]>
Commit | Line | Data |
---|---|---|
1227625a SP |
1 | /* |
2 | * Ported to Linux's Second Extended File System as part of the | |
3 | * dump and restore backup suit | |
4 | * Remy Card <card@Linux.EU.Org>, 1994, 1995, 1996 | |
5 | * | |
6 | */ | |
7 | ||
8 | /*- | |
9 | * Copyright (c) 1980, 1991, 1993 | |
10 | * The Regents of the University of California. All rights reserved. | |
11 | * | |
12 | * Redistribution and use in source and binary forms, with or without | |
13 | * modification, are permitted provided that the following conditions | |
14 | * are met: | |
15 | * 1. Redistributions of source code must retain the above copyright | |
16 | * notice, this list of conditions and the following disclaimer. | |
17 | * 2. Redistributions in binary form must reproduce the above copyright | |
18 | * notice, this list of conditions and the following disclaimer in the | |
19 | * documentation and/or other materials provided with the distribution. | |
20 | * 3. All advertising materials mentioning features or use of this software | |
21 | * must display the following acknowledgement: | |
22 | * This product includes software developed by the University of | |
23 | * California, Berkeley and its contributors. | |
24 | * 4. Neither the name of the University nor the names of its contributors | |
25 | * may be used to endorse or promote products derived from this software | |
26 | * without specific prior written permission. | |
27 | * | |
28 | * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND | |
29 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE | |
30 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE | |
31 | * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE | |
32 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL | |
33 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS | |
34 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) | |
35 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT | |
36 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY | |
37 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF | |
38 | * SUCH DAMAGE. | |
39 | */ | |
40 | ||
41 | #ifndef lint | |
42 | static char sccsid[] = "@(#)tape.c 8.4 (Berkeley) 5/1/95"; | |
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 | #ifdef __linux__ | |
50 | #include <linux/ext2_fs.h> | |
51 | #include <bsdcompat.h> | |
52 | #else /* __linux__ */ | |
53 | #ifdef sunos | |
54 | #include <sys/vnode.h> | |
55 | ||
56 | #include <ufs/fs.h> | |
57 | #include <ufs/inode.h> | |
58 | #else | |
59 | #include <ufs/ufs/dinode.h> | |
60 | #include <ufs/ffs/fs.h> | |
61 | #endif | |
62 | #endif /* __linux__ */ | |
63 | ||
64 | #include <protocols/dumprestore.h> | |
65 | ||
66 | #include <errno.h> | |
67 | #include <fcntl.h> | |
68 | #include <setjmp.h> | |
69 | #include <signal.h> | |
70 | #include <stdio.h> | |
71 | #ifdef __STDC__ | |
72 | #include <stdlib.h> | |
73 | #include <string.h> | |
74 | #include <unistd.h> | |
75 | #else | |
76 | int write(), read(); | |
77 | #endif | |
78 | ||
79 | #ifdef __linux__ | |
80 | #include <ext2fs/ext2fs.h> | |
81 | #endif | |
82 | #include "dump.h" | |
83 | #include "pathnames.h" | |
84 | ||
85 | int writesize; /* size of malloc()ed buffer for tape */ | |
86 | long lastspclrec = -1; /* tape block number of last written header */ | |
87 | int trecno = 0; /* next record to write in current block */ | |
88 | extern long blocksperfile; /* number of blocks per output file */ | |
89 | long blocksthisvol; /* number of blocks on current output file */ | |
90 | extern int ntrec; /* blocking factor on tape */ | |
91 | extern int cartridge; | |
92 | extern char *host; | |
93 | char *nexttape; | |
94 | ||
95 | static int atomic __P((int (*)(), int, char *, int)); | |
