]>
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 | 4 | * Remy Card <card@Linux.EU.Org>, 1994-1997 |
df9ae507 | 5 | * Stelian Pop <pop@cybercable.fr>, 1999 |
1227625a SP |
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 | ||
df9ae507 SP |
41 | #ifndef lint |
42 | static const char rcsid[] = | |
d1c73b9a | 43 | "$Id: tape.c,v 1.9 1999/11/21 16:01:47 tiniou Exp $"; |
df9ae507 SP |
44 | #endif /* not lint */ |
45 | ||
b45f51d6 SP |
46 | #ifdef __linux__ |
47 | #include <sys/types.h> | |
48 | #include <linux/types.h> | |
49 | #endif | |
1227625a SP |
50 | #include <sys/param.h> |
51 | #include <sys/socket.h> | |
52 | #include <sys/time.h> | |
53 | #include <sys/wait.h> | |
54 | #ifdef __linux__ | |
55 | #include <linux/ext2_fs.h> | |
56 | #include <bsdcompat.h> | |
57 | #else /* __linux__ */ | |
58 | #ifdef sunos | |
59 | #include <sys/vnode.h> | |
60 | ||
61 | #include <ufs/fs.h> | |
62 | #include <ufs/inode.h> | |
63 | #else | |
64 | #include <ufs/ufs/dinode.h> | |
65 | #include <ufs/ffs/fs.h> | |
66 | #endif | |
67 | #endif /* __linux__ */ | |
68 | ||
69 | #include <protocols/dumprestore.h> | |
70 | ||
71 | #include <errno.h> | |
72 | #include <fcntl.h> | |
73 | #include <setjmp.h> | |
74 | #include <signal.h> | |
75 | #include <stdio.h> | |
76 | #ifdef __STDC__ | |
77 | #include <stdlib.h> | |
78 | #include <string.h> | |
79 | #include <unistd.h> | |
80 | #else | |
81 | int write(), read(); | |
82 | #endif | |
83 | ||
c930abff | 84 | #ifdef __linux__ |
1227625a SP |
85 | #include <ext2fs/ext2fs.h> |
86 | #endif | |
c930abff | 87 | |
1227625a | 88 | #include "dump.h" |
1227625a SP |
89 | |
90 | int writesize; /* size of malloc()ed buffer for tape */ | |
91 | long lastspclrec = -1; /* tape block number of last written header */ | |
92 | int trecno = 0; /* next record to write in current block */ | |
93 | extern long blocksperfile; /* number of blocks per output file */ | |
94 | long blocksthisvol; /* number of blocks on current output file */ | |
95 | extern int ntrec; /* blocking factor on tape */ | |
96 | extern int cartridge; | |
97 | extern char *host; | |
98 | char *nexttape; | |
f34aca37 | 99 | extern pid_t rshpid; |
1227625a | 100 | |
ddd2ef55 SP |
101 | static ssize_t atomic_read __P((int, void *, size_t)); |
102 | static ssize_t atomic_write __P((int, const void *, size_t)); | |
1227625a SP |
103 | static void doslave __P((int, int)); |
104 | static void enslave __P((void)); | |
105 | static void flushtape __P((void)); | |
106 | static void killall __P((void)); | |
107 | static void rollforward __P((void)); | |
108 | ||
109 | /* | |
110 | * Concurrent dump mods (Caltech) - disk block reading and tape writing | |
111 | * are exported to several slave processes. While one slave writes the | |
112 | * tape, the others read disk blocks; they pass control of the tape in | |
113 | * a ring via signals. The parent process traverses the filesystem and | |
114 | * sends writeheader()'s and lists of daddr's to the slaves via pipes. | |
115 | * The following structure defines the instruction packets sent to slaves. | |
116 | */ | |
117 | struct req { | |
118 | daddr_t dblk; | |
119 | int count; | |
120 | }; | |
121 | int reqsiz; | |
122 | ||
123 | #define SLAVES 3 /* 1 slave writing, 1 reading, 1 for slack */ | |
124 | struct slave { | |
125 | int tapea; /* header number at start of this chunk */ | |
126 | int count; /* count to next header (used for TS_TAPE */ | |
127 | /* after EOT) */ | |
128 | int inode; /* inode that we are currently dealing with */ | |
129 | int fd; /* FD for this slave */ | |
130 | int pid; /* PID for this slave */ | |
131 | int sent; /* 1 == we've sent this slave requests */ | |
132 | int firstrec; /* record number of this block */ | |
133 | char (*tblock)[TP_BSIZE]; /* buffer for data blocks */ | |
134 | struct req *req; /* buffer for requests */ | |
135 | } slaves[SLAVES+1]; | |
136 | struct slave *slp; | |
137 | ||
138 | char (*nextblock)[TP_BSIZE]; | |
139 | ||
ddd2ef55 SP |
140 | static time_t tstart_volume; /* time of volume start */ |
141 | static int tapea_volume; /* value of spcl.c_tapea at volume start */ | |
142 | ||
1227625a SP |
143 | int master; /* pid of master, for sending error signals */ |
144 | int tenths; /* length of tape used per block written */ | |
