| 1 | #ifndef lint |
| 2 | static char *sccsid = "@(#)tape.c 1.10 (Berkeley) %G%"; |
| 3 | #endif |
| 4 | |
| 5 | #include "dump.h" |
| 6 | #include <signal.h> |
| 7 | |
| 8 | char (*tblock)[TP_BSIZE]; /* Pointer to malloc()ed buffer for tape */ |
| 9 | int writesize; /* Size of malloc()ed buffer for tape */ |
| 10 | int trecno = 0; |
| 11 | extern int ntrec; /* blocking factor on tape */ |
| 12 | |
| 13 | /* |
| 14 | * Streaming dump mods (Caltech) - disk block reading and tape writing |
| 15 | * are exported to several slave processes. While one slave writes the |
| 16 | * tape, the others read disk blocks; they pass control of the tape in |
| 17 | * a ring via pipes. The parent process traverses the filesystem and |
| 18 | * sends daddr's, inode records, etc, through pipes to each slave. |
| 19 | * Speed from Eagle to TU77 on VAX/780 is about 140 Kbytes/second. |
| 20 | * #ifdef RDUMP version is CPU-limited to about 40 Kbytes/second. |
| 21 | */ |
| 22 | struct req { /* instruction packets sent to slaves */ |
| 23 | daddr_t dblk; |
| 24 | int count; |
| 25 | } *req; |
| 26 | int reqsiz; |
| 27 | |
| 28 | #define SLAVES 3 /* 2 slaves read disk while 3rd writes tape */ |
| 29 | #define LAG 2 /* Write behind by LAG tape blocks (rdump) */ |
| 30 | int slavefd[SLAVES]; /* Pipes from master to each slave */ |
| 31 | int rotor; /* Current slave number */ |
| 32 | int master; /* Pid of master, for sending error signals */ |
| 33 | int trace = 0; /* Protocol trace; easily patchable with adb */ |
| 34 | #define tmsg if (trace) msg |
| 35 | |
| 36 | #ifdef RDUMP |
| 37 | extern int rmtape; |
| 38 | #endif |
| 39 | |
| 40 | /* |
| 41 | * Allocate tape buffer contiguous with the array of instruction packets, |
| 42 | * so they can be written with a single write call in flusht(). |
| 43 | */ |
| 44 | alloctape() |
| 45 | { |
| 46 | |
| 47 | writesize = ntrec * TP_BSIZE; |
| 48 | reqsiz = ntrec * sizeof(struct req); |
| 49 | req = (struct req *)malloc(reqsiz+writesize); /* array of packets */ |
| 50 | tblock = (char (*)[TP_BSIZE]) &req[ntrec]; /* Tape buffer */ |
| 51 | return (req != NULL); |
| 52 | } |
| 53 | |
| 54 | /* |
| 55 | * Send special record to be put on tape |
| 56 | */ |
| 57 | taprec(dp) |
| 58 | char *dp; |
| 59 | { |
| 60 | |
| 61 | tmsg("taprec %d\n", trecno); |
| 62 | req[trecno].dblk = (daddr_t)0; |
| 63 | req[trecno].count = 1; |
| 64 | *(union u_spcl *)(*tblock++) = *(union u_spcl *)dp; |
| 65 | spcl.c_tapea++; |
| 66 | if (++trecno >= ntrec) |
| 67 | flusht(); |
| 68 | } |
| 69 | |
| 70 | dmpblk(blkno, size) |
| 71 | daddr_t blkno; |
| 72 | int size; |
| 73 | { |
| 74 | int tpblks, dblkno; |
| 75 | register int avail; |
| 76 | |
| 77 | if (size % TP_BSIZE != 0) |
| 78 | msg("bad size to dmpblk: %d\n", size); |
| 79 | dblkno = fsbtodb(sblock, blkno); |