96 | static void doslave __P((int, int)); | |
97 | static void enslave __P((void)); | |
98 | static void flushtape __P((void)); | |
99 | static void killall __P((void)); | |
100 | static void rollforward __P((void)); | |
101 | ||
102 | /* | |
103 | * Concurrent dump mods (Caltech) - disk block reading and tape writing | |
104 | * are exported to several slave processes. While one slave writes the | |
105 | * tape, the others read disk blocks; they pass control of the tape in | |
106 | * a ring via signals. The parent process traverses the filesystem and | |
107 | * sends writeheader()'s and lists of daddr's to the slaves via pipes. | |
108 | * The following structure defines the instruction packets sent to slaves. | |
109 | */ | |
110 | struct req { | |
111 | daddr_t dblk; | |
112 | int count; | |
113 | }; | |
114 | int reqsiz; | |
115 | ||
116 | #define SLAVES 3 /* 1 slave writing, 1 reading, 1 for slack */ | |
117 | struct slave { | |
118 | int tapea; /* header number at start of this chunk */ | |
119 | int count; /* count to next header (used for TS_TAPE */ | |
120 | /* after EOT) */ | |
121 | int inode; /* inode that we are currently dealing with */ | |
122 | int fd; /* FD for this slave */ | |
123 | int pid; /* PID for this slave */ | |
124 | int sent; /* 1 == we've sent this slave requests */ | |
125 | int firstrec; /* record number of this block */ | |
126 | char (*tblock)[TP_BSIZE]; /* buffer for data blocks */ | |
127 | struct req *req; /* buffer for requests */ | |
128 | } slaves[SLAVES+1]; | |
129 | struct slave *slp; | |
130 | ||
131 | char (*nextblock)[TP_BSIZE]; | |
132 | ||
133 | int master; /* pid of master, for sending error signals */ | |
134 | int tenths; /* length of tape used per block written */ | |
135 | static int caught; /* have we caught the signal to proceed? */ | |
136 | static int ready; /* have we reached the lock point without having */ | |
137 | /* received the SIGUSR2 signal from the prev slave? */ | |
138 | static jmp_buf jmpbuf; /* where to jump to if we are ready when the */ | |
139 | /* SIGUSR2 arrives from the previous slave */ | |
140 | ||
141 | int | |
142 | alloctape() | |
143 | { | |
144 | int pgoff = getpagesize() - 1; | |
145 | char *buf; | |
146 | int i; | |
147 | ||
148 | writesize = ntrec * TP_BSIZE; | |
149 | reqsiz = (ntrec + 1) * sizeof(struct req); | |
150 | /* | |
151 | * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode | |
152 | * (see DEC TU80 User's Guide). The shorter gaps of 6250-bpi require | |
153 | * repositioning after stopping, i.e, streaming mode, where the gap is | |
154 | * variable, 0.30" to 0.45". The gap is maximal when the tape stops. | |
155 | */ | |
156 | if (blocksperfile == 0) | |
157 | tenths = writesize / density + | |
158 | (cartridge ? 16 : density == 625 ? 5 : 8); | |
159 | /* | |
160 | * Allocate tape buffer contiguous with the array of instruction | |
161 | * packets, so flushtape() can write them together with one write(). | |
162 | * Align tape buffer on page boundary to speed up tape write(). | |
163 | */ | |
164 | for (i = 0; i <= SLAVES; i++) { | |
165 | buf = (char *) | |
166 | malloc((unsigned)(reqsiz + writesize + pgoff + TP_BSIZE)); | |
167 | if (buf == NULL) | |
168 | return(0); | |
169 | slaves[i].tblock = (char (*)[TP_BSIZE]) | |
170 | #ifdef __linux__ | |
171 | (((long)&buf[reqsiz] + pgoff) &~ pgoff); | |
172 | #else | |
173 | (((long)&buf[ntrec + 1] + pgoff) &~ pgoff); | |
174 | #endif | |
175 | slaves[i].req = (struct req *)slaves[i].tblock - ntrec - 1; | |
176 | } | |
177 | slp = &slaves[0]; | |
178 | slp->count = 1; | |
179 | slp->tapea = 0; | |
180 | slp->firstrec = 0; | |
181 | nextblock = slp->tblock; | |
182 | return(1); | |
183 | } | |
