]>
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 | |
b45f51d6 SP |
4 | * Remy Card <card@Linux.EU.Org>, 1994-1997 |
5 | * Stelian Pop <pop@cybercable.fr>, 1999 | |
1227625a SP |
6 | * |
7 | */ | |
8 | ||
9 | /*- | |
10 | * Copyright (c) 1980, 1991, 1993 | |
11 | * The Regents of the University of California. All rights reserved. | |
12 | * | |
13 | * Redistribution and use in source and binary forms, with or without | |
14 | * modification, are permitted provided that the following conditions | |
15 | * are met: | |
16 | * 1. Redistributions of source code must retain the above copyright | |
17 | * notice, this list of conditions and the following disclaimer. | |
18 | * 2. Redistributions in binary form must reproduce the above copyright | |
19 | * notice, this list of conditions and the following disclaimer in the | |
20 | * documentation and/or other materials provided with the distribution. | |
21 | * 3. All advertising materials mentioning features or use of this software | |
22 | * must display the following acknowledgement: | |
23 | * This product includes software developed by the University of | |
24 | * California, Berkeley and its contributors. | |
25 | * 4. Neither the name of the University nor the names of its contributors | |
26 | * may be used to endorse or promote products derived from this software | |
27 | * without specific prior written permission. | |
28 | * | |
29 | * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND | |
30 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE | |
31 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE | |
32 | * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE | |
33 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL | |
34 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS | |
35 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) | |
36 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT | |
37 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY | |
38 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF | |
39 | * SUCH DAMAGE. | |
40 | */ | |
41 | ||
42 | #ifndef lint | |
b45f51d6 | 43 | #if 0 |
1227625a | 44 | static char sccsid[] = "@(#)tape.c 8.4 (Berkeley) 5/1/95"; |
b45f51d6 SP |
45 | #endif |
46 | static const char rcsid[] = | |
ddd2ef55 | 47 | "$Id: tape.c,v 1.3 1999/10/11 12:59:19 stelian Exp $"; |
1227625a SP |
48 | #endif /* not lint */ |
49 | ||
b45f51d6 SP |
50 | #ifdef __linux__ |
51 | #include <sys/types.h> | |
52 | #include <linux/types.h> | |
53 | #endif | |
1227625a SP |
54 | #include <sys/param.h> |
55 | #include <sys/socket.h> | |
56 | #include <sys/time.h> | |
57 | #include <sys/wait.h> | |
58 | #ifdef __linux__ | |
59 | #include <linux/ext2_fs.h> | |
60 | #include <bsdcompat.h> | |
61 | #else /* __linux__ */ | |
62 | #ifdef sunos | |
63 | #include <sys/vnode.h> | |
64 | ||
65 | #include <ufs/fs.h> | |
66 | #include <ufs/inode.h> | |
67 | #else | |
68 | #include <ufs/ufs/dinode.h> | |
69 | #include <ufs/ffs/fs.h> | |
70 | #endif | |
71 | #endif /* __linux__ */ | |
72 | ||
73 | #include <protocols/dumprestore.h> | |
74 | ||
75 | #include <errno.h> | |
76 | #include <fcntl.h> | |
77 | #include <setjmp.h> | |
78 | #include <signal.h> | |
79 | #include <stdio.h> | |
80 | #ifdef __STDC__ | |
81 | #include <stdlib.h> | |
82 | #include <string.h> | |
83 | #include <unistd.h> | |
84 | #else | |
85 | int write(), read(); | |
86 | #endif | |
87 | ||
88 | #ifdef __linux__ | |
89 | #include <ext2fs/ext2fs.h> | |
90 | #endif | |
91 | #include "dump.h" | |
1227625a SP |
92 | |
93 | int writesize; /* size of malloc()ed buffer for tape */ | |
94 | long lastspclrec = -1; /* tape block number of last written header */ | |
95 | int trecno = 0; /* next record to write in current block */ | |
96 | extern long blocksperfile; /* number of blocks per output file */ | |
97 | long blocksthisvol; /* number of blocks on current output file */ | |
98 | extern int ntrec; /* blocking factor on tape */ | |
99 | extern int cartridge; | |
100 | extern char *host; | |
101 | char *nexttape; | |
102 | ||
ddd2ef55 SP |
103 | static ssize_t atomic_read __P((int, void *, size_t)); |
104 | static ssize_t atomic_write __P((int, const void *, size_t)); | |
1227625a SP |
105 | static void doslave __P((int, int)); |
106 | static void enslave __P((void)); | |
107 | static void flushtape __P((void)); | |
108 | static void killall __P((void)); | |
109 | static void rollforward __P((void)); | |
110 | ||
111 | /* | |