| 80 | tpblks = size / TP_BSIZE; |
| 81 | while ((avail = MIN(tpblks, ntrec - trecno)) > 0) { |
| 82 | tmsg("dmpblk %d\n", avail); |
| 83 | req[trecno].dblk = dblkno; |
| 84 | req[trecno].count = avail; |
| 85 | trecno += avail; |
| 86 | spcl.c_tapea += avail; |
| 87 | if (trecno >= ntrec) |
| 88 | flusht(); |
| 89 | dblkno += avail * (TP_BSIZE / DEV_BSIZE); |
| 90 | tpblks -= avail; |
| 91 | } |
| 92 | } |
| 93 | |
| 94 | int nogripe = 0; |
| 95 | |
| 96 | tperror() { |
| 97 | if (pipeout) { |
| 98 | msg("Tape write error on %s\n", tape); |
| 99 | msg("Cannot recover\n"); |
| 100 | dumpabort(); |
| 101 | /* NOTREACHED */ |
| 102 | } |
| 103 | msg("Tape write error on tape %d\n", tapeno); |
| 104 | broadcast("TAPE ERROR!\n"); |
| 105 | if (!query("Do you want to restart?")) |
| 106 | dumpabort(); |
| 107 | msg("This tape will rewind. After it is rewound,\n"); |
| 108 | msg("replace the faulty tape with a new one;\n"); |
| 109 | msg("this dump volume will be rewritten.\n"); |
| 110 | nogripe = 1; |
| 111 | close_rewind(); |
| 112 | Exit(X_REWRITE); |
| 113 | } |
| 114 | |
| 115 | senderr() |
| 116 | { |
| 117 | |
| 118 | perror("dump: pipe error in command to slave"); |
| 119 | dumpabort(); |
| 120 | } |
| 121 | |
| 122 | #ifdef RDUMP |
| 123 | tflush(cnt) |
| 124 | int cnt; |
| 125 | { |
| 126 | int i; |
| 127 | |
| 128 | for (i = 0; i < ntrec; i++) |
| 129 | spclrec(); |
| 130 | } |
| 131 | #endif RDUMP |
| 132 | |
| 133 | flusht() |
| 134 | { |
| 135 | int sig, siz = (char *)tblock - (char *)req; |
| 136 | |
| 137 | tmsg("flusht %d\n", siz); |
| 138 | sig = sigblock(1<<SIGINT-1 | 1<<SIGIOT-1); /* Don't interrupt write */ |
| 139 | if (write(slavefd[rotor], req, siz) != siz) |
| 140 | senderr(); |
| 141 | sigsetmask(sig); |
| 142 | if (++rotor >= SLAVES) rotor = 0; |
| 143 | tblock = (char (*)[TP_BSIZE]) &req[ntrec]; |
| 144 | trecno = 0; |
| 145 | asize += writesize/density; |
| 146 | asize += 7; /* inter-record gap (why fixed?) */ |
| 147 | blockswritten += ntrec; |
| 148 | if (!pipeout && asize > tsize) { |
| 149 | close_rewind(); |
| 150 | otape(); |
| 151 | } |
| 152 | timeest(); |
| 153 | } |
| 154 | |
| 155 | rewind() |
| 156 | { |
| 157 | register int f; |
| 158 | |
| 159 | if (pipeout) |
| 160 | return; |
| 161 | for (f = 0; f < SLAVES; f++) |
| 162 | close(slavefd[f]); |
| 163 | while (wait(NULL) >= 0) ; /* wait for any signals from slaves */ |
| 164 | msg("Tape rewinding\n"); |
| 165 | #ifdef RDUMP |
| 166 | rmtclose(); |
| 167 | while (rmtopen(tape, 0) < 0) |
| 168 | sleep(10); |
| 169 | rmtclose(); |
| 170 | #else |
| 171 | close(to); |
| 172 | while ((f = open(tape, 0)) < 0) |
| 173 | sleep (10); |
| 174 | close(f); |
| 175 | #endif |
| 176 | } |
| 177 | |
| 178 | close_rewind() |
| 179 | { |
| 180 | rewind(); |
| 181 | if (!nogripe) { |
