupt 613 */ 614 signal(SIGINT, SIG_IGN); 615 signal(SIGINFO, SIG_IGN); /* only want child's stats */ 616 #ifdef TDEBUG 617 msg("Tape: %d; parent process: %d child process %d\n", 618 tapeno+1, parentpid, childpid); 619 #endif /* TDEBUG */ 620 while ((waitforpid = wait(&status)) != childpid) 621 msg("Parent %d waiting for child %d has another child %d return\n", 622 parentpid, childpid, waitforpid); 623 if (status & 0xFF) { 624 msg("Child %d returns LOB status %o\n", 625 childpid, status&0xFF); 626 } 627 status = (status >> 8) & 0xFF; 628 #ifdef TDEBUG 629 switch(status) { 630 case X_FINOK: 631 msg("Child %d finishes X_FINOK\n", childpid); 632 break; 633 case X_ABORT: 634 msg("Child %d finishes X_ABORT\n", childpid); 635 break; 636 case X_REWRITE: 637 msg("Child %d finishes X_REWRITE\n", childpid); 638 break; 639 default: 640 msg("Child %d finishes unknown %d\n", 641 childpid, status); 642 break; 643 } 644 #endif /* TDEBUG */ 645 switch(status) { 646 case X_FINOK: 647 Exit(X_FINOK); 648 case X_ABORT: 649 Exit(X_ABORT); 650 case X_REWRITE: 651 goto restore_check_point; 652 default: 653 msg("Bad return code from dump: %d\n", status); 654 Exit(X_ABORT); 655 } 656 /*NOTREACHED*/ 657 } else { /* we are the child; just continue */ 658 signal(SIGINFO, statussig); /* now want child's stats */ 659 #ifdef TDEBUG 660 sleep(4); /* allow time for parent's message to get out */ 661 msg("Child on Tape %d has parent %d, my pid = %d\n", 662 tapeno+1, parentpid, getpid()); 663 #endif /* TDEBUG */ 664 /* 665 * If we have a name like "/dev/rst0,/dev/rst1", 666 * use the name before the comma first, and save 667 * the remaining names for subsequent volumes. 668 */ 669 tapeno++; /* current tape sequence */ 670 if (nexttape || strchr(tape, ',')) { 671 if (nexttape && *nexttape) 672 tape = nexttape; 673 if ((p = strchr(tape, ',')) != NULL) { 674 *p = '\0'; 675 nexttape = p + 1; 676 } else 677 nexttape = NULL; 678 msg("Dumping volume %d on %s\n", tapeno, tape); 679 } 680 #ifdef RDUMP 681 while ((tapefd = (host ? rmtopen(tape, 2, 1) : 682 pipeout ? 1 : open(tape, O_WRONLY|O_CREAT, 0666))) < 0) 683 #else 684 while ((tapefd = (pipeout ? 1 : 685 open(tape, O_WRONLY|O_CREAT, 0666))) < 0) 686 #endif 687 { 688 msg("Cannot open output \"%s\".\n", tape); 689 if (!query("Do you want to retry the open?")) 690 dumpabort(0); 691 } 692 693 enslave(); /* Share open tape file descriptor with slaves */ 694 695 asize = 0; 696 blocksthisvol = 0; 697 if (top) 698 newtape++; /* new tape signal */ 699 spcl.c_count = iswap32(slp->count); 700 /* 701 * measure firstrec in TP_BSIZE units since restore doesn't 702 * know the correct ntrec value... 