184 | ||
185 | void | |
186 | writerec(dp, isspcl) | |
187 | char *dp; | |
188 | int isspcl; | |
189 | { | |
190 | ||
191 | slp->req[trecno].dblk = (daddr_t)0; | |
192 | slp->req[trecno].count = 1; | |
193 | *(union u_spcl *)(*(nextblock)++) = *(union u_spcl *)dp; | |
194 | if (isspcl) | |
195 | lastspclrec = spcl.c_tapea; | |
196 | trecno++; | |
197 | spcl.c_tapea++; | |
198 | if (trecno >= ntrec) | |
199 | flushtape(); | |
200 | } | |
201 | ||
202 | void | |
203 | dumpblock(blkno, size) | |
204 | daddr_t blkno; | |
205 | int size; | |
206 | { | |
207 | int avail, tpblks, dblkno; | |
208 | ||
209 | dblkno = fsbtodb(sblock, blkno); | |
210 | tpblks = size >> tp_bshift; | |
211 | while ((avail = MIN(tpblks, ntrec - trecno)) > 0) { | |
212 | slp->req[trecno].dblk = dblkno; | |
213 | slp->req[trecno].count = avail; | |
214 | trecno += avail; | |
215 | spcl.c_tapea += avail; | |
216 | if (trecno >= ntrec) | |
217 | flushtape(); | |
218 | dblkno += avail << (tp_bshift - dev_bshift); | |
219 | tpblks -= avail; | |
220 | } | |
221 | } | |
222 | ||
223 | int nogripe = 0; | |
224 | ||
225 | void | |
226 | tperror(signo) | |
227 | int signo; | |
228 | { | |
229 | ||
230 | if (pipeout) { | |
231 | msg("write error on %s\n", tape); | |
232 | quit("Cannot recover\n"); | |
233 | /* NOTREACHED */ | |
234 | } | |
235 | msg("write error %d blocks into volume %d\n", blocksthisvol, tapeno); | |
236 | broadcast("DUMP WRITE ERROR!\n"); | |
237 | if (!query("Do you want to restart?")) | |
238 | dumpabort(0); | |
239 | msg("Closing this volume. Prepare to restart with new media;\n"); | |
240 | msg("this dump volume will be rewritten.\n"); | |
241 | killall(); | |
242 | nogripe = 1; | |
243 | close_rewind(); | |
244 | Exit(X_REWRITE); | |
245 | } | |
246 | ||
247 | void | |
248 | sigpipe(signo) | |
249 | int signo; | |
250 | { | |
251 | ||
252 | quit("Broken pipe\n"); | |
253 | } | |
254 | ||
255 | static void | |
256 | flushtape() | |
257 | { | |
258 | int i, blks, got; | |
259 | long lastfirstrec; | |
260 | ||
261 | int siz = (char *)nextblock - (char *)slp->req; | |
262 | ||
263 | slp->req[trecno].count = 0; /* Sentinel */ | |
264 | ||
265 | if (atomic(write, slp->fd, (char *)slp->req, siz) != siz) | |
266 | quit("error writing command pipe: %s\n", strerror(errno)); | |
267 | slp->sent = 1; /* we sent a request, read the response later */ | |
268 | ||
269 | lastfirstrec = slp->firstrec; | |
270 | ||
271 | if (++slp >= &slaves[SLAVES]) | |
272 | slp = &slaves[0]; | |
273 | ||
274 | /* Read results back from next slave */ | |
275 | if (slp->sent) { | |
276 | if (atomic(read, slp->fd, (char *)&got, sizeof got) | |
277 | != sizeof got) { | |
278 | perror(" DUMP: error reading command pipe in master"); | |
279 | dumpabort(0); | |
280 | } | |
281 | slp->sent = 0; | |
282 | ||
283 | /* Check for end of tape */ | |
284 | if (got < writesize) { | |
285 | msg("End of tape detected\n"); | |
286 | ||
287 | /* | |
288 | * Drain the results, don't care what the values were. | |
289 | * If we read them here then trewind won't... | |
290 | */ | |
291 | for (i = 0; i < SLAVES; i++) { | |
292 | if (slaves[i].sent) { | |
293 | if (atomic(read, slaves[i].fd, | |
294 | (char *)&got, sizeof got) | |
295 | != sizeof got) { | |
296 | perror(" DUMP: error reading command pipe in master"); | |
297 | dumpabort(0); | |
298 | } | |
299 | slaves[i].sent = 0; | |
300 | } | |
301 | } | |
302 | ||
303 | close_rewind(); | |
304 | rollforward(); | |
305 | return; | |
306 | } | |
307 | } | |
308 | ||
309 | blks = 0; | |
310 | if (spcl.c_type != TS_END) { | |
311 | for (i = 0; i < spcl.c_count; i++) | |
312 | if (spcl.c_addr[i] != 0) | |
313 | blks++; | |
314 | } | |
315 | slp->count = lastspclrec + blks + 1 - spcl.c_tapea; | |