145 | static int caught; /* have we caught the signal to proceed? */ | |
146 | static int ready; /* have we reached the lock point without having */ | |
147 | /* received the SIGUSR2 signal from the prev slave? */ | |
ddd2ef55 | 148 | static sigjmp_buf jmpbuf; /* where to jump to if we are ready when the */ |
1227625a SP |
149 | /* SIGUSR2 arrives from the previous slave */ |
150 | ||
151 | int | |
ddd2ef55 | 152 | alloctape(void) |
1227625a SP |
153 | { |
154 | int pgoff = getpagesize() - 1; | |
155 | char *buf; | |
156 | int i; | |
157 | ||
158 | writesize = ntrec * TP_BSIZE; | |
159 | reqsiz = (ntrec + 1) * sizeof(struct req); | |
160 | /* | |
161 | * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode | |
162 | * (see DEC TU80 User's Guide). The shorter gaps of 6250-bpi require | |
163 | * repositioning after stopping, i.e, streaming mode, where the gap is | |
164 | * variable, 0.30" to 0.45". The gap is maximal when the tape stops. | |
165 | */ | |
b45f51d6 | 166 | if (blocksperfile == 0 && !unlimited) |
1227625a SP |
167 | tenths = writesize / density + |
168 | (cartridge ? 16 : density == 625 ? 5 : 8); | |
169 | /* | |
170 | * Allocate tape buffer contiguous with the array of instruction | |
171 | * packets, so flushtape() can write them together with one write(). | |
172 | * Align tape buffer on page boundary to speed up tape write(). | |
173 | */ | |
174 | for (i = 0; i <= SLAVES; i++) { | |
175 | buf = (char *) | |
176 | malloc((unsigned)(reqsiz + writesize + pgoff + TP_BSIZE)); | |
177 | if (buf == NULL) | |
178 | return(0); | |
179 | slaves[i].tblock = (char (*)[TP_BSIZE]) | |
180 | #ifdef __linux__ | |
181 | (((long)&buf[reqsiz] + pgoff) &~ pgoff); | |
182 | #else | |
183 | (((long)&buf[ntrec + 1] + pgoff) &~ pgoff); | |
184 | #endif | |
185 | slaves[i].req = (struct req *)slaves[i].tblock - ntrec - 1; | |
186 | } | |
187 | slp = &slaves[0]; | |
188 | slp->count = 1; | |
189 | slp->tapea = 0; | |
190 | slp->firstrec = 0; | |
191 | nextblock = slp->tblock; | |
192 | return(1); | |
193 | } | |
194 | ||
195 | void | |
ddd2ef55 | 196 | writerec(const void *dp, int isspcl) |
1227625a SP |
197 | { |
198 | ||
199 | slp->req[trecno].dblk = (daddr_t)0; | |
200 | slp->req[trecno].count = 1; | |
ddd2ef55 SP |
201 | /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */ |
202 | *(union u_spcl *)(*(nextblock)) = *(union u_spcl *)dp; | |
203 | nextblock++; | |
1227625a SP |
204 | if (isspcl) |
205 | lastspclrec = spcl.c_tapea; | |
206 | trecno++; | |
207 | spcl.c_tapea++; | |
208 | if (trecno >= ntrec) | |
209 | flushtape(); | |
210 | } | |
211 | ||
212 | void | |
ddd2ef55 | 213 | dumpblock(daddr_t blkno, int size) |
1227625a SP |
214 | { |
215 | int avail, tpblks, dblkno; | |
216 | ||
217 | dblkno = fsbtodb(sblock, blkno); | |
218 | tpblks = size >> tp_bshift; | |
219 | while ((avail = MIN(tpblks, ntrec - trecno)) > 0) { | |
220 | slp->req[trecno].dblk = dblkno; | |
221 | slp->req[trecno].count = avail; | |
222 | trecno += avail; | |
223 | spcl.c_tapea += avail; | |
224 | if (trecno >= ntrec) | |
225 | flushtape(); | |
226 | dblkno += avail << (tp_bshift - dev_bshift); | |
227 | tpblks -= avail; | |
228 | } | |
229 | } | |
230 | ||
231 | int nogripe = 0; | |
232 | ||
ddd2ef55 SP |
233 | static void |
234 | tperror(int signo) | |
1227625a SP |
235 | { |
236 | ||
237 | if (pipeout) { | |
238 | msg("write error on %s\n", tape); | |
239 | quit("Cannot recover\n"); | |
240 | /* NOTREACHED */ | |
241 | } | |
242 | msg("write error %d blocks into volume %d\n", blocksthisvol, tapeno); | |
243 | broadcast("DUMP WRITE ERROR!\n"); | |
244 | if (!query("Do you want to restart?")) | |
245 | dumpabort(0); | |
246 | msg("Closing this volume. Prepare to restart with new media;\n"); | |
247 | msg("this dump volume will be rewritten.\n"); | |
248 | killall(); | |
249 | nogripe = 1; | |
250 | close_rewind(); | |
251 | Exit(X_REWRITE); | |
252 | } | |
253 | ||
ddd2ef55 SP |
254 | static void |
255 | sigpipe(int signo) | |
1227625a SP |
256 | { |
257 | ||
258 | quit("Broken pipe\n"); | |
259 | } | |
260 | ||
ddd2ef55 SP |
261 | /* |
262 | * do_stats -- | |
263 | * Update xferrate stats | |
264 | */ | |
265 | time_t | |
266 | do_stats(void) | |
267 | { | |
268 | time_t tnow, ttaken; | |
269 | int blocks; | |
270 | ||
8d4197bb SP |
271 | #ifdef __linux__ |
272 | (void)time4(&tnow); | |