| 182 | msg("Change Tapes: Mount tape #%d\n", tapeno+1); |
| 183 | broadcast("CHANGE TAPES!\7\7\n"); |
| 184 | } |
| 185 | while (!query("Is the new tape mounted and ready to go?")) |
| 186 | if (query("Do you want to abort?")) |
| 187 | dumpabort(); |
| 188 | } |
| 189 | |
| 190 | /* |
| 191 | * We implement taking and restoring checkpoints on the tape level. |
| 192 | * When each tape is opened, a new process is created by forking; this |
| 193 | * saves all of the necessary context in the parent. The child |
| 194 | * continues the dump; the parent waits around, saving the context. |
| 195 | * If the child returns X_REWRITE, then it had problems writing that tape; |
| 196 | * this causes the parent to fork again, duplicating the context, and |
| 197 | * everything continues as if nothing had happened. |
| 198 | */ |
| 199 | |
| 200 | otape() |
| 201 | { |
| 202 | int parentpid; |
| 203 | int childpid; |
| 204 | int status; |
| 205 | int waitpid; |
| 206 | int interrupt(); |
| 207 | |
| 208 | parentpid = getpid(); |
| 209 | |
| 210 | restore_check_point: |
| 211 | signal(SIGINT, interrupt); |
| 212 | /* |
| 213 | * All signals are inherited... |
| 214 | */ |
| 215 | childpid = fork(); |
| 216 | if (childpid < 0) { |
| 217 | msg("Context save fork fails in parent %d\n", parentpid); |
| 218 | Exit(X_ABORT); |
| 219 | } |
| 220 | if (childpid != 0) { |
| 221 | /* |
| 222 | * PARENT: |
| 223 | * save the context by waiting |
| 224 | * until the child doing all of the work returns. |
| 225 | * don't catch the interrupt |
| 226 | */ |
| 227 | signal(SIGINT, SIG_IGN); |
| 228 | #ifdef TDEBUG |
| 229 | msg("Tape: %d; parent process: %d child process %d\n", |
| 230 | tapeno+1, parentpid, childpid); |
| 231 | #endif TDEBUG |
| 232 | while ((waitpid = wait(&status)) != childpid) |
| 233 | msg("Parent %d waiting for child %d has another child %d return\n", |
| 234 | parentpid, childpid, waitpid); |
| 235 | if (status & 0xFF) { |
| 236 | msg("Child %d returns LOB status %o\n", |
| 237 | childpid, status&0xFF); |
| 238 | } |
| 239 | status = (status >> 8) & 0xFF; |
| 240 | #ifdef TDEBUG |
| 241 | switch(status) { |
| 242 | case X_FINOK: |
| 243 | msg("Child %d finishes X_FINOK\n", childpid); |
| 244 | break; |
| 245 | case X_ABORT: |
| 246 | msg("Child %d finishes X_ABORT\n", childpid); |
| 247 | break; |
| 248 | case X_REWRITE: |
| 249 | msg("Child %d finishes X_REWRITE\n", childpid); |
| 250 | break; |
| 251 | default: |
| 252 | msg("Child %d finishes unknown %d\n", |
| 253 | childpid, status); |
| 254 | break; |
| 255 | } |
| 256 | #endif TDEBUG |
| 257 | switch(status) { |
| 258 | case X_FINOK: |
| 259 | Exit(X_FINOK); |
| 260 | case X_ABORT: |
| 261 | Exit(X_ABORT); |
| 262 | case X_REWRITE: |
| 263 | goto restore_check_point; |
| 264 | default: |
| 265 | msg("Bad return code from dump: %d\n", status); |