112 | * Concurrent dump mods (Caltech) - disk block reading and tape writing | |
113 | * are exported to several slave processes. While one slave writes the | |
114 | * tape, the others read disk blocks; they pass control of the tape in | |
115 | * a ring via signals. The parent process traverses the filesystem and | |
116 | * sends writeheader()'s and lists of daddr's to the slaves via pipes. | |
117 | * The following structure defines the instruction packets sent to slaves. | |
118 | */ | |
119 | struct req { | |
120 | daddr_t dblk; | |
121 | int count; | |
122 | }; | |
123 | int reqsiz; | |
124 | ||
125 | #define SLAVES 3 /* 1 slave writing, 1 reading, 1 for slack */ | |
126 | struct slave { | |
127 | int tapea; /* header number at start of this chunk */ | |
128 | int count; /* count to next header (used for TS_TAPE */ | |
129 | /* after EOT) */ | |
130 | int inode; /* inode that we are currently dealing with */ | |
131 | int fd; /* FD for this slave */ | |
132 | int pid; /* PID for this slave */ | |
133 | int sent; /* 1 == we've sent this slave requests */ | |
134 | int firstrec; /* record number of this block */ | |
135 | char (*tblock)[TP_BSIZE]; /* buffer for data blocks */ | |
136 | struct req *req; /* buffer for requests */ | |
137 | } slaves[SLAVES+1]; | |
138 | struct slave *slp; | |
139 | ||
140 | char (*nextblock)[TP_BSIZE]; | |
141 | ||
ddd2ef55 SP |
142 | static time_t tstart_volume; /* time of volume start */ |
143 | static int tapea_volume; /* value of spcl.c_tapea at volume start */ | |
144 | ||
1227625a SP |
145 | int master; /* pid of master, for sending error signals */ |
146 | int tenths; /* length of tape used per block written */ | |
147 | static int caught; /* have we caught the signal to proceed? */ | |
148 | static int ready; /* have we reached the lock point without having */ | |
149 | /* received the SIGUSR2 signal from the prev slave? */ | |
ddd2ef55 | 150 | static sigjmp_buf jmpbuf; /* where to jump to if we are ready when the */ |
1227625a SP |
151 | /* SIGUSR2 arrives from the previous slave */ |
152 | ||
153 | int | |
ddd2ef55 | 154 | alloctape(void) |
1227625a SP |
155 | { |
156 | int pgoff = getpagesize() - 1; | |
157 | char *buf; | |
158 | int i; | |
159 | ||
160 | writesize = ntrec * TP_BSIZE; | |
161 | reqsiz = (ntrec + 1) * sizeof(struct req); | |
162 | /* | |
163 | * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode | |
164 | * (see DEC TU80 User's Guide). The shorter gaps of 6250-bpi require | |
165 | * repositioning after stopping, i.e, streaming mode, where the gap is | |
166 | * variable, 0.30" to 0.45". The gap is maximal when the tape stops. | |
167 | */ | |
b45f51d6 | 168 | if (blocksperfile == 0 && !unlimited) |
1227625a SP |
169 | tenths = writesize / density + |
170 | (cartridge ? 16 : density == 625 ? 5 : 8); | |
171 | /* | |
172 | * Allocate tape buffer contiguous with the array of instruction | |
173 | * packets, so flushtape() can write them together with one write(). | |
174 | * Align tape buffer on page boundary to speed up tape write(). | |
175 | */ | |
176 | for (i = 0; i <= SLAVES; i++) { | |
177 | buf = (char *) | |
178 | malloc((unsigned)(reqsiz + writesize + pgoff + TP_BSIZE)); | |
179 | if (buf == NULL) | |
180 | return(0); | |
181 | slaves[i].tblock = (char (*)[TP_BSIZE]) | |
182 | #ifdef __linux__ | |
183 | (((long)&buf[reqsiz] + pgoff) &~ pgoff); | |
184 | #else | |
185 | (((long)&buf[ntrec + 1] + pgoff) &~ pgoff); | |
186 | #endif | |
187 | slaves[i].req = (struct req *)slaves[i].tblock - ntrec - 1; | |
188 | } | |
189 | slp = &slaves[0]; | |
190 | slp->count = 1; | |
191 | slp->tapea = 0; | |
192 | slp->firstrec = 0; | |
193 | nextblock = slp->tblock; | |
194 | return(1); | |
195 | } | |
196 | ||
197 | void | |
ddd2ef55 | 198 | writerec(const void *dp, int isspcl) |
1227625a SP |
199 | { |
200 | ||
201 | slp->req[trecno].dblk = (daddr_t)0; | |
202 | slp->req[trecno].count = 1; | |
ddd2ef55 SP |
203 | /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */ |
204 | *(union u_spcl *)(*(nextblock)) = *(union u_spcl *)dp; | |
205 | nextblock++; | |
1227625a SP |
206 | if (isspcl) |
207 | lastspclrec = spcl.c_tapea; | |
208 | trecno++; | |
209 | spcl.c_tapea++; | |
210 | if (trecno >= ntrec) | |
211 | flushtape(); | |
212 | } | |
213 | ||
214 | void | |
ddd2ef55 | 215 | dumpblock(daddr_t blkno, int size) |
1227625a SP |
216 | { |
217 | int avail, tpblks, dblkno; | |
218 | ||
219 | dblkno = fsbtodb(sblock, blkno); | |
220 | tpblks = size >> tp_bshift; | |