703 */ 704 spcl.c_firstrec = iswap32(slp->firstrec); 705 spcl.c_volume = iswap32(iswap32(spcl.c_volume) + 1); 706 spcl.c_type = iswap32(TS_TAPE); 707 spcl.c_flags = iswap32(iswap32(spcl.c_flags) | DR_NEWHEADER); 708 writeheader((ino_t)slp->inode); 709 spcl.c_flags = iswap32(iswap32(spcl.c_flags) & ~ DR_NEWHEADER); 710 msg("Volume %d started at: %s", tapeno, ctime(&tstart_volume)); 711 if (tapeno > 1) 712 msg("Volume %d begins with blocks from inode %d\n", 713 tapeno, slp->inode); 714 } 715 } 716 717 void 718 dumpabort(int signo) 719 { 720 721 if (master != 0 && master != getpid()) 722 /* Signals master to call dumpabort */ 723 (void) kill(master, SIGTERM); 724 else { 725 killall(); 726 msg("The ENTIRE dump is aborted.\n"); 727 } 728 #ifdef RDUMP 729 rmtclose(); 730 #endif 731 Exit(X_ABORT); 732 } 733 734 void 735 Exit(int status) 736 { 737 738 #ifdef TDEBUG 739 msg("pid = %d exits with status %d\n", getpid(), status); 740 #endif /* TDEBUG */ 741 exit(status); 742 } 743 744 /* 745 * proceed - handler for SIGUSR2, used to synchronize IO between the slaves. 746 */ 747 static void 748 proceed(int signo) 749 { 750 751 if (ready) 752 longjmp(jmpbuf, 1); 753 caught++; 754 } 755 756 void 757 enslave(void) 758 { 759 int cmd[2]; 760 int i, j; 761 762 master = getpid(); 763 764 signal(SIGTERM, dumpabort); /* Slave sends SIGTERM on dumpabort() */ 765 signal(SIGPIPE, sigpipe); 766 signal(SIGUSR1, tperror); /* Slave sends SIGUSR1 on tape errors */ 767 signal(SIGUSR2, proceed); /* Slave sends SIGUSR2 to next slave */ 768 769 for (i = 0; i < SLAVES; i++) { 770 if (i == slp - &slaves[0]) { 771 caught = 1; 772 } else { 773 caught = 0; 774 } 775 776 if (socketpair(AF_LOCAL, SOCK_STREAM, 0, cmd) < 0 || 777 (slaves[i].pid = fork()) < 0) 778 quit("too many slaves, %d (recompile smaller): %s\n", 779 i, strerror(errno)); 780 781 slaves[i].fd = cmd[1]; 782 slaves[i].sent = 0; 783 if (slaves[i].pid == 0) { /* Slave starts up here */ 784 for (j = 0; j <= i; j++) 785 (void) close(slaves[j].fd); 786 signal(SIGINT, SIG_IGN); /* Master handles this */ 787 signal(SIGINFO, SIG_IGN); 788 doslave(cmd[0], i); 789 Exit(X_FINOK); 790 } 791 } 792 793 for (i = 0; i < SLAVES; i++) 794 (void) atomic_write(slaves[i].fd, 795 (char *) &slaves[(i + 1) % SLAVES].pid, 796 sizeof slaves[0].pid); 797 798 master = 0; 799 } 800 801 void 802 killall(void) 803 { 804 int i; 805 806 for (i = 0; i < SLAVES; i++) 807 if (slaves[i].pid > 0) { 808 (void) kill(slaves[i].pid, SIGKILL); 809 slaves[i].sent = 0; 810 } 811 } 812 813 /* 814 * Synchronization - each process has a lockfile, and shares file 815 * descriptors to the following process's lockfile. When our write 816 * completes, we release our lock on the following process's lock- 817 * file, allowing the following process to lock it and proceed. We 818 * get the lock back for the next cycle by swapping descriptors. 819 */ 820 static void 821 doslave(int cmd, int slave_number) 822 { 823 int nread, nextslave, size, wrote, eot_count, werror; 824 sigset_t sigset; 825 826 /* 827 * Need our own seek pointer. 828 */ 829 (void) close(diskfd); 830 if ((diskfd = open(disk, O_RDONLY)) < 0) 831 quit("slave couldn't reopen disk: %s\n", strerror(errno)); 832 833 /* 834 * Need the pid of the next slave in the loop... 835 */ 836 if ((nread = atomic_read(cmd, (char *)&nextslave, sizeof nextslave)) 837 != sizeof nextslave) { 838 quit("master/slave protocol botched - didn't get pid of next slave.