| 266 | Exit(X_ABORT); |
| 267 | } |
| 268 | /*NOTREACHED*/ |
| 269 | } else { /* we are the child; just continue */ |
| 270 | #ifdef TDEBUG |
| 271 | sleep(4); /* allow time for parent's message to get out */ |
| 272 | msg("Child on Tape %d has parent %d, my pid = %d\n", |
| 273 | tapeno+1, parentpid, getpid()); |
| 274 | #endif |
| 275 | #ifdef RDUMP |
| 276 | while ((to = rmtopen(tape, 2)) < 0) |
| 277 | #else |
| 278 | while ((to = pipeout ? 1 : creat(tape, 0666)) < 0) |
| 279 | #endif |
| 280 | if (!query("Cannot open tape. Do you want to retry the open?")) |
| 281 | dumpabort(); |
| 282 | |
| 283 | enslave(); /* Share open tape file descriptor with slaves */ |
| 284 | |
| 285 | asize = 0; |
| 286 | tapeno++; /* current tape sequence */ |
| 287 | newtape++; /* new tape signal */ |
| 288 | spcl.c_volume++; |
| 289 | spcl.c_type = TS_TAPE; |
| 290 | spclrec(); |
| 291 | if (tapeno > 1) |
| 292 | msg("Tape %d begins with blocks from ino %d\n", |
| 293 | tapeno, ino); |
| 294 | } |
| 295 | } |
| 296 | |
| 297 | dumpabort() |
| 298 | { |
| 299 | if (master != 0 && master != getpid()) |
| 300 | kill(master, SIGIOT); |
| 301 | msg("The ENTIRE dump is aborted.\n"); |
| 302 | Exit(X_ABORT); |
| 303 | } |
| 304 | |
| 305 | Exit(status) |
| 306 | { |
| 307 | #ifdef TDEBUG |
| 308 | msg("pid = %d exits with status %d\n", getpid(), status); |
| 309 | #endif TDEBUG |
| 310 | exit(status); |
| 311 | } |
| 312 | |
| 313 | #define OK 020 |
| 314 | char tok = OK; |
| 315 | |
| 316 | enslave() |
| 317 | { |
| 318 | int prev[2], next[2], cmd[2]; /* file descriptors for pipes */ |
| 319 | int i, j, slavepid; |
| 320 | |
| 321 | master = getpid(); |
| 322 | signal(SIGPIPE, dumpabort); |
| 323 | signal(SIGIOT, tperror); /* SIGIOT asks for restart from checkpoint */ |
| 324 | pipe(prev); |
| 325 | for (i = rotor = 0; i < SLAVES; ++i) { |
| 326 | if ((i < SLAVES - 1 && pipe(next) < 0) || pipe(cmd) < 0 |
| 327 | || (slavepid = fork()) < 0) { |
| 328 | perror(" DUMP: too many slaves"); |
| 329 | dumpabort(); |
| 330 | } |
| 331 | if (i >= SLAVES - 1) |
| 332 | next[1] = prev[1]; /* Last slave loops back */ |
| 333 | slavefd[i] = cmd[1]; |
| 334 | if (slavepid == 0) { /* Slave starts up here */ |
| 335 | for (j = 0; j <= i; j++) |
| 336 | close(slavefd[j]); |
| 337 | if (i < SLAVES - 1) { |
| 338 | close(prev[1]); |
| 339 | close(next[0]); |
| 340 | } else { /* Insert initial token */ |
| 341 | if (write(next[1], &tok, 1) != 1) |
| 342 | ringerr(); |
| 343 | } |
| 344 | doslave(i, cmd[0], prev[0], next[1]); |
| 345 | close(next[1]); |
| 346 | j = read(prev[0], &tok, 1); /* Eat the final token */ |
| 347 | #ifdef RDUMP /* Read remaining acknowledges */ |
| 348 | for (; j > 0 && (tok &~ OK) > 0; tok--) { |
| 349 | if (rmtwrite2() != writesize && (tok & OK)) { |
| 350 | kill(master, SIGIOT); |