221 | while ((avail = MIN(tpblks, ntrec - trecno)) > 0) { | |
222 | slp->req[trecno].dblk = dblkno; | |
223 | slp->req[trecno].count = avail; | |
224 | trecno += avail; | |
225 | spcl.c_tapea += avail; | |
226 | if (trecno >= ntrec) | |
227 | flushtape(); | |
228 | dblkno += avail << (tp_bshift - dev_bshift); | |
229 | tpblks -= avail; | |
230 | } | |
231 | } | |
232 | ||
233 | int nogripe = 0; | |
234 | ||
ddd2ef55 SP |
235 | static void |
236 | tperror(int signo) | |
1227625a SP |
237 | { |
238 | ||
239 | if (pipeout) { | |
240 | msg("write error on %s\n", tape); | |
241 | quit("Cannot recover\n"); | |
242 | /* NOTREACHED */ | |
243 | } | |
244 | msg("write error %d blocks into volume %d\n", blocksthisvol, tapeno); | |
245 | broadcast("DUMP WRITE ERROR!\n"); | |
246 | if (!query("Do you want to restart?")) | |
247 | dumpabort(0); | |
248 | msg("Closing this volume. Prepare to restart with new media;\n"); | |
249 | msg("this dump volume will be rewritten.\n"); | |
250 | killall(); | |
251 | nogripe = 1; | |
252 | close_rewind(); | |
253 | Exit(X_REWRITE); | |
254 | } | |
255 | ||
ddd2ef55 SP |
256 | static void |
257 | sigpipe(int signo) | |
1227625a SP |
258 | { |
259 | ||
260 | quit("Broken pipe\n"); | |
261 | } | |
262 | ||
ddd2ef55 SP |
263 | /* |
264 | * do_stats -- | |
265 | * Update xferrate stats | |
266 | */ | |
267 | time_t | |
268 | do_stats(void) | |
269 | { | |
270 | time_t tnow, ttaken; | |
271 | int blocks; | |
272 | ||
273 | (void)time(&tnow); | |
274 | ttaken = tnow - tstart_volume; | |
275 | blocks = spcl.c_tapea - tapea_volume; | |
276 | msg("Volume %d completed at: %s", tapeno, ctime(&tnow)); | |
277 | if (ttaken > 0) { | |
278 | msg("Volume %d took %d:%02d:%02d\n", tapeno, | |
279 | ttaken / 3600, (ttaken % 3600) / 60, ttaken % 60); | |
280 | msg("Volume %d transfer rate: %ld KB/s\n", tapeno, | |
281 | blocks / ttaken); | |
282 | xferrate += blocks / ttaken; | |
283 | } | |
284 | return(tnow); | |
285 | } | |
286 | ||
287 | #if defined(SIGINFO) | |
288 | /* | |
289 | * statussig -- | |
290 | * information message upon receipt of SIGINFO | |
291 | * (derived from optr.c::timeest()) | |
292 | */ | |
293 | void | |
294 | statussig(int notused) | |
295 | { | |
296 | time_t tnow, deltat; | |
297 | char msgbuf[128]; | |
298 | int save_errno = errno; | |
299 | ||
300 | if (blockswritten < 500) | |
301 | return; | |
302 | (void) time((time_t *) &tnow); | |
303 | deltat = tstart_writing - tnow + (1.0 * (tnow - tstart_writing)) | |
304 | / blockswritten * tapesize; | |
305 | (void)snprintf(msgbuf, sizeof(msgbuf), | |
306 | "%3.2f%% done at %ld KB/s, finished in %d:%02d\n", | |
307 | (blockswritten * 100.0) / tapesize, | |
308 | (spcl.c_tapea - tapea_volume) / (tnow - tstart_volume), | |
309 | (int)(deltat / 3600), (int)((deltat % 3600) / 60)); | |
310 | write(STDERR_FILENO, msgbuf, strlen(msgbuf)); | |
311 | errno = save_errno; | |
312 | } | |
313 | #endif | |
314 | ||
1227625a | 315 | static void |
ddd2ef55 | 316 | flushtape(void) |
1227625a SP |
317 | { |
318 | int i, blks, got; | |
319 | long lastfirstrec; | |
320 | ||
321 | int siz = (char *)nextblock - (char *)slp->req; | |
322 | ||
323 | slp->req[trecno].count = 0; /* Sentinel */ | |
324 | ||
ddd2ef55 | 325 | if (atomic_write( slp->fd, (char *)slp->req, siz) != siz) |
1227625a SP |
326 | quit("error writing command pipe: %s\n", strerror(errno)); |
327 | slp->sent = 1; /* we sent a request, read the response later */ | |
328 | ||
329 | lastfirstrec = slp->firstrec; | |
330 | ||
331 | if (++slp >= &slaves[SLAVES]) | |
332 | slp = &slaves[0]; | |
333 | ||
334 | /* Read results back from next slave */ | |
335 | if (slp->sent) { | |
ddd2ef55 | 336 | if (atomic_read( slp->fd, (char *)&got, sizeof got) |
1227625a SP |
337 | != sizeof got) { |
338 | perror(" DUMP: error reading command pipe in master"); | |
339 | dumpabort(0); | |
340 | } | |
341 | slp->sent = 0; | |
342 | ||
343 | /* Check for end of tape */ | |
344 | if (got < writesize) { | |
345 | msg("End of tape detected\n"); | |
346 | ||
347 | /* | |
348 | * Drain the results, don't care what the values were. | |
349 | * If we read them here then trewind won't... | |
350 | */ | |
351 | for (i = 0; i < SLAVES; i++) { | |
352 | if (slaves[i].sent) { | |
ddd2ef55 | 353 | if (atomic_read( slaves[i].fd, |
1227625a SP |
354 | (char *)&got, sizeof got) |
355 | != sizeof got) { | |
356 | perror(" DUMP: error reading command pipe in master"); | |
357 | dumpabort(0); | |
358 | } | |