273 | #else | |
ddd2ef55 | 274 | (void)time(&tnow); |
8d4197bb | 275 | #endif |
ddd2ef55 SP |
276 | ttaken = tnow - tstart_volume; |
277 | blocks = spcl.c_tapea - tapea_volume; | |
8d4197bb SP |
278 | msg("Volume %d completed at: %s", tapeno, |
279 | #ifdef __linux__ | |
280 | ctime4(&tnow)); | |
281 | #else | |
282 | ctime(&tnow)); | |
283 | #endif | |
ddd2ef55 SP |
284 | if (ttaken > 0) { |
285 | msg("Volume %d took %d:%02d:%02d\n", tapeno, | |
286 | ttaken / 3600, (ttaken % 3600) / 60, ttaken % 60); | |
287 | msg("Volume %d transfer rate: %ld KB/s\n", tapeno, | |
288 | blocks / ttaken); | |
289 | xferrate += blocks / ttaken; | |
290 | } | |
291 | return(tnow); | |
292 | } | |
293 | ||
294 | #if defined(SIGINFO) | |
295 | /* | |
296 | * statussig -- | |
297 | * information message upon receipt of SIGINFO | |
298 | * (derived from optr.c::timeest()) | |
299 | */ | |
300 | void | |
301 | statussig(int notused) | |
302 | { | |
303 | time_t tnow, deltat; | |
304 | char msgbuf[128]; | |
305 | int save_errno = errno; | |
306 | ||
307 | if (blockswritten < 500) | |
308 | return; | |
8d4197bb SP |
309 | #ifdef __linux__ |
310 | (void) time4(&tnow); | |
311 | #else | |
ddd2ef55 | 312 | (void) time((time_t *) &tnow); |
8d4197bb | 313 | #endif |
ddd2ef55 SP |
314 | deltat = tstart_writing - tnow + (1.0 * (tnow - tstart_writing)) |
315 | / blockswritten * tapesize; | |
316 | (void)snprintf(msgbuf, sizeof(msgbuf), | |
317 | "%3.2f%% done at %ld KB/s, finished in %d:%02d\n", | |
318 | (blockswritten * 100.0) / tapesize, | |
319 | (spcl.c_tapea - tapea_volume) / (tnow - tstart_volume), | |
320 | (int)(deltat / 3600), (int)((deltat % 3600) / 60)); | |
321 | write(STDERR_FILENO, msgbuf, strlen(msgbuf)); | |
322 | errno = save_errno; | |
323 | } | |
324 | #endif | |
325 | ||
1227625a | 326 | static void |
ddd2ef55 | 327 | flushtape(void) |
1227625a SP |
328 | { |
329 | int i, blks, got; | |
330 | long lastfirstrec; | |
331 | ||
332 | int siz = (char *)nextblock - (char *)slp->req; | |
333 | ||
334 | slp->req[trecno].count = 0; /* Sentinel */ | |
335 | ||
ddd2ef55 | 336 | if (atomic_write( slp->fd, (char *)slp->req, siz) != siz) |
1227625a SP |
337 | quit("error writing command pipe: %s\n", strerror(errno)); |
338 | slp->sent = 1; /* we sent a request, read the response later */ | |
339 | ||
340 | lastfirstrec = slp->firstrec; | |
341 | ||
342 | if (++slp >= &slaves[SLAVES]) | |
343 | slp = &slaves[0]; | |
344 | ||
345 | /* Read results back from next slave */ | |
346 | if (slp->sent) { | |
ddd2ef55 | 347 | if (atomic_read( slp->fd, (char *)&got, sizeof got) |
1227625a SP |
348 | != sizeof got) { |
349 | perror(" DUMP: error reading command pipe in master"); | |
350 | dumpabort(0); | |
351 | } | |
352 | slp->sent = 0; | |
353 | ||
354 | /* Check for end of tape */ | |
355 | if (got < writesize) { | |
356 | msg("End of tape detected\n"); | |
357 | ||
358 | /* | |
359 | * Drain the results, don't care what the values were. | |
360 | * If we read them here then trewind won't... | |
361 | */ | |
362 | for (i = 0; i < SLAVES; i++) { | |
363 | if (slaves[i].sent) { | |
ddd2ef55 | 364 | if (atomic_read( slaves[i].fd, |
1227625a SP |
365 | (char *)&got, sizeof got) |
366 | != sizeof got) { | |
367 | perror(" DUMP: error reading command pipe in master"); | |
368 | dumpabort(0); | |
369 | } | |
370 | slaves[i].sent = 0; | |
371 | } | |
372 | } | |
373 | ||
374 | close_rewind(); | |
375 | rollforward(); | |
376 | return; | |
377 | } | |
378 | } | |
379 | ||
380 | blks = 0; | |
381 | if (spcl.c_type != TS_END) { | |
382 | for (i = 0; i < spcl.c_count; i++) | |
383 | if (spcl.c_addr[i] != 0) | |
384 | blks++; | |
385 | } | |
386 | slp->count = lastspclrec + blks + 1 - spcl.c_tapea; | |
387 | slp->tapea = spcl.c_tapea; | |
388 | slp->firstrec = lastfirstrec + ntrec; | |
389 | slp->inode = curino; | |
390 | nextblock = slp->tblock; | |
391 | trecno = 0; | |
392 | asize += tenths; | |
393 | blockswritten += ntrec; | |
394 | blocksthisvol += ntrec; | |
b45f51d6 | 395 | if (!pipeout && !unlimited && (blocksperfile ? |
1227625a SP |
396 | (blocksthisvol >= blocksperfile) : (asize > tsize))) { |
397 | close_rewind(); | |
398 | startnewtape(0); | |
399 | } | |
400 | timeest(); | |
401 | } | |
402 | ||
403 | void | |
ddd2ef55 | 404 | trewind(void) |
1227625a SP |