\n"); 839 } 840 841 /* 842 * Get list of blocks to dump, read the blocks into tape buffer 843 */ 844 while ((nread = atomic_read(cmd, (char *)slp->req, reqsiz)) == reqsiz) { 845 struct req *p = slp->req; 846 847 for (trecno = 0; trecno < ntrec; 848 trecno += p->count, p += p->count) { 849 if (p->dblk) { 850 bread(p->dblk, slp->tblock[trecno], 851 p->count * TP_BSIZE); 852 } else { 853 if (p->count != 1 || atomic_read(cmd, 854 (char *)slp->tblock[trecno], 855 TP_BSIZE) != TP_BSIZE) 856 quit("master/slave protocol botched.\n"); 857 } 858 } 859 if (setjmp(jmpbuf) == 0) { 860 ready = 1; 861 if (!caught) 862 (void) pause(); 863 } 864 ready = 0; 865 caught = 0; 866 867 /* Try to write the data... */ 868 eot_count = 0; 869 size = 0; 870 werror = 0; 871 872 while (eot_count < 10 && size < writesize) { 873 #ifdef RDUMP 874 if (host) 875 wrote = rmtwrite(slp->tblock[0]+size, 876 writesize-size); 877 else 878 #endif 879 wrote = write(tapefd, slp->tblock[0]+size, 880 writesize-size); 881 werror = errno; 882 #ifdef WRITEDEBUG 883 fprintf(stderr, "slave %d wrote %d werror %d\n", 884 slave_number, wrote, werror); 885 #endif 886 if (wrote < 0) 887 break; 888 if (wrote == 0) 889 eot_count++; 890 size += wrote; 891 } 892 893 #ifdef WRITEDEBUG 894 if (size != writesize) 895 fprintf(stderr, 896 "slave %d only wrote %d out of %d bytes and gave up.\n", 897 slave_number, size, writesize); 898 #endif 899 900 /* 901 * Handle ENOSPC as an EOT condition. 902 */ 903 if (wrote < 0 && werror == ENOSPC) { 904 wrote = 0; 905 eot_count++; 906 } 907 908 if (eot_count > 0) 909 size = 0; 910 911 if (wrote < 0) { 912 (void) kill(master, SIGUSR1); 913 sigemptyset(&sigset); 914 for (;;) 915 sigsuspend(&sigset); 916 } else { 917 /* 918 * pass size of write back to master 919 * (for EOT handling) 920 */ 921 (void) atomic_write(cmd, (char *)&size, sizeof size); 922 } 923 924 /* 925 * If partial write, don't want next slave to go. 926 * Also jolts him awake. 927 */ 928 (void) kill(nextslave, SIGUSR2); 929 } 930 printcachestats(); 931 if (nread != 0) 932 quit("error reading command pipe: %s\n", strerror(errno)); 933 } 934 935 /* 936 * Since a read from a pipe may not return all we asked for, 937 * loop until the count is satisfied (or error). 938 */ 939 static ssize_t 940 atomic_read(int fd, char *buf, int count) 941 { 942 ssize_t got, need = count; 943 944 while ((got = read(fd, buf, need)) > 0 && (need -= got) > 0) 945 buf += got; 946 return (got < 0 ? got : count - need); 947 } 948 949 /* 950 * Since a write may not write all we ask if we get a signal, 951 * loop until the count is satisfied (or error). 952 */ 953 static ssize_t 954 atomic_write(int fd, char *buf, int count) 955 { 956 ssize_t got, need = count; 957 958 while ((got = write(fd, buf, need)) > 0 && (need -= got) > 0) 959 buf += got; 960 return (got < 0 ? got : count - need); 961 } 962