359 | slaves[i].sent = 0; | |
360 | } | |
361 | } | |
362 | ||
363 | close_rewind(); | |
364 | rollforward(); | |
365 | return; | |
366 | } | |
367 | } | |
368 | ||
369 | blks = 0; | |
370 | if (spcl.c_type != TS_END) { | |
371 | for (i = 0; i < spcl.c_count; i++) | |
372 | if (spcl.c_addr[i] != 0) | |
373 | blks++; | |
374 | } | |
375 | slp->count = lastspclrec + blks + 1 - spcl.c_tapea; | |
376 | slp->tapea = spcl.c_tapea; | |
377 | slp->firstrec = lastfirstrec + ntrec; | |
378 | slp->inode = curino; | |
379 | nextblock = slp->tblock; | |
380 | trecno = 0; | |
381 | asize += tenths; | |
382 | blockswritten += ntrec; | |
383 | blocksthisvol += ntrec; | |
b45f51d6 | 384 | if (!pipeout && !unlimited && (blocksperfile ? |
1227625a SP |
385 | (blocksthisvol >= blocksperfile) : (asize > tsize))) { |
386 | close_rewind(); | |
387 | startnewtape(0); | |
388 | } | |
389 | timeest(); | |
390 | } | |
391 | ||
392 | void | |
ddd2ef55 | 393 | trewind(void) |
1227625a SP |
394 | { |
395 | int f; | |
396 | int got; | |
397 | ||
398 | for (f = 0; f < SLAVES; f++) { | |
399 | /* | |
b45f51d6 SP |
400 | * Drain the results, but unlike EOT we DO (or should) care |
401 | * what the return values were, since if we detect EOT after | |
402 | * we think we've written the last blocks to the tape anyway, | |
1227625a SP |
403 | * we have to replay those blocks with rollforward. |
404 | * | |
b45f51d6 | 405 | * fixme: punt for now. |
1227625a SP |
406 | */ |
407 | if (slaves[f].sent) { | |
ddd2ef55 | 408 | if (atomic_read( slaves[f].fd, (char *)&got, sizeof got) |
1227625a SP |
409 | != sizeof got) { |
410 | perror(" DUMP: error reading command pipe in master"); | |
411 | dumpabort(0); | |
412 | } | |
413 | slaves[f].sent = 0; | |
414 | if (got != writesize) { | |
415 | msg("EOT detected in last 2 tape records!\n"); | |
416 | msg("Use a longer tape, decrease the size estimate\n"); | |
417 | quit("or use no size estimate at all.\n"); | |
418 | } | |
419 | } | |
420 | (void) close(slaves[f].fd); | |
421 | } | |
422 | while (wait((int *)NULL) >= 0) /* wait for any signals from slaves */ | |
423 | /* void */; | |
424 | ||
425 | if (pipeout) | |
426 | return; | |
427 | ||
428 | msg("Closing %s\n", tape); | |
429 | ||
430 | #ifdef RDUMP | |
431 | if (host) { | |
432 | rmtclose(); | |
433 | while (rmtopen(tape, 0) < 0) | |
434 | sleep(10); | |
435 | rmtclose(); | |
436 | return; | |
437 | } | |
438 | #endif | |
439 | (void) close(tapefd); | |
440 | while ((f = open(tape, 0)) < 0) | |
441 | sleep (10); | |
442 | (void) close(f); | |
443 | } | |
444 | ||
445 | void | |
ddd2ef55 | 446 | close_rewind(void) |
1227625a | 447 | { |
b45f51d6 SP |
448 | time_t tstart_changevol, tend_changevol; |
449 | ||
1227625a | 450 | trewind(); |
ddd2ef55 | 451 | (void)do_stats(); |
1227625a SP |
452 | if (nexttape) |
453 | return; | |
b45f51d6 SP |
454 | #ifdef __linux__ |
455 | (void)time4(&(tstart_changevol)); | |
456 | #else | |
457 | (void)time((time_t *)&(tstart_changevol)); | |
458 | #endif | |
1227625a SP |
459 | if (!nogripe) { |
460 | msg("Change Volumes: Mount volume #%d\n", tapeno+1); | |
461 | broadcast("CHANGE DUMP VOLUMES!\7\7\n"); | |
462 | } | |
463 | while (!query("Is the new volume mounted and ready to go?")) | |
464 | if (query("Do you want to abort?")) { | |
465 | dumpabort(0); | |
466 | /*NOTREACHED*/ | |
467 | } | |
b45f51d6 SP |
468 | #ifdef __linux__ |
469 | (void)time4(&(tend_changevol)); | |
470 | #else | |
471 | (void)time((time_t *)&(tend_changevol)); | |
472 | #endif | |
473 | if ((tstart_changevol != (time_t)-1) && (tend_changevol != (time_t)-1)) | |
474 | tstart_writing += (tend_changevol - tstart_changevol); | |
1227625a SP |
475 | } |
476 | ||
477 | void | |
ddd2ef55 | 478 | rollforward(void) |
1227625a SP |
479 | { |
480 | register struct req *p, *q, *prev; | |
481 | register struct slave *tslp; | |
482 | int i, size, savedtapea, got; | |
483 | union u_spcl *ntb, *otb; | |
b45f51d6 SP |
484 | #ifdef __linux__ |
485 | int blks; | |
486 | long lastfirstrec; | |
487 | #endif | |
1227625a SP |
488 | tslp = &slaves[SLAVES]; |
489 | ntb = (union u_spcl *)tslp->tblock[1]; | |
490 | ||
491 | /* | |
b45f51d6 SP |
492 | * Each of the N slaves should have requests that need to |
493 | * be replayed on the next tape. Use the extra slave buffers | |
494 | * (slaves[SLAVES]) to construct request lists to be sent to | |
1227625a SP |
495 | * each slave in turn. |
496 | */ | |
497 | for (i = 0; i < SLAVES; i++) { | |
498 | q = &tslp->req[1]; | |