405 | { |
406 | int f; | |
407 | int got; | |
408 | ||
409 | for (f = 0; f < SLAVES; f++) { | |
410 | /* | |
b45f51d6 SP |
411 | * Drain the results, but unlike EOT we DO (or should) care |
412 | * what the return values were, since if we detect EOT after | |
413 | * we think we've written the last blocks to the tape anyway, | |
1227625a SP |
414 | * we have to replay those blocks with rollforward. |
415 | * | |
b45f51d6 | 416 | * fixme: punt for now. |
1227625a SP |
417 | */ |
418 | if (slaves[f].sent) { | |
ddd2ef55 | 419 | if (atomic_read( slaves[f].fd, (char *)&got, sizeof got) |
1227625a SP |
420 | != sizeof got) { |
421 | perror(" DUMP: error reading command pipe in master"); | |
422 | dumpabort(0); | |
423 | } | |
424 | slaves[f].sent = 0; | |
425 | if (got != writesize) { | |
426 | msg("EOT detected in last 2 tape records!\n"); | |
427 | msg("Use a longer tape, decrease the size estimate\n"); | |
428 | quit("or use no size estimate at all.\n"); | |
429 | } | |
430 | } | |
431 | (void) close(slaves[f].fd); | |
432 | } | |
433 | while (wait((int *)NULL) >= 0) /* wait for any signals from slaves */ | |
434 | /* void */; | |
435 | ||
436 | if (pipeout) | |
437 | return; | |
438 | ||
439 | msg("Closing %s\n", tape); | |
440 | ||
441 | #ifdef RDUMP | |
442 | if (host) { | |
443 | rmtclose(); | |
444 | while (rmtopen(tape, 0) < 0) | |
445 | sleep(10); | |
446 | rmtclose(); | |
447 | return; | |
448 | } | |
449 | #endif | |
450 | (void) close(tapefd); | |
451 | while ((f = open(tape, 0)) < 0) | |
452 | sleep (10); | |
453 | (void) close(f); | |
454 | } | |
455 | ||
456 | void | |
ddd2ef55 | 457 | close_rewind(void) |
1227625a SP |
458 | { |
459 | trewind(); | |
ddd2ef55 | 460 | (void)do_stats(); |
d1c73b9a | 461 | if (nexttape || Mflag) |
1227625a SP |
462 | return; |
463 | if (!nogripe) { | |
464 | msg("Change Volumes: Mount volume #%d\n", tapeno+1); | |
465 | broadcast("CHANGE DUMP VOLUMES!\7\7\n"); | |
466 | } | |
467 | while (!query("Is the new volume mounted and ready to go?")) | |
468 | if (query("Do you want to abort?")) { | |
469 | dumpabort(0); | |
470 | /*NOTREACHED*/ | |
471 | } | |
472 | } | |
473 | ||
474 | void | |
ddd2ef55 | 475 | rollforward(void) |
1227625a SP |
476 | { |
477 | register struct req *p, *q, *prev; | |
478 | register struct slave *tslp; | |
479 | int i, size, savedtapea, got; | |
480 | union u_spcl *ntb, *otb; | |
b45f51d6 SP |
481 | #ifdef __linux__ |
482 | int blks; | |
483 | long lastfirstrec; | |
484 | #endif | |
1227625a SP |
485 | tslp = &slaves[SLAVES]; |
486 | ntb = (union u_spcl *)tslp->tblock[1]; | |
487 | ||
488 | /* | |
b45f51d6 SP |
489 | * Each of the N slaves should have requests that need to |
490 | * be replayed on the next tape. Use the extra slave buffers | |
491 | * (slaves[SLAVES]) to construct request lists to be sent to | |
1227625a SP |
492 | * each slave in turn. |
493 | */ | |
494 | for (i = 0; i < SLAVES; i++) { | |
495 | q = &tslp->req[1]; | |
496 | otb = (union u_spcl *)slp->tblock; | |
497 | ||
498 | /* | |
b45f51d6 | 499 | * For each request in the current slave, copy it to tslp. |
1227625a SP |
500 | */ |
501 | ||
502 | prev = NULL; | |
503 | for (p = slp->req; p->count > 0; p += p->count) { | |
504 | *q = *p; | |
505 | if (p->dblk == 0) | |
506 | *ntb++ = *otb++; /* copy the datablock also */ | |
507 | prev = q; | |
508 | q += q->count; | |
509 | } | |
510 | if (prev == NULL) | |
511 | quit("rollforward: protocol botch"); | |
512 | if (prev->dblk != 0) | |
513 | prev->count -= 1; | |
514 | else | |
515 | ntb--; | |
516 | q -= 1; | |
517 | q->count = 0; | |
518 | q = &tslp->req[0]; | |
519 | if (i == 0) { | |
520 | q->dblk = 0; | |
521 | q->count = 1; | |
522 | trecno = 0; | |
523 | nextblock = tslp->tblock; | |
524 | savedtapea = spcl.c_tapea; | |
525 | spcl.c_tapea = slp->tapea; | |
526 | startnewtape(0); | |
527 | spcl.c_tapea = savedtapea; | |
528 | lastspclrec = savedtapea - 1; | |
529 | } | |
530 | size = (char *)ntb - (char *)q; | |
ddd2ef55 | 531 | if (atomic_write( slp->fd, (char *)q, size) != size) { |
1227625a SP |
532 | perror(" DUMP: error writing command pipe"); |
533 | dumpabort(0); | |
534 | } | |
535 | slp->sent = 1; | |
b45f51d6 SP |
536 | #ifdef __linux__ |
537 | lastfirstrec = slp->firstrec; | |
538 | #endif | |
1227625a SP |
539 | if (++slp >= &slaves[SLAVES]) |