316 | slp->tapea = spcl.c_tapea; | |
317 | slp->firstrec = lastfirstrec + ntrec; | |
318 | slp->inode = curino; | |
319 | nextblock = slp->tblock; | |
320 | trecno = 0; | |
321 | asize += tenths; | |
322 | blockswritten += ntrec; | |
323 | blocksthisvol += ntrec; | |
324 | if (!pipeout && (blocksperfile ? | |
325 | (blocksthisvol >= blocksperfile) : (asize > tsize))) { | |
326 | close_rewind(); | |
327 | startnewtape(0); | |
328 | } | |
329 | timeest(); | |
330 | } | |
331 | ||
332 | void | |
333 | trewind() | |
334 | { | |
335 | int f; | |
336 | int got; | |
337 | ||
338 | for (f = 0; f < SLAVES; f++) { | |
339 | /* | |
340 | * Drain the results, but unlike EOT we DO (or should) care | |
341 | * what the return values were, since if we detect EOT after | |
342 | * we think we've written the last blocks to the tape anyway, | |
343 | * we have to replay those blocks with rollforward. | |
344 | * | |
345 | * fixme: punt for now. | |
346 | */ | |
347 | if (slaves[f].sent) { | |
348 | if (atomic(read, slaves[f].fd, (char *)&got, sizeof got) | |
349 | != sizeof got) { | |
350 | perror(" DUMP: error reading command pipe in master"); | |
351 | dumpabort(0); | |
352 | } | |
353 | slaves[f].sent = 0; | |
354 | if (got != writesize) { | |
355 | msg("EOT detected in last 2 tape records!\n"); | |
356 | msg("Use a longer tape, decrease the size estimate\n"); | |
357 | quit("or use no size estimate at all.\n"); | |
358 | } | |
359 | } | |
360 | (void) close(slaves[f].fd); | |
361 | } | |
362 | while (wait((int *)NULL) >= 0) /* wait for any signals from slaves */ | |
363 | /* void */; | |
364 | ||
365 | if (pipeout) | |
366 | return; | |
367 | ||
368 | msg("Closing %s\n", tape); | |
369 | ||
370 | #ifdef RDUMP | |
371 | if (host) { | |
372 | rmtclose(); | |
373 | while (rmtopen(tape, 0) < 0) | |
374 | sleep(10); | |
375 | rmtclose(); | |
376 | return; | |
377 | } | |
378 | #endif | |
379 | (void) close(tapefd); | |
380 | while ((f = open(tape, 0)) < 0) | |
381 | sleep (10); | |
382 | (void) close(f); | |
383 | } | |
384 | ||
385 | void | |
386 | close_rewind() | |
387 | { | |
388 | trewind(); | |
389 | if (nexttape) | |
390 | return; | |
391 | if (!nogripe) { | |
392 | msg("Change Volumes: Mount volume #%d\n", tapeno+1); | |
393 | broadcast("CHANGE DUMP VOLUMES!\7\7\n"); | |
394 | } | |
395 | while (!query("Is the new volume mounted and ready to go?")) | |
396 | if (query("Do you want to abort?")) { | |
397 | dumpabort(0); | |
398 | /*NOTREACHED*/ | |
399 | } | |
400 | } | |
401 | ||
402 | void | |
403 | rollforward() | |
404 | { | |
405 | register struct req *p, *q, *prev; | |
406 | register struct slave *tslp; | |
407 | int i, size, savedtapea, got; | |
408 | union u_spcl *ntb, *otb; | |
409 | tslp = &slaves[SLAVES]; | |
410 | ntb = (union u_spcl *)tslp->tblock[1]; | |
411 | ||
412 | /* | |
413 | * Each of the N slaves should have requests that need to | |
414 | * be replayed on the next tape. Use the extra slave buffers | |
415 | * (slaves[SLAVES]) to construct request lists to be sent to | |
416 | * each slave in turn. | |
417 | */ | |
418 | for (i = 0; i < SLAVES; i++) { | |
419 | q = &tslp->req[1]; | |
420 | otb = (union u_spcl *)slp->tblock; | |
421 | ||
422 | /* | |
423 | * For each request in the current slave, copy it to tslp. | |
424 | */ | |
425 | ||
426 | prev = NULL; | |
427 | for (p = slp->req; p->count > 0; p += p->count) { | |
428 | *q = *p; | |
429 | if (p->dblk == 0) | |
430 | *ntb++ = *otb++; /* copy the datablock also */ | |
431 | prev = q; | |
432 | q += q->count; | |
433 | } | |
434 | if (prev == NULL) | |
435 | quit("rollforward: protocol botch"); | |
436 | if (prev->dblk != 0) | |
437 | prev->count -= 1; | |
438 | else | |
439 | ntb--; | |