499 | otb = (union u_spcl *)slp->tblock; | |
500 | ||
501 | /* | |
b45f51d6 | 502 | * For each request in the current slave, copy it to tslp. |
1227625a SP |
503 | */ |
504 | ||
505 | prev = NULL; | |
506 | for (p = slp->req; p->count > 0; p += p->count) { | |
507 | *q = *p; | |
508 | if (p->dblk == 0) | |
509 | *ntb++ = *otb++; /* copy the datablock also */ | |
510 | prev = q; | |
511 | q += q->count; | |
512 | } | |
513 | if (prev == NULL) | |
514 | quit("rollforward: protocol botch"); | |
515 | if (prev->dblk != 0) | |
516 | prev->count -= 1; | |
517 | else | |
518 | ntb--; | |
519 | q -= 1; | |
520 | q->count = 0; | |
521 | q = &tslp->req[0]; | |
522 | if (i == 0) { | |
523 | q->dblk = 0; | |
524 | q->count = 1; | |
525 | trecno = 0; | |
526 | nextblock = tslp->tblock; | |
527 | savedtapea = spcl.c_tapea; | |
528 | spcl.c_tapea = slp->tapea; | |
529 | startnewtape(0); | |
530 | spcl.c_tapea = savedtapea; | |
531 | lastspclrec = savedtapea - 1; | |
532 | } | |
533 | size = (char *)ntb - (char *)q; | |
ddd2ef55 | 534 | if (atomic_write( slp->fd, (char *)q, size) != size) { |
1227625a SP |
535 | perror(" DUMP: error writing command pipe"); |
536 | dumpabort(0); | |
537 | } | |
538 | slp->sent = 1; | |
b45f51d6 SP |
539 | #ifdef __linux__ |
540 | lastfirstrec = slp->firstrec; | |
541 | #endif | |
1227625a SP |
542 | if (++slp >= &slaves[SLAVES]) |
543 | slp = &slaves[0]; | |
544 | ||
545 | q->count = 1; | |
546 | ||
547 | if (prev->dblk != 0) { | |
548 | /* | |
b45f51d6 SP |
549 | * If the last one was a disk block, make the |
550 | * first of this one be the last bit of that disk | |
1227625a SP |
551 | * block... |
552 | */ | |
553 | q->dblk = prev->dblk + | |
554 | prev->count * (TP_BSIZE / DEV_BSIZE); | |
555 | ntb = (union u_spcl *)tslp->tblock; | |
556 | } else { | |
557 | /* | |
b45f51d6 | 558 | * It wasn't a disk block. Copy the data to its |
1227625a SP |
559 | * new location in the buffer. |
560 | */ | |
561 | q->dblk = 0; | |
562 | *((union u_spcl *)tslp->tblock) = *ntb; | |
563 | ntb = (union u_spcl *)tslp->tblock[1]; | |
564 | } | |
565 | } | |
566 | slp->req[0] = *q; | |
567 | nextblock = slp->tblock; | |
568 | if (q->dblk == 0) | |
b45f51d6 SP |
569 | #ifdef __linux__ |
570 | *(union u_spcl *)(*(nextblock)++) = *(union u_spcl *)tslp->tblock; | |
571 | #else | |
1227625a | 572 | nextblock++; |
b45f51d6 | 573 | #endif |
1227625a SP |
574 | trecno = 1; |
575 | ||
576 | /* | |
577 | * Clear the first slaves' response. One hopes that it | |
578 | * worked ok, otherwise the tape is much too short! | |
579 | */ | |
580 | if (slp->sent) { | |
ddd2ef55 | 581 | if (atomic_read( slp->fd, (char *)&got, sizeof got) |
1227625a SP |
582 | != sizeof got) { |
583 | perror(" DUMP: error reading command pipe in master"); | |
584 | dumpabort(0); | |
585 | } | |
586 | slp->sent = 0; | |
587 | ||
588 | if (got != writesize) { | |
589 | quit("EOT detected at start of the tape!\n"); | |
590 | } | |
591 | } | |
b45f51d6 SP |
592 | |
593 | #ifdef __linux__ | |
594 | blks = 0; | |
595 | if (spcl.c_type != TS_END) { | |
596 | for (i = 0; i < spcl.c_count; i++) | |
597 | if (spcl.c_addr[i] != 0) | |
598 | blks++; | |
599 | } | |
600 | ||
601 | slp->firstrec = lastfirstrec + ntrec; | |
602 | slp->count = lastspclrec + blks + 1 - spcl.c_tapea; | |
603 | slp->inode = curino; | |
604 | asize += tenths; | |
605 | blockswritten += ntrec; | |
606 | blocksthisvol += ntrec; | |
607 | #endif | |
1227625a SP |
608 | } |
609 | ||
610 | /* | |
611 | * We implement taking and restoring checkpoints on the tape level. | |
612 | * When each tape is opened, a new process is created by forking; this | |
613 | * saves all of the necessary context in the parent. The child | |
614 | * continues the dump; the parent waits around, saving the context. | |
615 | * If the child returns X_REWRITE, then it had problems writing that tape; | |
616 | * this causes the parent to fork again, duplicating the context, and | |
617 | * everything continues as if nothing had happened. | |
618 | */ | |
619 | void | |
ddd2ef55 | 620 | startnewtape(int top) |
1227625a SP |
621 | { |
622 | int parentpid; | |
623 | int childpid; | |
624 | int status; | |
625 | int waitpid; | |
626 | char *p; | |
627 | #ifdef __linux__ | |
ddd2ef55 | 628 | void (*interrupt_save) __P((int signo)); |
1227625a SP |
629 | #else /* __linux__ */ |
630 | #ifdef sunos | |
631 | void (*interrupt_save)(); | |
632 | #else | |
633 | sig_t interrupt_save; | |
634 | #endif | |