540 | slp = &slaves[0]; | |
541 | ||
542 | q->count = 1; | |
543 | ||
544 | if (prev->dblk != 0) { | |
545 | /* | |
b45f51d6 SP |
546 | * If the last one was a disk block, make the |
547 | * first of this one be the last bit of that disk | |
1227625a SP |
548 | * block... |
549 | */ | |
550 | q->dblk = prev->dblk + | |
551 | prev->count * (TP_BSIZE / DEV_BSIZE); | |
552 | ntb = (union u_spcl *)tslp->tblock; | |
553 | } else { | |
554 | /* | |
b45f51d6 | 555 | * It wasn't a disk block. Copy the data to its |
1227625a SP |
556 | * new location in the buffer. |
557 | */ | |
558 | q->dblk = 0; | |
559 | *((union u_spcl *)tslp->tblock) = *ntb; | |
560 | ntb = (union u_spcl *)tslp->tblock[1]; | |
561 | } | |
562 | } | |
563 | slp->req[0] = *q; | |
564 | nextblock = slp->tblock; | |
565 | if (q->dblk == 0) | |
b45f51d6 SP |
566 | #ifdef __linux__ |
567 | *(union u_spcl *)(*(nextblock)++) = *(union u_spcl *)tslp->tblock; | |
568 | #else | |
1227625a | 569 | nextblock++; |
b45f51d6 | 570 | #endif |
1227625a SP |
571 | trecno = 1; |
572 | ||
573 | /* | |
574 | * Clear the first slaves' response. One hopes that it | |
575 | * worked ok, otherwise the tape is much too short! | |
576 | */ | |
577 | if (slp->sent) { | |
ddd2ef55 | 578 | if (atomic_read( slp->fd, (char *)&got, sizeof got) |
1227625a SP |
579 | != sizeof got) { |
580 | perror(" DUMP: error reading command pipe in master"); | |
581 | dumpabort(0); | |
582 | } | |
583 | slp->sent = 0; | |
584 | ||
585 | if (got != writesize) { | |
586 | quit("EOT detected at start of the tape!\n"); | |
587 | } | |
588 | } | |
b45f51d6 SP |
589 | |
590 | #ifdef __linux__ | |
591 | blks = 0; | |
592 | if (spcl.c_type != TS_END) { | |
593 | for (i = 0; i < spcl.c_count; i++) | |
594 | if (spcl.c_addr[i] != 0) | |
595 | blks++; | |
596 | } | |
597 | ||
598 | slp->firstrec = lastfirstrec + ntrec; | |
599 | slp->count = lastspclrec + blks + 1 - spcl.c_tapea; | |
600 | slp->inode = curino; | |
601 | asize += tenths; | |
602 | blockswritten += ntrec; | |
603 | blocksthisvol += ntrec; | |
604 | #endif | |
1227625a SP |
605 | } |
606 | ||
607 | /* | |
608 | * We implement taking and restoring checkpoints on the tape level. | |
609 | * When each tape is opened, a new process is created by forking; this | |
610 | * saves all of the necessary context in the parent. The child | |
611 | * continues the dump; the parent waits around, saving the context. | |
612 | * If the child returns X_REWRITE, then it had problems writing that tape; | |
613 | * this causes the parent to fork again, duplicating the context, and | |
614 | * everything continues as if nothing had happened. | |
615 | */ | |
616 | void | |
ddd2ef55 | 617 | startnewtape(int top) |
1227625a SP |
618 | { |
619 | int parentpid; | |
620 | int childpid; | |
621 | int status; | |
622 | int waitpid; | |
623 | char *p; | |
624 | #ifdef __linux__ | |
ddd2ef55 | 625 | void (*interrupt_save) __P((int signo)); |
1227625a SP |
626 | #else /* __linux__ */ |
627 | #ifdef sunos | |
628 | void (*interrupt_save)(); | |
629 | #else | |
630 | sig_t interrupt_save; | |
631 | #endif | |
632 | #endif /* __linux__ */ | |
633 | ||
634 | interrupt_save = signal(SIGINT, SIG_IGN); | |
635 | parentpid = getpid(); | |
ddd2ef55 | 636 | tapea_volume = spcl.c_tapea; |
8d4197bb SP |
637 | #ifdef __linux__ |
638 | (void)time4(&tstart_volume); | |
639 | #else | |
640 | (void)time((&tstart_volume); | |
641 | #endif | |
1227625a SP |
642 | |
643 | restore_check_point: | |
644 | (void)signal(SIGINT, interrupt_save); | |
645 | /* | |
646 | * All signals are inherited... | |
647 | */ | |
648 | childpid = fork(); | |
649 | if (childpid < 0) { | |
650 | msg("Context save fork fails in parent %d\n", parentpid); | |
651 | Exit(X_ABORT); | |
652 | } | |
653 | if (childpid != 0) { | |
654 | /* | |
655 | * PARENT: | |
656 | * save the context by waiting | |
657 | * until the child doing all of the work returns. | |
658 | * don't catch the interrupt | |
659 | */ | |
660 | signal(SIGINT, SIG_IGN); | |
661 | #ifdef TDEBUG | |
662 | msg("Tape: %d; parent process: %d child process %d\n", | |
663 | tapeno+1, parentpid, childpid); | |
664 | #endif /* TDEBUG */ | |
665 | while ((waitpid = wait(&status)) != childpid) | |
f34aca37 SP |
666 | if (waitpid != rshpid) |