| 351 | tok &= ~OK; |
| 352 | } |
| 353 | } |
| 354 | #endif |
| 355 | Exit(X_FINOK); |
| 356 | } |
| 357 | close(cmd[0]); |
| 358 | close(next[1]); |
| 359 | close(prev[0]); |
| 360 | prev[0] = next[0]; |
| 361 | } |
| 362 | master = 0; |
| 363 | } |
| 364 | |
| 365 | /* |
| 366 | * Somebody must have died, should never happen |
| 367 | */ |
| 368 | ringerr() |
| 369 | { |
| 370 | perror(" DUMP: token passing error"); |
| 371 | kill(master, SIGPIPE); |
| 372 | Exit(X_ABORT); |
| 373 | } |
| 374 | |
| 375 | doslave(num, cmd, prev, next) |
| 376 | int num, cmd, prev, next; |
| 377 | { |
| 378 | tmsg("slave %d\n", num); |
| 379 | signal(SIGINT, SIG_IGN); /* Master handles it */ |
| 380 | signal(SIGTERM, SIG_IGN); |
| 381 | signal(SIGPIPE, ringerr); |
| 382 | close(fi); |
| 383 | if ((fi = open(disk, 0)) < 0) { /* Need our own seek pointer */ |
| 384 | perror(" DUMP: can't reopen disk"); |
| 385 | kill(master, SIGPIPE); |
| 386 | Exit(X_ABORT); |
| 387 | } |
| 388 | while (readpipe(cmd, req, reqsiz) == reqsiz) { |
| 389 | register struct req *p = req; |
| 390 | for (trecno = 0; trecno < ntrec; trecno += p->count, p += p->count) { |
| 391 | if (p->dblk) { |
| 392 | tmsg("%d READS %d\n", num, p->count); |
| 393 | bread(p->dblk, tblock[trecno], |
| 394 | p->count * TP_BSIZE); |
| 395 | } else { |
| 396 | tmsg("%d PIPEIN %d\n", num, p->count); |
| 397 | if (p->count != 1) |
| 398 | ringerr(); |
| 399 | if (readpipe(cmd, tblock[trecno], TP_BSIZE) != TP_BSIZE) |
| 400 | senderr(); |
| 401 | } |
| 402 | } |
| 403 | if (read(prev, &tok, 1) != 1) |
| 404 | ringerr(); /* Wait your turn */ |
| 405 | tmsg("%d WRITE\n", num); |
| 406 | #ifdef RDUMP |
| 407 | if (tok & OK) { |
| 408 | rmtwrite0(writesize); |
| 409 | rmtwrite1(tblock[0], writesize); |
| 410 | tok++; /* Number of writes in progress */ |
| 411 | } |
| 412 | if (tok > (LAG|OK) && (--tok, rmtwrite2() != writesize)) { |
| 413 | #else |
| 414 | if ((tok & OK) && |
| 415 | write(to, tblock[0], writesize) != writesize) { |
| 416 | perror(tape); |
| 417 | #endif |
| 418 | kill(master, SIGIOT); /* restart from checkpoint */ |
| 419 | tok &= ~OK; |
| 420 | } |
| 421 | if (write(next, &tok, 1) != 1) |
| 422 | ringerr(); /* Next slave's turn */ |
| 423 | } |
| 424 | tmsg("%d CLOSE\n", num); |
| 425 | } |
| 426 | |
| 427 | /* |
| 428 | * Since a read from a pipe may not return all we asked for |
| 429 | * we must loop until we get all we need |
| 430 | */ |
| 431 | readpipe(fd, buf, cnt) |
| 432 | int fd; |
| 433 | char *buf; |
| 434 | int cnt; |
| 435 | { |
| 436 | int rd, got; |
| 437 | |
| 438 | for (rd = cnt; rd > 0; rd -= got) { |
| 439 | got = read(fd, buf, rd); |
| 440 | if (got < 0) |
| 441 | return (got); |
| 442 | if (got == 0) |
| 443 | return (cnt - rd); |
| 444 | buf += got; |
| 445 | } |
| 446 | return (cnt); |
| 447 | } |