635 | #endif /* __linux__ */ | |
636 | ||
637 | interrupt_save = signal(SIGINT, SIG_IGN); | |
638 | parentpid = getpid(); | |
ddd2ef55 SP |
639 | tapea_volume = spcl.c_tapea; |
640 | (void)time(&tstart_volume); | |
1227625a SP |
641 | |
642 | restore_check_point: | |
643 | (void)signal(SIGINT, interrupt_save); | |
644 | /* | |
645 | * All signals are inherited... | |
646 | */ | |
647 | childpid = fork(); | |
648 | if (childpid < 0) { | |
649 | msg("Context save fork fails in parent %d\n", parentpid); | |
650 | Exit(X_ABORT); | |
651 | } | |
652 | if (childpid != 0) { | |
653 | /* | |
654 | * PARENT: | |
655 | * save the context by waiting | |
656 | * until the child doing all of the work returns. | |
657 | * don't catch the interrupt | |
658 | */ | |
659 | signal(SIGINT, SIG_IGN); | |
660 | #ifdef TDEBUG | |
661 | msg("Tape: %d; parent process: %d child process %d\n", | |
662 | tapeno+1, parentpid, childpid); | |
663 | #endif /* TDEBUG */ | |
664 | while ((waitpid = wait(&status)) != childpid) | |
665 | msg("Parent %d waiting for child %d has another child %d return\n", | |
666 | parentpid, childpid, waitpid); | |
667 | if (status & 0xFF) { | |
668 | msg("Child %d returns LOB status %o\n", | |
669 | childpid, status&0xFF); | |
670 | } | |
671 | status = (status >> 8) & 0xFF; | |
672 | #ifdef TDEBUG | |
673 | switch(status) { | |
674 | case X_FINOK: | |
675 | msg("Child %d finishes X_FINOK\n", childpid); | |
676 | break; | |
b45f51d6 | 677 | case X_ABORT: |
1227625a SP |
678 | msg("Child %d finishes X_ABORT\n", childpid); |
679 | break; | |
680 | case X_REWRITE: | |
681 | msg("Child %d finishes X_REWRITE\n", childpid); | |
682 | break; | |
683 | default: | |
684 | msg("Child %d finishes unknown %d\n", | |
685 | childpid, status); | |
686 | break; | |
687 | } | |
688 | #endif /* TDEBUG */ | |
689 | switch(status) { | |
690 | case X_FINOK: | |
691 | Exit(X_FINOK); | |
692 | case X_ABORT: | |
693 | Exit(X_ABORT); | |
694 | case X_REWRITE: | |
695 | goto restore_check_point; | |
696 | default: | |
697 | msg("Bad return code from dump: %d\n", status); | |
698 | Exit(X_ABORT); | |
699 | } | |
700 | /*NOTREACHED*/ | |
701 | } else { /* we are the child; just continue */ | |
702 | #ifdef TDEBUG | |
703 | sleep(4); /* allow time for parent's message to get out */ | |
704 | msg("Child on Tape %d has parent %d, my pid = %d\n", | |
705 | tapeno+1, parentpid, getpid()); | |
706 | #endif /* TDEBUG */ | |
707 | /* | |
708 | * If we have a name like "/dev/rmt0,/dev/rmt1", | |
709 | * use the name before the comma first, and save | |
710 | * the remaining names for subsequent volumes. | |
711 | */ | |
712 | tapeno++; /* current tape sequence */ | |
713 | if (nexttape || strchr(tape, ',')) { | |
714 | if (nexttape && *nexttape) | |
715 | tape = nexttape; | |
716 | if ((p = strchr(tape, ',')) != NULL) { | |
717 | *p = '\0'; | |
718 | nexttape = p + 1; | |
719 | } else | |
720 | nexttape = NULL; | |
721 | msg("Dumping volume %d on %s\n", tapeno, tape); | |
722 | } | |
723 | #ifdef RDUMP | |
724 | while ((tapefd = (host ? rmtopen(tape, 2) : | |
725 | pipeout ? 1 : open(tape, O_WRONLY|O_CREAT, 0666))) < 0) | |
726 | #else | |
b45f51d6 | 727 | while ((tapefd = (pipeout ? 1 : |
1227625a SP |
728 | open(tape, O_WRONLY|O_CREAT, 0666))) < 0) |
729 | #endif | |
730 | { | |
731 | msg("Cannot open output \"%s\".\n", tape); | |
732 | if (!query("Do you want to retry the open?")) | |
733 | dumpabort(0); | |
734 | } | |
735 | ||
736 | enslave(); /* Share open tape file descriptor with slaves */ | |
737 | ||
738 | asize = 0; | |
739 | blocksthisvol = 0; | |
740 | if (top) | |
741 | newtape++; /* new tape signal */ | |
b45f51d6 | 742 | spcl.c_count = slp->count; |
1227625a SP |
743 | /* |
744 | * measure firstrec in TP_BSIZE units since restore doesn't | |
745 | * know the correct ntrec value... | |
746 | */ | |
747 | spcl.c_firstrec = slp->firstrec; | |
748 | spcl.c_volume++; | |
749 | spcl.c_type = TS_TAPE; | |
750 | spcl.c_flags |= DR_NEWHEADER; | |
751 | writeheader((ino_t)slp->inode); | |
752 | spcl.c_flags &=~ DR_NEWHEADER; | |
ddd2ef55 | 753 | msg("Volume %d started at: %s", tapeno, ctime(&tstart_volume)); |
1227625a SP |
754 | if (tapeno > 1) |
755 | msg("Volume %d begins with blocks from inode %d\n", | |
756 | tapeno, slp->inode); | |
757 | } | |
758 | } | |
759 | ||
760 | void | |
ddd2ef55 | 761 | dumpabort(int signo) |
1227625a SP |
762 | { |
763 | ||
764 | if (master != 0 && master != getpid()) | |
765 | /* Signals master to call dumpabort */ | |