667 | msg("Parent %d waiting for child %d has another child %d return\n", | |
1227625a SP |
668 | parentpid, childpid, waitpid); |
669 | if (status & 0xFF) { | |
670 | msg("Child %d returns LOB status %o\n", | |
671 | childpid, status&0xFF); | |
672 | } | |
673 | status = (status >> 8) & 0xFF; | |
674 | #ifdef TDEBUG | |
675 | switch(status) { | |
676 | case X_FINOK: | |
677 | msg("Child %d finishes X_FINOK\n", childpid); | |
678 | break; | |
b45f51d6 | 679 | case X_ABORT: |
1227625a SP |
680 | msg("Child %d finishes X_ABORT\n", childpid); |
681 | break; | |
682 | case X_REWRITE: | |
683 | msg("Child %d finishes X_REWRITE\n", childpid); | |
684 | break; | |
685 | default: | |
686 | msg("Child %d finishes unknown %d\n", | |
687 | childpid, status); | |
688 | break; | |
689 | } | |
690 | #endif /* TDEBUG */ | |
691 | switch(status) { | |
692 | case X_FINOK: | |
693 | Exit(X_FINOK); | |
694 | case X_ABORT: | |
695 | Exit(X_ABORT); | |
696 | case X_REWRITE: | |
697 | goto restore_check_point; | |
698 | default: | |
699 | msg("Bad return code from dump: %d\n", status); | |
700 | Exit(X_ABORT); | |
701 | } | |
702 | /*NOTREACHED*/ | |
703 | } else { /* we are the child; just continue */ | |
704 | #ifdef TDEBUG | |
705 | sleep(4); /* allow time for parent's message to get out */ | |
706 | msg("Child on Tape %d has parent %d, my pid = %d\n", | |
707 | tapeno+1, parentpid, getpid()); | |
708 | #endif /* TDEBUG */ | |
709 | /* | |
710 | * If we have a name like "/dev/rmt0,/dev/rmt1", | |
711 | * use the name before the comma first, and save | |
712 | * the remaining names for subsequent volumes. | |
713 | */ | |
714 | tapeno++; /* current tape sequence */ | |
d1c73b9a SP |
715 | if (Mflag) { |
716 | snprintf(tape, NAME_MAX, "%s%03d", tapeprefix, tapeno); | |
717 | tape[NAME_MAX - 1] = '\0'; | |
718 | msg("Dumping volume %d on %s\n", tapeno, tape); | |
719 | } | |
720 | else if (nexttape || strchr(tapeprefix, ',')) { | |
1227625a | 721 | if (nexttape && *nexttape) |
d1c73b9a SP |
722 | tapeprefix = nexttape; |
723 | if ((p = strchr(tapeprefix, ',')) != NULL) { | |
1227625a SP |
724 | *p = '\0'; |
725 | nexttape = p + 1; | |
726 | } else | |
727 | nexttape = NULL; | |
d1c73b9a SP |
728 | strncpy(tape, tapeprefix, NAME_MAX); |
729 | tape[NAME_MAX - 1] = '\0'; | |
1227625a SP |
730 | msg("Dumping volume %d on %s\n", tapeno, tape); |
731 | } | |
732 | #ifdef RDUMP | |
733 | while ((tapefd = (host ? rmtopen(tape, 2) : | |
734 | pipeout ? 1 : open(tape, O_WRONLY|O_CREAT, 0666))) < 0) | |
735 | #else | |
b45f51d6 | 736 | while ((tapefd = (pipeout ? 1 : |
1227625a SP |
737 | open(tape, O_WRONLY|O_CREAT, 0666))) < 0) |
738 | #endif | |
739 | { | |
740 | msg("Cannot open output \"%s\".\n", tape); | |
741 | if (!query("Do you want to retry the open?")) | |
742 | dumpabort(0); | |
743 | } | |
744 | ||
745 | enslave(); /* Share open tape file descriptor with slaves */ | |
746 | ||
747 | asize = 0; | |
748 | blocksthisvol = 0; | |
749 | if (top) | |
750 | newtape++; /* new tape signal */ | |
b45f51d6 | 751 | spcl.c_count = slp->count; |
1227625a SP |
752 | /* |
753 | * measure firstrec in TP_BSIZE units since restore doesn't | |
754 | * know the correct ntrec value... | |
755 | */ | |
756 | spcl.c_firstrec = slp->firstrec; | |
757 | spcl.c_volume++; | |
758 | spcl.c_type = TS_TAPE; | |
759 | spcl.c_flags |= DR_NEWHEADER; | |
760 | writeheader((ino_t)slp->inode); | |
761 | spcl.c_flags &=~ DR_NEWHEADER; | |
8d4197bb SP |
762 | msg("Volume %d started at: %s", tapeno, |
763 | #ifdef __linux__ | |
764 | ctime4(&tstart_volume)); | |
765 | #else | |
766 | ctime(&tstart_volume)); | |
767 | #endif | |
1227625a SP |
768 | if (tapeno > 1) |
769 | msg("Volume %d begins with blocks from inode %d\n", | |
770 | tapeno, slp->inode); | |
771 | } | |
772 | } | |
773 | ||
774 | void | |
ddd2ef55 | 775 | dumpabort(int signo) |
1227625a SP |
776 | { |
777 | ||
778 | if (master != 0 && master != getpid()) | |
779 | /* Signals master to call dumpabort */ | |
780 | (void) kill(master, SIGTERM); | |
781 | else { | |
782 | killall(); | |
783 | msg("The ENTIRE dump is aborted.\n"); | |
784 | } | |
785 | #ifdef RDUMP | |
786 | rmtclose(); | |
787 | #endif | |
788 | Exit(X_ABORT); | |
789 | } | |
790 | ||
b45f51d6 | 791 | void |
ddd2ef55 | 792 | Exit(int status) |
1227625a SP |
793 | { |
794 | ||
795 | #ifdef TDEBUG | |