440 | q -= 1; | |
441 | q->count = 0; | |
442 | q = &tslp->req[0]; | |
443 | if (i == 0) { | |
444 | q->dblk = 0; | |
445 | q->count = 1; | |
446 | trecno = 0; | |
447 | nextblock = tslp->tblock; | |
448 | savedtapea = spcl.c_tapea; | |
449 | spcl.c_tapea = slp->tapea; | |
450 | startnewtape(0); | |
451 | spcl.c_tapea = savedtapea; | |
452 | lastspclrec = savedtapea - 1; | |
453 | } | |
454 | size = (char *)ntb - (char *)q; | |
455 | if (atomic(write, slp->fd, (char *)q, size) != size) { | |
456 | perror(" DUMP: error writing command pipe"); | |
457 | dumpabort(0); | |
458 | } | |
459 | slp->sent = 1; | |
460 | if (++slp >= &slaves[SLAVES]) | |
461 | slp = &slaves[0]; | |
462 | ||
463 | q->count = 1; | |
464 | ||
465 | if (prev->dblk != 0) { | |
466 | /* | |
467 | * If the last one was a disk block, make the | |
468 | * first of this one be the last bit of that disk | |
469 | * block... | |
470 | */ | |
471 | q->dblk = prev->dblk + | |
472 | prev->count * (TP_BSIZE / DEV_BSIZE); | |
473 | ntb = (union u_spcl *)tslp->tblock; | |
474 | } else { | |
475 | /* | |
476 | * It wasn't a disk block. Copy the data to its | |
477 | * new location in the buffer. | |
478 | */ | |
479 | q->dblk = 0; | |
480 | *((union u_spcl *)tslp->tblock) = *ntb; | |
481 | ntb = (union u_spcl *)tslp->tblock[1]; | |
482 | } | |
483 | } | |
484 | slp->req[0] = *q; | |
485 | nextblock = slp->tblock; | |
486 | if (q->dblk == 0) | |
487 | nextblock++; | |
488 | trecno = 1; | |
489 | ||
490 | /* | |
491 | * Clear the first slaves' response. One hopes that it | |
492 | * worked ok, otherwise the tape is much too short! | |
493 | */ | |
494 | if (slp->sent) { | |
495 | if (atomic(read, slp->fd, (char *)&got, sizeof got) | |
496 | != sizeof got) { | |
497 | perror(" DUMP: error reading command pipe in master"); | |
498 | dumpabort(0); | |
499 | } | |
500 | slp->sent = 0; | |
501 | ||
502 | if (got != writesize) { | |
503 | quit("EOT detected at start of the tape!\n"); | |
504 | } | |
505 | } | |
506 | } | |
507 | ||
508 | /* | |
509 | * We implement taking and restoring checkpoints on the tape level. | |
510 | * When each tape is opened, a new process is created by forking; this | |
511 | * saves all of the necessary context in the parent. The child | |
512 | * continues the dump; the parent waits around, saving the context. | |
513 | * If the child returns X_REWRITE, then it had problems writing that tape; | |
514 | * this causes the parent to fork again, duplicating the context, and | |
515 | * everything continues as if nothing had happened. | |
516 | */ | |
517 | void | |
518 | startnewtape(top) | |
519 | int top; | |
520 | { | |
521 | int parentpid; | |
522 | int childpid; | |
523 | int status; | |
524 | int waitpid; | |
525 | char *p; | |
526 | #ifdef __linux__ | |
527 | void (*interrupt_save)(); | |
528 | #else /* __linux__ */ | |
529 | #ifdef sunos | |
530 | void (*interrupt_save)(); | |
531 | #else | |
532 | sig_t interrupt_save; | |
533 | #endif | |
534 | #endif /* __linux__ */ | |
535 | ||
536 | interrupt_save = signal(SIGINT, SIG_IGN); | |
537 | parentpid = getpid(); | |
538 | ||
539 | restore_check_point: | |
540 | (void)signal(SIGINT, interrupt_save); | |
541 | /* | |
542 | * All signals are inherited... | |
543 | */ | |
544 | childpid = fork(); | |
545 | if (childpid < 0) { | |
546 | msg("Context save fork fails in parent %d\n", parentpid); | |
547 | Exit(X_ABORT); | |
548 | } | |
549 | if (childpid != 0) { | |
550 | /* | |
551 | * PARENT: | |
552 | * save the context by waiting | |
553 | * until the child doing all of the work returns. | |
554 | * don't catch the interrupt | |
555 | */ | |
556 | signal(SIGINT, SIG_IGN); | |
557 | #ifdef TDEBUG | |
558 | msg("Tape: %d; parent process: %d child process %d\n", | |