766 | (void) kill(master, SIGTERM); | |
767 | else { | |
768 | killall(); | |
769 | msg("The ENTIRE dump is aborted.\n"); | |
770 | } | |
771 | #ifdef RDUMP | |
772 | rmtclose(); | |
773 | #endif | |
774 | Exit(X_ABORT); | |
775 | } | |
776 | ||
b45f51d6 | 777 | void |
ddd2ef55 | 778 | Exit(int status) |
1227625a SP |
779 | { |
780 | ||
781 | #ifdef TDEBUG | |
782 | msg("pid = %d exits with status %d\n", getpid(), status); | |
783 | #endif /* TDEBUG */ | |
784 | exit(status); | |
785 | } | |
786 | ||
787 | /* | |
788 | * proceed - handler for SIGUSR2, used to synchronize IO between the slaves. | |
789 | */ | |
ddd2ef55 SP |
790 | static void |
791 | proceed(int signo) | |
1227625a SP |
792 | { |
793 | ||
794 | if (ready) | |
ddd2ef55 | 795 | siglongjmp(jmpbuf, 1); |
1227625a SP |
796 | caught++; |
797 | } | |
798 | ||
799 | void | |
ddd2ef55 | 800 | enslave(void) |
1227625a SP |
801 | { |
802 | int cmd[2]; | |
803 | #ifdef LINUX_FORK_BUG | |
804 | int i, j; | |
805 | #else | |
806 | register int i, j; | |
807 | #endif | |
808 | ||
809 | master = getpid(); | |
810 | ||
811 | signal(SIGTERM, dumpabort); /* Slave sends SIGTERM on dumpabort() */ | |
812 | signal(SIGPIPE, sigpipe); | |
813 | signal(SIGUSR1, tperror); /* Slave sends SIGUSR1 on tape errors */ | |
814 | signal(SIGUSR2, proceed); /* Slave sends SIGUSR2 to next slave */ | |
815 | ||
816 | for (i = 0; i < SLAVES; i++) { | |
817 | if (i == slp - &slaves[0]) { | |
818 | caught = 1; | |
819 | } else { | |
820 | caught = 0; | |
821 | } | |
822 | ||
823 | if (socketpair(AF_UNIX, SOCK_STREAM, 0, cmd) < 0 || | |
824 | (slaves[i].pid = fork()) < 0) | |
825 | quit("too many slaves, %d (recompile smaller): %s\n", | |
826 | i, strerror(errno)); | |
827 | ||
828 | slaves[i].fd = cmd[1]; | |
829 | slaves[i].sent = 0; | |
830 | if (slaves[i].pid == 0) { /* Slave starts up here */ | |
831 | for (j = 0; j <= i; j++) | |
832 | (void) close(slaves[j].fd); | |
833 | signal(SIGINT, SIG_IGN); /* Master handles this */ | |
ddd2ef55 SP |
834 | #if defined(SIGINFO) |
835 | signal(SIGINFO, SIG_IGN); | |
836 | #endif | |
837 | ||
1227625a | 838 | #ifdef LINUX_FORK_BUG |
ddd2ef55 | 839 | if (atomic_write( cmd[0], (char *) &i, sizeof i) |
1227625a SP |
840 | != sizeof i) |
841 | quit("master/slave protocol botched 3\n"); | |
842 | #endif | |
843 | doslave(cmd[0], i); | |
844 | Exit(X_FINOK); | |
845 | } | |
846 | } | |
847 | ||
848 | #ifdef LINUX_FORK_BUG | |
849 | /* | |
850 | * Wait for all slaves to _actually_ start to circumvent a bug in | |
851 | * Linux kernels >= 2.1.3 where a signal sent to a child that hasn't | |
852 | * returned from fork() causes a SEGV in the child process | |
853 | */ | |
854 | for (i = 0; i < SLAVES; i++) | |
ddd2ef55 | 855 | if (atomic_read( slaves[i].fd, (char *) &j, sizeof j) != sizeof j) |
1227625a SP |
856 | quit("master/slave protocol botched 4\n"); |
857 | #endif | |
858 | ||
1227625a | 859 | for (i = 0; i < SLAVES; i++) |
ddd2ef55 | 860 | (void) atomic_write( slaves[i].fd, |
1227625a SP |
861 | (char *) &slaves[(i + 1) % SLAVES].pid, |
862 | sizeof slaves[0].pid); | |
863 | ||
864 | master = 0; | |
865 | } | |
866 | ||
867 | void | |
ddd2ef55 | 868 | killall(void) |
1227625a SP |
869 | { |
870 | register int i; | |
871 | ||
872 | for (i = 0; i < SLAVES; i++) | |
b45f51d6 | 873 | if (slaves[i].pid > 0) { |
1227625a | 874 | (void) kill(slaves[i].pid, SIGKILL); |
b45f51d6 SP |
875 | slaves[i].sent = 0; |
876 | } | |
1227625a SP |
877 | } |
878 | ||
879 | /* | |
880 | * Synchronization - each process has a lockfile, and shares file | |
881 | * descriptors to the following process's lockfile. When our write | |
882 | * completes, we release our lock on the following process's lock- | |
883 | * file, allowing the following process to lock it and proceed. We | |
884 | * get the lock back for the next cycle by swapping descriptors. | |
885 | */ | |
886 | static void | |
ddd2ef55 | 887 | doslave(int cmd, int slave_number) |
1227625a SP |
888 | { |
889 | register int nread; | |
ddd2ef55 SP |
890 | int nextslave, size, eot_count; |
891 | volatile int wrote = 0; | |
892 | sigset_t sigset; | |
1227625a SP |
893 | #ifdef __linux__ |
894 | errcode_t retval; | |
895 | #endif | |
896 | ||
1227625a SP |
897 | /* |
898 | * Need our own seek pointer. | |
899 | */ | |
900 | (void) close(diskfd); | |
901 | if ((diskfd = open(disk, O_RDONLY)) < 0) | |
902 | quit("slave couldn't reopen disk: %s\n", strerror(errno)); | |
903 | #ifdef __linux__ | |