796 | msg("pid = %d exits with status %d\n", getpid(), status); | |
797 | #endif /* TDEBUG */ | |
798 | exit(status); | |
799 | } | |
800 | ||
801 | /* | |
802 | * proceed - handler for SIGUSR2, used to synchronize IO between the slaves. | |
803 | */ | |
ddd2ef55 SP |
804 | static void |
805 | proceed(int signo) | |
1227625a SP |
806 | { |
807 | ||
808 | if (ready) | |
ddd2ef55 | 809 | siglongjmp(jmpbuf, 1); |
1227625a SP |
810 | caught++; |
811 | } | |
812 | ||
813 | void | |
ddd2ef55 | 814 | enslave(void) |
1227625a SP |
815 | { |
816 | int cmd[2]; | |
817 | #ifdef LINUX_FORK_BUG | |
818 | int i, j; | |
819 | #else | |
820 | register int i, j; | |
821 | #endif | |
822 | ||
823 | master = getpid(); | |
824 | ||
825 | signal(SIGTERM, dumpabort); /* Slave sends SIGTERM on dumpabort() */ | |
826 | signal(SIGPIPE, sigpipe); | |
827 | signal(SIGUSR1, tperror); /* Slave sends SIGUSR1 on tape errors */ | |
828 | signal(SIGUSR2, proceed); /* Slave sends SIGUSR2 to next slave */ | |
829 | ||
830 | for (i = 0; i < SLAVES; i++) { | |
831 | if (i == slp - &slaves[0]) { | |
832 | caught = 1; | |
833 | } else { | |
834 | caught = 0; | |
835 | } | |
836 | ||
837 | if (socketpair(AF_UNIX, SOCK_STREAM, 0, cmd) < 0 || | |
838 | (slaves[i].pid = fork()) < 0) | |
839 | quit("too many slaves, %d (recompile smaller): %s\n", | |
840 | i, strerror(errno)); | |
841 | ||
842 | slaves[i].fd = cmd[1]; | |
843 | slaves[i].sent = 0; | |
844 | if (slaves[i].pid == 0) { /* Slave starts up here */ | |
845 | for (j = 0; j <= i; j++) | |
846 | (void) close(slaves[j].fd); | |
847 | signal(SIGINT, SIG_IGN); /* Master handles this */ | |
ddd2ef55 SP |
848 | #if defined(SIGINFO) |
849 | signal(SIGINFO, SIG_IGN); | |
850 | #endif | |
851 | ||
1227625a | 852 | #ifdef LINUX_FORK_BUG |
ddd2ef55 | 853 | if (atomic_write( cmd[0], (char *) &i, sizeof i) |
1227625a SP |
854 | != sizeof i) |
855 | quit("master/slave protocol botched 3\n"); | |
856 | #endif | |
857 | doslave(cmd[0], i); | |
858 | Exit(X_FINOK); | |
859 | } | |
860 | } | |
861 | ||
862 | #ifdef LINUX_FORK_BUG | |
863 | /* | |
864 | * Wait for all slaves to _actually_ start to circumvent a bug in | |
865 | * Linux kernels >= 2.1.3 where a signal sent to a child that hasn't | |
866 | * returned from fork() causes a SEGV in the child process | |
867 | */ | |
868 | for (i = 0; i < SLAVES; i++) | |
ddd2ef55 | 869 | if (atomic_read( slaves[i].fd, (char *) &j, sizeof j) != sizeof j) |
1227625a SP |
870 | quit("master/slave protocol botched 4\n"); |
871 | #endif | |
872 | ||
1227625a | 873 | for (i = 0; i < SLAVES; i++) |
ddd2ef55 | 874 | (void) atomic_write( slaves[i].fd, |
1227625a SP |
875 | (char *) &slaves[(i + 1) % SLAVES].pid, |
876 | sizeof slaves[0].pid); | |
877 | ||
878 | master = 0; | |
879 | } | |
880 | ||
881 | void | |
ddd2ef55 | 882 | killall(void) |
1227625a SP |
883 | { |
884 | register int i; | |
885 | ||
886 | for (i = 0; i < SLAVES; i++) | |
b45f51d6 | 887 | if (slaves[i].pid > 0) { |
1227625a | 888 | (void) kill(slaves[i].pid, SIGKILL); |
b45f51d6 SP |
889 | slaves[i].sent = 0; |
890 | } | |
1227625a SP |
891 | } |
892 | ||
893 | /* | |
894 | * Synchronization - each process has a lockfile, and shares file | |
895 | * descriptors to the following process's lockfile. When our write | |
896 | * completes, we release our lock on the following process's lock- | |
897 | * file, allowing the following process to lock it and proceed. We | |
898 | * get the lock back for the next cycle by swapping descriptors. | |
899 | */ | |
900 | static void | |
ddd2ef55 | 901 | doslave(int cmd, int slave_number) |
1227625a SP |
902 | { |
903 | register int nread; | |
ddd2ef55 SP |
904 | int nextslave, size, eot_count; |
905 | volatile int wrote = 0; | |
906 | sigset_t sigset; | |
1227625a SP |
907 | #ifdef __linux__ |
908 | errcode_t retval; | |
909 | #endif | |
910 | ||
1227625a SP |
911 | /* |
912 | * Need our own seek pointer. | |
913 | */ | |
914 | (void) close(diskfd); | |
915 | if ((diskfd = open(disk, O_RDONLY)) < 0) | |
916 | quit("slave couldn't reopen disk: %s\n", strerror(errno)); | |
917 | #ifdef __linux__ | |
918 | ext2fs_close(fs); | |
c930abff | 919 | retval = dump_fs_open(disk, &fs); |
1227625a | 920 | if (retval) |
c930abff | 921 | quit("slave couldn't reopen disk: %s\n", error_message(retval)); |
1227625a SP |