559 | tapeno+1, parentpid, childpid); | |
560 | #endif /* TDEBUG */ | |
561 | while ((waitpid = wait(&status)) != childpid) | |
562 | msg("Parent %d waiting for child %d has another child %d return\n", | |
563 | parentpid, childpid, waitpid); | |
564 | if (status & 0xFF) { | |
565 | msg("Child %d returns LOB status %o\n", | |
566 | childpid, status&0xFF); | |
567 | } | |
568 | status = (status >> 8) & 0xFF; | |
569 | #ifdef TDEBUG | |
570 | switch(status) { | |
571 | case X_FINOK: | |
572 | msg("Child %d finishes X_FINOK\n", childpid); | |
573 | break; | |
574 | case X_ABORT: | |
575 | msg("Child %d finishes X_ABORT\n", childpid); | |
576 | break; | |
577 | case X_REWRITE: | |
578 | msg("Child %d finishes X_REWRITE\n", childpid); | |
579 | break; | |
580 | default: | |
581 | msg("Child %d finishes unknown %d\n", | |
582 | childpid, status); | |
583 | break; | |
584 | } | |
585 | #endif /* TDEBUG */ | |
586 | switch(status) { | |
587 | case X_FINOK: | |
588 | Exit(X_FINOK); | |
589 | case X_ABORT: | |
590 | Exit(X_ABORT); | |
591 | case X_REWRITE: | |
592 | goto restore_check_point; | |
593 | default: | |
594 | msg("Bad return code from dump: %d\n", status); | |
595 | Exit(X_ABORT); | |
596 | } | |
597 | /*NOTREACHED*/ | |
598 | } else { /* we are the child; just continue */ | |
599 | #ifdef TDEBUG | |
600 | sleep(4); /* allow time for parent's message to get out */ | |
601 | msg("Child on Tape %d has parent %d, my pid = %d\n", | |
602 | tapeno+1, parentpid, getpid()); | |
603 | #endif /* TDEBUG */ | |
604 | /* | |
605 | * If we have a name like "/dev/rmt0,/dev/rmt1", | |
606 | * use the name before the comma first, and save | |
607 | * the remaining names for subsequent volumes. | |
608 | */ | |
609 | tapeno++; /* current tape sequence */ | |
610 | if (nexttape || strchr(tape, ',')) { | |
611 | if (nexttape && *nexttape) | |
612 | tape = nexttape; | |
613 | if ((p = strchr(tape, ',')) != NULL) { | |
614 | *p = '\0'; | |
615 | nexttape = p + 1; | |
616 | } else | |
617 | nexttape = NULL; | |
618 | msg("Dumping volume %d on %s\n", tapeno, tape); | |
619 | } | |
620 | #ifdef RDUMP | |
621 | while ((tapefd = (host ? rmtopen(tape, 2) : | |
622 | pipeout ? 1 : open(tape, O_WRONLY|O_CREAT, 0666))) < 0) | |
623 | #else | |
624 | while ((tapefd = (pipeout ? 1 : | |
625 | open(tape, O_WRONLY|O_CREAT, 0666))) < 0) | |
626 | #endif | |
627 | { | |
628 | msg("Cannot open output \"%s\".\n", tape); | |
629 | if (!query("Do you want to retry the open?")) | |
630 | dumpabort(0); | |
631 | } | |
632 | ||
633 | enslave(); /* Share open tape file descriptor with slaves */ | |
634 | ||
635 | asize = 0; | |
636 | blocksthisvol = 0; | |
637 | if (top) | |
638 | newtape++; /* new tape signal */ | |
639 | spcl.c_count = slp->count; | |
640 | /* | |
641 | * measure firstrec in TP_BSIZE units since restore doesn't | |
642 | * know the correct ntrec value... | |
643 | */ | |
644 | spcl.c_firstrec = slp->firstrec; | |
645 | spcl.c_volume++; | |
646 | spcl.c_type = TS_TAPE; | |
647 | spcl.c_flags |= DR_NEWHEADER; | |
648 | writeheader((ino_t)slp->inode); | |
649 | spcl.c_flags &=~ DR_NEWHEADER; | |
650 | if (tapeno > 1) | |
651 | msg("Volume %d begins with blocks from inode %d\n", | |
652 | tapeno, slp->inode); | |
653 | } | |
654 | } | |
655 | ||
656 | void | |
657 | dumpabort(signo) | |
658 | int signo; | |
659 | { | |
660 | ||
661 | if (master != 0 && master != getpid()) | |
662 | /* Signals master to call dumpabort */ | |
663 | (void) kill(master, SIGTERM); | |
664 | else { | |
665 | killall(); | |
666 | msg("The ENTIRE dump is aborted.\n"); | |
667 | } | |
668 | #ifdef RDUMP | |
669 | rmtclose(); | |
670 | #endif | |
671 | Exit(X_ABORT); | |
672 | } | |
673 | ||