904 | ext2fs_close(fs); | |
905 | retval = ext2fs_open(disk, 0, 0, 0, unix_io_manager, &fs); | |
906 | if (retval) | |
907 | quit("slave couldn't reopen disk: %s\n", strerror(errno)); | |
908 | #endif /* __linux__ */ | |
909 | ||
910 | /* | |
911 | * Need the pid of the next slave in the loop... | |
912 | */ | |
ddd2ef55 | 913 | if ((nread = atomic_read( cmd, (char *)&nextslave, sizeof nextslave)) |
1227625a SP |
914 | != sizeof nextslave) { |
915 | quit("master/slave protocol botched - didn't get pid of next slave.\n"); | |
916 | } | |
917 | ||
918 | /* | |
919 | * Get list of blocks to dump, read the blocks into tape buffer | |
920 | */ | |
ddd2ef55 | 921 | while ((nread = atomic_read( cmd, (char *)slp->req, reqsiz)) == reqsiz) { |
1227625a SP |
922 | register struct req *p = slp->req; |
923 | ||
924 | for (trecno = 0; trecno < ntrec; | |
925 | trecno += p->count, p += p->count) { | |
926 | if (p->dblk) { | |
927 | bread(p->dblk, slp->tblock[trecno], | |
928 | p->count * TP_BSIZE); | |
929 | } else { | |
ddd2ef55 | 930 | if (p->count != 1 || atomic_read( cmd, |
b45f51d6 | 931 | (char *)slp->tblock[trecno], |
1227625a SP |
932 | TP_BSIZE) != TP_BSIZE) |
933 | quit("master/slave protocol botched.\n"); | |
934 | } | |
935 | } | |
936 | if (setjmp(jmpbuf) == 0) { | |
937 | ready = 1; | |
938 | if (!caught) | |
939 | (void) pause(); | |
940 | } | |
941 | ready = 0; | |
942 | caught = 0; | |
943 | ||
944 | /* Try to write the data... */ | |
b45f51d6 | 945 | wrote = 0; |
1227625a SP |
946 | eot_count = 0; |
947 | size = 0; | |
948 | ||
949 | while (eot_count < 10 && size < writesize) { | |
950 | #ifdef RDUMP | |
951 | if (host) | |
952 | wrote = rmtwrite(slp->tblock[0]+size, | |
953 | writesize-size); | |
954 | else | |
955 | #endif | |
956 | wrote = write(tapefd, slp->tblock[0]+size, | |
957 | writesize-size); | |
958 | #ifdef WRITEDEBUG | |
b45f51d6 | 959 | printf("slave %d wrote %d\n", slave_number, wrote); |
1227625a | 960 | #endif |
b45f51d6 | 961 | if (wrote < 0) |
1227625a SP |
962 | break; |
963 | if (wrote == 0) | |
964 | eot_count++; | |
965 | size += wrote; | |
966 | } | |
967 | ||
968 | #ifdef WRITEDEBUG | |
b45f51d6 SP |
969 | if (size != writesize) |
970 | printf("slave %d only wrote %d out of %d bytes and gave up.\n", | |
1227625a SP |
971 | slave_number, size, writesize); |
972 | #endif | |
973 | ||
b45f51d6 SP |
974 | /* |
975 | * Handle ENOSPC as an EOT condition. | |
976 | */ | |
977 | if (wrote < 0 && errno == ENOSPC) { | |
978 | wrote = 0; | |
979 | eot_count++; | |
980 | } | |
981 | ||
1227625a SP |
982 | if (eot_count > 0) |
983 | size = 0; | |
984 | ||
b45f51d6 | 985 | if (wrote < 0) { |
1227625a | 986 | (void) kill(master, SIGUSR1); |
ddd2ef55 | 987 | sigemptyset(&sigset); |
1227625a | 988 | for (;;) |
ddd2ef55 | 989 | sigsuspend(&sigset); |
1227625a SP |
990 | } else { |
991 | /* | |
992 | * pass size of write back to master | |
993 | * (for EOT handling) | |
994 | */ | |
ddd2ef55 | 995 | (void) atomic_write( cmd, (char *)&size, sizeof size); |
b45f51d6 | 996 | } |
1227625a SP |
997 | |
998 | /* | |
999 | * If partial write, don't want next slave to go. | |
1000 | * Also jolts him awake. | |
1001 | */ | |
1002 | (void) kill(nextslave, SIGUSR2); | |
1003 | } | |
1004 | if (nread != 0) | |
1005 | quit("error reading command pipe: %s\n", strerror(errno)); | |
1006 | } | |
1007 | ||
1008 | /* | |
1009 | * Since a read from a pipe may not return all we asked for, | |
1010 | * or a write may not write all we ask if we get a signal, | |
1011 | * loop until the count is satisfied (or error). | |
1012 | */ | |
ddd2ef55 SP |
1013 | static ssize_t |
1014 | atomic_read(int fd, void *buf, size_t count) | |
1015 | { | |
1016 | int got, need = count; | |
1017 | ||
1018 | while ((got = read(fd, buf, need)) > 0 && (need -= got) > 0) | |
1019 | (char *)buf += got; | |
1020 | return (got < 0 ? got : count - need); | |
1021 | } | |
1022 | ||
1023 | /* | |
1024 | * Since a read from a pipe may not return all we asked for, | |
1025 | * or a write may not write all we ask if we get a signal, | |
1026 | * loop until the count is satisfied (or error). | |
1027 | */ | |
1028 | static ssize_t | |
1029 | atomic_write(int fd, const void *buf, size_t count) | |
1227625a SP |
1030 | { |
1031 | int got, need = count; | |
1032 | ||
ddd2ef55 SP |
1033 | while ((got = write(fd, buf, need)) > 0 && (need -= got) > 0) |
1034 | (char *)buf += got; | |
1227625a SP |
1035 | return (got < 0 ? got : count - need); |
1036 | } |