922 | #endif /* __linux__ */ |
923 | ||
924 | /* | |
925 | * Need the pid of the next slave in the loop... | |
926 | */ | |
ddd2ef55 | 927 | if ((nread = atomic_read( cmd, (char *)&nextslave, sizeof nextslave)) |
1227625a SP |
928 | != sizeof nextslave) { |
929 | quit("master/slave protocol botched - didn't get pid of next slave.\n"); | |
930 | } | |
931 | ||
932 | /* | |
933 | * Get list of blocks to dump, read the blocks into tape buffer | |
934 | */ | |
ddd2ef55 | 935 | while ((nread = atomic_read( cmd, (char *)slp->req, reqsiz)) == reqsiz) { |
1227625a SP |
936 | register struct req *p = slp->req; |
937 | ||
938 | for (trecno = 0; trecno < ntrec; | |
939 | trecno += p->count, p += p->count) { | |
940 | if (p->dblk) { | |
941 | bread(p->dblk, slp->tblock[trecno], | |
942 | p->count * TP_BSIZE); | |
943 | } else { | |
ddd2ef55 | 944 | if (p->count != 1 || atomic_read( cmd, |
b45f51d6 | 945 | (char *)slp->tblock[trecno], |
1227625a SP |
946 | TP_BSIZE) != TP_BSIZE) |
947 | quit("master/slave protocol botched.\n"); | |
948 | } | |
949 | } | |
950 | if (setjmp(jmpbuf) == 0) { | |
951 | ready = 1; | |
952 | if (!caught) | |
953 | (void) pause(); | |
954 | } | |
955 | ready = 0; | |
956 | caught = 0; | |
957 | ||
958 | /* Try to write the data... */ | |
b45f51d6 | 959 | wrote = 0; |
1227625a SP |
960 | eot_count = 0; |
961 | size = 0; | |
962 | ||
963 | while (eot_count < 10 && size < writesize) { | |
964 | #ifdef RDUMP | |
965 | if (host) | |
966 | wrote = rmtwrite(slp->tblock[0]+size, | |
967 | writesize-size); | |
968 | else | |
969 | #endif | |
970 | wrote = write(tapefd, slp->tblock[0]+size, | |
971 | writesize-size); | |
972 | #ifdef WRITEDEBUG | |
b45f51d6 | 973 | printf("slave %d wrote %d\n", slave_number, wrote); |
1227625a | 974 | #endif |
b45f51d6 | 975 | if (wrote < 0) |
1227625a SP |
976 | break; |
977 | if (wrote == 0) | |
978 | eot_count++; | |
979 | size += wrote; | |
980 | } | |
981 | ||
982 | #ifdef WRITEDEBUG | |
b45f51d6 SP |
983 | if (size != writesize) |
984 | printf("slave %d only wrote %d out of %d bytes and gave up.\n", | |
1227625a SP |
985 | slave_number, size, writesize); |
986 | #endif | |
987 | ||
b45f51d6 SP |
988 | /* |
989 | * Handle ENOSPC as an EOT condition. | |
990 | */ | |
991 | if (wrote < 0 && errno == ENOSPC) { | |
992 | wrote = 0; | |
993 | eot_count++; | |
994 | } | |
995 | ||
1227625a SP |
996 | if (eot_count > 0) |
997 | size = 0; | |
998 | ||
b45f51d6 | 999 | if (wrote < 0) { |
1227625a | 1000 | (void) kill(master, SIGUSR1); |
ddd2ef55 | 1001 | sigemptyset(&sigset); |
1227625a | 1002 | for (;;) |
ddd2ef55 | 1003 | sigsuspend(&sigset); |
1227625a SP |
1004 | } else { |
1005 | /* | |
1006 | * pass size of write back to master | |
1007 | * (for EOT handling) | |
1008 | */ | |
ddd2ef55 | 1009 | (void) atomic_write( cmd, (char *)&size, sizeof size); |
b45f51d6 | 1010 | } |
1227625a SP |
1011 | |
1012 | /* | |
1013 | * If partial write, don't want next slave to go. | |
1014 | * Also jolts him awake. | |
1015 | */ | |
1016 | (void) kill(nextslave, SIGUSR2); | |
1017 | } | |
1018 | if (nread != 0) | |
1019 | quit("error reading command pipe: %s\n", strerror(errno)); | |
1020 | } | |
1021 | ||
1022 | /* | |
1023 | * Since a read from a pipe may not return all we asked for, | |
1024 | * or a write may not write all we ask if we get a signal, | |
1025 | * loop until the count is satisfied (or error). | |
1026 | */ | |
ddd2ef55 SP |
1027 | static ssize_t |
1028 | atomic_read(int fd, void *buf, size_t count) | |
1029 | { | |
1030 | int got, need = count; | |
1031 | ||
1032 | while ((got = read(fd, buf, need)) > 0 && (need -= got) > 0) | |
1033 | (char *)buf += got; | |
1034 | return (got < 0 ? got : count - need); | |
1035 | } | |
1036 | ||
1037 | /* | |
1038 | * Since a read from a pipe may not return all we asked for, | |
1039 | * or a write may not write all we ask if we get a signal, | |
1040 | * loop until the count is satisfied (or error). | |
1041 | */ | |
1042 | static ssize_t | |
1043 | atomic_write(int fd, const void *buf, size_t count) | |
1227625a SP |
1044 | { |
1045 | int got, need = count; | |
1046 | ||
ddd2ef55 SP |
1047 | while ((got = write(fd, buf, need)) > 0 && (need -= got) > 0) |
1048 | (char *)buf += got; | |
1227625a SP |
1049 | return (got < 0 ? got : count - need); |
1050 | } |