674 | __dead void | |
675 | Exit(status) | |
676 | int status; | |
677 | { | |
678 | ||
679 | #ifdef TDEBUG | |
680 | msg("pid = %d exits with status %d\n", getpid(), status); | |
681 | #endif /* TDEBUG */ | |
682 | exit(status); | |
683 | } | |
684 | ||
685 | /* | |
686 | * proceed - handler for SIGUSR2, used to synchronize IO between the slaves. | |
687 | */ | |
688 | void | |
689 | proceed(signo) | |
690 | int signo; | |
691 | { | |
692 | ||
693 | if (ready) | |
694 | longjmp(jmpbuf, 1); | |
695 | caught++; | |
696 | } | |
697 | ||
698 | void | |
699 | enslave() | |
700 | { | |
701 | int cmd[2]; | |
702 | #ifdef LINUX_FORK_BUG | |
703 | int i, j; | |
704 | #else | |
705 | register int i, j; | |
706 | #endif | |
707 | ||
708 | master = getpid(); | |
709 | ||
710 | signal(SIGTERM, dumpabort); /* Slave sends SIGTERM on dumpabort() */ | |
711 | signal(SIGPIPE, sigpipe); | |
712 | signal(SIGUSR1, tperror); /* Slave sends SIGUSR1 on tape errors */ | |
713 | signal(SIGUSR2, proceed); /* Slave sends SIGUSR2 to next slave */ | |
714 | ||
715 | for (i = 0; i < SLAVES; i++) { | |
716 | if (i == slp - &slaves[0]) { | |
717 | caught = 1; | |
718 | } else { | |
719 | caught = 0; | |
720 | } | |
721 | ||
722 | if (socketpair(AF_UNIX, SOCK_STREAM, 0, cmd) < 0 || | |
723 | (slaves[i].pid = fork()) < 0) | |
724 | quit("too many slaves, %d (recompile smaller): %s\n", | |
725 | i, strerror(errno)); | |
726 | ||
727 | slaves[i].fd = cmd[1]; | |
728 | slaves[i].sent = 0; | |
729 | if (slaves[i].pid == 0) { /* Slave starts up here */ | |
730 | for (j = 0; j <= i; j++) | |
731 | (void) close(slaves[j].fd); | |
732 | signal(SIGINT, SIG_IGN); /* Master handles this */ | |
733 | #ifdef LINUX_FORK_BUG | |
734 | if (atomic(write, cmd[0], (char *) &i, sizeof i) | |
735 | != sizeof i) | |
736 | quit("master/slave protocol botched 3\n"); | |
737 | #endif | |
738 | doslave(cmd[0], i); | |
739 | Exit(X_FINOK); | |
740 | } | |
741 | } | |
742 | ||
743 | #ifdef LINUX_FORK_BUG | |
744 | /* | |
745 | * Wait for all slaves to _actually_ start to circumvent a bug in | |
746 | * Linux kernels >= 2.1.3 where a signal sent to a child that hasn't | |
747 | * returned from fork() causes a SEGV in the child process | |
748 | */ | |
749 | for (i = 0; i < SLAVES; i++) | |
750 | if (atomic(read, slaves[i].fd, (char *) &j, sizeof j) != sizeof j) | |
751 | quit("master/slave protocol botched 4\n"); | |
752 | #endif | |
753 | ||
754 | /* write pid of next slave to each slave */ | |
755 | for (i = 0; i < SLAVES; i++) | |
756 | (void) atomic(write, slaves[i].fd, | |
757 | (char *) &slaves[(i + 1) % SLAVES].pid, | |
758 | sizeof slaves[0].pid); | |
759 | ||
760 | master = 0; | |
761 | } | |
762 | ||
763 | void | |
764 | killall() | |
765 | { | |
766 | register int i; | |
767 | ||
768 | for (i = 0; i < SLAVES; i++) | |
769 | if (slaves[i].pid > 0) | |
770 | (void) kill(slaves[i].pid, SIGKILL); | |
771 | } | |
772 | ||
773 | /* | |
774 | * Synchronization - each process has a lockfile, and shares file | |
775 | * descriptors to the following process's lockfile. When our write | |
776 | * completes, we release our lock on the following process's lock- | |
777 | * file, allowing the following process to lock it and proceed. We | |
778 | * get the lock back for the next cycle by swapping descriptors. | |
779 | */ | |
780 | static void | |
781 | doslave(cmd, slave_number) | |
782 | register int cmd; | |
783 | int slave_number; | |
784 | { | |
785 | register int nread; | |
786 | int nextslave, size, wrote, eot_count; | |
787 | #ifdef __linux__ | |
788 | errcode_t retval; | |
789 | #endif | |
790 | ||
791 | #ifdef TDEBUG | |
792 | msg("slave %d, pid %d\n", slave_number, getpid()); | |
793 | #endif | |
794 | /* | |
795 | * Need our own seek pointer. | |
796 | */ | |
797 | (void) close(diskfd); | |
798 | if ((diskfd = open(disk, O_RDONLY)) < 0) | |
799 | quit("slave couldn't reopen disk: %s\n", strerror(errno)); | |
800 | #ifdef __linux__ | |
801 | ext2fs_close(fs); | |
802 | retval = ext2fs_open(disk, 0, 0, 0, unix_io_manager, &fs); | |
803 | if (retval) | |
804 | quit("slave couldn't reopen disk: %s\n", strerror(errno)); | |
805 | #endif /* __linux__ */ | |
806 | ||
807 | /* | |
808 | * Need the pid of the next slave in the loop... | |
809 | */ | |
810 | if ((nread = atomic(read, cmd, (char *)&nextslave, sizeof nextslave)) | |
811 | != sizeof nextslave) { | |
812 | quit("master/slave protocol botched - didn't get pid of next slave.\n"); | |
813 | } | |
814 | ||
815 | /* | |
816 | * Get list of blocks to dump, read the blocks into tape buffer | |
817 | */ | |
818 | while ((nread = atomic(read, cmd, (char *)slp->req, reqsiz)) == reqsiz) { | |
819 | register struct req *p = slp->req; | |
820 | ||
821 | for (trecno = 0; trecno < ntrec; | |
822 | trecno += p->count, p += p->count) { | |
823 | if (p->dblk) { | |
824 | bread(p->dblk, slp->tblock[trecno], | |
825 | p->count * TP_BSIZE); | |
826 | } else { | |
827 | if (p->count != 1 || atomic(read, cmd, | |
828 | (char *)slp->tblock[trecno], | |
829 | TP_BSIZE) != TP_BSIZE) | |
830 | quit("master/slave protocol botched.\n"); | |
831 | } | |
832 | } | |
833 | if (setjmp(jmpbuf) == 0) { | |
834 | ready = 1; | |
835 | if (!caught) | |
836 | (void) pause(); | |
837 | } | |
838 | ready = 0; | |
839 | caught = 0; | |
840 | ||
841 | /* Try to write the data... */ | |
842 | eot_count = 0; | |
843 | size = 0; | |
844 | ||
845 | while (eot_count < 10 && size < writesize) { | |
846 | #ifdef RDUMP | |
847 | if (host) | |
848 | wrote = rmtwrite(slp->tblock[0]+size, | |
849 | writesize-size); | |
850 | else | |
851 | #endif | |
852 | wrote = write(tapefd, slp->tblock[0]+size, | |
853 | writesize-size); | |
854 | #ifdef WRITEDEBUG | |
855 | msg("slave %d wrote %d\n", slave_number, wrote); | |
856 | #endif | |
857 | if (wrote < 0) | |
858 | break; | |
859 | if (wrote == 0) | |
860 | eot_count++; | |
861 | size += wrote; | |
862 | } | |
863 | ||
864 | #ifdef WRITEDEBUG | |
865 | if (size != writesize) | |
866 | msg("slave %d only wrote %d out of %d bytes and gave up.\n", | |
867 | slave_number, size, writesize); | |
868 | #endif | |
869 | ||
870 | if (eot_count > 0) | |
871 | size = 0; | |
872 | ||
873 | /* | |
874 | * fixme: Pyramids running OSx return ENOSPC | |
875 | * at EOT on 1/2 inch drives. | |
876 | */ | |
877 | if (size < 0) { | |
878 | (void) kill(master, SIGUSR1); | |
879 | for (;;) | |
880 | (void) sigpause(0); | |
881 | } else { | |
882 | /* | |
883 | * pass size of write back to master | |
884 | * (for EOT handling) | |
885 | */ | |
886 | (void) atomic(write, cmd, (char *)&size, sizeof size); | |
887 | } | |
888 | ||
889 | /* | |
890 | * If partial write, don't want next slave to go. | |
891 | * Also jolts him awake. | |
892 | */ | |
893 | (void) kill(nextslave, SIGUSR2); | |
894 | } | |
895 | if (nread != 0) | |
896 | quit("error reading command pipe: %s\n", strerror(errno)); | |
897 | } | |
898 | ||
899 | /* | |
900 | * Since a read from a pipe may not return all we asked for, | |
901 | * or a write may not write all we ask if we get a signal, | |
902 | * loop until the count is satisfied (or error). | |
903 | */ | |
904 | static int | |
905 | atomic(func, fd, buf, count) | |
906 | int (*func)(), fd; | |
907 | char *buf; | |
908 | int count; | |
909 | { | |
910 | int got, need = count; | |
911 | ||
912 | while ((got = (*func)(fd, buf, need)) > 0 && (need -= got) > 0) | |
913 | buf += got; | |
914 | return (got < 0 ? got : count - need); | |
915 | } |